skip torchbind in cosntant folding (#148993)

Summary:
Do not fold torchbind objects in constant folding

Any operation on these torchbind objects can have arbitrary side effects, so we can't effectively constant fold anything torchbind-obj-related anyway.

Test Plan:
```
buck run fbcode//mode/dev-nosan //caffe2/test/inductor:torchbind -- -r aot_compile_constant_folding
```

Reviewed By: angelayi

Differential Revision: D69946541

Pull Request resolved: https://github.com/pytorch/pytorch/pull/148993
Approved by: https://github.com/angelayi

test/inductor/test_torchbind.py

@@ -234,6 +234,13 @@ class TestTorchbind(TestCase):
         # TODO: add accuracy test after we support loading and running compiled models with
         # torchbind objects.
 
+    @torch._inductor.config.patch("aot_inductor.use_runtime_constant_folding", True)
+    def test_torchbind_aot_compile_constant_folding(self):
+        ep, inputs, _, _ = self.get_exported_model()
+        aot_compile(ep.module(), inputs, options={"aot_inductor.package": True})
+        # TODO: add accuracy test after we support loading and running compiled models with
+        # torchbind objects.
+
 
 if __name__ == "__main__":
     run_tests()

torch/_inductor/compile_fx.py

@@ -1070,7 +1070,16 @@ class _InProcessFxCompile(FxCompile):
                 const_kernel_code = None
 
                 if aot_mode and config.aot_inductor.use_runtime_constant_folding:
-                    const_gm, const_output_index = split_const_gm(gm)
+                    # torchbind objects have name that starts with _torchbind_obj
+                    # See caffe2/torch/fx/_symbolic_trace.py?lines=406
+                    # We don't use node.meta["val"] because we don't typically
+                    # attach meta["val"] for get_attr nodes.
+                    const_gm, const_output_index = split_const_gm(
+                        gm,
+                        skip_folding_node_fn=lambda node: node.op == "get_attr"
+                        and isinstance(node.target, str)
+                        and node.target.startswith("_torchbind_obj"),
+                    )
 
                     const_graph = GraphLowering(
                         const_gm,

torch/_inductor/constant_folding.py

@@ -233,6 +233,10 @@ class ConstantFolder(torch.fx.Interpreter):
             return self.unknown_value
 
         out = self._deduce_value(node)
+
+        if isinstance(out, torch._C.ScriptObject):
+            return out
+
         if out == self.unknown_value:
             return self.unknown_value