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

[Marek Sedláček via llvm-commits]

mark-sed wrote:

> OK, so a repro could be applying a PR and then running a specific lit test, correct?

Yes.
But is that really a bug in llvm or rather my implementation of BPI usage in runtime unrolling? Because I just understand it the way that I was supposed to handled the correct BPI state, or am I mistaken?

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