`make onnx_mlir_supported_ops_cpu` sets op HardSwish minimum opset to none

[christopherlmunoz]

HardSwish operation is supported in ONNX opset 14. After running make onnx_mlir_supported_ops_cpu it sets the minimum opset (inclusive) value to none in docs/SupportedONNXOps-cpu.md.

There are no tests for hardswish in https://github.com/onnx/onnx-mlir/blob/main/test/backend/inference_backend.py.

[AlexandreEichenberger]

@christopherlmunoz please close this issue if the PR #3120 fully addressed this issue, thanks

[christopherlmunoz]

@christopherlmunoz please close this issue if the PR #3120 fully addressed this issue, thanks

Hi Alex, PR #3120 was not created to fix this issue; however, while I was updating the docs for that PR, I noticed that after I ran the command make onnx_mlir_supported_ops_cpu, it would set the HardSwish operation to none.

#3107 Changed the support from none to min opset 14, but I think this was done through a manual change of the file instead of updating https://github.com/onnx/onnx-mlir/blob/main/test/backend/inference_backend.py.

[christopherlmunoz]

If you think this assessment is correct, I can create a section for it in infrence_backend.py.

[AlexandreEichenberger]

I definitely saw Hardswish tests in the third_party/onnx subdirectories

        "|test_hardswish_cpu"
        "|test_hardswish_expanded_cpu"

So maybe we have not updated our internal file to represent these. It is also ok to add entries to represent an op in our inference_backend.py file even if there are no test for it.