159759 2016-07-14 05:15:12 -0700 [ARM] Disable Inline Caching on ARMv7 traditional until proper fix 2016-08-02 08:38:47 -0700 1 1 1 Unclassified WebKit JavaScriptCore Other Unspecified Unspecified RESOLVED FIXED P2 Normal --- 159408 1 ossy ossy commit-queue keith_miller mark.lam msaboff ossy saam oldest_to_newest 1210950 0 ossy 2016-07-14 05:15:12 -0700 https://trac.webkit.org/changeset/202214 broke inline caching on ARMv7 traditional (ARM instruction set / ARMAssembler) bug159408 is to fix this regression step by step. But ARM traditional assembler is completely broken since a month ago, so I propose disabling inline caching on it until all fixes are reviewed and landed. 1210952 1 283632 ossy 2016-07-14 05:32:38 -0700 Created attachment 283632 Patch 1211368 2 283632 commit-queue 2016-07-15 02:44:22 -0700 Comment on attachment 283632 Patch Clearing flags on attachment: 283632 Committed r203272: <http://trac.webkit.org/changeset/203272> 1211369 3 commit-queue 2016-07-15 02:44:26 -0700 All reviewed patches have been landed. Closing bug. 1215513 4 ossy 2016-07-29 00:55:38 -0700 Reopening to attach new patch. 1215514 5 284851 ossy 2016-07-29 00:55:42 -0700 Created attachment 284851 Patch Let's disable MathIC's too until proper fix. The ARM traditional port have been broken 6 weeks ago and more fixes are needed, for example bug160295, bug159720. 1215963 6 284851 saam 2016-07-31 14:03:33 -0700 Comment on attachment 284851 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=284851&action=review > Source/JavaScriptCore/jit/JITMathIC.h:71 > +#if CPU(ARM_TRADITIONAL) > + // FIXME: Remove this workaround once the proper fixes are landed. > + // [ARM] Disable Inline Caching on ARMv7 traditional until proper fix > + // https://bugs.webkit.org/show_bug.cgi?id=159759 > + return false; > +#endif Actually, I realized a better solution is to force the result to be JITMathICInlineResult::GenerateFullSnippet. You can accomplish that by having a compile-time conditional around m_generator.generateInline(...) below. This will revert to the behavior of the old code. This will be much faster than forcing a call on every add/mul/etc. 1216073 7 ossy 2016-08-01 03:15:59 -0700 (In reply to comment #6) > Comment on attachment 284851 [details] > Patch > > View in context: > https://bugs.webkit.org/attachment.cgi?id=284851&action=review > > > Source/JavaScriptCore/jit/JITMathIC.h:71 > > +#if CPU(ARM_TRADITIONAL) > > + // FIXME: Remove this workaround once the proper fixes are landed. > > + // [ARM] Disable Inline Caching on ARMv7 traditional until proper fix > > + // https://bugs.webkit.org/show_bug.cgi?id=159759 > > + return false; > > +#endif > > Actually, I realized a better solution is to force the result to be > JITMathICInlineResult::GenerateFullSnippet. > You can accomplish that by having a compile-time conditional around > m_generator.generateInline(...) below. > This will revert to the behavior of the old code. This will be much faster > than forcing a call on every add/mul/etc. Did you mean a similar workaround? (of course with CPU(ARM_TRADITIONAL) guard) - JITMathICInlineResult result = m_generator.generateInline(jit, state); + JITMathICInlineResult result = JITMathICInlineResult::GenerateFullSnippet; I made a quick test, it decreased the number of crashes from 4500 to 3500, but it is so far from the optimal number zero. :( I'll check what causes the crashes on a debug build in 1-2 hours. 1216084 8 ossy 2016-08-01 04:29:34 -0700 I got the following crash backtraces with the patch you suggested: ASSERTION FAILED: linkBuffer.isValid() ../../Source/JavaScriptCore/jit/JITMathIC.h(130) : JSC::JITMathIC<Generator>::generateOutOfLine(JSC::VM&, JSC::CdeBlock*, JSC::FunctionPtr)::<lambda()> [with GeneratorType = JSC::JITAddGenerator] 1 0x41c53ad8 WTFCrash 2 0x41764034 JSC::JITMathIC<JSC::JITAddGenerator>::generateOutOfLine(JSC::VM&, JSC::CodeBlock*, JSC::FunctionPr)::{lambda()#1}::operator()() const 3 0x417643fc JSC::JITMathIC<JSC::JITAddGenerator>::generateOutOfLine(JSC::VM&, JSC::CodeBlock*, JSC::FunctionPr) 4 0x4175acb4 Illegal instruction ERROR: Unexpected exit code: 132 ---------------------- ASSERTION FAILED: static_cast<ptrdiff_t>(inlineSize) <= MacroAssembler::maxJumpReplacementSize() ../../Source/JavaScriptCore/jit/JITMathIC.h(81) : bool JSC::JITMathIC<Generator>::generateInline(JSC::CCallHelpers&, JSC::athICGenerationState&, bool) [with GeneratorType = JSC::JITAddGenerator] 1 0x41b65ad8 WTFCrash 2 0x41441788 JSC::JITMathIC<JSC::JITAddGenerator>::generateInline(JSC::CCallHelpers&, JSC::MathICGenerationState&, bool) 3 0x41643f50 void JSC::JIT::emitMathICFast<JSC::JITAddGenerator, long long (*)(JSC::ExecState*, long long, long long, JS::ArithProfile*), long long (*)(JSC::ExecState*, long long, long long)>(JSC::JITMathIC<JSC::JITAddGenerator>*, JSC::Instruction*, long long (*)(JSC::ExecStat*, long long, long long, JSC::ArithProfile*), long long (*)(JSC::ExecState*, long long, long long)) 4 0x41641148 JSC::JIT::emit_op_add(JSC::Instruction*) 5 0x4162f710 JSC::JIT::privateCompileMainPass() 6 0x41632868 JSC::JIT::compileWithoutLinking(JSC::JITCompilationEffort) 7 0x41633e34 JSC::JIT::privateCompile(JSC::JITCompilationEffort) 8 0x41691630 JSC::JIT::compile(JSC::VM*, JSC::CodeBlock*, JSC::JITCompilationEffort) 9 0x41861300 10 0x418614d8 JSC::ScriptExecutable::prepareForExecutionImpl(JSC::ExecState*, JSC::JSFunction*, JSC::JSScope*, JSC::CodeSecializationKind) 11 0x415fb35c JSC::ScriptExecutable::prepareForExecution(JSC::ExecState*, JSC::JSFunction*, JSC::JSScope*, JSC::CodeSpecilizationKind) 12 0x4166636c Illegal instruction ERROR: Unexpected exit code: 132 Do you have a better idea how to workaround this regression? Or should we push the previous change I suggested? 1216121 9 284851 saam 2016-08-01 08:56:39 -0700 Comment on attachment 284851 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=284851&action=review >>> Source/JavaScriptCore/jit/JITMathIC.h:71 >>> +#endif >> >> Actually, I realized a better solution is to force the result to be JITMathICInlineResult::GenerateFullSnippet. >> You can accomplish that by having a compile-time conditional around m_generator.generateInline(...) below. >> This will revert to the behavior of the old code. This will be much faster than forcing a call on every add/mul/etc. > > Did you mean a similar workaround? (of course with CPU(ARM_TRADITIONAL) guard) > - JITMathICInlineResult result = m_generator.generateInline(jit, state); > + JITMathICInlineResult result = JITMathICInlineResult::GenerateFullSnippet; > > I made a quick test, it decreased the number of crashes from > 4500 to 3500, but it is so far from the optimal number zero. :( > I'll check what causes the crashes on a debug build in 1-2 hours. Sorry, so you want to both do that and then also put the if (ArithProfile* ...) branch behind a compile flag so it doesn't execute in arm traditional 1216451 10 285097 ossy 2016-08-02 04:02:31 -0700 Created attachment 285097 Patch 1216456 11 ossy 2016-08-02 04:57:05 -0700 (In reply to comment #10) > Created attachment 285097 [details] > Patch Did you mean this workaround? Unfortunately it doesn't help too lot. There are still 1066 crashes with this patch, it means that all mozilla-tests.yaml/****.js.mozilla-baseline tests crashes because of "Illegal instruction". 1216460 12 ossy 2016-08-02 05:22:20 -0700 (In reply to comment #11) > (In reply to comment #10) > > Created attachment 285097 [details] > > Patch > > Did you mean this workaround? Unfortunately it doesn't help too lot. > There are still 1066 crashes with this patch, it means that all > mozilla-tests.yaml/****.js.mozilla-baseline tests crashes because > of "Illegal instruction". I still get the following assertions with this workaround: ASSERTION FAILED: linkBuffer.isValid() ../../Source/JavaScriptCore/jit/JITMathIC.h(137) : JSC::JITMathIC<Generator>::generateOutOfLine(JSC::VM&, JSC::CodeBlock*, JSC::FunctionPtr)::<lambda()> [with GeneratorType = JSC::JITAddGenerator] 1 0xb638eb44 WTFCrash 2 0xb5e9e0e4 JSC::JITMathIC<JSC::JITAddGenerator>::generateOutOfLine(JSC::VM&, JSC::CodeBlock*, JSC::FunctionPtr)::{lambda()#1}::operator()() const 3 0xb5e9e778 JSC::JITMathIC<JSC::JITAddGenerator>::generateOutOfLine(JSC::VM&, JSC::CodeBlock*, JSC::FunctionPtr) 4 0xb5e94ab4 Segmentation fault ERROR: Unexpected exit code: 139 1216461 13 284851 ossy 2016-08-02 05:30:09 -0700 Comment on attachment 284851 Patch The ARM assembler is broken since 6 weeks and I will take my holiday since now until the end of August and don't want to left ARM assembler broken for this long period. Please reconsider your previous r- on this workaround. 1216462 14 285097 ossy 2016-08-02 05:30:56 -0700 Comment on attachment 285097 Patch clearing r? , because it doesn't work 1216482 15 284851 commit-queue 2016-08-02 08:38:42 -0700 Comment on attachment 284851 Patch Clearing flags on attachment: 284851 Committed r204025: <http://trac.webkit.org/changeset/204025> 1216483 16 commit-queue 2016-08-02 08:38:47 -0700 All reviewed patches have been landed. Closing bug. 283632 2016-07-14 05:32:38 -0700 2016-07-29 00:55:31 -0700 Patch bug-159759-20160714053151.patch text/plain 1498 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMjAzMTA5CmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvQ2hhbmdlTG9nIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwppbmRleCBk NjA3MzI2NWQzM2JiMGViOThiMDk5YzQyYzhjZWU0NGRmMjJjZGJkLi44ZTQyODZmNjliZmVhNjFm NTY4ODFmZmYyNGVmNzc1Y2VhZDBiMTBjIDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNjcmlwdENv cmUvQ2hhbmdlTG9nCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKQEAgLTEs MyArMSwxMyBAQAorMjAxNi0wNy0xNCAgQ3NhYmEgT3N6dHJvZ29uw6FjICA8b3NzeUB3ZWJraXQu b3JnPgorCisgICAgICAgIFtBUk1dIERpc2FibGUgSW5saW5lIENhY2hpbmcgb24gQVJNdjcgdHJh ZGl0aW9uYWwgdW50aWwgcHJvcGVyIGZpeAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9y Zy9zaG93X2J1Zy5jZ2k/aWQ9MTU5NzU5CisKKyAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChP T1BTISkuCisKKyAgICAgICAgKiBqaXQvUmVwYXRjaC5jcHA6CisgICAgICAgIChKU0M6OmZvcmNl SUNGYWlsdXJlKToKKwogMjAxNi0wNy0xMSAgQ2FpbyBMaW1hICA8dGljYWlvbGltYUBnbWFpbC5j b20+CiAKICAgICAgICAgRUNNQVNjcmlwdCAyMDE2OiAlVHlwZWRBcnJheSUucHJvdG90eXBlLmlu Y2x1ZGVzIGltcGxlbWVudGF0aW9uCmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlwdENvcmUv aml0L1JlcGF0Y2guY3BwIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9SZXBhdGNoLmNwcApp bmRleCBhMDI5YzhhMjkyN2QxOWUzOThhYTY4NzcxNTg4Y2ZkZDg3YzI3YjBmLi5hYWNkNDU3OGEx NzMzODcxNzEyNTVlYWZhNWViOWZkMTE2MWUyYTA2IDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNj cmlwdENvcmUvaml0L1JlcGF0Y2guY3BwCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9qaXQv UmVwYXRjaC5jcHAKQEAgLTEyMSw3ICsxMjEsMTQgQEAgc3RhdGljIElubGluZUNhY2hlQWN0aW9u IGFjdGlvbkZvckNlbGwoVk0mIHZtLCBKU0NlbGwqIGNlbGwpCiAKIHN0YXRpYyBib29sIGZvcmNl SUNGYWlsdXJlKEV4ZWNTdGF0ZSopCiB7CisjaWYgQ1BVKEFSTV9UUkFESVRJT05BTCkKKyAgICAv LyBGSVhNRTogUmVtb3ZlIHRoaXMgd29ya2Fyb3VuZCBvbmNlIHRoZSBwcm9wZXIgZml4ZXMgYXJl IGxhbmRlZC4KKyAgICAvLyBbQVJNXSBEaXNhYmxlIElubGluZSBDYWNoaW5nIG9uIEFSTXY3IHRy YWRpdGlvbmFsIHVudGlsIHByb3BlciBmaXgKKyAgICAvLyBodHRwczovL2J1Z3Mud2Via2l0Lm9y Zy9zaG93X2J1Zy5jZ2k/aWQ9MTU5NzU5CisgICAgcmV0dXJuIHRydWU7CisjZWxzZQogICAgIHJl dHVybiBPcHRpb25zOjpmb3JjZUlDRmFpbHVyZSgpOworI2VuZGlmCiB9CiAKIGlubGluZSBKX0pJ VE9wZXJhdGlvbl9FU3NpSkkgYXBwcm9wcmlhdGVPcHRpbWl6aW5nR2V0QnlJZEZ1bmN0aW9uKEdl dEJ5SURLaW5kIGtpbmQpCg== 284851 2016-07-29 00:55:42 -0700 2016-08-02 08:38:42 -0700 Patch bug-159759-20160729005428.patch text/plain 1494 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMjAzODk2CmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvQ2hhbmdlTG9nIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwppbmRleCAz MWUzMWZmNjllMTcwNTM4ODFkZWUyMDNiNzA1MTE0ZGU3MWJjZDE0Li41OTc2YjU4MDAxN2ZmOTlk YWJkMjk0YmNmMTM0MjViMDMxZDJjZTQ1IDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNjcmlwdENv cmUvQ2hhbmdlTG9nCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKQEAgLTEs MyArMSwxMyBAQAorMjAxNi0wNy0yOSAgQ3NhYmEgT3N6dHJvZ29uw6FjICA8b3NzeUB3ZWJraXQu b3JnPgorCisgICAgICAgIFtBUk1dIERpc2FibGUgSW5saW5lIENhY2hpbmcgb24gQVJNdjcgdHJh ZGl0aW9uYWwgdW50aWwgcHJvcGVyIGZpeAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9y Zy9zaG93X2J1Zy5jZ2k/aWQ9MTU5NzU5CisKKyAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChP T1BTISkuCisKKyAgICAgICAgKiBqaXQvSklUTWF0aElDLmg6CisgICAgICAgIChKU0M6OkpJVE1h dGhJQzo6Z2VuZXJhdGVJbmxpbmUpOgorCiAyMDE2LTA3LTI4ICBZdXN1a2UgU3V6dWtpICA8dXRh dGFuZS50ZWFAZ21haWwuY29tPgogCiAgICAgICAgIEpTQzo6U3ltYm9sIHNob3VsZCBiZSBoYXNo LWNvbnNlZApkaWZmIC0tZ2l0IGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVRNYXRoSUMu aCBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9qaXQvSklUTWF0aElDLmgKaW5kZXggMDgyNDk4Yjhi YzRhYzAwOTU0NTNhYjBhMTRlYWNkOGFkMTJiNDQ2ZC4uYWY4ZDU5YzM3NGM4ZWIxNWYyOGRmNGJh MDIwMjBmODAzYzBjNjk2NSAxMDA2NDQKLS0tIGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9K SVRNYXRoSUMuaAorKysgYi9Tb3VyY2UvSmF2YVNjcmlwdENvcmUvaml0L0pJVE1hdGhJQy5oCkBA IC02Myw2ICs2MywxMyBAQCBwdWJsaWM6CiAKICAgICBib29sIGdlbmVyYXRlSW5saW5lKENDYWxs SGVscGVycyYgaml0LCBNYXRoSUNHZW5lcmF0aW9uU3RhdGUmIHN0YXRlLCBib29sIHNob3VsZEVt aXRQcm9maWxpbmcgPSB0cnVlKQogICAgIHsKKyNpZiBDUFUoQVJNX1RSQURJVElPTkFMKQorICAg ICAgICAvLyBGSVhNRTogUmVtb3ZlIHRoaXMgd29ya2Fyb3VuZCBvbmNlIHRoZSBwcm9wZXIgZml4 ZXMgYXJlIGxhbmRlZC4KKyAgICAgICAgLy8gW0FSTV0gRGlzYWJsZSBJbmxpbmUgQ2FjaGluZyBv biBBUk12NyB0cmFkaXRpb25hbCB1bnRpbCBwcm9wZXIgZml4CisgICAgICAgIC8vIGh0dHBzOi8v YnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD0xNTk3NTkKKyAgICAgICAgcmV0dXJuIGZh bHNlOworI2VuZGlmCisKICAgICAgICAgc3RhdGUuZmFzdFBhdGhTdGFydCA9IGppdC5sYWJlbCgp OwogICAgICAgICBzaXplX3Qgc3RhcnRTaXplID0gaml0Lm1fYXNzZW1ibGVyLmJ1ZmZlcigpLmNv ZGVTaXplKCk7CiAK 285097 2016-08-02 04:02:31 -0700 2016-08-02 05:30:56 -0700 Patch bug-159759-20160802040109.patch text/plain 1966 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMjA0MDIxCmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvQ2hhbmdlTG9nIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwppbmRleCA4 ODVjNWMwNTVhNDc4ODhmZTMwMzZhYWY2N2RiNDAyNTU4N2RmODRhLi5kMGM0ZDQ5M2FhZDUzYmI5 ZTIxOTFhYjRhYzI4NTRkMjFlZjJiYTBjIDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNjcmlwdENv cmUvQ2hhbmdlTG9nCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKQEAgLTEs MyArMSwxMyBAQAorMjAxNi0wOC0wMiAgQ3NhYmEgT3N6dHJvZ29uw6FjICA8b3NzeUB3ZWJraXQu b3JnPgorCisgICAgICAgIFtBUk1dIERpc2FibGUgSW5saW5lIENhY2hpbmcgb24gQVJNdjcgdHJh ZGl0aW9uYWwgdW50aWwgcHJvcGVyIGZpeAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9y Zy9zaG93X2J1Zy5jZ2k/aWQ9MTU5NzU5CisKKyAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChP T1BTISkuCisKKyAgICAgICAgKiBqaXQvSklUTWF0aElDLmg6CisgICAgICAgIChKU0M6OkpJVE1h dGhJQzo6Z2VuZXJhdGVJbmxpbmUpOgorCiAyMDE2LTA4LTAxICBGaWxpcCBQaXpsbyAgPGZwaXps b0BhcHBsZS5jb20+CiAKICAgICAgICAgUkVHUkVTU0lPTiAocjIwMzk5MCk6IEpTQyBEZWJ1ZyB0 ZXN0IHN0cmVzcy9hcml0eS1jaGVjay1mdGwtdGhyb3cuanMgZmFpbGluZwpkaWZmIC0tZ2l0IGEv U291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVRNYXRoSUMuaCBiL1NvdXJjZS9KYXZhU2NyaXB0 Q29yZS9qaXQvSklUTWF0aElDLmgKaW5kZXggNDA4YjRlY2RjNzkzMzFlN2JlYjE0MThkMzY1Yzhl NjI4ODAwMzI2Yy4uMTkzZDEyZDhlMGY3NWEwODdhNjE3OWQwNDgxMDUwMThlOTc5YTIwYyAxMDA2 NDQKLS0tIGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2ppdC9KSVRNYXRoSUMuaAorKysgYi9Tb3Vy Y2UvSmF2YVNjcmlwdENvcmUvaml0L0pJVE1hdGhJQy5oCkBAIC02Nyw2ICs2Nyw3IEBAIHB1Ymxp YzoKICAgICAgICAgc3RhdGUuZmFzdFBhdGhTdGFydCA9IGppdC5sYWJlbCgpOwogICAgICAgICBz aXplX3Qgc3RhcnRTaXplID0gaml0Lm1fYXNzZW1ibGVyLmJ1ZmZlcigpLmNvZGVTaXplKCk7CiAK KyNpZiAhQ1BVKEFSTV9UUkFESVRJT05BTCkKICAgICAgICAgaWYgKEFyaXRoUHJvZmlsZSogYXJp dGhQcm9maWxlID0gbV9nZW5lcmF0b3IuYXJpdGhQcm9maWxlKCkpIHsKICAgICAgICAgICAgIGlm IChhcml0aFByb2ZpbGUtPmxoc09ic2VydmVkVHlwZSgpLmlzRW1wdHkoKSB8fCBhcml0aFByb2Zp bGUtPnJoc09ic2VydmVkVHlwZSgpLmlzRW1wdHkoKSkgewogICAgICAgICAgICAgICAgIC8vIEl0 IGxvb2tzIGxpa2UgdGhlIE1hdGhJQyBoYXMgeWV0IHRvIGV4ZWN1dGUuIFdlIGRvbid0IHdhbnQg dG8gZW1pdCBjb2RlIGluIHRoaXMKQEAgLTg1LDYgKzg2LDEyIEBAIHB1YmxpYzoKICAgICAgICAg fQogCiAgICAgICAgIEpJVE1hdGhJQ0lubGluZVJlc3VsdCByZXN1bHQgPSBtX2dlbmVyYXRvci5n ZW5lcmF0ZUlubGluZShqaXQsIHN0YXRlKTsKKyNlbHNlCisgICAgICAgIC8vIEZJWE1FOiBSZW1v dmUgdGhpcyB3b3JrYXJvdW5kIG9uY2UgdGhlIHByb3BlciBmaXhlcyBhcmUgbGFuZGVkLgorICAg ICAgICAvLyBbQVJNXSBEaXNhYmxlIElubGluZSBDYWNoaW5nIG9uIEFSTXY3IHRyYWRpdGlvbmFs IHVudGlsIHByb3BlciBmaXgKKyAgICAgICAgLy8gaHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hv d19idWcuY2dpP2lkPTE1OTc1OQorICAgICAgICBKSVRNYXRoSUNJbmxpbmVSZXN1bHQgcmVzdWx0 ID0gSklUTWF0aElDSW5saW5lUmVzdWx0OjpHZW5lcmF0ZUZ1bGxTbmlwcGV0OworI2VuZGlmCiAK ICAgICAgICAgc3dpdGNoIChyZXN1bHQpIHsKICAgICAgICAgY2FzZSBKSVRNYXRoSUNJbmxpbmVS ZXN1bHQ6OkdlbmVyYXRlZEZhc3RQYXRoOiB7Cg==