nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ggml: fix HC_POST single-token CPU chunk count (#2357)

This commit is contained in:
Joel Farthing 2026-08-25 09:29:06 -05:00 committed by GitHub
parent c49f7db34a
commit 08b500b958
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml.c
View File

@ -24471,7 +24471,8 @@ static void ggml_compute_forward_hc_post_f32(
const float * post_r = (const float *)((const char *)post->data); const float * post_r = (const float *)((const char *)post->data);
const float * comb_r = (const float *)((const char *)comb->data); const float * comb_r = (const float *)((const char *)comb->data);
int nchunk = (ne0 + nth - 1)/nth; // chunks are 64 elements wide (see `first` below); threads stride over chunks
int nchunk = (ne0 + 63)/64;
for (int ic = ith; ic < nchunk; ic += nth) { for (int ic = ith; ic < nchunk; ic += nth) {
int first = 64*ic; int first = 64*ic;