Skip to content

GitLab

Commit 9b7093aa authored by Rostyslav Geyyer's avatar Rostyslav Geyyer
Browse files

Merge branch 'develop' into lwpck-791

parents 4cb63955 bba085d2
Hide whitespace changes
Inline Side-by-side
Showing with 202 additions and 218 deletions
CMakeLists.txt
View file @ 9b7093aa
cmake_minimum_required(VERSION 3.14) cmake_minimum_required(VERSION 3.14)
cmake_policy(SET CMP0140 NEW)
# This has to be initialized before the project() command appears # This has to be initialized before the project() command appears
# Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D. MSVC_IDE does not use CMAKE_BUILD_TYPE # Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D. MSVC_IDE does not use CMAKE_BUILD_TYPE
...@@ -105,6 +106,7 @@ if(NOT PROFILER_ONLY) ...@@ -105,6 +106,7 @@ if(NOT PROFILER_ONLY)
TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102") TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
else() else()
add_definitions(-DPROFILER_ONLY) add_definitions(-DPROFILER_ONLY)
set(GPU_TARGETS "" CACHE STRING "" FORCE)
if(GPU_TARGETS) if(GPU_TARGETS)
message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx9, gfx10, or gfx11") message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx9, gfx10, or gfx11")
endif() endif()
...@@ -117,11 +119,12 @@ else() ...@@ -117,11 +119,12 @@ else()
else() else()
message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx9, gfx10, or gfx11") message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx9, gfx10, or gfx11")
endif() endif()
set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
endif() endif()
message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}") message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " ") set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
if(GPU_TARGETS) if(GPU_TARGETS)
message("Building CK for the following targets: ${GPU_TARGETS}") message("Building CK for the following targets: ${GPU_TARGETS}")
...@@ -381,31 +384,31 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu ...@@ -381,31 +384,31 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
set(cmake_instance) set(cmake_instance)
file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance) file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
set(add_inst 0) set(add_inst 0)
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8") if(("${cmake_instance}" MATCHES "fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
#message("fp8 instance found!") #message("fp8 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf8\" " AND DTYPES MATCHES "bf8") if(("${cmake_instance}" MATCHES "bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
#message("bf8 instance found!") #message("bf8 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16") if(("${cmake_instance}" MATCHES "fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
#message("fp16 instance found!") #message("fp16 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32") if(("${cmake_instance}" MATCHES "fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
#message("fp32 instance found!") #message("fp32 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64") if(("${cmake_instance}" MATCHES "fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
#message("fp64 instance found!") #message("fp64 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16") if(("${cmake_instance}" MATCHES "bf16" OR "${cmake_instance}" MATCHES "_b16") AND DTYPES MATCHES "bf16")
#message("bf16 instance found!") #message("bf16 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8") if(("${cmake_instance}" MATCHES "int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
#message("int8 instance found!") #message("int8 instance found!")
set(add_inst 1) set(add_inst 1)
endif() endif()
......
Dockerfile
View file @ 9b7093aa
...@@ -16,25 +16,12 @@ RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl ...@@ -16,25 +16,12 @@ RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
RUN wget https://repo.radeon.com/amdgpu-install/5.6/ubuntu/focal/amdgpu-install_5.6.50600-1_all.deb --no-check-certificate RUN wget https://repo.radeon.com/amdgpu-install/5.7/ubuntu/focal/amdgpu-install_5.7.50700-1_all.deb --no-check-certificate
RUN apt-get update && \ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_5.7.50700-1_all.deb
DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
./amdgpu-install_5.6.50600-1_all.deb RUN 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] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
RUN if [ "$ROCMVERSION" != "5.7" ]; then \ 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'
wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "rc1" ]; then \
sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.7 rel-19 > /etc/apt/sources.list.d/rocm-build.list' && \
amdgpu-repo --amdgpu-build=1637781; \
fi
RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list" RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
RUN amdgpu-install -y --usecase=rocm --no-dkms RUN amdgpu-install -y --usecase=rocm --no-dkms
...@@ -114,7 +101,7 @@ RUN sh -c "echo compiler commit = '$compiler_commit'" ...@@ -114,7 +101,7 @@ RUN sh -c "echo compiler commit = '$compiler_commit'"
RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \
git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \ git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
cd llvm-project && mkdir build && cd build && \ cd llvm-project && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \ cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
make -j 8 ; \ make -j 8 ; \
else echo "using the release compiler"; \ else echo "using the release compiler"; \
fi fi
...@@ -122,7 +109,7 @@ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; ...@@ -122,7 +109,7 @@ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ];
RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \
git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \ git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \ cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \ cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
make -j 8 ; \ make -j 8 ; \
else echo "using the release compiler"; \ else echo "using the release compiler"; \
fi fi
......
Jenkinsfile
View file @ 9b7093aa
...@@ -33,7 +33,7 @@ def runShell(String command){ ...@@ -33,7 +33,7 @@ def runShell(String command){
def getDockerImageName(){ def getDockerImageName(){
def img def img
if (params.ROCMVERSION != "5.7"){ if (params.ROCMVERSION != "6.0"){
if (params.COMPILER_VERSION == "") { if (params.COMPILER_VERSION == "") {
img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}" img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
} }
...@@ -613,8 +613,8 @@ def process_results(Map conf=[:]){ ...@@ -613,8 +613,8 @@ 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 //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;ROCMVERSION=5.7;COMPILER_VERSION=rc1 CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=
0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT= 0 21 * * * % ROCMVERSION=5.7;COMPILER_VERSION=;COMPILER_COMMIT=
0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : "" 0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
pipeline { pipeline {
...@@ -632,8 +632,8 @@ pipeline { ...@@ -632,8 +632,8 @@ pipeline {
description: "Force building docker image (default: false), set to true if docker image needs to be updated.") description: "Force building docker image (default: false), set to true if docker image needs to be updated.")
string( string(
name: 'ROCMVERSION', name: 'ROCMVERSION',
defaultValue: '5.6', defaultValue: '5.7',
description: 'Specify which ROCM version to use: 5.6 (default).') description: 'Specify which ROCM version to use: 5.7 (default).')
string( string(
name: 'COMPILER_VERSION', name: 'COMPILER_VERSION',
defaultValue: '', defaultValue: '',
......
example/01_gemm/CMakeLists.txt
View file @ 9b7093aa
if(DL_KERNELS) add_custom_target(example_gemm_dl)
add_custom_target(example_gemm_dl)
add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp) add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
add_dependencies(example_gemm_dl example_gemm_dl_fp32) if(result EQUAL 0)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_dependencies(example_gemm_dl example_gemm_dl_fp32)
add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp) endif()
add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_dl example_gemm_dl_fp16) add_dependencies(example_gemm_dl example_gemm_dl_fp16)
add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp) endif()
endif() add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_dl example_gemm_dl_int8) add_dependencies(example_gemm_dl example_gemm_dl_int8)
endif() endif()
if(USE_BITINT_EXTENSION_INT4)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp) add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
add_dependencies(example_gemm_dl example_gemm_dl_int4) add_dependencies(example_gemm_dl example_gemm_dl_int4)
endif(USE_BITINT_EXTENSION_INT4) endif(USE_BITINT_EXTENSION_INT4)
endif()
add_custom_target(example_gemm_xdl) add_custom_target(example_gemm_xdl)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp) if(result EQUAL 0)
add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp) add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp16) endif()
add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16) add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16) add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
endif()
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102") add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
endif()
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
add_custom_target(example_gemm_wmma) add_custom_target(example_gemm_wmma)
add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp) add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
add_dependencies(example_gemm_wmma example_gemm_wmma_fp16) if(result EQUAL 0)
endif() add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
endif()
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_bf16) add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp) add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn) add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
endif() endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_int8) add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
endif() endif()
...@@ -54,29 +57,30 @@ if(USE_BITINT_EXTENSION_INT4) ...@@ -54,29 +57,30 @@ if(USE_BITINT_EXTENSION_INT4)
add_dependencies(example_gemm_xdl example_gemm_xdl_int4) add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4) endif(USE_BITINT_EXTENSION_INT4)
if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES) # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
# 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_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp64) add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
endif() endif()
add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp) add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942") if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp) add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_f8) add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
endif() endif()
endif() endif()
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "bf8") OR NOT DEFINED DTYPES) # if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
# if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942") add_example_executable(example_gemm_xdl_f8_bf8 gemm_xdl_f8_bf8.cpp)
add_example_executable(example_gemm_xdl_f8_bf8 gemm_xdl_f8_bf8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_f8_bf8) add_dependencies(example_gemm_xdl example_gemm_xdl_f8_bf8)
# endif() endif()
endif() # endif()
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES) add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8) add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
endif() endif()
example/02_gemm_bilinear/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102) list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
...@@ -19,4 +18,3 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -19,4 +18,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
example/03_gemm_bias_relu/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
...@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
example/04_gemm_add_add_fastgelu/CMakeLists.txt
View file @ 9b7093aa
...@@ -3,24 +3,24 @@ set(target 0) ...@@ -3,24 +3,24 @@ set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_gemm_add_add_fastgelu_xdl) add_custom_target(example_gemm_add_add_fastgelu_xdl)
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16) add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16) add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32) add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
endif() endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp) add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4) add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4) endif(USE_BITINT_EXTENSION_INT4)
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8) add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
endif() endif()
set(target 1) set(target 1)
......
example/09_convnd_fwd/CMakeLists.txt
View file @ 9b7093aa
...@@ -2,34 +2,16 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) ...@@ -2,34 +2,16 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp) add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp) add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp) add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp) add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
endif() # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp) add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
if(DL_KERNELS) add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp) add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
endif()
endif()
example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
View file @ 9b7093aa
...@@ -2,27 +2,27 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) ...@@ -2,27 +2,27 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_convnd_fwd_reduce_xdl) add_custom_target(example_convnd_fwd_reduce_xdl)
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8) add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16) add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16) add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp) if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32) add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
endif() endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp) add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4) add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4) endif(USE_BITINT_EXTENSION_INT4)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
\ No newline at end of file
example/13_pool2d_fwd/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp) add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
endif()
example/14_gemm_quantization/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
# dlops # dlops
if(DL_KERNELS) add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
endif()
# xdlops # xdlops
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
...@@ -14,4 +10,3 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -14,4 +10,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
\ No newline at end of file
example/15_grouped_gemm/CMakeLists.txt
View file @ 9b7093aa
add_custom_target(example_grouped_gemm_xdl) add_custom_target(example_grouped_gemm_xdl)
add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) if(result EQUAL 0)
add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp) if(result EQUAL 0)
add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp) endif()
add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp) add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
example_grouped_gemm_xdl_fp16 endif()
example_grouped_gemm_multiple_d_dl_fp16 add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
example_grouped_gemm_xdl_splitk_fp16 if(result EQUAL 0)
example_grouped_gemm_xdl_fixed_nk_fp16 add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
example_grouped_gemm_xdl_fixed_nk_bias_fp16) endif()
endif() add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) if(result EQUAL 0)
add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bfp16) endif()
endif() add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) if(result EQUAL 0)
add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
endif()
add_example_executable(example_grouped_gemm_xdl_bf16 grouped_gemm_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
endif()
add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
endif() endif()
if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES) add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp) if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8) add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
endif() endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp) add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4) if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
endif()
endif() endif()
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
View file @ 9b7093aa
...@@ -6,30 +6,43 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -6,30 +6,43 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_custom_target(example_gemm_reduce_xdl_max) add_custom_target(example_gemm_reduce_xdl_max)
add_custom_target(example_gemm_reduce_xdl_mean_meansquare) add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
add_custom_target(example_gemm_add_add_mean_meansquare_xdl) add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp) if(result EQUAL 0)
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16) add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
endif()
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16) add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16) add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp) add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8) add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
endif()
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8) add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
endif() endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp) add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32) add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32) add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp) add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16) add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16) add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
endif() endif()
...@@ -40,7 +53,9 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -40,7 +53,9 @@ foreach(gpu IN LISTS GPU_TARGETS)
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp) add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4) if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
endif()
endif() endif()
set(target 1) set(target 1)
endif() endif()
......
example/17_convnd_bwd_data/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp) add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility) if(result EQUAL 0)
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
if(DL_KERNELS)
add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp) add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility) if(result EQUAL 0)
endif() target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
endif() endif()
example/18_batched_gemm_reduce/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
...@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
example/20_grouped_conv_bwd_weight/CMakeLists.txt
View file @ 9b7093aa
...@@ -3,22 +3,20 @@ set(target 0) ...@@ -3,22 +3,20 @@ set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_grouped_conv_bwd_weight) add_custom_target(example_grouped_conv_bwd_weight)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16) add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16) add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
endif() endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_custom_target(example_grouped_conv_bwd_weight_dl)
if(DL_KERNELS) add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
add_custom_target(example_grouped_conv_bwd_weight_dl) if(result EQUAL 0)
add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp) add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16) endif()
endif()
endif()
\ No newline at end of file
example/21_gemm_layernorm/CMakeLists.txt
View file @ 9b7093aa
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
...@@ -10,4 +9,4 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -10,4 +9,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
example/22_cgemm/CMakeLists.txt
View file @ 9b7093aa
add_custom_target(example_cgemm_xdl) add_custom_target(example_cgemm_xdl)
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp) if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16) add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16) add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp) add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32) if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
endif() endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8) add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
endif() endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
......
example/24_batched_gemm/CMakeLists.txt
View file @ 9b7093aa
add_custom_target(example_batched_gemm_xdl) add_custom_target(example_batched_gemm_xdl)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp) if(result EQUAL 0)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp) if(result EQUAL 0)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES) add_example_executable(example_batched_gemm_xdl_bf16 batched_gemm_xdl_bf16.cpp)
add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp) if(result EQUAL 0)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bfp16) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bf16)
endif() endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp) if(result EQUAL 0)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
endif() endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp) add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4) if(result EQUAL 0)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
endif()
endif() endif()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment