mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-11 17:00:18 +00:00
* clean up * add mutilple thread scratch to ThreadwiseTensorSliceTransfer_v3r1 * add 2 stage prefetch * add more sanity check into transform_tensor_descriptor * tweak * enabling 2 stage prefetch to exsiting gridwise gemm; tweak * enabling 2 stage prefetch to exsiting gridwise gemm * move gridwise gemm pipeline in class; clean up * add some irregular tile size * update CalculateHasMainK0BlockLoop for multi-stage-prefetch * refactor gridwise gemm pipeline class
154 lines
11 KiB
CMake
154 lines
11 KiB
CMake
include_directories(BEFORE
|
|
include
|
|
${PROJECT_SOURCE_DIR}/host/host_tensor/include
|
|
${PROJECT_SOURCE_DIR}/device/include
|
|
${PROJECT_SOURCE_DIR}/device_operation/include
|
|
${PROJECT_SOURCE_DIR}/profiler/include
|
|
${PROJECT_SOURCE_DIR}/composable_kernel/include
|
|
${PROJECT_SOURCE_DIR}/composable_kernel/include/utility
|
|
${PROJECT_SOURCE_DIR}/composable_kernel/include/tensor_description
|
|
${PROJECT_SOURCE_DIR}/composable_kernel/include/tensor_operation
|
|
${PROJECT_SOURCE_DIR}/composable_kernel/include/problem_transform
|
|
${PROJECT_SOURCE_DIR}/external/rocm/include
|
|
)
|
|
|
|
# device_gemm_instance
|
|
set(DEVICE_GEMM_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp;
|
|
)
|
|
|
|
# device_gemm_bias_2d_instance
|
|
set(DEVICE_GEMM_BIAS_2D_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f32_f32_f32_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f32_f32_f32_km_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f32_f32_f32_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f32_f32_f32_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f16_f16_f16_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f16_f16_f16_km_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f16_f16_f16_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_2d_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
)
|
|
|
|
# device_gemm_bias_relu_instance
|
|
set(DEVICE_GEMM_BIAS_RELU_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_f16_f16_f16_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_f16_f16_f16_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_f16_f16_f16_km_nk_mn_instance.cpp;
|
|
)
|
|
|
|
# device_gemm_bias_relu_add_instance
|
|
set(DEVICE_GEMM_BIAS_RELU_ADD_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_add_f16_f16_f16_mk_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_add_f16_f16_f16_mk_nk_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_add_f16_f16_f16_km_kn_mn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_gemm_xdl_c_shuffle_bias_relu_add_f16_f16_f16_km_nk_mn_instance.cpp;
|
|
)
|
|
|
|
set(DEVICE_BATCHED_GEMM_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp;
|
|
)
|
|
|
|
# device_conv2d_fwd_instance
|
|
set(DEVICE_CONV2D_FWD_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp;
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp;
|
|
)
|
|
|
|
# device_conv1d_fwd_instance
|
|
set(DEVICE_CONV1D_FWD_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instance.cpp;
|
|
)
|
|
|
|
# device_conv2d_fwd_bias_relu_instance
|
|
set(DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp;
|
|
)
|
|
|
|
# device_conv2d_fwd_bias_relu_add_instance
|
|
set(DEVICE_CONV2D_FWD_BIAS_RELU_ADD_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp;
|
|
)
|
|
|
|
# device_conv2d_fwd_bias_relu_atomic_add_instance
|
|
set(DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE
|
|
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_bias_relu_atomic_add_nhwc_kyxc_nhwk_f16_instance.cpp;
|
|
)
|
|
|
|
add_library(device_gemm_instance SHARED ${DEVICE_GEMM_INSTANCE_SOURCE})
|
|
add_library(device_gemm_bias_2d_instance SHARED ${DEVICE_GEMM_BIAS_2D_INSTANCE_SOURCE})
|
|
add_library(device_gemm_bias_relu_instance SHARED ${DEVICE_GEMM_BIAS_RELU_INSTANCE_SOURCE})
|
|
add_library(device_gemm_bias_relu_add_instance SHARED ${DEVICE_GEMM_BIAS_RELU_ADD_INSTANCE_SOURCE})
|
|
add_library(device_batched_gemm_instance SHARED ${DEVICE_BATCHED_GEMM_INSTANCE_SOURCE})
|
|
add_library(device_conv1d_fwd_instance SHARED ${DEVICE_CONV1D_FWD_INSTANCE_SOURCE})
|
|
add_library(device_conv2d_fwd_instance SHARED ${DEVICE_CONV2D_FWD_INSTANCE_SOURCE})
|
|
add_library(device_conv2d_fwd_bias_relu_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE})
|
|
add_library(device_conv2d_fwd_bias_relu_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ADD_INSTANCE_SOURCE})
|
|
add_library(device_conv2d_fwd_bias_relu_atomic_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE})
|
|
|
|
target_include_directories(device_gemm_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_gemm_bias_2d_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_gemm_bias_relu_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_gemm_bias_relu_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_batched_gemm_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_conv1d_fwd_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_conv2d_fwd_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_conv2d_fwd_bias_relu_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_conv2d_fwd_bias_relu_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
target_include_directories(device_conv2d_fwd_bias_relu_atomic_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
|
|
|
|
target_compile_features(device_gemm_instance PUBLIC)
|
|
target_compile_features(device_gemm_bias_2d_instance PUBLIC)
|
|
target_compile_features(device_gemm_bias_relu_instance PUBLIC)
|
|
target_compile_features(device_gemm_bias_relu_add_instance PUBLIC)
|
|
target_compile_features(device_batched_gemm_instance PUBLIC)
|
|
target_compile_features(device_conv1d_fwd_instance PUBLIC)
|
|
target_compile_features(device_conv2d_fwd_instance PUBLIC)
|
|
target_compile_features(device_conv2d_fwd_bias_relu_instance PUBLIC)
|
|
target_compile_features(device_conv2d_fwd_bias_relu_add_instance PUBLIC)
|
|
target_compile_features(device_conv2d_fwd_bias_relu_atomic_add_instance PUBLIC)
|
|
|
|
set_target_properties(device_gemm_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_gemm_bias_2d_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_gemm_bias_relu_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_gemm_bias_relu_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_batched_gemm_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_conv1d_fwd_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_conv2d_fwd_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_conv2d_fwd_bias_relu_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_conv2d_fwd_bias_relu_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
set_target_properties(device_conv2d_fwd_bias_relu_atomic_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
|
|
|
install(TARGETS device_gemm_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_gemm_bias_2d_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_gemm_bias_relu_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_gemm_bias_relu_add_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_batched_gemm_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_conv1d_fwd_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_conv2d_fwd_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_conv2d_fwd_bias_relu_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_conv2d_fwd_bias_relu_add_instance LIBRARY DESTINATION lib)
|
|
install(TARGETS device_conv2d_fwd_bias_relu_atomic_add_instance LIBRARY DESTINATION lib)
|