69717 2011-10-09 01:02:13 -0700 Fix value profiling in 32_64 JIT 2011-10-09 12:11:50 -0700 1 1 1 Unclassified WebKit JavaScriptCore 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED P2 Normal --- 0 yuqiang.xian webkit-unassigned barraclough fpizlo webkit.review.bot oldest_to_newest 480670 0 yuqiang.xian 2011-10-09 01:02:13 -0700 Current value profiling for 32_64 JIT is broken and cannot record correct predicated types, which results in many speculation failures in the 32_64 DFG JIT, fallbacks to baseline JIT, and re-optimizations again and again. With this fix 32_64 DFG JIT can demonstrate real performance gains. 480673 1 110292 yuqiang.xian 2011-10-09 01:12:20 -0700 Created attachment 110292 the patch 480675 2 110292 fpizlo 2011-10-09 01:29:00 -0700 Comment on attachment 110292 the patch View in context: https://bugs.webkit.org/attachment.cgi?id=110292&action=review It would be great to see the performance impact, if any, on JSVALUE64 and JSVALUE32_64. > Source/JavaScriptCore/bytecode/ValueProfile.cpp:64 > + if (!value) { Have you made sure that these changes don't perturb SunSpider, V8, and Kraken performance on JSVALUE64? 480683 3 yuqiang.xian 2011-10-09 02:03:11 -0700 Performance result on Mac 64 - no difference with this change. Kraken TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: - 4142.9ms +/- 0.2% 4146.3ms +/- 0.4% ============================================================================= ai: - 802.3ms +/- 0.1% 802.7ms +/- 0.1% astar: - 802.3ms +/- 0.1% 802.7ms +/- 0.1% audio: ?? 1091.6ms +/- 0.3% 1093.9ms +/- 0.8% not conclusive: might be *1.002x as slow* beat-detection: - 222.0ms +/- 0.3% 221.3ms +/- 0.2% dft: ?? 440.9ms +/- 0.1% 445.0ms +/- 1.9% not conclusive: might be *1.009x as slow* fft: - 152.0ms +/- 0.2% 152.0ms +/- 0.0% oscillator: - 276.7ms +/- 1.2% 275.6ms +/- 0.3% imaging: ?? 1342.4ms +/- 0.8% 1343.9ms +/- 1.1% not conclusive: might be *1.001x as slow* gaussian-blur: ?? 604.2ms +/- 0.2% 606.8ms +/- 0.6% not conclusive: might be *1.004x as slow* darkroom: - 503.3ms +/- 2.1% 502.1ms +/- 2.6% desaturate: - 234.9ms +/- 0.1% 235.0ms +/- 0.1% json: 1.006x as fast 158.6ms +/- 0.2% 157.7ms +/- 0.3% significant parse-financial: - 69.2ms +/- 0.4% 68.9ms +/- 0.3% stringify-tinderbox: 1.007x as fast 89.4ms +/- 0.4% 88.8ms +/- 0.5% significant stanford: - 748.0ms +/- 0.6% 748.1ms +/- 0.4% crypto-aes: - 153.5ms +/- 0.5% 153.0ms +/- 0.5% crypto-ccm: - 137.7ms +/- 1.1% 137.7ms +/- 1.0% crypto-pbkdf2: ?? 358.0ms +/- 1.1% 358.4ms +/- 0.5% not conclusive: might be *1.001x as slow* crypto-sha256-iterative: ?? 98.8ms +/- 0.5% 99.0ms +/- 0.3% not conclusive: might be *1.002x as slow* SunSpider TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: - 185.4ms +/- 0.4% 184.6ms +/- 0.4% ============================================================================= 3d: - 27.1ms +/- 1.5% 26.9ms +/- 2.0% cube: - 10.0ms +/- 0.0% 10.0ms +/- 0.0% morph: - 8.0ms +/- 0.0% 8.0ms +/- 0.0% raytrace: - 9.1ms +/- 4.5% 8.9ms +/- 5.9% access: - 17.2ms +/- 3.3% 16.7ms +/- 4.5% binary-trees: - 2.5ms +/- 15.1% 2.5ms +/- 20.2% fannkuch: - 7.5ms +/- 6.7% 7.1ms +/- 3.2% nbody: - 4.2ms +/- 10.8% 4.0ms +/- 0.0% nsieve: ?? 3.0ms +/- 0.0% 3.1ms +/- 7.3% not conclusive: might be *1.033x as slow* bitops: - 14.3ms +/- 2.4% 14.0ms +/- 0.0% 3bit-bits-in-byte: - 1.0ms +/- 0.0% 1.0ms +/- 0.0% bits-in-byte: - 5.0ms +/- 0.0% 5.0ms +/- 0.0% bitwise-and: - 3.2ms +/- 9.4% 3.0ms +/- 0.0% nsieve-bits: - 5.1ms +/- 4.4% 5.0ms +/- 0.0% controlflow: - 2.0ms +/- 0.0% 2.0ms +/- 0.0% recursive: - 2.0ms +/- 0.0% 2.0ms +/- 0.0% crypto: - 12.0ms +/- 0.0% 12.0ms +/- 0.0% aes: - 7.0ms +/- 0.0% 7.0ms +/- 0.0% md5: - 3.0ms +/- 0.0% 3.0ms +/- 0.0% sha1: - 2.0ms +/- 0.0% 2.0ms +/- 0.0% date: - 22.0ms +/- 0.0% 22.0ms +/- 0.0% format-tofte: - 12.0ms +/- 0.0% 12.0ms +/- 0.0% format-xparb: - 10.0ms +/- 0.0% 10.0ms +/- 0.0% math: - 20.0ms +/- 0.0% 20.0ms +/- 0.0% cordic: - 7.0ms +/- 0.0% 7.0ms +/- 0.0% partial-sums: - 10.0ms +/- 0.0% 10.0ms +/- 0.0% spectral-norm: - 3.0ms +/- 0.0% 3.0ms +/- 0.0% regexp: ?? 12.8ms +/- 2.4% 13.0ms +/- 0.0% not conclusive: might be *1.016x as slow* dna: ?? 12.8ms +/- 2.4% 13.0ms +/- 0.0% not conclusive: might be *1.016x as slow* string: - 58.0ms +/- 0.0% 58.0ms +/- 0.0% base64: - 6.0ms +/- 0.0% 6.0ms +/- 0.0% fasta: - 7.0ms +/- 0.0% 7.0ms +/- 0.0% tagcloud: - 13.0ms +/- 0.0% 13.0ms +/- 0.0% unpack-code: - 25.0ms +/- 0.0% 25.0ms +/- 0.0% validate-input: - 7.0ms +/- 0.0% 7.0ms +/- 0.0% V8 Before Richards: 5786 DeltaBlue: 4463 Crypto: 13952 RayTrace: 7770 EarleyBoyer: 8285 RegExp: 1741 Splay: 5025 ---- Score (version 6): 5730 After Richards: 5952 DeltaBlue: 4450 Crypto: 13942 RayTrace: 8083 EarleyBoyer: 8240 RegExp: 1738 Splay: 5039 ---- Score (version 6): 5779 480684 4 yuqiang.xian 2011-10-09 02:06:10 -0700 Performance result on Linux IA32 with DFG JIT turned on. Only part of the kraken result is available. There are still some issues with v8 benchmarks. TEST COMPARISON FROM (ToT) TO (Mod) DETAILS ============================================================================= ** TOTAL **: 1.47x as fast 10246.2ms +/- 0.3% 6970.3ms +/- 0.1% significant ============================================================================= ai: 2.88x as fast 2270.8ms +/- 0.3% 789.7ms +/- 0.1% significant astar: 2.88x as fast 2270.8ms +/- 0.3% 789.7ms +/- 0.1% significant audio: 1.071x as fast 2427.6ms +/- 0.6% 2266.4ms +/- 0.2% significant beat-detection: 1.116x as fast 728.1ms +/- 0.5% 652.5ms +/- 0.5% significant dft: 1.081x as fast 602.8ms +/- 0.8% 557.6ms +/- 0.3% significant fft: 1.164x as fast 552.6ms +/- 0.5% 474.6ms +/- 0.1% significant oscillator: *1.069x as slow* 544.1ms +/- 0.8% 581.7ms +/- 0.1% significant imaging: 1.49x as fast 4376.0ms +/- 0.5% 2929.3ms +/- 0.1% significant gaussian-blur: 2.21x as fast 2843.7ms +/- 1.2% 1284.9ms +/- 0.1% significant darkroom: *1.135x as slow* 634.1ms +/- 1.8% 719.5ms +/- 0.6% significant desaturate: *1.030x as slow* 898.2ms +/- 0.3% 924.9ms +/- 0.2% significant json: - 215.1ms +/- 0.4% 214.8ms +/- 0.5% parse-financial: 1.021x as fast 98.5ms +/- 0.4% 96.5ms +/- 0.5% significant stringify-tinderbox: *1.015x as slow* 116.6ms +/- 0.5% 118.3ms +/- 0.7% significant stanford: 1.24x as fast 956.7ms +/- 0.4% 770.1ms +/- 0.3% significant crypto-aes: 1.105x as fast 189.7ms +/- 1.3% 171.6ms +/- 1.2% significant crypto-pbkdf2: 1.28x as fast 767.0ms +/- 0.4% 598.5ms +/- 0.1% significant 480685 5 fpizlo 2011-10-09 02:14:44 -0700 This looks great! By the way, you might find it easier to run all of the benchmarks in one go using Tools/Scripts/bencher. You need two checkouts, which you build separately: one for baseline (no changes), one with your changes. You also need ruby installed, and the json gem. And a ~/.bencher file. When you run bencher the first time, it will tell you how to set this file up. Then do: Tools/Scripts/bencher TipOfTree:<path-to-baseline-jsc> MyChanges:<path-to-your-jsc> Example: Tools/Scripts/bencher TipOfTree:/Volumes/Data/pizlo/OpenSource/WebKitBuild/Release/jsc CFA:/Volumes/Data/pizlo/secondary/OpenSource/WebKitBuild/Release/jsc I find that this makes it a bit faster to run benchmarks. But the way you ran them is fine, as well. 480691 6 110292 webkit.review.bot 2011-10-09 03:18:29 -0700 Comment on attachment 110292 the patch Clearing flags on attachment: 110292 Committed r97025: <http://trac.webkit.org/changeset/97025> 480692 7 webkit.review.bot 2011-10-09 03:18:33 -0700 All reviewed patches have been landed. Closing bug. 480711 8 yuqiang.xian 2011-10-09 07:09:58 -0700 Thanks for your hints, Filip. I will try the approach you suggested next time. BTW, one thing I think need to clarify is that the performance result of 32-bit DFG I pasted before is for the comparison between DFG w/o and w/ this patch (i.e. both have DFG turned on). Here's the comparsion between w/o DFG and w/ DFG after this change, tested on Linux ia32. We still have two cases failed to run (stanford-crypto-ccm produces wrong results and stanford-crypto-sha256-iterative crashes) and under investigations. TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: 1.39x as fast 9678.1ms +/- 0.0% 6970.3ms +/- 0.1% significant ============================================================================= ai: 2.68x as fast 2114.8ms +/- 0.1% 789.7ms +/- 0.1% significant astar: 2.68x as fast 2114.8ms +/- 0.1% 789.7ms +/- 0.1% significant audio: 1.037x as fast 2349.3ms +/- 0.1% 2266.4ms +/- 0.2% significant beat-detection: 1.090x as fast 711.5ms +/- 0.3% 652.5ms +/- 0.5% significant dft: 1.045x as fast 582.5ms +/- 0.4% 557.6ms +/- 0.3% significant fft: 1.113x as fast 528.4ms +/- 0.2% 474.6ms +/- 0.1% significant oscillator: *1.104x as slow* 526.9ms +/- 0.1% 581.7ms +/- 0.1% significant imaging: 1.39x as fast 4082.4ms +/- 0.1% 2929.3ms +/- 0.1% significant gaussian-blur: 1.98x as fast 2543.3ms +/- 0.1% 1284.9ms +/- 0.1% significant darkroom: *1.142x as slow* 630.0ms +/- 0.1% 719.5ms +/- 0.6% significant desaturate: *1.017x as slow* 909.1ms +/- 0.0% 924.9ms +/- 0.2% significant json: *1.013x as slow* 212.0ms +/- 0.4% 214.8ms +/- 0.5% significant parse-financial: *1.010x as slow* 95.5ms +/- 0.4% 96.5ms +/- 0.5% significant stringify-tinderbox: *1.015x as slow* 116.5ms +/- 0.5% 118.3ms +/- 0.7% significant stanford: 1.194x as fast 919.6ms +/- 0.2% 770.1ms +/- 0.3% significant crypto-aes: ?? 170.1ms +/- 0.3% 171.6ms +/- 1.2% not conclusive: might be *1.009x as slow* crypto-pbkdf2: 1.25x as fast 749.5ms +/- 0.2% 598.5ms +/- 0.1% significant 480730 9 fpizlo 2011-10-09 12:11:50 -0700 (In reply to comment #8) > Thanks for your hints, Filip. I will try the approach you suggested next time. > > BTW, one thing I think need to clarify is that the performance result of 32-bit DFG I pasted before is for the comparison between DFG w/o and w/ this patch (i.e. both have DFG turned on). > > Here's the comparsion between w/o DFG and w/ DFG after this change, tested on Linux ia32. We still have two cases failed to run (stanford-crypto-ccm produces wrong results and stanford-crypto-sha256-iterative crashes) and under investigations. Ah! Thanks for the clarification. > ** TOTAL **: 1.39x as fast 9678.1ms +/- 0.0% 6970.3ms +/- 0.1% significant This is starting to look good! 110292 2011-10-09 01:12:20 -0700 2011-10-09 03:18:29 -0700 the patch 69717.patch text/plain 8520 yuqiang.xian SW5kZXg6IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkocmV2aXNpb24gOTcwMjEpCisrKyBTb3VyY2Uv SmF2YVNjcmlwdENvcmUvQ2hhbmdlTG9nCSh3b3JraW5nIGNvcHkpCkBAIC0xLDMgKzEsMzYgQEAK KzIwMTEtMTAtMDkgIFl1cWlhbmcgWGlhbiAgPHl1cWlhbmcueGlhbkBpbnRlbC5jb20+CisKKyAg ICAgICAgRml4IHZhbHVlIHByb2ZpbGluZyBpbiAzMl82NCBKSVQKKyAgICAgICAgaHR0cHM6Ly9i dWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTY5NzE3CisKKyAgICAgICAgUmV2aWV3ZWQg YnkgTk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgQ3VycmVudCB2YWx1ZSBwcm9maWxpbmcgZm9y IDMyXzY0IEpJVCBpcyBicm9rZW4gYW5kIGNhbm5vdCByZWNvcmQKKyAgICAgICAgY29ycmVjdCBw cmVkaWNhdGVkIHR5cGVzLCB3aGljaCByZXN1bHRzIGluIG1hbnkgc3BlY3VsYXRpb24gZmFpbHVy ZXMKKyAgICAgICAgaW4gdGhlIDMyXzY0IERGRyBKSVQsIGZhbGxiYWNrcyB0byBiYXNlbGluZSBK SVQsIGFuZCByZS1vcHRpbWl6YXRpb25zCisgICAgICAgIGFnYWluIGFuZCBhZ2Fpbi4gCisgICAg ICAgIFdpdGggdGhpcyBmaXggMzJfNjQgREZHIEpJVCBjYW4gZGVtb25zdHJhdGUgcmVhbCBwZXJm b3JtYW5jZSBnYWlucy4KKworICAgICAgICAqIGJ5dGVjb2RlL1ZhbHVlUHJvZmlsZS5jcHA6Cisg ICAgICAgIChKU0M6OlZhbHVlUHJvZmlsZTo6Y29tcHV0ZVN0YXRpc3RpY3MpOgorICAgICAgICAq IGJ5dGVjb2RlL1ZhbHVlUHJvZmlsZS5oOgorICAgICAgICAoSlNDOjpWYWx1ZVByb2ZpbGU6OmNs YXNzSW5mbyk6CisgICAgICAgIChKU0M6OlZhbHVlUHJvZmlsZTo6bnVtYmVyT2ZTYW1wbGVzKToK KyAgICAgICAgKEpTQzo6VmFsdWVQcm9maWxlOjppc0xpdmUpOgorICAgICAgICAoSlNDOjpWYWx1 ZVByb2ZpbGU6Om51bWJlck9mSW50MzJzKToKKyAgICAgICAgKEpTQzo6VmFsdWVQcm9maWxlOjpu dW1iZXJPZkRvdWJsZXMpOgorICAgICAgICAoSlNDOjpWYWx1ZVByb2ZpbGU6Om51bWJlck9mQm9v bGVhbnMpOgorICAgICAgICAoSlNDOjpWYWx1ZVByb2ZpbGU6OmR1bXApOgorICAgICAgICAgICAg RW1wdHkgdmFsdWUgY2hlY2sgc2hvdWxkIGJlIHBlcmZvcm1lZCBvbiBkZWNvZGVkIEpTVmFsdWUs CisgICAgICAgICAgICBhcyBmb3IgMzJfNjQgZW1wdHkgdmFsdWUgaXMgbm90IGlkZW50aWNhbCB0 byBlbmNvZGVkIDAuCisgICAgICAgICogaml0L0pJVC5jcHA6CisgICAgICAgIChKU0M6OkpJVDo6 cHJpdmF0ZUNvbXBpbGUpOgorICAgICAgICAqIGppdC9KSVRJbmxpbmVNZXRob2RzLmg6CisgICAg ICAgIChKU0M6OkpJVDo6ZW1pdFZhbHVlUHJvZmlsaW5nU2l0ZSk6CisgICAgICAgICogaml0L0pJ VFN0dWJDYWxsLmg6CisgICAgICAgIChKU0M6OkpJVFN0dWJDYWxsOjpjYWxsV2l0aFZhbHVlUHJv ZmlsaW5nKToKKyAgICAgICAgICAgIFJlY29yZCB0aGUgcmlnaHQgcHJvZmlsaW5nIHJlc3VsdCBm b3IgMzJfNjQuCisKIDIwMTEtMTAtMDggIE1hcmsgSGFobmVuYmVyZyAgPG1oYWhuZW5iZXJnQGFw cGxlLmNvbT4KIAogICAgICAgICBBZGQgc3RhdGljIHZlcnNpb24gb2YgSlNDZWxsOjpnZXRDb25z dHJ1Y3REYXRhCkluZGV4OiBTb3VyY2UvSmF2YVNjcmlwdENvcmUvYnl0ZWNvZGUvVmFsdWVQcm9m aWxlLmNwcAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09Ci0tLSBTb3VyY2UvSmF2YVNjcmlwdENvcmUvYnl0ZWNvZGUvVmFs dWVQcm9maWxlLmNwcAkocmV2aXNpb24gOTcwMTcpCisrKyBTb3VyY2UvSmF2YVNjcmlwdENvcmUv Ynl0ZWNvZGUvVmFsdWVQcm9maWxlLmNwcAkod29ya2luZyBjb3B5KQpAQCAtNjAsNyArNjAsOCBA QCB2b2lkIFZhbHVlUHJvZmlsZTo6Y29tcHV0ZVN0YXRpc3RpY3MoY29uCiB2b2lkIFZhbHVlUHJv ZmlsZTo6Y29tcHV0ZVN0YXRpc3RpY3MoU3RhdGlzdGljcyYgc3RhdGlzdGljcykgY29uc3QKIHsK ICAgICBmb3IgKHVuc2lnbmVkIGkgPSAwOyBpIDwgbnVtYmVyT2ZCdWNrZXRzOyArK2kpIHsKLSAg ICAgICAgaWYgKCFtX2J1Y2tldHNbaV0pIHsKKyAgICAgICAgSlNWYWx1ZSB2YWx1ZSA9IEpTVmFs dWU6OmRlY29kZShtX2J1Y2tldHNbaV0pOworICAgICAgICBpZiAoIXZhbHVlKSB7CiAgICAgICAg ICAgICBXZWFrQnVja2V0IHdlYWtCdWNrZXQgPSBtX3dlYWtCdWNrZXRzW2ldOwogICAgICAgICAg ICAgaWYgKCEhd2Vha0J1Y2tldCkgewogICAgICAgICAgICAgICAgIHN0YXRpc3RpY3Muc2FtcGxl cysrOwpAQCAtNzIsNyArNzMsNiBAQCB2b2lkIFZhbHVlUHJvZmlsZTo6Y29tcHV0ZVN0YXRpc3Rp Y3MoU3RhCiAgICAgICAgIAogICAgICAgICBzdGF0aXN0aWNzLnNhbXBsZXMrKzsKICAgICAgICAg Ci0gICAgICAgIEpTVmFsdWUgdmFsdWUgPSBKU1ZhbHVlOjpkZWNvZGUobV9idWNrZXRzW2ldKTsK ICAgICAgICAgaWYgKHZhbHVlLmlzSW50MzIoKSkKICAgICAgICAgICAgIHN0YXRpc3RpY3MuaW50 MzJzKys7CiAgICAgICAgIGVsc2UgaWYgKHZhbHVlLmlzRG91YmxlKCkpCkluZGV4OiBTb3VyY2Uv SmF2YVNjcmlwdENvcmUvYnl0ZWNvZGUvVmFsdWVQcm9maWxlLmgKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL2J5dGVjb2RlL1ZhbHVlUHJvZmlsZS5oCShyZXZpc2lvbiA5NzAx NykKKysrIFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9ieXRlY29kZS9WYWx1ZVByb2ZpbGUuaAkod29y a2luZyBjb3B5KQpAQCAtNTUsOCArNTUsOCBAQCBzdHJ1Y3QgVmFsdWVQcm9maWxlIHsKICAgICAK ICAgICBjb25zdCBDbGFzc0luZm8qIGNsYXNzSW5mbyh1bnNpZ25lZCBidWNrZXQpIGNvbnN0CiAg ICAgewotICAgICAgICBpZiAoISFtX2J1Y2tldHNbYnVja2V0XSkgewotICAgICAgICAgICAgSlNW YWx1ZSB2YWx1ZSA9IEpTVmFsdWU6OmRlY29kZShtX2J1Y2tldHNbYnVja2V0XSk7CisgICAgICAg IEpTVmFsdWUgdmFsdWUgPSBKU1ZhbHVlOjpkZWNvZGUobV9idWNrZXRzW2J1Y2tldF0pOworICAg ICAgICBpZiAoISF2YWx1ZSkgewogICAgICAgICAgICAgaWYgKCF2YWx1ZS5pc0NlbGwoKSkKICAg ICAgICAgICAgICAgICByZXR1cm4gMDsKICAgICAgICAgICAgIHJldHVybiB2YWx1ZS5hc0NlbGwo KS0+c3RydWN0dXJlKCktPmNsYXNzSW5mbygpOwpAQCAtNjgsNyArNjgsNyBAQCBzdHJ1Y3QgVmFs dWVQcm9maWxlIHsKICAgICB7CiAgICAgICAgIHVuc2lnbmVkIHJlc3VsdCA9IDA7CiAgICAgICAg IGZvciAodW5zaWduZWQgaSA9IDA7IGkgPCBudW1iZXJPZkJ1Y2tldHM7ICsraSkgewotICAgICAg ICAgICAgaWYgKCEhbV9idWNrZXRzW2ldIHx8ICEhbV93ZWFrQnVja2V0c1tpXSkKKyAgICAgICAg ICAgIGlmICghIUpTVmFsdWU6OmRlY29kZShtX2J1Y2tldHNbaV0pIHx8ICEhbV93ZWFrQnVja2V0 c1tpXSkKICAgICAgICAgICAgICAgICByZXN1bHQrKzsKICAgICAgICAgfQogICAgICAgICByZXR1 cm4gcmVzdWx0OwpAQCAtODIsNyArODIsNyBAQCBzdHJ1Y3QgVmFsdWVQcm9maWxlIHsKICAgICBi b29sIGlzTGl2ZSgpIGNvbnN0CiAgICAgewogICAgICAgICBmb3IgKHVuc2lnbmVkIGkgPSAwOyBp IDwgbnVtYmVyT2ZCdWNrZXRzOyArK2kpIHsKLSAgICAgICAgICAgIGlmICghIW1fYnVja2V0c1tp XSB8fCAhIW1fd2Vha0J1Y2tldHNbaV0pCisgICAgICAgICAgICBpZiAoISFKU1ZhbHVlOjpkZWNv ZGUobV9idWNrZXRzW2ldKSB8fCAhIW1fd2Vha0J1Y2tldHNbaV0pCiAgICAgICAgICAgICAgICAg cmV0dXJuIHRydWU7CiAgICAgICAgIH0KICAgICAgICAgcmV0dXJuIGZhbHNlOwpAQCAtOTksNyAr OTksNyBAQCBzdHJ1Y3QgVmFsdWVQcm9maWxlIHsKICAgICB7CiAgICAgICAgIHVuc2lnbmVkIHJl c3VsdCA9IDA7CiAgICAgICAgIGZvciAodW5zaWduZWQgaSA9IDA7IGkgPCBudW1iZXJPZkJ1Y2tl dHM7ICsraSkgewotICAgICAgICAgICAgaWYgKCEhbV9idWNrZXRzW2ldICYmIEpTVmFsdWU6OmRl Y29kZShtX2J1Y2tldHNbaV0pLmlzSW50MzIoKSkKKyAgICAgICAgICAgIGlmIChKU1ZhbHVlOjpk ZWNvZGUobV9idWNrZXRzW2ldKS5pc0ludDMyKCkpCiAgICAgICAgICAgICAgICAgcmVzdWx0Kys7 CiAgICAgICAgIH0KICAgICAgICAgcmV0dXJuIHJlc3VsdDsKQEAgLTEwOSw3ICsxMDksNyBAQCBz dHJ1Y3QgVmFsdWVQcm9maWxlIHsKICAgICB7CiAgICAgICAgIHVuc2lnbmVkIHJlc3VsdCA9IDA7 CiAgICAgICAgIGZvciAodW5zaWduZWQgaSA9IDA7IGkgPCBudW1iZXJPZkJ1Y2tldHM7ICsraSkg ewotICAgICAgICAgICAgaWYgKCEhbV9idWNrZXRzW2ldICYmIEpTVmFsdWU6OmRlY29kZShtX2J1 Y2tldHNbaV0pLmlzRG91YmxlKCkpCisgICAgICAgICAgICBpZiAoSlNWYWx1ZTo6ZGVjb2RlKG1f YnVja2V0c1tpXSkuaXNEb3VibGUoKSkKICAgICAgICAgICAgICAgICByZXN1bHQrKzsKICAgICAg ICAgfQogICAgICAgICByZXR1cm4gcmVzdWx0OwpAQCAtMTcwLDcgKzE3MCw3IEBAIHN0cnVjdCBW YWx1ZVByb2ZpbGUgewogICAgIHsKICAgICAgICAgdW5zaWduZWQgcmVzdWx0ID0gMDsKICAgICAg ICAgZm9yICh1bnNpZ25lZCBpID0gMDsgaSA8IG51bWJlck9mQnVja2V0czsgKytpKSB7Ci0gICAg ICAgICAgICBpZiAoISFtX2J1Y2tldHNbaV0gJiYgSlNWYWx1ZTo6ZGVjb2RlKG1fYnVja2V0c1tp XSkuaXNCb29sZWFuKCkpCisgICAgICAgICAgICBpZiAoSlNWYWx1ZTo6ZGVjb2RlKG1fYnVja2V0 c1tpXSkuaXNCb29sZWFuKCkpCiAgICAgICAgICAgICAgICAgcmVzdWx0Kys7CiAgICAgICAgIH0K ICAgICAgICAgcmV0dXJuIHJlc3VsdDsKQEAgLTIzOCw3ICsyMzgsOCBAQCBzdHJ1Y3QgVmFsdWVQ cm9maWxlIHsKICAgICAgICAgICAgICAgICBwcmVkaWN0aW9uVG9TdHJpbmcobV9wcmVkaWN0aW9u KSwgbV9udW1iZXJPZlNhbXBsZXNJblByZWRpY3Rpb24pOwogICAgICAgICBib29sIGZpcnN0ID0g dHJ1ZTsKICAgICAgICAgZm9yICh1bnNpZ25lZCBpID0gMDsgaSA8IG51bWJlck9mQnVja2V0czsg KytpKSB7Ci0gICAgICAgICAgICBpZiAoISFtX2J1Y2tldHNbaV0gfHwgISFtX3dlYWtCdWNrZXRz W2ldKSB7CisgICAgICAgICAgICBKU1ZhbHVlIHZhbHVlID0gSlNWYWx1ZTo6ZGVjb2RlKG1fYnVj a2V0c1tpXSk7CisgICAgICAgICAgICBpZiAoISF2YWx1ZSB8fCAhIW1fd2Vha0J1Y2tldHNbaV0p IHsKICAgICAgICAgICAgICAgICBpZiAoZmlyc3QpIHsKICAgICAgICAgICAgICAgICAgICAgZnBy aW50ZihvdXQsICI6ICIpOwogICAgICAgICAgICAgICAgICAgICBmaXJzdCA9IGZhbHNlOwpAQCAt MjQ2LDggKzI0Nyw4IEBAIHN0cnVjdCBWYWx1ZVByb2ZpbGUgewogICAgICAgICAgICAgICAgICAg ICBmcHJpbnRmKG91dCwgIiwgIik7CiAgICAgICAgICAgICB9CiAgICAgICAgICAgICAKLSAgICAg ICAgICAgIGlmICghIW1fYnVja2V0c1tpXSkKLSAgICAgICAgICAgICAgICBmcHJpbnRmKG91dCwg IiVzIiwgSlNWYWx1ZTo6ZGVjb2RlKG1fYnVja2V0c1tpXSkuZGVzY3JpcHRpb24oKSk7CisgICAg ICAgICAgICBpZiAoISF2YWx1ZSkKKyAgICAgICAgICAgICAgICBmcHJpbnRmKG91dCwgIiVzIiwg dmFsdWUuZGVzY3JpcHRpb24oKSk7CiAgICAgICAgICAgICAKICAgICAgICAgICAgIGlmICghIW1f d2Vha0J1Y2tldHNbaV0pCiAgICAgICAgICAgICAgICAgZnByaW50ZihvdXQsICJEZWFkQ2VsbCIp OwpJbmRleDogU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVQuY3BwCj09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0t IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9qaXQvSklULmNwcAkocmV2aXNpb24gOTcwMTcpCisrKyBT b3VyY2UvSmF2YVNjcmlwdENvcmUvaml0L0pJVC5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTU3Myw3 ICs1NzMsMTIgQEAgSklUQ29kZSBKSVQ6OnByaXZhdGVDb21waWxlKENvZGVQdHIqIGZ1bgogICAg ICAgICAgICAgICAgICAgICAmJiBhcmd1bWVudFJlZ2lzdGVyID09IC1SZWdpc3RlckZpbGU6OkNh bGxGcmFtZUhlYWRlclNpemUgLSBtX2NvZGVCbG9jay0+bV9udW1QYXJhbWV0ZXJzKQogICAgICAg ICAgICAgICAgICAgICBtX2NvZGVCbG9jay0+YWRkVmFsdWVQcm9maWxlKC0xKTsKICAgICAgICAg ICAgICAgICBlbHNlIHsKKyNpZiBVU0UoSlNWQUxVRTY0KQogICAgICAgICAgICAgICAgICAgICBs b2FkUHRyKEFkZHJlc3MoY2FsbEZyYW1lUmVnaXN0ZXIsIGFyZ3VtZW50UmVnaXN0ZXIgKiBzaXpl b2YoUmVnaXN0ZXIpKSwgcmVnVDApOworI2VsaWYgVVNFKEpTVkFMVUUzMl82NCkKKyAgICAgICAg ICAgICAgICAgICAgbG9hZDMyKEFkZHJlc3MoY2FsbEZyYW1lUmVnaXN0ZXIsIGFyZ3VtZW50UmVn aXN0ZXIgKiBzaXplb2YoUmVnaXN0ZXIpICsgT0JKRUNUX09GRlNFVE9GKEpTVmFsdWUsIHUuYXNC aXRzLnBheWxvYWQpKSwgcmVnVDApOworICAgICAgICAgICAgICAgICAgICBsb2FkMzIoQWRkcmVz cyhjYWxsRnJhbWVSZWdpc3RlciwgYXJndW1lbnRSZWdpc3RlciAqIHNpemVvZihSZWdpc3Rlcikg KyBPQkpFQ1RfT0ZGU0VUT0YoSlNWYWx1ZSwgdS5hc0JpdHMudGFnKSksIHJlZ1QxKTsKKyNlbmRp ZgogICAgICAgICAgICAgICAgICAgICBlbWl0VmFsdWVQcm9maWxpbmdTaXRlKEZpcnN0UHJvZmls aW5nU2l0ZSk7CiAgICAgICAgICAgICAgICAgfQogICAgICAgICAgICAgfQpJbmRleDogU291cmNl L0phdmFTY3JpcHRDb3JlL2ppdC9KSVRJbmxpbmVNZXRob2RzLmgKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVRJbmxpbmVNZXRob2RzLmgJKHJldmlzaW9uIDk3MDE3 KQorKysgU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVRJbmxpbmVNZXRob2RzLmgJKHdvcmtp bmcgY29weSkKQEAgLTQ3MSw3ICs0NzEsMTMgQEAgaW5saW5lIHZvaWQgSklUOjplbWl0VmFsdWVQ cm9maWxpbmdTaXRlKAogICAgICAgICBhZGQzMihJbW0zMigzKSwgYnVja2V0Q291bnRlclJlZ2lz dGVyKTsKICAgICBhbmQzMihJbW0zMihWYWx1ZVByb2ZpbGU6OmJ1Y2tldEluZGV4TWFzayksIGJ1 Y2tldENvdW50ZXJSZWdpc3Rlcik7CiAgICAgbW92ZShJbW1QdHIodmFsdWVQcm9maWxlLT5tX2J1 Y2tldHMpLCBzY3JhdGNoKTsKKyNpZiBVU0UoSlNWQUxVRTY0KQogICAgIHN0b3JlUHRyKHZhbHVl LCBCYXNlSW5kZXgoc2NyYXRjaCwgYnVja2V0Q291bnRlclJlZ2lzdGVyLCBUaW1lc0VpZ2h0KSk7 CisjZWxpZiBVU0UoSlNWQUxVRTMyXzY0KQorICAgIGNvbnN0IFJlZ2lzdGVySUQgdmFsdWVUYWcg PSByZWdUMTsKKyAgICBzdG9yZTMyKHZhbHVlLCBCYXNlSW5kZXgoc2NyYXRjaCwgYnVja2V0Q291 bnRlclJlZ2lzdGVyLCBUaW1lc0VpZ2h0LCBPQkpFQ1RfT0ZGU0VUT0YoSlNWYWx1ZSwgdS5hc0Jp dHMucGF5bG9hZCkpKTsKKyAgICBzdG9yZTMyKHZhbHVlVGFnLCBCYXNlSW5kZXgoc2NyYXRjaCwg YnVja2V0Q291bnRlclJlZ2lzdGVyLCBUaW1lc0VpZ2h0LCBPQkpFQ1RfT0ZGU0VUT0YoSlNWYWx1 ZSwgdS5hc0JpdHMudGFnKSkpOworI2VuZGlmCiB9CiAjZW5kaWYKIApJbmRleDogU291cmNlL0ph dmFTY3JpcHRDb3JlL2ppdC9KSVRTdHViQ2FsbC5oCj09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFNvdXJjZS9KYXZh U2NyaXB0Q29yZS9qaXQvSklUU3R1YkNhbGwuaAkocmV2aXNpb24gOTcwMTcpCisrKyBTb3VyY2Uv SmF2YVNjcmlwdENvcmUvaml0L0pJVFN0dWJDYWxsLmgJKHdvcmtpbmcgY29weSkKQEAgLTIwNSw2 ICsyMDUsOSBAQCBuYW1lc3BhY2UgSlNDIHsKICAgICAgICAgewogICAgICAgICAgICAgQVNTRVJU KG1fcmV0dXJuVHlwZSA9PSBWYWx1ZSB8fCBtX3JldHVyblR5cGUgPT0gQ2VsbCk7CiAgICAgICAg ICAgICBKSVQ6OkNhbGwgY2FsbCA9IHRoaXMtPmNhbGwoKTsKKyAgICAgICAgICAgIEFTU0VSVChK SVQ6OnJldHVyblZhbHVlUmVnaXN0ZXIgPT0gSklUOjpyZWdUMCk7CisgICAgICAgICAgICBpZiAo bV9yZXR1cm5UeXBlID09IENlbGwpCisgICAgICAgICAgICAgICAgbV9qaXQtPm1vdmUoSklUOjpU cnVzdGVkSW1tMzIoSlNWYWx1ZTo6Q2VsbFRhZyksIEpJVDo6cmVnVDEpOwogICAgICAgICAgICAg bV9qaXQtPmVtaXRWYWx1ZVByb2ZpbGluZ1NpdGUoa2luZCk7CiAgICAgICAgICAgICBpZiAobV9y ZXR1cm5UeXBlID09IFZhbHVlKQogICAgICAgICAgICAgICAgIG1faml0LT5lbWl0U3RvcmUoZHN0 LCBKSVQ6OnJlZ1QxLCBKSVQ6OnJlZ1QwKTsK