ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-01 20:19:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

iqk_mul_mat(bitnet): fix typo

Browse Source 
With the last change (which added the typo), I'm now getting
PP-512 = 300 t/s on the Ryzen-5975WX.
This commit is contained in:
Kawrakow
2024-06-17 16:50:11 +03:00
parent 7b3cb2b96c
commit 0efd620d01
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
iqk_mul_mat.cpp
View File

@@ -1381,7 +1381,7 @@ IQK_NOINLINE void mul_mat_iq1bn_q8_K64(int n, const void * vx, size_t bx, const
#if defined __AVX512VNNI__ && defined __AVX512VL__
auto dot = _mm256_dpbusd_epi32(_mm256_dpbusd_epi32(_mm256_setzero_si256(), m1_8, dot1), m1_8, dot2);
#else
auto dot = _mm256_madd_epi16(m1_16, _mm256_add_api16(_mm256_maddubs_epi16(m1_8, dot1), _mm256_maddubs_epi16(m1_8, dot2)));
auto dot = _mm256_madd_epi16(m1_16, _mm256_add_epi16(_mm256_maddubs_epi16(m1_8, dot1), _mm256_maddubs_epi16(m1_8, dot2)));
#endif
accd[0] = _mm256_fmadd_ps(_mm256_set1_ps(q8.scale(0, i)), _mm256_cvtepi32_ps(dot), accd[0]);
} else {