EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

ggml : automatic selection of best CPU backend (#10606)

Browse Source 
* ggml : automatic selection of best CPU backend

* amx : minor opt

* add GGML_AVX_VNNI to enable avx-vnni, fix checks
This commit is contained in:
Diego Devesa
2024-12-01 16:12:41 +01:00
committed by GitHub
parent 86dc11c5bc
commit 3420909dff
12 changed files with 599 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cpu/ggml-cpu-aarch64.c
View File

@@ -128,7 +128,7 @@ static inline __m512i sum_i16_pairs_int_32x16(const __m512i x) {
}
static inline __m512i mul_sum_us8_pairs_int32x16(const __m512i ax, const __m512i sy) {
#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
#if defined(__AVX512VNNI__)
const __m512i zero = _mm512_setzero_si512();
return _mm512_dpbusd_epi32(zero, ax, sy);
#else