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

[Mark Rousskov via llvm-commits]

Mark-Simulacrum wrote:

I split the cache creation out into a dedicated function. I think in at least the librustc_driver.so case, none of the other usages show up as hot, so it's potentially not worth building the cache upfront for them.

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