[PATCH] D42717: [JumpThreading] sync DT for LVI analysis (PR 36133)

Tue Feb 6 15:49:23 PST 2018

brzycki added a comment.

I'm seeing about 3% run variation on sqlite with the upstream compiler:

  brzycki at cc01 /work/brzycki/test-suite $ ~/git/me/llvm/litdiff build.upstream/results.json build.upstream2/results.json compile_time
  #  left column: build.upstream/results.json
  # right column: build.upstream2/results.json
  #  metric name: compile_time
      336.3600 <- 347.4120     [  3.29%]  CTMark/sqlite3/sqlite3.test
      493.8600 -> 487.0560     [  1.40%]  CTMark/tramp3d-v4/tramp3d-v4.test
      373.0960 -> 368.5280     [  1.24%]  CTMark/consumer-typeset/consumer-typeset.test
      387.2960 <- 389.8320     [  0.65%]  CTMark/kimwitu++/kc.test
      459.5960 <- 462.5440     [  0.64%]  CTMark/SPASS/SPASS.test
      924.1160 -> 921.2680     [  0.31%]  CTMark/Bullet/bullet.test
      497.1560 <- 498.5240     [  0.28%]  CTMark/ClamAV/clamscan.test
      589.8000 -> 588.3840     [  0.24%]  CTMark/lencod/lencod.test
     1265.6800 <- 1267.2280    [  0.12%]  CTMark/7zip/7zip-benchmark.test
      256.1360 -> 256.1000     [  0.01%]  CTMark/mafft/pairlocalalign.test

That still puts the patch around 10% slower.

Repository:
  rL LLVM

https://reviews.llvm.org/D42717