Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : add NVPL BLAS support (#8329) (#8425)

Browse Source 
* ggml : add NVPL BLAS support

* ggml : replace `<BLASLIB>_ENABLE_CBLAS` with `GGML_BLAS_USE_<BLASLIB>`

---------

Co-authored-by: ntukanov <ntukanov@nvidia.com>
This commit is contained in:
Nicholai Tukanov 2024-07-11 11:49:15 -05:00 committed by GitHub
parent b078c619aa
commit 368645698a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 16 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Makefile
View File

@ -547,11 +547,17 @@ ifdef GGML_OPENBLAS64
endif # GGML_OPENBLAS64
ifdef GGML_BLIS
MK_CPPFLAGS += -DGGML_USE_BLAS -I/usr/local/include/blis -I/usr/include/blis
MK_CPPFLAGS += -DGGML_USE_BLAS -DGGML_BLAS_USE_BLIS -I/usr/local/include/blis -I/usr/include/blis
MK_LDFLAGS += -lblis -L/usr/local/lib
OBJ_GGML += ggml/src/ggml-blas.o
endif # GGML_BLIS
ifdef GGML_NVPL
MK_CPPFLAGS += -DGGML_USE_BLAS -DGGML_BLAS_USE_NVPL -DNVPL_ILP64 -I/usr/local/include/nvpl_blas -I/usr/include/nvpl_blas
MK_LDFLAGS += -L/usr/local/lib -lnvpl_blas_core -lnvpl_blas_ilp64_gomp
OBJ_GGML += ggml/src/ggml-blas.o
endif # GGML_NVPL
ifndef GGML_NO_LLAMAFILE
MK_CPPFLAGS += -DGGML_USE_LLAMAFILE
OBJ_GGML += ggml/src/llamafile/sgemm.o

13
ggml/src/ggml-blas.cpp
View File

@ -8,11 +8,12 @@
# include <Accelerate/Accelerate.h>
#elif defined(GGML_BLAS_USE_MKL)
# include <mkl.h>
#elif defined(GGML_BLAS_USE_BLIS)
# include <blis.h>
#elif defined(GGML_BLAS_USE_NVPL)
# include <nvpl_blas.h>
#else
# include <cblas.h>
# ifdef BLIS_ENABLE_CBLAS
# include <blis.h>
# endif
#endif
struct ggml_backend_blas_context {
@ -140,10 +141,14 @@ static void ggml_backend_blas_mul_mat(ggml_backend_blas_context * ctx, struct gg
openblas_set_num_threads(ctx->n_threads);
#endif
#if defined(BLIS_ENABLE_CBLAS)
#if defined(GGML_BLAS_USE_BLIS)
bli_thread_set_num_threads(ctx->n_threads);
#endif
#if defined(GGML_BLAS_USE_NVPL)
nvpl_blas_set_num_threads(ctx->n_threads);
#endif
for (int64_t i13 = 0; i13 < ne13; i13++) {
for (int64_t i12 = 0; i12 < ne12; i12++) {
const int64_t i03 = i13/r3;