[dynamo] Fix bug in hasattr(tensor, "size") (#152883)

Fixes https://github.com/pytorch/pytorch/issues/135696

Pull Request resolved: https://github.com/pytorch/pytorch/pull/152883
Approved by: https://github.com/StrongerXi

test/dynamo/test_repros.py

@@ -6832,6 +6832,19 @@ class ReproTestsDevice(torch._dynamo.test_case.TestCase):
         out2, _ = torch.compile(moe_mlp, backend="eager")(x)
         self.assertEqual(out1, out2)
 
+    def test_tensor_size_hasattr(self):
+        def fn(x):
+            if hasattr(x, "size"):
+                x = x * 2
+            if hasattr(x, "stride"):
+                x = x * 3
+            return x * 5
+
+        x = torch.ones(4)
+
+        opt_fn = torch.compile(fn, backend="eager", fullgraph=True)
+        self.assertEqual(fn(x), opt_fn(x))
+
     @requires_cuda
     def test_memleak_when_graph_input_has_tensor_attr(self, device):
         @torch.compile(backend="eager")

torch/_dynamo/variables/tensor.py

@@ -394,6 +394,13 @@ class TensorVariable(VariableTracker):
         from . import GetAttrVariable
         from .builtin import BuiltinVariable
 
+        # TODO - This is not a good solution but solves an accuracy issue.
+        # Today, var_getattr returns GetAttrVariable for both non-existent
+        # attributes and existing attributes. This is a bug and requires more
+        # deep dive.
+        if name in ("size", "stride"):
+            return ConstantVariable(True)
+
         try:
             var = BuiltinVariable(getattr).call_function(
                 tx, [self, ConstantVariable(name)], {}