fix regression which creates a new fake tensor (#111864)

Summary:
Fixes regression identified here: https://github.com/pytorch/pytorch/pull/111565/files/ccd6b373b5fba81b6141a6b93b2ffb6595dd49ae#r1369334484

Now that `get_fake_value` will identify aliases, we should not try to wrap the fake value again.

X-link: pytorch/pytorch#111864
Approved by: https://github.com/eellison

Reviewed By: izaitsevfb

Differential Revision: D50627345

fbshipit-source-id: a02ae99d1cf523db38f25675a8baa7a853792850

userbenchmark/dynamo/dynamobench/_dynamo/utils.py

@@ -1360,7 +1360,7 @@ def get_fake_value(node, tx):
 
     op = node.op
 
-    # FX Node should always return the same value
+    # FX Node should always return the same fake value
     if "example_value" in node.meta and is_fake(node.meta["example_value"]):
         return node.meta["example_value"]