Bug 162879

Summary: Creating pcToOriginMap in FTL shouldn't insert unnecessary NOPs
Product: WebKit Reporter: Michael Saboff <msaboff>
Component: JavaScriptCoreAssignee: Michael Saboff <msaboff>
Status: RESOLVED FIXED    
Severity: Normal CC: commit-queue, keith_miller, mark.lam, saam
Priority: P2    
Version: Other   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
Patch fpizlo: review+

Michael Saboff
Reported 2016-10-03 14:38:18 PDT
In AirGenerate.cpp near line 196, we use standard macro assembler labels to get the current PC: pcToOriginMap.appendItem(jit.label(), ...); If we recently created a watchpoint label, via labelForWatchpoint(), then we'll not pad when creating the label for pcToOriginMap. This isn't necessary. Instead, we should use jit.labelIgnoringWatchpoints() for pcToOriginMap's labels.
Attachments
Patch (3.41 KB, patch)
2016-10-03 14:44 PDT, Michael Saboff 		fpizlo: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Michael Saboff
Comment 1 2016-10-03 14:44:48 PDT
Created attachment 290524 [details] Patch
Michael Saboff
Comment 2 2016-10-03 14:47:56 PDT
Performance results of patch. This shows a neutral to possible slight speedup on Octane. VMs tested: "Baseline" at /Volumes/Data/src/webkit.baseline/WebKitBuild/Release/JavaScriptCore.framework/Versions/A/Resources/jsc "ReducedNops" at /Volumes/Data/src/webkit/WebKitBuild/Release/JavaScriptCore.framework/Versions/A/Resources/jsc Collected 10 samples per benchmark/VM, with 10 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. Baseline ReducedNops SunSpider: 3d-cube 8.4927+-0.2201 8.4569+-0.3008 3d-morph 8.2876+-0.0970 8.2708+-0.1228 3d-raytrace 8.4772+-0.1849 ? 8.5529+-0.1157 ? access-binary-trees 3.1561+-0.1318 3.0818+-0.0988 might be 1.0241x faster access-fannkuch 8.5353+-0.3750 ? 8.5602+-0.2261 ? access-nbody 4.3588+-0.0829 ? 4.4079+-0.1106 ? might be 1.0113x slower access-nsieve 4.6929+-0.1068 ? 4.7037+-0.1082 ? bitops-3bit-bits-in-byte 1.7280+-0.0614 ? 1.7440+-0.0493 ? bitops-bits-in-byte 5.3063+-0.0667 ? 5.4083+-0.0655 ? might be 1.0192x slower bitops-bitwise-and 2.7582+-0.0843 ? 2.8231+-0.0581 ? might be 1.0235x slower bitops-nsieve-bits 4.7888+-0.1007 4.7659+-0.0748 controlflow-recursive 3.5892+-0.0747 ? 3.6187+-0.0766 ? crypto-aes 7.2152+-0.1444 7.1528+-0.1348 crypto-md5 4.4246+-0.1020 4.4095+-0.1500 crypto-sha1 4.4526+-0.0679 ! 4.6137+-0.0745 ! definitely 1.0362x slower date-format-tofte 14.2963+-0.3398 14.2701+-0.2102 date-format-xparb 7.7327+-0.0951 ? 7.7615+-0.1554 ? math-cordic 4.1716+-0.0670 ? 4.1983+-0.1158 ? math-partial-sums 9.0604+-0.0999 9.0447+-0.0844 math-spectral-norm 3.2773+-0.0997 3.2567+-0.0873 regexp-dna 10.5064+-0.1815 10.4241+-0.1226 string-base64 6.8975+-0.0620 ? 6.9925+-0.0856 ? might be 1.0138x slower string-fasta 9.2126+-0.1079 9.1795+-0.1092 string-tagcloud 13.5916+-0.1649 ? 13.6450+-0.2125 ? string-unpack-code 26.8644+-0.3241 26.8144+-0.2583 string-validate-input 6.6886+-0.0874 6.6695+-0.0890 <arithmetic> 7.4063+-0.0279 ? 7.4164+-0.0330 ? might be 1.0014x slower Baseline ReducedNops Octane: encrypt 0.27579+-0.00106 ? 0.27656+-0.00036 ? decrypt 5.10833+-0.03781 5.07763+-0.05787 deltablue x2 0.21972+-0.00619 ? 0.22088+-0.00790 ? earley 0.45393+-0.00336 0.45152+-0.00161 boyer 8.05405+-0.06061 ? 8.06141+-0.05228 ? navier-stokes x2 6.48237+-0.01023 ? 6.48366+-0.01257 ? raytrace x2 1.26735+-0.01108 1.25816+-0.00740 richards x2 0.14584+-0.00255 0.14379+-0.00139 might be 1.0142x faster splay x2 0.55314+-0.00466 0.54653+-0.00251 might be 1.0121x faster regexp x2 28.82025+-0.24989 28.71947+-0.38194 pdfjs x2 65.33183+-0.38564 65.30303+-0.17080 mandreel x2 69.00627+-0.34301 68.73104+-0.22256 gbemu x2 64.55220+-6.32640 60.84211+-1.77502 might be 1.0610x faster closure 0.79790+-0.00121 ! 0.80622+-0.00523 ! definitely 1.0104x slower jquery 10.56091+-0.04150 10.53725+-0.02327 box2d x2 16.89407+-0.03999 ^ 16.66270+-0.06511 ^ definitely 1.0139x faster zlib x2 543.56729+-16.39174 ? 547.80662+-13.38795 ? typescript x2 1138.51814+-9.00853 ? 1141.69075+-8.76392 ? <geometric> 8.47344+-0.04978 8.42189+-0.02698 might be 1.0061x faster Baseline ReducedNops Kraken: ai-astar 141.622+-1.036 ? 142.413+-0.553 ? audio-beat-detection 64.743+-0.508 64.150+-0.496 audio-dft 129.453+-1.837 ? 129.635+-0.468 ? audio-fft 52.363+-0.448 51.915+-0.294 audio-oscillator 76.678+-0.236 ? 76.769+-0.250 ? imaging-darkroom 96.780+-0.331 96.472+-0.131 imaging-desaturate 91.117+-0.322 ? 91.216+-0.274 ? imaging-gaussian-blur 112.888+-4.577 ? 114.224+-4.722 ? might be 1.0118x slower json-parse-financial 61.056+-0.294 ? 61.089+-0.386 ? json-stringify-tinderbox 39.834+-0.172 ? 39.834+-0.146 ? stanford-crypto-aes 64.406+-0.813 63.362+-0.810 might be 1.0165x faster stanford-crypto-ccm 59.174+-1.134 58.481+-2.039 might be 1.0119x faster stanford-crypto-pbkdf2 155.361+-2.938 152.626+-3.681 might be 1.0179x faster stanford-crypto-sha256-iterative 49.942+-0.505 49.711+-0.496 <arithmetic> 85.387+-0.313 85.135+-0.517 might be 1.0030x faster Baseline ReducedNops Geomean of preferred means: <scaled-result> 17.4989+-0.0423 17.4542+-0.0506 might be 1.0026x faster
Filip Pizlo
Comment 3 2016-10-03 14:49:38 PDT
Comment on attachment 290524 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=290524&action=review > Source/JavaScriptCore/b3/testb3.cpp:13295 > + AllowMacroScratchRegisterUsage allowScratch(jit); Remove > Source/JavaScriptCore/b3/testb3.cpp:13302 > + AllowMacroScratchRegisterUsage allowScratch(jit); Remove.
Michael Saboff
Comment 4 2016-10-03 14:53:37 PDT
Committed r206752: <http://trac.webkit.org/changeset/206752>
Note You need to log in before you can comment on or make changes to this bug.