158906 2016-06-18 12:28:50 -0700 REGRESSION(concurrent baseline JIT): Kraken/ai-astar runs 20% slower 2016-06-20 18:59:34 -0700 1 1 1 Unclassified WebKit JavaScriptCore WebKit Nightly Build All All RESOLVED FIXED P2 Normal --- 1 fpizlo fpizlo barraclough benjamin ggaren keith_miller mark.lam mhahnenb msaboff oliver rniwa saam sam oldest_to_newest 1203585 0 fpizlo 2016-06-18 12:28:50 -0700 So far this looks like it's because the baseline compile of the search# is delayed by >10ms. I'm still investigating why I do know that it's only the decision to compile baseline code concurrently that causes the slow-down, and not any of the other changes in the patch. 1203586 1 fpizlo 2016-06-18 12:32:03 -0700 Looks like we enqueue the search# plan and then it takes ~10ms for the compiler thread to wake up! 1203588 2 fpizlo 2016-06-18 12:43:18 -0700 This is fascinating! Here's what's happening: we enqueued a compile of <global>, which is a ginormous piece of global code, during the benchmark's unmeansured initialization phase. Then at the time that we try to compile search#, the JIT thread is busy compiling <global>. That takes a long time to compile and it generates a disgusting amount of code. <global> takes 16ms to compile and generates 8MB!! There are a bunch of solutions we could attempt: 1) If we want to compile something and the concurrent JIT thread is busy then compile on the main thread instead. 2) Launch multiple JIT threads. 3) Delay the compilation of <global> based on size heuristics the same way that we do for DFG and FTL tier-up. This might be hard because I think that <global> actually spends a significant amount of time in its silly loop. 4) Don't compile all of <global> with the baseline JIT. It tiers up because of a tiny loop at the end. I think I'll attempt (1). That seems like the easiest way of recovering this regression. I think it would be great to consider the other solutions, too. 1203590 3 fpizlo 2016-06-18 13:15:53 -0700 Attempt (1) seems to work for Kraken/ai-astar! Trunk (including concurrent baseline JIT): 111.2 ms With my fix: 92.6 ms With concurrent baseline JIT totally disabled: 87.9 ms I'll do more tests. 1203591 4 fpizlo 2016-06-18 13:22:19 -0700 I believe that this resolves the Kraken issue. Here are the numbers now. FROM = concurrent baseline JIT completely disabled. TO = my change, which uses synchronous compilation of the thread is busy. TEST COMPARISON FROM TO DETAILS ==================================================================================== ** TOTAL **: 1.013x as fast 826.1ms +/- 0.6% 815.5ms +/- 1.1% significant ==================================================================================== ai: *1.043x as slow* 85.3ms +/- 2.0% 89.0ms +/- 1.6% significant astar: *1.043x as slow* 85.3ms +/- 2.0% 89.0ms +/- 1.6% significant audio: 1.037x as fast 259.1ms +/- 1.1% 249.8ms +/- 2.9% significant beat-detection: ?? 58.1ms +/- 3.7% 59.1ms +/- 11.5% might be *1.017x as slow* dft: 1.100x as fast 103.1ms +/- 2.1% 93.7ms +/- 2.1% significant fft: - 43.9ms +/- 1.8% 42.9ms +/- 2.4% oscillator: ?? 54.0ms +/- 1.2% 54.1ms +/- 2.1% might be *1.002x as slow* imaging: ?? 185.1ms +/- 1.0% 185.6ms +/- 2.5% might be *1.003x as slow* gaussian-blur: - 64.2ms +/- 2.2% 62.2ms +/- 3.4% darkroom: ?? 72.9ms +/- 1.0% 73.5ms +/- 0.8% might be *1.008x as slow* desaturate: ?? 48.0ms +/- 5.1% 49.9ms +/- 6.5% might be *1.040x as slow* json: - 58.7ms +/- 1.3% 57.8ms +/- 1.3% parse-financial: - 35.2ms +/- 1.6% 34.8ms +/- 1.6% stringify-tinderbox: 1.022x as fast 23.5ms +/- 1.6% 23.0ms +/- 1.5% significant stanford: 1.020x as fast 237.9ms +/- 0.8% 233.3ms +/- 1.2% significant crypto-aes: 1.045x as fast 51.5ms +/- 1.9% 49.3ms +/- 1.2% significant crypto-ccm: - 48.4ms +/- 1.9% 48.3ms +/- 4.7% crypto-pbkdf2: 1.014x as fast 99.6ms +/- 1.3% 98.2ms +/- 0.6% significant crypto-sha256-iterative: 1.024x as fast 38.4ms +/- 1.3% 37.5ms +/- 1.3% significant 1203592 5 281618 fpizlo 2016-06-18 13:50:33 -0700 Created attachment 281618 the patch 1203605 6 281618 benjamin 2016-06-18 23:51:22 -0700 Comment on attachment 281618 the patch A bit nasty but let's try anyway. Especially since ARM hurts a lot and we don't have more cores. 1203613 7 fpizlo 2016-06-19 09:40:25 -0700 (In reply to comment #6) > Comment on attachment 281618 [details] > the patch > > A bit nasty but let's try anyway. Especially since ARM hurts a lot and we > don't have more cores. We had a similar thing arise with DFG vs FTL compilation, and we found that even on two cores, it's profitable to have DFG and FTL be in separate threads so that they interleave with each other in case the FTL takes a long time. I think that a similar thing applies here. Having additional baseline threads would likely help. It would be a generalization of what this patch does. But maybe what this patch does is even better, since most baseline compiles are fast enough that it's OK to just do them on the main thread when the helper is busy. 1203614 8 fpizlo 2016-06-19 09:45:04 -0700 Landed in r202205. I think. 1203986 9 rniwa 2016-06-20 18:59:34 -0700 I think we're ~3% regressed on iPhone 6 Plus. Will email you details. 281618 2016-06-18 13:50:33 -0700 2016-06-18 23:51:22 -0700 the patch blah.patch text/plain 7423 fpizlo SW5kZXg6IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkocmV2aXNpb24gMjAyMTk4KQorKysgU291cmNl L0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkod29ya2luZyBjb3B5KQpAQCAtMSwzICsxLDU0IEBA CisyMDE2LTA2LTE4ICBGaWxpcCBQaXpsbyAgPGZwaXpsb0BhcHBsZS5jb20+CisKKyAgICAgICAg UkVHUkVTU0lPTihjb25jdXJyZW50IGJhc2VsaW5lIEpJVCk6IEtyYWtlbi9haS1hc3RhciBydW5z IDIwJSBzbG93ZXIKKyAgICAgICAgaHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dp P2lkPTE1ODkwNgorCisgICAgICAgIFJldmlld2VkIGJ5IE5PQk9EWSAoT09QUyEpLgorICAgICAg ICAKKyAgICAgICAgVGhlIGNvbmN1cnJlbnQgYmFzZWxpbmUgSklUIHdhcyBhIDItMyUgcHJvZ3Jl c3Npb24gb24gSlNCZW5jaCwgcG9zc2libHkgYSAxJSBwcm9ncmVzc2lvbgorICAgICAgICBvbiBQ TFQzLCBidXQgYSAyLTUlIHJlZ3Jlc3Npb24gb24gS3Jha2VuLiBUaGlzIHBhdGNoIGZpeGVzIHRo ZSBLcmFrZW4gcmVncmVzc2lvbiB3aXRob3V0CisgICAgICAgIGFmZmVjdGluZyB0aGUgb3RoZXIg dGVzdHMuCisgICAgICAgIAorICAgICAgICBUaGUgcHJvYmxlbSBpcyB0aGF0IEtyYWtlbi9haS1h c3RhcidzIGluaXRpYWxpemF0aW9uIGNvZGUgaGFkIGEgZ2lub3Jtb3VzIHBpZWNlIG9mIGluaXQK KyAgICAgICAgY29kZSB0aGF0IHRvb2sgYWJvdXQgMTZtcyB0byBjb21waWxlIGluIGJhc2VsaW5l LiBUaGVyZSdzIG5vIGdvb2Qgd2F5IHRvIGF2b2lkIGxldHRpbmcgaXQKKyAgICAgICAgdGllci11 cCBpbnRvIGJhc2VsaW5lIHNpbmNlIGl0IGhhcyBhIGNvbXB1dGUgbG9vcC4gVGhlIHRpbWUgaXQg dGFrZXMgdG8gcnVuIHRoaXMgY29kZSBpcworICAgICAgICBuZXZlciBtZWFzdXJlZC4gVGhlIGNv bmN1cnJlbnQgYmFzZWxpbmUgSklUIGNhdXNlZCB1cyB0byBzY2hlZHVsZSB0aGUgY29tcGlsYXRp b24gb2YgdGhpcworICAgICAgICBodWdlIGNvZGUgcmF0aGVyIHRoYW4gZG9pbmcgaXQgZWFnZXJs eS4gVGhpcyBtZWFudCB0aGF0IGFmdGVyIGluaXRpYWxpemF0aW9uIHdhcyBkb25lIGFuZAorICAg ICAgICB3ZSBzdGFydGVkIGFjdHVhbGx5IHJ1bm5pbmcgcmVhbCBzdHVmZiwgYWxsIG9mIHRoZSBy ZWFsIHN0dWZmJ3MgY29tcGlsZXMgd291bGQgYmUgY29udm95ZWQKKyAgICAgICAgYmVoaW5kIHRo aXMgc3VwZXItZXhwZW5zaXZlIGJhc2VsaW5lIGNvbXBpbGUuIE5vdGUgdGhhdCBERkcgYW5kIEZU TCBjb21waWxlcyBjb252b3kgYmVoaW5kCisgICAgICAgIGJhc2VsaW5lIGNvbXBpbGVzLCBzaW5j ZSB5b3UgY2FuJ3Qgc2NoZWR1bGUgYSBERkcgY29tcGlsZSBmb3IgYSBjb2RlIGJsb2NrIHVudGls IHRoYXQgY29kZQorICAgICAgICBibG9jayBpcyBpbiBiYXNlbGluZS4KKyAgICAgICAgCisgICAg ICAgIFRoaXMgdXNlcyB0aGUgc2ltcGxlc3QgZml4OiBpZiB3ZSBhcmUgdGhpbmtpbmcgYWJvdXQg c2NoZWR1bGluZyBzb21lIGNvbXBpbGUgYW5kIHRoZQorICAgICAgICB0aHJlYWQgaXMgYnVzeSwg ZG8gdGhlIGNvbXBpbGUgb24gdGhlIG1haW4gdGhyZWFkIGluc3RlYWQuIFRoaXMgZG9lc24ndCBj b21wbGV0ZWx5CisgICAgICAgIGVsaW1pbmF0ZSB0aGUgYWktYXN0YXIgcmVncmVzc2lvbiAod2Ug c3RpbGwgaGF2ZSBhIDQlIHJlZ3Jlc3Npb24gb24gdGhhdCB0ZXN0KSBidXQgaXQgbm93CisgICAg ICAgIHJlc3VsdHMgaW4gY29uY3VycmVudCBiYXNlbGluZSBKSVQgYmVpbmcgYW4gb3ZlcmFsbCBw cm9ncmVzc2lvbiBvbiBLcmFrZW4gYXMgYSB3aG9sZSAoMSUKKyAgICAgICAgb24gbXkgbWFjaGlu ZSkuIFRoaXMgaXMgYmVjYXVzZSBjb25jdXJyZW50IGJhc2VsaW5lIGFwcGVhcnMgdG8gaGVscCBv biBvdGhlciB0ZXN0cy4KKworICAgICAgICBJbiB0aGUgZnV0dXJlLCB3ZSBjb3VsZCBmaXggdGhp cyBldmVuIGJldHRlciBieSBhbGxvd2luZyB0aGUgSklUV29ya2xpc3QgdG8gc3Bhd24gbW9yZQor ICAgICAgICB0aHJlYWRzIG9yIGJ5IGJlaW5nIHNtYXJ0ZXIgYWJvdXQgYmFzZWxpbmUgY29tcGls YXRpb24uIEkgdGhpbmsgaXQncyBuYXN0eSB0aGF0IGlmIGEgZ2lhbnQKKyAgICAgICAgcGllY2Ug b2YgaW5pdGlhbGl6YXRpb24gY29kZSBlbmRzIGluIGEgY29tcHV0ZSBsb29wLCB3ZSBjb21waWxl IGFsbCBvZiB0aGUgY29kZSBpbnN0ZWFkIG9mCisgICAgICAgIGp1c3QgdGhlIGxvb3AuIEl0J3Mg YWxzbyBncm9zcyB0aGF0IGEgY29uc3RhbnQtbGlrZSBvYmplY3QgY3JlYXRpb24gZXhwcmVzc2lv biB3aWxsIHJlc3VsdAorICAgICAgICBpbiBzbyBtdWNoIGNvZGUuIEl0IHdvdWxkIHJlc3VsdCBp biBsZXNzIGNvZGUgaWYgd2UgYWxsb3dlZCBvdXJzZWx2ZXMgdG8gZG8gYSBiaXQgbW9yZQorICAg ICAgICBzdGF0aWMgcmVhc29uaW5nIGFib3V0IG9iamVjdCBsaXRlcmFscy4KKyAgICAgICAgCisg ICAgICAgIEJ1dCBmb3Igbm93LCBJIHRoaW5rIHRoYXQgdGhpcyBpcyBhIGdyZWF0IHdheSB0byBy ZWNvdmVyIHRoZSBLcmFrZW4gcmVncmVzc2lvbiB3aGlsZSBzdGlsbAorICAgICAgICBrZWVwaW5n IHRoZSBvdGhlciBwcm9ncmVzc2lvbnMgZnJvbSBjb25jdXJyZW50IGJhc2VsaW5lLgorCisgICAg ICAgICogaml0L0pJVFdvcmtsaXN0LmNwcDoKKyAgICAgICAgKEpTQzo6SklUV29ya2xpc3Q6OlBs YW46OlBsYW4pOgorICAgICAgICAoSlNDOjpKSVRXb3JrbGlzdDo6UGxhbjo6Y29tcGlsZUluVGhy ZWFkKToKKyAgICAgICAgKEpTQzo6SklUV29ya2xpc3Q6OlBsYW46OmZpbmFsaXplKToKKyAgICAg ICAgKEpTQzo6SklUV29ya2xpc3Q6OlBsYW46OmNvZGVCbG9jayk6CisgICAgICAgIChKU0M6OkpJ VFdvcmtsaXN0OjpQbGFuOjppc0ZpbmlzaGVkQ29tcGlsaW5nKToKKyAgICAgICAgKEpTQzo6SklU V29ya2xpc3Q6OlBsYW46OmNvbXBpbGVOb3cpOgorICAgICAgICAoSlNDOjpKSVRXb3JrbGlzdDo6 SklUV29ya2xpc3QpOgorICAgICAgICAoSlNDOjpKSVRXb3JrbGlzdDo6Y29tcGlsZUxhdGVyKToK KyAgICAgICAgKEpTQzo6SklUV29ya2xpc3Q6OmNvbXBpbGVOb3cpOgorICAgICAgICAoSlNDOjpK SVRXb3JrbGlzdDo6cnVuVGhyZWFkKToKKyAgICAgICAgKEpTQzo6SklUV29ya2xpc3Q6OlBsYW46 OmlzRmluYWxpemVkKTogRGVsZXRlZC4KKyAgICAgICAgKiBqaXQvSklUV29ya2xpc3QuaDoKKwog MjAxNi0wNi0xNyAgQ29tbWl0IFF1ZXVlICA8Y29tbWl0LXF1ZXVlQHdlYmtpdC5vcmc+CiAKICAg ICAgICAgVW5yZXZpZXdlZCwgcm9sbGluZyBvdXQgcjIwMjE1Mi4KSW5kZXg6IFNvdXJjZS9KYXZh U2NyaXB0Q29yZS9qaXQvSklUV29ya2xpc3QuY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFNvdXJjZS9KYXZh U2NyaXB0Q29yZS9qaXQvSklUV29ya2xpc3QuY3BwCShyZXZpc2lvbiAyMDIxNzcpCisrKyBTb3Vy Y2UvSmF2YVNjcmlwdENvcmUvaml0L0pJVFdvcmtsaXN0LmNwcAkod29ya2luZyBjb3B5KQpAQCAt NzEsOSArNzEsNiBAQCBwdWJsaWM6CiAgICAgICAgICAgICBSRUxFQVNFX0FTU0VSVF9OT1RfUkVB Q0hFRCgpOwogICAgICAgICAgICAgcmV0dXJuOwogICAgICAgICB9Ci0gICAgICAgIAotICAgICAg ICBMb2NrSG9sZGVyIGxvY2tlcihtX2xvY2spOwotICAgICAgICBtX2lzRmluYWxpemVkID0gdHJ1 ZTsKICAgICB9CiAgICAgCiAgICAgQ29kZUJsb2NrKiBjb2RlQmxvY2soKSB7IHJldHVybiBtX2Nv ZGVCbG9jazsgfQpAQCAtODQsMTAgKzgxLDEyIEBAIHB1YmxpYzoKICAgICAgICAgTG9ja0hvbGRl ciBsb2NrZXIobV9sb2NrKTsKICAgICAgICAgcmV0dXJuIG1faXNGaW5pc2hlZENvbXBpbGluZzsK ICAgICB9Ci0gICAgYm9vbCBpc0ZpbmFsaXplZCgpCisgICAgCisgICAgc3RhdGljIHZvaWQgY29t cGlsZU5vdyhDb2RlQmxvY2sqIGNvZGVCbG9jaykKICAgICB7Ci0gICAgICAgIExvY2tIb2xkZXIg bG9ja2VyKG1fbG9jayk7Ci0gICAgICAgIHJldHVybiBtX2lzRmluYWxpemVkOworICAgICAgICBQ bGFuIHBsYW4oY29kZUJsb2NrKTsKKyAgICAgICAgcGxhbi5jb21waWxlSW5UaHJlYWQoKTsKKyAg ICAgICAgcGxhbi5maW5hbGl6ZSgpOwogICAgIH0KICAgICAKIHByaXZhdGU6CkBAIC05NSw3ICs5 NCw2IEBAIHByaXZhdGU6CiAgICAgSklUIG1faml0OwogICAgIExvY2sgbV9sb2NrOwogICAgIGJv b2wgbV9pc0ZpbmlzaGVkQ29tcGlsaW5nIHsgZmFsc2UgfTsKLSAgICBib29sIG1faXNGaW5hbGl6 ZWQgeyBmYWxzZSB9OwogfTsKIAogSklUV29ya2xpc3Q6OkpJVFdvcmtsaXN0KCkKQEAgLTE3Nywy NCArMTc1LDQ1IEBAIHZvaWQgSklUV29ya2xpc3Q6OmNvbXBpbGVMYXRlcihDb2RlQmxvY2sKICAg ICB9CiAgICAgCiAgICAgaWYgKCFPcHRpb25zOjp1c2VDb25jdXJyZW50SklUKCkpIHsKLSAgICAg ICAgUGxhbiBwbGFuKGNvZGVCbG9jayk7Ci0gICAgICAgIHBsYW4uY29tcGlsZUluVGhyZWFkKCk7 Ci0gICAgICAgIHBsYW4uZmluYWxpemUoKTsKKyAgICAgICAgUGxhbjo6Y29tcGlsZU5vdyhjb2Rl QmxvY2spOwogICAgICAgICByZXR1cm47CiAgICAgfQogICAgIAogICAgIGNvZGVCbG9jay0+aml0 U29vbigpOwotICAgIAotICAgIFBsYW5zIG15UGxhbnM7Ci0gICAgTG9ja0hvbGRlciBsb2NrZXIo bV9sb2NrKTsKLSAgICAKLSAgICBpZiAoIW1fcGxhbm5lZC5hZGQoY29kZUJsb2NrKS5pc05ld0Vu dHJ5KQotICAgICAgICByZXR1cm47Ci0gICAgCi0gICAgUmVmUHRyPFBsYW4+IHBsYW4gPSBhZG9w dFJlZihuZXcgUGxhbihjb2RlQmxvY2spKTsKLSAgICBtX3BsYW5zLmFwcGVuZChwbGFuKTsKLSAg ICBtX3F1ZXVlLmFwcGVuZChwbGFuKTsKLSAgICBtX2NvbmRpdGlvbi5ub3RpZnlBbGwoKTsKKwor ICAgIHsKKyAgICAgICAgTG9ja0hvbGRlciBsb2NrZXIobV9sb2NrKTsKKyAgICAgICAgCisgICAg ICAgIGlmIChtX3BsYW5uZWQuY29udGFpbnMoY29kZUJsb2NrKSkKKyAgICAgICAgICAgIHJldHVy bjsKKyAgICAgICAgCisgICAgICAgIGlmIChtX251bUF2YWlsYWJsZVRocmVhZHMpIHsKKyAgICAg ICAgICAgIG1fcGxhbm5lZC5hZGQoY29kZUJsb2NrKTsKKyAgICAgICAgICAgIFJlZlB0cjxQbGFu PiBwbGFuID0gYWRvcHRSZWYobmV3IFBsYW4oY29kZUJsb2NrKSk7CisgICAgICAgICAgICBtX3Bs YW5zLmFwcGVuZChwbGFuKTsKKyAgICAgICAgICAgIG1fcXVldWUuYXBwZW5kKHBsYW4pOworICAg ICAgICAgICAgbV9jb25kaXRpb24ubm90aWZ5QWxsKCk7CisgICAgICAgICAgICByZXR1cm47Cisg ICAgICAgIH0KKyAgICB9CisKKyAgICAvLyBDb21waWxpbmcgb24gdGhlIG1haW4gdGhyZWFkIGlm IHRoZSBoZWxwZXIgdGhyZWFkIGlzbid0IGF2YWlsYWJsZSBpcyBhIGRlZmVuc2UgYWdhaW5zdCB0 aGlzCisgICAgLy8gcGF0aG9sb2d5OgorICAgIC8vCisgICAgLy8gMSkgRG8gc29tZXRoaW5nIHRo YXQgaXMgYWxsb3dlZCB0byB0YWtlIGEgd2hpbGUsIGxpa2UgbG9hZCBhIGdpYW50IHBpZWNlIG9m IGluaXRpYWxpemF0aW9uCisgICAgLy8gICAgY29kZS4gVGhpcyBwbGFucyB0aGUgY29tcGlsZSBv ZiB0aGUgaW5pdCBjb2RlLCBidXQgZG9lc24ndCBmaW5pc2ggaXQuIEl0IHdpbGwgdGFrZSBhCisg ICAgLy8gICAgd2hpbGUuCisgICAgLy8KKyAgICAvLyAyKSBEbyBzb21ldGhpbmcgdGhhdCBpcyBz dXBwb3NlZCB0byBiZSBxdWljay4gTm93IGFsbCBiYXNlbGluZSBjb21waWxlcywgYW5kIHNvIGFs bCBERkcgYW5kCisgICAgLy8gICAgRlRMIGNvbXBpbGVzLCBvZiBldmVyeXRoaW5nIGlzIGJsb2Nr ZWQgb24gdGhlIGxvbmctcnVubmluZyBiYXNlbGluZSBjb21waWxlIG9mIHRoYXQKKyAgICAvLyAg ICBpbml0aWFsaXphdGlvbiBjb2RlLgorICAgIC8vCisgICAgLy8gVGhlIHNpbmdsZS10aHJlYWRl ZCBjb25jdXJyZW50IEpJVCBoYXMgdGhpcyB0ZW5kZW5jeSB0byBjb252b3kgZXZlcnl0aGluZyB3 aGlsZSBhdCB0aGUgc2FtZQorICAgIC8vIHRpbWUgcG9zdHBvbmluZyB3aGVuIGl0IGhhcHBlbnMs IHdoaWNoIG1lYW5zIHRoYXQgdGhlIGNvbnZveSBkZWxheXMgYXJlIGxlc3MgcHJlZGljdGFibGUu CisgICAgLy8gVGhpcyB3b3JrcyBhcm91bmQgdGhlIGlzc3VlLiBJZiB0aGUgY29uY3VycmVudCBK SVQgdGhyZWFkIGlzIGNvbnZveWVkLCB3ZSByZXZlcnQgdG8gbWFpbgorICAgIC8vIHRocmVhZCBj b21waWxlcy4gVGhpcyBpcyBwcm9iYWJseSBub3QgYXMgZ29vZCBhcyBpZiB3ZSBoYWQgbXVsdGlw bGUgSklUIHRocmVhZHMuIE1heWJlIHdlCisgICAgLy8gY2FuIGRvIHRoYXQgc29tZWRheS4KKyAg ICBQbGFuOjpjb21waWxlTm93KGNvZGVCbG9jayk7CiB9CiAKIHZvaWQgSklUV29ya2xpc3Q6OmNv bXBpbGVOb3coQ29kZUJsb2NrKiBjb2RlQmxvY2spCkBAIC0yMzUsOCArMjU0LDEwIEBAIHZvaWQg SklUV29ya2xpc3Q6OnJ1blRocmVhZCgpCiAgICAgICAgIFBsYW5zIG15UGxhbnM7CiAgICAgICAg IHsKICAgICAgICAgICAgIExvY2tIb2xkZXIgbG9ja2VyKG1fbG9jayk7CisgICAgICAgICAgICBt X251bUF2YWlsYWJsZVRocmVhZHMrKzsKICAgICAgICAgICAgIHdoaWxlIChtX3F1ZXVlLmlzRW1w dHkoKSkKICAgICAgICAgICAgICAgICBtX2NvbmRpdGlvbi53YWl0KG1fbG9jayk7CisgICAgICAg ICAgICBtX251bUF2YWlsYWJsZVRocmVhZHMtLTsKICAgICAgICAgICAgIAogICAgICAgICAgICAg Ly8gVGhpcyBpcyBhIGZ1biB3YXkgdG8gZGVxdWV1ZS4gSSBkb24ndCBrbm93IGlmIGl0J3MgYW55 IGJldHRlciBvciB3b3JzZSB0aGFuIGRlcXVldWluZwogICAgICAgICAgICAgLy8gb25lIHRoaW5n IGF0IGEgdGltZS4KSW5kZXg6IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9qaXQvSklUV29ya2xpc3Qu aAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09Ci0tLSBTb3VyY2UvSmF2YVNjcmlwdENvcmUvaml0L0pJVFdvcmtsaXN0LmgJ KHJldmlzaW9uIDIwMjE3NykKKysrIFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9qaXQvSklUV29ya2xp c3QuaAkod29ya2luZyBjb3B5KQpAQCAtNzMsNiArNzMsOCBAQCBwcml2YXRlOgogICAgIAogICAg IExvY2sgbV9sb2NrOwogICAgIENvbmRpdGlvbiBtX2NvbmRpdGlvbjsgLy8gV2UgdXNlIE9uZSBU cnVlIENvbmRpdGlvbiBmb3IgZXZlcnl0aGluZyBiZWNhdXNlIHRoYXQncyBlYXNpZXIuCisgICAg CisgICAgdW5zaWduZWQgbV9udW1BdmFpbGFibGVUaHJlYWRzIHsgMCB9OwogfTsKIAogfSAvLyBu YW1lc3BhY2UgSlNDCg==