This is an archive of the discontinued LLVM Phabricator instance.

[CSSPGO] Undoing the concept of dangling pseudo probe
ClosedPublic

Authored by hoy on Jun 17 2021, 11:21 AM.

Details

Summary

As a follow-up to https://reviews.llvm.org/D104129, I'm cleaning up the danling probe related code in both the compiler and llvm-profgen.

I'm seeing a 5% size win for the pseudo_probe section for SPEC2017 and 10% for Ciner. Certain benchmark such as 602.gcc has a 20% size win. No obvious difference seen on build time for SPEC2017 and Cinder.

Diff Detail

Event Timeline

hoy created this revision.Jun 17 2021, 11:21 AM
hoy requested review of this revision.Jun 17 2021, 11:21 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 17 2021, 11:21 AM
hoy updated this revision to Diff 352803.Jun 17 2021, 11:27 AM

Updating D104477: [CSSPGO] Undoing the concept of dangling pseudo probe

Thanks for the clean up.

llvm/test/tools/llvm-profgen/merge-cold-profile.test
25

I thought the changes for removing dangling probes are later, and checksum is computed earlier? What triggered cfg checksum changes?

hoy added inline comments.Jun 18 2021, 1:45 PM
llvm/test/tools/llvm-profgen/merge-cold-profile.test
25

It' probably due to different version of compilers used to build the test binary. Previous compiler didn't include number of direct calls in CFGChecksum.

wenlei accepted this revision.Jun 18 2021, 2:49 PM

lgtm, thanks.

This revision is now accepted and ready to land.Jun 18 2021, 2:49 PM
This revision was landed with ongoing or failed builds.Jun 18 2021, 3:14 PM
This revision was automatically updated to reflect the committed changes.
llvm/test/tools/llvm-profgen/fname-canonicalization.test