Alexandre Eichenberger

I believe this is done during the Conversion/KrnlToLLVM in the KrnlGlobal.cpp file

In general, Krnl operations are converted into Affine and also into LLVM phases

Let me know if this is ready for review

@negiyas let me know if it is ready to be reviewed.

@sstamenova any suggestions for @sazzad15-1779 ?

@etiotto let me know if you need the full traces. Or you may simply run `valgrind Debug/bin/TestMatMul2D`.

Hi @sjarus Thanks for starting this discussion. We have thought of ways to interface with MLIR in different ways, right now we do that at the Affine/SCF/Vector... level. An approach...

For ref [tosa_spec.pdf](https://github.com/onnx/onnx-mlir/files/8120502/tosa_spec.pdf)

Lots of information, at some time we will need to go in depth step by steps. On your last question, the ONNX runtime, which is really just creating/deleting/querying of OMTensor...

@sjarus Trying to rephrase your answer to make sure I understood well. For TOSA, you have to ways to implement it via MLIR. 1. One way is to emit for...