furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 02:05:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix clang-cl AVX-VNNI always_inline target feature mismatch (#2100)

Browse Source 
When building with clang-cl (MSVC + Clang), the CMake MSVC branch defined
__AVXVNNI__ as a preprocessor macro alongside /arch:AVX2, but clang-cl
requires the actual -mavxvnni target feature flag to enable AVX-VNNI
codegen. Without it, clang-cl refused to inline _mm256_dpbusd_avx_epi32
and _mm256_dpwssd_avx_epi32 into functions compiled under /arch:AVX2,
causing 'requires target feature avxvnni' errors in:
  - ggml-quants.c (mul_sum_us8_pairs_float)
  - iqk_gemm_iquants.cpp (mul_mat_iq3_xxs_r4_q8_k)
  - iqk_gemm_kquants.cpp (mul_mat_q3_k_r4_q8_k)
  - iqk_gemm_legacy_quants.cpp (dot, accum_q4_0_quants, operator())

Fix: Detect clang-cl via CMAKE_CXX_COMPILER_ID STREQUAL 'Clang' and
append -mavxvnni to ARCH_FLAGS instead of manual __AVXVNNI__ define.

Also add missing GGML_AVXVNNI handling for the non-MSVC (GCC/Clang on
Linux) branch, passing -mavxvnni as expected.
This commit is contained in:
Nexes the Elder
2026-07-09 09:07:32 +03:00
committed by GitHub
parent 9647246458
commit b2f263a0c4
1 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/CMakeLists.txt
+12 -2
View File
@@ -1139,8 +1139,15 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
elseif (GGML_AVX2)
list(APPEND ARCH_FLAGS /arch:AVX2)
if (GGML_AVXVNNI)
add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVXVNNI__>)
add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVXVNNI__>)
# clang-cl supports -m flags alongside /arch: flags.
# Using -mavxvnni enables the actual target feature for clang,
# which is required for intrinsic functions like _mm256_dpbusd_avx_epi32.
if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
list(APPEND ARCH_FLAGS -mavxvnni)
else()
add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVXVNNI__>)
add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVXVNNI__>)
endif()
endif()
elseif (GGML_AVX)
list(APPEND ARCH_FLAGS /arch:AVX)
@@ -1161,6 +1168,9 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
if (GGML_AVX2)
list(APPEND ARCH_FLAGS -mavx2)
endif()
if (GGML_AVXVNNI)
list(APPEND ARCH_FLAGS -mavxvnni)
endif()
if (GGML_AVX512)
list(APPEND ARCH_FLAGS -mavx512f)
list(APPEND ARCH_FLAGS -mavx512bw)