[PATCH] D143631: [LTO] Don't let InstCombine re-sink the vastly more expensive fdiv
Nikita Popov via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 4 08:10:04 PDT 2023
nikic added reviewers: fhahn, dmgreen.
nikic added a comment.
Compile-time: http://llvm-compile-time-tracker.com/compare.php?from=6c5ba7cddfa2210e9ddb12ab6f016b84db9a8b23&to=ef44d3f482b284ee11a5335d8c714f7eceb3dea7&stat=instructions:u
Post-link numbers are:
kc.link 30147M 30326M (+0.59%)
sqlite3.link 40914M 41303M (+0.95%)
consumer-typeset.link 36935M 37405M (+1.27%)
bullet.link 27582M 27914M (+1.20%)
tramp3d-v4.link 137504M 139271M (+1.29%)
pairlocalalign.link 19476M 19686M (+1.08%)
clamscan.link 56907M 57439M (+0.93%)
lencod.link 102321M 102803M (+0.47%)
SPASS.link 51517M 52669M (+2.24%)
7zip-benchmark.link 106003M 105851M (-0.14%)
Despite that, I'm on board with this change.
================
Comment at: llvm/lib/Passes/PassBuilderPipelines.cpp:1259
+ /*AllowSpeculation=*/true),
+ /*UseMemorySSA=*/true, /*UseBlockFrequencyInfo=*/true));
+
----------------
UseBlockFrequencyInfo should stay false.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D143631/new/
https://reviews.llvm.org/D143631
More information about the llvm-commits
mailing list