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

cuda : fallback to CPU on host buffer alloc fail

Browse source
This commit is contained in:
slaren 2023-12-24 11:59:15 +01:00
parent ab6ad5e6e8
commit 5acc9e50f5
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-cuda.cu
View file

@ -9797,8 +9797,10 @@ static void ggml_backend_cuda_host_buffer_free_buffer(ggml_backend_buffer_t buff
static ggml_backend_buffer_t ggml_backend_cuda_host_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) { static ggml_backend_buffer_t ggml_backend_cuda_host_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
void * ptr = ggml_cuda_host_malloc(size); void * ptr = ggml_cuda_host_malloc(size);
if (ptr == nullptr) { if (ptr == nullptr) {
return nullptr; // fallback to cpu buffer
return ggml_backend_buft_alloc_buffer(ggml_backend_cpu_buffer_type(), size);
} }
// FIXME: this is a hack to avoid having to implement a new buffer type // FIXME: this is a hack to avoid having to implement a new buffer type