Merge branch 'develop' into wavelet_model

.gitignore

@@ -46,3 +46,4 @@ build*
 
 # GDB temporary files
 .gdb_history
+install.dir*

CMakeLists.txt

@@ -21,6 +21,19 @@ 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)
 
+option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
+
+if(USE_BITINT_EXTENSION_INT4)
+    add_compile_definitions(CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4)
+    add_compile_options(-Wno-bit-int-extension)
+    message("CK compiled with USE_BITINT_EXTENSION_INT4 set to ${USE_BITINT_EXTENSION_INT4}")
+endif()
+
+## Threads
+set(THREADS_PREFER_PTHREAD_FLAG ON)
+find_package(Threads REQUIRED)
+link_libraries(Threads::Threads)
+
 ## C++
 enable_language(CXX)
 set(CMAKE_CXX_STANDARD 17)
@@ -70,6 +83,8 @@ if( DEFINED CK_OVERRIDE_HIP_VERSION_PATCH )
     message(STATUS "CK_HIP_VERSION_PATCH overriden with ${CK_OVERRIDE_HIP_VERSION_PATCH}")
 endif()
 message(STATUS "Build with HIP ${HIP_VERSION}")
+link_libraries(hip::device)
+add_compile_definitions(__HIP_PLATFORM_HCC__=1)
 
 ## tidy
 include(EnableCompilerWarnings)
@@ -219,6 +234,7 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)
 include_directories(BEFORE
     ${PROJECT_SOURCE_DIR}/include
     ${PROJECT_SOURCE_DIR}/library/include
+    ${HIP_INCLUDE_DIRS}
 )
 
 

Dockerfile

-FROM ubuntu:18.04
+FROM ubuntu:20.04
 
-ARG ROCMVERSION=5.1
-ARG OSDB_BKC_VERSION
+ARG ROCMVERSION=5.2.3
 ARG compiler_version
 
 RUN set -xe
 
-ARG BUILD_THREADS=8
 ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
 # Add rocm repository
 RUN apt-get update
@@ -20,8 +18,8 @@ RUN sh -c "echo deb https://apt.kitware.com/ubuntu/ bionic main | tee -a /etc/ap
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     apt-utils \
     build-essential \
-    cmake-data=3.15.1-0kitware1 \
-    cmake=3.15.1-0kitware1 \
+    cmake-data \
+    cmake \
     curl \
     git \
     hip-rocclr \
@@ -33,13 +31,11 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     llvm-amdgpu \
     pkg-config \
     python \
-    python3.8 \
+    python3 \
     python-dev \
     python3-dev \
-    python-pip \
     python3-pip \
     software-properties-common \
-    wget \
     rocm-dev \
     rocm-device-libs \
     rocm-cmake \

Jenkinsfile

@@ -19,10 +19,26 @@ def runShell(String command){
 }
 
 def getDockerImageName(){
-    def img = "${env.MIOPEN_IMAGE_URL}:composable_kernels_${params.COMPILER_VERSION}"
+    def img = "${env.CK_IMAGE_URL}:ck_ub20.04_rocm5.2.3_${params.COMPILER_VERSION}"
     return img
 }
 
+def build_compiler(){
+    def compiler
+    if (params.BUILD_COMPILER == "hipcc"){
+        compiler = '/opt/rocm/bin/hipcc'
+    }
+    else{
+        if (params.COMPILER_VERSION == "release"){
+            compiler = "/opt/rocm/llvm/bin/clang++"
+        }
+        else{
+            compiler = "/llvm-project/build/bin/clang++"
+        }        
+    }
+    return compiler
+}
+
 def getDockerImage(Map conf=[:]){
     env.DOCKER_BUILDKIT=1
     def prefixpath = conf.get("prefixpath", "/opt/rocm") // prefix:/opt/rocm
@@ -103,7 +119,7 @@ def buildDocker(install_prefix){
 
 def cmake_build(Map conf=[:]){
 
-    def compiler = conf.get("compiler","/opt/rocm/bin/hipcc")
+    def compiler = build_compiler()
     def config_targets = conf.get("config_targets","check")
     def debug_flags = "-g -fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize-recover=undefined " + conf.get("extradebugflags", "")
     def build_envs = "CTEST_PARALLEL_LEVEL=4 " + conf.get("build_env","")
@@ -185,7 +201,6 @@ def buildHipClangJob(Map conf=[:]){
         if (conf.get("enforce_xnack_on", false)) {
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 --env GPU_ARCH='${gpu_arch}' "
         }
-        //def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg GPU_ARCH='${gpu_arch}' --build-arg compiler_version='${params.COMPILER_VERSION}' "
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' "
         if (params.COMPILER_VERSION != "release"){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
@@ -337,6 +352,8 @@ def runCKProfiler(Map conf=[:]){
                             archiveArtifacts "perf_conv_bwd_data_${gpu_arch}.log"
                             archiveArtifacts "perf_gemm_bilinear_${gpu_arch}.log"
                             archiveArtifacts "perf_reduction_${gpu_arch}.log"
+                            archiveArtifacts "perf_splitK_gemm_${gpu_arch}.log"
+                            archiveArtifacts "perf_onnx_gemm_${gpu_arch}.log"
                            // stash perf files to master
                             stash name: "perf_gemm_${gpu_arch}.log"
                             stash name: "perf_resnet50_N256_${gpu_arch}.log"
@@ -347,6 +364,8 @@ def runCKProfiler(Map conf=[:]){
                             stash name: "perf_conv_bwd_data_${gpu_arch}.log"
                             stash name: "perf_gemm_bilinear_${gpu_arch}.log"
                             stash name: "perf_reduction_${gpu_arch}.log"
+                            stash name: "perf_splitK_gemm_${gpu_arch}.log"
+                            stash name: "perf_onnx_gemm_${gpu_arch}.log"
                             //we will process results on the master node
                         }
                         else{
@@ -427,6 +446,8 @@ def process_results(Map conf=[:]){
                         unstash "perf_conv_bwd_data_${gpu_arch}.log"
                         unstash "perf_gemm_bilinear_${gpu_arch}.log"
                         unstash "perf_reduction_${gpu_arch}.log"
+                        unstash "perf_splitK_gemm_${gpu_arch}.log"
+                        unstash "perf_onnx_gemm_${gpu_arch}.log"
                         sh "./process_qa_data.sh ${gpu_arch}"
                     }
                     else{
@@ -467,6 +488,10 @@ pipeline {
             name: 'COMPILER_VERSION', 
             defaultValue: 'ck-9110', 
             description: 'Specify which version of compiler to use: ck-9110 (default), release, or amd-stg-open.')
+        string(
+            name: 'BUILD_COMPILER', 
+            defaultValue: 'hipcc', 
+            description: 'Specify whether to build CK with hipcc (default) or with clang.')
         booleanParam(
             name: "RUN_FULL_QA",
             defaultValue: false,
@@ -549,7 +574,8 @@ pipeline {
                 {
                     agent{ label rocmnode("gfx908")}
                     environment{
-                        setup_args = """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
+                        //setup_args = """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
+                        setup_args = "${params.COMPILER_VERSION == "release" ? """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 " -DBUILD_DEV=On """ : """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """}"
                     }
                     steps{
                         buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "check", no_reboot:true, build_type: 'Release', gpu_arch: "gfx908")
@@ -561,9 +587,11 @@ pipeline {
                         beforeAgent true
                         expression { params.RUN_FULL_QA.toBoolean() }
                     }
+                    options { retry(2) }
                     agent{ label rocmnode("gfx90a")}
                     environment{
-                        setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx90a -O3 " -DBUILD_DEV=On """
+                        //setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx90a -O3 " -DBUILD_DEV=On """
+                        setup_args = "${params.COMPILER_VERSION == "release" ? """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx90a -O3 " -DBUILD_DEV=On """ : """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx90a -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """}"
                     }
                     steps{
                         buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "check", no_reboot:true, build_type: 'Release', gpu_arch: "gfx90a")
@@ -583,8 +611,11 @@ pipeline {
                 {
                     agent{ label rocmnode("gfx908")}
                     environment{
-                        setup_args = """ -D  -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DCMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc .. && make -j """ 
+                        //setup_args = """ -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " """
+                        setup_args = "${params.COMPILER_VERSION == "release" ? """ -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " """ : """ -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" """ }"
+                        //execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        execute_args = "${params.COMPILER_VERSION == "release" ? """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ : """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ }"
+
                     }
                     steps{
                         buildHipClangJobAndReboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
@@ -602,9 +633,11 @@ pipeline {
                         beforeAgent true
                         expression { !params.RUN_FULL_QA.toBoolean() && !params.TEST_NODE_PERFORMANCE.toBoolean() }
                     }
+                    options { retry(2) }
                     agent{ label rocmnode("gfx908")}
                     environment{
-                        setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
+                        //setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
+                        setup_args = "${params.COMPILER_VERSION == "release" ? """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 " -DBUILD_DEV=On """ : """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """}"
                    }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release', gpu_arch: "gfx908")
@@ -616,10 +649,12 @@ pipeline {
                         beforeAgent true
                         expression { params.RUN_FULL_QA.toBoolean() || params.TEST_NODE_PERFORMANCE.toBoolean() }
                     }
+                    options { retry(2) }
                     agent{ label rocmnode("gfx90a")}
                     environment{
-                        setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx90a -O3 " -DBUILD_DEV=On """
-                   }
+                        //setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx90a -O3 " -DBUILD_DEV=On """
+                        setup_args = "${params.COMPILER_VERSION == "release" ? """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx90a -O3 " -DBUILD_DEV=On """ : """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx90a -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """}"
+                    }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release', gpu_arch: "gfx90a")
                     }

client_example/03_gemm_layernorm/gemm_add_add_layernorm.cpp

@@ -128,11 +128,14 @@ bool RunDeviceNormalize2D(normalize_op_ptr& p_op,
     std::array<void*, 1> output      = {p_y};
     auto normalize_functor           = ck::tensor_operation::element_wise::Normalize{};
 
-    auto argument_ptr = p_op->MakeArgumentPointer(input,
+    std::array<ck::index_t, 2> xyLengths = {M, N};
+    std::array<ck::index_t, 2> xyStrides = {StrideX, 1};
+
+    auto argument_ptr = p_op->MakeArgumentPointer(xyLengths,
+                                                  {xyStrides, {1, 0}, {1, 0}, {0, 1}, {0, 1}},
+                                                  {xyStrides},
+                                                  input,
                                                   output,
-                                                  {M, N},
-                                                  {{StrideX, 1}, {1, 0}, {1, 0}, {0, 1}, {0, 1}},
-                                                  {{StrideX, 1}},
                                                   ck::tensor_operation::element_wise::Normalize{});
 
     if(p_op->IsSupportedArgument(argument_ptr.get()))

client_example/05_layernorm/CMakeLists.txt

+add_executable(client_layernorm2d layernorm2d.cpp)
+target_link_libraries(client_layernorm2d PRIVATE composable_kernel::device_operations)

client_example/05_layernorm/layernorm2d.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/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;
+
+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 Stride = 1024;
+
+    auto xy_size = (M - 1) * Stride + N;
+
+    SimpleDeviceMem x_device_buf(sizeof(XDataType) * xy_size);
+    SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * N);
+    SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * N);
+    SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceLayernorm<XDataType,
+                                                                   GammaDataType,
+                                                                   BetaDataType,
+                                                                   AccDataType,
+                                                                   YDataType,
+                                                                   PassThrough,
+                                                                   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;
+
+    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({M, N},      // lengths
+                                                        {Stride, 1}, // xStrides
+                                                        {1},         // gammaStrides
+                                                        {1},         // betaStrides
+                                                        {Stride, 1}, // yStrides
+                                                        {1},         // reduceDims
+                                                        1e-4,
+                                                        x_device_buf.GetDeviceBuffer(),
+                                                        gamma_device_buf.GetDeviceBuffer(),
+                                                        beta_device_buf.GetDeviceBuffer(),
+                                                        y_device_buf.GetDeviceBuffer(),
+                                                        PassThrough{});
+
+        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) * M * N + sizeof(GammaDataType) * N +
+                                   sizeof(BetaDataType) * N + sizeof(YDataType) * M * 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({M, N},      // lengths
+                                                        {Stride, 1}, // xStrides
+                                                        {1},         // gammaStrides
+                                                        {1},         // betaStrides
+                                                        {Stride, 1}, // yStrides
+                                                        {1},         // reduceDims
+                                                        1e-4,
+                                                        x_device_buf.GetDeviceBuffer(),
+                                                        gamma_device_buf.GetDeviceBuffer(),
+                                                        beta_device_buf.GetDeviceBuffer(),
+                                                        y_device_buf.GetDeviceBuffer(),
+                                                        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 0;
+}

client_example/06_softmax/CMakeLists.txt

+add_executable(client_softmax4d softmax4d.cpp)
+target_link_libraries(client_softmax4d PRIVATE composable_kernel::device_operations)

client_example/06_softmax/softmax4d.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, 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/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/softmax.hpp"
+
+using InDataType  = ck::half_t;
+using OutDataType = ck::half_t;
+using AccDataType = float;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr int Rank         = 4;
+constexpr int NumReduceDim = 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[])
+{
+    std::vector<ck::index_t> in_lengths{2, 8, 128, 1024};
+    std::vector<ck::index_t> in_strides{8 * 128 * 1024, 128 * 1024, 1024, 1};
+    std::vector<ck::index_t> reduce_dims{2, 3};
+
+    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};
+
+    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>;
+    // 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];
+
+        if(op_ptr->GetRank() != Rank || op_ptr->GetNumReduceDim() != NumReduceDim)
+        {
+            continue;
+        }
+
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in_lengths,
+                                                        in_strides,
+                                                        reduce_dims,
+                                                        &alpha,
+                                                        &beta,
+                                                        in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        PassThrough{},
+                                                        PassThrough{});
+        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_elements * sizeof(InDataType) +
+                                    (beta == 0.0f ? 1 : 2) * num_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
+    {
+        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,
+                                                        reduce_dims,
+                                                        &alpha,
+                                                        &beta,
+                                                        in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        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 0;
+}
\ No newline at end of file

client_example/CMakeLists.txt

@@ -10,3 +10,5 @@ add_subdirectory(01_gemm)
 add_subdirectory(02_gemm_add_add_fastgelu)
 add_subdirectory(03_gemm_layernorm)
 add_subdirectory(04_contraction)
+add_subdirectory(05_layernorm)
+add_subdirectory(06_softmax)

cmake/googletest.cmake

@@ -42,3 +42,8 @@ target_compile_options(gtest PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
 target_compile_options(gtest_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
 target_compile_options(gmock PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
 target_compile_options(gmock_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
+
+set_target_properties(gtest PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(gtest_main PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(gmock PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(gmock_main PROPERTIES POSITION_INDEPENDENT_CODE ON)

example/01_gemm/CMakeLists.txt

+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(USE_BITINT_EXTENSION_INT4)
+  add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
+  add_dependencies(example_gemm_dl example_gemm_dl_int4)
+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_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)
+
+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)
+endif(USE_BITINT_EXTENSION_INT4)
+
 add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
 # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
 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)

example/01_gemm/common.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <numeric>
+
+#include "ck/ck.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 "ck/utility/data_type.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/fill.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"
+
+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 StrideC = 4096;
+};
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+};
+
+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;
+
+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 == 10)
+    {
+        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.StrideC = std::stoi(argv[9]);
+    }
+    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 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl;
+        return false;
+    }
+
+    return true;
+}

example/01_gemm/gemm_dl_fp16.cpp

 // 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 "common.hpp"
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_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/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   = ck::half_t;
 using BDataType   = ck::half_t;
@@ -37,174 +14,24 @@ using ALayout = Col;
 using BLayout = Row;
 using CLayout = Row;
 
-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 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::
-        //  ########| 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| 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|               |      |      |      |      |   |           |           |       |              |              |        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,     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>;
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
+// ######|     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| 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|               |      |      |      |      |   |           |           |       |              |              |        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|                |                   |
+// ######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    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>;
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::
     ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    if(argc == 1)
-    {
-        // do nothing
-    }
-    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 == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    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");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(1);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    case 2:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = AElementOp{};
-    auto b_element_op = BElementOp{};
-    auto c_element_op = CElementOp{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-    bool pass = true;
-
-    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, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        pass = ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
-    }
+#include "run_gemm_example.inc"
 
-    return pass ? 0 : 1;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_dl_fp32.cpp

 // 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 "common.hpp"
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_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/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-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   = float;
 using BDataType   = float;
@@ -36,174 +14,24 @@ using ALayout = Col;
 using BLayout = Row;
 using CLayout = Row;
 
-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 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::
-        //  ########| 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| 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|               |      |      |      |      |   |           |           |       |              |              |        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<   F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 1>,      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, 1>,      S<2, 1, 4, 1>,      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, 1>, S<0, 1, 2, 3, 4, 5>,                5,                  4>;
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
+// ######|     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| 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|               |      |      |      |      |   |           |           |       |              |              |        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|                |                   |
+// ######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 1>,       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, 1>,      S<2, 1, 4, 1>,       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, 1>, S<0, 1, 2, 3, 4, 5>,               5,                  4>;
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::
     ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    if(argc == 1)
-    {
-        // do nothing
-    }
-    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 == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    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");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(1);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    case 2:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = AElementOp{};
-    auto b_element_op = BElementOp{};
-    auto c_element_op = CElementOp{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-    bool pass = true;
-
-    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, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        pass = ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
-    }
+#include "run_gemm_example.inc"
 
-    return pass ? 0 : 1;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_dl_int4.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
+#error Should compile this file with ck::int4_t support
+#endif
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
+
+using ADataType       = ck::int4_t;
+using BDataType       = ck::int4_t;
+using CDataType       = ck::int4_t;
+using KernelADataType = int8_t;
+using KernelBDataType = int8_t;
+using KernelCDataType = int8_t;
+using AccDataType     = int32_t;
+
+using ALayout = Col;
+using BLayout = Row;
+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::DeviceGemmDl
+// ######|           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| 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|               |      |      |      |      |   |           |           |       |              |              |        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|                |                   |
+// ######|                |                |                |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < KernelADataType, KernelBDataType, KernelCDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    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>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#define BUILD_INT4_EXAMPLE
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_dl_int8.cpp

 // 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 "common.hpp"
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_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/reference_tensor_operation/cpu/reference_gemm.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 ADataType   = int8_t;
 using BDataType   = int8_t;
@@ -34,174 +14,24 @@ using ALayout = Col;
 using BLayout = Row;
 using CLayout = Row;
 
-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 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::
-        // #########|  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| 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|               |      |      |      |      |   |           |           |       |              |              |        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>;
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
+// ######|     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| 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|               |      |      |      |      |   |           |           |       |              |              |        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|                |                   |
+// ######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    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>;
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::
     ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    if(argc == 1)
-    {
-        // do nothing
-    }
-    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 == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    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");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(1);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    case 2:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = AElementOp{};
-    auto b_element_op = BElementOp{};
-    auto c_element_op = CElementOp{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-    bool pass = true;
-
-    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, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        pass = ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
-    }
+#include "run_gemm_example.inc"
 
-    return pass ? 0 : 1;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_bf16.cpp

 // 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 "common.hpp"
 
-#include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.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/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/utility/check_err.hpp"
+using ADataType        = ck::bhalf_t;
+using BDataType        = ck::bhalf_t;
+using CDataType        = ck::bhalf_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::bhalf_t;
 
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
 
-using BF16 = ck::bhalf_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   = BF16;
-using BDataType   = BF16;
-using CDataType   = BF16;
-using AccDataType = F32;
-
-using ALayout = ck::tensor_layout::gemm::RowMajor;
-using BLayout = ck::tensor_layout::gemm::ColumnMajor;
-using CLayout = ck::tensor_layout::gemm::RowMajor;
+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,                    // typename ALayout
-     BLayout,                    // typename BLayout
-     CLayout,                    // typename CLayout
-     ADataType,                  // typename ADataType
-     BDataType,                  // typename BDataType
-     CDataType,                  // typename CDataType
-     AccDataType,                // typename GemmAccDataType
-     CDataType,                  // typename CShuffleDataType
-     PassThrough,                // typename AElementwiseOperation
-     PassThrough,                // typename BElementwiseOperation
-     PassThrough,                // typename CElementwiseOperation
-     GemmDefault,                // GemmSpecialization GemmSpec
-     1,                          // index_t NumGemmKPrefetchStage
-     256,                        // index_t BlockSize
-     256,                        // index_t MPerBlock
-     128,                        // index_t NPerBlock
-     32,                         // index_t KPerBlock
-     8,                          // index_t AK1
-     8,                          // index_t BK1
-     32,                         // index_t MPerXDL
-     32,                         // index_t NPerXDL
-     4,                          // index_t MXdlPerWave
-     2,                          // index_t NXdlPerWave
-     S<4, 64, 1>,                // typename ABlockTransferThreadClusterLengths_AK0_M_AK1
-     S<1, 0, 2>,                 // typename ABlockTransferThreadClusterArrangeOrder
-     S<1, 0, 2>,                 // typename ABlockTransferSrcAccessOrder
-     2,                          // index_t ABlockTransferSrcVectorDim
-     8,                          // index_t ABlockTransferSrcScalarPerVector
-     8,                          // index_t ABlockTransferDstScalarPerVector_AK1
-     1,                          // index_t ABlockLdsExtraM
-     S<4, 64, 1>,                // typename BBlockTransferThreadClusterLengths_BK0_N_BK1
-     S<1, 0, 2>,                 // typename BBlockTransferThreadClusterArrangeOrder
-     S<1, 0, 2>,                 // typename BBlockTransferSrcAccessOrder
-     2,                          // index_t BBlockTransferSrcVectorDim
-     8,                          // index_t BBlockTransferSrcScalarPerVector
-     8,                          // index_t BBlockTransferDstScalarPerVector_BK1
-     1,                          // index_t BBlockLdsExtraN
-     1,                          // index_t CShuffleMXdlPerWavePerShuffle
-     1,                          // index_t CShuffleNXdlPerWavePerShuffle
-     S<1, 32, 1, 8>,             // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
-     8>;                         // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
+// ######| 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,    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 ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
-                                                                        BDataType,
-                                                                        CDataType,
-                                                                        AccDataType,
-                                                                        PassThrough,
-                                                                        PassThrough,
-                                                                        PassThrough>;
-
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    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");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(0);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = PassThrough{};
-    auto b_element_op = PassThrough{};
-    auto c_element_op = PassThrough{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-        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_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
 
-        return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
-    }
+#include "run_gemm_example.inc"
 
-    return 0;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_fp16.cpp

 // 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 "common.hpp"
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_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/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 = F32;
-using CDataType        = F16;
+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;
@@ -48,18 +25,20 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 
 // clang-format off
 using DeviceGemmInstance0 = ck::tensor_operation::device::DeviceGemmXdl
-//######|     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|
-//######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  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|
-//######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
-        < 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>;
+// ######|     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|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  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|
+// ######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+         < 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 off
 using DeviceGemmInstance1 = 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,    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>;
+// ######| 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,    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>;
 
 using DeviceGemmInstance2 = ck::tensor_operation::device::DeviceGemm_Xdl_WaveletModel_CShuffle
 //######| 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|
@@ -74,162 +53,6 @@ using DeviceGemmInstance = DeviceGemmInstance2;
 using ReferenceGemmInstance = ck::tensor_operation::host::
     ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    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 == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    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, StrideC\n");
-        exit(0);
-    }
-
-    const ck::index_t PackedStrideA = ck::is_same_v<ALayout, Row> ? K : M;
-    const ck::index_t PackedStrideB = ck::is_same_v<BLayout, Row> ? N : K;
-    const ck::index_t PackedStrideC = ck::is_same_v<CLayout, Row> ? N : M;
-
-    StrideA = (StrideA < 0) ? PackedStrideA : StrideA;
-    StrideB = (StrideB < 0) ? PackedStrideB : StrideB;
-    StrideC = (StrideC < 0) ? PackedStrideC : StrideC;
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    case 2:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_Sequential<0>{});
-        b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = AElementOp{};
-    auto b_element_op = BElementOp{};
-    auto c_element_op = CElementOp{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_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, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
-    }
+#include "run_gemm_example.inc"
 
-    return 0;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_fp64.cpp

 // 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 "common.hpp"
 
-#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_gemm_xdl.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/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 F64 = double;
 
 using ADataType   = double;
 using BDataType   = double;
 using CDataType   = double;
 using AccDataType = double;
 
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using ALayout = ck::tensor_layout::gemm::RowMajor;
-using BLayout = ck::tensor_layout::gemm::ColumnMajor;
-using CLayout = ck::tensor_layout::gemm::RowMajor;
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
 
-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 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::DeviceGemmXdl
-//##########| 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|
-//##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  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|
-//##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+// ######|     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|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  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|
+// ######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
 #if 0
-             <  F64,   F64,   F64,     F64,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   64,    32,    32,     4,  1,   16,   16,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,      true,               7,               1>;
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,   GemmDefault,    64,    32,    32,     4,  1,   16,   16,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,      true,               7,               1>;
 #else
-             <  F64,   F64,   F64,     F64,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,  256,   128,   128,     4,  2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1>;
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,   GemmDefault,   256,   128,   128,     4,  2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1>;
 #endif
     // clang-format on
 
@@ -64,176 +41,6 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl
                                                                             BElementOp,
                                                                             CElementOp>;
 
-template <typename DataType>
-std::ostream& show_2d_matrix(std::ostream& os, Tensor<DataType>& matrix)
-{
-    os << "[" << std::endl;
-    for(int x = 0; x < matrix.mDesc.GetLengths()[0]; x++)
-    {
-        os << "[";
-        for(int y = 0; y < matrix.mDesc.GetLengths()[1]; y++)
-        {
-            os << std::setw(4) << static_cast<float>(matrix(x, y));
-        }
-        os << "]" << std::endl;
-    }
-    os << "]";
-    return os;
-}
-
-int main(int argc, char* argv[])
-{
-    bool do_verification = 0;
-    int init_method      = 0;
-    bool time_kernel     = false;
-
-    // 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 StrideC = 4096;
-
-    if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 10)
-    {
-        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]);
-        StrideC = std::stoi(argv[9]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: run kernel # of times (>1)\n");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(0);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            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}));
-            }
-            else
-            {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, 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<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-
-    std::cout << "data type: " << typeid(ADataType{}).name() << std::endl;
-    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_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});
-        break;
-    case 2:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_1<ADataType>{1});
-        b_k_n.GenerateTensorValue(GeneratorTensor_1<BDataType>{1});
-    }
-
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
-
-    auto a_element_op = AElementOp{};
-    auto b_element_op = BElementOp{};
-    auto c_element_op = CElementOp{};
-
-    // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-                                      M,
-                                      N,
-                                      K,
-                                      StrideA,
-                                      StrideB,
-                                      StrideC,
-                                      a_element_op,
-                                      b_element_op,
-                                      c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
-    {
-        std::cout << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    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(CDataType) * 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;
-
-    c_m_n_device_buf.FromDevice(c_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, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-#if 0
-        {
-            show_2d_matrix(std::cout << "a : ", a_m_k) << std::endl;
-            show_2d_matrix(std::cout << "b: ", b_k_n) << std::endl;
-            show_2d_matrix(std::cout << "c_device: ", c_m_n_device_result) << std::endl;
-            show_2d_matrix(std::cout << "c_host  :", c_m_n_host_result) << std::endl;
-        }
-#endif
-        return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
-    }
+#include "run_gemm_example.inc"
 
-    return 0;
-}
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }