From 5a8ae474f071ddeb8e9925889c081f984d47eab7 Mon Sep 17 00:00:00 2001
From: Nicholai Tukanov <nicholaitukanov@gmail.com>
Date: Thu, 11 Jul 2024 11:49:15 -0500
Subject: [PATCH] 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>
---
 ggml/src/ggml-blas.cpp | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/ggml/src/ggml-blas.cpp b/ggml/src/ggml-blas.cpp
index d709a357..a37aa407 100644
--- a/ggml/src/ggml-blas.cpp
+++ b/ggml/src/ggml-blas.cpp
@@ -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;