@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 717a5f82cd
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 16, 10:45 UTC

[Bugfix][CI/Build] Fix CUDA 11.8 Build (#9386)

Lucas Wilkinson committed Oct 15, 2024 at 20:15 UTC 717a5f82cda6dd6a52be6504179adaa64bbdc67a
1 file changed +5 -5
CMakeLists.txt
+5 -5
@@ -286,10 +286,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
286 list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_C3X=1")
287 message(STATUS "Building scaled_mm_c3x for archs: ${SCALED_MM_3X_ARCHS}")
288 else()
289 - # clear SCALED_MM_3X_ARCHS so the scaled_mm_c2x kernels know we didn't
290 - # build any 3x kernels
291 - set(SCALED_MM_3X_ARCHS)
292 -
289 if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_3X_ARCHS)
290 message(STATUS "Not building scaled_mm_c3x as CUDA Compiler version is "
291 "not >= 12.0, we recommend upgrading to CUDA 12.0 or "
@@ -299,13 +295,17 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
295 message(STATUS "Not building scaled_mm_c3x as no compatible archs found "
296 "in CUDA target architectures")
297 endif()
298 +
299 + # clear SCALED_MM_3X_ARCHS so the scaled_mm_c2x kernels know we didn't
300 + # build any 3x kernels
301 + set(SCALED_MM_3X_ARCHS)
302 endif()
303
304 #
305 # For the cutlass_scaled_mm kernels we want to build the c2x (CUTLASS 2.x)
306 # kernels for the remaining archs that are not already built for 3x.
307 cuda_archs_loose_intersection(SCALED_MM_2X_ARCHS
308 - "7.5;8.0;8.6;8.9;9.0;9.0a" "${CUDA_ARCHS}")
308 + "7.5;8.0;8.6;8.9;9.0" "${CUDA_ARCHS}")
309 # subtract out the archs that are already built for 3x
310 list(REMOVE_ITEM SCALED_MM_2X_ARCHS ${SCALED_MM_3X_ARCHS})
311 if (SCALED_MM_2X_ARCHS)