Allow symints to be created for arguments

exir/pass_base.py

@@ -1,6 +1,6 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
@@ -191,6 +191,11 @@ def create_arg(self, a: Argument) -> torch.fx.Node:
                 if not hasattr(a, "constant") or a.constant is None:
                     raise ExportPassBaseError(f"Cannot add {a} to graph.")
                 a = a.constant
+            elif isinstance(a, torch.SymInt):
+                if a.node.constant is not None:
+                    return a.node.constant
+                else:
+                    return a
             node = super().create_arg(a)
             if (
                 isinstance(a, torch.Tensor)

exir/tests/test_dynamic_shape_propagation.py

@@ -1,16 +1,26 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
+# Copyright 2026 Arm Limited and/or its affiliates.
+#
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
 # pyre-unsafe
 
 from unittest import TestCase
 
+import torch
+
 from executorch import exir
 from executorch.exir import to_edge
-from executorch.exir.passes import DebugPass, HintBasedSymShapeEvalPass, SpecPropPass
+from executorch.exir.passes import (
+    DebugPass,
+    ExportPass,
+    HintBasedSymShapeEvalPass,
+    SpecPropPass,
+)
+from executorch.exir.passes.sym_shape_eval_pass import ConstraintBasedSymShapeEvalPass
 from executorch.exir.tests.models import Repeat, TensorItem
 from torch.export import export
 
@@ -67,3 +77,98 @@ def test_unbacked_symint(self):
         self.assertEqual(
             speclist[0].shape, [100, 100]
         )  # upper bound of TensorItem model
+
+
+class TestSymIntViewArgs(TestCase):
+    class Conv1dToConv2d(torch.nn.Module):
+        def __init__(self) -> None:
+            super().__init__()
+
+        def forward(self, input: torch.Tensor) -> torch.Tensor:
+            # Use view to make sure edge view handle symint shapes correctly.
+            # input = input.view(input.size(0), input.size(1), input.size(2), 1)  # (N, C, H, W)
+            # weight = torch.randn(1, 16, 3, 1)  # (out_channels, in_channels, kH, kW)
+            # return torch.nn.functional.conv2d(input, weight)
+
+            return torch.nn.functional.conv1d(
+                input, torch.randn(1, 16, 3)
+            )  # (out_channels, in_channels, kW)
+
+        def get_random_inputs(self) -> tuple[torch.Tensor]:
+            return (torch.randn(1, 16, 50),)  # (batch_size, channels, width)
+
+        def get_dynamic_shape(self) -> tuple[dict[int, torch.export.Dim]]:
+            dim = torch.export.Dim("width", min=10, max=100)
+            return ({2: dim},)
+
+    def test_symint_viewargs(self):
+        eager_model = TestSymIntViewArgs.Conv1dToConv2d()
+        inputs = eager_model.get_random_inputs()
+
+        class TestViewCopyPass(ExportPass):
+            def call_operator(self, op, args, kwargs, meta):
+                from executorch.exir.dialects._ops import ops as exir_ops
+
+                if op != exir_ops.edge.aten.convolution.default:
+                    return super().call_operator(op, args, kwargs, meta)
+
+                x = args[0]
+                x = super().call_operator(
+                    exir_ops.edge.aten.view_copy.default,
+                    (x, list(x.data.shape) + [1]),
+                    {},
+                    meta,
+                )
+
+                w = args[1]
+                w = super().call_operator(
+                    exir_ops.edge.aten.view_copy.default,
+                    (w, list(w.data.shape) + [1]),
+                    {},
+                    meta,
+                )
+
+                new_args = (
+                    x,
+                    w,
+                    args[2],
+                    args[3] + [1],  # stride
+                    args[4] + [0],  # padding
+                    args[5] + [1],  # dilation
+                    args[6],
+                    args[7] + [0],
+                    args[8],
+                )
+                x = super().call_operator(
+                    exir_ops.edge.aten.convolution.default, new_args, kwargs, meta
+                )
+                x = super().call_operator(
+                    exir_ops.edge.aten.view_copy.default,
+                    (x, list(x.data.shape)[:-1]),
+                    {},
+                    meta,
+                )
+
+                return x
+
+        prog = to_edge(
+            export(
+                eager_model,
+                inputs,
+                dynamic_shapes=eager_model.get_dynamic_shape(),
+                strict=True,
+            ),
+            compile_config=exir.EdgeCompileConfig(_check_ir_validity=False),
+        )
+        new_prog = prog.transform(
+            [SpecPropPass(), ConstraintBasedSymShapeEvalPass(), TestViewCopyPass()]
+        )
+        gm = new_prog.exported_program().graph_module
+        DebugPass(show_spec=True)(gm)
+        *_, return_node = gm.graph.nodes
+        speclist = return_node.meta["spec"]
+
+        self.assertEqual(len(speclist), 1)
+        out_spec = speclist[0]
+        self.assertTrue(out_spec.is_upper_bound_tensor)
+        self.assertEqual(out_spec.shape, [1, 1, 98])