nephele
/
haikuwebkit
1
0
Fork 0
Code Issues Releases Activity
haikuwebkit/LayoutTests/js/dom/prototype-chain-caching-wit...

11 lines
313 B
HTML
Raw Permalink Normal View History

Baseline JIT should be concurrent https://bugs.webkit.org/show_bug.cgi?id=158755 Reviewed by Geoffrey Garen. Source/JavaScriptCore: This makes the baseline JIT concurrent. We want it to be concurrent because it takes up about 1% of PLT3 and 10% of JSBench (though the JSBench number might be down from recent optimizations). The idea is really simple: I separated the compile and link phases of JIT::privateCompile(), and arranged to call the compile phase from another thread. This doesn't reuse the old DFG::Worklist code, because that code does things we don't need (like compilation plan cancellation to allow GC to interleave with compilations) and is structured in a way that would have required more changes to the baseline JIT. Also, I think that code uses the wrong API, and as a result, clients of that API have a bad time. For example, it's never clear who has the responsibility of setting the JIT thresholds and the DFG::Worklist goes to great lengths to try to help its client set those things correctly, but since it doesn't set them directly, the client then has to have additional complex logic to combine what it learned from the Worklist and what it knows to set the thresholds. This patch takes a simpler approach: the JITWorklist takes complete control over scheduling compilations. It's like a combination of DFG::Worklist and operationOptimize(). Because the baseline JIT runs quickly, we can take some shortcuts. The JITWorklist requires that all of its plans complete before a GC begins. This ensures that we don't have to worry about interactions between the concurrent baseline JIT and the GC. I needed to do a bunch of minor changes to the JIT to handle the races that emerged. For example, I needed to do things to opcodes that read profiling both in the main path code generator and the slow path one. One trick I used was to create a copy of the instruction stream and provide that for anyone interested in the original value of the profiles. Most code still uses the CodeBlock's instruction stream because it may emit JIT code that points at the stream. This also fixes a LLInt bug in prototype caching. This bug was revealed by this change because more of our LayoutTests now run in LLInt. This looks like it might be a ~1% Octane speed-up (on command line) and a ~0.7% PLT3 speed-up. This also looks like a ~2% JSBench speed-up. * CMakeLists.txt: * JavaScriptCore.xcodeproj/project.pbxproj: * debugger/Debugger.cpp: (JSC::Debugger::setSteppingMode): (JSC::Debugger::toggleBreakpoint): (JSC::Debugger::clearBreakpoints): (JSC::Debugger::clearDebuggerRequests): * dfg/DFGOSRExitPreparation.cpp: (JSC::DFG::prepareCodeOriginForOSRExit): * heap/Heap.cpp: (JSC::Heap::didFinishIterating): (JSC::Heap::completeAllJITPlans): (JSC::Heap::deleteAllCodeBlocks): (JSC::Heap::collectImpl): (JSC::Heap::completeAllDFGPlans): Deleted. * heap/Heap.h: * heap/HeapInlines.h: (JSC::Heap::forEachCodeBlock): * jit/JIT.cpp: (JSC::JIT::emitNotifyWrite): (JSC::JIT::privateCompileMainPass): (JSC::JIT::privateCompileSlowCases): (JSC::JIT::compileWithoutLinking): (JSC::JIT::link): (JSC::JIT::privateCompile): (JSC::JIT::privateCompileExceptionHandlers): * jit/JIT.h: (JSC::JIT::compile): (JSC::JIT::getSlowCase): (JSC::JIT::linkSlowCase): (JSC::JIT::linkDummySlowCase): * jit/JITInlines.h: (JSC::JIT::emitTagBool): (JSC::JIT::originalInstruction): * jit/JITPropertyAccess32_64.cpp: (JSC::JIT::emitSlow_op_put_to_scope): * jit/JITPropertyAccess.cpp: (JSC::JIT::emitSlow_op_put_by_val): (JSC::JIT::emit_op_resolve_scope): (JSC::JIT::emitSlow_op_resolve_scope): (JSC::JIT::emit_op_get_from_scope): (JSC::JIT::emitSlow_op_get_from_scope): (JSC::JIT::emit_op_put_to_scope): (JSC::JIT::emitSlow_op_put_to_scope): * jit/JITWorklist.cpp: Added. (JSC::JITWorklist::Plan::Plan): (JSC::JITWorklist::Plan::compileInThread): (JSC::JITWorklist::Plan::finalize): (JSC::JITWorklist::Plan::codeBlock): (JSC::JITWorklist::Plan::vm): (JSC::JITWorklist::Plan::isFinishedCompiling): (JSC::JITWorklist::Plan::isFinalized): (JSC::JITWorklist::JITWorklist): (JSC::JITWorklist::~JITWorklist): (JSC::JITWorklist::completeAllForVM): (JSC::JITWorklist::poll): (JSC::JITWorklist::compileLater): (JSC::JITWorklist::compileNow): (JSC::JITWorklist::runThread): (JSC::JITWorklist::finalizePlans): (JSC::JITWorklist::instance): * jit/JITWorklist.h: Added. * llint/LLIntSlowPaths.cpp: (JSC::LLInt::jitCompileAndSetHeuristics): * runtime/CommonSlowPaths.cpp: (JSC::SLOW_PATH_DECL): * runtime/CommonSlowPaths.h: (JSC::CommonSlowPaths::tryCachePutToScopeGlobal): (JSC::CommonSlowPaths::tryCacheGetFromScopeGlobal): * runtime/VM.cpp: (JSC::VM::~VM): Source/WTF: The concurrent baseline JIT needs to be able to clone bytecode to get a consistent snapshot. So, this adds such a method. * wtf/RefCountedArray.h: (WTF::RefCountedArray::RefCountedArray): (WTF::RefCountedArray::clone): Tools: Need to disable concurrent JIT when running profiler tests. We should have been doing this all along. * Scripts/run-jsc-stress-tests: LayoutTests: This change revealed a latent bug in the LLInt. The non-llint version of this new test would usually fail because it was still in LLInt. This new test always fails. * js/dom/prototype-chain-caching-with-impure-get-own-property-slot-traps-llint.html: Added. * js/dom/prototype-chain-caching-with-impure-get-own-property-slot-traps-llint-expected.txt: Added. * js/dom/script-tests/prototype-chain-caching-with-impure-get-own-property-slot-traps-llint.js: Added. Canonical link: https://commits.webkit.org/176935@main git-svn-id: https://svn.webkit.org/repository/webkit/trunk@202157 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-06-17 04:48:47 +00:00
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html>
<head>
<script src="../../resources/js-test-pre.js"></script>
</head>
<body>
<script src="script-tests/prototype-chain-caching-with-impure-get-own-property-slot-traps-llint.js"></script>
<script src="../../resources/js-test-post.js"></script>
</body>
</html>