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

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

Browse source
This commit is contained in:
Slaren 2023-04-29 22:25:00 +02:00
parent 08e539d5e4
commit 476f46f7cc
1 changed files with 5 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml-cuda.cu
View file

@ -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 * 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); cudaError_t err = cudaMallocHost((void **) &ptr, size);
if (err != cudaSuccess) { if (err != cudaSuccess) {
fprintf(stderr, "WARNING: failed to allocate %.2f MB of pinned memory: %s\n", fprintf(stderr, "WARNING: failed to allocate %.2f MB of pinned memory: %s\n",
size/1024.0/1024.0, cudaGetErrorString(err)); size/1024.0/1024.0, cudaGetErrorString(err));
return NULL; return nullptr;
} }
return ptr; return ptr;