diff --git a/Jenkinsfile b/Jenkinsfile index ec453a99ca..a40bd97f3a 100644 --- a/Jenkinsfile +++ b/Jenkinsfile @@ -229,8 +229,11 @@ def cmake_build(Map conf=[:]){ if (setup_args.contains("gfx10")){ invocation_tag="gfx10" } - if (setup_args.contains("gfx90")){ - invocation_tag="gfx90" + if (setup_args.contains("gfx908")){ + invocation_tag="gfx908" + } + if (setup_args.contains("gfx90a")){ + invocation_tag="gfx90a" } if (setup_args.contains("gfx94")){ invocation_tag="gfx94" @@ -314,9 +317,13 @@ def cmake_build(Map conf=[:]){ if (setup_args.contains("gfx90a") && params.NINJA_BUILD_TRACE){ sh "/ninjatracing/ninjatracing .ninja_log > ck_build_trace.json" archiveArtifacts "ck_build_trace.json" - sh "ninja test" + // do not run unit tests when building instances only + if(!params.BUILD_INSTANCES_ONLY){ + sh "ninja test" + } } else{ + // run unit tests sh "make check" } } @@ -511,6 +518,9 @@ def Build_CK(Map conf=[:]){ else if ( runShell('grep -n "gfx1201" rocminfo.log') ) { arch_type = 5 } + else if ( runShell('grep -n "gfx908" rocminfo.log') ) { + arch_type = 6 + } cmake_build(conf) if ( !params.BUILD_LEGACY_OS && arch_type == 1 ){ echo "Run inductor codegen tests" @@ -582,7 +592,17 @@ def Build_CK(Map conf=[:]){ sh "./run_gemm_performance_tests.sh 0 CI_${params.COMPILER_VERSION} ${env.BRANCH_NAME} ${NODE_NAME} gfx12" archiveArtifacts "perf_onnx_gemm_gfx12.log" stash includes: "perf_onnx_gemm_gfx12.log", name: "perf_log_gfx12" - } + } + else if ( arch_type == 6 ){ + // run standard tests on gfx908 + echo "Run performance tests" + sh "./run_performance_tests.sh 0 CI_${params.COMPILER_VERSION} ${env.BRANCH_NAME} ${NODE_NAME}" + archiveArtifacts "perf_gemm_gfx908.log" + archiveArtifacts "perf_onnx_gemm_gfx908.log" + archiveArtifacts "perf_resnet50_N256_gfx908.log" + archiveArtifacts "perf_resnet50_N4_gfx908.log" + stash includes: "perf_**.log", name: "perf_log_gfx908" + } } } if (params.hipTensor_test && arch_type == 1 ){ @@ -718,11 +738,12 @@ def process_results(Map conf=[:]){ //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;DISABLE_DL_KERNELS=true;ROCMVERSION=6.3;RUN_CK_TILE_FMHA_TESTS=true;RUN_CK_TILE_GEMM_TESTS=true + 0 22 * * * % ROCMVERSION=6.3;BUILD_GFX908=true;BUILD_GFX12=false;RUN_PERFORMANCE_TESTS=false 0 21 * * * % ROCMVERSION=6.3;hipTensor_test=true;RUN_CODEGEN_TESTS=true 0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-staging;BUILD_COMPILER=/llvm-project/build/bin/clang++;USE_SCCACHE=false;NINJA_BUILD_TRACE=true 0 17 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-mainline;BUILD_COMPILER=/llvm-project/build/bin/clang++;USE_SCCACHE=false;NINJA_BUILD_TRACE=true 0 15 * * * % BUILD_INSTANCES_ONLY=true;RUN_PERFORMANCE_TESTS=false;USE_SCCACHE=false - 0 13 * * * % BUILD_LEGACY_OS=true''' : "" + 0 13 * * * % BUILD_LEGACY_OS=true;USE_SCCACHE=false;RUN_PERFORMANCE_TESTS=false''' : "" pipeline { agent none @@ -805,6 +826,10 @@ pipeline { name: "BUILD_INSTANCES_ONLY", defaultValue: false, description: "Test building instances for various architectures simultaneously (default: OFF)") + booleanParam( + name: "BUILD_GFX908", + defaultValue: false, + description: "Build CK and run tests on gfx908 (default: OFF)") booleanParam( name: "BUILD_GFX12", defaultValue: true, @@ -1117,6 +1142,26 @@ pipeline { cleanWs() } } + stage("Build CK and run Tests on gfx908") + { + when { + beforeAgent true + expression { params.BUILD_GFX908.toBoolean() && !params.RUN_FULL_QA.toBoolean() && !params.BUILD_INSTANCES_ONLY.toBoolean() && !params.BUILD_LEGACY_OS.toBoolean() } + } + agent{ label rocmnode("gfx908") } + environment{ + setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908" -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" \ + -DGPU_TARGETS="gfx908" \ + -DCMAKE_CXX_COMPILER="${build_compiler()}" \ + -DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """ + } + steps{ + Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local') + cleanWs() + } + } stage("Build CK and run Tests on gfx90a") { when { diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt index ab1b0c68a7..09288cd4ad 100755 --- a/example/01_gemm/CMakeLists.txt +++ b/example/01_gemm/CMakeLists.txt @@ -46,9 +46,6 @@ foreach(gpu IN LISTS GPU_TARGETS) endif() endforeach() -add_example_executable(example_gemm_xdl_bf16_streamk_v3 gemm_xdl_bf16_streamk_v3.cpp) -add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_streamk_v3) - add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp) add_example_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16) @@ -80,6 +77,12 @@ foreach(gpu IN LISTS GPU_TARGETS) add_example_executable(example_gemm_xdl_lds_direct_load_fp16 gemm_xdl_lds_direct_load_fp16.cpp) add_example_dependencies(example_gemm_xdl example_gemm_xdl_lds_direct_load_fp16) + + add_example_executable(example_gemm_xdl_bf16_streamk_v3 gemm_xdl_bf16_streamk_v3.cpp) + add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_streamk_v3) + + add_example_executable(example_gemm_xdl_fp8_streamk_v3 gemm_xdl_fp8_streamk_v3.cpp) + add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_streamk_v3) set(target 1) endif() endforeach() @@ -90,9 +93,6 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8) add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp) add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8) -add_example_executable(example_gemm_xdl_fp8_streamk_v3 gemm_xdl_fp8_streamk_v3.cpp) -add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_streamk_v3) - add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp) add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8) diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt index 8a295d14c4..91c072aef7 100644 --- a/example/09_convnd_fwd/CMakeLists.txt +++ b/example/09_convnd_fwd/CMakeLists.txt @@ -3,7 +3,6 @@ add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp) add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp) add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp) add_example_executable(example_convnd_fwd_xdl_fp8 convnd_fwd_xdl_fp8.cpp) -add_example_executable(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp) add_example_executable(example_convnd_fwd_xdl_bf8 convnd_fwd_xdl_bf8.cpp) add_example_executable(example_convnd_fwd_xdl_fp16_comp_fp8 convnd_fwd_xdl_fp16_comp_fp8.cpp) add_example_executable(example_convnd_fwd_xdl_fp8_bf8 convnd_fwd_xdl_fp8_bf8.cpp) @@ -11,3 +10,13 @@ add_example_executable(example_convnd_fwd_xdl_bf8_fp8 convnd_fwd_xdl_bf8_fp8.cpp add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp) add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp) add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp) + +# only build fp64 example for the following targets +list(APPEND gpu_list gfx90a gfx942 gfx950) +set(target 0) +foreach(gpu IN LISTS GPU_TARGETS) + if(gpu IN_LIST gpu_list AND target EQUAL 0) + add_example_executable(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp) + set(target 1) + endif() +endforeach() \ No newline at end of file