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

Fix issues with SSE3 version for vec_dot_q4_0_b16_q8_0_b16

Browse source
This commit is contained in:
Srihari-mcw 2024-05-24 10:33:54 -07:00
parent 9e5174ce5d
commit c480818d97
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-quants.c
View file

@ -4486,7 +4486,7 @@ void ggml_vec_dot_q4_0_b16_q8_0_b16(int n, float * restrict s, size_t bs, const
_mm_prefetch(&y[0] + sizeof(block_q8_0), _MM_HINT_T0);
// Compute combined scale for the block 0 and 1
const __m128 d_0_1 = _mm_set1_ps( GGML_BF16_TO_FP32(ggml_make_bf16(x[i].d)) * GGML_BF16_TO_FP32(ggml_make_bf16(y[i].d)));
const __m128 d_0_1 = _mm_set1_ps( GGML_BF16_TO_FP32(ggml_make_bf16(x[0].d)) * GGML_BF16_TO_FP32(ggml_make_bf16(y[0].d)));
const __m128i tmp_0_1 = _mm_loadu_si128((const __m128i *)x[0].qs);