[all-commits] [llvm/llvm-project] 93b74f: [ctxprof] Scale up everything under a root by its ...

[Mircea Trofin via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 93b74f7178e510a12f0eae700a25f7d3eb226265
      https://github.com/llvm/llvm-project/commit/93b74f7178e510a12f0eae700a25f7d3eb226265
  Author: Mircea Trofin <mtrofin at google.com>
  Date:   2025-04-21 (Mon, 21 Apr 2025)

  Changed paths:
    M llvm/lib/Analysis/CtxProfAnalysis.cpp
    M llvm/test/Analysis/CtxProfAnalysis/flatten-and-annotate.ll
    M llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll
    M llvm/test/Analysis/CtxProfAnalysis/flatten-insert-icp-mdprof.ll
    M llvm/test/Analysis/CtxProfAnalysis/flatten-zero-path.ll
    M llvm/test/Analysis/CtxProfAnalysis/full-cycle.ll
    M llvm/test/Analysis/CtxProfAnalysis/inline.ll
    M llvm/test/Analysis/CtxProfAnalysis/load.ll

  Log Message:
  -----------
  [ctxprof] Scale up everything under a root by its `TotalRootEntryCount` (#136015)

`TotalRootEntryCount` captures how many times that root was entered - regardless if a profile was also collected or not (profile collection for a given root happens on only one thread at a time).

We don't do this in compiler_rt because the goal there is to flush out the data as fast as possible, so traversing and multiplying vectors is punted to the profile user.

We really just need to do this when flattening the profile so that the values across roots and flat profiles match. We could do it earlier, too - like when loading the profile - but it seems beneficial (at least for debugging) to keep the counter values the same as the loaded ones. We can revisit this later.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications