Skip to content

GitLab

Unverified Commit 7965d66a authored by Illia Silin's avatar Illia Silin Committed by GitHub
Browse files

Split the static library into several files. (#1044) 

* spolit the static library into several

* update lib paths and fix client example

* do not use device_mha_operarions for client examples

* use appropriate libs to link to client examples

* remove the gpu/transpose path from the list

* try fixing clinet examples 3,4,9

* add necessary libs for client examples

* fix the layernorm client example

* fix the client examples 23 and 24

* fix typo

* add interface library and refresh clang format
parent 6ef034f6
Show whitespace changes
Inline Side-by-side
Showing with 175 additions and 72 deletions
client_example/18_groupnorm/CMakeLists.txt
View file @ 7965d66a
add_executable(client_groupnorm_swish groupnorm_swish.cpp)
target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_operations)
target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_other_operations)
client_example/19_pool/CMakeLists.txt
View file @ 7965d66a
add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations)
target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_max_pool2d_bwd max_pool2d_bwd.cpp)
target_link_libraries(client_max_pool2d_bwd PRIVATE composable_kernel::device_operations)
target_link_libraries(client_max_pool2d_bwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations)
target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_avg_pool3d_bwd avg_pool3d_bwd.cpp)
target_link_libraries(client_avg_pool3d_bwd PRIVATE composable_kernel::device_operations)
target_link_libraries(client_avg_pool3d_bwd PRIVATE composable_kernel::device_other_operations)
client_example/20_splitk_gemm/CMakeLists.txt
View file @ 7965d66a
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
add_executable(client_splitK_gemm splitK_gemm_fp16_f8.cpp)
target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_operations)
target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/21_grouped_gemm_bias/CMakeLists.txt
View file @ 7965d66a
add_executable(client_grouped_gemm_fixed_nk_bias_fp16 grouped_gemm_fixed_nk_bias_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_bias_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_gemm_fixed_nk_bias_fp16 PRIVATE composable_kernel::device_gemm_operations)
client_example/22_grouped_gemm/CMakeLists.txt
View file @ 7965d66a
add_executable(client_grouped_gemm_fixed_nk_fp16 grouped_gemm_fixed_nk_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_gemm_fixed_nk_fp16 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_fp8 grouped_gemm_fixed_nk_fp8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp8 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_gemm_fixed_nk_fp8 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_i8 grouped_gemm_fixed_nk_i8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_i8 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_gemm_fixed_nk_i8 PRIVATE composable_kernel::device_gemm_operations)
client_example/22_im2col_col2im/CMakeLists.txt
View file @ 7965d66a
add_executable(client_image_to_column image_to_column.cpp)
target_link_libraries(client_image_to_column PRIVATE composable_kernel::device_operations)
target_link_libraries(client_image_to_column PRIVATE composable_kernel::device_other_operations)
add_executable(client_column_to_image column_to_image.cpp)
target_link_libraries(client_column_to_image PRIVATE composable_kernel::device_operations)
target_link_libraries(client_column_to_image PRIVATE composable_kernel::device_other_operations)
client_example/23_elementwise_transpose/CMakeLists.txt
View file @ 7965d66a
add_executable(client_elementwise_transpose3d elementwise_transpose_3d.cpp)
target_link_libraries(client_elementwise_transpose3d PRIVATE composable_kernel::device_operations)
target_link_libraries(client_elementwise_transpose3d PRIVATE composable_kernel::device_other_operations)
client_example/23_grouped_convnd_fwd_scaleadd_scaleadd_relu/CMakeLists.txt
View file @ 7965d66a
add_executable(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp32 grouped_conv_fwd_scaleadd_scaleadd_relu_fp32.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp32 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp32 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp16 grouped_conv_fwd_scaleadd_scaleadd_relu_fp16.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_fp16 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_bf16 grouped_conv_fwd_scaleadd_scaleadd_relu_bf16.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_bf16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_bf16 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_int8 grouped_conv_fwd_scaleadd_scaleadd_relu_int8.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_int8 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_scaleadd_relu_int8 PRIVATE composable_kernel::device_conv_operations)
client_example/24_grouped_convnd_fwd_scaleadd_ab/CMakeLists.txt
View file @ 7965d66a
add_executable(client_grouped_convnd_fwd_scaleadd_ab_fp32 grouped_conv_fwd_scaleadd_ab_fp32.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_fp32 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_fp32 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_ab_fp16 grouped_conv_fwd_scaleadd_ab_fp16.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_fp16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_fp16 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_ab_bf16 grouped_conv_fwd_scaleadd_ab_bf16.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_bf16 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_bf16 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_convnd_fwd_scaleadd_ab_int8 grouped_conv_fwd_scaleadd_ab_int8.cpp)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_int8 PRIVATE composable_kernel::device_operations)
target_link_libraries(client_grouped_convnd_fwd_scaleadd_ab_int8 PRIVATE composable_kernel::device_conv_operations)
client_example/CMakeLists.txt
View file @ 7965d66a
......@@ -48,7 +48,7 @@ else()
endif()
endif()
find_package(composable_kernel COMPONENTS device_operations)
find_package(composable_kernel COMPONENTS device_other_operations device_gemm_operations device_conv_operations device_contraction_operations device_reduction_operations)
find_package(hip REQUIRED PATHS /opt/rocm)
message(STATUS "Build with HIP ${hip_VERSION}")
......
library/src/tensor_operation_instance/gpu/CMakeLists.txt
View file @ 7965d66a
......@@ -58,7 +58,12 @@ endfunction(add_instance_library INSTANCE_NAME)
file(GLOB dir_list LIST_DIRECTORIES true *)
set(CK_DEVICE_INSTANCES)
set(CK_DEVICE_OTHER_INSTANCES)
set(CK_DEVICE_GEMM_INSTANCES)
set(CK_DEVICE_CONV_INSTANCES)
set(CK_DEVICE_MHA_INSTANCES)
set(CK_DEVICE_CONTRACTION_INSTANCES)
set(CK_DEVICE_REDUCTION_INSTANCES)
FOREACH(subdir_path ${dir_list})
set(target_dir)
IF(IS_DIRECTORY "${subdir_path}")
......@@ -122,7 +127,19 @@ FOREACH(subdir_path ${dir_list})
if((add_inst EQUAL 1))
get_filename_component(target_dir ${subdir_path} NAME)
add_subdirectory(${target_dir})
list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
if("${cmake_instance}" MATCHES "gemm")
list(APPEND CK_DEVICE_GEMM_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
elseif("${cmake_instance}" MATCHES "conv")
list(APPEND CK_DEVICE_CONV_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
elseif("${cmake_instance}" MATCHES "mha")
list(APPEND CK_DEVICE_MHA_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
elseif("${cmake_instance}" MATCHES "contr")
list(APPEND CK_DEVICE_CONTRACTION_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
elseif("${cmake_instance}" MATCHES "reduce")
list(APPEND CK_DEVICE_REDUCTION_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
else()
list(APPEND CK_DEVICE_OTHER_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
endif()
message("add_instance_directory ${subdir_path}")
else()
message("skip_instance_directory ${subdir_path}")
......@@ -130,18 +147,14 @@ FOREACH(subdir_path ${dir_list})
ENDIF()
ENDFOREACH()
add_library(device_operations STATIC ${CK_DEVICE_INSTANCES})
add_library(composablekernels::device_operations ALIAS device_operations)
set(DEV_OPS_INC_DIRS
${PROJECT_SOURCE_DIR}/include/ck/
${PROJECT_SOURCE_DIR}/library/include/ck/
)
target_compile_features(device_operations PUBLIC)
set_target_properties(device_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_operations PUBLIC
if(CK_DEVICE_OTHER_INSTANCES)
add_library(device_other_operations STATIC ${CK_DEVICE_OTHER_INSTANCES})
add_library(composablekernels::device_other_operations ALIAS device_other_operations)
target_compile_features(device_other_operations PUBLIC)
set_target_properties(device_other_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_other_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/utility>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/tensor_description>
......@@ -157,23 +170,115 @@ target_include_directories(device_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/utility>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/quantization>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/softmax>
)
rocm_install(TARGETS device_other_operations
EXPORT device_other_operationsTargets)
rocm_install(EXPORT device_other_operationsTargets
FILE composable_kerneldevice_other_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
if(CK_DEVICE_GEMM_INSTANCES)
add_library(device_gemm_operations STATIC ${CK_DEVICE_GEMM_INSTANCES})
add_library(composablekernels::device_gemm_operations ALIAS device_gemm_operations)
target_compile_features(device_gemm_operations PUBLIC)
set_target_properties(device_gemm_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_gemm_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu>
)
rocm_install(TARGETS device_gemm_operations
EXPORT device_gemm_operationsTargets)
rocm_install(EXPORT device_gemm_operationsTargets
FILE composable_kerneldevice_gemm_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
if(CK_DEVICE_CONV_INSTANCES)
add_library(device_conv_operations STATIC ${CK_DEVICE_CONV_INSTANCES})
add_library(composablekernels::device_conv_operations ALIAS device_conv_operations)
target_compile_features(device_conv_operations PUBLIC)
set_target_properties(device_conv_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_conv_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd>
)
rocm_install(TARGETS device_conv_operations
EXPORT device_conv_operationsTargets)
rocm_install(EXPORT device_conv_operationsTargets
FILE composable_kerneldevice_conv_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
if(CK_DEVICE_MHA_INSTANCES)
add_library(device_mha_operations STATIC ${CK_DEVICE_MHA_INSTANCES})
add_library(composablekernels::device_mha_operations ALIAS device_mha_operations)
target_compile_features(device_mha_operations PUBLIC)
set_target_properties(device_mha_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_mha_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/mha>
)
rocm_install(TARGETS device_mha_operations
EXPORT device_mha_operationsTargets)
rocm_install(EXPORT device_mha_operationsTargets
FILE composable_kerneldevice_mha_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
if(CK_DEVICE_CONTRACTION_INSTANCES)
add_library(device_contraction_operations STATIC ${CK_DEVICE_CONTRACTION_INSTANCES})
add_library(composablekernels::device_contraction_operations ALIAS device_contraction_operations)
target_compile_features(device_contraction_operations PUBLIC)
set_target_properties(device_contraction_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_contraction_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu>
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/contraction>
)
rocm_install(TARGETS device_contraction_operations
EXPORT device_contraction_operationsTargets)
rocm_install(EXPORT device_contraction_operationsTargets
FILE composable_kerneldevice_contraction_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
if(CK_DEVICE_REDUCTION_INSTANCES)
add_library(device_reduction_operations STATIC ${CK_DEVICE_REDUCTION_INSTANCES})
add_library(composablekernels::device_reduction_operations ALIAS device_reduction_operations)
target_compile_features(device_reduction_operations PUBLIC)
set_target_properties(device_reduction_operations PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(device_reduction_operations PUBLIC
$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/ck/library/tensor_operation_instance/gpu/reduce>
)
#once new arches are enabled make this an option on the main cmake file
# and pass down here to be exported
target_compile_options(device_operations PRIVATE
--offload-arch=gfx908
--offload-arch=gfx90a
)
# install(TARGETS device_operations LIBRARY DESTINATION lib)
rocm_install(TARGETS device_operations
EXPORT device_operationsTargets)
rocm_install(DIRECTORY ${DEV_OPS_INC_DIRS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck)
rocm_install(EXPORT device_operationsTargets
FILE composable_kerneldevice_operationsTargets.cmake
)
rocm_install(TARGETS device_reduction_operations
EXPORT device_reduction_operationsTargets)
rocm_install(EXPORT device_reduction_operationsTargets
FILE composable_kerneldevice_reduction_operationsTargets.cmake
NAMESPACE composable_kernel::
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
endif()
add_library(device_operations INTERFACE)
target_link_libraries(device_operations INTERFACE
device_contraction_operations
device_conv_operations
device_gemm_operations
device_other_operations
device_reduction_operations
utility)
set(DEV_OPS_INC_DIRS
${PROJECT_SOURCE_DIR}/include/ck/
${PROJECT_SOURCE_DIR}/library/include/ck/
)
rocm_install(DIRECTORY ${DEV_OPS_INC_DIRS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck)
test/transpose/test_transpose_ut_cases.inc
View file @ 7965d66a
......@@ -14,7 +14,6 @@ TYPED_TEST(TestTranspose, Test1)
this->Run();
}
TYPED_TEST(TestTranpose, Test2)
{
std::vector<int> Ms{127, 255, 312, 799, 1573};
......@@ -27,4 +26,3 @@ TYPED_TEST(TestTranpose, Test2)
this->Run();
}
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