Disable inlining on cudagraph fallback tests

ghstack-source-id: de457d9cf9374657da7b65ba12e5a86d7b69b02e
Pull Request resolved: https://github.com/pytorch/pytorch/pull/131557

test/inductor/test_cudagraph_trees.py

@@ -1973,6 +1973,7 @@ if HAS_CUDA and not TEST_WITH_ASAN:
             self.run_static_input_param_test(fn, 6)
 
         @torch._dynamo.config.patch("error_on_recompile", True)
+        @torch._dynamo.config.patch("inline_inbuilt_nn_modules", False)
         @torch._inductor.config.patch("triton.cudagraph_support_input_mutation", True)
         @torch._inductor.config.patch("triton.cudagraph_unexpected_rerecord_limit", 0)
         def test_fallback_to_eager_if_recompiling_too_many_times(self):
@@ -2008,6 +2009,7 @@ if HAS_CUDA and not TEST_WITH_ASAN:
             self.assertEqual(counters["inductor"]["cudagraph_skips"], 1)
 
         @torch._dynamo.config.patch("error_on_recompile", True)
+        @torch._dynamo.config.patch("inline_inbuilt_nn_modules", False)
         @torch._inductor.config.patch("triton.cudagraph_support_input_mutation", True)
         @torch._inductor.config.patch("triton.cudagraph_unexpected_rerecord_limit", 0)
         def test_fallback_to_eager_if_recompiling_too_many_times_warn_only_once(self):
@@ -2052,6 +2054,7 @@ if HAS_CUDA and not TEST_WITH_ASAN:
             )
             self.assertEqual(counters["inductor"]["cudagraph_skips"], 2)
 
+        @torch._dynamo.config.patch("inline_inbuilt_nn_modules", False)
         @torch._inductor.config.patch("triton.cudagraph_support_input_mutation", True)
         @torch._inductor.config.patch("triton.cudagraph_unexpected_rerecord_limit", 0)
         def test_fallback_to_eager_if_recompiling_too_many_times_due_to_cudagraph_managed_tensor(
@@ -2096,6 +2099,7 @@ if HAS_CUDA and not TEST_WITH_ASAN:
             ).run(captured_output[0])
             self.assertEqual(counters["inductor"]["cudagraph_skips"], 1)
 
+        @torch._dynamo.config.patch("inline_inbuilt_nn_modules", False)
         @torch._dynamo.config.patch("error_on_recompile", True)
         @torch._dynamo.config.patch("inline_inbuilt_nn_modules", True)
         @torch._inductor.config.patch("triton.cudagraph_unexpected_rerecord_limit", 1)

torch/_dynamo/variables/builder.py

@@ -1133,6 +1133,9 @@ class VariableBuilder:
                 source_i = GetItemSource(base=source, index=i, index_is_slice=False)
                 # access unpacked tensor from this list instead of from a lifted arg
                 self.tx.output.input_source_to_var[source_i] = tensor_variable
+                tensor_variable.proxy.node.meta["tensor_dict"] = value[
+                    i
+                ].__dict__.copy()
 
                 guard = functools.partial(
                     GuardBuilder.TENSOR_MATCH, value=TensorWeakRef(value[i])