Bug 140948 – REGRESSION(r178591): 20% regression in Octane box2d

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Michael Saboff]

I believe that I made willResolveToArgumentsRegister() too restrictive on when we can't use the arguments register.

[Michael Saboff]

rdar://problem/19603062

[Michael Saboff]

Created attachment 245459 [details]
Patch for Landing - reviewed in person.

Performance test results:

Generating benchmark report at /Volumes/Data/src/webkit/Baseline_ArgsSpeedup_V8SpiderOctane_msaboff-pro_20150127_1207_report.txt
And raw data at /Volumes/Data/src/webkit/Baseline_ArgsSpeedup_V8SpiderOctane_msaboff-pro_20150127_1207.json

Benchmark report for V8Spider and Octane on msaboff-pro (MacPro5,1).

VMs tested:
"Baseline" at /Volumes/Data/src/webkit.baseline/WebKitBuild/Release/JavaScriptCore.framework/Versions/A/Resources/jsc
"ArgsSpeedup" at /Volumes/Data/src/webkit/WebKitBuild/Release/JavaScriptCore.framework/Versions/A/Resources/jsc

Collected 4 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.

                               Baseline                ArgsSpeedup                                    
V8Spider:
   crypto                  72.3723+-1.7669     ?     72.6451+-1.2317        ?
   deltablue              112.3093+-4.1750          109.4640+-1.1189          might be 1.0260x faster
   earley-boyer            62.8467+-2.6278           62.0949+-1.1430          might be 1.0121x faster
   raytrace                46.8278+-0.5360     ^     40.7990+-2.2191        ^ definitely 1.1478x faster
   regexp                  95.0038+-1.2239           94.0562+-0.2997          might be 1.0101x faster
   richards               115.2778+-5.3954          113.8145+-1.8537          might be 1.0129x faster
   splay                   47.5782+-3.6035     ?     48.3935+-1.4790        ? might be 1.0171x slower

   <arithmetic>            78.8880+-0.8313     ^     77.3239+-0.3591        ^ definitely 1.0202x faster
   <geometric> *           74.2509+-0.4071     ^     72.4026+-0.6576        ^ definitely 1.0255x faster
   <harmonic>              69.8161+-0.3718     ^     67.5540+-1.0545        ^ definitely 1.0335x faster

                               Baseline                ArgsSpeedup                                    
Octane:
   encrypt                 0.31826+-0.00621    ?     0.32239+-0.00534       ? might be 1.0130x slower
   decrypt                 5.67187+-0.03230    ?     5.67794+-0.02059       ?
   deltablue      x2       0.26549+-0.00733          0.26494+-0.00487       
   earley                  0.98706+-0.01422          0.98068+-0.00719       
   boyer                   8.84213+-0.23500          8.79115+-0.09154       
   navier-stokes  x2       6.45035+-0.04989          6.41486+-0.01337       
   raytrace       x2       1.89617+-0.00534          1.86997+-0.12254         might be 1.0140x faster
   richards       x2       0.16955+-0.00206          0.16665+-0.00476         might be 1.0174x faster
   splay          x2       0.52388+-0.01477          0.52011+-0.00737       
   regexp         x2      45.04862+-0.64073    ?    45.71838+-0.87198       ? might be 1.0149x slower
   pdfjs          x2      70.36549+-0.29415    ?    71.52163+-2.49187       ? might be 1.0164x slower
   mandreel       x2      73.32669+-1.18600    ?    73.33962+-0.63429       ?
   gbemu          x2      63.27088+-1.78430         62.73810+-0.55299       
   closure                 0.74960+-0.00221    ?     0.75456+-0.00875       ?
   jquery                  9.45089+-0.06624    ?     9.45695+-0.19931       ?
   box2d          x2      22.77361+-0.25336    ^    18.65764+-0.25051       ^ definitely 1.2206x faster
   zlib           x2     683.33309+-4.18352    ?   686.91012+-19.77018      ?
   typescript     x2    1107.77979+-23.04321   ?  1128.86298+-22.23050      ? might be 1.0190x slower

   <arithmetic>          139.21423+-1.45920    ?   140.66512+-2.23907       ? might be 1.0104x slower
   <geometric> *           9.82902+-0.02897    ^     9.70225+-0.05371       ^ definitely 1.0131x faster
   <harmonic>              0.97964+-0.00566          0.97194+-0.01243         might be 1.0079x faster

                               Baseline                ArgsSpeedup                                    
All benchmarks:
   <arithmetic>           127.8012+-1.0760     ?    128.6816+-1.8686        ? might be 1.0069x slower
   <geometric>             14.4097+-0.0258     ^     14.1909+-0.0474        ^ definitely 1.0154x faster
   <harmonic>               1.2043+-0.0069            1.1947+-0.0152          might be 1.0080x faster

                               Baseline                ArgsSpeedup                                    
Geomean of preferred means:
   <scaled-result>         85.4289+-0.1672     ^     83.8128+-0.2473        ^ definitely 1.0193x faster

[Michael Saboff]

Committed r179202: <http://trac.webkit.org/changeset/179202>