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

Unroll loops in dmmv shader

Browse source
This commit is contained in:
0cc4m 2023-07-22 17:46:52 +02:00
parent 2859562501
commit 3452095089
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
vk_shaders/dequant_mul_mat_vec_q4_0.glsl
View file

@ -33,11 +33,11 @@ void main() {
const int row = int(gl_WorkGroupID.x); const int row = int(gl_WorkGroupID.x);
const int tid = int(gl_LocalInvocationID.x); const int tid = int(gl_LocalInvocationID.x);
const int y_offset = QUANT_R == 1 ? 1 : QUANT_K/2; const int y_offset = QUANT_K/2;
tmp[tid] = 0; tmp[tid] = 0;
for (int i = 0; i < p.ncols/block_size; i += 2) { [[unroll]] for (int i = 0; i < p.ncols/block_size; i += 2) {
const int col = i*block_size + 2*tid; const int col = i*block_size + 2*tid;
const int ib = (row*p.ncols + col)/QUANT_K; // block index const int ib = (row*p.ncols + col)/QUANT_K; // block index
const int iqs = (col%QUANT_K)/QUANT_R; // quant index const int iqs = (col%QUANT_K)/QUANT_R; // quant index
@ -61,7 +61,7 @@ void main() {
// sum up partial sums and write back result // sum up partial sums and write back result
barrier(); barrier();
for (int s=block_size/2; s>0; s>>=1) { [[unroll]] for (int s=block_size/2; s>0; s>>=1) {
if (tid < s) { if (tid < s) {
tmp[tid] += tmp[tid + s]; tmp[tid] += tmp[tid + s];
} }