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

ggml : add NVPL BLAS support

Browse source
This commit is contained in:
ntukanov 2024-07-10 15:59:24 -07:00
parent dd07a123b7
commit 0b6de3a135
2 changed files with 14 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
Makefile
View file

@ -552,6 +552,12 @@ ifdef GGML_BLIS
OBJ_GGML += ggml/src/ggml-blas.o
endif # GGML_BLIS
ifdef GGML_NVPL
MK_CPPFLAGS += -DGGML_USE_BLAS -DNVPL_ENABLE_CBLAS -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

11
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(BLIS_ENABLE_CBLAS)
# include <blis.h>
#elif defined(NVPL_ENABLE_CBLAS)
# include <nvpl_blas.h>
#else
# include <cblas.h>
# ifdef BLIS_ENABLE_CBLAS
# include <blis.h>
# endif
#endif
struct ggml_backend_blas_context {
@ -144,6 +145,10 @@ static void ggml_backend_blas_mul_mat(ggml_backend_blas_context * ctx, struct gg
bli_thread_set_num_threads(ctx->n_threads);
#endif
#if defined(NVPL_ENABLE_CBLAS)
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;