From 0b6de3a1357cb920fd25bd1139541ca6a4c8663e Mon Sep 17 00:00:00 2001
From: ntukanov <ntukanov@nvidia.com>
Date: Wed, 10 Jul 2024 15:59:24 -0700
Subject: [PATCH] ggml : add NVPL BLAS support

---
 Makefile               |  6 ++++++
 ggml/src/ggml-blas.cpp | 11 ++++++++---
 2 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/Makefile b/Makefile
index 668b38b99..fbfca7610 100644
--- a/Makefile
+++ b/Makefile
@@ -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
diff --git a/ggml/src/ggml-blas.cpp b/ggml/src/ggml-blas.cpp
index d709a357b..f7604e580 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(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;