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

disable for arm

Browse source
This commit is contained in:
slaren 2025-02-09 16:22:56 +01:00
parent 0f0d8c3ae7
commit 1b90527d78
1 changed files with 6 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml/src/ggml-cpu/ggml-cpu.c
View file

@ -7832,11 +7832,16 @@ static void ggml_compute_forward_mul_mat_id(
const int64_t nr0 = ne01;
const int64_t nr1 = cne1;
//int chunk_size = (nr0 + nr1) / nth;
#if defined(__aarch64__)
// disable for ARM
int chunk_size = (nr0 + nr1) / nth;
#else
int chunk_size = 16;
if (nr0 == 1 || nr1 == 1) {
chunk_size = 64;
}
#endif // defined(__aarch64__)
int64_t nchunk0 = (nr0 + chunk_size - 1) / chunk_size;
int64_t nchunk1 = (nr1 + chunk_size - 1) / chunk_size;