Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 22:08:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : fix q4_1 dot product types

Browse Source
This commit is contained in:
Georgi Gerganov 2023-04-14 09:45:42 +03:00
parent c5d70f5c9e
commit 0f07cacb05
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ggml.c
View File

@ -2344,14 +2344,14 @@ static void ggml_vec_dot_q4_1(const int n, float * restrict s, const void * rest
#if defined(__ARM_FEATURE_DOTPROD) #if defined(__ARM_FEATURE_DOTPROD)
// dot product into int32x4_t // dot product into int32x4_t
int32x4_t p_0 = vdotq_s32(vdupq_n_s32(0), v0_0l, v1_0l); uint32x4_t p_0 = vdotq_u32(vdupq_n_u32(0), v0_0l, v1_0l);
int32x4_t p_1 = vdotq_s32(vdupq_n_s32(0), v0_1l, v1_1l); uint32x4_t p_1 = vdotq_u32(vdupq_n_u32(0), v0_1l, v1_1l);
p_0 = vdotq_s32(p_0, v0_0h, v1_0h); p_0 = vdotq_u32(p_0, v0_0h, v1_0h);
p_1 = vdotq_s32(p_1, v0_1h, v1_1h); p_1 = vdotq_u32(p_1, v0_1h, v1_1h);
sum11 += x0->d*y0->d*vaddvq_s32(p_0); sum11 += x0->d*y0->d*vaddvq_u32(p_0);
sum11 += x1->d*y1->d*vaddvq_s32(p_1); sum11 += x1->d*y1->d*vaddvq_u32(p_1);
#else #else
const uint16x8_t pl0l = vmull_u8(vget_low_u8 (v0_0l), vget_low_u8 (v1_0l)); const uint16x8_t pl0l = vmull_u8(vget_low_u8 (v0_0l), vget_low_u8 (v1_0l));
const uint16x8_t pl0h = vmull_u8(vget_high_u8(v0_0l), vget_high_u8(v1_0l)); const uint16x8_t pl0h = vmull_u8(vget_high_u8(v0_0l), vget_high_u8(v1_0l));