[llvm] [LoopUnroll] Use branch probability in multi-exit loop unrolling (PR #164799)

[Marek Sedláček via llvm-commits]

mark-sed wrote:

> This is what troubles me: it sounds like you found a problem.

Well it arises only when BPI is introduced into runtime unrolling. I have tried reproducing this bug with upstream, but I was unable to do so (I tried using unittest and invoking BPI on a case that failed downstream). Only when I introduce BPI into unrolling and from my understanding of such analysis when introducing it I should also make sure the state of it is correct, which I suppose would be the more difficult task.


https://github.com/llvm/llvm-project/pull/164799