[PATCH] D54730: [DomTree] Fix order of domtree updates in MergeBlockIntoPredecessor.

[Brian Rzycki via Phabricator via llvm-commits]

brzycki added a comment.

@efriedma I ran CTMark on the patch with the following patch as tip (both compilers are Release):

  commit 9b6b09070e777106e6e91c15406e197c7d18e1fb (origin/master, origin/HEAD)
  Author: Ilya Biryukov <ibiryukov at google.com>
  Date:   Mon Nov 26 15:38:01 2018 +0000
  
      [clangd] Add type boosting in code completion
  
      Reviewers: sammccall, ioeric
  
      Reviewed By: sammccall
  
      Subscribers: MaskRay, jkorous, arphaman, kadircet, cfe-commits
  
      Differential Revision: https://reviews.llvm.org/D52276

The results show it's mostly good, except for tramp3d-v4 which regresses about 9%:

  #  left column: D54730/results.json
  # right column: tip/results.json
  #  metric name: compile_time
       41.3040 -> 37.9320      [  8.89%]  CTMark/tramp3d-v4/tramp3d-v4.test
      125.0240 <- 132.6200     [  6.08%]  CTMark/7zip/7zip-benchmark.test
       23.3760 -> 22.8840      [  2.15%]  CTMark/mafft/pairlocalalign.test
       43.6280 <- 44.3680      [  1.70%]  CTMark/SPASS/SPASS.test
       52.6560 <- 53.0440      [  0.74%]  CTMark/lencod/lencod.test
       31.9480 <- 32.1480      [  0.63%]  CTMark/consumer-typeset/consumer-typeset.test
       35.1400 <- 35.3160      [  0.50%]  CTMark/kimwitu++/kc.test
       51.0240 <- 51.2560      [  0.45%]  CTMark/ClamAV/clamscan.test
       26.1120 <- 26.2080      [  0.37%]  CTMark/sqlite3/sqlite3.test
       86.8120 <- 86.9920      [  0.21%]  CTMark/Bullet/bullet.test

I'm slightly concerned this patch may regress compile time for others whose code shape resembles tramp3d.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D54730/new/

https://reviews.llvm.org/D54730