[llvm] [VPlan] Plumb scalable register size through narrowInterleaveGroups (PR #167505)

[Florian Hahn via llvm-commits]

https://github.com/fhahn approved this pull request.

LGTM, thanks.

> > I think these changes look sound, but I'm worried about the mysterious unknown failure test case.
> 
> There's a reproducer for the crash in that PR, it's to do with VPVectorPointerRecipes still hanging around. I think that's likely related to the other changes in that PR related to moving the transform itself. This PR doesn't change any of the codegen, it just enables more cases

The currnet changes should be fine and unrelated the the reason for the revert. There already should be test cases for the issue that caused the revert, they were independent of changes for scalable vectors.



https://github.com/llvm/llvm-project/pull/167505