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

Faster CPU PP performance for Qwen3-Next - optimize concat (#1276)

This commit is contained in:
Kawrakow 2026-02-17 11:46:27 +01:00 committed by GitHub
parent 35c99f9f41
commit 16fe459a49
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ggml/src/ggml.c
View File

@ -14661,6 +14661,18 @@ static void ggml_compute_forward_concat_f32(
return;
}
if (ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && ggml_is_contiguous(dst) && dim == 2 && dst->ne[3] > 1 && src1->ne[2] == 1) {
for (int i3 = ith; i3 < (int)dst->ne[3]; i3 += nth) {
char * dst_ptr = (char *)dst->data + i3*dst->nb[3];
char * src_ptr = (char *)src0->data + i3*src0->nb[3];
memcpy(dst_ptr, src_ptr, src0->nb[3]);
dst_ptr += src0->nb[3];
src_ptr = (char *)src1->data + i3*src1->nb[3];
memcpy(dst_ptr, src_ptr, src1->nb[3]);
}
return;
}
int64_t o[4] = {0, 0, 0, 0};
o[dim] = src0->ne[dim];