[pass] Update DCE passes

[justinchuby]

• Remove the remove_initialized_inputs option in dce because the contract of the pass it that it does not modify model signature. Fixed bugs where initializers are removed. Instead, users can use Add add/remove initializers passes #2253 to remove the initialized inputs first.
• Additionally updated RemoveUnusedOpsetsPass to always retain the default opset.

[codecov]

❌ 8 Tests Failed:

Tests completed	Failed	Passed	Skipped
15373	8	15365	2338

View the top 3 failed test(s) by shortest run time

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0211_test_bitshift_right_uint8

Stack Traces | 0.003s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_bitshift_right_uint8'

The above exception was the direct cause of the following exception:
.nox\test_onnx_weekly\Lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_bitshift_right_uint8' (e=No module named 'tests.onnx_backend_test_code.test_bitshift_right_uint8') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_bitshift_right_uint8.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_bitshift_right_uint8.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import UINT8
E   from onnxscript.onnx_opset import opset11
E   
E   @script()
E   def bck_test_bitshift_right_uint8(x: UINT8[3], y: UINT8[3]) -> (UINT8[3]):
E       z = opset11.BitShift(x, y, direction='RIGHT')
E       return z

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_1181_test_sqrt

Stack Traces | 0.003s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_sqrt'

The above exception was the direct cause of the following exception:
.nox\test\lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_sqrt' (e=No module named 'tests.onnx_backend_test_code.test_sqrt') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_sqrt.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_sqrt.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import FLOAT
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_sqrt(x: FLOAT[3,4,5]) -> (FLOAT[3,4,5]):
E       y = opset13.Sqrt(x)
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0745_test_optional_has_element_empty_no_input_name_optional_input

Stack Traces | 0.004s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_optional_has_element_empty_no_input_name_optional_input'

The above exception was the direct cause of the following exception:
.nox\test\lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_optional_has_element_empty_no_input_name_optional_input' (e=No module named 'tests.onnx_backend_test_code.test_optional_has_element_empty_no_input_name_optional_input') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_optional_has_element_empty_no_input_name_optional_input.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_optional_has_element_empty_no_input_name_optional_input.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import BOOL
E   from onnxscript.onnx_opset import opset18
E   
E   @script()
E   def bck_test_optional_has_element_empty_no_input_name_optional_input() -> (BOOL):
E       output = opset18.OptionalHasElement(None)
E       return output

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[Copilot]

Pull Request Overview

This pull request refactors the dead code elimination passes to prevent unintended modifications to the model signature by removing the now-irrelevant remove_initialized_inputs option and ensuring the default opset is always retained.

• Removed the removal of initialized inputs and associated constructor parameters in RemoveUnusedNodesPass.
• Modified initializer removal logic to protect inputs in the model signature.
• Updated domain handling in _process_graph_like to always retain the default opset.

[shubhambhokare1]

Might need to update tests such as

onnxscript/onnxscript/ir/passes/common/unused_removal_test.py

Lines 105 to 107 in b63ba43

	model = self.remove_unused_nodes(
	model, remove_initialized_inputs=remove_initialized_inputs
	)

and/or the function called in these tests

onnxscript/onnxscript/optimizer/__init__.py

Lines 115 to 127 in b63ba43

	def remove_unused_nodes(
	model: ir.Model | onnx.ModelProto, remove_initialized_inputs: bool = False
	) -> None:
	"""Removes unused nodes from a model inplace."""
	if isinstance(model, ir.Model):
	onnxscript.ir.passes.common.unused_removal.RemoveUnusedNodesPass(
	remove_initialized_inputs=remove_initialized_inputs
	)(model)
	else:
	model_ir = ir.serde.deserialize_model(model)
	model_ir = onnxscript.ir.passes.common.unused_removal.RemoveUnusedNodesPass(
	remove_initialized_inputs=remove_initialized_inputs
	)(model_ir).model

to reflect the removal of the attribute

[justinchuby]

@shubhambhokare1 thanks! Done.

[xadupre]

There should a test with a model including subgraphs.

[justinchuby]

Thanks - will update

[justinchuby]

#2264

[titaiwangms]

From the discussion with Kunal on phi4_mm/builder.py, It seems from the user side, they don't really benefit from delicately/cautiously split passes. The end of the day, they just call the higher level API. So that means these delicately/cautiously split passes only serve as debugging usage for ourselves. From debugging point of view, would it be too trivial to separate unused initializers from DCE?

[justinchuby]

I don't think the passes are designed for debugging purposes only. There can be users who want a high level api which we can provide, but we will also support usages that are more fine grained & users can design their own transformation pipelines. Each pass should be modularized: it should have a clear contract and a self contained behavior.