[PATCH] D53245: Teach the DominatorTree fallback to recalculation when applying updates to speedup JT (PR37929)
Chijun Sima via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 16 06:55:04 PDT 2018
NutshellySima added a comment.
@kuhar , here is the statistics for `k*α>n`:
Let `y=time used by DT.applyUpdates/Times used by all DomTree Updating operations`.
raw statistics:
clang: https://paste.ubuntu.com/p/yC8sSQqbsx/
SQLite: https://paste.ubuntu.com/p/cTpqDthsNC/
oggenc: https://paste.ubuntu.com/p/4RJkCxjhkY/
| α | y/clang | y/sqlite | y/oggenc |
| 0 | 33.20% | 24.90% | 28.40% |
| 5 | 22.10% | 19.60% | 20.80% |
| 10 | 20.90% | 20.00% | 20.30% |
| 20 | 20.70% | 21.70% | 22.50% |
| 30 | 13.10% | 23.10% | 23.70% |
| 40 | 13.60% | 24.10% | 24.90% |
| 55 | 14.20% | 26.00% | 28.00% |
| 65 | 13.80% | 27% | 28.70% |
| 70 | 14.10% | 27.30% | 29.30% |
| 75 | 14.00% | 28.80% | 30.10% |
| 80 | 14.20% | 28.10% | 30.20% |
| 85 | 14.20% | 29.50% | 30.80% |
| 100 | 14.70% | 30.60% | 31.00% |
| 300 | 17.60% | 42.90% | 34.60% |
| 1000 | 56.40% | 43.00% | 34.50% |
|
I think maybe "40" is a better choice?
https://reviews.llvm.org/D53245
More information about the llvm-commits
mailing list