Revert "[dynamo] Fix issue with tensors passed as view() shapes (pytorch#156928)"

This reverts commit 75f3e5a.

Reverted pytorch#156928 on behalf of https://github.com/jeanschmidt due to Breaks a internal test, more details can be found on D77449971 ([comment](pytorch#156928 (comment)))

Author: pytorchmergebot

test/dynamo/test_view.py

@@ -33,86 +33,6 @@ def f(t, _n):
         t = torch.tensor([2, 4], dtype=torch.int32)
         f(t, 8)
 
-    def test_view_with_tensor_shape_params(self):
-        # Test for issue #156720: aten.view.default with tensor shape parameters
-        class TestModel(torch.nn.Module):
-            def forward(self, x, shape_params):
-                return torch.ops.aten.view.default(x, shape_params)
-
-        x = torch.randn(24)
-        shape_params = [
-            torch.tensor(2, dtype=torch.int32),
-            torch.tensor(3, dtype=torch.int32),
-            torch.tensor(4, dtype=torch.int32),
-        ]
-
-        model = TestModel()
-        expected = model(x, shape_params)
-
-        compiled_model = torch.compile(model, backend="eager")
-        result = compiled_model(x, shape_params)
-
-        torch.testing.assert_close(result, expected)
-
-    def test_tensor_view_with_tensor_shape_params(self):
-        # Test tensor.view() method with tensor shape parameters (list version)
-        class TestModel(torch.nn.Module):
-            def forward(self, x, shape_params):
-                return x.view(shape_params)
-
-        x = torch.randn(24)
-        shape_params = (
-            torch.tensor(2, dtype=torch.int32),
-            torch.tensor(3, dtype=torch.int32),
-            torch.tensor(4, dtype=torch.int32),
-        )
-
-        model = TestModel()
-        expected = model(x, shape_params)
-
-        compiled_model = torch.compile(model, backend="eager")
-        result = compiled_model(x, shape_params)
-
-        torch.testing.assert_close(result, expected)
-
-    def test_tensor_view_with_tensor_args(self):
-        # Test tensor.view() method with individual tensor arguments
-        class TestModel(torch.nn.Module):
-            def forward(self, x, dim1, dim2, dim3):
-                return x.view(dim1, dim2, dim3)
-
-        x = torch.randn(24)
-        dim1 = torch.tensor(2, dtype=torch.int32)
-        dim2 = torch.tensor(3, dtype=torch.int32)
-        dim3 = torch.tensor(4, dtype=torch.int32)
-
-        model = TestModel()
-        expected = model(x, dim1, dim2, dim3)
-
-        compiled_model = torch.compile(model, backend="eager")
-        result = compiled_model(x, dim1, dim2, dim3)
-
-        torch.testing.assert_close(result, expected)
-
-    def test_torch_reshape_with_tensor_shape_params(self):
-        # Test torch.reshape() function with tensor shape parameters
-        def test_fn(x, shape_params):
-            return torch.reshape(x, shape_params)
-
-        x = torch.randn(24)
-        shape_params = [
-            torch.tensor(2, dtype=torch.int32),
-            torch.tensor(3, dtype=torch.int32),
-            torch.tensor(4, dtype=torch.int32),
-        ]
-
-        expected = test_fn(x, shape_params)
-
-        compiled_fn = torch.compile(test_fn, backend="eager")
-        result = compiled_fn(x, shape_params)
-
-        torch.testing.assert_close(result, expected)
-
 
 if __name__ == "__main__":
     from torch._dynamo.test_case import run_tests

torch/_dynamo/variables/torch.py

@@ -746,37 +746,6 @@ def handle_full(self, tx, size, fill_value, **kwargs):
                     tx, [size, result], kwargs
                 )
 
-        @register(torch.ops.aten.view.default)
-        def handle_view_default(self, tx, tensor, shape):
-            from ..utils import proxy_args_kwargs
-            from .builder import wrap_fx_proxy
-            from .lists import ListVariable
-
-            def convert_tensor_to_scalar(item):
-                if isinstance(item, TensorVariable):
-                    return TorchInGraphFunctionVariable(
-                        torch.ops.aten._local_scalar_dense
-                    ).call_function(tx, [item], {})
-                return item
-
-            if isinstance(shape, ListVariable):
-                converted_items = [
-                    convert_tensor_to_scalar(item) for item in shape.items
-                ]
-                shape = ListVariable(converted_items)
-            elif isinstance(shape, TensorVariable):
-                shape = convert_tensor_to_scalar(shape)
-
-            # Create proxy directly to avoid recursion
-            return wrap_fx_proxy(
-                tx,
-                tx.output.create_proxy(
-                    "call_function",
-                    torch.ops.aten.view.default,
-                    *proxy_args_kwargs([tensor, shape], {}),
-                ),
-            )
-
         @register(torch._foreach_lerp_)
         def handle_inplace_foreach_lerp_scalar(
             _, tx: "InstructionTranslator", *args, **kwargs