Skip to content

GitLab

Commit 4dae6d81 authored by Adam Osewski's avatar Adam Osewski
Browse files

Clean debug changes.

parent 235903ed
Hide whitespace changes
Inline Side-by-side
Showing with 328 additions and 328 deletions
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
View file @ 4dae6d81
...@@ -17,83 +17,83 @@ namespace instance { ...@@ -17,83 +17,83 @@ namespace instance {
#if defined(CK_USE_XDL) #if defined(CK_USE_XDL)
#if defined(CK_ENABLE_FP16) #if defined(CK_ENABLE_FP16)
// void add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances( void add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances( void add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col, Col,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances( void add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Col, std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances( void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Col, std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
// Col, Col,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances( void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col, Col,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances( void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances( void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
...@@ -108,116 +108,116 @@ void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances ...@@ -108,116 +108,116 @@ void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances
PassThrough, PassThrough,
PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances( void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances( void add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
#endif #endif
// #if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP8) #if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP8)
// void add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances( void add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F16, F16,
// F8, F8,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances( void add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// F8, F8,
// F16, F16,
// Empty_Tuple, Empty_Tuple,
// F16, F16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// #endif #endif
// #if defined(CK_ENABLE_BF16) #if defined(CK_ENABLE_BF16)
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances( void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// BF16, BF16,
// BF16, BF16,
// Empty_Tuple, Empty_Tuple,
// BF16, BF16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances( void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col, Col,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// BF16, BF16,
// BF16, BF16,
// Empty_Tuple, Empty_Tuple,
// BF16, BF16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// #endif #endif
// #if defined(CK_ENABLE_BF16) && defined(CK_ENABLE_INT8) #if defined(CK_ENABLE_BF16) && defined(CK_ENABLE_INT8)
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances( void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row, Row,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// BF16, BF16,
// I8, I8,
// Empty_Tuple, Empty_Tuple,
// BF16, BF16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances( void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row, std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col, Col,
// Empty_Tuple, Empty_Tuple,
// Row, Row,
// BF16, BF16,
// I8, I8,
// Empty_Tuple, Empty_Tuple,
// BF16, BF16,
// PassThrough, PassThrough,
// PassThrough, PassThrough,
// PassThrough>>>& instances); PassThrough>>>& instances);
// #endif #endif
#endif // CK_USE_XDL #endif // CK_USE_XDL
template <typename ALayout, template <typename ALayout,
typename BLayout, typename BLayout,
...@@ -260,30 +260,30 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -260,30 +260,30 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances( add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
// op_ptrs); op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances( add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
// op_ptrs); op_ptrs);
} }
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> && else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances( add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
op_ptrs); op_ptrs);
} }
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> && else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
} }
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> && else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs); add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
} }
} }
#endif #endif
...@@ -294,7 +294,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -294,7 +294,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(op_ptrs); add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(op_ptrs);
} }
} }
else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> && else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
...@@ -303,7 +303,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -303,7 +303,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(op_ptrs); add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
} }
} }
#endif #endif
...@@ -314,14 +314,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -314,14 +314,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances( add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
// op_ptrs); op_ptrs);
} }
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> && else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances( add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
// op_ptrs); op_ptrs);
} }
} }
#endif #endif
...@@ -332,14 +332,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -332,14 +332,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances( add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances(
// op_ptrs); op_ptrs);
} }
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> && else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>) is_same_v<ELayout, Row>)
{ {
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances( add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
// op_ptrs); op_ptrs);
} }
} }
#endif #endif
......
profiler/src/CMakeLists.txt
View file @ 4dae6d81
# ckProfiler # ckProfiler
set(PROFILER_SOURCES set(PROFILER_SOURCES
profiler.cpp profiler.cpp
# profile_gemm.cpp profile_gemm.cpp
# profile_reduce.cpp profile_reduce.cpp
# profile_groupnorm_bwd_data.cpp profile_groupnorm_bwd_data.cpp
# profile_groupnorm_fwd.cpp profile_groupnorm_fwd.cpp
# profile_layernorm_bwd_data.cpp profile_layernorm_bwd_data.cpp
# profile_layernorm_bwd_gamma_beta.cpp profile_layernorm_bwd_gamma_beta.cpp
# profile_groupnorm_bwd_gamma_beta.cpp profile_groupnorm_bwd_gamma_beta.cpp
# profile_layernorm_fwd.cpp profile_layernorm_fwd.cpp
# profile_max_pool3d_fwd.cpp profile_max_pool3d_fwd.cpp
# profile_avg_pool3d_bwd.cpp profile_avg_pool3d_bwd.cpp
# profile_max_pool3d_bwd.cpp profile_max_pool3d_bwd.cpp
# profile_softmax.cpp profile_softmax.cpp
# profile_batchnorm_fwd.cpp profile_batchnorm_fwd.cpp
# profile_batchnorm_bwd.cpp profile_batchnorm_bwd.cpp
# profile_batchnorm_infer.cpp profile_batchnorm_infer.cpp
# profile_conv_tensor_rearrange.cpp profile_conv_tensor_rearrange.cpp
# profile_transpose.cpp profile_transpose.cpp
# profile_permute_scale.cpp profile_permute_scale.cpp
) )
# if(GPU_TARGETS MATCHES "gfx9") if(GPU_TARGETS MATCHES "gfx9")
# if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
# list(APPEND PROFILER_SOURCES profile_contraction_bilinear.cpp) list(APPEND PROFILER_SOURCES profile_contraction_bilinear.cpp)
# list(APPEND PROFILER_SOURCES profile_contraction_scale.cpp) list(APPEND PROFILER_SOURCES profile_contraction_scale.cpp)
# endif() endif()
# if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
# list(APPEND PROFILER_SOURCES profile_gemm_reduce.cpp) list(APPEND PROFILER_SOURCES profile_gemm_reduce.cpp)
# list(APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp) list(APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp)
# list(APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp) list(APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_gemm.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_streamk.cpp) list(APPEND PROFILER_SOURCES profile_gemm_streamk.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp) list(APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_relu.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_relu.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_silu.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_silu.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_gemm_fixed_nk.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm_fixed_nk.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_gemm_two_stage.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm_two_stage.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp)
list(APPEND PROFILER_SOURCES profile_grouped_gemm_multiple_d_splitk.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm_multiple_d_splitk.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_gemm_tile_loop.cpp) list(APPEND PROFILER_SOURCES profile_grouped_gemm_tile_loop.cpp)
# endif() endif()
# list(APPEND PROFILER_SOURCES profile_gemm_multiply_add.cpp) list(APPEND PROFILER_SOURCES profile_gemm_multiply_add.cpp)
# list(APPEND PROFILER_SOURCES profile_batched_gemm.cpp) list(APPEND PROFILER_SOURCES profile_batched_gemm.cpp)
# list(APPEND PROFILER_SOURCES profile_batched_gemm_reduce.cpp) list(APPEND PROFILER_SOURCES profile_batched_gemm_reduce.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_add_multiply.cpp) list(APPEND PROFILER_SOURCES profile_gemm_add_multiply.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_bias_add_reduce.cpp) list(APPEND PROFILER_SOURCES profile_gemm_bias_add_reduce.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_splitk.cpp) list(APPEND PROFILER_SOURCES profile_gemm_splitk.cpp)
# list(APPEND PROFILER_SOURCES profile_gemm_universal.cpp) list(APPEND PROFILER_SOURCES profile_gemm_universal.cpp)
# list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu.cpp) list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu.cpp)
# list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu_add.cpp) list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu_add.cpp)
# list(APPEND PROFILER_SOURCES profile_conv_bwd_data.cpp) list(APPEND PROFILER_SOURCES profile_conv_bwd_data.cpp)
# list(APPEND PROFILER_SOURCES profile_conv_fwd.cpp) list(APPEND PROFILER_SOURCES profile_conv_fwd.cpp)
# endif() endif()
# if(GPU_TARGETS MATCHES "gfx11" OR GPU_TARGETS MATCHES "gfx9") if(GPU_TARGETS MATCHES "gfx11" OR GPU_TARGETS MATCHES "gfx9")
# if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
# list(APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp) list(APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp)
# endif() endif()
# list(APPEND PROFILER_SOURCES profile_grouped_conv_fwd.cpp) list(APPEND PROFILER_SOURCES profile_grouped_conv_fwd.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_data.cpp) list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_data.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp) list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp)
# endif() endif()
# if(DL_KERNELS) if(DL_KERNELS)
# list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp) list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
# list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp) list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp)
# endif() endif()
set(PROFILER_EXECUTABLE ckProfiler) set(PROFILER_EXECUTABLE ckProfiler)
...@@ -79,78 +79,78 @@ add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES}) ...@@ -79,78 +79,78 @@ add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES})
target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors) target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility getopt::getopt) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility getopt::getopt)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_gamma_beta_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_gamma_beta_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_column_to_image_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_column_to_image_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_transpose_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_transpose_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_permute_scale_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_permute_scale_instance)
# if(GPU_TARGETS MATCHES "gfx9") if(GPU_TARGETS MATCHES "gfx9")
# if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
# endif() endif()
# if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_silu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_silu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fixed_nk_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fixed_nk_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_multiple_d_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_multiple_d_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_tile_loop_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_tile_loop_instance)
# endif() endif()
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_multiply_add_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_multiply_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_universal_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_universal_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
# endif() endif()
# if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11") if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11")
# if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
# endif() endif()
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
# endif() endif()
# if(DL_KERNELS) if(DL_KERNELS)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
# endif() endif()
rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler) rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
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