Test that broadcast doesn't copy when dst and src devices are the same (

pytorch#8803)

* test that broadcast doesn't copy when dst and src devices are the same

* only test if input is cuda

test/test_cuda.py

@@ -860,6 +860,8 @@ def _test_broadcast(self, input):
         for i, t in enumerate(result):
             self.assertEqual(t.get_device(), i)
             self.assertEqual(t, input)
+            if input.is_cuda and input.get_device() == i:
+                self.assertEqual(t.data_ptr(), input.data_ptr())
 
     def test_broadcast_cpu(self):
         self._test_broadcast(torch.randn(5, 5))