[Mlir-commits] [mlir] [mlir][linalg] Implement `LinalgGroupedConvolutionOpInterface` to unify grouped convs (PR #94796)

[llvmlistbot at llvm.org]

ivangarcia44 wrote:

> > > Hi @ivangarcia44, sure I'd be willing to bring it back to life if there's some need for it. I'd definitely hate for another conv op to be added and is why I started this in the first place. I just lost momentum and time to finish it. It might help to get some initial comments on what I have. I'll try to find some time this weekend to update it as it's pretty old by now
> > 
> > 
> > Thank you for taking this back. This could help in networks like the one in this paper for MRI's/CT scans: "A 3D Grouped Convolutional Network Fused with Conditional Random Field and Its Application in Image Multi‑target Fine Segmentation". Please let me know if any help is needed. Thanks
> 
> sorry @ivangarcia44 . I guess no one gives a shit enough about this to even respond. I'm kinda sick of dealing of dealing with the linalg community so I have no plans on continuing to push on this.

I understand and felt the same multiple times in the linalg area. The delayed or lack of response in this area is a blocker. I think supporting 3D grouped convolution is an important use case. I am trying to get help on this.

https://github.com/llvm/llvm-project/pull/94796