ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-12-21 05:33:06 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

llamafile : improve sgemm.cpp (llama/6796)

Browse Source 
* llamafile : improve sgemm.cpp

- Re-enable by default
- Fix issue described in #6716
- Make code more abstract, elegant, and maintainable
- Faster handling of weirdly shaped `m` an `n` edge cases

* Address review comments

* Help clang produce fma instructions

* Address review comments
This commit is contained in:
Justine Tunney 2024-04-22 15:00:36 -04:00 committed by Georgi Gerganov
parent a6d264f331
commit 6c3971b29b
1 changed files with 3 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ggml.c
View File

@ -10887,7 +10887,7 @@ static void ggml_compute_forward_mul_mat(
#endif
#if GGML_USE_LLAMAFILE
if (nb10 == ggml_type_size(src1->type)) {
if (src1_cont) {
for (int64_t i13 = 0; i13 < ne13; i13++)
for (int64_t i12 = 0; i12 < ne12; i12++)
if (!llamafile_sgemm(ne01, ne11, ne00/ggml_blck_size(src0->type),
@ -10940,15 +10940,13 @@ UseGgmlGemm1:;
const size_t row_size = ggml_row_size(vec_dot_type, ne10);
#if GGML_USE_LLAMAFILE
if (nb10 == ggml_type_size(src1->type) || src1->type != vec_dot_type) {
if (src1->type != vec_dot_type) {
for (int64_t i13 = 0; i13 < ne13; i13++)
for (int64_t i12 = 0; i12 < ne12; i12++)
if (!llamafile_sgemm(ne01, ne11, ne00/ggml_blck_size(src0->type),
(const char *)src0->data + i12/r2*nb02 + i13/r3*nb03,
nb01/ggml_type_size(src0->type),
(const char *)wdata + ggml_row_size(vec_dot_type,
nb12/ggml_type_size(src1->type)*i12 +
nb13/ggml_type_size(src1->type)*i13),
(const char *)wdata + (i12*ne11 + i13*ne12*ne11)*row_size,
row_size/ggml_type_size(vec_dot_type),
(char *)dst->data + i12*nb2 + i13*nb3,
nb1/ggml_type_size(dst->type),