Cast-cast elimination

[gramalingam]

Enable the cast-cast simplification to a single cast in a couple of cases where it is valid. This shows up in examples like SmolLM (FP16) and is needed for fusion-pattern to work.

Also: add display of replaced and replacing nodes in fusion in verbose mode.

[codecov]

❌ 24 Tests Failed:

Tests completed	Failed	Passed	Skipped
16406	24	16382	2339

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

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0313_test_depthtospace_example

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\3.11.9\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_depthtospace_example'

The above exception was the direct cause of the following exception:
.nox\test_ort_nightly\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_depthtospace_example' (e=No module named 'tests.onnx_backend_test_code.test_depthtospace_example') (file: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_depthtospace_example.py', absolute path: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_depthtospace_example.py', current folder: C:\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_depthtospace_example(x: FLOAT[1,8,2,3]) -> (FLOAT[1,2,4,6]):
E       y = opset13.DepthToSpace(x, blocksize=2, mode='DCR')
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0598_test_lrn

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\3.11.9\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_lrn'

The above exception was the direct cause of the following exception:
.nox\test_ort_nightly\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_lrn' (e=No module named 'tests.onnx_backend_test_code.test_lrn') (file: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_lrn.py', absolute path: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_lrn.py', current folder: C:\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_lrn(x: FLOAT[5,5,5,5]) -> (FLOAT[5,5,5,5]):
E       y = opset13.LRN(x, alpha=0.00019999999494757503, beta=0.5, bias=2.0, size=3)
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_1098_test_shape_clip_end

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\3.11.9\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_shape_clip_end'

The above exception was the direct cause of the following exception:
.nox\test_ort_nightly\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_shape_clip_end' (e=No module named 'tests.onnx_backend_test_code.test_shape_clip_end') (file: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_shape_clip_end.py', absolute path: 'C:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_shape_clip_end.py', current folder: C:\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, INT64
E   from onnxscript.onnx_opset import opset21
E   
E   @script()
E   def bck_test_shape_clip_end(x: FLOAT[3,4,5]) -> (INT64[3]):
E       y = opset21.Shape(x, end=10)
E       return y

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

[justinchuby]

Maybe when type2==type3 or when both .is_floating_point() and type2.itemsize > type3.itemsize?

[gramalingam]

The first one is already covered by redundant cast elimination. The second one is what I think too (at least for float types; something similar for int types too may hold), but am keeping it simple here. (Do these methods is_floating_point / itemsize exist in ir.DataType?)

[justinchuby]

Yes the methods exist. I think it’s pretty simple to implement

[justinchuby]

Suggested change

	type2 = to_ignored.as_int()
	type2 = it.DataType(to_ignored.as_int())

[justinchuby]

Suggested change

	type3 = to.as_int()
	type3 = ir.DataType(to.as_int())