happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ggml-blas: band-aid fix 

this took unnecessarily long to debug

Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo 2025-12-24 20:49:22 +08:00
parent ebef650f88
commit 51c069a913
No known key found for this signature in database
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-blas/ggml-blas.cpp
View File

@ -500,7 +500,7 @@ static bool ggml_backend_blas_device_supports_op(ggml_backend_dev_t dev, const g
const int64_t ne1 = dst->ne[1];
// TODO: find the optimal value
const int64_t min_batch = 32;
const int64_t min_batch = 1024;
return ggml_is_contiguous(src0)
&& ggml_is_contiguous(src1)
@ -508,7 +508,7 @@ static bool ggml_backend_blas_device_supports_op(ggml_backend_dev_t dev, const g
// NOTE: llama-bench creates views that somehow does not go through init_tensor
// this prevents the uninitialized views from being used in BLAS
&& src0->view_src == nullptr && src1->view_src == nullptr
&& (ne0 >= min_batch || ne1 >= min_batch || ne10 >= min_batch)
&& (ne0 >= min_batch && ne1 >= min_batch && ne10 >= min_batch)
&& (src0->type == GGML_TYPE_F32 || ggml_get_type_traits(src0->type)->to_float != NULL);
}