69322 – DFG should be capable of a broader range of speculations on branch and not

RESOLVED FIXED 69322

DFG should be capable of a broader range of speculations on branch and not

https://bugs.webkit.org/show_bug.cgi?id=69322

Summary DFG should be capable of a broader range of speculations on branch and not

Filip Pizlo

Reported 2011-10-04 00:04:02 PDT

The Branch and LogicalNot nodes may take any JS value. Currently they are only capable of speculating that the value is a boolean. But there are cases where these nodes are used for null checks, among other things. DFG should be able to make the appropriate speculation based on the prediction of the child node.

Attachments
work in progress (27.14 KB, patch) 2011-10-04 01:05 PDT, Filip Pizlo	no flags	Details Formatted Diff Diff
the patch (21.08 KB, patch) 2011-10-04 17:15 PDT, Filip Pizlo	oliver: review+	Details Formatted Diff Diff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.

Filip Pizlo

Comment 1 2011-10-04 01:05:06 PDT

Created attachment 109589 [details] work in progress This boosts V8 performance by a decent amount. Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc "BetterBranch" 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 BetterBranch SunSpider: 3d-cube 7.4862+-0.2303 7.4407+-0.1867 3d-morph 7.4372+-0.2109 7.3692+-0.1329 3d-raytrace 7.5381+-0.2667 7.5366+-0.1559 access-binary-trees 1.7708+-0.0529 1.7450+-0.0765 might be 1.0148x faster access-fannkuch 6.4389+-0.1258 6.3115+-0.0905 might be 1.0202x faster access-nbody 3.4741+-0.0653 ? 3.5484+-0.0954 ? might be 1.0214x slower access-nsieve 2.6006+-0.0750 ? 2.6779+-0.0623 ? might be 1.0297x slower bitops-3bit-bits-in-byte 1.7211+-0.0251 1.7178+-0.0313 bitops-bits-in-byte 2.7789+-0.0697 2.6863+-0.0669 might be 1.0345x faster bitops-bitwise-and 3.5273+-0.1354 3.3771+-0.1045 might be 1.0445x faster bitops-nsieve-bits 5.4515+-0.1457 5.4010+-0.1522 controlflow-recursive 2.0884+-0.0502 2.0741+-0.0493 crypto-aes 6.6547+-0.2610 6.5233+-0.2116 might be 1.0201x faster crypto-md5 2.7261+-0.0626 ? 2.7558+-0.0760 ? might be 1.0109x slower crypto-sha1 2.4994+-0.0769 2.4907+-0.0650 date-format-tofte 9.9591+-0.2039 ? 10.2103+-0.2621 ? might be 1.0252x slower date-format-xparb 9.4223+-0.3017 9.3798+-0.2827 math-cordic 6.3661+-0.0935 ? 6.5290+-0.1993 ? might be 1.0256x slower math-partial-sums 7.6179+-0.1193 ? 7.6684+-0.1775 ? math-spectral-norm 2.7889+-0.0816 ? 2.8738+-0.0771 ? might be 1.0304x slower regexp-dna 10.8016+-0.2973 ? 10.8941+-0.2345 ? string-base64 5.4812+-0.1156 ? 5.5595+-0.2529 ? might be 1.0143x slower string-fasta 6.8691+-0.2140 6.6995+-0.1728 might be 1.0253x faster string-tagcloud 11.7154+-0.3687 ? 11.8161+-0.3781 ? string-unpack-code 21.2979+-0.4546 21.0217+-0.4061 might be 1.0131x faster string-validate-input 6.6042+-0.2719 6.3652+-0.2468 might be 1.0376x faster <arithmetic> * 6.2737+-0.0372 6.2567+-0.0333 <geometric> 5.1451+-0.0318 5.1313+-0.0249 <harmonic> 4.2180+-0.0337 4.2059+-0.0305 TipOfTree BetterBranch V8: crypto 72.0374+-0.3624 ? 72.1185+-0.3399 ? deltablue 214.6783+-1.0184 ? 216.0820+-1.2420 ? earley-boyer 87.7386+-0.8656 ? 88.2202+-1.4012 ? raytrace 61.5332+-0.3629 ^ 57.8440+-0.4749 ^ definitely 1.0638x faster regexp 103.0834+-0.4982 ? 103.4798+-0.5179 ? richards 185.3879+-0.5735 ? 186.2984+-0.8325 ? splay 91.3363+-0.6340 90.5186+-0.5669 <arithmetic> 116.5422+-0.2563 116.3659+-0.1903 <geometric> * 105.5238+-0.2539 ^ 104.7846+-0.1695 ^ definitely 1.0071x faster <harmonic> 96.9026+-0.2719 ^ 95.6197+-0.2223 ^ definitely 1.0134x faster TipOfTree BetterBranch Kraken: ai-astar 491.9563+-4.0277 ? 492.9658+-2.1950 ? audio-beat-detection 190.2766+-0.9526 189.2783+-1.2611 audio-dft 270.8751+-2.6753 270.4489+-2.6225 audio-fft 125.8853+-0.8666 125.3651+-0.4597 audio-oscillator 246.5480+-1.7126 246.3990+-1.6485 imaging-darkroom 418.2428+-1.2805 ? 420.7685+-1.8173 ? imaging-desaturate 224.6155+-0.5200 ! 230.8769+-0.6319 ! definitely 1.0279x slower imaging-gaussian-blur 581.5427+-1.2182 581.2523+-2.1482 json-parse-financial 48.7181+-0.3893 48.4358+-0.3936 json-stringify-tinderbox 68.1482+-0.2456 ^ 67.6601+-0.2085 ^ definitely 1.0072x faster stanford-crypto-aes 130.5927+-1.3596 129.1095+-1.3129 might be 1.0115x faster stanford-crypto-ccm 101.0150+-0.6277 ? 101.9456+-1.5832 ? stanford-crypto-pbkdf2 192.8562+-1.5205 192.3827+-1.4093 stanford-crypto-sha256-iterative 74.6896+-0.2253 ? 77.4587+-2.8886 ? might be 1.0371x slower <arithmetic> * 226.1401+-0.6133 ? 226.7391+-0.6994 ? <geometric> 175.8822+-0.4818 ? 176.3939+-0.7048 ? <harmonic> 136.0794+-0.4157 ? 136.4451+-0.7318 ? TipOfTree BetterBranch All benchmarks: <arithmetic> 88.1888+-0.1780 ? 88.3315+-0.2029 ? <geometric> 23.1031+-0.0880 23.0647+-0.0682 <harmonic> 7.4143+-0.0578 7.3928+-0.0523 TipOfTree BetterBranch Geomean of preferred means: <scaled-result> 53.0982+-0.1242 52.9724+-0.0768

Filip Pizlo

Comment 2 2011-10-04 17:15:39 PDT

Created attachment 109721 [details] the patch Updated performance after merging. Benchmark report for SunSpider, V8, and Kraken. VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc "BetterBranch" 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 BetterBranch SunSpider: 3d-cube 7.4812+-0.1460 ? 7.5067+-0.1737 ? 3d-morph 7.3919+-0.1335 7.3708+-0.1453 3d-raytrace 7.5996+-0.1626 ? 7.7081+-0.1909 ? might be 1.0143x slower access-binary-trees 1.7566+-0.0562 1.7154+-0.0455 might be 1.0240x faster access-fannkuch 6.3631+-0.0820 ? 6.3847+-0.1062 ? access-nbody 3.6076+-0.0756 3.5630+-0.0802 might be 1.0125x faster access-nsieve 2.7525+-0.1543 2.5887+-0.0714 might be 1.0633x faster bitops-3bit-bits-in-byte 1.7318+-0.0244 1.7168+-0.0479 bitops-bits-in-byte 2.7818+-0.0539 2.7355+-0.0740 might be 1.0169x faster bitops-bitwise-and 3.4696+-0.0807 ? 3.9148+-0.5005 ? might be 1.1283x slower bitops-nsieve-bits 5.4481+-0.0925 5.4103+-0.1025 controlflow-recursive 2.0624+-0.0400 2.0529+-0.0372 crypto-aes 6.4679+-0.1676 ? 6.5114+-0.1765 ? crypto-md5 2.7490+-0.0700 ? 2.8099+-0.0835 ? might be 1.0222x slower crypto-sha1 2.4743+-0.0590 2.4562+-0.0486 date-format-tofte 10.1093+-0.2269 9.8557+-0.1560 might be 1.0257x faster date-format-xparb 9.5077+-0.2055 9.2729+-0.1515 might be 1.0253x faster math-cordic 6.3826+-0.1270 6.3111+-0.1125 might be 1.0113x faster math-partial-sums 7.6905+-0.1866 7.4815+-0.1294 might be 1.0279x faster math-spectral-norm 2.7907+-0.0630 ? 2.8314+-0.0866 ? might be 1.0146x slower regexp-dna 10.7219+-0.2144 10.7126+-0.1027 string-base64 5.5707+-0.1089 5.4329+-0.0964 might be 1.0254x faster string-fasta 6.5750+-0.1426 ? 6.6665+-0.1379 ? might be 1.0139x slower string-tagcloud 11.6204+-0.1419 ? 11.8170+-0.1822 ? might be 1.0169x slower string-unpack-code 21.1400+-0.3545 20.9919+-0.2841 string-validate-input 6.2364+-0.1875 ? 6.2398+-0.1595 ? <arithmetic> * 6.2493+-0.0159 6.2330+-0.0383 <geometric> 5.1367+-0.0177 5.1242+-0.0332 <harmonic> 4.2210+-0.0260 4.2030+-0.0362 TipOfTree BetterBranch V8: crypto 72.8145+-0.5199 72.6473+-0.4037 deltablue 214.1124+-1.0016 ! 217.9888+-1.8622 ! definitely 1.0181x slower earley-boyer 88.2514+-0.5703 ^ 87.2301+-0.2962 ^ definitely 1.0117x faster raytrace 62.1892+-0.5513 ^ 58.1012+-0.4594 ^ definitely 1.0704x faster regexp 102.9545+-0.5633 ? 104.1702+-0.7198 ? might be 1.0118x slower richards 187.3573+-0.4727 ? 188.5787+-0.8794 ? splay 90.9779+-0.8013 90.6149+-1.0160 <arithmetic> 116.9510+-0.1704 ? 117.0473+-0.2610 ? <geometric> * 105.9743+-0.1602 ^ 105.2228+-0.1759 ^ definitely 1.0071x faster <harmonic> 97.4046+-0.1800 ^ 95.9244+-0.2019 ^ definitely 1.0154x faster TipOfTree BetterBranch Kraken: ai-astar 495.4805+-2.3816 ! 502.2475+-3.9661 ! definitely 1.0137x slower audio-beat-detection 190.5182+-0.9920 ? 191.6489+-1.1051 ? audio-dft 270.7026+-2.4880 269.3619+-3.1474 audio-fft 125.9513+-0.6101 ? 126.4527+-0.9342 ? audio-oscillator 247.7809+-1.8248 247.3469+-1.9099 imaging-darkroom 427.8471+-7.5061 426.0058+-2.2012 imaging-desaturate 225.0271+-1.6814 ! 232.7494+-1.0558 ! definitely 1.0343x slower imaging-gaussian-blur 585.0237+-2.6857 583.4547+-3.3415 json-parse-financial 49.2309+-0.2701 ^ 48.5261+-0.4178 ^ definitely 1.0145x faster json-stringify-tinderbox 68.9585+-0.7374 ? 70.1904+-1.0000 ? might be 1.0179x slower stanford-crypto-aes 130.2263+-1.5949 ? 132.4506+-1.9081 ? might be 1.0171x slower stanford-crypto-ccm 100.7143+-0.7437 ! 103.0105+-1.1431 ! definitely 1.0228x slower stanford-crypto-pbkdf2 190.2549+-1.9168 ? 190.8994+-2.4199 ? stanford-crypto-sha256-iterative 76.2882+-1.1701 75.8700+-1.1122 <arithmetic> * 227.4289+-0.7199 ? 228.5868+-0.4489 ? <geometric> 176.7279+-0.5544 ! 177.7903+-0.3950 ! definitely 1.0060x slower <harmonic> 136.9439+-0.4958 ? 137.5425+-0.4616 ? TipOfTree BetterBranch All benchmarks: <arithmetic> 88.6201+-0.2244 ! 88.9703+-0.1221 ! definitely 1.0040x slower <geometric> 23.1301+-0.0572 23.1154+-0.0733 <harmonic> 7.4206+-0.0446 7.3890+-0.0619 TipOfTree BetterBranch Geomean of preferred means: <scaled-result> 53.2057+-0.0926 53.1228+-0.0961

Filip Pizlo

Comment 3 2011-10-05 14:35:57 PDT

Landed in r96750.

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

2011-10-04 00:04 PDT

Modified

2011-10-05 14:35 PDT History

CC List

1 user Show

URL

Keywords

Depends on

69321

Blocks

Dependencies

tree graph