vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix non-vectorised q4_1 vec mul

Browse source
This commit is contained in:
Matvey Soloviev 2023-03-16 00:36:49 +01:00
parent edbd4a0534
commit e13991e7cc
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml.c
View file

@ -1684,11 +1684,11 @@ inline static void ggml_vec_dot_q4_1(const int n, float * restrict s, const void
#else #else
// scalar // scalar
for (int i = 0; i < nb; i++) { for (int i = 0; i < nb; i++) {
const float * m0 = (const float *) (pm0 + i*bs); const float m0 = *(const float *) (pm0 + i*bs);
const float * m1 = (const float *) (pm1 + i*bs); const float m1 = *(const float *) (pm1 + i*bs);
const float * d0 = (const float *) (pd0 + i*bs); const float d0 = *(const float *) (pd0 + i*bs);
const float * d1 = (const float *) (pd1 + i*bs); const float d1 = *(const float *) (pd1 + i*bs);
const uint8_t * restrict p0 = pb0 + i*bs; const uint8_t * restrict p0 = pb0 + i*bs;
const uint8_t * restrict p1 = pb1 + i*bs; const uint8_t * restrict p1 = pb1 + i*bs;