[Mlir-commits] [mlir] [MLIR][Linalg] Remove/update failing obsolete OpDSL tests for linalg.matmul. (PR #115319)

[Renato Golin]

rengolin wrote:

> I was afraid of such a thing when the decision was to call the new op added to TableGen the same as the OpDSL name. Previous suggestion was that we change the name of the TableGen version of the op to something else, but that wasnt accepted. I would still suggest doing that to avoid such breakages.

Having two matmuls will create more confusion than this. We need to move away from OpDSL and the first op is the one that was bound to break things. If we leave the ops in OpDSL, then the Python code will use one set and the C++ code another, and that's way way worse than any temporary code churn.

It seems obvious to me that Python should be able to call other ODS operations. If so, we need to tie the same mechanism to `matmul`, and then the follow up operations. This seems like a mechanical solution would work here.

I don't know the Python code well enough, @makslevental can you help us here?



https://github.com/llvm/llvm-project/pull/115319