Skip to content

GitLab

Commit e983c804 authored by zhuwenwen's avatar zhuwenwen
Browse files

add --gpu-max-threads-per-block=1024 to fix rms_norm_kernel accuracy

parent 0dc4c8e9
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 3 deletions
cmake/utils.cmake
View file @ e983c804
...@@ -119,10 +119,11 @@ function (get_torch_gpu_compiler_flags OUT_GPU_FLAGS GPU_LANG) ...@@ -119,10 +119,11 @@ function (get_torch_gpu_compiler_flags OUT_GPU_FLAGS GPU_LANG)
list(APPEND GPU_FLAGS list(APPEND GPU_FLAGS
"-DUSE_ROCM" "-DUSE_ROCM"
#"-DENABLE_FP8" # "-DENABLE_FP8"
"-U__HIP_NO_HALF_CONVERSIONS__" "-U__HIP_NO_HALF_CONVERSIONS__"
"-U__HIP_NO_HALF_OPERATORS__" "-U__HIP_NO_HALF_OPERATORS__"
"-fno-gpu-rdc") "-fno-gpu-rdc"
"--gpu-max-threads-per-block=1024")
endif() endif()
set(${OUT_GPU_FLAGS} ${GPU_FLAGS} PARENT_SCOPE) set(${OUT_GPU_FLAGS} ${GPU_FLAGS} PARENT_SCOPE)
...@@ -451,4 +452,4 @@ function (define_gpu_extension_target GPU_MOD_NAME) ...@@ -451,4 +452,4 @@ function (define_gpu_extension_target GPU_MOD_NAME)
endif() endif()
install(TARGETS ${GPU_MOD_NAME} LIBRARY DESTINATION ${GPU_DESTINATION} COMPONENT ${GPU_MOD_NAME}) install(TARGETS ${GPU_MOD_NAME} LIBRARY DESTINATION ${GPU_DESTINATION} COMPONENT ${GPU_MOD_NAME})
endfunction() endfunction()
\ No newline at end of file
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