[MLIR][TORCH] Add OnnxToTorch support for BlackmanWindow function

[vinayakdsci]

Implements OnnxToTorch lowering for the BlackmanWindow Function.

[vinayakdsci]

cc @vivekkhandelwal1 @renxida @zjgarvey.

[vinayakdsci]

@vivekkhandelwal1 gentle ping.

[renxida]

Also needs an e2e test.

This can either be done in torch-mlir/projects/pt1 or Shark-TestSuite

SharkTestSuite is what we're moving towards, and in the past, for ONNX models, it has given me less trouble.

All you need to do is:

1. Make a new folder named after your op in
   https://github.com/nod-ai/SHARK-TestSuite/tree/main/e2eshark/onnx/operators
2. Add a file named 'model.py' inside that folder. Reference https://github.com/nod-ai/SHARK-TestSuite/blob/main/e2eshark/onnx/operators/If/model.py for an example i just pushed today
3. run it with

PYTHONPATH="/home/azureuser/torch-mlir/build/tools/torch-mlir/python_packages/torch_mlir:$PYTHONPATH" python ./run.py -c ~/torch-mlir/build -i ~/iree-build --tests onnx/operators/If --verbose --cachedir /tmp/ --verbose --torchtolinalg

Note that you have to
cd Shark-TestSuite/e2eshark
and modify the string after --tests to match the directory of the oprerator you're testing. Here, i'm testing onnx.If

[vinayakdsci]

Note that you have to cd Shark-TestSuite/e2eshark and modify the string after --tests to match the directory of the oprerator you're testing. Here, i'm testing onnx.If

I have written the test, but would it require to be pushed to the SHARK-TestSuite repository?

[vivekkhandelwal1]

Hi @vinayakdsci, I think you have adopted the PyTorch Blackman Window instead of Onnx Blackman Window implementation for this lowering. Although they both might be same but I think it's better for an onnx op to be lowered as per their implementation, and it's more readable as well to compare the algo.

[vivekkhandelwal1]

Apart from a small nit, everything else looks fine.

[vivekkhandelwal1]

@vinayakdsci, you need an approval from @renxida too to merge this.