CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

HIP: enable vec fattn on RDNA4 (#14323)

Browse Source
This commit is contained in:
uvos
2025-06-22 16:51:23 +02:00
committed by GitHub
parent 5d5c066de8
commit af3373f1ad
2 changed files with 14 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@@ -100,8 +100,7 @@ int ggml_cuda_get_device() {
static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device) {
ggml_cuda_set_device(device);
cudaError_t err;
if (getenv("GGML_CUDA_ENABLE_UNIFIED_MEMORY") != nullptr)
{
if (getenv("GGML_CUDA_ENABLE_UNIFIED_MEMORY") != nullptr) {
err = cudaMallocManaged(ptr, size);
#if defined(GGML_USE_HIP)
if (err == hipSuccess) {
@@ -119,9 +118,7 @@ static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device)
err = cudaMalloc(ptr, size);
}
#endif // defined(GGML_USE_HIP)
}
else
{
} else {
err = cudaMalloc(ptr, size);
}
return err;