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 14, 05:30 UTC
fix(nix): Allow local oneDNN path to fix vLLM CPU build failure (#26401)
Signed-off-by: lyd1992 <liuyudong@iscas.ac.cn> Signed-off-by: ihb2032 <1355790728@qq.com>
ihb2032 committed
Oct 11, 2025 at 17:12 UTC
086609de64456bfcfa44e1f9236a940122156b02
1 file changed
+19
-8
cmake/cpu_extension.cmake
+19
-8
@@ -198,13 +198,24 @@ else()
198
endif()
199
200
if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON_FOUND) OR POWER9_FOUND OR POWER10_FOUND OR POWER11_FOUND)
201
- FetchContent_Declare(
202
- oneDNN
203
- GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
204
- GIT_TAG v3.9
205
- GIT_PROGRESS TRUE
206
- GIT_SHALLOW TRUE
207
- )
201
+ set(FETCHCONTENT_SOURCE_DIR_ONEDNN "$ENV{FETCHCONTENT_SOURCE_DIR_ONEDNN}" CACHE PATH "Path to a local oneDNN source directory.")
202
+
203
+ if(FETCHCONTENT_SOURCE_DIR_ONEDNN)
204
+ message(STATUS "Using oneDNN from specified source directory: ${FETCHCONTENT_SOURCE_DIR_ONEDNN}")
205
+ FetchContent_Declare(
206
+ oneDNN
207
+ SOURCE_DIR ${FETCHCONTENT_SOURCE_DIR_ONEDNN}
208
+ )
209
+ else()
210
+ message(STATUS "Downloading oneDNN from GitHub")
211
+ FetchContent_Declare(
212
+ oneDNN
213
+ GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
214
+ GIT_TAG v3.9
215
+ GIT_PROGRESS TRUE
216
+ GIT_SHALLOW TRUE
217
+ )
218
+ endif()
219
220
if(USE_ACL)
221
find_library(ARM_COMPUTE_LIBRARY NAMES arm_compute PATHS $ENV{ACL_ROOT_DIR}/build/)
@@ -227,7 +238,7 @@ if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON
238
set(ONEDNN_ENABLE_ITT_TASKS "OFF")
239
set(ONEDNN_ENABLE_MAX_CPU_ISA "OFF")
240
set(ONEDNN_ENABLE_CPU_ISA_HINTS "OFF")
230
- set(ONEDNN_VERBOSE "ON")
241
+ set(ONEDNN_VERBOSE "OFF")
242
set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)
243
244
FetchContent_MakeAvailable(oneDNN)