Skip to content

GitLab

"examples/community/pipeline_stable_diffusion_pag.py" did not exist on "bdd16116f319fdc4b299f248519c17a5aeb9dd51"
Unverified Commit 140d2fa6 authored by Illia Silin's avatar Illia Silin Committed by GitHub
Browse files

Merge pull request #197 from ROCm/merge_from_public 

Merge from public
parents 87ea11d0 d4d83037
Show whitespace changes
Inline Side-by-side
Showing with 239 additions and 114 deletions
CMakeLists.txt
View file @ 140d2fa6
...@@ -97,10 +97,9 @@ if(DL_KERNELS) ...@@ -97,10 +97,9 @@ if(DL_KERNELS)
add_definitions(-DDL_KERNELS) add_definitions(-DDL_KERNELS)
set(CK_ENABLE_DL_KERNELS "ON") set(CK_ENABLE_DL_KERNELS "ON")
endif() endif()
option(CK_USE_CODEGEN "Enable codegen library" OFF)
if(INSTANCES_ONLY) if(CK_USE_CODEGEN)
add_definitions(-DINSTANCES_ONLY) add_definitions(-DCK_USE_CODEGEN)
set(CK_ENABLE_INSTANCES_ONLY "ON")
endif() endif()
include(getopt) include(getopt)
...@@ -127,7 +126,17 @@ rocm_setup_version(VERSION ${version}) ...@@ -127,7 +126,17 @@ rocm_setup_version(VERSION ${version})
list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip "$ENV{ROCM_PATH}" "$ENV{HIP_PATH}") list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip "$ENV{ROCM_PATH}" "$ENV{HIP_PATH}")
message("GPU_TARGETS= ${GPU_TARGETS}") message("GPU_TARGETS= ${GPU_TARGETS}")
message("GPU_ARCHS= ${GPU_ARCHS}")
if(GPU_ARCHS)
#disable GPU_TARGETS to avoid conflicts, this needs to happen before we call hip package
unset(GPU_TARGETS CACHE)
unset(AMDGPU_TARGETS CACHE)
endif()
if(GPU_TARGETS)
set(USER_GPU_TARGETS 1)
else()
set(USER_GPU_TARGETS 0)
endif()
find_package(hip) find_package(hip)
# No assumption that HIP kernels are launched with uniform block size for backward compatibility # No assumption that HIP kernels are launched with uniform block size for backward compatibility
# SWDEV-413293 and https://reviews.llvm.org/D155213 # SWDEV-413293 and https://reviews.llvm.org/D155213
...@@ -135,56 +144,39 @@ math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR}) ...@@ -135,56 +144,39 @@ math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR})
message("hip_version_flat=${hip_VERSION_FLAT}") message("hip_version_flat=${hip_VERSION_FLAT}")
message("checking which targets are supported") message("checking which targets are supported")
#This is the list of targets to be used in case GPU_TARGETS is not set on command line #In order to build just the CK library (without tests and examples) for all supported GPU targets
#These targets will be filtered and only supported ones will be used #use -D GPU_ARCHS="gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
#Setting GPU_TARGETS on command line will override this list #the GPU_TARGETS flag will be reset in this case in order to avoid conflicts.
if(NOT PROFILER_ONLY) #
if(NOT ENABLE_ASAN_PACKAGING) #In order to build CK along with all tests and examples it should be OK to set GPU_TARGETS to just 1 or 2 similar architectures.
#build CK for all supported targets if(NOT ENABLE_ASAN_PACKAGING)
if(NOT WIN32 AND ${hip_VERSION_FLAT} LESS 600300000) if(NOT WIN32 AND ${hip_VERSION_FLAT} LESS 600300000)
# WORKAROUND: compiler does not yet fully support gfx12 targets, need to fix version above # WORKAROUND: compiler does not yet fully support gfx12 targets, need to fix version above
rocm_check_target_ids(DEFAULT_GPU_TARGETS set(CK_GPU_TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx950")
TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx950")
else() else()
rocm_check_target_ids(DEFAULT_GPU_TARGETS set(CK_GPU_TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950")
TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx950")
endif()
else()
#build CK only for xnack-supported targets
rocm_check_target_ids(DEFAULT_GPU_TARGETS
TARGETS "gfx908:xnack+;gfx90a:xnack+;gfx940:xnack+;gfx941:xnack+;gfx942:xnack+")
set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
endif() endif()
else() else()
add_definitions(-DPROFILER_ONLY) #build CK only for xnack-supported targets when using ASAN
set(GPU_TARGETS "" CACHE STRING "" FORCE) set(CK_GPU_TARGETS "gfx908:xnack+;gfx90a:xnack+;gfx940:xnack+;gfx941:xnack+;gfx942:xnack+")
if(GPU_TARGETS)
message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, gfx11 or gfx12")
endif()
if(GPU_ARCH MATCHES "gfx90")
rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx908;gfx90a")
elseif(GPU_ARCH MATCHES "gfx94")
rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx940;gfx941;gfx942")
elseif(GPU_ARCH MATCHES "gfx10")
rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1030")
elseif(GPU_ARCH MATCHES "gfx11")
rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1100;gfx1101;gfx1102")
elseif(GPU_ARCH MATCHES "gfx12")
rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1200;gfx1201")
else()
message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx90, gfx94, gfx10, gfx11 or gfx12")
endif()
set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
endif() endif()
message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}") #if user set GPU_ARCHS on the cmake command line, overwrite default target list with user's list
#otherwise, if user set GPU_TARGETS, use that set of targets
if(GPU_TARGETS) if(GPU_ARCHS)
message("Building CK for the following targets: ${GPU_TARGETS}") set(CK_GPU_TARGETS ${GPU_ARCHS})
else() else()
message("Building CK for the default targets: ${DEFAULT_GPU_TARGETS}") if(USER_GPU_TARGETS)
set(CK_GPU_TARGETS ${GPU_TARGETS})
endif()
endif() endif()
#make sure all the targets on the list are actually supported by the current compiler
rocm_check_target_ids(SUPPORTED_GPU_TARGETS
TARGETS ${CK_GPU_TARGETS})
message("Building CK for the following targets: ${SUPPORTED_GPU_TARGETS}")
if (GPU_TARGETS) if (GPU_TARGETS)
if (GPU_TARGETS MATCHES "gfx9") if (GPU_TARGETS MATCHES "gfx9")
add_definitions(-DCK_USE_XDL) add_definitions(-DCK_USE_XDL)
...@@ -557,8 +549,7 @@ ENDFOREACH() ...@@ -557,8 +549,7 @@ ENDFOREACH()
add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES} SOURCES ${INSTANCE_FILES}) add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES} SOURCES ${INSTANCE_FILES})
add_subdirectory(library) add_subdirectory(library)
if(NOT DEFINED INSTANCES_ONLY) if(NOT GPU_ARCHS AND USER_GPU_TARGETS)
if(NOT DEFINED PROFILER_ONLY)
rocm_package_setup_component(tests rocm_package_setup_component(tests
LIBRARY_NAME composablekernel LIBRARY_NAME composablekernel
PACKAGE_NAME tests # Prevent -static suffix on package name PACKAGE_NAME tests # Prevent -static suffix on package name
...@@ -569,24 +560,18 @@ if(NOT DEFINED INSTANCES_ONLY) ...@@ -569,24 +560,18 @@ if(NOT DEFINED INSTANCES_ONLY)
PACKAGE_NAME examples PACKAGE_NAME examples
) )
add_subdirectory(example) add_subdirectory(example)
if(BUILD_TESTING)
add_subdirectory(test) add_subdirectory(test)
endif()
endif()
rocm_package_setup_component(profiler rocm_package_setup_component(profiler
LIBRARY_NAME composablekernel LIBRARY_NAME composablekernel
PACKAGE_NAME ckprofiler PACKAGE_NAME ckprofiler
) )
add_subdirectory(profiler) add_subdirectory(profiler)
else()
#When building PROFILER_ONLY, label the package with GPU_ARCH
rocm_package_setup_component(profiler
LIBRARY_NAME composablekernel
PACKAGE_NAME ckprofiler_${GPU_ARCH}
)
add_subdirectory(profiler)
endif()
endif()
if(NOT DEFINED PROFILER_ONLY AND (GPU_TARGETS MATCHES "gfx9" OR DEFINED INSTANCES_ONLY)) if(CK_USE_CODEGEN AND (GPU_TARGETS MATCHES "gfx9" OR GPU_ARCHS))
add_subdirectory(codegen) add_subdirectory(codegen)
endif() endif()
......
Jenkinsfile
View file @ 140d2fa6
...@@ -320,7 +320,7 @@ def cmake_build(Map conf=[:]){ ...@@ -320,7 +320,7 @@ def cmake_build(Map conf=[:]){
if (package_build == true && (env.BRANCH_NAME == "develop" || env.BRANCH_NAME == "amd-master")) { if (package_build == true && (env.BRANCH_NAME == "develop" || env.BRANCH_NAME == "amd-master")) {
archiveArtifacts artifacts: "build/*.deb", allowEmptyArchive: true, fingerprint: true archiveArtifacts artifacts: "build/*.deb", allowEmptyArchive: true, fingerprint: true
} }
if (params.RUN_CK_TILE_TESTS){ if (params.RUN_CK_TILE_FMHA_TESTS){
try{ try{
archiveArtifacts "perf_fmha_fwd_*.log" archiveArtifacts "perf_fmha_fwd_*.log"
archiveArtifacts "perf_fmha_bwd_*.log" archiveArtifacts "perf_fmha_bwd_*.log"
...@@ -353,7 +353,7 @@ def buildHipClangJob(Map conf=[:]){ ...@@ -353,7 +353,7 @@ def buildHipClangJob(Map conf=[:]){
def prefixpath = conf.get("prefixpath", "/opt/rocm") def prefixpath = conf.get("prefixpath", "/opt/rocm")
// Jenkins is complaining about the render group // Jenkins is complaining about the render group
def dockerOpts="--rm --device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined" def dockerOpts="--device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined"
if (conf.get("enforce_xnack_on", false)) { if (conf.get("enforce_xnack_on", false)) {
dockerOpts = dockerOpts + " --env HSA_XNACK=1 " dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
} }
...@@ -412,7 +412,7 @@ def runCKProfiler(Map conf=[:]){ ...@@ -412,7 +412,7 @@ def runCKProfiler(Map conf=[:]){
def prefixpath = conf.get("prefixpath", "/opt/rocm") def prefixpath = conf.get("prefixpath", "/opt/rocm")
// Jenkins is complaining about the render group // Jenkins is complaining about the render group
def dockerOpts="--rm --device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined" def dockerOpts="--device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined"
if (conf.get("enforce_xnack_on", false)) { if (conf.get("enforce_xnack_on", false)) {
dockerOpts = dockerOpts + " --env HSA_XNACK=1 " dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
} }
...@@ -544,7 +544,7 @@ def Build_CK(Map conf=[:]){ ...@@ -544,7 +544,7 @@ def Build_CK(Map conf=[:]){
def prefixpath = conf.get("prefixpath", "/opt/rocm") def prefixpath = conf.get("prefixpath", "/opt/rocm")
// Jenkins is complaining about the render group // Jenkins is complaining about the render group
def dockerOpts="--rm --device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined" def dockerOpts="--device=/dev/kfd --device=/dev/dri --group-add video --group-add render --cap-add=SYS_PTRACE --security-opt seccomp=unconfined"
if (conf.get("enforce_xnack_on", false)) { if (conf.get("enforce_xnack_on", false)) {
dockerOpts = dockerOpts + " --env HSA_XNACK=1 " dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
} }
...@@ -660,7 +660,7 @@ def process_results(Map conf=[:]){ ...@@ -660,7 +660,7 @@ def process_results(Map conf=[:]){
def prefixpath = "/opt/rocm" def prefixpath = "/opt/rocm"
// Jenkins is complaining about the render group // Jenkins is complaining about the render group
def dockerOpts="--rm --cap-add=SYS_PTRACE --security-opt seccomp=unconfined" def dockerOpts="--cap-add=SYS_PTRACE --security-opt seccomp=unconfined"
if (conf.get("enforce_xnack_on", false)) { if (conf.get("enforce_xnack_on", false)) {
dockerOpts = dockerOpts + " --env HSA_XNACK=1 " dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
} }
...@@ -682,7 +682,7 @@ def process_results(Map conf=[:]){ ...@@ -682,7 +682,7 @@ def process_results(Map conf=[:]){
timeout(time: 1, unit: 'HOURS'){ timeout(time: 1, unit: 'HOURS'){
try{ try{
dir("script"){ dir("script"){
if (params.RUN_CK_TILE_TESTS){ if (params.RUN_CK_TILE_FMHA_TESTS){
try{ try{
unstash "perf_fmha_fwd_gfx942.log" unstash "perf_fmha_fwd_gfx942.log"
unstash "perf_fmha_bwd_gfx942.log" unstash "perf_fmha_bwd_gfx942.log"
...@@ -1127,8 +1127,8 @@ pipeline { ...@@ -1127,8 +1127,8 @@ pipeline {
execute_args = """ cmake -D CMAKE_PREFIX_PATH=/opt/rocm \ execute_args = """ cmake -D CMAKE_PREFIX_PATH=/opt/rocm \
-D CMAKE_CXX_COMPILER="${build_compiler()}" \ -D CMAKE_CXX_COMPILER="${build_compiler()}" \
-D CMAKE_BUILD_TYPE=Release \ -D CMAKE_BUILD_TYPE=Release \
-D INSTANCES_ONLY=ON \ -D GPU_ARCHS="gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102" \
-DCMAKE_CXX_FLAGS=" -O3 " .. && make -j64 """ -D CMAKE_CXX_FLAGS=" -O3 " .. && make -j64 """
} }
steps{ steps{
buildHipClangJobAndReboot(setup_cmd: "", build_cmd: "", no_reboot:true, build_type: 'Release', execute_cmd: execute_args) buildHipClangJobAndReboot(setup_cmd: "", build_cmd: "", no_reboot:true, build_type: 'Release', execute_cmd: execute_args)
......
README.md
View file @ 140d2fa6
...@@ -91,6 +91,12 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa ...@@ -91,6 +91,12 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa
If you don't set `GPU_TARGETS` on the cmake command line, CK is built for all GPU targets If you don't set `GPU_TARGETS` on the cmake command line, CK is built for all GPU targets
supported by the current compiler (this may take a long time). supported by the current compiler (this may take a long time).
Tests and examples will only get built if the GPU_TARGETS is set by the user on the cmake command line.
NOTE: If you try setting `GPU_TARGETS` to a list of architectures, the build will only work if the
architectures are similar, e.g., `gfx908;gfx90a`, or `gfx1100;gfx1101;gfx11012`. Otherwise, if you
want to build the library for a list of different architectures,
you should use the `GPU_ARCHS` build argument, for example `GPU_ARCHS=gfx908;gfx1030;gfx1100;gfx942`.
4. Build the entire CK library: 4. Build the entire CK library:
...@@ -137,10 +143,6 @@ crash. In such cases, you can reduce the number of threads to 32 by using `-j32` ...@@ -137,10 +143,6 @@ crash. In such cases, you can reduce the number of threads to 32 by using `-j32`
Additional cmake flags can be used to significantly speed-up the build: Additional cmake flags can be used to significantly speed-up the build:
* `INSTANCES_ONLY` (default is OFF) must be set to ON in order to build only the instances and library
while skipping all tests, examples, and profiler. This is useful in cases when you plan to use CK as a
dependency and don't plan to run any examples or tests.
* `DTYPES` (default is not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build * `DTYPES` (default is not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build
instances of select data types only. The main default data types are fp32 and fp16; you can safely skip instances of select data types only. The main default data types are fp32 and fp16; you can safely skip
other data types. other data types.
......
cmake/Embed.cmake
View file @ 140d2fa6
...@@ -233,6 +233,8 @@ function(add_embed_library EMBED_NAME) ...@@ -233,6 +233,8 @@ function(add_embed_library EMBED_NAME)
else() else()
target_sources(${EMBED_NAME} INTERFACE $<TARGET_OBJECTS:${INTERNAL_EMBED_LIB}>) target_sources(${EMBED_NAME} INTERFACE $<TARGET_OBJECTS:${INTERNAL_EMBED_LIB}>)
endif() endif()
target_include_directories(${EMBED_NAME} INTERFACE "${EMBED_DIR}/include") target_include_directories(${EMBED_NAME} INTERFACE
$<BUILD_INTERFACE:${EMBED_DIR}/include>
$<INSTALL_INTERFACE:include/ck>)
endfunction() endfunction()
codegen/CMakeLists.txt
View file @ 140d2fa6
...@@ -39,6 +39,7 @@ set_target_properties(ck_host PROPERTIES ...@@ -39,6 +39,7 @@ set_target_properties(ck_host PROPERTIES
target_include_directories(ck_host PUBLIC target_include_directories(ck_host PUBLIC
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include> $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
$<INSTALL_INTERFACE:include>
) )
add_executable(ck-template-driver driver/main.cpp) add_executable(ck-template-driver driver/main.cpp)
...@@ -48,6 +49,12 @@ rocm_install( ...@@ -48,6 +49,12 @@ rocm_install(
TARGETS ck_host ck_headers TARGETS ck_host ck_headers
EXPORT ck_hostTargets EXPORT ck_hostTargets
) )
rocm_install(EXPORT ck_hostTargets
FILE composable_kernelck_hostTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel)
rocm_install(DIRECTORY include/ck DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}) rocm_install(DIRECTORY include/ck DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
add_subdirectory(test) if(BUILD_TESTING)
add_subdirectory(test)
endif()
codegen/test/CMakeLists.txt
View file @ 140d2fa6
list(APPEND CMAKE_PREFIX_PATH /opt/rocm) list(APPEND CMAKE_PREFIX_PATH /opt/rocm)
add_subdirectory(rtc) add_subdirectory(rtc)
file(GLOB TEST_SRCS CONFIGURE_DEPENDS *.cpp) file(GLOB TEST_SRCS CONFIGURE_DEPENDS *.cpp)
if(NOT INSTANCES_ONLY) # do not build the tests when we build the library for various targets
if(NOT GPU_ARCHS)
foreach(TEST_SRC ${TEST_SRCS}) foreach(TEST_SRC ${TEST_SRCS})
set_source_files_properties(${TEST_SRC} PROPERTIES LANGUAGE HIP) set_source_files_properties(${TEST_SRC} PROPERTIES LANGUAGE HIP)
get_filename_component(BASE_NAME ${TEST_SRC} NAME_WE) get_filename_component(BASE_NAME ${TEST_SRC} NAME_WE)
......
docs/reference/API_Reference_Guide.rst
View file @ 140d2fa6
...@@ -12,12 +12,6 @@ API reference guide ...@@ -12,12 +12,6 @@ API reference guide
This document contains details of the APIs for the Composable Kernel (CK) library and introduces This document contains details of the APIs for the Composable Kernel (CK) library and introduces
some of the key design principles that are used to write new classes that extend CK functionality. some of the key design principles that are used to write new classes that extend CK functionality.
=================
Using CK API
=================
This section describes how to use the CK library API.
================= =================
CK Datatypes CK Datatypes
================= =================
......
example/01_gemm/common.hpp
View file @ 140d2fa6
...@@ -21,6 +21,7 @@ ...@@ -21,6 +21,7 @@
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp" #include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
struct ProblemSize final struct ProblemSize final
{ {
...@@ -28,9 +29,9 @@ struct ProblemSize final ...@@ -28,9 +29,9 @@ struct ProblemSize final
ck::index_t N = 4096; ck::index_t N = 4096;
ck::index_t K = 4096; ck::index_t K = 4096;
ck::index_t StrideA = 4096; ck::index_t StrideA = 0;
ck::index_t StrideB = 4096; ck::index_t StrideB = 0;
ck::index_t StrideC = 4096; ck::index_t StrideC = 0;
}; };
struct ProblemSizeStreamK final struct ProblemSizeStreamK final
...@@ -39,9 +40,9 @@ struct ProblemSizeStreamK final ...@@ -39,9 +40,9 @@ struct ProblemSizeStreamK final
ck::index_t N = 4096; ck::index_t N = 4096;
ck::index_t K = 4096; ck::index_t K = 4096;
ck::index_t StrideA = 4096; ck::index_t StrideA = 0;
ck::index_t StrideB = 4096; ck::index_t StrideB = 0;
ck::index_t StrideC = 4096; ck::index_t StrideC = 0;
ck::index_t NumSKBlocks = -1; ck::index_t NumSKBlocks = -1;
}; };
...@@ -51,9 +52,9 @@ struct ProblemSizeStreamK_universal final ...@@ -51,9 +52,9 @@ struct ProblemSizeStreamK_universal final
ck::index_t N = 4096; ck::index_t N = 4096;
ck::index_t K = 4096; ck::index_t K = 4096;
ck::index_t StrideA = 4096; ck::index_t StrideA = 0;
ck::index_t StrideB = 4096; ck::index_t StrideB = 0;
ck::index_t StrideC = 4096; ck::index_t StrideC = 0;
ck::index_t Grid_size = -1; // defaults to max occupancy ck::index_t Grid_size = -1; // defaults to max occupancy
ck::index_t Streamk_sel = 1; // defaults to 1-tile SK ck::index_t Streamk_sel = 1; // defaults to 1-tile SK
...@@ -65,9 +66,9 @@ struct ProblemSizeSplitK final ...@@ -65,9 +66,9 @@ struct ProblemSizeSplitK final
ck::index_t N = 4096; ck::index_t N = 4096;
ck::index_t K = 4096; ck::index_t K = 4096;
ck::index_t StrideA = 4096; ck::index_t StrideA = 0;
ck::index_t StrideB = 4096; ck::index_t StrideB = 0;
ck::index_t StrideC = 4096; ck::index_t StrideC = 0;
ck::index_t KBatch = 1; ck::index_t KBatch = 1;
}; };
...@@ -125,7 +126,7 @@ bool parse_cmd_args<ProblemSize>(int argc, ...@@ -125,7 +126,7 @@ bool parse_cmd_args<ProblemSize>(int argc,
} }
else else
{ {
std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl std::cerr << "arg1: verification (0=no, 1=CPU and GPU)" << std::endl
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
<< std::endl << std::endl
<< "arg3: time kernel (0=no, 1=yes)" << std::endl << "arg3: time kernel (0=no, 1=yes)" << std::endl
...@@ -175,7 +176,7 @@ bool parse_cmd_args<ProblemSizeStreamK_universal>(int argc, ...@@ -175,7 +176,7 @@ bool parse_cmd_args<ProblemSizeStreamK_universal>(int argc,
else else
{ {
std::cerr std::cerr
<< "arg1: verification (0=no, 1=yes)" << std::endl << "arg1: verification (0=no, 1=CPU and GPU)" << std::endl
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << std::endl
<< "arg3: time kernel (0=no, 1=yes)" << std::endl << "arg3: time kernel (0=no, 1=yes)" << std::endl
<< "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
...@@ -224,7 +225,7 @@ bool parse_cmd_args<ProblemSizeStreamK>(int argc, ...@@ -224,7 +225,7 @@ bool parse_cmd_args<ProblemSizeStreamK>(int argc,
} }
else else
{ {
std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl std::cerr << "arg1: verification (0=no, 1=CPU and GPU)" << std::endl
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
<< std::endl << std::endl
<< "arg3: time kernel (0=no, 1=yes)" << std::endl << "arg3: time kernel (0=no, 1=yes)" << std::endl
...@@ -274,7 +275,7 @@ bool parse_cmd_args<ProblemSizeSplitK>(int argc, ...@@ -274,7 +275,7 @@ bool parse_cmd_args<ProblemSizeSplitK>(int argc,
} }
else else
{ {
std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl std::cerr << "arg1: verification (0=no, 1=CPU and GPU)" << std::endl
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)" << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
<< std::endl << std::endl
<< "arg3: time kernel (0=no, 1=yes)" << std::endl << "arg3: time kernel (0=no, 1=yes)" << std::endl
......
example/01_gemm/gemm_dl_fp16.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl ...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_dl_fp32.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl ...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_dl_int8.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl ...@@ -32,6 +32,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_dpp_fp16.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -34,6 +34,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp ...@@ -34,6 +34,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::
ReferenceGemm<ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_wmma_fp16.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -68,6 +68,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle ...@@ -68,6 +68,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_bf16.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -33,6 +33,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -33,6 +33,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceComputeType = float;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp,
ReferenceComputeType,
ReferenceComputeType>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_bf16_rtn.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -34,6 +34,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -34,6 +34,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceComputeType = float;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp,
ReferenceComputeType,
ReferenceComputeType>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp16.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -47,6 +47,17 @@ using DeviceGemmInstance = DeviceGemmInstance1; ...@@ -47,6 +47,17 @@ using DeviceGemmInstance = DeviceGemmInstance1;
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp16_fp8.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -42,6 +42,17 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataTyp ...@@ -42,6 +42,17 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataTyp
CElementOp, CElementOp,
ComputeType>; ComputeType>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp16_v2.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -46,6 +46,17 @@ using DeviceGemmInstance = ...@@ -46,6 +46,17 @@ using DeviceGemmInstance =
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp64.cpp
View file @ 140d2fa6
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -41,6 +41,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl ...@@ -41,6 +41,17 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl
BElementOp, BElementOp,
CElementOp>; CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp8.cpp
View file @ 140d2fa6
...@@ -37,6 +37,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -37,6 +37,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceComputeType = float;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp,
ReferenceComputeType,
ReferenceComputeType>;
#include "run_gemm_example.inc" #include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); } int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
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