[llvm] [SCEV] Consider IVMayOverflow to use cheap BE-count formula. (PR #218694)

[Nikita Popov via llvm-commits]

================

----------------
nikic wrote:

This whole comment looks outdated. We're not actually using End, we're using RHS. So the argument about the RHS <= Start doesn't match what we actually do, or our preconditions (which has Start-Stride < RHS only, which is not quite the same).

I think these comments need an update to reflect reality, it's hard to verify whether the logic is correct otherwise.

https://github.com/llvm/llvm-project/pull/218694