ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-21 05:34:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix nb4

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-06-10 17:56:55 +02:00
parent 9593e163db
commit 3593891f39
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml-quants.c
View File

@@ -870,13 +870,13 @@ void quantize_row_q8_0(const float * restrict x, void * restrict vy, int64_t k)
block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
block_q8_0_x4 * y4 = (block_q8_0_x4 *)vy;
#if GGML_USE_IQK_MULMAT
const int nb4 = 4*(nb/4);
#else
const int nb4 = -1;
#endif
#if defined(__ARM_NEON)
block_q8_0_x4 * y4 = (block_q8_0_x4 *)vy;
for (int i = 0; i < nb; i++) {
int i4 = i/4, ir = i%4;
float32x4_t srcv [8];
@@ -953,7 +953,6 @@ void quantize_row_q8_0(const float * restrict x, void * restrict vy, int64_t k)
}
#elif defined(__AVX2__) || defined(__AVX__)
block_q8_0_x4 * y4 = (block_q8_0_x4 *)vy;
int nb4 = 4*(nb/4);
#ifdef __AVX2__
const bool pack = true;
#else