ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 09:16:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix merge from upstream

Browse Source
This commit is contained in:
carlushuang
2024-03-26 14:09:54 +00:00
parent c94b545747
commit 04ee01191a
105 changed files with 16558 additions and 2285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
example/01_gemm/CMakeLists.txt
View File

@@ -27,7 +27,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
if(GPU_TARGETS MATCHES "gfx11")
add_custom_target(example_gemm_wmma)
add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
@@ -53,13 +53,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
list(APPEND gpu_list gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -72,5 +66,12 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif()
endforeach()
add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)

2
example/02_gemm_bilinear/CMakeLists.txt
View File

@@ -1,5 +1,5 @@
list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0)

2
example/03_gemm_bias_relu/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

4
example/04_gemm_add_add_fastgelu/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -33,4 +33,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_lds_direct_load_fp32)
set(target 1)
endif()
endforeach()
endforeach()

4
example/09_convnd_fwd/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -6,6 +6,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
add_example_executable(example_convnd_fwd_xdl_fp8 convnd_fwd_xdl_fp8.cpp)
add_example_executable(example_convnd_fwd_xdl_bf8 convnd_fwd_xdl_bf8.cpp)
# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
set(target 1)

2
example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/14_gemm_quantization/CMakeLists.txt
View File

@@ -1,7 +1,7 @@
# dlops
add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
# xdlops
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/17_convnd_bwd_data/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/18_batched_gemm_reduce/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/20_grouped_conv_bwd_weight/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)

2
example/21_gemm_layernorm/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
set(target 0)

4
example/31_batched_gemm_gemm/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif()
endforeach()
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx95" AND NOT GPU_TARGETS MATCHES "gfx1")
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
endif()

5
example/35_splitK_gemm/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -10,6 +10,9 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
add_example_executable(example_splitK_gemm_xdl_fp16_fp8 splitK_gemm_xdl_fp16_fp8.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16_fp8)
add_example_executable(example_splitK_gemm_xdl_lds_direct_load_fp16 splitK_gemm_xdl_lds_direct_load_fp16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_lds_direct_load_fp16)

2
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)

2
example/40_conv2d_fwd_quantization/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

4
example/41_grouped_conv_conv_fwd/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list2 gfx908 gfx90a)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif()
endforeach()
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx95" AND NOT GPU_TARGETS MATCHES "gfx1")
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
endif()

2
example/44_elementwise_permute/CMakeLists.txt
View File

@@ -5,6 +5,6 @@ add_example_executable(example_elementwise_permute_4D_fp16_row elementwise_permu
add_example_executable(example_elementwise_permute_4D_fp32_col elementwise_permute_4D_fp32_col.cpp)
add_example_executable(example_elementwise_permute_4D_fp16_col elementwise_permute_4D_fp16_col.cpp)
add_example_executable(example_elementwise_permute elementwise_permute.cpp)
if((NOT GPU_TARGETS MATCHES "gfx940") AND (NOT GPU_TARGETS MATCHES "gfx941") AND (NOT GPU_TARGETS MATCHES "gfx942") AND (NOT GPU_TARGETS MATCHES "gfx950"))
if((NOT GPU_TARGETS MATCHES "gfx940") AND (NOT GPU_TARGETS MATCHES "gfx941") AND (NOT GPU_TARGETS MATCHES "gfx942"))
add_example_executable(example_elementwise_permute_3d elementwise_permute_3d.cpp)
endif()

2
example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/52_im2col_col2im/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)

2
example/60_gemm_multi_ABD/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0)

2
example/61_contraction_multi_ABD/CMakeLists.txt
View File

@@ -1,4 +1,4 @@
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0)