[all-commits] [llvm/llvm-project] 6a33b6: [SampleProfile] Support MD5-based ProfileSymbolLis...

[Kazu Hirata via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 6a33b69d8bae737e5e1d60c7b7776a0b1f885bfa
      https://github.com/llvm/llvm-project/commit/6a33b69d8bae737e5e1d60c7b7776a0b1f885bfa
  Author: Kazu Hirata <kazu at google.com>
  Date:   2026-07-17 (Fri, 17 Jul 2026)

  Changed paths:
    M llvm/include/llvm/ProfileData/SampleProf.h
    M llvm/include/llvm/ProfileData/SampleProfReader.h
    M llvm/include/llvm/ProfileData/SampleProfWriter.h
    M llvm/lib/ProfileData/SampleProfReader.cpp
    M llvm/lib/ProfileData/SampleProfWriter.cpp
    M llvm/test/tools/llvm-profdata/profile-symbol-list.test
    M llvm/unittests/ProfileData/SampleProfTest.cpp

  Log Message:
  -----------
  [SampleProfile] Support MD5-based ProfileSymbolList (#210235)

This patch speeds up sample profile loading by introducing an
MD5-based ProfileSymbolList (cold symbol list) in extensible binary
profiles.

The sample profile loader spends about a third of its time on loading
and decoding the symbol names in the Profile Symbol List section, yet
we use them only for membership checking purposes via
ProfileSymbolList::contains.

This patch teaches the writer to emit the section as an array of
64-bit GUIDs in the Eytzinger layout.  The reader checks the
SecFlagMD5 section flag and sets up an EytzingerTableSpan
pointing into the mmap memory.  This achieves both space efficiency
and runtime efficiency.

The new flag, md5-prof-sym-list, is off by default for now.

Profile merging is supported only from strings to an MD5-based
Eytzinger array.

RFC:
https://discourse.llvm.org/t/rfc-faster-sample-profile-loading/90957/7

Assisted-by: Antitygravity



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