@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 45a1a69b98
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 17, 00:30 UTC

[Build] Disable sm_90a in cu11 (#5141)

Simon Mo committed May 30, 2024 at 16:37 UTC 45a1a69b9841a4cb7cc70788cf7dea1a2d3ec3d6
1 file changed +8 -6
CMakeLists.txt
+8 -6
@@ -177,7 +177,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
177 include(FetchContent)
178 SET(CUTLASS_ENABLE_HEADERS_ONLY=ON)
179 FetchContent_Declare(
180 - cutlass
180 + cutlass
181 GIT_REPOSITORY https://github.com/nvidia/cutlass.git
182 # CUTLASS 3.5.0
183 GIT_TAG 7d49e6c7e2f8896c47f586706e67e1fb215529dc
@@ -200,11 +200,13 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
200 # The CUTLASS kernels for Hopper require sm90a to be enabled.
201 # This is done via the below gencode option, BUT that creates kernels for both sm90 and sm90a.
202 # That adds an extra 17MB to compiled binary, so instead we selectively enable it.
203 - set_source_files_properties(
204 - "csrc/quantization/cutlass_w8a8/scaled_mm_dq_c3x.cu"
205 - PROPERTIES
206 - COMPILE_FLAGS
207 - "-gencode arch=compute_90a,code=sm_90a")
203 + if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 11)
204 + set_source_files_properties(
205 + "csrc/quantization/cutlass_w8a8/scaled_mm_dq_c3x.cu"
206 + PROPERTIES
207 + COMPILE_FLAGS
208 + "-gencode arch=compute_90a,code=sm_90a")
209 + endif()
210
211 endif()
212