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, 03:00 UTC
CMake build elf without PTX (#3739)
Simon Mo committed
Mar 29, 2024 at 18:53 UTC
51c31bc10ca7c48b580cd58fcd741ba4d6db4447
1 file changed
+4
-1
cmake/utils.cmake
+4
-1
@@ -240,9 +240,12 @@ macro(override_gpu_arches GPU_ARCHES GPU_LANG GPU_SUPPORTED_ARCHES)
240
endif()
241
242
if (_SM)
243
- set(_VIRT "")
243
+ # -real suffix let CMake to only generate elf code for the kernels.
244
+ # we want this, otherwise the added ptx (default) will increase binary size.
245
+ set(_VIRT "-real")
246
set(_CODE_ARCH ${_SM})
247
else()
248
+ # -virtual suffix let CMake to generate ptx code for the kernels.
249
set(_VIRT "-virtual")
250
set(_CODE_ARCH ${_CODE})
251
endif()