Guard for prefetchAsync GPU capability. #470 #451 #477

7be5f2c7 · Tim Dettmers · f3232d13 · 7be5f2c7
Commit 7be5f2c7 authored Jul 16, 2023 by Tim Dettmers
Show whitespace changes
Inline Side-by-side

Showing with 5 additions and 0 deletions

csrc/pythonInterface.c csrc/pythonInterface.c +5 -0

No files found.
--- a/csrc/pythonInterface.c
+++ b/csrc/pythonInterface.c
@@ -385,6 +385,11 @@ extern "C"
 	void cprefetch(void *ptr, size_t bytes, int device)
 	{
+		int hasPrefetch = 0;
+		CUDA_CHECK_RETURN(cudaDeviceGetAttribute(&hasPrefetch, cudaDevAttrConcurrentManagedAccess, device)); // 40ns overhead
+		if (hasPrefetch == 0) return;
 		CUDA_CHECK_RETURN(cudaMemPrefetchAsync(ptr, bytes, device, 0));
 		CUDA_CHECK_RETURN(cudaPeekAtLastError());
 	}