Fix bf8 conversion issues (#1003)

* Fix the conversion

* Add bf8 functionality

* Enable example on MI200 as well

[ROCm/composable_kernel commit: 1fd27d520f]
This commit is contained in:
Rostyslav Geyyer
2023-10-20 08:00:45 -05:00
committed by GitHub
parent 75e7948ac8
commit 07d78c032b
5 changed files with 116 additions and 75 deletions

View File

@@ -10,10 +10,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
endif()
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
set(target 1)
endif()