ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-25 15:44:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove unnecessary barrier in ggml_compute_forward_mul_mat

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-09-09 12:53:23 +03:00
parent e9bb1a54ee
commit 237a2380ee
1 changed files with 6 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml.c
View File

@@ -13094,14 +13094,14 @@ UseGgmlGemm1:;
int64_t t2 = ggml_time_us();
if (ith == 0) printf("quantize(%s): %d us\n", dst->name, (int)(t2 - t1));
#endif
}
if (ith == 0) {
// Every thread starts at ith, so the first unprocessed chunk is nth. This save a bit of coordination right at the start.
atomic_store(&params->shared->current_chunk, nth);
}
if (ith == 0) {
// Every thread starts at ith, so the first unprocessed chunk is nth. This save a bit of coordination right at the start.
atomic_store(&params->shared->current_chunk, nth);
}
ggml_barrier(params->shared);
ggml_barrier(params->shared);
}
const void * wdata = (src1->type == vec_dot_type) ? src1->data : params->wdata;
@@ -13120,8 +13120,6 @@ UseGgmlGemm1:;
IQK_MulMat_Not_Available2:;
#endif
ggml_barrier(params->shared);
#if GGML_USE_LLAMAFILE
if (src1->type != vec_dot_type) {
const size_t row_size = ggml_row_size(vec_dot_type, ne10);