ggml : add NVPL BLAS support (ggml/8329) (llama/8425)
* 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:
parent
84493d7f3e
commit
5a8ae474f0
@ -8,11 +8,12 @@
|
|||||||
# include <Accelerate/Accelerate.h>
|
# include <Accelerate/Accelerate.h>
|
||||||
#elif defined(GGML_BLAS_USE_MKL)
|
#elif defined(GGML_BLAS_USE_MKL)
|
||||||
# include <mkl.h>
|
# include <mkl.h>
|
||||||
|
#elif defined(GGML_BLAS_USE_BLIS)
|
||||||
|
# include <blis.h>
|
||||||
|
#elif defined(GGML_BLAS_USE_NVPL)
|
||||||
|
# include <nvpl_blas.h>
|
||||||
#else
|
#else
|
||||||
# include <cblas.h>
|
# include <cblas.h>
|
||||||
# ifdef BLIS_ENABLE_CBLAS
|
|
||||||
# include <blis.h>
|
|
||||||
# endif
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
struct ggml_backend_blas_context {
|
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);
|
openblas_set_num_threads(ctx->n_threads);
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#if defined(BLIS_ENABLE_CBLAS)
|
#if defined(GGML_BLAS_USE_BLIS)
|
||||||
bli_thread_set_num_threads(ctx->n_threads);
|
bli_thread_set_num_threads(ctx->n_threads);
|
||||||
#endif
|
#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 i13 = 0; i13 < ne13; i13++) {
|
||||||
for (int64_t i12 = 0; i12 < ne12; i12++) {
|
for (int64_t i12 = 0; i12 < ne12; i12++) {
|
||||||
const int64_t i03 = i13/r3;
|
const int64_t i03 = i13/r3;
|
||||||
|
Loading…
Reference in New Issue
Block a user