Detune DFG::PhantomInsertionPhase to insert Phantoms maximally late so as to avoid querying bytecode liveness
https://bugs.webkit.org/show_bug.cgi?id=144524
Summary Detune DFG::PhantomInsertionPhase to insert Phantoms maximally late so as to ...
Filip Pizlo
Reported 2015-05-02 10:21:46 PDT
This would be a reduction in compile times but also a reduction in throughput of DFG-compiled code.
Attachments
work in progress (8.27 KB, patch)
2015-05-02 14:54 PDT, Filip Pizlo
no flags
Filip Pizlo
Comment 1 2015-05-02 14:54:35 PDT
Created attachment 252249 [details] work in progress
Filip Pizlo
Comment 2 2015-05-02 15:12:16 PDT
This doesn't appear to be better. Benchmark report for SunSpider on dethklok (MacBookPro9,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/secondary/OpenSource/WebKitBuild/Release/jsc (r183718) "DetunedPhantoms" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc (r183721) Collected 100 samples per benchmark/VM, with 100 VM invocations per benchmark. Emitted a call to gc() between sample measurements. Used 1 benchmark iteration per VM invocation for warm-up. Used the jsc-specific preciseTime() function to get microsecond-level timing. Reporting benchmark execution times with 95% confidence intervals in milliseconds. TipOfTree DetunedPhantoms 3d-cube 5.3233+-0.1250 5.3042+-0.0305 3d-morph 5.8984+-0.0865 ? 6.0443+-0.1219 ? might be 1.0247x slower 3d-raytrace 6.6861+-0.1187 ? 6.7050+-0.1079 ? access-binary-trees 2.3390+-0.0453 ? 2.3773+-0.0768 ? might be 1.0163x slower access-fannkuch 6.1774+-0.0777 ? 6.2872+-0.1330 ? might be 1.0178x slower access-nbody 3.0041+-0.0607 ! 3.1383+-0.0699 ! definitely 1.0447x slower access-nsieve 3.4737+-0.0548 ^ 3.3286+-0.0287 ^ definitely 1.0436x faster bitops-3bit-bits-in-byte 1.6361+-0.0336 ? 1.6433+-0.0260 ? bitops-bits-in-byte 3.3886+-0.0184 ? 3.4225+-0.0354 ? might be 1.0100x slower bitops-bitwise-and 2.2676+-0.0617 2.2475+-0.0544 bitops-nsieve-bits 3.1754+-0.0249 ! 3.5718+-0.0273 ! definitely 1.1248x slower controlflow-recursive 2.3645+-0.0415 ? 2.3792+-0.0828 ? crypto-aes 4.5778+-0.0557 ? 4.6347+-0.0830 ? might be 1.0124x slower crypto-md5 2.9661+-0.0220 ^ 2.7761+-0.0197 ^ definitely 1.0685x faster crypto-sha1 2.7874+-0.0541 ? 2.8665+-0.0833 ? might be 1.0284x slower date-format-tofte 9.9556+-0.3853 9.5920+-0.0543 might be 1.0379x faster date-format-xparb 6.0086+-0.1248 ? 6.0508+-0.1489 ? math-cordic 3.3141+-0.0825 ^ 3.1275+-0.0485 ^ definitely 1.0596x faster math-partial-sums 5.4121+-0.0999 ? 5.5348+-0.1023 ? might be 1.0227x slower math-spectral-norm 2.0055+-0.0378 ! 2.2018+-0.1111 ! definitely 1.0979x slower regexp-dna 7.7860+-0.2137 7.7783+-0.2062 string-base64 4.9737+-0.1368 4.8204+-0.1130 might be 1.0318x faster string-fasta 6.8986+-0.1180 ? 7.0080+-0.1314 ? might be 1.0159x slower string-tagcloud 10.1560+-0.2007 10.0309+-0.2097 might be 1.0125x faster string-unpack-code 20.0669+-0.3698 ? 20.3773+-0.5218 ? might be 1.0155x slower string-validate-input 4.9601+-0.0374 ? 4.9744+-0.0369 ? <arithmetic> 5.2924+-0.0265 ? 5.3163+-0.0297 ? might be 1.0045x slower
Note You need to log in before you can comment on or make changes to this bug.