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

[Antonio Frighetto via llvm-commits]

antoniofrighetto wrote:

> There are improvements from analyzing sdiv instructions in IR

It may be helpful to elaborate on which ones. As already suggested, it looks like part of the existing sdiv reasoning in the codebase has seemingly worked out well directly working on the IR. How would having sdiv as first-class SCEV expressions help ameliorate these (and other) use cases? At least from the main motivation, this appears specifically tailored for computing pointer strides; I wonder if a more targeted approach (e.g., directly extending the stride computation to handle signed division) might well work better.

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