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

[Maksim Panchenko via llvm-commits]

maksfb wrote:

Thanks for checking.

30 MiB is hard to explain, unless it's a number gathered across multiple processes, e.g. in a case of a parallel build. If it's the number for a single process, it must be some side effect. I find it hard to believe it's a direct impact from a code layout optimization.

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