Merge branch 'develop' into add_fp16_wmma_conv_instance

.pre-commit-config.yaml

+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++]

CMakeLists.txt

@@ -5,6 +5,31 @@ project(composable_kernel)
 
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
 
+if (DTYPES)
+        add_definitions(-DDTYPES)
+        if (DTYPES MATCHES "int8")
+                add_definitions(-D__int8__)
+        endif()
+        if (DTYPES MATCHES "fp8")
+                add_definitions(-D__fp8__)
+        endif()
+        if (DTYPES MATCHES "fp16")
+                add_definitions(-D__fp16__)
+        endif()
+        if (DTYPES MATCHES "fp32")
+                add_definitions(-D__fp32__)
+        endif()
+        if (DTYPES MATCHES "fp64")
+                add_definitions(-D__fp64__)
+        endif()
+        if (DTYPES MATCHES "bf16")
+                add_definitions(-D__bf16__)
+        endif()
+        message("DTYPES macro set to ${DTYPES}")
+else()
+        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+endif()
+
 enable_testing()
 
 set(ROCM_SYMLINK_LIBS OFF)
@@ -16,11 +41,24 @@ include(ROCMSetupVersion)
 include(ROCMInstallSymlinks)
 include(ROCMCreatePackage)
 include(CheckCXXCompilerFlag)
-
+include(ROCMCheckTargetIds)
 rocm_setup_version(VERSION 0.2.0)
 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)
 
+message("GPU_TARGETS= ${GPU_TARGETS}")
+
+message("checking which targets are supported")
+#This is the list of targets to be used in case GPU_TARGETS is not set on command line
+#These targets will be filtered and only supported ones will be used
+#Setting GPU_TARGETS on command line will override this list
+rocm_check_target_ids(DEFAULT_GPU_TARGETS
+    TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102"
+)
+message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
+set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " ")
+find_package(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)
 

Dockerfile

@@ -12,28 +12,27 @@ RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 RUN chmod 1777 /tmp
 RUN apt-get update
 RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-RUN if [ "$ROCMVERSION" != "5.6" ] && [ "$ROCMVERSION" != "5.7" ]; then \
+
+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] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
-         amdgpu-install -y --usecase=rocm --no-dkms; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc4" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
-         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-62  > /etc/apt/sources.list.d/rocm-build.list' && \
-         amdgpu-repo --amdgpu-build=1609671 && amdgpu-install -y --usecase=rocm --no-dkms; \
+        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 && amdgpu-install -y --usecase=rocm --no-dkms; \
+         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
     fi
 
-RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
-RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+RUN amdgpu-install -y --usecase=rocm --no-dkms
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \

Jenkinsfile

@@ -11,6 +11,20 @@ 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")
@@ -19,7 +33,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.6" && params.ROCMVERSION != "5.7"){
+    if (params.ROCMVERSION != "5.7"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }
@@ -219,7 +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", """
@@ -461,7 +476,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -482,7 +497,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -494,7 +509,8 @@ def Build_CK(Map conf=[:]){
                     cmake_build(conf)
                     dir("build"){
                         //run tests and examples
-                        sh 'make -j\$(( \$(nproc) / 2 )) check'
+                        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
@@ -598,7 +614,7 @@ 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
-                                              0 21 * * * % ROCMVERSION=5.5;COMPILER_VERSION=release;COMPILER_COMMIT=
+                                              0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {
@@ -717,7 +733,7 @@ pipeline {
                         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 Navi")
+                stage("Build CK and run Tests on Navi21")
                 {
                     when {
                         beforeAgent true
@@ -733,6 +749,22 @@ pipeline {
                         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 Navi32")
+                {
+                    when {
+                        beforeAgent true
+                        expression { !params.RUN_FULL_QA.toBoolean() }
+                    }
+                    agent{ label rocmnode("navi32") }
+                    environment{
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DDTYPES="fp16;fp32;bf16" -DGPU_TARGETS="gfx1101" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -DDTYPES="fp16;fp32;bf16" -D CMAKE_CXX_COMPILER="${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')
+                    }
+                }
             }
         }
 

README.md

@@ -109,6 +109,24 @@ make install
 
 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
 

client_example/09_quantization/CMakeLists.txt

+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 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)
 
@@ -18,3 +19,4 @@ target_link_libraries(client_conv2d_fwd_perlayer_quantization PRIVATE composable
 
 add_executable(client_gemm_quantization gemm_quantization.cpp)
 target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_operations)
+endif()

client_example/11_grouped_conv_bwd_weight/common.hpp

@@ -101,13 +101,15 @@ template <ck::index_t NumDimSpatial,
           typename WeiLayout,
           typename OutLayout>
 bool run_grouped_conv_bwd_weight(
-    ck::index_t G,
-    ck::index_t N,
-    ck::index_t K,
-    ck::index_t C,
+    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,
@@ -157,6 +159,8 @@ bool run_grouped_conv_bwd_weight(
                                                         input_spatial_lengths,
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
                                                         input_left_pads,
@@ -224,6 +228,8 @@ bool run_grouped_conv_bwd_weight(
                                                         input_spatial_lengths,
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
                                                         input_left_pads,

client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp

@@ -22,6 +22,15 @@ 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()
 {
@@ -31,7 +40,19 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G, N, K, C, {Wi}, {X}, {Wo}, {1}, {1}, {1}, {1})
+                                       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;
 }

client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp

@@ -25,6 +25,17 @@ 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()
 {
@@ -34,8 +45,19 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(
-               G, N, K, C, {Hi, Wi}, {Y, X}, {Ho, Wo}, {1, 1}, {1, 1}, {1, 1}, {1, 1})
+                                       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;
 }

client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp

@@ -28,6 +28,17 @@ 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()
 {
@@ -41,13 +52,15 @@ int main()
                                                   N,
                                                   K,
                                                   C,
-                                                  {Di, Hi, Wi},
-                                                  {Z, Y, X},
-                                                  {Do, Ho, Wo},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1})
+                                                  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;
 }

client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp

@@ -28,6 +28,17 @@ 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()
 {
@@ -37,13 +48,16 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G,
+                                       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},

client_example/CMakeLists.txt

@@ -2,6 +2,31 @@ cmake_minimum_required(VERSION 3.15)
 project(ck_app)
 add_compile_options(-std=c++17)
 
+if (DTYPES)
+        add_definitions(-DDTYPES)
+        if (DTYPES MATCHES "int8")
+                add_definitions(-D__int8__)
+        endif()
+        if (DTYPES MATCHES "fp8")
+                add_definitions(-D__fp8__)
+        endif()
+        if (DTYPES MATCHES "fp16")
+                add_definitions(-D__fp16__)
+        endif()
+        if (DTYPES MATCHES "fp32")
+                add_definitions(-D__fp32__)
+        endif()
+        if (DTYPES MATCHES "fp64")
+                add_definitions(-D__fp64__)
+        endif()
+        if (DTYPES MATCHES "bf16")
+                add_definitions(-D__bf16__)
+        endif()
+        message("DTYPES macro set to ${DTYPES}")
+else()
+        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+endif()
+
 find_package(composable_kernel 1.0.0 COMPONENTS device_operations)
 find_package(hip REQUIRED PATHS /opt/rocm)
 message(STATUS "Build with HIP ${hip_VERSION}")

cmake/EnableCompilerWarnings.cmake

@@ -67,6 +67,7 @@ else()
             -Wunused
             -Wno-reserved-identifier
             -Werror
+            -Wno-option-ignored
             -Wsign-compare
             -Wno-extra-semi-stmt
         )

example/01_gemm/CMakeLists.txt

@@ -2,11 +2,14 @@ add_custom_target(example_gemm_dl)
 
 add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
 add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
-add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
 
 add_dependencies(example_gemm_dl example_gemm_dl_fp32)
 add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-add_dependencies(example_gemm_dl example_gemm_dl_int8)
+
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+   add_dependencies(example_gemm_dl example_gemm_dl_int8)
+endif()
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
@@ -19,13 +22,16 @@ 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(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
+endif()
+
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
@@ -44,3 +50,7 @@ if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS
   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()

example/01_gemm/gemm_xdl_f8.cpp

+// 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); }

example/01_gemm/gemm_xdl_wavelet_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 
 using ADataType        = ck::half_t;
 using BDataType        = ck::half_t;

example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp

@@ -3,7 +3,7 @@
 
 #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"
 

example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp

@@ -3,7 +3,7 @@
 
 #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"
 

example/09_convnd_fwd/convnd_fwd_dl_int8.cpp

@@ -3,7 +3,7 @@
 
 #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"
 

example/14_gemm_quantization/CMakeLists.txt

+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 # dlops
 add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 
@@ -11,3 +12,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
\ No newline at end of file