cuBLAS: do not use pinned memory if env variable GGML_CUDA_NO_PINNED is set

Author: slaren

ggml-cuda.cu

@@ -355,13 +355,16 @@ cudaError_t ggml_cuda_h2d_tensor_2d(void * dst, const struct ggml_tensor * src,
 }
 
 void * ggml_cuda_host_malloc(size_t size) {
-    void * ptr = nullptr;
+    if (getenv("GGML_CUDA_NO_PINNED") != nullptr) {
+        return nullptr;
+    }
 
+    void * ptr = nullptr;
     cudaError_t err = cudaMallocHost((void **) &ptr, size);
     if (err != cudaSuccess) {
         fprintf(stderr, "WARNING: failed to allocate %.2f MB of pinned memory: %s\n",
             size/1024.0/1024.0, cudaGetErrorString(err));
-        return NULL;
+        return nullptr;
     }
 
     return ptr;