Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 15:18:26 +01:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Minor arithmetic improvement to mmvq wrapper kernel (#7172)"

Browse Source 
This reverts commit 8c570c9496.
This commit is contained in:
Joe Todd 2024-06-14 22:22:57 +01:00
parent 18133cab40
commit 6f6612570e
1 changed files with 9 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ggml-sycl.cpp
View File

@ -7984,15 +7984,13 @@ static void mul_mat_vec_q(const void * __restrict__ vx, const void * __restrict_
const int blocks_per_row = ncols / qk; const int blocks_per_row = ncols / qk;
const int blocks_per_warp = vdr * WARP_SIZE / qi; const int blocks_per_warp = vdr * WARP_SIZE / qi;
const int qi_vdr = (qi / vdr); // N_threads processing 1 qk block
// partial sum for each thread // partial sum for each thread
float tmp = 0.0f; float tmp = 0.0f;
const block_q_t * x = (const block_q_t *) vx; const block_q_t * x = (const block_q_t *) vx;
const block_q8_1 * y = (const block_q8_1 *) vy; const block_q8_1 * y = (const block_q8_1 *) vy;
for (int i = item_ct1.get_local_id(2) / qi_vdr; i < blocks_per_row; for (int i = item_ct1.get_local_id(2) / (qi / vdr); i < blocks_per_row;
i += blocks_per_warp) { i += blocks_per_warp) {
const int ibx = row*blocks_per_row + i; // x block index const int ibx = row*blocks_per_row + i; // x block index
@ -8000,8 +7998,8 @@ static void mul_mat_vec_q(const void * __restrict__ vx, const void * __restrict_
const int iqs = const int iqs =
vdr * vdr *
(item_ct1.get_local_id(2) - (item_ct1.get_local_id(2) %
i * qi_vdr); // x block quant index when casting the quants to int (qi / vdr)); // x block quant index when casting the quants to int
tmp += vec_dot_q_sycl(&x[ibx], &y[iby], iqs); tmp += vec_dot_q_sycl(&x[ibx], &y[iby], iqs);
} }