ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-22 22:24:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix #230 (#231)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-02-24 09:29:58 +02:00
committed by GitHub
parent ac1d259b93
commit 547eee81d9
1 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -4062,7 +4062,7 @@ static void mul_mat_q4_0_r8_q8_1(int n, const void * vx, size_t bx, const DataIn
template <int nrc_y>
static void mul_mat_q5_0_r4_q8_1_avx2(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%8 == 0);
GGML_ASSERT(nrc_x%4 == 0);
Q8<nrc_y, block_q8_1_x4> q8(info);
auto m4 = _mm256_set1_epi8(0xf);
auto m5 = _mm256_set1_epi8(0x10);
@@ -4232,7 +4232,7 @@ static void mul_mat_q5_0_r4_q8_1(int n, const void * vx, size_t bx, const DataIn
template <int nrc_y>
static void mul_mat_q6_0_r4_q8_1_avx2(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%8 == 0);
GGML_ASSERT(nrc_x%4 == 0);
Q8<nrc_y, block_q8_1_x4> q8(info);
auto m4 = _mm256_set1_epi8(0xf);
auto m6 = _mm256_set1_epi8(0x30);
@@ -6493,7 +6493,6 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI
#endif
return;
}
GGML_ASSERT(nrc_x%8 == 0);
__m256i qx[2];
__m256i acc[2*nrc_y] = {};
float dy[nrc_y];
@@ -6566,7 +6565,7 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI
template <int nrc_y>
static void mul_mat_q8_KV_q8_KV(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%8 == 0);
GGML_ASSERT(nrc_x%4 == 0);
GGML_ASSERT(n%32 == 0);
__m256i qx[4];
#ifndef HAVE_FANCY_SIMD