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

[Brian Rzycki via Phabricator via llvm-commits]

brzycki added a comment.

The variance of tramp3d-v4 is considerable, as much as 18.5%. I ran CTMark on the pre-DDT compiler and here are the results:

  $ cat bar.sh
  #!/bin/bash
  
  for i in 1 2 3 4 5 6; do
      for j in 1 2 3 4 5 6; do
          ~/git/me/llvm/litdiff \
              build.pre_dt_$i/results.json \
              build.pre_dt_$j/results.json \
              compile_time  | grep tramp3d
      done
  done
  brzycki at cc01 /work/brzycki/test-suite $ ./bar.sh
      524.8640 -> 474.9400     [ 10.51%]  CTMark/tramp3d-v4/tramp3d-v4.test
      524.8640 -> 486.9560     [  7.78%]  CTMark/tramp3d-v4/tramp3d-v4.test
      524.8640 -> 482.9680     [  8.67%]  CTMark/tramp3d-v4/tramp3d-v4.test
      524.8640 -> 484.6760     [  8.29%]  CTMark/tramp3d-v4/tramp3d-v4.test
      524.8640 <- 562.8000     [  7.23%]  CTMark/tramp3d-v4/tramp3d-v4.test
      474.9400 <- 524.8640     [ 10.51%]  CTMark/tramp3d-v4/tramp3d-v4.test
      474.9400 <- 486.9560     [  2.53%]  CTMark/tramp3d-v4/tramp3d-v4.test
      474.9400 <- 482.9680     [  1.69%]  CTMark/tramp3d-v4/tramp3d-v4.test
      474.9400 <- 484.6760     [  2.05%]  CTMark/tramp3d-v4/tramp3d-v4.test
      474.9400 <- 562.8000     [ 18.50%]  CTMark/tramp3d-v4/tramp3d-v4.test
      486.9560 <- 524.8640     [  7.78%]  CTMark/tramp3d-v4/tramp3d-v4.test
      486.9560 -> 474.9400     [  2.53%]  CTMark/tramp3d-v4/tramp3d-v4.test
      486.9560 -> 482.9680     [  0.83%]  CTMark/tramp3d-v4/tramp3d-v4.test
      486.9560 -> 484.6760     [  0.47%]  CTMark/tramp3d-v4/tramp3d-v4.test
      486.9560 <- 562.8000     [ 15.58%]  CTMark/tramp3d-v4/tramp3d-v4.test
      482.9680 <- 524.8640     [  8.67%]  CTMark/tramp3d-v4/tramp3d-v4.test
      482.9680 -> 474.9400     [  1.69%]  CTMark/tramp3d-v4/tramp3d-v4.test
      482.9680 <- 486.9560     [  0.83%]  CTMark/tramp3d-v4/tramp3d-v4.test
      482.9680 <- 484.6760     [  0.35%]  CTMark/tramp3d-v4/tramp3d-v4.test
      482.9680 <- 562.8000     [ 16.53%]  CTMark/tramp3d-v4/tramp3d-v4.test
      484.6760 <- 524.8640     [  8.29%]  CTMark/tramp3d-v4/tramp3d-v4.test
      484.6760 -> 474.9400     [  2.05%]  CTMark/tramp3d-v4/tramp3d-v4.test
      484.6760 <- 486.9560     [  0.47%]  CTMark/tramp3d-v4/tramp3d-v4.test
      484.6760 -> 482.9680     [  0.35%]  CTMark/tramp3d-v4/tramp3d-v4.test
      484.6760 <- 562.8000     [ 16.12%]  CTMark/tramp3d-v4/tramp3d-v4.test
      562.8000 -> 524.8640     [  7.23%]  CTMark/tramp3d-v4/tramp3d-v4.test
      562.8000 -> 474.9400     [ 18.50%]  CTMark/tramp3d-v4/tramp3d-v4.test
      562.8000 -> 486.9560     [ 15.58%]  CTMark/tramp3d-v4/tramp3d-v4.test
      562.8000 -> 482.9680     [ 16.53%]  CTMark/tramp3d-v4/tramp3d-v4.test
      562.8000 -> 484.6760     [ 16.12%]  CTMark/tramp3d-v4/tramp3d-v4.test

Best time was `474.94`, worst was `562.80`.  I'll perform the same experiment with proposed patch in this differential.


https://reviews.llvm.org/D42717