Bug 68177

Summary: The DFG non-speculative JIT is no longer used and should be removed.
Product: WebKit Reporter: Filip Pizlo <fpizlo>
Component: JavaScriptCoreAssignee: Nobody <webkit-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: barraclough, fpizlo, ggaren
Priority: P2    
Version: 528+ (Nightly build)   
Hardware: All   
OS: All   
Bug Depends on: 68186    
Bug Blocks:    
Attachments:
Description Flags
the patch ggaren: review+

Filip Pizlo
Reported 2011-09-15 12:01:31 PDT
Tiered compilation is enabled in ToT and produces a pure win. Tiered compilation does not use the non-speculative JIT. The non-speculative JIT is not used by anyone. It should be removed. This also implies removing the spec->non-spec failure code (since we now use OSR instead), the SpeculationCheck code (since we now use OSRExit instead), and any configuration options in DFGNode or elsewhere that pertain to non-speculative. As part of its removal, any useful functionality that it did have that we intend on using in the speculative JIT should be transformed into a helper function on JITCodeGenerator.
Attachments
the patch (99.24 KB, patch)
2011-09-15 15:36 PDT, Filip Pizlo 		ggaren: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Filip Pizlo
Comment 1 2011-09-15 15:36:47 PDT
Created attachment 107557 [details] the patch Will let this sit here for the bots to chew on. I don't know if I killed the references to non-spec in a way that non-Mac platforms will like.
Filip Pizlo
Comment 2 2011-09-15 15:41:48 PDT
Performance tests indicate that this refactoring didn't break anything big. Still running layout tests. Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/septenary/OpenSource/WebKitBuild/Release/jsc "TheNonSpecThe" at /Volumes/Data/pizlo/quartary/OpenSource/WebKitBuild/Release/jsc Collected 12 samples per benchmark/VM, with 4 VM invocations per benchmark. 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 TheNonSpecThe SunSpider: 3d-cube 7.6177+-0.1156 ? 7.7606+-0.1962 ? might be 1.0188x slower 3d-morph 7.5393+-0.2320 ? 7.6219+-0.1830 ? might be 1.0110x slower 3d-raytrace 7.7714+-0.2578 7.5804+-0.2339 might be 1.0252x faster access-binary-trees 2.3403+-0.1317 ? 2.3491+-0.0605 ? access-fannkuch 11.5210+-0.2266 11.4543+-0.2352 access-nbody 4.2669+-0.0950 4.1249+-0.0852 might be 1.0344x faster access-nsieve 2.6167+-0.0547 2.5910+-0.0570 bitops-3bit-bits-in-byte 1.7149+-0.0711 1.6641+-0.0414 might be 1.0305x faster bitops-bits-in-byte 2.7302+-0.0770 ? 2.7872+-0.0475 ? might be 1.0209x slower bitops-bitwise-and 3.6835+-0.1097 3.5397+-0.0759 might be 1.0406x faster bitops-nsieve-bits 5.2826+-0.1478 5.2773+-0.0838 controlflow-recursive 1.9993+-0.0516 1.9780+-0.0702 might be 1.0108x faster crypto-aes 7.0485+-0.4404 6.9846+-0.2883 crypto-md5 2.8006+-0.0772 ? 3.3002+-0.6366 ? might be 1.1784x slower crypto-sha1 2.1984+-0.0615 ? 2.2656+-0.0758 ? might be 1.0305x slower date-format-tofte 10.6388+-0.4510 10.2994+-0.2232 might be 1.0330x faster date-format-xparb 9.0774+-0.6078 8.8395+-0.2180 might be 1.0269x faster math-cordic 6.2369+-0.1227 6.2271+-0.0884 math-partial-sums 7.4144+-0.1098 7.3608+-0.1278 math-spectral-norm 2.6606+-0.1141 2.6382+-0.0876 regexp-dna 11.1858+-0.4029 11.0782+-0.3356 string-base64 5.8766+-0.3127 5.7979+-0.2424 might be 1.0136x faster string-fasta 7.1525+-0.1607 7.0287+-0.2103 might be 1.0176x faster string-tagcloud 12.1184+-0.3133 ? 12.1203+-0.4079 ? string-unpack-code 18.6965+-0.5085 ? 19.0846+-0.5797 ? might be 1.0208x slower string-validate-input 6.7369+-0.2145 6.7058+-0.2513 <arithmetic> 6.4972+-0.0399 6.4792+-0.0494 <geometric> 5.3523+-0.0373 5.3443+-0.0431 <harmonic> 4.3619+-0.0507 4.3610+-0.0394 TipOfTree TheNonSpecThe V8: crypto 83.4792+-0.6764 83.1459+-0.4669 deltablue 245.5458+-1.6084 244.7773+-2.4858 earley-boyer 95.4794+-1.0822 ? 95.8919+-0.8258 ? raytrace 73.3252+-0.5046 73.1796+-0.7989 regexp 107.4445+-0.6514 ? 107.7198+-0.5681 ? richards 221.3328+-1.8472 220.4779+-1.2757 splay 100.5540+-1.1823 99.2863+-1.3507 might be 1.0128x faster <arithmetic> 132.4516+-0.4749 132.0684+-0.7092 <geometric> 119.2531+-0.4279 118.9338+-0.5794 <harmonic> 109.4740+-0.4200 109.1951+-0.5315 TipOfTree TheNonSpecThe Kraken: ai-astar 633.2072+-3.1198 ? 634.0155+-5.6007 ? audio-beat-detection 483.5001+-6.8955 480.5581+-7.4540 audio-dft 431.6491+-6.5128 429.2882+-5.2847 audio-fft 372.8691+-1.7360 369.3022+-1.9860 audio-oscillator 317.6334+-1.8057 ? 317.8665+-1.2358 ? imaging-darkroom 428.3435+-9.1983 419.3040+-1.7876 might be 1.0216x faster imaging-desaturate 207.9498+-0.5476 ? 209.7226+-1.7717 ? imaging-gaussian-blur 1730.3700+-2.7285 ? 1734.7773+-5.0711 ? json-parse-financial 49.0744+-0.4401 ? 49.1756+-0.2212 ? json-stringify-tinderbox 69.9297+-0.5986 ? 70.2837+-1.3321 ? stanford-crypto-aes 145.6043+-0.9866 145.1874+-0.8767 stanford-crypto-ccm 113.0345+-0.9881 112.1558+-0.6815 stanford-crypto-pbkdf2 397.3920+-2.2566 ? 398.4926+-2.9769 ? stanford-crypto-sha256-iterative 150.0585+-1.5669 149.3581+-0.5932 <arithmetic> 395.0440+-0.9378 394.2491+-1.1606 <geometric> 264.2923+-0.5908 263.6392+-0.6825 <harmonic> 176.9877+-0.5693 176.8618+-0.5881 TipOfTree TheNonSpecThe All benchmarks: <arithmetic> 140.9937+-0.2860 140.6899+-0.3304 <geometric> 27.1488+-0.1023 27.0951+-0.1256 <harmonic> 7.7001+-0.0873 7.6982+-0.0679
Filip Pizlo
Comment 3 2011-09-15 15:53:19 PDT
Comment on attachment 107557 [details] the patch Tests pass. It appears that it's good.
Geoffrey Garen
Comment 4 2011-09-15 16:19:03 PDT
Comment on attachment 107557 [details] the patch r=me
Filip Pizlo
Comment 5 2011-09-15 16:23:56 PDT
Landed in r95240.
Note You need to log in before you can comment on or make changes to this bug.