torch-mlir Add transposed case for at::convolution

Here's transposed support for convolution, both strided and unstrided.

Jun 08 '22 07:06 gpetters94

Can you squash the commits

Jun 13 '22 16:06 dan-garvey

@gpetters94 Could you please resolve the conflicts?

Done.

Aug 06 '22 06:08 gpetters94

Can we approve and merge this PR? @vivekkhandelwal1 @silvasean

Aug 08 '22 04:08 ZihengJiang

It looks like the shape function is not correct (failing PyTorch tests), see https://github.com/pytorch/pytorch/pull/80860

Aug 08 '22 22:08 silvasean

Oh, sorry, it is an "incorrect shape compute mapping function schema name" not incorrect shape function. Yes, we in general wait for these change to land upstream. Is there an urgent reason to merge it now?

Aug 08 '22 22:08 silvasean

We are working on the transposed convolution on the Torch2MHLO lowering side so checked this PR to see whether we can merge it. It's ok to wait the upstream PR to be merged though.

Aug 08 '22 23:08 ZihengJiang

It's holding up U-Net support, but it isn't the only thing blocking that.

Aug 08 '22 23:08 gpetters94

The PR on the PyTorch side is accepted. Can we merge this PR now? @vivekkhandelwal1 @silvasean

Aug 15 '22 16:08 ZihengJiang

@ZihengJiang @silvasean I made a small change in the shape logic (I forgot that two of the dims needed transposing) so I need to make another PR in Pytorch that will hopefully not take long to get upstreamed this time. Otherwise this is passing all the tests.

Aug 16 '22 20:08 gpetters94

The PR on the PyTorch side is accepted. Can we merge this PR now? @vivekkhandelwal1 @silvasean

Sure. Once this PR (https://github.com/pytorch/pytorch/pull/83557) is merged, we can get this patch merged.

Aug 17 '22 06:08 vivekkhandelwal1

@vivekkhandelwal1 @silvasean The upstream shape code is merged, so this one should be good to go.

Aug 22 '22 19:08 gpetters94

@vivekkhandelwal1 Done.

Aug 23 '22 06:08 gpetters94