mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-14 18:17:44 +00:00
Add -fno-slp-vectorize option for building hstu kernels on gfx950
This commit is contained in:
@@ -17,7 +17,8 @@ if (DEFINED ENV{ASSUME_HIGHLY_VARIED_SEQLEN})
|
||||
endif()
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx95" AND NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx90")
|
||||
list(APPEND EXAMPLE_HSTU_ATTENTION_COMPILE_OPTIONS -DBUILD_HSTU_FOR_GFX95_ONLY)
|
||||
## disable slp-vectorize improve pipelines performance on gfx950
|
||||
list(APPEND EXAMPLE_HSTU_ATTENTION_COMPILE_OPTIONS -DBUILD_HSTU_FOR_GFX95_ONLY -fno-slp-vectorize)
|
||||
endif()
|
||||
|
||||
target_compile_options(${EXAMPLE_HSTU_ATTENTION} PRIVATE ${EXAMPLE_HSTU_ATTENTION_COMPILE_OPTIONS})
|
||||
|
||||
Reference in New Issue
Block a user