ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-03 10:30:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix AVX2

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-06-14 11:20:35 +03:00
parent 13ce76a85c
commit b02a73c1ec
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/iqk/iqk_gemm_kquants.cpp
View File

@@ -953,10 +953,10 @@ static void mul_mat_qY_K_q8_2_X4_T(int n, const void * vx, size_t bx, const Data
// b0 b2 b4 b6 b1 b3 b5 b7
sumi1 = _mm256_add_epi32(_mm256_unpacklo_epi64(sumi1, sumi3), _mm256_unpackhi_epi64(sumi1, sumi3));
#else
auto sumi1 = _mm256_maddubs_epi16(deq.bits.values[0], _mm256_loadu_si256((const __m256i*)y.qs+0));
auto sumi2 = _mm256_maddubs_epi16(deq.bits.values[1], _mm256_loadu_si256((const __m256i*)y.qs+1));
auto sumi3 = _mm256_maddubs_epi16(deq.bits.values[2], _mm256_loadu_si256((const __m256i*)y.qs+2));
auto sumi4 = _mm256_maddubs_epi16(deq.bits.values[3], _mm256_loadu_si256((const __m256i*)y.qs+3));
auto sumi1 = _mm256_maddubs_epi16(us[0], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i*)qs+0), deq.bits.values[0]));
auto sumi2 = _mm256_maddubs_epi16(us[1], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i*)qs+1), deq.bits.values[1]));
auto sumi3 = _mm256_maddubs_epi16(us[2], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i*)qs+2), deq.bits.values[2]));
auto sumi4 = _mm256_maddubs_epi16(us[3], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i*)qs+3), deq.bits.values[3]));
sumi1 = _mm256_add_epi16(_mm256_unpacklo_epi32(sumi1, sumi2), _mm256_unpackhi_epi32(sumi1, sumi2));
sumi3 = _mm256_add_epi16(_mm256_unpacklo_epi32(sumi3, sumi4), _mm256_unpackhi_epi32(sumi3, sumi4));
sumi1 = _mm256_add_epi16(_mm256_unpacklo_epi64(sumi1, sumi3), _mm256_unpackhi_epi64(sumi1, sumi3));