Unverified Commit 5be81251 authored by Tim Moon's avatar Tim Moon Committed by GitHub
Browse files

Fix bug where CUTLASS kernel was not being compiled for SM90a (#2235)


Signed-off-by: default avatarTim Moon <tmoon@nvidia.com>
parent dfe5b7df
......@@ -155,18 +155,12 @@ add_library(transformer_engine SHARED ${transformer_engine_SOURCES})
target_include_directories(transformer_engine PUBLIC
"${CMAKE_CURRENT_SOURCE_DIR}/include")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0)
set_source_files_properties(
"gemm/cutlass_grouped_gemm.cu"
PROPERTIES
COMPILE_FLAGS
"-gencode arch=compute_90a,code=sm_90a")
else()
message(FATAL_ERROR "cutlass gemm/cutlass_grouped_gemm.cu kernel required sm 90a")
endif()
# Disable debug build for cutlass due to hang.
set_source_files_properties("gemm/cutlass_grouped_gemm.cu" PROPERTIES COMPILE_FLAGS "-g0")
# CUTLASS kernels require SM90a and cause hang in debug build
set_property(
SOURCE gemm/cutlass_grouped_gemm.cu
APPEND
PROPERTY
COMPILE_OPTIONS "--generate-code=arch=compute_90a,code=sm_90a;-g0")
# Configure dependencies
target_link_libraries(transformer_engine PUBLIC
......
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