[llvm] [polly] [ConstraintElimination] Use SCEV ranges information for Loop counter (PR #91457)

[via llvm-commits]

v01dXYZ wrote:

I think you're right about not using the range information in this pass, as this pass seems to simplify using mainly a forward approach which traverses the CFG while keeping track of the branch conditions associated to the dominating edges (among other things). 

Supporting this case would be against this simple forward approach.

There are other passes that do a range analysis but do not simplify icmp.

* SCCP: lattice -> set NUW/NSW for add
* Correlated Propagation: Lazy Value Info / Value Lattice -> set NUW/NSW for add
* Ind Var: SCEV





https://github.com/llvm/llvm-project/pull/91457