67554 – DFG variable predictions only work for local variables, not temporaries

RESOLVED FIXED 67554

DFG variable predictions only work for local variables, not temporaries

https://bugs.webkit.org/show_bug.cgi?id=67554

Summary DFG variable predictions only work for local variables, not temporaries

Filip Pizlo

Reported 2011-09-02 23:14:29 PDT

The local variable prediction code in DFG::Graph does not work for temporaries, even though it easily could, and probably should. This causes speculation logic to be overly conservative if a temporary is live across multiple basic blocks.

Attachments
the patch (1.84 KB, patch) 2011-09-02 23:18 PDT, Filip Pizlo	no flags	Details Formatted Diff Diff
View All Add attachment proposed patch, testcase, etc.

Filip Pizlo

Comment 1 2011-09-02 23:18:14 PDT

Created attachment 106247 [details] the patch

Filip Pizlo

Comment 2 2011-09-02 23:21:39 PDT

This is the performance: Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc "ResizeVarPreds" at /Volumes/Data/pizlo/septenary/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 ResizeVarPreds SunSpider: 3d-cube 8.0769+-0.2633 7.8421+-0.1866 might be 1.0299x faster 3d-morph 7.9217+-0.1421 ^ 7.6166+-0.1522 ^ definitely 1.0401x faster 3d-raytrace 7.5407+-0.2822 ? 7.5764+-0.1596 ? access-binary-trees 2.3041+-0.0434 2.2708+-0.0502 might be 1.0147x faster access-fannkuch 11.9338+-0.2432 11.9160+-0.2582 access-nbody 4.2317+-0.0457 ? 4.3202+-0.0630 ? might be 1.0209x slower access-nsieve 2.6031+-0.1330 2.5357+-0.0514 might be 1.0266x faster bitops-3bit-bits-in-byte 1.7767+-0.0823 1.7109+-0.0568 might be 1.0385x faster bitops-bits-in-byte 4.5004+-0.1841 ? 4.6038+-0.2552 ? might be 1.0230x slower bitops-bitwise-and 3.7526+-0.1117 3.7345+-0.0547 bitops-nsieve-bits 5.4707+-0.1478 ? 5.5343+-0.1609 ? might be 1.0116x slower controlflow-recursive 2.0275+-0.0386 ? 2.1304+-0.0970 ? might be 1.0507x slower crypto-aes 6.5957+-0.1562 6.5791+-0.1279 crypto-md5 2.7703+-0.0633 2.7343+-0.0570 might be 1.0132x faster crypto-sha1 2.3163+-0.0783 2.2194+-0.0506 might be 1.0437x faster date-format-tofte 10.3854+-0.2497 ? 10.3987+-0.3102 ? date-format-xparb 8.8015+-0.2505 8.7822+-0.2139 math-cordic 6.4468+-0.0492 ? 6.4918+-0.0753 ? math-partial-sums 8.1159+-0.2666 7.8487+-0.2188 might be 1.0340x faster math-spectral-norm 2.5621+-0.0720 2.5355+-0.0520 might be 1.0105x faster regexp-dna 10.3926+-0.0760 ? 10.4463+-0.1122 ? string-base64 6.1697+-0.2178 ? 6.2009+-0.2772 ? string-fasta 7.5724+-0.1146 7.4252+-0.1236 might be 1.0198x faster string-tagcloud 12.3140+-0.4933 12.2323+-0.2761 string-unpack-code 18.9475+-0.4868 ? 19.3867+-0.4740 ? might be 1.0232x slower string-validate-input 7.4412+-0.2151 ? 7.4739+-0.2596 ? <arithmetic> 6.6527+-0.0453 6.6364+-0.0350 <geometric> 5.5332+-0.0355 5.5064+-0.0282 <harmonic> 4.5285+-0.0383 4.4957+-0.0263 TipOfTree ResizeVarPreds V8: crypto 92.3651+-0.3348 92.3497+-0.5405 deltablue 144.0207+-0.8783 143.1391+-1.0990 earley-boyer 100.5560+-0.5754 ? 100.5589+-0.5720 ? raytrace 52.9002+-0.3640 52.4050+-0.3120 regexp 111.1890+-0.4171 ? 111.7704+-0.8885 ? richards 246.2668+-0.7304 ? 249.7471+-4.0842 ? might be 1.0141x slower splay 104.8678+-0.6789 ? 104.9036+-0.6532 ? <arithmetic> 121.7380+-0.2460 ? 122.1249+-0.5218 ? <geometric> 110.6561+-0.2208 ? 110.7121+-0.2636 ? <harmonic> 101.1378+-0.2360 100.9628+-0.2184 TipOfTree ResizeVarPreds Kraken: ai-astar 1117.1826+-7.3669 1107.8483+-4.7591 audio-beat-detection 479.5615+-1.5012 ? 481.1826+-1.6781 ? audio-dft 437.8244+-5.0489 ? 439.3640+-3.1897 ? audio-fft 373.5425+-0.4485 ? 373.8119+-1.7038 ? audio-oscillator 386.2704+-2.5483 385.2089+-1.5145 imaging-darkroom 560.4767+-1.9983 ^ 553.5504+-1.7509 ^ definitely 1.0125x faster imaging-desaturate 601.5663+-7.9396 600.6047+-6.4242 imaging-gaussian-blur 1756.0249+-3.8117 ? 1760.0847+-4.2614 ? json-parse-financial 48.0583+-0.4829 ? 48.1131+-0.2504 ? json-stringify-tinderbox 73.4140+-0.3952 ? 73.4151+-0.3149 ? stanford-crypto-aes 148.1695+-2.6645 146.9939+-0.8603 stanford-crypto-ccm 118.6461+-4.4542 ^ 113.2232+-0.5431 ^ definitely 1.0479x faster stanford-crypto-pbkdf2 351.8983+-2.5204 348.3458+-1.5261 might be 1.0102x faster stanford-crypto-sha256-iterative 134.3091+-0.8045 ^ 132.5642+-0.6250 ^ definitely 1.0132x faster <arithmetic> 470.4960+-1.0338 ^ 468.8793+-0.5025 ^ definitely 1.0034x faster <geometric> 304.3833+-1.0155 ^ 302.4315+-0.3470 ^ definitely 1.0065x faster <harmonic> 187.9824+-0.9233 ^ 186.6012+-0.3609 ^ definitely 1.0074x faster TipOfTree ResizeVarPreds All benchmarks: <arithmetic> 161.9592+-0.3159 161.5262+-0.1479 <geometric> 28.5211+-0.1086 28.3923+-0.0794 <harmonic> 7.9863+-0.0660 7.9289+-0.0452

Filip Pizlo

Comment 3 2011-09-02 23:33:44 PDT

Comment on attachment 106247 [details] the patch Passes tests.

WebKit Review Bot

Comment 4 2011-09-03 00:48:21 PDT

Comment on attachment 106247 [details] the patch Clearing flags on attachment: 106247 Committed r94484: <http://trac.webkit.org/changeset/94484>

WebKit Review Bot

Comment 5 2011-09-03 00:48:25 PDT

All reviewed patches have been landed. Closing bug.

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

2011-09-02 23:14 PDT

Modified

2011-09-03 00:48 PDT History

CC List

1 user Show

URL

Keywords

Depends on

Blocks