| Summary: | fourthTier: DFG should be able to run on a separate thread | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | WebKit | Reporter: | Filip Pizlo <fpizlo> | ||||||||||||||||
| Component: | JavaScriptCore | Assignee: | Filip Pizlo <fpizlo> | ||||||||||||||||
| Status: | RESOLVED FIXED | ||||||||||||||||||
| Severity: | Normal | CC: | barraclough, ggaren, mark.lam, mhahnenberg, msaboff, oliver, sam | ||||||||||||||||
| Priority: | P2 | ||||||||||||||||||
| Version: | 528+ (Nightly build) | ||||||||||||||||||
| Hardware: | All | ||||||||||||||||||
| OS: | All | ||||||||||||||||||
| Bug Depends on: | 114707, 114708, 114762, 114906, 114909, 114987, 115083, 115297, 115299, 115300, 115301, 115445, 115582, 115594, 115598, 116060, 116126, 116350 | ||||||||||||||||||
| Bug Blocks: | 112836 | ||||||||||||||||||
| Attachments: |
|
||||||||||||||||||
|
Description
Filip Pizlo
2013-03-20 13:43:52 PDT
Created attachment 202219 [details]
work in progress
Created attachment 202222 [details]
it runs things
Wow. Benchmark report for SunSpider on oldmac (MacPro4,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/fourthTier/OpenSource/WebKitBuild/Release/jsc (r150349) "CoCo" at /Volumes/Data/fromMiniMe/fourthTier/secondary/OpenSource/WebKitBuild/Release/jsc (r150349) Collected 12 samples per benchmark/VM, with 4 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 CoCo 3d-cube 9.7970+-0.3084 ^ 7.0279+-0.5858 ^ definitely 1.3940x faster 3d-morph 9.0063+-0.2394 8.8302+-0.5370 might be 1.0199x faster 3d-raytrace 10.8543+-0.3490 ^ 8.3311+-0.1109 ^ definitely 1.3029x faster access-binary-trees 1.9811+-0.0081 ? 2.1782+-0.5912 ? might be 1.0994x slower access-fannkuch 7.8915+-0.1643 ? 8.5080+-2.2365 ? might be 1.0781x slower access-nbody 4.7534+-0.0515 ^ 3.9523+-0.0584 ^ definitely 1.2027x faster access-nsieve 4.9548+-0.0678 ! 6.1717+-0.1144 ! definitely 1.2456x slower bitops-3bit-bits-in-byte 1.8517+-0.0107 ^ 1.7721+-0.0183 ^ definitely 1.0449x faster bitops-bits-in-byte 6.6758+-0.0920 6.6059+-0.1225 might be 1.0106x faster bitops-bitwise-and 2.7389+-0.0519 ! 4.2159+-1.1382 ! definitely 1.5393x slower bitops-nsieve-bits 4.7877+-0.0547 ? 7.3282+-3.2467 ? might be 1.5306x slower controlflow-recursive 3.2102+-0.0488 3.1139+-0.1090 might be 1.0309x faster crypto-aes 7.8632+-0.1244 ^ 5.1990+-0.0911 ^ definitely 1.5125x faster crypto-md5 4.4320+-0.0377 ^ 3.0869+-0.0767 ^ definitely 1.4357x faster crypto-sha1 3.3562+-0.0248 ^ 2.8028+-0.0159 ^ definitely 1.1974x faster date-format-tofte 15.4000+-0.3783 ? 15.5954+-1.3100 ? might be 1.0127x slower date-format-xparb 9.6366+-0.1786 ^ 8.2300+-0.1334 ^ definitely 1.1709x faster math-cordic 4.0619+-0.0539 ^ 3.9165+-0.0129 ^ definitely 1.0371x faster math-partial-sums 12.6170+-0.1246 ^ 12.1648+-0.0850 ^ definitely 1.0372x faster math-spectral-norm 3.1937+-0.0119 ^ 2.7017+-0.0070 ^ definitely 1.1821x faster regexp-dna 13.2100+-0.3825 ^ 12.6349+-0.1839 ^ definitely 1.0455x faster string-base64 5.0036+-0.0991 ? 5.0320+-0.0483 ? string-fasta 11.1450+-0.2891 ^ 10.6305+-0.1452 ^ definitely 1.0484x faster string-tagcloud 14.3686+-0.2043 14.2502+-0.2312 string-unpack-code 27.9101+-0.1143 ? 28.0778+-0.6290 ? string-validate-input 7.3840+-0.2730 ^ 6.5804+-0.1159 ^ definitely 1.1221x faster <arithmetic> * 8.0032+-0.0840 ^ 7.6515+-0.1927 ^ definitely 1.0460x faster <geometric> 6.4521+-0.0570 ^ 6.0837+-0.1248 ^ definitely 1.0605x faster <harmonic> 5.2095+-0.0315 ^ 4.8993+-0.1004 ^ definitely 1.0633x faster Created attachment 202227 [details]
it's a speed-up, for the most part
Still have some slow-downs to investigate. But it's starting to really work.
Current results. Notice that some benchmarks, like navier-stokes, exhibit ridiculous slow-down. This is almost certainly a bug. Benchmark report for SunSpider, V8Spider, Octane, Kraken, and JSRegress on oldmac (MacPro4,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/fourthTier/OpenSource/WebKitBuild/Release/jsc (r150349) "CoCo" at /Volumes/Data/fromMiniMe/fourthTier/secondary/OpenSource/WebKitBuild/Release/jsc (r150349) Collected 12 samples per benchmark/VM, with 4 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 CoCo SunSpider: 3d-cube 9.7786+-0.2486 ^ 6.7297+-0.0589 ^ definitely 1.4531x faster 3d-morph 8.8167+-0.0967 ? 8.9819+-0.5222 ? might be 1.0187x slower 3d-raytrace 10.8967+-0.3513 ^ 8.2750+-0.1397 ^ definitely 1.3168x faster access-binary-trees 1.9855+-0.0096 ^ 1.8045+-0.0477 ^ definitely 1.1003x faster access-fannkuch 7.8129+-0.1508 ? 9.3907+-2.8883 ? might be 1.2020x slower access-nbody 4.7573+-0.0358 4.4756+-1.1266 might be 1.0630x faster access-nsieve 4.9352+-0.0712 ! 6.1566+-0.1039 ! definitely 1.2475x slower bitops-3bit-bits-in-byte 1.8525+-0.0102 ? 1.9083+-0.2077 ? might be 1.0301x slower bitops-bits-in-byte 6.6271+-0.0720 ? 6.7302+-0.1056 ? might be 1.0156x slower bitops-bitwise-and 2.7114+-0.0832 ! 3.9366+-1.1098 ! definitely 1.4519x slower bitops-nsieve-bits 4.7344+-0.0604 ? 5.9164+-2.4752 ? might be 1.2497x slower controlflow-recursive 3.1815+-0.0339 3.1131+-0.1030 might be 1.0220x faster crypto-aes 7.9432+-0.1245 ^ 5.1281+-0.0759 ^ definitely 1.5490x faster crypto-md5 4.4196+-0.0349 ^ 3.0194+-0.0297 ^ definitely 1.4637x faster crypto-sha1 3.3745+-0.0404 ^ 2.7956+-0.0178 ^ definitely 1.2071x faster date-format-tofte 15.3180+-0.2270 15.0403+-0.2231 might be 1.0185x faster date-format-xparb 9.5845+-0.1476 ^ 8.0790+-0.1620 ^ definitely 1.1863x faster math-cordic 4.0867+-0.0078 ? 4.3198+-0.9029 ? might be 1.0570x slower math-partial-sums 12.4959+-0.0957 ^ 12.1021+-0.0746 ^ definitely 1.0325x faster math-spectral-norm 3.1934+-0.0110 ^ 2.7070+-0.0119 ^ definitely 1.1797x faster regexp-dna 12.7434+-0.2371 12.6044+-0.2043 might be 1.0110x faster string-base64 5.0709+-0.0966 ? 5.1673+-0.2849 ? might be 1.0190x slower string-fasta 10.8532+-0.1253 10.6744+-0.1372 might be 1.0168x faster string-tagcloud 14.5169+-0.2767 14.2301+-0.1903 might be 1.0202x faster string-unpack-code 28.0144+-0.1109 27.6866+-0.5252 might be 1.0118x faster string-validate-input 7.1890+-0.1689 ^ 6.6443+-0.1905 ^ definitely 1.0820x faster <arithmetic> * 7.9574+-0.0694 ^ 7.6007+-0.1121 ^ definitely 1.0469x faster <geometric> 6.4203+-0.0483 ^ 6.0430+-0.0871 ^ definitely 1.0624x faster <harmonic> 5.1931+-0.0283 ^ 4.8566+-0.0738 ^ definitely 1.0693x faster TipOfTree CoCo V8Spider: crypto 247.3585+-0.4020 ^ 242.7377+-0.5920 ^ definitely 1.0190x faster deltablue 126.7658+-0.1396 ^ 106.2332+-7.5506 ^ definitely 1.1933x faster earley-boyer 83.7641+-0.4134 ^ 74.7976+-5.4524 ^ definitely 1.1199x faster raytrace 62.7493+-0.2024 ^ 37.6050+-0.1536 ^ definitely 1.6686x faster regexp 102.2530+-0.1857 ^ 100.3067+-0.4839 ^ definitely 1.0194x faster richards 118.3580+-0.2858 ^ 113.2298+-1.0252 ^ definitely 1.0453x faster splay 48.7244+-0.3366 ^ 47.4327+-0.4667 ^ definitely 1.0272x faster <arithmetic> 112.8533+-0.1439 ^ 103.1918+-1.7441 ^ definitely 1.0936x faster <geometric> * 99.5922+-0.1724 ^ 87.3317+-1.5862 ^ definitely 1.1404x faster <harmonic> 89.1031+-0.2119 ^ 74.8030+-1.1203 ^ definitely 1.1912x faster TipOfTree CoCo Octane and V8v7: encrypt 1.49987+-0.00933 1.49929+-0.00906 decrypt 28.16765+-0.06705 28.15719+-0.04085 deltablue x2 0.56801+-0.00090 ! 0.57283+-0.00201 ! definitely 1.0085x slower earley 0.88494+-0.00446 0.88448+-0.00576 boyer 12.78958+-0.09568 12.77414+-0.04585 raytrace x2 4.46236+-0.04225 ? 4.54341+-0.05089 ? might be 1.0182x slower regexp x2 32.43851+-0.17818 32.35575+-0.08666 richards x2 0.30187+-0.00008 ! 0.31085+-0.00573 ! definitely 1.0297x slower splay x2 0.62218+-0.01490 ? 0.65646+-0.04560 ? might be 1.0551x slower navier-stokes x2 10.79146+-0.01235 ! 23.36345+-8.28389 ! definitely 2.1650x slower closure 0.31881+-0.03402 0.31669+-0.03423 jquery 4.47099+-0.54795 ? 4.51045+-0.55188 ? gbemu x2 263.87720+-10.81013 258.18663+-11.14612 might be 1.0220x faster box2d x2 33.77843+-0.13454 ? 33.86534+-0.28680 ? V8v7: <arithmetic> 8.85693+-0.02612 ! 10.43254+-1.03001 ! definitely 1.1779x slower <geometric> * 2.78715+-0.01020 ! 3.05287+-0.15261 ! definitely 1.0953x slower <harmonic> 1.00189+-0.00451 ! 1.03018+-0.01599 ! definitely 1.0282x slower Octane including V8v7: <arithmetic> 33.71872+-1.00995 ? 34.35690+-1.24531 ? might be 1.0189x slower <geometric> * 4.88991+-0.07278 ! 5.21323+-0.20748 ! definitely 1.0661x slower <harmonic> 1.13031+-0.02225 ? 1.15546+-0.03132 ? might be 1.0222x slower TipOfTree CoCo Kraken: ai-astar 493.841+-0.650 ? 496.357+-4.487 ? audio-beat-detection 1204.313+-3.631 ! 1224.124+-10.203 ! definitely 1.0164x slower audio-dft 310.932+-0.909 309.026+-1.527 audio-fft 1144.335+-7.195 1131.977+-6.795 might be 1.0109x faster audio-oscillator 233.131+-1.077 232.778+-1.102 imaging-darkroom 292.055+-0.850 291.050+-0.829 imaging-desaturate 853.947+-3.313 852.413+-3.945 imaging-gaussian-blur 406.844+-5.471 403.461+-0.323 json-parse-financial 82.815+-1.210 81.831+-0.319 might be 1.0120x faster json-stringify-tinderbox 99.140+-0.318 ! 100.808+-0.249 ! definitely 1.0168x slower stanford-crypto-aes 183.598+-0.868 183.096+-1.188 stanford-crypto-ccm 215.641+-1.535 190.852+-24.833 might be 1.1299x faster stanford-crypto-pbkdf2 270.211+-0.635 269.042+-2.759 stanford-crypto-sha256-iterative 116.998+-0.515 ^ 115.881+-0.389 ^ definitely 1.0096x faster <arithmetic> * 421.986+-0.899 420.192+-1.980 might be 1.0043x faster <geometric> 301.550+-0.579 297.993+-3.142 might be 1.0119x faster <harmonic> 222.969+-0.643 219.542+-3.109 might be 1.0156x faster TipOfTree CoCo JSRegress: adapt-to-double-divide 22.4739+-0.0930 ? 31.6828+-10.5863 ? might be 1.4098x slower aliased-arguments-getbyval 0.9335+-0.0125 ^ 0.8175+-0.0079 ^ definitely 1.1419x faster allocate-big-object 2.5427+-0.0354 ? 2.6232+-0.2472 ? might be 1.0317x slower arity-mismatch-inlining 0.7796+-0.0117 ? 1.0926+-0.3755 ? might be 1.4015x slower array-access-polymorphic-structure 7.0968+-0.0844 ^ 6.8144+-0.0696 ^ definitely 1.0414x faster array-with-double-add 5.8303+-0.0842 5.7995+-0.0644 array-with-double-increment 4.1078+-0.0123 ? 4.1340+-0.0686 ? array-with-double-mul-add 7.0954+-0.0899 ? 15.1081+-11.8846 ? might be 2.1293x slower array-with-double-sum 7.8917+-0.1094 ? 15.5021+-8.7747 ? might be 1.9644x slower array-with-int32-add-sub 10.4537+-0.1057 ? 10.5442+-0.1097 ? array-with-int32-or-double-sum 7.9409+-0.0891 ? 10.4671+-5.4904 ? might be 1.3181x slower big-int-mul 4.9434+-0.0658 ? 5.6440+-1.2370 ? might be 1.1417x slower boolean-test 4.4148+-0.0760 4.3702+-0.0616 might be 1.0102x faster cast-int-to-double 13.9257+-0.1357 ? 14.8823+-2.1077 ? might be 1.0687x slower cell-argument 14.4188+-0.1139 ? 17.8431+-3.9468 ? might be 1.2375x slower cfg-simplify 3.9719+-0.0662 ? 3.9803+-0.0439 ? cmpeq-obj-to-obj-other 12.2042+-0.4054 ? 13.1207+-2.1311 ? might be 1.0751x slower constant-test 8.3285+-0.1645 ? 11.8873+-7.3971 ? might be 1.4273x slower direct-arguments-getbyval 0.8539+-0.0080 ^ 0.7389+-0.0307 ^ definitely 1.1557x faster double-pollution-getbyval 10.7442+-0.1385 ? 14.5226+-4.6349 ? might be 1.3517x slower double-pollution-putbyoffset 5.1270+-0.0994 ? 5.5488+-1.1070 ? might be 1.0823x slower empty-string-plus-int 10.9205+-0.1937 10.8740+-0.1591 external-arguments-getbyval 2.2125+-0.0260 ^ 1.8978+-0.0153 ^ definitely 1.1658x faster external-arguments-putbyval 3.3016+-0.0091 3.2219+-0.1288 might be 1.0247x faster Float32Array-matrix-mult 14.2612+-0.0958 ^ 13.7907+-0.1085 ^ definitely 1.0341x faster fold-double-to-int 21.9474+-0.1485 ? 23.3149+-2.9695 ? might be 1.0623x slower function-dot-apply 3.2041+-0.0132 ? 3.4739+-0.5573 ? might be 1.0842x slower function-test 4.9854+-0.0949 ? 5.6177+-1.4156 ? might be 1.1268x slower get-by-id-chain-from-try-block 6.7581+-0.1894 ? 6.8534+-0.1469 ? might be 1.0141x slower HashMap-put-get-iterate-keys 96.1959+-1.2993 ^ 93.8202+-0.9095 ^ definitely 1.0253x faster HashMap-put-get-iterate 99.3947+-1.0722 ^ 97.4217+-0.6628 ^ definitely 1.0203x faster HashMap-string-put-get-iterate 75.1716+-0.3378 ^ 71.5885+-0.7680 ^ definitely 1.0501x faster indexed-properties-in-objects 4.5383+-0.0646 ? 4.7673+-0.3401 ? might be 1.0505x slower inline-arguments-access 1.2664+-0.0086 ? 1.9950+-0.8140 ? might be 1.5753x slower inline-arguments-local-escape 23.1776+-0.1920 22.9687+-0.1958 inline-get-scoped-var 6.5421+-0.1403 ? 9.4513+-4.0656 ? might be 1.4447x slower inlined-put-by-id-transition 16.7348+-0.3361 16.3492+-0.5430 might be 1.0236x faster int-or-other-abs-then-get-by-val 8.7916+-0.0755 ? 9.6336+-1.9292 ? might be 1.0958x slower int-or-other-abs-zero-then-get-by-val 37.3649+-0.2140 ? 38.8876+-2.7263 ? might be 1.0408x slower int-or-other-add-then-get-by-val 10.2339+-0.0866 ! 15.3102+-4.7637 ! definitely 1.4960x slower int-or-other-add 10.4608+-0.0814 ! 17.8036+-4.0116 ! definitely 1.7019x slower int-or-other-div-then-get-by-val 7.8719+-0.0881 7.8416+-0.0912 int-or-other-max-then-get-by-val 9.9540+-0.2137 ? 11.8540+-2.9076 ? might be 1.1909x slower int-or-other-min-then-get-by-val 8.0446+-0.0652 ? 8.9856+-1.3212 ? might be 1.1170x slower int-or-other-mod-then-get-by-val 8.0506+-0.1008 7.9150+-0.1011 might be 1.0171x faster int-or-other-mul-then-get-by-val 7.1369+-0.0927 ? 7.8557+-1.6019 ? might be 1.1007x slower int-or-other-neg-then-get-by-val 8.1312+-0.1002 ! 10.1972+-1.6909 ! definitely 1.2541x slower int-or-other-neg-zero-then-get-by-val 36.5502+-0.2997 36.3447+-0.1276 int-or-other-sub-then-get-by-val 10.2345+-0.1038 ? 11.5472+-2.9679 ? might be 1.1283x slower int-or-other-sub 8.2688+-0.0986 ? 11.0039+-3.2494 ? might be 1.3308x slower int-overflow-local 12.8705+-0.0951 ? 13.7738+-2.2198 ? might be 1.0702x slower Int16Array-bubble-sort 49.4250+-0.1755 ? 52.1285+-4.3410 ? might be 1.0547x slower Int16Array-load-int-mul 1.8956+-0.0095 ? 2.6325+-1.5924 ? might be 1.3888x slower Int8Array-load 4.8114+-0.0689 ? 5.1702+-0.5741 ? might be 1.0746x slower integer-divide 15.1066+-0.1226 ? 17.8787+-6.1291 ? might be 1.1835x slower integer-modulo 2.1554+-0.0116 ^ 2.0299+-0.0368 ^ definitely 1.0618x faster make-indexed-storage 3.9397+-0.0472 3.8425+-0.2482 might be 1.0253x faster method-on-number 25.9195+-0.5861 25.4819+-0.3081 might be 1.0172x faster nested-function-parsing-random 379.4199+-13.0925 ? 381.5733+-13.0563 ? nested-function-parsing 47.9182+-1.0874 47.4300+-1.0438 might be 1.0103x faster new-array-buffer-dead 3.6458+-0.0122 ? 6.2953+-5.8220 ? might be 1.7267x slower new-array-buffer-push 10.4308+-0.2114 10.1843+-0.1701 might be 1.0242x faster new-array-dead 28.2093+-0.0816 ! 42.3392+-9.3632 ! definitely 1.5009x slower new-array-push 6.9565+-0.1293 6.9478+-0.3606 number-test 4.3422+-0.0524 4.2916+-0.0562 might be 1.0118x faster object-closure-call 8.4023+-0.1510 ! 11.8350+-1.7579 ! definitely 1.4085x slower object-test 4.9035+-0.0348 ? 5.4789+-1.3989 ? might be 1.1173x slower poly-stricteq 984.3646+-13.4675 ? 1028.7016+-49.3085 ? might be 1.0450x slower polymorphic-structure 20.0156+-0.1019 ! 58.9155+-14.8085 ! definitely 2.9435x slower polyvariant-monomorphic-get-by-id 12.5454+-0.1149 ? 15.0380+-5.4914 ? might be 1.1987x slower rare-osr-exit-on-local 20.6400+-0.1368 ! 26.4702+-3.7327 ! definitely 1.2825x slower register-pressure-from-osr 31.5666+-0.1310 ? 34.1304+-5.3456 ? might be 1.0812x slower simple-activation-demo 34.3976+-0.1046 ? 61.1678+-58.8882 ? might be 1.7783x slower slow-array-profile-convergence 4.4311+-0.0428 ^ 4.1532+-0.1402 ^ definitely 1.0669x faster slow-convergence 3.8460+-0.0511 ^ 3.4631+-0.0825 ^ definitely 1.1105x faster sparse-conditional 1.3262+-0.0136 ? 1.4397+-0.3124 ? might be 1.0856x slower splice-to-remove 50.3002+-0.1648 ^ 49.2651+-0.1984 ^ definitely 1.0210x faster string-concat-object 2.7010+-0.0456 2.6934+-0.0286 string-concat-pair-object 2.6667+-0.0202 2.6593+-0.0438 string-concat-pair-simple 17.0172+-0.2936 ? 18.2761+-1.7396 ? might be 1.0740x slower string-concat-simple 16.8689+-0.2328 ? 17.5145+-1.2728 ? might be 1.0383x slower string-cons-repeat 10.0805+-0.0302 ? 12.7000+-2.9505 ? might be 1.2599x slower string-cons-tower 10.9122+-0.0330 10.8540+-0.0572 string-equality 106.5795+-0.1729 ^ 104.4463+-0.1456 ^ definitely 1.0204x faster string-hash 2.6712+-0.0127 ^ 2.5469+-0.0123 ^ definitely 1.0488x faster string-repeat-arith 45.3087+-0.6742 ^ 44.2748+-0.1749 ^ definitely 1.0234x faster string-sub 88.7269+-1.3198 87.4230+-1.7861 might be 1.0149x faster string-test 4.3118+-0.0364 ? 4.8124+-1.2970 ? might be 1.1161x slower structure-hoist-over-transitions 3.2898+-0.0278 ? 3.3188+-0.1845 ? tear-off-arguments-simple 1.8101+-0.0077 ? 1.9340+-0.2435 ? might be 1.0684x slower tear-off-arguments 3.3889+-0.0155 3.2447+-0.1354 might be 1.0444x faster temporal-structure 20.7727+-0.0501 ! 49.8408+-19.4181 ! definitely 2.3993x slower to-int32-boolean 30.5883+-0.1113 ? 30.6623+-0.1350 ? undefined-test 4.5833+-0.0375 4.5686+-0.0860 <arithmetic> 30.4665+-0.2127 ! 33.0164+-1.0522 ! definitely 1.0837x slower <geometric> * 9.6806+-0.0289 ! 10.4774+-0.1859 ! definitely 1.0823x slower <harmonic> 5.2218+-0.0194 ? 5.3190+-0.0787 ? might be 1.0186x slower TipOfTree CoCo All benchmarks: <arithmetic> 64.4805+-0.2557 ? 65.4113+-0.7401 ? might be 1.0144x slower <geometric> 12.2789+-0.0595 ! 12.7496+-0.1594 ! definitely 1.0383x slower <harmonic> 3.8243+-0.0368 ? 3.8577+-0.0445 ? might be 1.0087x slower TipOfTree CoCo Geomean of preferred means: <scaled-result> 27.5350+-0.1518 27.3131+-0.2670 might be 1.0081x faster Created attachment 202238 [details] fixed more buggies Still investigating things, but we're now up to 8% SunSpider speed-up. Benchmark report for SunSpider, V8Spider, Octane, Kraken, and JSRegress on oldmac (MacPro4,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/fourthTier/OpenSource/WebKitBuild/Release/jsc (r150349) "CoCo" at /Volumes/Data/fromMiniMe/fourthTier/secondary/OpenSource/WebKitBuild/Release/jsc (r150349) Collected 12 samples per benchmark/VM, with 4 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 CoCo SunSpider: 3d-cube 9.5285+-0.1540 ^ 6.6626+-0.1032 ^ definitely 1.4301x faster 3d-morph 8.8888+-0.1291 ^ 8.5282+-0.1103 ^ definitely 1.0423x faster 3d-raytrace 10.5736+-0.1471 ^ 8.3586+-0.1267 ^ definitely 1.2650x faster access-binary-trees 1.9797+-0.0078 ^ 1.8267+-0.0135 ^ definitely 1.0838x faster access-fannkuch 7.8151+-0.1464 ^ 7.4967+-0.0599 ^ definitely 1.0425x faster access-nbody 4.7413+-0.0605 ^ 3.9856+-0.0109 ^ definitely 1.1896x faster access-nsieve 4.9422+-0.0293 ^ 4.8525+-0.0294 ^ definitely 1.0185x faster bitops-3bit-bits-in-byte 1.8526+-0.0120 ^ 1.7759+-0.0164 ^ definitely 1.0432x faster bitops-bits-in-byte 6.6707+-0.1072 ? 6.6850+-0.1330 ? bitops-bitwise-and 2.7080+-0.0744 ? 2.7749+-0.0337 ? might be 1.0247x slower bitops-nsieve-bits 4.7892+-0.0502 ^ 4.4798+-0.0357 ^ definitely 1.0691x faster controlflow-recursive 3.1806+-0.0350 ^ 3.0745+-0.0192 ^ definitely 1.0345x faster crypto-aes 7.8931+-0.1327 ^ 5.1704+-0.0682 ^ definitely 1.5266x faster crypto-md5 4.4070+-0.0595 ^ 3.0655+-0.0275 ^ definitely 1.4376x faster crypto-sha1 3.3563+-0.0149 ^ 2.8171+-0.0185 ^ definitely 1.1914x faster date-format-tofte 15.2275+-0.1845 14.9006+-0.1817 might be 1.0219x faster date-format-xparb 9.7970+-0.2375 ^ 8.3792+-0.1627 ^ definitely 1.1692x faster math-cordic 4.0941+-0.0115 ^ 3.9618+-0.0486 ^ definitely 1.0334x faster math-partial-sums 12.5590+-0.1206 ^ 12.2155+-0.1270 ^ definitely 1.0281x faster math-spectral-norm 3.1948+-0.0121 ^ 2.7176+-0.0178 ^ definitely 1.1756x faster regexp-dna 12.7943+-0.2196 12.6545+-0.1833 might be 1.0110x faster string-base64 5.0671+-0.0789 5.0279+-0.0450 string-fasta 10.8400+-0.1384 10.7529+-0.1644 string-tagcloud 14.4488+-0.1885 14.1776+-0.2011 might be 1.0191x faster string-unpack-code 28.0919+-0.1948 ^ 27.5249+-0.1342 ^ definitely 1.0206x faster string-validate-input 7.1768+-0.1531 ^ 6.4514+-0.1304 ^ definitely 1.1124x faster <arithmetic> * 7.9468+-0.0666 ^ 7.3199+-0.0543 ^ definitely 1.0856x faster <geometric> 6.4135+-0.0449 ^ 5.8037+-0.0351 ^ definitely 1.1051x faster <harmonic> 5.1892+-0.0250 ^ 4.6901+-0.0211 ^ definitely 1.1064x faster TipOfTree CoCo V8Spider: crypto 247.7954+-0.5886 ^ 243.3546+-0.8866 ^ definitely 1.0182x faster deltablue 126.7398+-0.0859 ^ 102.6285+-1.0971 ^ definitely 1.2349x faster earley-boyer 83.8980+-0.3234 ^ 69.5889+-0.2082 ^ definitely 1.2056x faster raytrace 62.7492+-0.2910 ^ 38.0090+-0.1596 ^ definitely 1.6509x faster regexp 102.9754+-0.9878 ^ 99.7671+-0.1285 ^ definitely 1.0322x faster richards 118.3877+-0.2179 ^ 112.6261+-0.8601 ^ definitely 1.0512x faster splay 48.7556+-0.1596 ^ 47.0282+-0.3036 ^ definitely 1.0367x faster <arithmetic> 113.0430+-0.2012 ^ 101.8575+-0.2796 ^ definitely 1.1098x faster <geometric> * 99.7493+-0.1809 ^ 86.0376+-0.2319 ^ definitely 1.1594x faster <harmonic> 89.2255+-0.1756 ^ 73.9141+-0.1959 ^ definitely 1.2072x faster TipOfTree CoCo Octane and V8v7: encrypt 1.49364+-0.00896 ? 1.49674+-0.00804 ? decrypt 28.19836+-0.04289 28.16538+-0.05527 deltablue x2 0.56748+-0.00049 ! 0.57325+-0.00166 ! definitely 1.0102x slower earley 0.88295+-0.00333 0.87931+-0.00275 boyer 12.74352+-0.01894 ! 12.81610+-0.04487 ! definitely 1.0057x slower raytrace x2 4.43320+-0.02796 ! 4.51827+-0.03663 ! definitely 1.0192x slower regexp x2 32.55321+-0.15627 32.54845+-0.17454 richards x2 0.30373+-0.00066 ! 0.31527+-0.00570 ! definitely 1.0380x slower splay x2 0.61463+-0.00495 ? 0.64555+-0.02756 ? might be 1.0503x slower navier-stokes x2 10.78794+-0.01333 10.77550+-0.00845 closure 0.31921+-0.03402 0.31802+-0.03442 jquery 4.48681+-0.54602 ? 4.49695+-0.54853 ? gbemu x2 262.74903+-10.90715 258.06415+-12.03657 might be 1.0182x faster box2d x2 33.82863+-0.11753 33.76532+-0.21236 V8v7: <arithmetic> 8.86493+-0.02362 ? 8.88188+-0.02486 ? might be 1.0019x slower <geometric> * 2.78219+-0.00647 ! 2.82190+-0.01711 ! definitely 1.0143x slower <harmonic> 1.00140+-0.00210 ! 1.02883+-0.01185 ! definitely 1.0274x slower Octane including V8v7: <arithmetic> 33.62728+-1.01897 33.20837+-1.12328 might be 1.0126x faster <geometric> * 4.88340+-0.07030 ? 4.92456+-0.07864 ? might be 1.0084x slower <harmonic> 1.13010+-0.02092 ? 1.15460+-0.02540 ? might be 1.0217x slower TipOfTree CoCo Kraken: ai-astar 494.364+-0.295 ^ 491.928+-0.577 ^ definitely 1.0050x faster audio-beat-detection 1210.120+-7.781 ? 1225.708+-18.890 ? might be 1.0129x slower audio-dft 310.353+-0.997 ^ 308.155+-0.674 ^ definitely 1.0071x faster audio-fft 1132.330+-5.151 ? 1230.324+-98.202 ? might be 1.0865x slower audio-oscillator 233.828+-1.205 232.605+-0.973 imaging-darkroom 291.906+-0.870 291.580+-0.877 imaging-desaturate 853.856+-3.697 852.350+-3.563 imaging-gaussian-blur 403.180+-0.329 ? 403.314+-0.203 ? json-parse-financial 81.987+-0.429 81.485+-0.206 json-stringify-tinderbox 99.899+-0.374 99.860+-0.293 stanford-crypto-aes 185.453+-1.308 184.623+-0.956 stanford-crypto-ccm 213.829+-1.239 ? 215.180+-2.589 ? stanford-crypto-pbkdf2 270.560+-0.884 ! 272.451+-0.965 ! definitely 1.0070x slower stanford-crypto-sha256-iterative 116.885+-0.365 ^ 115.510+-0.158 ^ definitely 1.0119x faster <arithmetic> * 421.325+-1.027 ? 428.934+-7.186 ? might be 1.0181x slower <geometric> 301.263+-0.526 ? 302.555+-1.703 ? might be 1.0043x slower <harmonic> 222.795+-0.509 222.298+-0.468 might be 1.0022x faster TipOfTree CoCo JSRegress: adapt-to-double-divide 22.4466+-0.0983 ? 22.5301+-0.1122 ? aliased-arguments-getbyval 0.9305+-0.0109 ^ 0.8241+-0.0078 ^ definitely 1.1291x faster allocate-big-object 2.6085+-0.0957 ^ 2.4611+-0.0276 ^ definitely 1.0599x faster arity-mismatch-inlining 0.7793+-0.0115 ? 0.7802+-0.0195 ? array-access-polymorphic-structure 7.0565+-0.1269 ^ 6.8076+-0.0788 ^ definitely 1.0366x faster array-with-double-add 5.8062+-0.0914 ? 5.8666+-0.0731 ? might be 1.0104x slower array-with-double-increment 4.1395+-0.0935 4.1098+-0.0347 array-with-double-mul-add 7.0991+-0.1008 7.0900+-0.0834 array-with-double-sum 7.8402+-0.1096 ? 7.9001+-0.0895 ? array-with-int32-add-sub 10.3817+-0.0805 ? 10.4571+-0.0570 ? array-with-int32-or-double-sum 7.9751+-0.0966 7.9568+-0.0838 big-int-mul 4.9398+-0.0774 ^ 4.7980+-0.0588 ^ definitely 1.0295x faster boolean-test 4.4610+-0.0665 4.3665+-0.0777 might be 1.0216x faster cast-int-to-double 13.8484+-0.1551 ? 13.8849+-0.1852 ? cell-argument 14.4327+-0.1460 14.4141+-0.0964 cfg-simplify 3.9975+-0.0456 3.9727+-0.0631 cmpeq-obj-to-obj-other 12.1682+-0.2698 12.1626+-0.3441 constant-test 8.4068+-0.1362 ? 8.5511+-0.0826 ? might be 1.0172x slower direct-arguments-getbyval 0.8562+-0.0086 ^ 0.7316+-0.0112 ^ definitely 1.1703x faster double-pollution-getbyval 10.7168+-0.1099 ? 10.9031+-0.1260 ? might be 1.0174x slower double-pollution-putbyoffset 5.0900+-0.1512 4.9592+-0.0735 might be 1.0264x faster empty-string-plus-int 10.9669+-0.2028 10.8820+-0.1666 external-arguments-getbyval 2.2131+-0.0352 ^ 1.9324+-0.0121 ^ definitely 1.1453x faster external-arguments-putbyval 3.2895+-0.0200 ^ 3.1656+-0.0221 ^ definitely 1.0391x faster Float32Array-matrix-mult 14.2250+-0.0722 ^ 13.5711+-0.1048 ^ definitely 1.0482x faster fold-double-to-int 21.9252+-0.3220 ? 22.0347+-0.1511 ? function-dot-apply 3.2002+-0.0133 ^ 3.1235+-0.0254 ^ definitely 1.0245x faster function-test 4.9808+-0.0608 ? 5.0180+-0.0878 ? get-by-id-chain-from-try-block 6.7345+-0.1870 6.7021+-0.1603 HashMap-put-get-iterate-keys 96.9098+-1.6110 ^ 92.4773+-0.8929 ^ definitely 1.0479x faster HashMap-put-get-iterate 100.5368+-0.5532 ^ 95.4911+-0.9847 ^ definitely 1.0528x faster HashMap-string-put-get-iterate 75.2821+-0.5074 ^ 72.5220+-0.7001 ^ definitely 1.0381x faster indexed-properties-in-objects 4.5818+-0.0126 ^ 4.4151+-0.0072 ^ definitely 1.0378x faster inline-arguments-access 1.2674+-0.0093 ! 1.2855+-0.0085 ! definitely 1.0143x slower inline-arguments-local-escape 23.2293+-0.1921 23.1472+-0.2079 inline-get-scoped-var 6.5505+-0.0923 ? 6.6310+-0.0851 ? might be 1.0123x slower inlined-put-by-id-transition 16.8299+-0.2828 ^ 16.1305+-0.2523 ^ definitely 1.0434x faster int-or-other-abs-then-get-by-val 8.8349+-0.0854 8.8140+-0.0954 int-or-other-abs-zero-then-get-by-val 37.2495+-0.1369 ^ 36.9133+-0.1011 ^ definitely 1.0091x faster int-or-other-add-then-get-by-val 10.2423+-0.1246 10.2387+-0.1324 int-or-other-add 10.5475+-0.1212 ? 10.6019+-0.0853 ? int-or-other-div-then-get-by-val 7.8563+-0.0869 7.7964+-0.0981 int-or-other-max-then-get-by-val 10.0489+-0.1153 ? 10.0814+-0.1695 ? int-or-other-min-then-get-by-val 8.2059+-0.1147 ^ 8.0141+-0.0687 ^ definitely 1.0239x faster int-or-other-mod-then-get-by-val 8.0701+-0.0709 7.9355+-0.0870 might be 1.0170x faster int-or-other-mul-then-get-by-val 7.2558+-0.0939 ^ 7.0464+-0.0838 ^ definitely 1.0297x faster int-or-other-neg-then-get-by-val 8.1465+-0.1022 8.0454+-0.0954 might be 1.0126x faster int-or-other-neg-zero-then-get-by-val 36.8138+-0.3771 36.4259+-0.2154 might be 1.0106x faster int-or-other-sub-then-get-by-val 10.3160+-0.1392 10.2053+-0.0971 might be 1.0108x faster int-or-other-sub 8.2569+-0.0999 8.2380+-0.0979 int-overflow-local 12.8636+-0.1229 12.7774+-0.1217 Int16Array-bubble-sort 49.3466+-0.1220 49.1553+-0.1455 Int16Array-load-int-mul 1.8942+-0.0112 ! 1.9136+-0.0067 ! definitely 1.0102x slower Int8Array-load 4.8423+-0.0599 4.7814+-0.0618 might be 1.0127x faster integer-divide 15.1039+-0.1330 ? 15.2086+-0.1146 ? integer-modulo 2.1546+-0.0110 ^ 2.0247+-0.0153 ^ definitely 1.0641x faster make-indexed-storage 3.9221+-0.0345 ^ 3.5742+-0.0347 ^ definitely 1.0973x faster method-on-number 25.3716+-0.4273 25.3462+-0.4099 nested-function-parsing-random 379.8008+-12.9035 ? 379.9561+-13.0431 ? nested-function-parsing 47.9680+-1.0905 47.5330+-1.0243 new-array-buffer-dead 3.6434+-0.0123 ? 3.6564+-0.0304 ? new-array-buffer-push 10.4945+-0.2015 10.2637+-0.1779 might be 1.0225x faster new-array-dead 28.2327+-0.1104 ? 28.3446+-0.0945 ? new-array-push 6.9503+-0.1022 6.8054+-0.0983 might be 1.0213x faster number-test 4.3402+-0.0535 4.2790+-0.0582 might be 1.0143x faster object-closure-call 8.4685+-0.0987 ! 10.5277+-1.9032 ! definitely 1.2432x slower object-test 4.9035+-0.0603 4.8286+-0.0750 might be 1.0155x faster poly-stricteq 980.3060+-7.8710 ! 1029.5500+-41.2130 ! definitely 1.0502x slower polymorphic-structure 20.0159+-0.1304 ! 20.8507+-0.0808 ! definitely 1.0417x slower polyvariant-monomorphic-get-by-id 12.5403+-0.0946 12.4817+-0.0682 rare-osr-exit-on-local 20.4776+-0.0831 ? 20.5206+-0.0833 ? register-pressure-from-osr 31.4887+-0.1135 31.2958+-0.1150 simple-activation-demo 34.5363+-0.2762 34.2328+-0.0794 slow-array-profile-convergence 4.4793+-0.0917 ^ 4.0832+-0.0466 ^ definitely 1.0970x faster slow-convergence 3.8507+-0.0066 ^ 3.4197+-0.0096 ^ definitely 1.1261x faster sparse-conditional 1.3163+-0.0188 ? 1.3168+-0.0166 ? splice-to-remove 50.1300+-0.1299 ? 50.2866+-0.1812 ? string-concat-object 2.7650+-0.0503 2.7096+-0.0393 might be 1.0204x faster string-concat-pair-object 2.6786+-0.0516 2.6283+-0.0225 might be 1.0192x faster string-concat-pair-simple 17.2471+-0.2528 ^ 16.7335+-0.2530 ^ definitely 1.0307x faster string-concat-simple 16.9704+-0.3704 ? 17.0389+-0.3626 ? string-cons-repeat 10.0613+-0.0222 ! 10.1398+-0.0250 ! definitely 1.0078x slower string-cons-tower 10.9021+-0.0357 10.8617+-0.0354 string-equality 106.4793+-0.1430 105.6643+-1.4763 string-hash 2.6725+-0.0152 ^ 2.5458+-0.0092 ^ definitely 1.0497x faster string-repeat-arith 45.2100+-0.3428 ? 45.2340+-0.2505 ? string-sub 87.5054+-0.8567 ? 89.4133+-1.1788 ? might be 1.0218x slower string-test 4.3056+-0.0354 ^ 4.2396+-0.0269 ^ definitely 1.0156x faster structure-hoist-over-transitions 3.2815+-0.0216 ^ 3.1741+-0.0264 ^ definitely 1.0338x faster tear-off-arguments-simple 1.8062+-0.0077 ^ 1.7394+-0.0074 ^ definitely 1.0384x faster tear-off-arguments 3.3933+-0.0155 ^ 3.1772+-0.0196 ^ definitely 1.0680x faster temporal-structure 20.9042+-0.0911 ? 20.9465+-0.0642 ? to-int32-boolean 30.5509+-0.0859 ? 30.6098+-0.1244 ? undefined-test 4.5289+-0.0791 4.5066+-0.0441 <arithmetic> 30.4389+-0.1196 ? 30.7956+-0.3653 ? might be 1.0117x slower <geometric> * 9.6932+-0.0261 ^ 9.5507+-0.0353 ^ definitely 1.0149x faster <harmonic> 5.2284+-0.0169 ^ 5.0416+-0.0231 ^ definitely 1.0371x faster TipOfTree CoCo All benchmarks: <arithmetic> 64.4020+-0.2359 ? 64.6243+-0.5400 ? might be 1.0035x slower <geometric> 12.2837+-0.0565 ^ 11.9286+-0.0628 ^ definitely 1.0298x faster <harmonic> 3.8258+-0.0348 3.7547+-0.0388 might be 1.0189x faster TipOfTree CoCo Geomean of preferred means: <scaled-result> 27.5276+-0.1477 ^ 26.3489+-0.1858 ^ definitely 1.0447x faster Created attachment 202256 [details]
moar
Still playing around with alternatives.
Created attachment 202329 [details]
more things
Changed how GC is deferred. This appears to be an improvement and it fixes some crashes. Still testing though.
Created attachment 202380 [details]
the patch
It passes tests.
Comment on attachment 202380 [details] the patch View in context: https://bugs.webkit.org/attachment.cgi?id=202380&action=review r=me > Source/JavaScriptCore/jit/JITStubs.cpp:959 > + worklistState = > + stackFrame.vm->worklist->completeAllReadyPlansForVM(*stackFrame.vm, codeBlock); This seems OK for now, but it might become a problem if LLVM compilation sticks at ~4ms / function. If you had 4 functions queued up, one call to cti_optimize would cause you to drop an animation frame. If you had 16 functions queued up, you'd drop 4 frames. Basically, there's a tradeoff here between startup time and responsiveness, and the nob is turned all the way toward startup time right now, which is probably the wrong balance. > Source/JavaScriptCore/runtime/JSSegmentedVariableObject.h:97 > + typedef ConcurrentJITLock Lock; > + typedef ConcurrentJITLocker Locker; Actually, I think this reads even better without the typedef, since it calls out the other client you're locking against. (In reply to comment #10) > (From update of attachment 202380 [details]) > View in context: https://bugs.webkit.org/attachment.cgi?id=202380&action=review > > r=me > > > Source/JavaScriptCore/jit/JITStubs.cpp:959 > > + worklistState = > > + stackFrame.vm->worklist->completeAllReadyPlansForVM(*stackFrame.vm, codeBlock); > > This seems OK for now, but it might become a problem if LLVM compilation sticks at ~4ms / function. If you had 4 functions queued up, one call to cti_optimize would cause you to drop an animation frame. If you had 16 functions queued up, you'd drop 4 frames. Basically, there's a tradeoff here between startup time and responsiveness, and the nob is turned all the way toward startup time right now, which is probably the wrong balance. No it won't, there's no trade-off. This method, "complete all ready [sic] plans for VM", does not wait for the compilation thread. It just completes the ones that are done. The only waiting it does is on the Worklist::m_lock, which is only held when modifying the queue/map/finished data structures. Those modifications are pretty quick - at worst, grab the lock, loop over the list, release the lock. The compilation thread only holds the lock for a very short time - at worst to dequeue or to append. This is distinct from completeAllPlansForVM(), which does wait for the compilation thread to finish. > > > Source/JavaScriptCore/runtime/JSSegmentedVariableObject.h:97 > > + typedef ConcurrentJITLock Lock; > > + typedef ConcurrentJITLocker Locker; > > Actually, I think this reads even better without the typedef, since it calls out the other client you're locking against. Do you advocate making this change throughout? There are a bunch of places that use the typedef-ConcurrentJITLock-to-Lock idiom. I kind of like how this makes most code terse but still lets you find what the typedef means. But I kind of like the change you're proposing. Since that would be a larger patch (all of the places that use ConcurrentJITLock, and there are a bunch of them), how about I do it in a separate bug? Most up-to-date numbers: Benchmark report for SunSpider, V8Spider, Octane, Kraken, and JSRegress on oldmac (MacPro4,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/fourthTier/OpenSource/WebKitBuild/Release/jsc (r150398) "CoCo" at /Volumes/Data/fromMiniMe/fourthTier/secondary/OpenSource/WebKitBuild/Release/jsc (r150398) Collected 12 samples per benchmark/VM, with 4 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 CoCo SunSpider: 3d-cube 9.6499+-0.2056 ^ 6.7133+-0.1048 ^ definitely 1.4374x faster 3d-morph 8.9624+-0.1326 ^ 8.5918+-0.1158 ^ definitely 1.0431x faster 3d-raytrace 10.6995+-0.1522 ^ 8.3448+-0.1436 ^ definitely 1.2822x faster access-binary-trees 1.9843+-0.0099 1.9258+-0.0948 might be 1.0304x faster access-fannkuch 7.8671+-0.1521 ^ 7.5370+-0.0595 ^ definitely 1.0438x faster access-nbody 4.7480+-0.0643 ^ 4.0090+-0.0225 ^ definitely 1.1844x faster access-nsieve 4.9202+-0.0510 ^ 4.7887+-0.0721 ^ definitely 1.0275x faster bitops-3bit-bits-in-byte 1.8576+-0.0104 ^ 1.7855+-0.0192 ^ definitely 1.0404x faster bitops-bits-in-byte 6.6267+-0.0955 6.5624+-0.0832 bitops-bitwise-and 2.8060+-0.0186 2.7799+-0.0239 bitops-nsieve-bits 4.7735+-0.0727 ^ 4.4933+-0.0226 ^ definitely 1.0624x faster controlflow-recursive 3.1904+-0.0353 ^ 3.0758+-0.0189 ^ definitely 1.0372x faster crypto-aes 8.0045+-0.1763 ^ 5.1721+-0.0808 ^ definitely 1.5476x faster crypto-md5 4.3211+-0.0290 ^ 3.1718+-0.0213 ^ definitely 1.3624x faster crypto-sha1 3.3981+-0.0154 ^ 2.8351+-0.0177 ^ definitely 1.1986x faster date-format-tofte 15.3863+-0.1878 ^ 14.5663+-0.1854 ^ definitely 1.0563x faster date-format-xparb 9.4774+-0.1809 ^ 8.2224+-0.1213 ^ definitely 1.1526x faster math-cordic 4.1031+-0.0089 ^ 3.9208+-0.0402 ^ definitely 1.0465x faster math-partial-sums 12.5822+-0.1089 ^ 12.2123+-0.1300 ^ definitely 1.0303x faster math-spectral-norm 3.2215+-0.0147 ^ 2.7153+-0.0182 ^ definitely 1.1864x faster regexp-dna 12.5889+-0.1631 ? 12.7145+-0.1816 ? string-base64 5.0933+-0.0910 5.0118+-0.0700 might be 1.0163x faster string-fasta 10.7299+-0.0893 10.7248+-0.1361 string-tagcloud 14.3621+-0.1760 14.2318+-0.1842 string-unpack-code 28.3396+-0.1400 ^ 27.6466+-0.1300 ^ definitely 1.0251x faster string-validate-input 7.1897+-0.1548 ^ 6.5280+-0.0847 ^ definitely 1.1014x faster <arithmetic> * 7.9571+-0.0665 ^ 7.3185+-0.0530 ^ definitely 1.0873x faster <geometric> 6.4260+-0.0464 ^ 5.8174+-0.0378 ^ definitely 1.1046x faster <harmonic> 5.2093+-0.0267 ^ 4.7217+-0.0327 ^ definitely 1.1033x faster TipOfTree CoCo V8Spider: crypto 252.3019+-6.3185 ^ 242.0084+-0.3358 ^ definitely 1.0425x faster deltablue 127.7899+-0.8304 ^ 102.0635+-0.6066 ^ definitely 1.2521x faster earley-boyer 84.0409+-0.2748 ^ 69.8289+-0.3471 ^ definitely 1.2035x faster raytrace 63.6212+-0.3199 ^ 38.1306+-0.1315 ^ definitely 1.6685x faster regexp 102.3147+-0.5922 ^ 99.8834+-0.3194 ^ definitely 1.0243x faster richards 118.7354+-0.2458 ^ 112.7906+-0.8365 ^ definitely 1.0527x faster splay 48.7566+-0.2879 ^ 47.2623+-0.5742 ^ definitely 1.0316x faster <arithmetic> 113.9372+-0.9922 ^ 101.7097+-0.1476 ^ definitely 1.1202x faster <geometric> * 100.2889+-0.4861 ^ 86.0756+-0.1392 ^ definitely 1.1651x faster <harmonic> 89.6062+-0.2913 ^ 74.0569+-0.1679 ^ definitely 1.2100x faster TipOfTree CoCo Octane and V8v7: encrypt 1.49828+-0.00917 1.49704+-0.00886 decrypt 28.21896+-0.04017 28.14959+-0.14971 deltablue x2 0.57479+-0.00275 ? 0.57546+-0.00286 ? earley 0.88702+-0.00528 ? 0.89320+-0.00329 ? boyer 12.76316+-0.02460 ? 12.81662+-0.04635 ? raytrace x2 4.43644+-0.00628 ! 4.47048+-0.02746 ! definitely 1.0077x slower regexp x2 32.48712+-0.13902 ? 32.61359+-0.18111 ? richards x2 0.30402+-0.00062 ! 0.31225+-0.00339 ! definitely 1.0271x slower splay x2 0.61994+-0.00764 ? 0.62812+-0.01887 ? might be 1.0132x slower navier-stokes x2 10.78609+-0.01360 10.77549+-0.00675 closure 0.31899+-0.03415 ? 0.31931+-0.03499 ? jquery 4.49077+-0.55383 4.47929+-0.54674 gbemu x2 264.59222+-10.11509 258.36364+-12.90555 might be 1.0241x faster box2d x2 33.82537+-0.14673 33.74638+-0.16887 V8v7: <arithmetic> 8.86151+-0.02077 ? 8.88170+-0.02389 ? might be 1.0023x slower <geometric> * 2.79117+-0.00590 ! 2.81027+-0.01164 ! definitely 1.0068x slower <harmonic> 1.00681+-0.00232 ! 1.02130+-0.00737 ! definitely 1.0144x slower Octane including V8v7: <arithmetic> 33.79223+-0.94658 33.23299+-1.19610 might be 1.0168x faster <geometric> * 4.89788+-0.06747 ? 4.90984+-0.07650 ? might be 1.0024x slower <harmonic> 1.13487+-0.02029 ? 1.14839+-0.02437 ? might be 1.0119x slower TipOfTree CoCo Kraken: ai-astar 494.467+-0.810 ^ 492.339+-0.497 ^ definitely 1.0043x faster audio-beat-detection 1201.365+-6.093 ? 1251.031+-68.854 ? might be 1.0413x slower audio-dft 310.148+-0.726 ? 310.679+-0.772 ? audio-fft 1138.103+-13.501 ? 1194.025+-59.244 ? might be 1.0491x slower audio-oscillator 234.181+-1.042 233.800+-0.967 imaging-darkroom 290.995+-1.136 290.596+-0.884 imaging-desaturate 849.050+-3.729 ? 849.382+-4.051 ? imaging-gaussian-blur 403.118+-0.605 ? 404.140+-2.560 ? json-parse-financial 81.772+-0.193 ^ 80.663+-0.203 ^ definitely 1.0137x faster json-stringify-tinderbox 99.190+-0.263 ? 99.664+-0.355 ? stanford-crypto-aes 183.103+-0.577 ^ 181.488+-0.813 ^ definitely 1.0089x faster stanford-crypto-ccm 214.439+-1.003 191.840+-25.989 might be 1.1178x faster stanford-crypto-pbkdf2 269.982+-1.299 ^ 265.512+-2.326 ^ definitely 1.0168x faster stanford-crypto-sha256-iterative 115.778+-0.193 115.552+-0.124 <arithmetic> * 420.406+-1.448 ? 425.765+-9.608 ? might be 1.0127x slower <geometric> 300.374+-0.569 298.448+-4.175 might be 1.0065x faster <harmonic> 221.866+-0.313 ^ 218.375+-3.120 ^ definitely 1.0160x faster TipOfTree CoCo JSRegress: adapt-to-double-divide 22.4341+-0.0826 ? 22.5966+-0.1355 ? aliased-arguments-getbyval 0.9417+-0.0102 ^ 0.8408+-0.0091 ^ definitely 1.1200x faster allocate-big-object 2.5568+-0.0350 2.5014+-0.0249 might be 1.0222x faster arity-mismatch-inlining 0.7822+-0.0136 ? 0.7833+-0.0211 ? array-access-polymorphic-structure 7.2140+-0.1301 ^ 6.7576+-0.0936 ^ definitely 1.0675x faster array-with-double-add 5.7864+-0.1013 ? 5.8422+-0.0693 ? array-with-double-increment 4.0798+-0.0531 ? 4.2154+-0.0862 ? might be 1.0332x slower array-with-double-mul-add 7.1538+-0.0799 ? 7.1720+-0.0855 ? array-with-double-sum 7.8619+-0.0979 ? 7.8726+-0.0719 ? array-with-int32-add-sub 10.6305+-0.1364 10.5590+-0.0803 array-with-int32-or-double-sum 7.9269+-0.1150 ? 8.0773+-0.1024 ? might be 1.0190x slower big-int-mul 4.8795+-0.0737 4.8129+-0.0411 might be 1.0138x faster boolean-test 4.4385+-0.0589 4.3904+-0.0627 might be 1.0110x faster cast-int-to-double 13.8579+-0.0780 ? 13.9432+-0.1146 ? cell-argument 14.3010+-0.0312 ! 14.4514+-0.0801 ! definitely 1.0105x slower cfg-simplify 4.0578+-0.0474 4.0042+-0.0364 might be 1.0134x faster cmpeq-obj-to-obj-other 12.2170+-0.3348 ? 12.3393+-0.3465 ? might be 1.0100x slower constant-test 8.5006+-0.1432 ? 8.5705+-0.1229 ? direct-arguments-getbyval 0.8609+-0.0087 ^ 0.7323+-0.0073 ^ definitely 1.1756x faster double-pollution-getbyval 10.7288+-0.1148 ? 10.8311+-0.1005 ? double-pollution-putbyoffset 5.1052+-0.1032 4.9191+-0.0994 might be 1.0378x faster empty-string-plus-int 10.8354+-0.1787 ? 10.8436+-0.1678 ? external-arguments-getbyval 2.2258+-0.0301 ^ 1.9093+-0.0095 ^ definitely 1.1658x faster external-arguments-putbyval 3.3275+-0.0165 3.2827+-0.1412 might be 1.0137x faster Float32Array-matrix-mult 14.2371+-0.0983 ^ 13.7713+-0.2161 ^ definitely 1.0338x faster fold-double-to-int 21.9802+-0.3169 ? 22.0115+-0.2034 ? function-dot-apply 3.2082+-0.0135 ^ 3.1038+-0.0120 ^ definitely 1.0336x faster function-test 5.1082+-0.0752 5.0986+-0.0716 get-by-id-chain-from-try-block 6.8377+-0.1457 6.7745+-0.1775 HashMap-put-get-iterate-keys 94.8348+-0.8266 93.9742+-1.9139 HashMap-put-get-iterate 98.8796+-1.1131 97.1083+-0.9762 might be 1.0182x faster HashMap-string-put-get-iterate 74.8393+-0.4343 ^ 71.7546+-0.4862 ^ definitely 1.0430x faster indexed-properties-in-objects 4.5585+-0.0447 ^ 4.3592+-0.0602 ^ definitely 1.0457x faster inline-arguments-access 1.2686+-0.0085 ! 1.2946+-0.0073 ! definitely 1.0205x slower inline-arguments-local-escape 22.8639+-0.1066 ? 23.0111+-0.2137 ? inline-get-scoped-var 6.5241+-0.0904 ? 6.5947+-0.0744 ? might be 1.0108x slower inlined-put-by-id-transition 16.9741+-0.3131 ^ 16.3867+-0.1678 ^ definitely 1.0358x faster int-or-other-abs-then-get-by-val 8.9580+-0.1331 8.8304+-0.0711 might be 1.0145x faster int-or-other-abs-zero-then-get-by-val 37.2762+-0.2962 37.0678+-0.2655 int-or-other-add-then-get-by-val 10.1959+-0.0883 10.1918+-0.0833 int-or-other-add 10.4838+-0.0927 10.4806+-0.0976 int-or-other-div-then-get-by-val 7.9154+-0.1208 7.7420+-0.0836 might be 1.0224x faster int-or-other-max-then-get-by-val 10.0459+-0.1317 9.9063+-0.1830 might be 1.0141x faster int-or-other-min-then-get-by-val 8.1799+-0.1150 8.0404+-0.0963 might be 1.0174x faster int-or-other-mod-then-get-by-val 8.0952+-0.1000 ^ 7.8596+-0.0957 ^ definitely 1.0300x faster int-or-other-mul-then-get-by-val 7.1842+-0.1142 7.0712+-0.0996 might be 1.0160x faster int-or-other-neg-then-get-by-val 8.0313+-0.0966 7.9997+-0.1039 int-or-other-neg-zero-then-get-by-val 36.5081+-0.1659 ? 36.5956+-0.3472 ? int-or-other-sub-then-get-by-val 10.2755+-0.0902 10.2391+-0.1035 int-or-other-sub 8.2252+-0.0955 8.2185+-0.1083 int-overflow-local 12.8977+-0.0772 12.7714+-0.1342 Int16Array-bubble-sort 49.4208+-0.1271 49.1982+-0.1818 Int16Array-load-int-mul 1.8965+-0.0104 ! 1.9175+-0.0086 ! definitely 1.0111x slower Int8Array-load 4.8578+-0.0608 4.8493+-0.0222 integer-divide 15.1363+-0.0848 15.0918+-0.0723 integer-modulo 2.1610+-0.0169 ^ 2.0387+-0.0145 ^ definitely 1.0600x faster make-indexed-storage 4.1687+-0.0329 ^ 3.7362+-0.0354 ^ definitely 1.1158x faster method-on-number 25.9305+-0.4858 25.5215+-0.5658 might be 1.0160x faster nested-function-parsing-random 382.7938+-13.2770 382.5710+-13.3033 nested-function-parsing 48.0004+-1.1081 47.6056+-0.9559 new-array-buffer-dead 3.6478+-0.0126 ? 3.6654+-0.0320 ? new-array-buffer-push 10.4742+-0.1599 10.3453+-0.1926 might be 1.0125x faster new-array-dead 28.2037+-0.0834 ? 28.3341+-0.0778 ? new-array-push 6.9144+-0.1146 6.8513+-0.0614 number-test 4.2870+-0.0863 ? 4.3301+-0.0699 ? might be 1.0101x slower object-closure-call 8.4225+-0.1019 ! 10.5617+-1.9059 ! definitely 1.2540x slower object-test 4.8792+-0.0772 ? 4.9431+-0.0602 ? might be 1.0131x slower poly-stricteq 970.9461+-2.8302 ! 998.5610+-12.9388 ! definitely 1.0284x slower polymorphic-structure 19.9364+-0.0806 ! 21.2821+-0.6624 ! definitely 1.0675x slower polyvariant-monomorphic-get-by-id 12.4956+-0.0837 ? 12.5439+-0.0887 ? rare-osr-exit-on-local 20.4981+-0.1298 20.4968+-0.0769 register-pressure-from-osr 31.5793+-0.1081 ^ 31.3128+-0.1067 ^ definitely 1.0085x faster simple-activation-demo 34.4835+-0.1380 ^ 34.2321+-0.0932 ^ definitely 1.0073x faster slow-array-profile-convergence 4.4407+-0.0591 ^ 4.1033+-0.0225 ^ definitely 1.0822x faster slow-convergence 3.8817+-0.0047 ^ 3.4549+-0.0263 ^ definitely 1.1235x faster sparse-conditional 1.3439+-0.0099 1.3207+-0.0219 might be 1.0175x faster splice-to-remove 50.2737+-0.2501 ^ 49.5566+-0.1260 ^ definitely 1.0145x faster string-concat-object 2.7306+-0.0457 2.7267+-0.0512 string-concat-pair-object 2.7407+-0.0789 2.6530+-0.0241 might be 1.0331x faster string-concat-pair-simple 16.8871+-0.3057 ? 16.9288+-0.2450 ? string-concat-simple 17.0945+-0.2832 ? 17.2314+-0.2649 ? string-cons-repeat 10.0806+-0.0248 ! 10.1844+-0.0261 ! definitely 1.0103x slower string-cons-tower 10.9453+-0.0229 ^ 10.8847+-0.0327 ^ definitely 1.0056x faster string-equality 108.0721+-1.9209 ^ 105.1292+-0.8394 ^ definitely 1.0280x faster string-hash 2.6790+-0.0128 ^ 2.5520+-0.0106 ^ definitely 1.0497x faster string-repeat-arith 43.9696+-0.4475 ? 44.3525+-0.1472 ? string-sub 85.6510+-1.0897 ! 87.2145+-0.3700 ! definitely 1.0183x slower string-test 4.2977+-0.0257 4.2428+-0.0657 might be 1.0129x faster structure-hoist-over-transitions 3.3313+-0.0247 ^ 3.1696+-0.0227 ^ definitely 1.0510x faster tear-off-arguments-simple 1.8326+-0.0105 ^ 1.7574+-0.0085 ^ definitely 1.0428x faster tear-off-arguments 3.4296+-0.0153 ^ 3.1928+-0.0189 ^ definitely 1.0742x faster temporal-structure 21.2194+-0.5357 21.0447+-0.1684 to-int32-boolean 30.6567+-0.2084 ? 30.7856+-0.2886 ? undefined-test 4.5862+-0.0360 4.4873+-0.0853 might be 1.0220x faster <arithmetic> 30.3231+-0.1327 ? 30.5050+-0.1677 ? might be 1.0060x slower <geometric> * 9.7065+-0.0245 ^ 9.5811+-0.0356 ^ definitely 1.0131x faster <harmonic> 5.2523+-0.0209 ^ 5.0702+-0.0228 ^ definitely 1.0359x faster TipOfTree CoCo All benchmarks: <arithmetic> 64.3186+-0.2581 64.1813+-0.8267 might be 1.0021x faster <geometric> 12.3018+-0.0546 ^ 11.9361+-0.0614 ^ definitely 1.0306x faster <harmonic> 3.8423+-0.0339 ^ 3.7581+-0.0380 ^ definitely 1.0224x faster TipOfTree CoCo Geomean of preferred means: <scaled-result> 27.5761+-0.1454 ^ 26.3109+-0.1840 ^ definitely 1.0481x faster (In reply to comment #11) > (In reply to comment #10) > > (From update of attachment 202380 [details] [details]) > > View in context: https://bugs.webkit.org/attachment.cgi?id=202380&action=review > > > > r=me > > > > > Source/JavaScriptCore/jit/JITStubs.cpp:959 > > > + worklistState = > > > + stackFrame.vm->worklist->completeAllReadyPlansForVM(*stackFrame.vm, codeBlock); > > > > This seems OK for now, but it might become a problem if LLVM compilation sticks at ~4ms / function. If you had 4 functions queued up, one call to cti_optimize would cause you to drop an animation frame. If you had 16 functions queued up, you'd drop 4 frames. Basically, there's a tradeoff here between startup time and responsiveness, and the nob is turned all the way toward startup time right now, which is probably the wrong balance. > > No it won't, there's no trade-off. This method, "complete all ready [sic] plans for VM", does not wait for the compilation thread. It just completes the ones that are done. The only waiting it does is on the Worklist::m_lock, which is only held when modifying the queue/map/finished data structures. Those modifications are pretty quick - at worst, grab the lock, loop over the list, release the lock. The compilation thread only holds the lock for a very short time - at worst to dequeue or to append. > > This is distinct from completeAllPlansForVM(), which does wait for the compilation thread to finish. > > > > > > Source/JavaScriptCore/runtime/JSSegmentedVariableObject.h:97 > > > + typedef ConcurrentJITLock Lock; > > > + typedef ConcurrentJITLocker Locker; > > > > Actually, I think this reads even better without the typedef, since it calls out the other client you're locking against. > > Do you advocate making this change throughout? There are a bunch of places that use the typedef-ConcurrentJITLock-to-Lock idiom. I kind of like how this makes most code terse but still lets you find what the typedef means. > > But I kind of like the change you're proposing. Since that would be a larger patch (all of the places that use ConcurrentJITLock, and there are a bunch of them), how about I do it in a separate bug? Filed: https://bugs.webkit.org/show_bug.cgi?id=116561 > > But I kind of like the change you're proposing. Since that would be a larger patch (all of the places that use ConcurrentJITLock, and there are a bunch of them), how about I do it in a separate bug?
Sounds great!
Landed in http://trac.webkit.org/changeset/150465 |