67920 – JavaScriptCore does not have baseline->speculative OSR

RESOLVED FIXED Bug 67920

JavaScriptCore does not have baseline->speculative OSR

https://bugs.webkit.org/show_bug.cgi?id=67920

Summary JavaScriptCore does not have baseline->speculative OSR

Filip Pizlo

Reported 2011-09-11 21:17:38 PDT

Tiered compilation in JavaScriptCore only produces a benefit if the code being optimized does most of its work in functions that get called repeatedly. Hot loops in functions that don't return for a long time continue to execute old JIT code. Fixing that requires on-stack-replacement from baseline (old JIT) code to speculative (DFG JIT) code.

Attachments
work in progress (140.28 KB, patch) 2011-09-11 21:19 PDT, Filip Pizlo	no flags	Details Formatted Diff Diff
the patch (80.19 KB, patch) 2011-09-12 19:01 PDT, Filip Pizlo	no flags	Details Formatted Diff Diff
the patch - fix style (80.50 KB, patch) 2011-09-12 19:09 PDT, Filip Pizlo	no flags	Details Formatted Diff Diff
the patch - fix style (80.50 KB, patch) 2011-09-12 19:12 PDT, Filip Pizlo	webkit-ews: commit-queue-	Details Formatted Diff Diff
the patch - fix build (80.08 KB, patch) 2011-09-12 19:30 PDT, Filip Pizlo	oliver: review+ fpizlo: commit-queue-	Details Formatted Diff Diff
Show Obsolete (4) View All Add attachment proposed patch, testcase, etc.

Filip Pizlo

Comment 1 2011-09-11 21:19:21 PDT

Created attachment 107019 [details] work in progress Still working on it, but it's starting to work.

Filip Pizlo

Comment 2 2011-09-12 01:03:37 PDT

It appears that with this patch, tiered compilation is a pure win, and is probably good enough to enable in ToT. Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc "OSRExit" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc "OSREnter" at /Volumes/Data/pizlo/octonary/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 OSRExit OSREnter OSREnter v. TipOfTree SunSpider: 3d-cube 8.6538+-0.2445 ! 9.8197+-0.2429 9.7211+-0.2500 ! definitely 1.1233x slower 3d-morph 7.4333+-0.1905 ? 7.5068+-0.2221 7.4015+-0.1159 3d-raytrace 8.0800+-0.2070 ? 8.1327+-0.1275 8.0972+-0.1711 ? access-binary-trees 3.3529+-0.0715 3.3493+-0.0639 ? 3.4034+-0.1113 ? might be 1.0151x slower access-fannkuch 11.7457+-0.1979 ? 11.8288+-0.2726 11.5642+-0.2437 might be 1.0157x faster access-nbody 4.2957+-0.0658 4.2564+-0.1403 4.1282+-0.0654 ^ definitely 1.0406x faster access-nsieve 2.5446+-0.0719 ! 2.8076+-0.0829 ^ 2.6584+-0.0650 ? might be 1.0447x slower bitops-3bit-bits-in-byte 1.7152+-0.0510 ! 1.8190+-0.0393 1.7013+-0.0950 bitops-bits-in-byte 3.3163+-0.0906 ! 5.3742+-0.1637 ^ 3.1249+-0.0470 ^ definitely 1.0613x faster bitops-bitwise-and 3.6230+-0.1118 3.6229+-0.0979 3.5523+-0.1120 might be 1.0199x faster bitops-nsieve-bits 5.4752+-0.1264 5.2639+-0.1630 ? 5.2954+-0.1245 might be 1.0340x faster controlflow-recursive 1.9879+-0.0458 ? 2.0770+-0.0553 1.9895+-0.0495 ? crypto-aes 7.6468+-0.2855 ? 8.0935+-0.3318 7.6487+-0.1677 ? crypto-md5 2.8797+-0.1083 2.8015+-0.0864 ? 2.8928+-0.1564 ? crypto-sha1 2.2917+-0.0576 ^ 2.1669+-0.0464 ? 2.2158+-0.0414 might be 1.0343x faster date-format-tofte 12.4361+-0.2173 12.3944+-0.3561 ? 12.4613+-0.3561 ? date-format-xparb 10.7844+-0.2669 ? 10.8425+-0.3045 10.6484+-0.2471 might be 1.0128x faster math-cordic 6.4941+-0.1730 6.2070+-0.1304 ? 6.2946+-0.1683 might be 1.0317x faster math-partial-sums 7.5474+-0.1556 7.4442+-0.2342 7.4153+-0.1500 might be 1.0178x faster math-spectral-norm 2.4813+-0.0507 ! 2.6118+-0.0580 2.5514+-0.0353 ? might be 1.0283x slower regexp-dna 10.7326+-0.1828 ? 10.7935+-0.2052 ? 10.9701+-0.3206 ? might be 1.0221x slower string-base64 7.9637+-0.2596 7.8483+-0.2541 ? 7.9952+-0.2541 ? string-fasta 8.5115+-0.3501 ! 9.3982+-0.2687 9.2106+-0.2616 ! definitely 1.0821x slower string-tagcloud 12.6699+-0.3420 ? 12.9489+-0.4807 12.6656+-0.3720 string-unpack-code 20.9736+-0.5323 ? 21.0942+-0.6218 20.7491+-0.4228 might be 1.0108x faster string-validate-input 9.4274+-0.3451 ^ 8.6838+-0.3201 ? 8.8235+-0.3177 might be 1.0684x faster <arithmetic> 7.1178+-0.0319 ! 7.2764+-0.0511 ^ 7.1223+-0.0569 ? <geometric> 5.8095+-0.0246 ! 5.9878+-0.0418 ^ 5.8014+-0.0390 <harmonic> 4.6563+-0.0306 ! 4.8303+-0.0449 ^ 4.6390+-0.0377 TipOfTree OSRExit OSREnter OSREnter v. TipOfTree V8: crypto 91.3270+-0.5783 ^ 84.8944+-0.7944 ? 85.1714+-0.5082 ^ definitely 1.0723x faster deltablue 280.4096+-1.6747 ^ 267.4857+-2.5117 ? 271.4189+-1.8715 ^ definitely 1.0331x faster earley-boyer 107.2856+-1.4692 106.8650+-0.3474 106.8528+-0.7143 raytrace 87.1412+-0.5858 ! 88.5495+-0.4171 ^ 87.3666+-0.6869 ? regexp 116.8110+-0.5595 ? 117.8736+-0.9676 116.5011+-1.3460 richards 228.9993+-0.9538 ^ 215.3427+-0.2853 ! 222.4097+-1.0724 ^ definitely 1.0296x faster splay 179.7157+-1.0183 ? 181.3913+-0.7650 180.0017+-0.9076 ? <arithmetic> 155.9556+-0.4211 ^ 151.7718+-0.4730 ! 152.8175+-0.2995 ^ definitely 1.0205x faster <geometric> 141.7613+-0.4067 ^ 138.7326+-0.4030 ? 139.0711+-0.2878 ^ definitely 1.0193x faster <harmonic> 130.0867+-0.4418 ^ 127.7069+-0.4133 127.5706+-0.3208 ^ definitely 1.0197x faster TipOfTree OSRExit OSREnter OSREnter v. TipOfTree Kraken: ai-astar 1085.3365+-5.9033 ! 1118.8661+-6.9293 1114.5020+-5.7857 ! definitely 1.0269x slower audio-beat-detection 480.7979+-2.1540 474.5989+-6.6917 468.2753+-2.6514 ^ definitely 1.0267x faster audio-dft 415.4874+-2.1583 ! 422.8530+-3.5811 420.1431+-2.9278 ? might be 1.0112x slower audio-fft 369.3367+-1.4110 ^ 365.8327+-1.2752 ^ 361.9488+-1.3125 ^ definitely 1.0204x faster audio-oscillator 377.5589+-1.8711 ^ 357.7255+-1.5519 ^ 352.3107+-1.0252 ^ definitely 1.0717x faster imaging-darkroom 529.9963+-3.4289 ^ 508.8806+-1.1331 ? 512.4397+-7.4576 ^ definitely 1.0343x faster imaging-desaturate 622.7195+-5.3329 ^ 571.2379+-3.0810 ^ 216.0253+-0.5649 ^ definitely 2.8826x faster imaging-gaussian-blur 1722.3304+-6.0498 ! 2271.0239+-4.6600 ^ 1722.6744+-7.9226 ? json-parse-financial 55.3235+-0.9510 55.1895+-0.5653 ? 55.4367+-0.6541 ? json-stringify-tinderbox 69.0833+-0.3308 68.8264+-0.3208 ^ 67.9154+-0.4609 ^ definitely 1.0172x faster stanford-crypto-aes 150.6488+-0.9144 ! 152.5331+-0.9238 ^ 150.7415+-0.4415 ? stanford-crypto-ccm 123.7896+-1.0019 ? 124.2190+-0.7717 ? 124.4001+-0.7065 ? stanford-crypto-pbkdf2 352.5917+-4.7548 ! 407.2319+-2.8103 ? 410.2262+-3.7950 ! definitely 1.1635x slower stanford-crypto-sha256-iterative 133.9557+-0.6957 ! 152.1153+-0.5327 151.8553+-0.4600 ! definitely 1.1336x slower <arithmetic> 463.4969+-0.6681 ! 503.6524+-1.1248 ^ 437.7782+-1.2406 ^ definitely 1.0587x faster <geometric> 304.0857+-0.6485 ! 312.9511+-0.5859 ^ 285.0960+-0.6719 ^ definitely 1.0666x faster <harmonic> 193.3201+-0.9482 ! 196.2345+-0.3049 ^ 187.5629+-0.6077 ^ definitely 1.0307x faster TipOfTree OSRExit OSREnter OSREnter v. TipOfTree All benchmarks: <arithmetic> 165.2279+-0.2263 ! 176.6537+-0.2824 ^ 157.1021+-0.3760 ^ definitely 1.0517x faster <geometric> 30.3929+-0.0773 ! 31.0708+-0.1171 ^ 29.7067+-0.1209 ^ definitely 1.0231x faster <harmonic> 8.2311+-0.0533 ! 8.5317+-0.0776 ^ 8.1964+-0.0654

Filip Pizlo

Comment 3 2011-09-12 19:01:11 PDT

Created attachment 107125 [details] the patch

WebKit Review Bot

Comment 4 2011-09-12 19:03:52 PDT

Attachment 107125 [details] did not pass style-queue: Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/ChangeLog', u'Source..." exit_code: 1 Source/JavaScriptCore/jit/CompactJITCodeMap.h:160: The parameter name "jitCodeMap" adds no information, so it should be removed. [readability/parameter_name] [5] Source/JavaScriptCore/jit/JITStubs.cpp:40: Alphabetical sorting problem. [build/include_order] [4] Source/JavaScriptCore/jit/JITStubs.cpp:1860: Should have only a single space after a punctuation in a comment. [whitespace/comments] [5] Source/JavaScriptCore/bytecode/CodeBlock.h:596: Should have only a single space after a punctuation in a comment. [whitespace/comments] [5] Total errors found: 4 in 27 files If any of these errors are false positives, please file a bug against check-webkit-style.

Filip Pizlo

Comment 5 2011-09-12 19:09:19 PDT

Created attachment 107126 [details] the patch - fix style

WebKit Review Bot

Comment 6 2011-09-12 19:11:45 PDT

Attachment 107126 [details] did not pass style-queue: Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/ChangeLog', u'Source..." exit_code: 1 Source/JavaScriptCore/jit/JITStubs.cpp:1860: Should have only a single space after a punctuation in a comment. [whitespace/comments] [5] Total errors found: 1 in 27 files If any of these errors are false positives, please file a bug against check-webkit-style.

Filip Pizlo

Comment 7 2011-09-12 19:12:51 PDT

Created attachment 107127 [details] the patch - fix style

Filip Pizlo

Comment 8 2011-09-12 19:23:23 PDT

Latest numbers, after merging. TipOfTree: r94996 TipOfTreeDyn: r94996 with tiered compilation enabled OSREntry: r94996 with this patch and tiered compilation enabled Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc "TipOfTreeDyn" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc "OSREntry" at /Volumes/Data/pizlo/octonary/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 TipOfTreeDyn OSREntry OSREntry v. TipOfTree SunSpider: 3d-cube 7.7184+-0.1100 ! 8.9085+-0.1620 ? 8.9214+-0.1325 ! definitely 1.1559x slower 3d-morph 7.4503+-0.1056 ! 7.7622+-0.1635 7.6214+-0.1253 ? might be 1.0230x slower 3d-raytrace 7.4272+-0.1336 ? 7.5459+-0.1919 ? 7.6250+-0.2150 ? might be 1.0266x slower access-binary-trees 2.2947+-0.0509 ? 2.3706+-0.1106 ? 2.4490+-0.0621 ! definitely 1.0672x slower access-fannkuch 11.9483+-0.1975 11.6984+-0.2259 ? 11.9258+-1.0304 access-nbody 4.3891+-0.0669 4.2874+-0.0599 4.2159+-0.0917 ^ definitely 1.0411x faster access-nsieve 2.5623+-0.0878 ! 2.8594+-0.0228 ^ 2.5834+-0.0431 ? bitops-3bit-bits-in-byte 1.7385+-0.0336 ! 1.8276+-0.0365 ^ 1.6624+-0.0317 ^ definitely 1.0458x faster bitops-bits-in-byte 3.2493+-0.0690 ! 5.5825+-0.1160 ^ 3.2049+-0.0624 might be 1.0139x faster bitops-bitwise-and 3.5795+-0.0882 ? 3.6996+-0.0938 3.6282+-0.0949 ? might be 1.0136x slower bitops-nsieve-bits 5.6046+-0.1285 5.3794+-0.1261 ? 5.4322+-0.0559 might be 1.0317x faster controlflow-recursive 2.0711+-0.0439 ^ 1.9854+-0.0307 ? 2.0288+-0.0526 might be 1.0208x faster crypto-aes 6.6405+-0.1539 ! 7.1225+-0.2580 7.0693+-0.2434 ! definitely 1.0646x slower crypto-md5 2.8088+-0.0418 2.7695+-0.0662 ? 2.8314+-0.0446 ? crypto-sha1 2.3172+-0.0678 2.2083+-0.0452 2.1598+-0.0469 ^ definitely 1.0729x faster date-format-tofte 10.0744+-0.2586 ? 10.3379+-0.0644 ^ 10.0557+-0.1534 date-format-xparb 8.6629+-0.1886 ? 8.8287+-0.1612 8.5590+-0.1303 might be 1.0121x faster math-cordic 6.4963+-0.1488 ^ 6.2470+-0.0928 ? 6.3014+-0.0690 might be 1.0309x faster math-partial-sums 7.7424+-0.1417 ^ 7.4190+-0.1262 ? 7.4701+-0.1282 ^ definitely 1.0364x faster math-spectral-norm 2.4844+-0.0485 ! 2.5765+-0.0323 2.5679+-0.0335 ! definitely 1.0336x slower regexp-dna 10.7656+-0.1135 ? 10.9759+-0.1922 ? 11.2254+-0.2329 ! definitely 1.0427x slower string-base64 6.1072+-0.1020 5.8784+-0.1417 5.8057+-0.1375 ^ definitely 1.0519x faster string-fasta 7.5601+-0.1272 ! 8.3642+-0.1315 8.2778+-0.1067 ! definitely 1.0949x slower string-tagcloud 11.9666+-0.2187 ? 12.1449+-0.1819 11.9751+-0.1859 ? string-unpack-code 18.7544+-0.1948 ? 18.7719+-0.1832 18.5954+-0.2264 string-validate-input 7.1293+-0.1916 ^ 6.6377+-0.1308 6.4927+-0.0986 ^ definitely 1.0981x faster <arithmetic> 6.5209+-0.0327 ! 6.6996+-0.0190 ^ 6.5648+-0.0423 ? <geometric> 5.4079+-0.0257 ! 5.5923+-0.0125 ^ 5.4196+-0.0216 ? <harmonic> 4.4293+-0.0268 ! 4.5850+-0.0154 ^ 4.4125+-0.0261 TipOfTree TipOfTreeDyn OSREntry OSREntry v. TipOfTree V8: crypto 93.9862+-1.1192 ^ 85.1213+-0.7738 ! 88.3021+-1.3944 ^ definitely 1.0644x faster deltablue 275.9727+-2.3166 ^ 260.9344+-2.1377 ? 263.8965+-1.5998 ^ definitely 1.0458x faster earley-boyer 95.7358+-1.1461 ? 96.4736+-0.5928 95.6193+-0.8177 raytrace 77.9518+-0.8902 ? 78.7043+-0.8503 ? 79.5421+-1.2368 ? might be 1.0204x slower regexp 108.6761+-0.6543 ? 109.7936+-0.5208 109.4807+-0.9202 ? richards 236.1509+-2.6112 ^ 219.8697+-2.3076 218.7715+-1.1891 ^ definitely 1.0794x faster splay 102.6694+-1.0482 ? 103.6818+-1.0776 ? 104.9136+-1.1978 ? might be 1.0219x slower <arithmetic> 141.5919+-0.5536 ^ 136.3684+-0.5139 ? 137.2180+-0.4638 ^ definitely 1.0319x faster <geometric> 126.1949+-0.5441 ^ 122.8319+-0.3350 ! 123.7718+-0.4899 ^ definitely 1.0196x faster <harmonic> 115.2729+-0.5685 ^ 112.9720+-0.3575 ! 114.0106+-0.5369 ^ definitely 1.0111x faster TipOfTree TipOfTreeDyn OSREntry OSREntry v. TipOfTree Kraken: ai-astar 1132.2278+-9.3623 ? 1139.4513+-10.9565 ! 1181.6296+-13.2666 ! definitely 1.0436x slower audio-beat-detection 485.6322+-4.3795 482.8488+-4.4589 479.7418+-5.5265 might be 1.0123x faster audio-dft 447.0208+-8.0453 439.9793+-11.6465 ? 444.7427+-11.4764 audio-fft 375.8554+-3.4366 ? 377.7592+-2.5715 ^ 371.3852+-3.1369 might be 1.0120x faster audio-oscillator 386.6605+-3.4668 ^ 364.1778+-4.7367 358.4429+-2.7246 ^ definitely 1.0787x faster imaging-darkroom 540.9640+-4.1834 ^ 523.6350+-4.1609 520.7076+-6.5455 ^ definitely 1.0389x faster imaging-desaturate 638.5479+-6.2265 ^ 586.9443+-6.0193 ^ 223.2449+-2.3036 ^ definitely 2.8603x faster imaging-gaussian-blur 1781.5950+-9.5280 ! 2322.6440+-11.0342 ^ 1753.7993+-8.3234 ^ definitely 1.0158x faster json-parse-financial 50.2136+-0.3357 ? 50.6477+-0.6864 49.6699+-0.3513 might be 1.0109x faster json-stringify-tinderbox 70.5986+-0.3327 ? 71.2043+-0.7247 ^ 68.5845+-0.9333 ^ definitely 1.0294x faster stanford-crypto-aes 146.3031+-0.8386 ! 149.2882+-0.9315 147.8578+-0.5439 ! definitely 1.0106x slower stanford-crypto-ccm 113.4581+-1.6943 ? 115.7532+-1.7936 ? 115.7774+-1.2888 ? might be 1.0204x slower stanford-crypto-pbkdf2 342.3863+-3.1070 ! 403.7150+-4.6479 ? 408.9552+-3.7587 ! definitely 1.1944x slower stanford-crypto-sha256-iterative 135.4929+-1.4215 ! 153.7509+-0.6486 152.6394+-1.2845 ! definitely 1.1265x slower <arithmetic> 474.7826+-1.4206 ! 512.9857+-1.9518 ^ 448.3699+-1.8529 ^ definitely 1.0589x faster <geometric> 304.8701+-0.7144 ! 314.2936+-1.0548 ^ 286.2978+-0.7210 ^ definitely 1.0649x faster <harmonic> 188.0597+-0.5032 ! 192.3389+-0.4486 ^ 182.3855+-0.4791 ^ definitely 1.0311x faster TipOfTree TipOfTreeDyn OSREntry OSREntry v. TipOfTree All benchmarks: <arithmetic> 166.1201+-0.4098 ! 176.8206+-0.5782 ^ 157.6253+-0.5166 ^ definitely 1.0539x faster <geometric> 28.7324+-0.0862 ! 29.4183+-0.0425 ^ 28.1519+-0.0594 ^ definitely 1.0206x faster <harmonic> 7.8265+-0.0462 ! 8.0960+-0.0269 ^ 7.7936+-0.0451

Early Warning System Bot

Comment 9 2011-09-12 19:28:17 PDT

Comment on attachment 107127 [details] the patch - fix style Attachment 107127 [details] did not pass qt-ews (qt): Output: http://queues.webkit.org/results/9649293

Filip Pizlo

Comment 10 2011-09-12 19:30:24 PDT

Created attachment 107129 [details] the patch - fix build This now passes all tests with tiering enabled. The web also works. Hopefully I've got all the kinks out for when tiering is disabled, as well.

Filip Pizlo

Comment 11 2011-09-12 22:34:03 PDT

Comment on attachment 107129 [details] the patch - fix build Committing manually.

Filip Pizlo

Comment 12 2011-09-12 22:35:50 PDT

Landed in r95016.

Note You need to log in before you can comment on or make changes to this bug.

Status RESOLVED

Resolution FIXED

Priority P2

Severity Normal

Classification Unclassified

Version 528+ (Nightly build)

Hardware All

OS All

Product WebKit

Component JavaScriptCore

Assignee

Nobody

Reported

2011-09-11 21:17 PDT

Modified

2011-09-13 00:20 PDT History

CC List

1 user Show

URL

Keywords

Depends on

64279

Blocks

Dependancies

tree graph