ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Better gemm/gemv on AVX2 fr q4_0_r8 (#331)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-04-15 17:18:50 +02:00
committed by GitHub
parent 1a786850e6
commit 5a98a66b5c
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -3620,8 +3620,7 @@ inline __m256i accum_q4_0_quants(const __m256i * v, const int8_t * qs) {
_mm256_maddubs_epi16(v[5], _mm256_shuffle_epi32(yh, 0x55)));
auto sumi4 = _mm256_add_epi16(_mm256_maddubs_epi16(v[6], _mm256_shuffle_epi32(yh, 0xaa)),
_mm256_maddubs_epi16(v[7], _mm256_shuffle_epi32(yh, 0xff)));
auto sumi = _mm256_add_epi32(_mm256_madd_epi16(_mm256_set1_epi16(1), _mm256_add_epi16(sumi1, sumi2)),
_mm256_madd_epi16(_mm256_set1_epi16(1), _mm256_add_epi16(sumi3, sumi4)));
auto sumi = _mm256_madd_epi16(_mm256_set1_epi16(1), _mm256_add_epi16(_mm256_add_epi16(sumi1, sumi2), _mm256_add_epi16(sumi3, sumi4)));
#endif
return sumi;
}