[preprocess] Pad is not folded in Conv when opset_import is > 20

[Johansmm]

Describe the issue

Preprocess does not fold 'Pad' in 'Conv' when model has an opset version higher than 20.

To reproduce

def test_pad(op_version=22):
    path = "opt_model.onnx"
    model = onnx.parser.parse_model(
        """
        <ir_version: 9, opset_import: ["": {op_version}]>
        agraph (float[1, 3, 10, 10] X) => (float[1, ?, ?, ?] Y){{
            pad_values = Constant<value_ints=[0,0,1,1,0,0,1,1]>()
            pad = Pad(X, pad_values)
            Y = Conv(pad, W)
        }}
        """.format(
            op_version=op_version
        )
    )
    w = onnx.numpy_helper.from_array(np.ones((3, 3, 3, 3), "float32"), "W")
    model.graph.initializer.extend([w])

    ort_quant.quant_pre_process(model, path, skip_symbolic_shape=True)

    opt_model = onnx.load(path)
    current_nodes = [node.op_type for node in opt_model.graph.node]
    print([node.op_type for node in opt_model.graph.node])
    assert current_nodes == ["Conv"], f"Wrong. Current nodes: {current_nodes}."

test_pad(op_version=18) # This pass
test_pad(op_version=21) # This raises an exception

Urgency

No response

Platform

Linux

OS Version

Ubuntu 22.04

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.21.0

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.