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

Fixed dmmv dequant for k<= GGML_SYCL_DMMV_X

Browse source
This commit is contained in:
OuadiElfarouki 2024-09-02 14:06:58 +01:00
parent b60074f1c2
commit 3ba9d04c2a
1 changed files with 2 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-sycl/dmmv.cpp
View file

@ -47,6 +47,7 @@ static void dequantize_mul_mat_vec(const void * __restrict__ vx, const dfloat *
for (int i = 0; i < ncols; i += iter_stride) { for (int i = 0; i < ncols; i += iter_stride) {
const int col = i + vals_per_iter*tid; const int col = i + vals_per_iter*tid;
if (col >= ncols) break;
const int ib = (row*ncols + col)/qk; // x block index const int ib = (row*ncols + col)/qk; // x block index
const int iqs = (col%qk)/qr; // x quant index const int iqs = (col%qk)/qr; // x quant index
const int iybs = col - col%qk; // y block start index const int iybs = col - col%qk; // y block start index
@ -54,6 +55,7 @@ static void dequantize_mul_mat_vec(const void * __restrict__ vx, const dfloat *
// processing >2 values per i iter is faster for fast GPUs // processing >2 values per i iter is faster for fast GPUs
#pragma unroll #pragma unroll
for (int j = 0; j < vals_per_iter; j += 2) { for (int j = 0; j < vals_per_iter; j += 2) {
if (col + j >= ncols) break;
// process 2 vals per j iter // process 2 vals per j iter
// dequantize // dequantize