Merge branch 'origin/gfx950' into 'andriy/lwpck-2243'.

CMakeLists.txt

@@ -26,11 +26,15 @@ set(version 1.1.0)
 project(composable_kernel VERSION ${version} LANGUAGES CXX HIP)
 include(CTest)
 
+# Usage: for customized Python location cmake -DCK_USE_ALTERNATIVE_PYTHON="/opt/Python-3.8.13/bin/python3.8"
+# CK Codegen requires dataclass which is added in Python 3.7
+# Python version 3.8 is required for general good practice as it is default for Ubuntu 20.04
 if(NOT CK_USE_ALTERNATIVE_PYTHON)
-   find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)
+   find_package(Python3 3.8 COMPONENTS Interpreter REQUIRED)
 else()
    message("Using alternative python version")
    set(EXTRA_PYTHON_PATH)
+   # this is overly restrictive, we may need to be more flexible on the following
    string(REPLACE "/bin/python3.8" "" EXTRA_PYTHON_PATH "${CK_USE_ALTERNATIVE_PYTHON}")
    message("alternative python path is: ${EXTRA_PYTHON_PATH}")
    find_package(Python3 3.6 COMPONENTS Interpreter REQUIRED)

Dockerfile

@@ -23,6 +23,11 @@ RUN if [ "$ROCMVERSION" != "6.3" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
         sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
         sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
+    elif [ "$ROCMVERSION" = "6.3" ] && [ "$compiler_version" = "rc1" ]; then \
+        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_6.3.0.1-20.04-1_all.deb --no-check-certificate" && \
+        apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install dialog libpopt0 rsync && DEBIAN_FRONTEND=noninteractive apt-get install ./amdgpu-install-internal_6.3.0.1-20.04-1_all.deb && \
+        sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 6.3.0.1 rel-5 > /etc/apt/sources.list.d/rocm-build.list' && \
+        amdgpu-repo --amdgpu-build=2033700; \
     fi
 
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
@@ -130,6 +135,8 @@ ENV compiler_commit=$compiler_commit
 RUN sh -c "echo compiler version = '$compiler_version'"
 RUN sh -c "echo compiler commit = '$compiler_commit'"
 
+ARG DISABLE_CACHE=0
+
 RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline-open" ] ) && [ "$compiler_commit" = "" ]; then \
         git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
         cd llvm-project && mkdir build && cd build && \

Jenkinsfile

@@ -94,13 +94,21 @@ def getDockerImage(Map conf=[:]){
     env.DOCKER_BUILDKIT=1
     def prefixpath = conf.get("prefixpath", "/opt/rocm")
     def no_cache = conf.get("no_cache", false)
-    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${prefixpath} --build-arg CK_SCCACHE='${env.CK_SCCACHE}' --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
+    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${prefixpath} --build-arg CK_SCCACHE='${env.CK_SCCACHE}' --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' --build-arg DISABLE_CACHE='git rev-parse ${params.COMPILER_VERSION}' "
     if(no_cache)
     {
         dockerArgs = dockerArgs + " --no-cache "
     }
     echo "Docker Args: ${dockerArgs}"
-    def image = getDockerImageName()
+    def image
+    if ( params.BUILD_LEGACY_OS && conf.get("docker_name", "") != "" ){
+        image = conf.get("docker_name", "")
+        echo "Using legacy docker: ${image}"
+    }
+    else{
+        image = getDockerImageName()
+        echo "Using default docker: ${image}"
+    }
     //Check if image exists 
     def retimage
     try 
@@ -124,8 +132,10 @@ def buildDocker(install_prefix){
     checkout scm
     def image_name = getDockerImageName()
     echo "Building Docker for ${image_name}"
-    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${install_prefix} --build-arg CK_SCCACHE='${env.CK_SCCACHE}' --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-
+    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${install_prefix} --build-arg CK_SCCACHE='${env.CK_SCCACHE}' --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' --build-arg DISABLE_CACHE='git rev-parse ${params.COMPILER_VERSION}' "
+    if(params.COMPILER_VERSION == "amd-staging" || params.COMPILER_VERSION == "amd-mainline-open" || params.COMPILER_COMMIT != ""){
+        dockerArgs = dockerArgs + " --no-cache "
+    }
     echo "Build Args: ${dockerArgs}"
     try{
         if(params.BUILD_DOCKER){
@@ -259,6 +269,7 @@ def cmake_build(Map conf=[:]){
             """)
         sh cmd3
     }
+
     // reduce parallelism when compiling, clang uses too much memory
     def nt = nthreads()
     def cmd
@@ -273,7 +284,7 @@ def cmake_build(Map conf=[:]){
         }
         else{
             setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
-            build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make -j${nt} ${config_targets}")
+            build_cmd = conf.get("build_cmd", "${build_envs} make -j${nt} ${config_targets}")
         }
         cmd = conf.get("cmd", """
             ${setup_cmd}
@@ -292,8 +303,8 @@ def cmake_build(Map conf=[:]){
     dir("build"){
         //build CK
         sh cmd
-        //run tests
-        if(!setup_args.contains("NO_CK_BUILD")){
+        //run tests except when NO_CK_BUILD or BUILD_LEGACY_OS are set
+        if(!setup_args.contains("NO_CK_BUILD") && !params.BUILD_LEGACY_OS){
             if (setup_args.contains("gfx90a") && params.NINJA_BUILD_TRACE){
                 sh "/ninjatracing/ninjatracing .ninja_log > ck_build_trace.json"
                 archiveArtifacts "ck_build_trace.json"
@@ -330,7 +341,15 @@ def buildHipClangJob(Map conf=[:]){
         env.HSA_ENABLE_SDMA=0
         checkout scm
 
-        def image = getDockerImageName() 
+        def image
+        if ( params.BUILD_LEGACY_OS  && conf.get("docker_name", "") != "" ){
+            image = conf.get("docker_name", "")
+            echo "Using legacy docker: ${image}"
+        }
+        else{
+            image = getDockerImageName()
+            echo "Using default docker: ${image}"
+        }
         def prefixpath = conf.get("prefixpath", "/opt/rocm")
 
         // Jenkins is complaining about the render group 
@@ -352,7 +371,7 @@ def buildHipClangJob(Map conf=[:]){
         def retimage
         (retimage, image) = getDockerImage(conf)
 
-        gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
+        gitStatusWrapper(credentialsId: "${env.ck_git_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
             withDockerContainer(image: image, args: dockerOpts + ' -v=/var/jenkins/:/var/jenkins') {
                 timeout(time: 48, unit: 'HOURS')
                 {
@@ -407,7 +426,7 @@ def runCKProfiler(Map conf=[:]){
         def variant = env.STAGE_NAME
         def retimage
 
-        gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
+        gitStatusWrapper(credentialsId: "${env.ck_git_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
             try {
                 (retimage, image) = getDockerImage(conf)
                 withDockerContainer(image: image, args: dockerOpts) {
@@ -512,7 +531,16 @@ def Build_CK(Map conf=[:]){
         env.DOCKER_BUILDKIT=1
         checkout scm
 
-        def image = getDockerImageName() 
+        def image
+        if ( params.BUILD_LEGACY_OS  && conf.get("docker_name", "") != "" ){
+            image = conf.get("docker_name", "")
+            echo "Using legacy docker: ${image}"
+        }
+        else{
+            image = getDockerImageName()
+            echo "Using default docker: ${image}"
+        }
+
         def prefixpath = conf.get("prefixpath", "/opt/rocm")
 
         // Jenkins is complaining about the render group 
@@ -524,6 +552,9 @@ def Build_CK(Map conf=[:]){
         if (params.COMPILER_VERSION == "amd-staging" || params.COMPILER_VERSION == "amd-mainline-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
+        if(params.BUILD_LEGACY_OS){
+            dockerOpts = dockerOpts + " --env LD_LIBRARY_PATH='/opt/Python-3.8.13/lib' "
+        }
         def video_id = sh(returnStdout: true, script: 'getent group video | cut -d: -f3')
         def render_id = sh(returnStdout: true, script: 'getent group render | cut -d: -f3')
         dockerOpts = dockerOpts + " --group-add=${video_id} --group-add=${render_id} "
@@ -532,7 +563,7 @@ def Build_CK(Map conf=[:]){
         def variant = env.STAGE_NAME
         def retimage
 
-        gitStatusWrapper(credentialsId: "${env.status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
+        gitStatusWrapper(credentialsId: "${env.ck_git_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
             try {
                 (retimage, image) = getDockerImage(conf)
                 withDockerContainer(image: image, args: dockerOpts) {
@@ -637,7 +668,7 @@ def process_results(Map conf=[:]){
     def variant = env.STAGE_NAME
     def retimage
 
-    gitStatusWrapper(credentialsId: "${env.status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
+    gitStatusWrapper(credentialsId: "${env.ck_git_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel-internal') {
         try {
             (retimage, image) = getDockerImage(conf)
         }
@@ -765,9 +796,13 @@ pipeline {
             defaultValue: false,
             description: "Run the grouped conv large cases tests (default: OFF)")
         booleanParam(
-            name: "RUN_CK_TILE_TESTS",
+            name: "RUN_CK_TILE_FMHA_TESTS",
             defaultValue: false,
-            description: "Run the ck_tile tests (default: OFF)")
+            description: "Run the ck_tile FMHA tests (default: OFF)")
+        booleanParam(
+            name: "RUN_CK_TILE_GEMM_TESTS",
+            defaultValue: false,
+            description: "Run the ck_tile GEMM tests (default: OFF)")
         booleanParam(
             name: "BUILD_INSTANCES_ONLY",
             defaultValue: false,
@@ -780,6 +815,10 @@ pipeline {
             name: "NINJA_BUILD_TRACE",
             defaultValue: false,
             description: "Generate a ninja build trace (default: OFF)")
+        booleanParam(
+            name: "BUILD_LEGACY_OS",
+            defaultValue: false,
+            description: "Try building CK with legacy OS dockers: RHEL8 and SLES15 (default: OFF)")
     }
     environment{
         dbuser = "${dbuser}"
@@ -788,7 +827,7 @@ pipeline {
         dbsshport = "${dbsshport}"
         dbsshuser = "${dbsshuser}"
         dbsshpassword = "${dbsshpassword}"
-        status_wrapper_creds = "${status_wrapper_creds}"
+        ck_git_creds = "${ck_git_creds}"
         gerrit_cred="${gerrit_cred}"
         DOCKER_BUILDKIT = "1"
     }
@@ -884,15 +923,15 @@ pipeline {
                 }
             }
         }
-        stage("Run CK_TILE Tests")
+        stage("Run CK_TILE_FMHA Tests")
         {
             parallel
             {
-                stage("Run CK_TILE Tests on gfx90a")
+                stage("Run CK_TILE_FMHA Tests on gfx90a")
                 {
                     when {
                         beforeAgent true
-                        expression { params.RUN_CK_TILE_TESTS.toBoolean() }
+                        expression { params.RUN_CK_TILE_FMHA_TESTS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx90a") }
                     environment{
@@ -907,11 +946,11 @@ pipeline {
                         cleanWs()
                     }
                 }
-                stage("Run CK_TILE Tests on gfx942")
+                stage("Run CK_TILE_FMHA Tests on gfx942")
                 {
                     when {
                         beforeAgent true
-                        expression { params.RUN_CK_TILE_TESTS.toBoolean() }
+                        expression { params.RUN_CK_TILE_FMHA_TESTS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx942") }
                     environment{
@@ -928,21 +967,103 @@ pipeline {
                 }
             }
         }
+        stage("Run CK_TILE_GEMM Tests")
+        {
+            parallel
+            {
+                stage("Run CK_TILE_GEMM Tests on gfx90a")
+                {
+                    when {
+                        beforeAgent true
+                        expression { params.RUN_CK_TILE_GEMM_TESTS.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx90a") }
+                    environment{
+                        setup_args = "NO_CK_BUILD"
+                        execute_args = """ ../script/cmake-ck-dev.sh  ../ gfx90a && \
+                                           make -j64 tile_example_gemm_basic && \
+                                           cd ../ &&
+                                           example/ck_tile/03_gemm/script/run_full_test.sh "CI_${params.COMPILER_VERSION}" "${env.BRANCH_NAME}" "${NODE_NAME}" gfx90a """
+                   }
+                    steps{
+                        buildHipClangJobAndReboot(setup_args:setup_args, no_reboot:true, build_type: 'Release', execute_cmd: execute_args)
+                        cleanWs()
+                    }
+                }
+                stage("Run CK_TILE_GEMM Tests on gfx942")
+                {
+                    when {
+                        beforeAgent true
+                        expression { params.RUN_CK_TILE_GEMM_TESTS.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx942") }
+                    environment{
+                        setup_args = "NO_CK_BUILD"
+                        execute_args = """ ../script/cmake-ck-dev.sh  ../ gfx942 && \
+                                           make -j64 tile_example_gemm_basic && \
+                                           cd ../ &&
+                                           example/ck_tile/03_gemm/script/run_full_test.sh "CI_${params.COMPILER_VERSION}" "${env.BRANCH_NAME}" "${NODE_NAME}" gfx942 """
+                   }
+                    steps{
+                        buildHipClangJobAndReboot(setup_args:setup_args, no_reboot:true, build_type: 'Release', execute_cmd: execute_args)
+                        cleanWs()
+                    }
+                }
+            }
+        }
+
 		stage("Build CK and run Tests")
         {
             parallel
             {
+                stage("Build CK with RHEL8")
+                {
+                    when {
+                        beforeAgent true
+                        expression { params.BUILD_LEGACY_OS.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx90a") }
+                    environment{
+                        def docker_name = "${env.CK_DOCKERHUB_PRIVATE}:ck_rhel8_rocm6.3"
+                        setup_args = """ -DGPU_TARGETS="gfx942" \
+                                         -DCMAKE_CXX_FLAGS=" -O3 " \
+                                         -DCK_USE_ALTERNATIVE_PYTHON=/opt/Python-3.8.13/bin/python3.8 """
+                        execute_args = " "
+                   }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: " ", no_reboot:true, build_type: 'Release', docker_name: docker_name)
+                        cleanWs()
+                    }
+                }
+                stage("Build CK with SLES15")
+                {
+                    when {
+                        beforeAgent true
+                        expression { params.BUILD_LEGACY_OS.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx90a") }
+                    environment{
+                        def docker_name = "${env.CK_DOCKERHUB_PRIVATE}:ck_sles15_rocm6.3"
+                        setup_args = """ -DGPU_TARGETS="gfx942" \
+                                         -DCMAKE_CXX_FLAGS=" -O3 " \
+                                         -DCK_USE_ALTERNATIVE_PYTHON=/opt/Python-3.8.13/bin/python3.8 """
+                        execute_args = " "
+                   }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: " ", no_reboot:true, build_type: 'Release', docker_name: docker_name)
+                        cleanWs()
+                    }
+                }
                 stage("Build CK for all gfx9 targets")
                 {
                     when {
                         beforeAgent true
-                        expression { params.RUN_FULL_QA.toBoolean() }
+                        expression { params.RUN_FULL_QA.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx90a") }
                     environment{
                         setup_args = """ -DCMAKE_INSTALL_PREFIX=../install \
                                          -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" \
-                                         -DCMAKE_EXE_LINKER_FLAGS=" -L ${env.WORKSPACE}/script -T hip_fatbin_insert " \
                                          -DCMAKE_CXX_FLAGS=" -O3 " """
                         execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
                                            cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
@@ -959,7 +1080,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { params.RUN_FULL_QA.toBoolean() }
+                        expression { params.RUN_FULL_QA.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx942") }
                     environment{
@@ -979,7 +1100,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() }
+                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx90a") }
                     environment{
@@ -999,7 +1120,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { params.BUILD_INSTANCES_ONLY.toBoolean() && !params.RUN_FULL_QA.toBoolean() }
+                        expression { params.BUILD_INSTANCES_ONLY.toBoolean() && !params.RUN_FULL_QA.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx90a") }
                     environment{
@@ -1018,7 +1139,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() }
+                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx1030") }
                     environment{
@@ -1038,7 +1159,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() }
+                        expression { !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx1101") }
                     environment{
@@ -1058,7 +1179,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { params.BUILD_GFX12.toBoolean() && !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() }
+                        expression { params.BUILD_GFX12.toBoolean() && !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent{ label rocmnode("gfx1201") }
                     environment{
@@ -1085,7 +1206,7 @@ pipeline {
                 {
                     when {
                         beforeAgent true
-                        expression { params.RUN_PERFORMANCE_TESTS.toBoolean() }
+                        expression { params.RUN_PERFORMANCE_TESTS.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     options { retry(1) }
                     agent{ label rocmnode("gfx90a")}
@@ -1106,7 +1227,7 @@ pipeline {
                 stage("Process results"){
                     when {
                         beforeAgent true
-                        expression { params.RUN_PERFORMANCE_TESTS.toBoolean() }
+                        expression { params.RUN_PERFORMANCE_TESTS.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() }
                     }
                     agent { label 'mici' }
                     steps{

codegen/test/CMakeLists.txt

@@ -6,6 +6,9 @@ if(NOT INSTANCES_ONLY)
     set_source_files_properties(${TEST_SRC} PROPERTIES LANGUAGE HIP)
     get_filename_component(BASE_NAME ${TEST_SRC} NAME_WE)
     add_executable(codegen_test_${BASE_NAME} ${TEST_SRC})
+    if(CK_USE_ALTERNATIVE_PYTHON)
+      target_link_options(codegen_test_${BASE_NAME} PRIVATE -lstdc++fs)
+    endif()
     add_dependencies(codegen codegen_test_${BASE_NAME})
     add_dependencies(tests codegen_test_${BASE_NAME})
     add_dependencies(check codegen_test_${BASE_NAME})

codegen/test/rtc/include/rtc/compile_kernel.hpp

@@ -2,14 +2,14 @@
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_COMPILE_KERNEL
 
 #include <rtc/kernel.hpp>
-#include <filesystem>
+#include <ck/filesystem.hpp>
 #include <string>
 
 namespace rtc {
 
 struct src_file
 {
-    std::filesystem::path path;
+    CK::fs::path path;
     std::string_view content;
 };
 

codegen/test/rtc/include/rtc/tmp_dir.hpp

@@ -2,13 +2,13 @@
 #define GUARD_HOST_TEST_RTC_INCLUDE_RTC_TMP_DIR
 
 #include <string>
-#include <filesystem>
+#include <ck/filesystem.hpp>
 
 namespace rtc {
 
 struct tmp_dir
 {
-    std::filesystem::path path;
+    CK::fs::path path;
     tmp_dir(const std::string& prefix = "");
 
     void execute(const std::string& cmd) const;

codegen/test/rtc/src/compile_kernel.cpp

@@ -70,9 +70,9 @@ kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options
 
     for(const auto& src : srcs)
     {
-        std::filesystem::path full_path   = td.path / src.path;
-        std::filesystem::path parent_path = full_path.parent_path();
-        std::filesystem::create_directories(parent_path);
+        CK::fs::path full_path   = td.path / src.path;
+        CK::fs::path parent_path = full_path.parent_path();
+        CK::fs::create_directories(parent_path);
         write_string(full_path.string(), src.content);
         if(src.path.extension().string() == ".cpp")
         {
@@ -86,7 +86,7 @@ kernel compile_kernel(const std::vector<src_file>& srcs, compile_options options
     td.execute(compiler() + options.flags);
 
     auto out_path = td.path / out;
-    if(not std::filesystem::exists(out_path))
+    if(not CK::fs::exists(out_path))
         throw std::runtime_error("Output file missing: " + out);
 
     auto obj = read_buffer(out_path.string());

codegen/test/rtc/src/tmp_dir.cpp

@@ -31,10 +31,10 @@ std::string unique_string(const std::string& prefix)
 }
 
 tmp_dir::tmp_dir(const std::string& prefix)
-    : path(std::filesystem::temp_directory_path() /
+    : path(CK::fs::temp_directory_path() /
            unique_string(prefix.empty() ? "ck-rtc" : "ck-rtc-" + prefix))
 {
-    std::filesystem::create_directories(this->path);
+    CK::fs::create_directories(this->path);
 }
 
 void tmp_dir::execute(const std::string& cmd) const
@@ -43,6 +43,6 @@ void tmp_dir::execute(const std::string& cmd) const
     std::system(s.c_str());
 }
 
-tmp_dir::~tmp_dir() { std::filesystem::remove_all(this->path); }
+tmp_dir::~tmp_dir() { CK::fs::remove_all(this->path); }
 
 } // namespace rtc

docs/sphinx/requirements.in

-rocm-docs-core==1.7.2
-sphinxcontrib-bibtex==2.6.2
+rocm-docs-core==1.8.2
+sphinxcontrib-bibtex==2.6.3

docs/sphinx/requirements.txt

@@ -103,7 +103,7 @@ requests==2.32.3
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==1.7.2
+rocm-docs-core==1.8.2
     # via -r requirements.in
 six==1.16.0
     # via pybtex
@@ -137,7 +137,7 @@ sphinx-notfound-page==1.0.3
     # via rocm-docs-core
 sphinxcontrib-applehelp==2.0.0
     # via sphinx
-sphinxcontrib-bibtex==2.6.2
+sphinxcontrib-bibtex==2.6.3
     # via -r requirements.in
 sphinxcontrib-devhelp==2.0.0
     # via sphinx

example/01_gemm/run_gemm_example.inc

@@ -305,6 +305,14 @@ bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
         }
 #endif
     }
+    else
+    {
+        // When the Problem Type and Problem Size does not fit.
+
+        std::cerr << gemm.GetTypeString() << ": the instance does not support the problem config."
+                  << std::endl;
+        return true;
+    }
 
     std::size_t flop = 2_uz * M * N * K;
     std::size_t num_btype =

example/01_gemm/run_gemm_example_v2.inc

@@ -161,18 +161,6 @@ bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
         a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
         b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
     }
-#if 0
-    printf("B matrix:\n");
-    for (int in = 0; in < N; in++)
-    {
-        for (int ik = 0; ik < K; ik++)
-        {
-            printf("%02x ", *(reinterpret_cast<uint8_t*>(&b_k_n(ik,in))));
-            if(ik%8==7) printf("|");
-        }
-        printf("\n");
-    }
-#endif
 
     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{}));
@@ -272,7 +260,8 @@ bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
 
     if(config.time_kernel)
     {
-        ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        ave_time =
+            invoker.Run(argument, StreamConfig{nullptr, config.time_kernel, 0, 5, 10, true, 4});
 
         std::size_t flop = 2_uz * M * N * K;
         std::size_t num_btype =

example/66_complex_contraction_bilinear/CMakeLists.txt

+add_example_executable(example_complex_contraction_bilinear_xdl_fp32 complex_contraction_bilinear_xdl_fp32.cpp)
+add_example_executable(example_complex_contraction_bilinear_xdl_fp64 complex_contraction_bilinear_xdl_fp64.cpp)
+

example/66_complex_contraction_bilinear/README.md

+# Instructions for ```example_complex_contraction_bilinear_xdl_fp32```
+
+## Run
+```bash
+#arg1: verification (0=no, 1=yes)
+#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
+#arg3: time kernel (0=no, 1=yes)
+./bin/example_contraction_bilinear_xdl_fp32 1 1 1
+```
+
+

example/66_complex_contraction_bilinear/complex_contraction_bilinear_xdl_fp32.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F32;
+using BDataType        = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F32;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F32;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_complex_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_complex_contraction_bilinear_example(argc, argv); }

example/66_complex_contraction_bilinear/complex_contraction_bilinear_xdl_fp64.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F64;
+using CShuffleDataType = F64;
+using DDataType        = F64;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F64;
+using ComputeDataType  = F64;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_complex_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_complex_contraction_bilinear_example(argc, argv); }

example/ck_tile/01_fmha/CMakeLists.txt

@@ -22,12 +22,20 @@ string(REPLACE ";" "," FMHA_FWD_APIS "${FMHA_FWD_ENABLE_APIS}")
 execute_process(
   COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_LIST_DIR}/generate.py
   --api ${FMHA_FWD_APIS} --list_blobs ${CMAKE_CURRENT_BINARY_DIR}/fwd_blob_list.txt
+  RESULT_VARIABLE ret
 )
+if(ret AND NOT ret EQUAL 0)
+  message( FATAL_ERROR "CK Tile FMHA FAILED to genrate a list of FWD kernels via Python.")
+endif()
 
 execute_process(
   COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_LIST_DIR}/generate.py
   --api bwd --list_blobs ${CMAKE_CURRENT_BINARY_DIR}/bwd_blob_list.txt --receipt 3
+  RESULT_VARIABLE ret
 )
+if(ret AND NOT ret EQUAL 0)
+  message( FATAL_ERROR "CK Tile FMHA FAILED to genrate a list of BWD kernels via Python.")
+endif()
 
 # NOTE: for cmake, the FMHA_FWD_GEN_BLOBS/FMHA_BWD_GEN_BLOBS files must be in the same directory
 #       as current cmake list, otherwise will not figure out the dependency properly

example/ck_tile/01_fmha/fmha_bwd.cpp

@@ -99,13 +99,26 @@ auto create_args(int argc, char* argv[])
 
 // different threshold for different dtype
 template <typename DataType>
-auto get_elimit(int /*init_method*/)
+auto get_elimit(ck_tile::index_t /*hdim_q*/, ck_tile::index_t /*hdim_v*/)
 {
     double rtol = 1e-2;
     double atol = 1e-2;
     return ck_tile::make_tuple(rtol, atol);
 }
 
+template <>
+auto get_elimit<ck_tile::bf16_t>(ck_tile::index_t hdim_q, ck_tile::index_t hdim_v)
+{
+    double rtol = 1e-2;
+    double atol = 1e-2;
+    if(hdim_q > 128 && hdim_v > 128) // 3.2 for RTZ/1.5 for RTN
+    {
+        rtol = 3.2e-2;
+        atol = 3.2e-2;
+    }
+    return ck_tile::make_tuple(rtol, atol);
+}
+
 template <typename DataType>
 bool run(const ck_tile::ArgParser& arg_parser)
 {
@@ -899,7 +912,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
         }
         // clang-format on
 
-        auto [rtol, atol] = get_elimit<DataType>(init_method);
+        auto [rtol, atol] = get_elimit<DataType>(hdim_q, hdim_v);
         bool dq_cur_pass  = ck_tile::check_err(dq_host_result,
                                               dq_host_ref,
                                               std::string("Error: QGrad Incorrect results!"),