74835 – DFG is too sloppy with register allocation

Filip Pizlo

Reported 2011-12-18 20:25:58 PST

The DFG compiler sometimes keeps things in registers longer than it should, and doesn't assert aggressively enough that its register allocation discipline makes any sense. <rdar://problem/10599330>

Filip Pizlo

Comment 1 2011-12-18 20:26:34 PST

Performance on 64-bit Benchmark report for SunSpider, V8, and Kraken on bigmac (MacPro5,1). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc (r103205) "FixRegAllox" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc (r103205) Collected 12 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. TipOfTree FixRegAllox SunSpider: 3d-cube 6.2415+-0.0296 6.2226+-0.0214 3d-morph 8.4986+-0.1461 8.3180+-0.0478 might be 1.0217x faster 3d-raytrace 7.5847+-0.0637 ? 7.6225+-0.1090 ? access-binary-trees 1.6184+-0.0109 1.6118+-0.0045 access-fannkuch 7.5383+-0.0054 ? 7.5451+-0.0073 ? access-nbody 3.9539+-0.0091 3.9465+-0.0128 access-nsieve 3.2099+-0.0431 ? 3.2124+-0.0376 ? bitops-3bit-bits-in-byte 1.2578+-0.0161 1.2466+-0.0105 bitops-bits-in-byte 5.0772+-0.0226 5.0664+-0.0239 bitops-bitwise-and 3.2899+-0.0053 3.2888+-0.0043 bitops-nsieve-bits 5.6547+-0.0446 5.6138+-0.0276 controlflow-recursive 2.3483+-0.0259 ? 2.3617+-0.0223 ? crypto-aes 7.5875+-0.0427 7.5685+-0.0386 crypto-md5 2.4906+-0.0224 ? 2.5017+-0.0261 ? crypto-sha1 2.2310+-0.0161 ? 2.2370+-0.0231 ? date-format-tofte 10.5841+-0.0644 10.4793+-0.0767 might be 1.0100x faster date-format-xparb 10.1877+-0.2496 ^ 9.7576+-0.1500 ^ definitely 1.0441x faster math-cordic 7.1462+-0.0251 7.1226+-0.0195 math-partial-sums 10.3879+-0.0173 ? 10.4673+-0.0986 ? math-spectral-norm 2.6406+-0.0035 ? 2.6631+-0.0277 ? regexp-dna 8.8241+-0.0191 ? 8.8605+-0.0825 ? string-base64 4.2716+-0.0189 ? 4.2797+-0.0151 ? string-fasta 7.3972+-0.1232 7.2546+-0.0344 might be 1.0197x faster string-tagcloud 12.4045+-0.0676 ? 12.4642+-0.0452 ? string-unpack-code 20.3879+-0.0666 ? 20.5254+-0.0992 ? string-validate-input 5.5163+-0.0353 ? 5.5835+-0.0390 ? might be 1.0122x slower <arithmetic> * 6.4742+-0.0195 6.4547+-0.0185 might be 1.0030x faster <geometric> 5.2756+-0.0171 5.2631+-0.0167 might be 1.0024x faster <harmonic> 4.1889+-0.0164 4.1826+-0.0157 might be 1.0015x faster TipOfTree FixRegAllox V8: crypto 76.9500+-0.1898 ? 77.0115+-0.2233 ? deltablue 169.8515+-0.5094 ? 170.5953+-0.9752 ? earley-boyer 99.2693+-1.3206 99.0097+-1.2394 raytrace 52.3763+-0.3498 ? 52.5079+-0.3731 ? regexp 122.5183+-0.4655 ! 123.5123+-0.4026 ! definitely 1.0081x slower richards 140.3176+-0.8017 139.5973+-0.8194 splay 91.5850+-3.0501 90.0657+-0.8446 might be 1.0169x faster <arithmetic> 107.5526+-0.4720 107.4714+-0.2578 might be 1.0008x faster <geometric> * 100.9514+-0.5162 100.8392+-0.2470 might be 1.0011x faster <harmonic> 94.2317+-0.5136 94.1332+-0.2461 might be 1.0010x faster TipOfTree FixRegAllox Kraken: ai-astar 807.8613+-12.3133 807.7542+-13.0338 audio-beat-detection 191.4026+-2.9822 189.8475+-1.1770 audio-dft 278.2161+-1.9174 275.7080+-2.9470 audio-fft 118.6202+-1.6812 118.2027+-0.8468 audio-oscillator 281.4815+-2.6395 279.0403+-0.3853 imaging-darkroom 299.5302+-5.5458 ? 300.7372+-5.8153 ? imaging-desaturate 237.7154+-0.1280 237.5794+-0.0961 imaging-gaussian-blur 520.4992+-0.4187 520.3533+-0.4624 json-parse-financial 71.3050+-0.4063 70.9569+-0.1153 json-stringify-tinderbox 82.1325+-0.3451 82.0580+-0.2758 stanford-crypto-aes 117.2481+-0.8390 ? 117.7502+-0.7447 ? stanford-crypto-ccm 112.3163+-0.4701 ! 113.9508+-0.7557 ! definitely 1.0146x slower stanford-crypto-pbkdf2 229.0775+-1.2323 ? 229.5788+-0.8270 ? stanford-crypto-sha256-iterative 95.3843+-0.2308 ? 95.8788+-0.4132 ? <arithmetic> * 245.9136+-0.9030 245.6711+-1.1304 might be 1.0010x faster <geometric> 191.5584+-0.3163 191.4799+-0.4247 might be 1.0004x faster <harmonic> 155.5540+-0.2637 ? 155.6007+-0.2172 ? might be 1.0003x slower TipOfTree FixRegAllox All benchmarks: <arithmetic> 92.8508+-0.2536 92.7557+-0.3255 might be 1.0010x faster <geometric> 23.8709+-0.0549 23.8329+-0.0536 might be 1.0016x faster <harmonic> 7.3769+-0.0282 7.3661+-0.0271 might be 1.0015x faster TipOfTree FixRegAllox Geomean of preferred means: <scaled-result> 54.3696+-0.1325 54.2769+-0.1142 might be 1.0017x faster Running locally... 376/376 Generating benchmark report at TipOfTree_FixRegAllox_SunSpiderV8Kraken_nitroflex_20111218_2023_benchReport.txt Benchmark report for SunSpider, V8, and Kraken on nitroflex (MacBookPro8,2). VMs tested: "TipOfTree" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc (r103205) "FixRegAllox" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc (r103205) Collected 12 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. TipOfTree FixRegAllox SunSpider: 3d-cube 5.3674+-0.0597 ? 5.4565+-0.1200 ? might be 1.0166x slower 3d-morph 7.6145+-0.1208 ? 7.6215+-0.1253 ? 3d-raytrace 7.0994+-0.1581 7.0386+-0.1572 access-binary-trees 1.5602+-0.0401 ^ 1.4707+-0.0299 ^ definitely 1.0609x faster access-fannkuch 6.2365+-0.0922 6.1130+-0.0728 might be 1.0202x faster access-nbody 3.2109+-0.0681 ? 3.2299+-0.0703 ? access-nsieve 2.6344+-0.0724 2.6069+-0.0746 might be 1.0106x faster bitops-3bit-bits-in-byte 1.2275+-0.0263 ? 1.2434+-0.0286 ? might be 1.0129x slower bitops-bits-in-byte 2.3385+-0.0443 ? 2.3758+-0.0547 ? might be 1.0160x slower bitops-bitwise-and 3.3384+-0.0514 ? 3.4514+-0.0628 ? might be 1.0339x slower bitops-nsieve-bits 5.3581+-0.1394 ? 5.5527+-0.1103 ? might be 1.0363x slower controlflow-recursive 1.9907+-0.0436 1.9422+-0.0145 might be 1.0249x faster crypto-aes 7.3454+-0.1617 ? 7.4329+-0.1966 ? might be 1.0119x slower crypto-md5 2.2993+-0.0415 2.2786+-0.0300 crypto-sha1 2.1155+-0.0484 2.0971+-0.0461 date-format-tofte 9.8219+-0.1826 9.6931+-0.1243 might be 1.0133x faster date-format-xparb 9.1419+-0.1737 9.0915+-0.1871 math-cordic 6.3434+-0.0873 ? 6.4302+-0.1598 ? might be 1.0137x slower math-partial-sums 7.3916+-0.1143 7.3809+-0.1216 math-spectral-norm 2.2987+-0.0196 ? 2.3135+-0.0403 ? regexp-dna 7.9644+-0.1101 7.7743+-0.1024 might be 1.0245x faster string-base64 4.2075+-0.1145 ? 4.3032+-0.0729 ? might be 1.0228x slower string-fasta 6.5694+-0.1144 6.5625+-0.1204 string-tagcloud 10.9646+-0.1829 10.8515+-0.1448 might be 1.0104x faster string-unpack-code 18.7656+-0.2652 18.5289+-0.2323 might be 1.0128x faster string-validate-input 5.2800+-0.1388 5.2085+-0.1322 might be 1.0137x faster <arithmetic> * 5.7110+-0.0285 5.6942+-0.0188 might be 1.0030x faster <geometric> 4.6281+-0.0269 4.6196+-0.0137 might be 1.0018x faster <harmonic> 3.7038+-0.0261 3.6892+-0.0120 might be 1.0039x faster TipOfTree FixRegAllox V8: crypto 69.2721+-0.7413 ? 69.3614+-0.6678 ? deltablue 147.2022+-1.4357 146.4766+-0.9339 earley-boyer 79.1693+-1.2043 ? 79.7949+-1.1389 ? raytrace 47.4207+-0.2673 ? 47.4259+-0.5865 ? regexp 103.4519+-0.5066 102.7928+-0.8345 richards 118.0912+-0.6863 ? 118.3173+-1.1495 ? splay 72.4382+-0.7795 ^ 70.9343+-0.5434 ^ definitely 1.0212x faster <arithmetic> 91.0065+-0.2999 90.7290+-0.4134 might be 1.0031x faster <geometric> * 85.6656+-0.2896 85.4076+-0.4116 might be 1.0030x faster <harmonic> 80.5495+-0.2752 80.3162+-0.4426 might be 1.0029x faster TipOfTree FixRegAllox Kraken: ai-astar 474.6751+-2.0640 ? 477.1846+-5.2130 ? audio-beat-detection 169.2471+-0.8500 ? 170.2460+-1.4331 ? audio-dft 276.9173+-3.8333 276.1273+-2.6584 audio-fft 105.8960+-0.4160 ? 106.1126+-0.9759 ? audio-oscillator 250.0479+-1.6387 ? 250.1671+-1.1911 ? imaging-darkroom 267.9879+-5.8848 ? 268.6597+-6.0591 ? imaging-desaturate 224.2193+-0.6828 ! 227.6462+-0.8253 ! definitely 1.0153x slower imaging-gaussian-blur 481.6816+-1.9163 ? 484.9727+-3.5730 ? json-parse-financial 57.5064+-0.6289 ? 57.8643+-0.8253 ? json-stringify-tinderbox 73.3214+-0.4450 ? 73.8877+-0.5348 ? stanford-crypto-aes 94.6605+-0.4635 ! 95.7534+-0.2978 ! definitely 1.0115x slower stanford-crypto-ccm 97.7390+-0.7350 ? 98.4418+-0.9552 ? stanford-crypto-pbkdf2 188.3791+-0.9875 188.3673+-1.2585 stanford-crypto-sha256-iterative 82.6176+-0.7911 82.4370+-0.2675 <arithmetic> * 203.2069+-0.6381 ? 204.1334+-0.8902 ? might be 1.0046x slower <geometric> 164.1801+-0.4580 ? 164.9492+-0.6033 ? might be 1.0047x slower <harmonic> 133.8590+-0.3944 ? 134.5314+-0.4795 ? might be 1.0050x slower TipOfTree FixRegAllox All benchmarks: <arithmetic> 77.2431+-0.2058 ? 77.4685+-0.2900 ? might be 1.0029x slower <geometric> 20.6937+-0.0668 20.6922+-0.0486 might be 1.0001x faster <harmonic> 6.5175+-0.0447 6.4928+-0.0206 might be 1.0038x faster TipOfTree FixRegAllox Geomean of preferred means: <scaled-result> 46.3248+-0.0971 46.3029+-0.1112 might be 1.0005x faster

Filip Pizlo

Comment 2 2011-12-18 20:41:26 PST

Performance on 32-bit [pizlo@nitroflex bencher] ./bencher TipOfTree32:/Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc FixRegAllox32:/Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc --remote bigmac --local Copying TipOfTree32 into /Volumes/Data/pizlo/bencher/temp/benchdata... Copying FixRegAllox32 into /Volumes/Data/pizlo/bencher/temp/benchdata... All VMs are in place. Packaging benchmarking directory for remote hosts... Sending benchmark payload to bigmac... Running on bigmac... 376/376 Generating benchmark report at TipOfTree32_FixRegAllox32_SunSpiderV8Kraken_bigmac_20111218_2038_benchReport.txt Benchmark report for SunSpider, V8, and Kraken on bigmac (MacPro5,1). VMs tested: "TipOfTree32" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc (r103205) "FixRegAllox32" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc (r103205) Collected 12 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. TipOfTree32 FixRegAllox32 SunSpider: 3d-cube 7.2415+-0.0118 ? 7.2659+-0.0156 ? 3d-morph 10.9712+-0.0257 ? 10.9964+-0.0376 ? 3d-raytrace 9.6943+-0.0333 ^ 9.6388+-0.0128 ^ definitely 1.0058x faster access-binary-trees 1.9335+-0.0074 ? 1.9368+-0.0089 ? access-fannkuch 9.3437+-0.0223 ? 9.3513+-0.0325 ? access-nbody 5.7192+-0.0043 ^ 5.6822+-0.0146 ^ definitely 1.0065x faster access-nsieve 3.1192+-0.0071 ? 3.1231+-0.0084 ? bitops-3bit-bits-in-byte 1.2905+-0.0207 ? 1.2965+-0.0202 ? bitops-bits-in-byte 4.9757+-0.0094 4.9749+-0.0161 bitops-bitwise-and 3.8653+-0.0306 ? 3.8788+-0.0191 ? bitops-nsieve-bits 6.7604+-0.0087 ? 6.7777+-0.0268 ? controlflow-recursive 3.0331+-0.0239 3.0131+-0.0056 crypto-aes 10.0957+-0.0489 ! 10.3150+-0.0261 ! definitely 1.0217x slower crypto-md5 3.2807+-0.0198 3.2748+-0.0179 crypto-sha1 2.7549+-0.0208 ? 2.7590+-0.0099 ? date-format-tofte 11.4431+-0.0481 ? 11.5128+-0.0441 ? date-format-xparb 11.8692+-0.0576 ! 12.1184+-0.1233 ! definitely 1.0210x slower math-cordic 9.9647+-0.0341 9.9379+-0.0299 math-partial-sums 12.4512+-0.0185 12.4244+-0.0265 math-spectral-norm 2.7283+-0.0069 2.7254+-0.0108 regexp-dna 9.3073+-0.1241 ^ 9.1720+-0.0105 ^ definitely 1.0148x faster string-base64 4.8903+-0.0073 4.8867+-0.0063 string-fasta 8.6363+-0.0148 ? 8.6560+-0.0262 ? string-tagcloud 13.7226+-0.0610 13.6375+-0.0284 string-unpack-code 22.0759+-0.0648 21.9853+-0.0526 string-validate-input 6.3142+-0.0366 ? 6.3586+-0.0208 ? <arithmetic> * 7.5955+-0.0134 ? 7.6038+-0.0087 ? might be 1.0011x slower <geometric> 6.1988+-0.0130 ? 6.2053+-0.0096 ? might be 1.0010x slower <harmonic> 4.8555+-0.0184 ? 4.8606+-0.0159 ? might be 1.0011x slower TipOfTree32 FixRegAllox32 V8: crypto 101.2954+-0.1853 101.0174+-0.3112 deltablue 187.6014+-1.2169 186.7793+-1.0042 earley-boyer 118.6513+-1.4013 ? 119.1959+-1.3526 ? raytrace 56.8082+-0.2555 ? 58.2214+-1.8180 ? might be 1.0249x slower regexp 128.3248+-0.6396 127.6049+-0.3558 richards 199.3216+-0.5879 ? 199.4225+-1.2734 ? splay 98.4823+-1.1111 ? 99.1035+-1.7769 ? <arithmetic> 127.2121+-0.4190 ? 127.3350+-0.5843 ? might be 1.0010x slower <geometric> * 118.2102+-0.3626 ? 118.5846+-0.7172 ? might be 1.0032x slower <harmonic> 109.0275+-0.2954 ? 109.7254+-0.9855 ? might be 1.0064x slower TipOfTree32 FixRegAllox32 Kraken: ai-astar 899.0276+-7.1020 ? 900.7698+-0.5249 ? audio-beat-detection 362.4061+-2.3057 ? 364.5499+-0.7262 ? audio-dft 363.2819+-2.3767 361.4780+-2.6151 audio-fft 220.6278+-0.6800 ? 221.3411+-0.5022 ? audio-oscillator 442.6988+-2.9341 441.9102+-2.8373 imaging-darkroom 391.0075+-4.0946 ? 391.0196+-4.1839 ? imaging-desaturate 297.6189+-0.2173 297.3822+-0.2328 imaging-gaussian-blur 706.8123+-0.4998 ? 707.9762+-1.3653 ? json-parse-financial 72.9125+-0.1172 ? 73.2383+-0.2330 ? json-stringify-tinderbox 106.0381+-0.3153 105.7179+-0.2628 stanford-crypto-aes 140.6319+-0.5437 ? 141.2936+-0.7772 ? stanford-crypto-ccm 129.0145+-0.7118 128.1255+-0.5148 stanford-crypto-pbkdf2 259.4363+-0.8636 ! 261.3396+-0.8399 ! definitely 1.0073x slower stanford-crypto-sha256-iterative 109.2524+-0.3084 ? 109.3277+-0.3363 ? <arithmetic> * 321.4833+-0.5407 ? 321.8193+-0.5402 ? might be 1.0010x slower <geometric> 250.2567+-0.3287 ? 250.4803+-0.4151 ? might be 1.0009x slower <harmonic> 195.0791+-0.2745 ? 195.2465+-0.3022 ? might be 1.0009x slower TipOfTree32 FixRegAllox32 All benchmarks: <arithmetic> 118.9092+-0.1616 ? 119.0322+-0.2140 ? might be 1.0010x slower <geometric> 28.9337+-0.0514 ? 28.9716+-0.0502 ? might be 1.0013x slower <harmonic> 8.5599+-0.0318 ? 8.5694+-0.0276 ? might be 1.0011x slower TipOfTree32 FixRegAllox32 Geomean of preferred means: <scaled-result> 66.0878+-0.0957 ? 66.2044+-0.1591 ? might be 1.0018x slower Running locally... 376/376 Generating benchmark report at TipOfTree32_FixRegAllox32_SunSpiderV8Kraken_nitroflex_20111218_2040_benchReport.txt Benchmark report for SunSpider, V8, and Kraken on nitroflex (MacBookPro8,2). VMs tested: "TipOfTree32" at /Volumes/Data/pizlo/quinary/OpenSource/WebKitBuild/Release/jsc (r103205) "FixRegAllox32" at /Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc (r103205) Collected 12 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. TipOfTree32 FixRegAllox32 SunSpider: 3d-cube 6.3514+-0.1064 6.2673+-0.0873 might be 1.0134x faster 3d-morph 10.3341+-0.0950 10.3068+-0.1198 3d-raytrace 9.0755+-0.1247 ? 9.0998+-0.1234 ? access-binary-trees 1.7146+-0.0474 ? 1.7928+-0.0606 ? might be 1.0456x slower access-fannkuch 7.7941+-0.1011 7.7734+-0.1096 access-nbody 5.1546+-0.0793 5.0752+-0.0657 might be 1.0156x faster access-nsieve 2.5397+-0.0571 ? 2.5791+-0.0493 ? might be 1.0155x slower bitops-3bit-bits-in-byte 1.2588+-0.0312 1.2378+-0.0237 might be 1.0170x faster bitops-bits-in-byte 2.4702+-0.0677 ? 2.4708+-0.0587 ? bitops-bitwise-and 2.9361+-0.1600 2.8068+-0.0503 might be 1.0461x faster bitops-nsieve-bits 6.2598+-0.1007 6.2304+-0.0836 controlflow-recursive 2.5292+-0.1050 ? 2.5748+-0.0540 ? might be 1.0180x slower crypto-aes 9.0541+-0.1422 ? 9.2731+-0.2058 ? might be 1.0242x slower crypto-md5 2.9867+-0.0680 ? 2.9954+-0.0684 ? crypto-sha1 2.5955+-0.0531 ? 2.6003+-0.0610 ? date-format-tofte 10.5304+-0.2213 10.2841+-0.1304 might be 1.0239x faster date-format-xparb 10.1926+-0.1257 ? 10.2508+-0.1510 ? math-cordic 8.1121+-0.1336 8.0873+-0.1531 math-partial-sums 9.6628+-0.1622 ? 9.6865+-0.1445 ? math-spectral-norm 2.4549+-0.0393 2.4202+-0.0499 might be 1.0143x faster regexp-dna 7.8236+-0.1302 7.7456+-0.1457 might be 1.0101x faster string-base64 4.6609+-0.1067 4.5756+-0.0896 might be 1.0186x faster string-fasta 7.7347+-0.1063 7.7169+-0.1646 string-tagcloud 12.0710+-0.1818 12.0534+-0.1728 string-unpack-code 20.1773+-0.3363 20.0840+-0.3095 string-validate-input 5.9710+-0.1366 5.9646+-0.1215 <arithmetic> * 6.6325+-0.0266 6.6136+-0.0246 might be 1.0029x faster <geometric> 5.3542+-0.0227 5.3423+-0.0230 might be 1.0022x faster <harmonic> 4.1960+-0.0238 4.1936+-0.0215 might be 1.0006x faster TipOfTree32 FixRegAllox32 V8: crypto 93.0857+-0.4705 ? 93.1162+-0.4067 ? deltablue 155.1923+-1.3671 154.8117+-1.5066 earley-boyer 93.4563+-0.9280 ? 93.5306+-1.1499 ? raytrace 51.0585+-0.8245 ? 51.4678+-0.4252 ? regexp 111.2716+-0.3969 ^ 108.3468+-0.4182 ^ definitely 1.0270x faster richards 166.2101+-1.0434 ? 167.0656+-0.5824 ? splay 80.6801+-1.1339 79.6174+-0.9456 might be 1.0133x faster <arithmetic> 107.2792+-0.3074 106.8509+-0.4148 might be 1.0040x faster <geometric> * 100.3933+-0.3439 99.9947+-0.3987 might be 1.0040x faster <harmonic> 93.4734+-0.4482 93.2024+-0.3899 might be 1.0029x faster TipOfTree32 FixRegAllox32 Kraken: ai-astar 518.0567+-4.2396 ? 522.9223+-3.8962 ? audio-beat-detection 332.5109+-5.2239 ? 339.3258+-5.9447 ? might be 1.0205x slower audio-dft 363.2800+-2.9691 362.3718+-2.9151 audio-fft 207.1526+-1.5889 ^ 204.0062+-0.5000 ^ definitely 1.0154x faster audio-oscillator 408.0943+-1.8188 406.8593+-2.2099 imaging-darkroom 362.6742+-3.5852 362.5699+-4.2088 imaging-desaturate 262.4619+-0.6562 261.4084+-0.8725 imaging-gaussian-blur 586.2568+-2.6599 ? 587.0692+-2.7495 ? json-parse-financial 59.4232+-0.3383 ! 61.4743+-0.2087 ! definitely 1.0345x slower json-stringify-tinderbox 93.6780+-0.2445 ^ 90.4414+-0.2895 ^ definitely 1.0358x faster stanford-crypto-aes 111.6194+-1.2202 110.6155+-0.5579 stanford-crypto-ccm 110.7395+-0.6315 ? 110.9770+-0.4050 ? stanford-crypto-pbkdf2 213.6550+-0.7054 ! 215.7886+-0.8343 ! definitely 1.0100x slower stanford-crypto-sha256-iterative 96.1172+-0.3435 ^ 95.2770+-0.2565 ^ definitely 1.0088x faster <arithmetic> * 266.1228+-0.5818 ? 266.5076+-0.6539 ? might be 1.0014x slower <geometric> 213.7936+-0.3803 213.7788+-0.4147 might be 1.0001x faster <harmonic> 167.0125+-0.3691 ? 167.1128+-0.2168 ? might be 1.0006x slower TipOfTree32 FixRegAllox32 All benchmarks: <arithmetic> 98.9174+-0.1807 ? 98.9578+-0.2273 ? might be 1.0004x slower <geometric> 24.8469+-0.0550 24.8010+-0.0530 might be 1.0019x faster <harmonic> 7.3956+-0.0410 7.3912+-0.0369 might be 1.0006x faster TipOfTree32 FixRegAllox32 Geomean of preferred means: <scaled-result> 56.1674+-0.0887 56.0664+-0.0733 might be 1.0018x faster [pizlo@nitroflex bencher]

Filip Pizlo

Comment 3 2011-12-18 20:54:41 PST

Created attachment 119805 [details] the patch

Filip Pizlo

Comment 4 2011-12-18 22:36:26 PST

Landed in http://trac.webkit.org/changeset/103218

Attachments
the patch (10.77 KB, patch) 2011-12-18 20:54 PST, Filip Pizlo	barraclough: review+	Details Formatted Diff Diff
View All Add attachment proposed patch, testcase, etc.