[SYCL][CUDA] Add managed mem check to enqueue prefetch (#5446)

This PR adds a managed memory check in cuda_piextUSMEnqueuePrefetch.
cuMemPrefetchAsync only works on managed memory and returns an error if host or device memory is passed.

This check ignores the prefetch hint if the memory provided is not managed.
A warning is emitted to the user if the prefetch is ignored.

This is a partial fix to: #5209

Author: AidanBeltonS

sycl/plugins/cuda/pi_cuda.cpp

@@ -4762,6 +4762,15 @@ pi_result cuda_piextUSMEnqueuePrefetch(pi_queue queue, const void *ptr,
     return PI_PLUGIN_SPECIFIC_ERROR;
   }
 
+  unsigned int is_managed;
+  PI_CHECK_ERROR(cuPointerGetAttribute(
+      &is_managed, CU_POINTER_ATTRIBUTE_IS_MANAGED, (CUdeviceptr)ptr));
+  if (!is_managed) {
+    setErrorMessage("Prefetch hint ignored as prefetch only works with USM",
+                    PI_SUCCESS);
+    return PI_PLUGIN_SPECIFIC_ERROR;
+  }
+
   // flags is currently unused so fail if set
   if (flags != 0)
     return PI_INVALID_VALUE;