Disables cudaMemPrefetchAsync() in unsupported systems (e.g. Windows). …

…Fixes artidoro/qlora#73 and bitsandbytes-foundation#453 (cherry picked from commit e02f078)
jllllll · Jul 8, 2023 · a04320b · a04320b
1 parent 57b56a6
commit a04320b
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/csrc/pythonInterface.c b/csrc/pythonInterface.c
@@ -356,6 +356,10 @@ 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());
 	}