FTL is already concurrent but we should allow parallel FTL compiles also.
Created attachment 220697 [details] the patch
The short story: Benchmark report for Octane on dethklok (MacBookPro9,1). VMs tested: "DFG" at /Volumes/Data/pizlo/cStack/OpenSource/WebKitBuild/Release/jsc (r161543) "FTL" at /Volumes/Data/pizlo/cStack/OpenSource/WebKitBuild/Release/jsc (r161543) export JSC_useExperimentalFTL=true "FTLParallel" at /Volumes/Data/pizlo/cStack/OpenSource/WebKitBuild/Release/jsc (r161543) export JSC_useExperimentalFTL=true export JSC_numberOfCompilerThreads=7 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. DFG FTL FTLParallel FTLParallel v. DFG Octane and V8v7: gbemu x2 37.09279+-0.16632 ! 57.01742+-3.35945 ^ 46.94429+-0.28350 ! definitely 1.2656x slower V8v7: <arithmetic> ERROR ERROR ERROR <geometric> * ERROR ERROR ERROR <harmonic> ERROR ERROR ERROR Octane including V8v7: <arithmetic> 37.09279+-0.16632 ! 57.01742+-3.35945 ^ 46.94429+-0.28350 ! definitely 1.2656x slower <geometric> * 37.09279+-0.16632 ! 57.01742+-3.35945 ^ 46.94429+-0.28350 ! definitely 1.2656x slower <harmonic> 37.09279+-0.16632 ! 57.01742+-3.35945 ^ 46.94429+-0.28350 ! definitely 1.2656x slower
More numbers, note that by default we still disable parallel JIT, but making it available seems like a good thing: Benchmark report for SunSpider, Octane, and Kraken on oldmac (MacPro4,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/cStack/OpenSource/WebKitBuild/Release/jsc (r161543) export JSC_useExperimentalFTL=true "Parallel" at /Volumes/Data/fromMiniMe/cStack/OpenSource/WebKitBuild/Release/jsc (r161543) export JSC_useExperimentalFTL=true export JSC_numberOfCompilerThreads=7 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. TipOfTree Parallel SunSpider: 3d-cube 7.7071+-0.0627 ? 7.7599+-0.0630 ? 3d-morph 8.7913+-0.0855 ! 9.1583+-0.1142 ! definitely 1.0417x slower 3d-raytrace 9.3980+-0.1465 ? 9.5145+-0.1579 ? might be 1.0124x slower access-binary-trees 2.4403+-0.0295 ^ 2.3333+-0.0685 ^ definitely 1.0458x faster access-fannkuch 7.9568+-0.1316 ! 8.4934+-0.3379 ! definitely 1.0674x slower access-nbody 4.2729+-0.0189 ! 4.4826+-0.0303 ! definitely 1.0491x slower access-nsieve 5.5803+-0.0928 ? 5.6215+-0.0719 ? bitops-3bit-bits-in-byte 1.9870+-0.0047 ! 2.1859+-0.0122 ! definitely 1.1001x slower bitops-bits-in-byte 6.7047+-0.0682 ! 6.8630+-0.0448 ! definitely 1.0236x slower bitops-bitwise-and 3.0412+-0.0320 ! 3.1931+-0.0858 ! definitely 1.0499x slower bitops-nsieve-bits 5.8168+-0.0552 ! 6.1890+-0.0612 ! definitely 1.0640x slower controlflow-recursive 4.5129+-0.0739 ^ 3.9195+-0.0261 ^ definitely 1.1514x faster crypto-aes 5.8784+-0.0767 ! 6.1094+-0.0306 ! definitely 1.0393x slower crypto-md5 3.5477+-0.0188 ? 3.5822+-0.0360 ? crypto-sha1 3.6050+-0.0223 ! 3.7944+-0.0272 ! definitely 1.0525x slower date-format-tofte 11.6571+-0.1033 ! 11.9440+-0.1331 ! definitely 1.0246x slower date-format-xparb 8.6383+-0.1450 ? 8.8529+-0.2685 ? might be 1.0248x slower math-cordic 4.7751+-0.0291 ! 4.9111+-0.0260 ! definitely 1.0285x slower math-partial-sums 10.1289+-0.1080 ! 10.3787+-0.0792 ! definitely 1.0247x slower math-spectral-norm 4.6044+-0.0323 ^ 3.2050+-0.0165 ^ definitely 1.4366x faster regexp-dna 13.0012+-0.1077 12.9501+-0.1136 string-base64 5.5442+-0.0405 ! 5.9641+-0.0575 ! definitely 1.0757x slower string-fasta 11.3186+-0.1675 ^ 10.8574+-0.0923 ^ definitely 1.0425x faster string-tagcloud 15.0780+-0.1324 ! 15.5265+-0.1503 ! definitely 1.0297x slower string-unpack-code 32.4747+-0.4656 ? 32.9790+-0.2077 ? might be 1.0155x slower string-validate-input 7.1373+-0.0561 ! 7.2932+-0.0750 ! definitely 1.0218x slower <arithmetic> * 7.9076+-0.0206 ! 8.0024+-0.0196 ! definitely 1.0120x slower <geometric> 6.5238+-0.0140 ! 6.5626+-0.0162 ! definitely 1.0059x slower <harmonic> 5.5163+-0.0139 ? 5.5209+-0.0174 ? might be 1.0008x slower TipOfTree Parallel Octane and V8v7: encrypt 0.43707+-0.00054 ? 0.43721+-0.00042 ? decrypt 8.07201+-0.01921 ! 8.17635+-0.06581 ! definitely 1.0129x slower deltablue x2 0.55410+-0.00355 0.54569+-0.01000 might be 1.0154x faster earley 0.89137+-0.01115 0.88520+-0.00610 boyer 11.92959+-0.05730 ? 12.02242+-0.06071 ? raytrace x2 3.95029+-0.04078 3.94056+-0.01886 regexp x2 30.88141+-0.08180 ? 30.97033+-0.05840 ? richards x2 0.22391+-0.00123 ? 0.22518+-0.00330 ? splay x2 0.63047+-0.00325 0.62737+-0.00320 navier-stokes x2 8.24171+-0.00417 ^ 8.22514+-0.00855 ^ definitely 1.0020x faster closure 0.77295+-0.00243 ? 0.77768+-0.00246 ? jquery 10.88617+-0.07203 ? 11.05496+-0.12591 ? might be 1.0155x slower gbemu x2 94.17431+-0.30754 ^ 83.98530+-0.59289 ^ definitely 1.1213x faster mandreel x2 102.53284+-0.44228 101.71001+-0.63318 pdfjs x2 100.40156+-0.33543 ? 100.56494+-0.24322 ? box2d x2 34.75103+-0.13026 ^ 34.16445+-0.16824 ^ definitely 1.0172x faster V8v7: <arithmetic> 6.89336+-0.01301 ? 6.91186+-0.01271 ? might be 1.0027x slower <geometric> * 2.16438+-0.00520 2.16183+-0.00799 might be 1.0012x faster <harmonic> 0.79419+-0.00231 0.79301+-0.00622 might be 1.0015x faster Octane including V8v7: <arithmetic> 30.21817+-0.04109 ^ 29.35661+-0.07554 ^ definitely 1.0293x faster <geometric> * 6.62176+-0.01072 ^ 6.55245+-0.01652 ^ definitely 1.0106x faster <harmonic> 1.20092+-0.00332 1.19956+-0.00877 might be 1.0011x faster TipOfTree Parallel Kraken: ai-astar 494.728+-0.926 494.582+-0.917 audio-beat-detection 222.562+-1.220 ? 223.112+-1.602 ? audio-dft 292.784+-1.339 ? 293.696+-2.648 ? audio-fft 129.641+-0.122 ? 129.743+-0.207 ? audio-oscillator 530.735+-0.598 ! 537.215+-3.845 ! definitely 1.0122x slower imaging-darkroom 293.979+-2.450 ^ 289.578+-0.685 ^ definitely 1.0152x faster imaging-desaturate 109.575+-0.117 ? 109.999+-0.386 ? imaging-gaussian-blur 192.666+-7.232 ? 203.863+-26.531 ? might be 1.0581x slower json-parse-financial 79.570+-0.179 ! 81.684+-0.424 ! definitely 1.0266x slower json-stringify-tinderbox 107.233+-2.463 106.180+-1.573 stanford-crypto-aes 92.793+-0.701 92.711+-0.393 stanford-crypto-ccm 101.024+-1.560 ? 104.447+-4.284 ? might be 1.0339x slower stanford-crypto-pbkdf2 260.186+-0.474 ? 260.354+-0.869 ? stanford-crypto-sha256-iterative 113.937+-0.544 ! 115.170+-0.659 ! definitely 1.0108x slower <arithmetic> * 215.815+-0.647 ? 217.310+-1.820 ? might be 1.0069x slower <geometric> 178.208+-0.642 ? 179.650+-1.418 ? might be 1.0081x slower <harmonic> 151.420+-0.633 ? 152.836+-0.932 ? might be 1.0094x slower TipOfTree Parallel All benchmarks: <arithmetic> 60.7982+-0.1331 ? 60.8132+-0.3843 ? might be 1.0002x slower <geometric> 13.2356+-0.0099 13.2342+-0.0324 might be 1.0001x faster <harmonic> 2.4947+-0.0056 2.4928+-0.0158 might be 1.0007x faster TipOfTree Parallel Geomean of preferred means: <scaled-result> 22.4405+-0.0235 ? 22.5024+-0.0734 ? might be 1.0028x slower
Landed on branch in http://trac.webkit.org/changeset/161546
Review status updated in r161938: <http://trac.webkit.org/r161938>.