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

The yield shouldn't be necessary.

Browse source
This commit is contained in:
Kunnis 2024-05-09 23:23:13 -05:00
parent c0557fa20a
commit 4762d79d3d
1 changed files with 0 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml.c
View file

@ -12106,12 +12106,6 @@ UseGgmlGemm2:;
const int64_t ir1_start = dr1 * ith1; const int64_t ir1_start = dr1 * ith1;
const int64_t ir1_end = MIN(ir1_start + dr1, nr1); const int64_t ir1_end = MIN(ir1_start + dr1, nr1);
// threads with no work simply yield (not sure if it helps)
if (ir0_start >= ir0_end || ir1_start >= ir1_end) {
sched_yield();
return;
}
ggml_compute_forward_mul_mat_one_chunk(params, dst, num_rows_per_vec_dot, ir0_start, ir0_end, ir1_start, ir1_end); ggml_compute_forward_mul_mat_one_chunk(params, dst, num_rows_per_vec_dot, ir0_start, ir0_end, ir1_start, ir1_end);
#ifdef GGML_PERF #ifdef GGML_PERF