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 13, 08:15 UTC
[Hardware][PowerPC] Fix fp16 compilation error for Power in cpu attention backend and bump oneDNN version (#28535)
Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com>
Akash kaothalkar committed
Nov 13, 2025 at 19:02 UTC
86d15bfd8d681a2ca2f3b2e550149a5ba3282ef1
2 files changed
+4
-2
cmake/cpu_extension.cmake
+2
-2
@@ -242,7 +242,7 @@ if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON
242
SUBBUILD_DIR "${FETCHCONTENT_BASE_DIR}/arm_compute-subbuild"
243
SOURCE_DIR "${FETCHCONTENT_BASE_DIR}/arm_compute-src"
244
GIT_REPOSITORY https://github.com/ARM-software/ComputeLibrary.git
245
- GIT_TAG v52.2.0
245
+ GIT_TAG v52.6.0
246
GIT_SHALLOW TRUE
247
GIT_PROGRESS TRUE
248
)
@@ -310,7 +310,7 @@ if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON
310
FetchContent_Declare(
311
oneDNN
312
GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
313
- GIT_TAG v3.9
313
+ GIT_TAG v3.10
314
GIT_PROGRESS TRUE
315
GIT_SHALLOW TRUE
316
)
csrc/cpu/cpu_attn_impl.hpp
+2
@@ -821,10 +821,12 @@ struct VecTypeTrait<c10::BFloat16> {
821
using vec_t = vec_op::BF16Vec16;
822
};
823
824
+#if !defined(__powerpc__)
825
template <>
826
struct VecTypeTrait<c10::Half> {
827
using vec_t = vec_op::FP16Vec16;
828
};
829
+#endif
830
831
template <typename T>
832
void print_logits(const char* name, T* ptr, int32_t row, int32_t col,