[llvm] [BOLT] Optimize basic block loops to avoid n^2 loop (PR #156243)

[Jakub Beránek via llvm-commits]

Kobzol wrote:

We haven't tried this exact combination, I'll try it, thanks. The issue we had with `--use-old-text` was that it was "unstable", i.e. in one commit the old text segment was reused, and in the next it wasn't, so the binary size was jumping all over the place, so we just disabled `--use-old-text`.

https://github.com/llvm/llvm-project/pull/156243