Suraj Sudhir

> Thanks again @sjarus @AlexandreEichenberger for reviewing! I tried to address the comments (`notifyMatchFailure` + removal of macros) in the last commit. Not quite sure if it is possible to...

> Hi, > > There're three main frontends in the deep learning world, TensorFlow, PyTorch and ONNX. And all of them are currently embracing MLIR/LLVM. However, here comes the version...

Thanks for the detailed response @AlexandreEichenberger! >An approach build on TOSA may indeed provide another path to optimized code. Yes it can indeed be an option - when we implemented...

> What support does TOSA use. In the MLIR IREE presentation last week, there were references to a "generic" implementation for tensors and lists. This is a question with multiple...

> Is that correct? If so, which solution are you interested in? 1, 2, or a combination of 1 & 2? It may have implications on what needs to be...

@AlexandreEichenberger yes I am! I recently rebased against existing onnx-mlir top of tree and was able to get the previously implemented proof of concept functionality to compile and run. My...

Yes let me get a sense of the size of the PR after rebasing against the current head, and we can discuss. Perhaps it makes sense to do an initial...

PR is in https://github.com/onnx/onnx-mlir/pull/1379 . I've attempted to keep it to a manageable size focused on build integration.

Great to see this convergence @sstamenova ! TOSA has common code used in the TorchToTosa and ONNXToTOSA efforts and alignment here makes it much more reasonable to proceed with putting...

I've encountered this already @Svoch - it also impacts MobilenetsV3 . Working on a fix internally but am getting some BERT ones out first.