[PATCH] D155103: [FuncSpec] Split the specialization bonus into CodeSize and Latency.

[Alexandros Lamprineas via Phabricator via llvm-commits]

labrinea added a comment.

I run some experiments to measure compilation time. It seems if we return `std::pair<Cost,Cost>` instead of caching CodeSize and Latency to the InstCostVisitor it is slightly faster (perhaps a litle uglier too).
This patch improves geomean of instruction count for llvm-test-suite by -0.016% at O3 <https://reviews.llvm.org/owners/package/3/> and regresses it by +0.07% at LTO. The alternative is -0.036 and +0.064% respectively.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D155103/new/

https://reviews.llvm.org/D155103