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: add note 

Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo 2025-12-14 19:22:14 +08:00
parent aae6d1e9b0
commit 717531b1a7
No known key found for this signature in database
1 changed files with 8 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
ggml/src/ggml-blas/ggml-blas.cpp
View File

@ -513,13 +513,14 @@ static bool ggml_backend_blas_device_supports_op(ggml_backend_dev_t dev, const g
// TODO: find the optimal value // TODO: find the optimal value
const int64_t min_batch = 32; const int64_t min_batch = 32;
return ggml_is_contiguous(src0) && return ggml_is_contiguous(src0)
ggml_is_contiguous(src1) && && ggml_is_contiguous(src1)
src0->view_src == nullptr && && src1->type == GGML_TYPE_F32
src1->view_src == nullptr && // NOTE: llama-bench creates views that somehow does not go through init_tensor
src1->type == GGML_TYPE_F32 && // this prevents the uninitialized views from being used in BLAS
(ne0 >= min_batch && ne1 >= min_batch && ne10 >= min_batch) && && src0->view_src == nullptr && src1->view_src == nullptr
(src0->type == GGML_TYPE_F32 || ggml_get_type_traits(src0->type)->to_float != NULL); && (ne0 >= min_batch && ne1 >= min_batch && ne10 >= min_batch)
&& (src0->type == GGML_TYPE_F32 || ggml_get_type_traits(src0->type)->to_float != NULL);
} }
default: default: