root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 19:04:35 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

ggml : fix vld1q_s8_x4 32-bit compat (#4828)

Browse Source 
* ggml : fix vld1q_s8_x4 32-bit compat

ggml-ci

* ggml : fix 32-bit ARM compat (cont)

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-01-09 10:42:06 +02:00 committed by GitHub
parent 8f900abfc0
commit 18c2e1752c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ggml-quants.c
View File

@ -7250,9 +7250,9 @@ void ggml_vec_dot_iq2_xxs_q8_K(const int n, float * restrict s, const void * res
uint32_t aux32[4]; uint32_t aux32[4];
const uint8_t * aux8 = (const uint8_t *)aux32; const uint8_t * aux8 = (const uint8_t *)aux32;
int8x16x4_t q2u; ggml_int8x16x4_t q2u;
int8x16x4_t q2s; ggml_int8x16x4_t q2s;
int8x16x4_t q8b; ggml_int8x16x4_t q8b;
float sumf = 0; float sumf = 0;
for (int i = 0; i < nb; ++i) { for (int i = 0; i < nb; ++i) {
@ -7261,7 +7261,7 @@ void ggml_vec_dot_iq2_xxs_q8_K(const int n, float * restrict s, const void * res
const int8_t * restrict q8 = y[i].qs; const int8_t * restrict q8 = y[i].qs;
float sumf1 = 0, sumf2 = 0; float sumf1 = 0, sumf2 = 0;
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) { for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
q8b = vld1q_s8_x4(q8); q8 += 64; q8b = ggml_vld1q_s8_x4(q8); q8 += 64;
memcpy(aux32, q2, 4*sizeof(uint32_t)); q2 += 8; memcpy(aux32, q2, 4*sizeof(uint32_t)); q2 += 8;
q2u.val[0] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 0])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 1]))); q2u.val[0] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 0])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 1])));
q2u.val[1] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 2])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 3]))); q2u.val[1] = vcombine_s8(vld1_s8((const void *)(iq2xxs_grid + aux8[ 2])), vld1_s8((const void *)(iq2xxs_grid + aux8[ 3])));