[llvm] [SCEV] Directly use wrap flags in getUDivExpr (PR #217133)

[Nikita Popov via llvm-commits]

nikic wrote:

The zext based approach is generally more powerful, though the gap has been somewhat closed due to recent changes to how the nowrap inference works. If we can replace it with a plain flag check without losing any practical benefit, that's great. I'd prefer to change and evaluate these one-by-one though. Otherwise it's hard to track where the diffs come from. I also don't really get what you're doing with inferNoWrapViaConstantRanges here.

https://github.com/llvm/llvm-project/pull/217133