When looking at TierUpInjectionPhase recently, it often seemed that the outermost loop would not be an OSR entry candidate. We should see why this is happening. This could probably help in some benchmarks.
Fixing the systematic issue looks like a speedup on a kraken subtest: og change ai-astar 85.692+-0.627 85.500+-0.840 audio-beat-detection 39.132+-2.003 38.364+-0.357 might be 1.0200x faster audio-dft 95.293+-0.681 ? 96.690+-1.584 ? might be 1.0147x slower audio-fft 28.706+-0.107 ? 28.733+-0.360 ? audio-oscillator 44.681+-0.460 ? 44.772+-0.427 ? imaging-darkroom 58.974+-0.416 58.969+-0.539 imaging-desaturate 45.945+-1.371 ? 46.142+-1.571 ? imaging-gaussian-blur 61.139+-1.573 ^ 55.691+-1.727 ^ definitely 1.0978x faster json-parse-financial 29.528+-0.587 ^ 28.671+-0.200 ^ definitely 1.0299x faster json-stringify-tinderbox 19.068+-0.402 18.986+-0.365 stanford-crypto-aes 44.608+-0.564 ? 45.408+-0.493 ? might be 1.0179x slower stanford-crypto-ccm 40.929+-1.719 40.911+-1.484 stanford-crypto-pbkdf2 60.492+-1.020 59.223+-1.505 might be 1.0214x faster stanford-crypto-sha256-iterative 18.558+-0.220 ? 18.672+-0.295 ? <arithmetic> 48.053+-0.265 47.624+-0.200 might be 1.0090x faster
Created attachment 342138 [details] WIP
Created attachment 342191 [details] patch
Comment on attachment 342191 [details] patch Clearing flags on attachment: 342191 Committed r232602: <https://trac.webkit.org/changeset/232602>
All reviewed patches have been landed. Closing bug.
<rdar://problem/40912318>