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

[Florian Hahn via llvm-commits]

fhahn wrote:

Computing the ranges unconditionally requires computing the BTC for all loops, which is quite expensive https://llvm-compile-time-tracker.com/compare.php?from=53d6cfb1d184127a57132657f41104b21e7c0f2b&to=86f4583def5ebd87a1ac08a819fee85f6fc53ae7&stat=instructions:u

Not sure if the benefit warrants the increase; do you have any statistics on larger software how beneficial this is?

Also it would probably be good to split off c41b5b55860b5afaef6d7dca643d77a817bdb1c9 at least to a separate PR


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