Add comment to the reason of adding flag

e521c0e7 · Po-Yen, Chen · 5a48bfa5 · e521c0e7
Commit e521c0e7 authored Oct 25, 2023 by Po-Yen, Chen
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt ...ary/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt +2 -0

No files found.
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -108,6 +108,8 @@ if (ENABLE_PIPELINE_V2_OPT)
       CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
    )
+    # TODO: The "-vectorize-slp=false" LLVM option is a workaround to prevent inefficient instruction scheduling
+    #       caused by SLP Vectorizer. Remove this option after fix the SLP Vectorizer issue.
    # layout=NT
    set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
       COMPILE_OPTIONS ";-mllvm;-vectorize-slp=false"