[llvm] [polly] [SCEV] Introduce SDiv expressions (PR #216862)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

> > If you're interested, you can take a look at #212816 for the stride speculation on the LAA side, and #216760 for digging out the stride for strided accesses on the VPlan side.
> 
> For the motivation, I think it would be more effective to provide a concrete example of a case where an `SDiv` SCEV expression is needed. Ideally this would be part of the description.
> 
> This would be clearer than providing links to the patches; AFAICT they only show what we can currently handle, with a bit of extra work to dig out the stride.

Included in description, thanks. There are improvements from analyzing sdiv instructions in IR, independently of any stride expressions, and the compile-time increase is in noise anyway.

https://github.com/llvm/llvm-project/pull/216862