[Mlir-commits] [mlir] [mlir][linalg] Add a test for inferConvolutionDimsImpl (PR #90057)

[Andrzej Warzyński]

banach-space wrote:

Thanks for taking a look, Quinn! I've rephrased this patch - adding a test for depthwise convs should be sufficient.

> Also I'm not necessarily opposed to a change in classification, I just wrote the current impl to be more in line with the inferContractionDims impl and I haven't worked much with depthwise convs.

Your implementation is perfectly fine. I think that what we are really missing is some high-level reference doc that would classify all the convs in Linalg (and normalize the terminology). I don't have the right overview as I've been mostly working with depthwise convs. But this is on my TODO list and I would also be more than happy to review PRs for this. We'll get there eventually!

https://github.com/llvm/llvm-project/pull/90057