[PATCH] D30651: [InlineCost, -Oz] Don't take into account the penalty of a fast call of frequently used functions
Evgeny Astigeevich via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 7 14:41:28 PST 2017
eastig added a comment.
LNT data I've got at the moment:
- x86
FrequentFuncThreshold: 2+ callers
| | Code size change | Exe time |
| MultiSource/Applications/SPASS/SPASS | -6.21% | +0..+1% |
| MultiSource/Benchmarks/FreeBench/distray/distray | -2.79% | 0% |
| SingleSource/Benchmarks/Misc/perlin | -2.32% | 0% |
| SingleSource/Benchmarks/BenchmarkGame/Large/fasta | -1.75% | 0% |
| MultiSource/Applications/sqlite3/sqlite3 | -1.09% | +0..1% |
| SingleSource/Benchmarks/Misc-C++/Large/sphereflake | +1.54% | 0% |
| MultiSource/Benchmarks/Olden/tsp/tsp | +1.16% | 0% |
FrequentFuncThreshold: 3+ callers
| | Code size change | Exe time |
| MultiSource/Applications/SPASS/SPASS | -6.29% | +0..+1% |
| MultiSource/Benchmarks/FreeBench/distray/distray | -2.79% | 0% |
| MultiSource/Applications/sqlite3/sqlite3 | -1.27% | +0..1% |
- AArch64
FrequentFuncThreshold: 2+ callers
| | Code size change | Exe time |
| MultiSource/Benchmarks/Olden/perimeter/perimeter | -4.0% | +5% |
| MultiSource/Applications/SPASS/SPASS | -3.97% | +4% |
| MultiSource/Benchmarks/FreeBench/distray/distray | -1.17% | 0% |
| MultiSource/Applications/sqlite3/sqlite3 | -1.05% | +2% |
FrequentFuncThreshold: 3+ callers
| | Code size change | Exe time |
| MultiSource/Applications/SPASS/SPASS | -4.04% | +4% |
| MultiSource/Benchmarks/Olden/perimeter/perimeter | -4.0% | 4% |
| MultiSource/Benchmarks/FreeBench/distray/distray | -1.17% | 0% |
| MultiSource/Applications/sqlite3/sqlite3 | -1.11% | +3% |
https://reviews.llvm.org/D30651
More information about the llvm-commits
mailing list