Google Git
Sign in
webkit / WebKit / 3e0a1a0c405cc4019ec09663859003b5c1ec4f26 / . / Source / JavaScriptCore / ChangeLog
blob: 5afd9a0e90f84df7ee8a8c4f905874b67a1a4af3 [file] [log] [blame]
2012-06-18 Mark Lam <mark.lam@apple.com>
Changed JSC to always record line number information so that error.stack
and window.onerror() can report proper line numbers.
https://bugs.webkit.org/show_bug.cgi?id=89410
Reviewed by Geoffrey Garen.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::CodeBlock):
(JSC::CodeBlock::lineNumberForBytecodeOffset):
(JSC::CodeBlock::shrinkToFit): m_lineInfo is now available unconditionally.
* bytecode/CodeBlock.h:
(JSC::CodeBlock::addLineInfo):
(JSC::CodeBlock::hasLineInfo): Unused. Now removed.
(JSC::CodeBlock::needsCallReturnIndices):
(CodeBlock):
(RareData): Hoisted m_lineInfo out of m_rareData. m_lineInfo is now
filled in unconditionally.
* bytecompiler/BytecodeGenerator.h:
(JSC::BytecodeGenerator::addLineInfo):
2012-06-18 Andy Estes <aestes@apple.com>
Fix r120663, which didn't land the change that was reviewed.
2012-06-18 Andy Estes <aestes@apple.com>
[JSC] In JSGlobalData.cpp, enableAssembler() sometimes leaks two CF objects
https://bugs.webkit.org/show_bug.cgi?id=89415
Reviewed by Sam Weinig.
In the case where canUseJIT was a non-NULL CFBooleanRef,
enableAssembler() would leak both canUseJITKey and canUseJIT by
returning before calling CFRelease. Fix this by using RetainPtr.
* runtime/JSGlobalData.cpp:
(JSC::enableAssembler):
2012-06-17 Geoffrey Garen <ggaren@apple.com>
GC copy phase spends needless cycles zero-filling blocks
https://bugs.webkit.org/show_bug.cgi?id=89128
Reviewed by Gavin Barraclough.
We only need to zero-fill when we're allocating memory that might not
get fully initialized before GC.
* heap/CopiedBlock.h:
(JSC::CopiedBlock::createNoZeroFill):
(JSC::CopiedBlock::create): Added a way to create without zero-filling.
This is our optimization.
(JSC::CopiedBlock::zeroFillToEnd):
(JSC::CopiedBlock::CopiedBlock): Split zero-filling out from creation,
so we can sometimes create without zero-filling.
* heap/CopiedSpace.cpp:
(JSC::CopiedSpace::init):
(JSC::CopiedSpace::tryAllocateSlowCase):
(JSC::CopiedSpace::doneCopying): Renamed addNewBlock to allocateBlock()
to clarify that the new block is always newly-allocated.
(JSC::CopiedSpace::doneFillingBlock): Make sure to zero-fill to the end
of a block that might be used in the future for allocation. (Most of the
time, this is a no-op, since we've already filled the block completely.)
(JSC::CopiedSpace::getFreshBlock): Removed this function because the
abstraction of "allocation must succeed" is no longer useful.
* heap/CopiedSpace.h: Updated declarations to match.
* heap/CopiedSpaceInlineMethods.h:
(JSC::CopiedSpace::allocateBlockForCopyingPhase): New function, which
knows that it can skip zero-filling.
Added tighter scoping to our lock, to improve parallelism.
(JSC::CopiedSpace::allocateBlock): Folded getFreshBlock functionality
into this function, for simplicity.
* heap/MarkStack.cpp:
(JSC::SlotVisitor::startCopying):
(JSC::SlotVisitor::allocateNewSpace): Use our new zero-fill-free helper
function for great good.
2012-06-17 Filip Pizlo <fpizlo@apple.com>
DFG should attempt to use structure watchpoints for all inlined get_by_id's and put_by_id's
https://bugs.webkit.org/show_bug.cgi?id=89316
Reviewed by Oliver Hunt.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::addStructureTransitionCheck):
(ByteCodeParser):
(JSC::DFG::ByteCodeParser::handleGetById):
(JSC::DFG::ByteCodeParser::parseBlock):
2012-06-15 Yong Li <yoli@rim.com>
[BlackBerry] Put platform-specific GC policy in GCActivityCallback
https://bugs.webkit.org/show_bug.cgi?id=89236
Reviewed by Rob Buis.
Add GCActivityCallbackBlackBerry.cpp and implement platform-specific
low memory GC policy there.
* PlatformBlackBerry.cmake:
* heap/Heap.h:
(JSC::Heap::isSafeToCollect): Added.
* runtime/GCActivityCallbackBlackBerry.cpp: Added.
(JSC):
(JSC::DefaultGCActivityCallbackPlatformData::DefaultGCActivityCallbackPlatformData):
(DefaultGCActivityCallbackPlatformData):
(JSC::DefaultGCActivityCallback::DefaultGCActivityCallback):
(JSC::DefaultGCActivityCallback::~DefaultGCActivityCallback):
(JSC::DefaultGCActivityCallback::didAllocate):
(JSC::DefaultGCActivityCallback::willCollect):
(JSC::DefaultGCActivityCallback::synchronize):
(JSC::DefaultGCActivityCallback::cancel):
2012-06-15 Filip Pizlo <fpizlo@apple.com>
DFG should be able to set watchpoints on structure transitions in the
method check prototype chain
https://bugs.webkit.org/show_bug.cgi?id=89058
Adding the same assertion to 32-bit that I added to 64-bit. This change
does not affect correctness but it's a good thing for assertion coverage.
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-06-13 Filip Pizlo <fpizlo@apple.com>
DFG should be able to set watchpoints on structure transitions in the
method check prototype chain
https://bugs.webkit.org/show_bug.cgi?id=89058
Reviewed by Gavin Barraclough.
This adds the ability to set watchpoints on Structures, and then does
the most modest thing we can do with this ability: the DFG now sets
watchpoints on structure transitions in the prototype chain of method
checks.
This appears to be a >1% speed-up on V8.
* bytecode/PutByIdStatus.cpp:
(JSC::PutByIdStatus::computeFromLLInt):
(JSC::PutByIdStatus::computeFor):
* bytecode/StructureSet.h:
(JSC::StructureSet::containsOnly):
(StructureSet):
* bytecode/Watchpoint.cpp:
(JSC::WatchpointSet::WatchpointSet):
(JSC::InlineWatchpointSet::add):
(JSC):
(JSC::InlineWatchpointSet::inflateSlow):
(JSC::InlineWatchpointSet::freeFat):
* bytecode/Watchpoint.h:
(WatchpointSet):
(JSC):
(InlineWatchpointSet):
(JSC::InlineWatchpointSet::InlineWatchpointSet):
(JSC::InlineWatchpointSet::~InlineWatchpointSet):
(JSC::InlineWatchpointSet::hasBeenInvalidated):
(JSC::InlineWatchpointSet::isStillValid):
(JSC::InlineWatchpointSet::startWatching):
(JSC::InlineWatchpointSet::notifyWrite):
(JSC::InlineWatchpointSet::isFat):
(JSC::InlineWatchpointSet::fat):
(JSC::InlineWatchpointSet::inflate):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::addStructureTransitionCheck):
(ByteCodeParser):
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::structureTransitionWatchpointElimination):
(CSEPhase):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCommon.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::dump):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::isCellConstant):
* dfg/DFGJITCompiler.h:
(JSC::DFG::JITCompiler::addWeakReferences):
(JITCompiler):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasStructure):
(Node):
(JSC::DFG::Node::structure):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGRepatch.cpp:
(JSC::DFG::emitPutTransitionStub):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* jit/JITStubs.cpp:
(JSC::JITThunks::tryCachePutByID):
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* runtime/Structure.cpp:
(JSC::Structure::Structure):
* runtime/Structure.h:
(JSC::Structure::transitionWatchpointSetHasBeenInvalidated):
(Structure):
(JSC::Structure::transitionWatchpointSetIsStillValid):
(JSC::Structure::addTransitionWatchpoint):
(JSC::Structure::notifyTransitionFromThisStructure):
(JSC::JSCell::setStructure):
* runtime/SymbolTable.cpp:
(JSC::SymbolTableEntry::attemptToWatch):
2012-06-13 Filip Pizlo <fpizlo@apple.com>
DFG should be able to set watchpoints on global variables
https://bugs.webkit.org/show_bug.cgi?id=88692
Reviewed by Geoffrey Garen.
Rolling back in after fixing Windows build issues, and implementing
branchTest8 for the Qt port's strange assemblers.
This implements global variable constant folding by allowing the optimizing
compiler to set a "watchpoint" on globals that it wishes to constant fold.
If the watchpoint fires, then an OSR exit is forced by overwriting the
machine code that the optimizing compiler generated with a jump.
As such, this patch is adding quite a bit of stuff:
- Jump replacement on those hardware targets supported by the optimizing
JIT. It is now possible to patch in a jump instruction over any recorded
watchpoint label. The jump must be "local" in the sense that it must be
within the range of the largest jump distance supported by a one
instruction jump.
- WatchpointSets and Watchpoints. A Watchpoint is a doubly-linked list node
that records the location where a jump must be inserted and the
destination to which it should jump. Watchpoints can be added to a
WatchpointSet. The WatchpointSet can be fired all at once, which plants
all jumps. WatchpointSet also remembers if it had ever been invalidated,
which allows for monotonicity: we typically don't want to optimize using
watchpoints on something for which watchpoints had previously fired. The
act of notifying a WatchpointSet has a trivial fast path in case no
Watchpoints are registered (one-byte load+branch).
- SpeculativeJIT::speculationWatchpoint(). It's like speculationCheck(),
except that you don't have to emit branches. But, you need to know what
WatchpointSet to add the resulting Watchpoint to. Not everything that
you could write a speculationCheck() for will have a WatchpointSet that
would get notified if the condition you were speculating against became
invalid.
- SymbolTableEntry now has the ability to refer to a WatchpointSet. It can
do so without incurring any space overhead for those entries that don't
have WatchpointSets.
- The bytecode generator infers all global function variables to be
watchable, and makes all stores perform the WatchpointSet's write check,
and marks all loads as being potentially watchable (i.e. you can compile
them to a watchpoint and a constant).
Put together, this allows for fully sleazy inlining of calls to globally
declared functions. The inline prologue will no longer contain the load of
the function, or any checks of the function you're calling. I.e. it's
pretty much like the kind of inlining you would see in Java or C++.
Furthermore, the watchpointing functionality is built to be fairly general,
and should allow setting watchpoints on all sorts of interesting things
in the future.
The sleazy inlining means that we will now sometimes inline in code paths
that have never executed. Previously, to inline we would have either had
to have executed the call (to read the call's inline cache) or have
executed the method check (to read the method check's inline cache). Now,
we might inline when the callee is a watched global variable. This
revealed some humorous bugs. First, constant folding disagreed with CFA
over what kinds of operations can clobber (example: code path A is dead
but stores a String into variable X, all other code paths store 0 into
X, and then you do CompareEq(X, 0) - CFA will say that this is a non-
clobbering constant, but constant folding thought it was clobbering
because it saw the String prediction). Second, inlining would crash if
the inline callee had not been compiled. This patch fixes both bugs,
since otherwise run-javascriptcore-tests would report regressions.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* assembler/ARMv7Assembler.h:
(ARMv7Assembler):
(JSC::ARMv7Assembler::ARMv7Assembler):
(JSC::ARMv7Assembler::labelForWatchpoint):
(JSC::ARMv7Assembler::label):
(JSC::ARMv7Assembler::replaceWithJump):
(JSC::ARMv7Assembler::maxJumpReplacementSize):
* assembler/AbstractMacroAssembler.h:
(JSC):
(AbstractMacroAssembler):
(Label):
(JSC::AbstractMacroAssembler::watchpointLabel):
(JSC::AbstractMacroAssembler::readPointer):
* assembler/AssemblerBuffer.h:
* assembler/MacroAssemblerARM.h:
(JSC::MacroAssemblerARM::branchTest8):
(MacroAssemblerARM):
(JSC::MacroAssemblerARM::replaceWithJump):
(JSC::MacroAssemblerARM::maxJumpReplacementSize):
* assembler/MacroAssemblerARMv7.h:
(JSC::MacroAssemblerARMv7::load8Signed):
(JSC::MacroAssemblerARMv7::load16Signed):
(MacroAssemblerARMv7):
(JSC::MacroAssemblerARMv7::replaceWithJump):
(JSC::MacroAssemblerARMv7::maxJumpReplacementSize):
(JSC::MacroAssemblerARMv7::branchTest8):
(JSC::MacroAssemblerARMv7::jump):
(JSC::MacroAssemblerARMv7::makeBranch):
* assembler/MacroAssemblerMIPS.h:
(JSC::MacroAssemblerMIPS::branchTest8):
(MacroAssemblerMIPS):
(JSC::MacroAssemblerMIPS::replaceWithJump):
(JSC::MacroAssemblerMIPS::maxJumpReplacementSize):
* assembler/MacroAssemblerSH4.h:
(JSC::MacroAssemblerSH4::branchTest8):
(MacroAssemblerSH4):
(JSC::MacroAssemblerSH4::replaceWithJump):
(JSC::MacroAssemblerSH4::maxJumpReplacementSize):
* assembler/MacroAssemblerX86.h:
(MacroAssemblerX86):
(JSC::MacroAssemblerX86::branchTest8):
* assembler/MacroAssemblerX86Common.h:
(JSC::MacroAssemblerX86Common::replaceWithJump):
(MacroAssemblerX86Common):
(JSC::MacroAssemblerX86Common::maxJumpReplacementSize):
* assembler/MacroAssemblerX86_64.h:
(MacroAssemblerX86_64):
(JSC::MacroAssemblerX86_64::branchTest8):
* assembler/X86Assembler.h:
(JSC::X86Assembler::X86Assembler):
(X86Assembler):
(JSC::X86Assembler::cmpb_im):
(JSC::X86Assembler::testb_im):
(JSC::X86Assembler::labelForWatchpoint):
(JSC::X86Assembler::label):
(JSC::X86Assembler::replaceWithJump):
(JSC::X86Assembler::maxJumpReplacementSize):
(JSC::X86Assembler::X86InstructionFormatter::memoryModRM):
* bytecode/CodeBlock.cpp:
(JSC):
(JSC::CodeBlock::printGetByIdCacheStatus):
(JSC::CodeBlock::dump):
* bytecode/CodeBlock.h:
(JSC::CodeBlock::appendOSRExit):
(JSC::CodeBlock::appendSpeculationRecovery):
(CodeBlock):
(JSC::CodeBlock::appendWatchpoint):
(JSC::CodeBlock::numberOfWatchpoints):
(JSC::CodeBlock::watchpoint):
(DFGData):
* bytecode/DFGExitProfile.h:
(JSC::DFG::exitKindToString):
(JSC::DFG::exitKindIsCountable):
* bytecode/GetByIdStatus.cpp:
(JSC::GetByIdStatus::computeForChain):
* bytecode/Instruction.h:
(Instruction):
(JSC::Instruction::Instruction):
* bytecode/Opcode.h:
(JSC):
(JSC::padOpcodeName):
* bytecode/Watchpoint.cpp: Added.
(JSC):
(JSC::Watchpoint::~Watchpoint):
(JSC::Watchpoint::correctLabels):
(JSC::Watchpoint::fire):
(JSC::WatchpointSet::WatchpointSet):
(JSC::WatchpointSet::~WatchpointSet):
(JSC::WatchpointSet::add):
(JSC::WatchpointSet::notifyWriteSlow):
(JSC::WatchpointSet::fireAllWatchpoints):
* bytecode/Watchpoint.h: Added.
(JSC):
(Watchpoint):
(JSC::Watchpoint::Watchpoint):
(JSC::Watchpoint::setDestination):
(WatchpointSet):
(JSC::WatchpointSet::isStillValid):
(JSC::WatchpointSet::hasBeenInvalidated):
(JSC::WatchpointSet::startWatching):
(JSC::WatchpointSet::notifyWrite):
(JSC::WatchpointSet::addressOfIsWatched):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::ResolveResult::checkValidity):
(JSC::BytecodeGenerator::addGlobalVar):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::resolve):
(JSC::BytecodeGenerator::emitResolve):
(JSC::BytecodeGenerator::emitResolveWithBase):
(JSC::BytecodeGenerator::emitResolveWithThis):
(JSC::BytecodeGenerator::emitGetStaticVar):
(JSC::BytecodeGenerator::emitPutStaticVar):
* bytecompiler/BytecodeGenerator.h:
(BytecodeGenerator):
* bytecompiler/NodesCodegen.cpp:
(JSC::FunctionCallResolveNode::emitBytecode):
(JSC::PostfixResolveNode::emitBytecode):
(JSC::PrefixResolveNode::emitBytecode):
(JSC::ReadModifyResolveNode::emitBytecode):
(JSC::AssignResolveNode::emitBytecode):
(JSC::ConstDeclNode::emitCodeSingle):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::clobberStructures):
* dfg/DFGAbstractState.h:
(AbstractState):
(JSC::DFG::AbstractState::didClobber):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCCallHelpers.h:
(CCallHelpers):
(JSC::DFG::CCallHelpers::setupArguments):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::globalVarWatchpointElimination):
(CSEPhase):
(JSC::DFG::CSEPhase::globalVarStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGCorrectableJumpPoint.h:
(JSC::DFG::CorrectableJumpPoint::isSet):
(CorrectableJumpPoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::linkOSRExits):
(JSC::DFG::JITCompiler::link):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasIdentifierNumberForCheck):
(Node):
(JSC::DFG::Node::identifierNumberForCheck):
(JSC::DFG::Node::hasRegisterPointer):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOSRExit.cpp:
(JSC::DFG::OSRExit::OSRExit):
* dfg/DFGOSRExit.h:
(OSRExit):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
(JSC::DFG::SpeculativeJIT::appendCall):
(SpeculativeJIT):
(JSC::DFG::SpeculativeJIT::speculationWatchpoint):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* interpreter/Interpreter.cpp:
(JSC::Interpreter::privateExecute):
* jit/JIT.cpp:
(JSC::JIT::privateCompileMainPass):
(JSC::JIT::privateCompileSlowCases):
* jit/JIT.h:
* jit/JITPropertyAccess.cpp:
(JSC::JIT::emit_op_put_global_var_check):
(JSC):
(JSC::JIT::emitSlow_op_put_global_var_check):
* jit/JITPropertyAccess32_64.cpp:
(JSC::JIT::emit_op_put_global_var_check):
(JSC):
(JSC::JIT::emitSlow_op_put_global_var_check):
* jit/JITStubs.cpp:
(JSC::DEFINE_STUB_FUNCTION):
(JSC):
* jit/JITStubs.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
(LLInt):
* llint/LLIntSlowPaths.h:
(LLInt):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
* runtime/JSObject.cpp:
(JSC::JSObject::removeDirect):
* runtime/JSObject.h:
(JSObject):
* runtime/JSSymbolTableObject.h:
(JSC::symbolTableGet):
(JSC::symbolTablePut):
(JSC::symbolTablePutWithAttributes):
* runtime/SymbolTable.cpp: Added.
(JSC):
(JSC::SymbolTableEntry::copySlow):
(JSC::SymbolTableEntry::freeFatEntrySlow):
(JSC::SymbolTableEntry::couldBeWatched):
(JSC::SymbolTableEntry::attemptToWatch):
(JSC::SymbolTableEntry::addressOfIsWatched):
(JSC::SymbolTableEntry::addWatchpoint):
(JSC::SymbolTableEntry::notifyWriteSlow):
(JSC::SymbolTableEntry::inflateSlow):
* runtime/SymbolTable.h:
(JSC):
(SymbolTableEntry):
(Fast):
(JSC::SymbolTableEntry::Fast::Fast):
(JSC::SymbolTableEntry::Fast::isNull):
(JSC::SymbolTableEntry::Fast::getIndex):
(JSC::SymbolTableEntry::Fast::isReadOnly):
(JSC::SymbolTableEntry::Fast::getAttributes):
(JSC::SymbolTableEntry::Fast::isFat):
(JSC::SymbolTableEntry::SymbolTableEntry):
(JSC::SymbolTableEntry::~SymbolTableEntry):
(JSC::SymbolTableEntry::operator=):
(JSC::SymbolTableEntry::isNull):
(JSC::SymbolTableEntry::getIndex):
(JSC::SymbolTableEntry::getFast):
(JSC::SymbolTableEntry::getAttributes):
(JSC::SymbolTableEntry::isReadOnly):
(JSC::SymbolTableEntry::watchpointSet):
(JSC::SymbolTableEntry::notifyWrite):
(FatEntry):
(JSC::SymbolTableEntry::FatEntry::FatEntry):
(JSC::SymbolTableEntry::isFat):
(JSC::SymbolTableEntry::fatEntry):
(JSC::SymbolTableEntry::inflate):
(JSC::SymbolTableEntry::bits):
(JSC::SymbolTableEntry::freeFatEntry):
(JSC::SymbolTableEntry::pack):
(JSC::SymbolTableEntry::isValidIndex):
2012-06-13 Sheriff Bot <webkit.review.bot@gmail.com>
Unreviewed, rolling out r120172.
http://trac.webkit.org/changeset/120172
https://bugs.webkit.org/show_bug.cgi?id=88976
The patch causes compilation failures on Gtk, Qt and Apple Win
bots (Requested by zdobersek on #webkit).
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* assembler/ARMv7Assembler.h:
(JSC::ARMv7Assembler::nop):
(JSC::ARMv7Assembler::label):
(JSC::ARMv7Assembler::readPointer):
(ARMv7Assembler):
* assembler/AbstractMacroAssembler.h:
(JSC):
(AbstractMacroAssembler):
(Label):
* assembler/AssemblerBuffer.h:
* assembler/MacroAssemblerARM.h:
* assembler/MacroAssemblerARMv7.h:
(JSC::MacroAssemblerARMv7::nop):
(JSC::MacroAssemblerARMv7::jump):
(JSC::MacroAssemblerARMv7::makeBranch):
* assembler/MacroAssemblerMIPS.h:
* assembler/MacroAssemblerSH4.h:
* assembler/MacroAssemblerX86.h:
(MacroAssemblerX86):
(JSC::MacroAssemblerX86::moveWithPatch):
* assembler/MacroAssemblerX86Common.h:
* assembler/MacroAssemblerX86_64.h:
(JSC::MacroAssemblerX86_64::branchTest8):
* assembler/X86Assembler.h:
(JSC::X86Assembler::cmpb_im):
(JSC::X86Assembler::codeSize):
(JSC::X86Assembler::label):
(JSC::X86Assembler::X86InstructionFormatter::memoryModRM):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
* bytecode/CodeBlock.h:
(JSC::CodeBlock::appendOSRExit):
(JSC::CodeBlock::appendSpeculationRecovery):
(DFGData):
* bytecode/DFGExitProfile.h:
(JSC::DFG::exitKindToString):
(JSC::DFG::exitKindIsCountable):
* bytecode/Instruction.h:
* bytecode/Opcode.h:
(JSC):
(JSC::padOpcodeName):
* bytecode/Watchpoint.cpp: Removed.
* bytecode/Watchpoint.h: Removed.
* bytecompiler/BytecodeGenerator.cpp:
(JSC::ResolveResult::checkValidity):
(JSC::BytecodeGenerator::addGlobalVar):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::resolve):
(JSC::BytecodeGenerator::emitResolve):
(JSC::BytecodeGenerator::emitResolveWithBase):
(JSC::BytecodeGenerator::emitResolveWithThis):
(JSC::BytecodeGenerator::emitGetStaticVar):
(JSC::BytecodeGenerator::emitPutStaticVar):
* bytecompiler/BytecodeGenerator.h:
(BytecodeGenerator):
* bytecompiler/NodesCodegen.cpp:
(JSC::FunctionCallResolveNode::emitBytecode):
(JSC::PostfixResolveNode::emitBytecode):
(JSC::PrefixResolveNode::emitBytecode):
(JSC::ReadModifyResolveNode::emitBytecode):
(JSC::AssignResolveNode::emitBytecode):
(JSC::ConstDeclNode::emitCodeSingle):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::clobberStructures):
* dfg/DFGAbstractState.h:
(AbstractState):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCCallHelpers.h:
(JSC::DFG::CCallHelpers::setupArguments):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::globalVarStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGCorrectableJumpPoint.h:
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::linkOSRExits):
(JSC::DFG::JITCompiler::link):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasRegisterPointer):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOSRExit.cpp:
(JSC::DFG::OSRExit::OSRExit):
* dfg/DFGOSRExit.h:
(OSRExit):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
(JSC::DFG::SpeculativeJIT::appendCallSetResult):
(JSC::DFG::SpeculativeJIT::speculationCheck):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* jit/JIT.cpp:
(JSC::JIT::privateCompileMainPass):
(JSC::JIT::privateCompileSlowCases):
* jit/JIT.h:
* jit/JITPropertyAccess.cpp:
* jit/JITPropertyAccess32_64.cpp:
* jit/JITStubs.cpp:
* jit/JITStubs.h:
* llint/LLIntSlowPaths.cpp:
* llint/LLIntSlowPaths.h:
(LLInt):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
* runtime/JSObject.cpp:
(JSC::JSObject::removeDirect):
* runtime/JSObject.h:
(JSObject):
* runtime/JSSymbolTableObject.h:
(JSC::symbolTableGet):
(JSC::symbolTablePut):
(JSC::symbolTablePutWithAttributes):
* runtime/SymbolTable.cpp: Removed.
* runtime/SymbolTable.h:
(JSC):
(JSC::SymbolTableEntry::isNull):
(JSC::SymbolTableEntry::getIndex):
(SymbolTableEntry):
(JSC::SymbolTableEntry::getAttributes):
(JSC::SymbolTableEntry::isReadOnly):
(JSC::SymbolTableEntry::pack):
(JSC::SymbolTableEntry::isValidIndex):
2012-06-12 Filip Pizlo <fpizlo@apple.com>
DFG should be able to set watchpoints on global variables
https://bugs.webkit.org/show_bug.cgi?id=88692
Reviewed by Geoffrey Garen.
This implements global variable constant folding by allowing the optimizing
compiler to set a "watchpoint" on globals that it wishes to constant fold.
If the watchpoint fires, then an OSR exit is forced by overwriting the
machine code that the optimizing compiler generated with a jump.
As such, this patch is adding quite a bit of stuff:
- Jump replacement on those hardware targets supported by the optimizing
JIT. It is now possible to patch in a jump instruction over any recorded
watchpoint label. The jump must be "local" in the sense that it must be
within the range of the largest jump distance supported by a one
instruction jump.
- WatchpointSets and Watchpoints. A Watchpoint is a doubly-linked list node
that records the location where a jump must be inserted and the
destination to which it should jump. Watchpoints can be added to a
WatchpointSet. The WatchpointSet can be fired all at once, which plants
all jumps. WatchpointSet also remembers if it had ever been invalidated,
which allows for monotonicity: we typically don't want to optimize using
watchpoints on something for which watchpoints had previously fired. The
act of notifying a WatchpointSet has a trivial fast path in case no
Watchpoints are registered (one-byte load+branch).
- SpeculativeJIT::speculationWatchpoint(). It's like speculationCheck(),
except that you don't have to emit branches. But, you need to know what
WatchpointSet to add the resulting Watchpoint to. Not everything that
you could write a speculationCheck() for will have a WatchpointSet that
would get notified if the condition you were speculating against became
invalid.
- SymbolTableEntry now has the ability to refer to a WatchpointSet. It can
do so without incurring any space overhead for those entries that don't
have WatchpointSets.
- The bytecode generator infers all global function variables to be
watchable, and makes all stores perform the WatchpointSet's write check,
and marks all loads as being potentially watchable (i.e. you can compile
them to a watchpoint and a constant).
Put together, this allows for fully sleazy inlining of calls to globally
declared functions. The inline prologue will no longer contain the load of
the function, or any checks of the function you're calling. I.e. it's
pretty much like the kind of inlining you would see in Java or C++.
Furthermore, the watchpointing functionality is built to be fairly general,
and should allow setting watchpoints on all sorts of interesting things
in the future.
The sleazy inlining means that we will now sometimes inline in code paths
that have never executed. Previously, to inline we would have either had
to have executed the call (to read the call's inline cache) or have
executed the method check (to read the method check's inline cache). Now,
we might inline when the callee is a watched global variable. This
revealed some humorous bugs. First, constant folding disagreed with CFA
over what kinds of operations can clobber (example: code path A is dead
but stores a String into variable X, all other code paths store 0 into
X, and then you do CompareEq(X, 0) - CFA will say that this is a non-
clobbering constant, but constant folding thought it was clobbering
because it saw the String prediction). Second, inlining would crash if
the inline callee had not been compiled. This patch fixes both bugs,
since otherwise run-javascriptcore-tests would report regressions.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* assembler/ARMv7Assembler.h:
(ARMv7Assembler):
(JSC::ARMv7Assembler::ARMv7Assembler):
(JSC::ARMv7Assembler::labelForWatchpoint):
(JSC::ARMv7Assembler::label):
(JSC::ARMv7Assembler::replaceWithJump):
(JSC::ARMv7Assembler::maxJumpReplacementSize):
* assembler/AbstractMacroAssembler.h:
(JSC):
(AbstractMacroAssembler):
(Label):
(JSC::AbstractMacroAssembler::watchpointLabel):
* assembler/AssemblerBuffer.h:
* assembler/MacroAssemblerARM.h:
(JSC::MacroAssemblerARM::replaceWithJump):
(MacroAssemblerARM):
(JSC::MacroAssemblerARM::maxJumpReplacementSize):
* assembler/MacroAssemblerARMv7.h:
(MacroAssemblerARMv7):
(JSC::MacroAssemblerARMv7::replaceWithJump):
(JSC::MacroAssemblerARMv7::maxJumpReplacementSize):
(JSC::MacroAssemblerARMv7::branchTest8):
(JSC::MacroAssemblerARMv7::jump):
(JSC::MacroAssemblerARMv7::makeBranch):
* assembler/MacroAssemblerMIPS.h:
(JSC::MacroAssemblerMIPS::replaceWithJump):
(MacroAssemblerMIPS):
(JSC::MacroAssemblerMIPS::maxJumpReplacementSize):
* assembler/MacroAssemblerSH4.h:
(JSC::MacroAssemblerSH4::replaceWithJump):
(MacroAssemblerSH4):
(JSC::MacroAssemblerSH4::maxJumpReplacementSize):
* assembler/MacroAssemblerX86.h:
(MacroAssemblerX86):
(JSC::MacroAssemblerX86::branchTest8):
* assembler/MacroAssemblerX86Common.h:
(JSC::MacroAssemblerX86Common::replaceWithJump):
(MacroAssemblerX86Common):
(JSC::MacroAssemblerX86Common::maxJumpReplacementSize):
* assembler/MacroAssemblerX86_64.h:
(MacroAssemblerX86_64):
(JSC::MacroAssemblerX86_64::branchTest8):
* assembler/X86Assembler.h:
(JSC::X86Assembler::X86Assembler):
(X86Assembler):
(JSC::X86Assembler::cmpb_im):
(JSC::X86Assembler::testb_im):
(JSC::X86Assembler::labelForWatchpoint):
(JSC::X86Assembler::label):
(JSC::X86Assembler::replaceWithJump):
(JSC::X86Assembler::maxJumpReplacementSize):
(JSC::X86Assembler::X86InstructionFormatter::memoryModRM):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
* bytecode/CodeBlock.h:
(JSC::CodeBlock::appendOSRExit):
(JSC::CodeBlock::appendSpeculationRecovery):
(CodeBlock):
(JSC::CodeBlock::appendWatchpoint):
(JSC::CodeBlock::numberOfWatchpoints):
(JSC::CodeBlock::watchpoint):
(DFGData):
* bytecode/DFGExitProfile.h:
(JSC::DFG::exitKindToString):
(JSC::DFG::exitKindIsCountable):
* bytecode/Instruction.h:
(Instruction):
(JSC::Instruction::Instruction):
* bytecode/Opcode.h:
(JSC):
(JSC::padOpcodeName):
* bytecode/Watchpoint.cpp: Added.
(JSC):
(JSC::Watchpoint::~Watchpoint):
(JSC::Watchpoint::correctLabels):
(JSC::Watchpoint::fire):
(JSC::WatchpointSet::WatchpointSet):
(JSC::WatchpointSet::~WatchpointSet):
(JSC::WatchpointSet::add):
(JSC::WatchpointSet::notifyWriteSlow):
(JSC::WatchpointSet::fireAllWatchpoints):
* bytecode/Watchpoint.h: Added.
(JSC):
(Watchpoint):
(JSC::Watchpoint::Watchpoint):
(JSC::Watchpoint::setDestination):
(WatchpointSet):
(JSC::WatchpointSet::isStillValid):
(JSC::WatchpointSet::hasBeenInvalidated):
(JSC::WatchpointSet::startWatching):
(JSC::WatchpointSet::notifyWrite):
(JSC::WatchpointSet::addressOfIsWatched):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::ResolveResult::checkValidity):
(JSC::BytecodeGenerator::addGlobalVar):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::resolve):
(JSC::BytecodeGenerator::emitResolve):
(JSC::BytecodeGenerator::emitResolveWithBase):
(JSC::BytecodeGenerator::emitResolveWithThis):
(JSC::BytecodeGenerator::emitGetStaticVar):
(JSC::BytecodeGenerator::emitPutStaticVar):
* bytecompiler/BytecodeGenerator.h:
(BytecodeGenerator):
* bytecompiler/NodesCodegen.cpp:
(JSC::FunctionCallResolveNode::emitBytecode):
(JSC::PostfixResolveNode::emitBytecode):
(JSC::PrefixResolveNode::emitBytecode):
(JSC::ReadModifyResolveNode::emitBytecode):
(JSC::AssignResolveNode::emitBytecode):
(JSC::ConstDeclNode::emitCodeSingle):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::clobberStructures):
* dfg/DFGAbstractState.h:
(AbstractState):
(JSC::DFG::AbstractState::didClobber):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCCallHelpers.h:
(CCallHelpers):
(JSC::DFG::CCallHelpers::setupArguments):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::globalVarWatchpointElimination):
(CSEPhase):
(JSC::DFG::CSEPhase::globalVarStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGCorrectableJumpPoint.h:
(JSC::DFG::CorrectableJumpPoint::isSet):
(CorrectableJumpPoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::linkOSRExits):
(JSC::DFG::JITCompiler::link):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasIdentifierNumberForCheck):
(Node):
(JSC::DFG::Node::identifierNumberForCheck):
(JSC::DFG::Node::hasRegisterPointer):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOSRExit.cpp:
(JSC::DFG::OSRExit::OSRExit):
* dfg/DFGOSRExit.h:
(OSRExit):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
(JSC::DFG::SpeculativeJIT::appendCall):
(SpeculativeJIT):
(JSC::DFG::SpeculativeJIT::speculationWatchpoint):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* jit/JIT.cpp:
(JSC::JIT::privateCompileMainPass):
(JSC::JIT::privateCompileSlowCases):
* jit/JIT.h:
* jit/JITPropertyAccess.cpp:
(JSC::JIT::emit_op_put_global_var_check):
(JSC):
(JSC::JIT::emitSlow_op_put_global_var_check):
* jit/JITPropertyAccess32_64.cpp:
(JSC::JIT::emit_op_put_global_var_check):
(JSC):
(JSC::JIT::emitSlow_op_put_global_var_check):
* jit/JITStubs.cpp:
(JSC::JITThunks::JITThunks):
(JSC::DEFINE_STUB_FUNCTION):
(JSC):
* jit/JITStubs.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
(LLInt):
* llint/LLIntSlowPaths.h:
(LLInt):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
* runtime/JSObject.cpp:
(JSC::JSObject::removeDirect):
* runtime/JSObject.h:
(JSObject):
* runtime/JSSymbolTableObject.h:
(JSC::symbolTableGet):
(JSC::symbolTablePut):
(JSC::symbolTablePutWithAttributes):
* runtime/SymbolTable.cpp: Added.
(JSC):
(JSC::SymbolTableEntry::copySlow):
(JSC::SymbolTableEntry::freeFatEntrySlow):
(JSC::SymbolTableEntry::couldBeWatched):
(JSC::SymbolTableEntry::attemptToWatch):
(JSC::SymbolTableEntry::addressOfIsWatched):
(JSC::SymbolTableEntry::addWatchpoint):
(JSC::SymbolTableEntry::notifyWriteSlow):
(JSC::SymbolTableEntry::inflateSlow):
* runtime/SymbolTable.h:
(JSC):
(SymbolTableEntry):
(Fast):
(JSC::SymbolTableEntry::Fast::Fast):
(JSC::SymbolTableEntry::Fast::isNull):
(JSC::SymbolTableEntry::Fast::getIndex):
(JSC::SymbolTableEntry::Fast::isReadOnly):
(JSC::SymbolTableEntry::Fast::getAttributes):
(JSC::SymbolTableEntry::Fast::isFat):
(JSC::SymbolTableEntry::SymbolTableEntry):
(JSC::SymbolTableEntry::~SymbolTableEntry):
(JSC::SymbolTableEntry::operator=):
(JSC::SymbolTableEntry::isNull):
(JSC::SymbolTableEntry::getIndex):
(JSC::SymbolTableEntry::getFast):
(JSC::SymbolTableEntry::getAttributes):
(JSC::SymbolTableEntry::isReadOnly):
(JSC::SymbolTableEntry::watchpointSet):
(JSC::SymbolTableEntry::notifyWrite):
(FatEntry):
(JSC::SymbolTableEntry::FatEntry::FatEntry):
(JSC::SymbolTableEntry::isFat):
(JSC::SymbolTableEntry::fatEntry):
(JSC::SymbolTableEntry::inflate):
(JSC::SymbolTableEntry::bits):
(JSC::SymbolTableEntry::freeFatEntry):
(JSC::SymbolTableEntry::pack):
(JSC::SymbolTableEntry::isValidIndex):
2012-06-12 Filip Pizlo <fpizlo@apple.com>
Unreviewed build fix for ARMv7 debug builds.
* jit/JITStubs.cpp:
(JSC::JITThunks::JITThunks):
2012-06-12 Geoffrey Garen <ggaren@apple.com>
Build fix for case-sensitive file systems: use the right case.
* heap/ListableHandler.h:
2012-06-11 Geoffrey Garen <ggaren@apple.com>
GC should be 1.7X faster
https://bugs.webkit.org/show_bug.cgi?id=88840
Reviewed by Oliver Hunt.
I profiled, and removed anything that showed up as a concurrency
bottleneck. Then, I added 3 threads to our max thread count, since we
can scale up to more threads now.
* heap/BlockAllocator.cpp:
(JSC::BlockAllocator::BlockAllocator):
(JSC::BlockAllocator::~BlockAllocator):
(JSC::BlockAllocator::releaseFreeBlocks):
(JSC::BlockAllocator::waitForRelativeTimeWhileHoldingLock):
(JSC::BlockAllocator::waitForRelativeTime):
(JSC::BlockAllocator::blockFreeingThreadMain):
* heap/BlockAllocator.h:
(BlockAllocator):
(JSC::BlockAllocator::allocate):
(JSC::BlockAllocator::deallocate): Use a spin lock for the common case
where we're just popping a linked list. (A pthread mutex would sleep our
thread even if the lock were only contended for a microsecond.)
Scope the lock to avoid holding it while allocating VM, since that's a
slow activity and it doesn't modify any of our data structures.
We still use a pthread mutex to handle our condition variable since we
have to, and it's not a hot path.
* heap/CopiedSpace.cpp:
(JSC::CopiedSpace::CopiedSpace):
(JSC::CopiedSpace::doneFillingBlock):
* heap/CopiedSpace.h:
(JSC::CopiedSpace::CopiedSpace): Use a spin lock for the to space lock,
since it just guards linked list and hash table manipulation.
* heap/MarkStack.cpp:
(JSC::MarkStackSegmentAllocator::MarkStackSegmentAllocator):
(JSC::MarkStackSegmentAllocator::allocate):
(JSC::MarkStackSegmentAllocator::release):
(JSC::MarkStackSegmentAllocator::shrinkReserve): Use a spin lock, since
we're just managing a linked list.
(JSC::MarkStackArray::donateSomeCellsTo): Changed donation to be proportional
to our current stack size. This fixes cases where we used to donate too
much. Interestingly, donating too much was starving the donor (when it
ran out of work later) *and* the recipient (since it had to wait on a
long donation operation to complete before it could acquire the lock).
In the worst case, we're still guaranteed to donate N cells in roughly log N time.
This change also fixes cases where we used to donate too little, since
we would always keep a fixed minimum number of cells. In the worst case,
with N marking threads, would could have N large object graph roots in
our stack for the duration of GC, and scale to only 1 thread.
It's an interesting observation that a single object in the mark stack
might represent an arbitrarily large object graph -- and only the act
of marking can find out.
(JSC::MarkStackArray::stealSomeCellsFrom): Steal in proportion to idle
threads. Once again, this fixes cases where constants could cause us
to steal too much or too little.
(JSC::SlotVisitor::donateKnownParallel): Always wake up other threads
if they're idle. We can afford to do this because we're conservative
about when we donate.
(JSC::SlotVisitor::drainFromShared):
* heap/MarkStack.h:
(MarkStackSegmentAllocator):
(MarkStackArray):
(JSC):
* heap/SlotVisitor.h: Merged the "should I donate?" decision into a
single function, for simplicity.
* runtime/Options.cpp:
(minimumNumberOfScansBetweenRebalance): Reduced the delay before donation
a lot. We can afford to do this because, in the common case, donation is
a single branch that decides not to donate.
(cpusToUse): Use more CPUs now, since we scale better now.
* runtime/Options.h:
(Options): Removed now-unused variables.
2012-06-12 Filip Pizlo <fpizlo@apple.com>
REGRESSION(120121): inspector tests crash in DFG
https://bugs.webkit.org/show_bug.cgi?id=88941
Reviewed by Geoffrey Garen.
The CFG simplifier has two different ways of fixing up GetLocal, Phantom, and Flush. If we've
already fixed up the node one way, we shouldn't try the other way. The reason why we shouldn't
is that the second way depends on the node referring to other nodes in the to-be-jettisoned
block. After fixup they potentially will refer to nodes in the block being merged to.
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::fixPossibleGetLocal):
(JSC::DFG::CFGSimplificationPhase::mergeBlocks):
2012-06-12 Leo Yang <leo.yang@torchmobile.com.cn>
Dynamic hash table in DOMObjectHashTableMap is wrong in multiple threads
https://bugs.webkit.org/show_bug.cgi?id=87334
Reviewed by Geoffrey Garen.
Add a copy member function to JSC::HasTable. This function will copy all data
members except for *table* which contains thread specific data that prevents
up copying it. When you want to copy a JSC::HashTable that was constructed
on another thread you should call JSC::HashTable::copy().
* runtime/Lookup.h:
(JSC::HashTable::copy):
(HashTable):
2012-06-12 Filip Pizlo <fpizlo@apple.com>
DFG should not ASSERT if you have a double use of a variable that is not revealed to be a double
until after CFG simplification
https://bugs.webkit.org/show_bug.cgi?id=88927
<rdar://problem/11513971>
Reviewed by Geoffrey Garen.
Speculation fixup needs to run if simplification did things, because simplification can change
predictions - particularly if you had a control flow path that stored weird things into a
variable, but that path got axed by the simplifier.
Running fixup in the fixpoint requires making it idempotent, which it previously wasn't. Only
one place needed to be changed, namely the un-MustGenerate-ion of ValueToInt32.
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
2012-06-12 Filip Pizlo <fpizlo@apple.com>
REGRESSION (r119779): Javascript TypeError: 'undefined' is not an object
https://bugs.webkit.org/show_bug.cgi?id=88783
<rdar://problem/11640299>
Reviewed by Geoffrey Garen.
If you don't keep alive the base of an object access over the various checks
you do for the prototype chain, you're going to have a bad time.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleGetById):
2012-06-12 Hojong Han <hojong.han@samsung.com>
Property names of the built-in object cannot be retrieved
after trying to delete one of its properties
https://bugs.webkit.org/show_bug.cgi?id=86461
Reviewed by Gavin Barraclough.
* runtime/JSObject.cpp:
(JSC::getClassPropertyNames):
(JSC::JSObject::getOwnPropertyNames):
2012-06-11 Gyuyoung Kim <gyuyoung.kim@samsung.com>
[CMAKE][EFL] Remove duplicated executable output path
https://bugs.webkit.org/show_bug.cgi?id=88765
Reviewed by Daniel Bates.
CMake files for EFL port have redefined executable output path. However, EFL port doesn't
need to define again because it is already defined in top-level CMake file.
* shell/CMakeLists.txt:
2012-06-11 Carlos Garcia Campos <cgarcia@igalia.com>
Unreviewed. Fix make distcheck issues.
* GNUmakefile.list.am: Remove non existent header file.
2012-06-10 Patrick Gansterer <paroga@webkit.org>
Unreviewed. Build fix for !ENABLE(JIT) after r119844 and r119925.
* runtime/Executable.h:
(ExecutableBase):
(JSC::ExecutableBase::clearCodeVirtual):
2012-06-10 Patrick Gansterer <paroga@webkit.org>
Unreviewed. Build fix for !ENABLE(JIT) after r119844.
* runtime/Executable.h:
(ExecutableBase):
(JSC):
2012-06-09 Dominic Cooney <dominicc@chromium.org>
[Chromium] Remove JavaScriptCore dependencies from gyp
https://bugs.webkit.org/show_bug.cgi?id=88510
Reviewed by Adam Barth.
Chromium doesn't support JSC any more and there doesn't seem to be
a strong interest in using GYP as the common build system in other
ports.
* JavaScriptCore.gyp/JavaScriptCore.gyp: WebCore still depends on YARR interpreter.
* JavaScriptCore.gypi: Only include YARR source.
* gyp/JavaScriptCore.gyp: Removed.
* gyp/gtk.gyp: Removed.
2012-06-09 Geoffrey Garen <ggaren@apple.com>
Unreviewed, rolling back in part2 of r118646.
This patch removes eager finalization.
Weak pointer finalization should be lazy
https://bugs.webkit.org/show_bug.cgi?id=87599
Reviewed by Sam Weinig.
* heap/Heap.cpp:
(JSC::Heap::collect): Don't finalize eagerly -- we'll do it lazily.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep): Do sweep weak sets when sweeping a block,
since we won't get another chance.
* heap/MarkedBlock.h:
(JSC::MarkedBlock::sweepWeakSet):
* heap/MarkedSpace.cpp:
(MarkedSpace::WeakSetSweep):
* heap/MarkedSpace.h:
(JSC::MarkedSpace::sweepWeakSets): Removed now-unused code.
2012-06-09 Sukolsak Sakshuwong <sukolsak@google.com>
Add UNDO_MANAGER flag
https://bugs.webkit.org/show_bug.cgi?id=87908
Reviewed by Tony Chang.
* Configurations/FeatureDefines.xcconfig:
2012-06-08 Geoffrey Garen <ggaren@apple.com>
Unreviewed, rolling back in part1 of r118646.
This patch includes everything necessary for lazy finalization, but
keeps eager finalization enabled for the time being.
Weak pointer finalization should be lazy
https://bugs.webkit.org/show_bug.cgi?id=87599
Reviewed by Sam Weinig.
* heap/MarkedBlock.cpp:
* heap/MarkedBlock.h:
(JSC::MarkedBlock::resetAllocator):
* heap/MarkedSpace.cpp:
(JSC::MarkedSpace::resetAllocators):
* heap/MarkedSpace.h:
(JSC::MarkedSpace::resetAllocators): Don't force allocator reset anymore.
It will happen automatically when a weak set is swept. It's simpler to
have only one canonical way for this to happen, and it wasn't buying
us anything to do it eagerly.
* heap/WeakBlock.cpp:
(JSC::WeakBlock::sweep): Don't short-circuit a sweep unless we know
the sweep would be a no-op. If even one finalizer is pending, we need to
run it, since we won't get another chance.
* heap/WeakSet.cpp:
(JSC::WeakSet::sweep): This loop can be simpler now that
WeakBlock::sweep() does what we mean.
Reset our allocator after a sweep because this is the optimal time to
start trying to recycle old weak pointers.
(JSC::WeakSet::tryFindAllocator): Don't sweep when searching for an
allocator because we've swept already, and forcing a new sweep would be
wasteful.
* heap/WeakSet.h:
(JSC::WeakSet::shrink): Be sure to reset our allocator after a shrink
because the shrink may have removed the block the allocator was going to
allocate out of.
2012-06-08 Gavin Barraclough <barraclough@apple.com>
Unreviewed roll out r119795.
This broke jquery/core.html
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::emitAllocateBasicJSObject):
* jit/JITInlineMethods.h:
(JSC::JIT::emitAllocateBasicJSObject):
* llint/LowLevelInterpreter.asm:
* runtime/JSGlobalData.h:
(JSGlobalData):
* runtime/JSGlobalThis.cpp:
(JSC::JSGlobalThis::setUnwrappedObject):
* runtime/JSObject.cpp:
(JSC::JSObject::visitChildren):
(JSC::JSObject::createInheritorID):
* runtime/JSObject.h:
(JSObject):
(JSC::JSObject::resetInheritorID):
(JSC):
(JSC::JSObject::offsetOfInheritorID):
(JSC::JSObject::inheritorID):
2012-06-08 Filip Pizlo <fpizlo@apple.com>
PredictedType should be called SpeculatedType
https://bugs.webkit.org/show_bug.cgi?id=88477
Unreviewed, fix a renaming goof from http://trac.webkit.org/changeset/119660.
I accidentally renamed ByteCodeParser::getPrediction to
ByteCodeParser::getSpeculation. That was not the intent. This changes it
back.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::addCall):
(JSC::DFG::ByteCodeParser::getPredictionWithoutOSRExit):
(JSC::DFG::ByteCodeParser::getPrediction):
(JSC::DFG::ByteCodeParser::handleCall):
(JSC::DFG::ByteCodeParser::parseBlock):
2012-06-08 Andy Wingo <wingo@igalia.com>
Explictly mark stubs called by JIT as being internal
https://bugs.webkit.org/show_bug.cgi?id=88552
Reviewed by Filip Pizlo.
* dfg/DFGOSRExitCompiler.h:
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* jit/HostCallReturnValue.h:
* jit/JITStubs.cpp:
* jit/JITStubs.h:
* jit/ThunkGenerators.cpp:
* llint/LLIntSlowPaths.h: Mark a bunch of stubs as being
WTF_INTERNAL. Change most calls to SYMBOL_STRING_RELOCATION to
LOCAL_REFERENCE, or GLOBAL_REFERENCE in the case of the wrappers
to truly global symbols.
* offlineasm/asm.rb: Generate LOCAL_REFERENCE instead of
SYMBOL_STRING_RELOCATION.
2012-06-08 Geoffrey Garen <ggaren@apple.com>
Don't rely on weak pointers for eager CodeBlock finalization
https://bugs.webkit.org/show_bug.cgi?id=88465
Reviewed by Gavin Barraclough.
This is incompatible with lazy weak pointer finalization.
I considered just making CodeBlock finalization lazy-friendly, but it
turns out that the heap is already way up in CodeBlock's business when
it comes to finalization, so I decided to finish the job and move full
responsibility for CodeBlock finalization into the heap.
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def: Maybe this
will build.
* debugger/Debugger.cpp: Updated for rename.
* heap/Heap.cpp:
(JSC::Heap::deleteAllCompiledCode): Renamed for consistency. Fixed a bug
where we would not delete code for a code block that had been previously
jettisoned. I don't know if this happens in practice -- I mostly did
this to improve consistency with deleteUnmarkedCompiledCode.
(JSC::Heap::deleteUnmarkedCompiledCode): New function, responsible for
eager finalization of unmarked code blocks.
(JSC::Heap::collect): Updated for rename. Updated to call
deleteUnmarkedCompiledCode(), which takes care of jettisoned DFG code
blocks too.
(JSC::Heap::addCompiledCode): Renamed, since this points to all code
now, not just functions.
* heap/Heap.h:
(Heap): Keep track of all user code, not just functions. This is a
negligible additional overhead, since most code is function code.
* runtime/Executable.cpp:
(JSC::*::finalize): Removed these functions, since we don't rely on
weak pointer finalization anymore.
(JSC::FunctionExecutable::FunctionExecutable): Moved linked-list stuff
into base class so all executables can be in the list.
(JSC::EvalExecutable::clearCode):
(JSC::ProgramExecutable::clearCode):
(JSC::FunctionExecutable::clearCode): All we need to do is delete our
CodeBlock -- that will delete all of its internal data structures.
(JSC::FunctionExecutable::clearCodeIfNotCompiling): Factored out a helper
function to improve clarity.
* runtime/Executable.h:
(JSC::ExecutableBase): Moved linked-list stuff
into base class so all executables can be in the list.
(JSC::NativeExecutable::create):
(NativeExecutable):
(ScriptExecutable):
(JSC::ScriptExecutable::finishCreation):
(JSC::EvalExecutable::create):
(EvalExecutable):
(JSC::ProgramExecutable::create):
(ProgramExecutable):
(FunctionExecutable):
(JSC::FunctionExecutable::create): Don't use a finalizer -- the heap
will call us back to destroy our code block.
(JSC::FunctionExecutable::discardCode): Renamed to clearCodeIfNotCompiling()
for clarity.
(JSC::FunctionExecutable::isCompiling): New helper function, for clarity.
(JSC::ScriptExecutable::clearCodeVirtual): New helper function, since
the heap needs to make polymorphic calls to clear code.
* runtime/JSGlobalData.cpp:
(JSC::StackPreservingRecompiler::operator()):
* runtime/JSGlobalObject.cpp:
(JSC::DynamicGlobalObjectScope::DynamicGlobalObjectScope): Updated for
renames.
2012-06-07 Filip Pizlo <fpizlo@apple.com>
DFG should inline prototype chain accesses, and do the right things if the
specific function optimization is available
https://bugs.webkit.org/show_bug.cgi?id=88594
Reviewed by Gavin Barraclough.
Looks like a 3% win on V8.
* bytecode/CodeBlock.h:
(JSC::Structure::prototypeForLookup):
(JSC):
* bytecode/GetByIdStatus.cpp:
(JSC::GetByIdStatus::computeFromLLInt):
(JSC):
(JSC::GetByIdStatus::computeForChain):
(JSC::GetByIdStatus::computeFor):
* bytecode/GetByIdStatus.h:
(JSC::GetByIdStatus::GetByIdStatus):
(JSC::GetByIdStatus::isSimple):
(JSC::GetByIdStatus::chain):
(JSC::GetByIdStatus::specificValue):
(GetByIdStatus):
* bytecode/StructureSet.h:
(StructureSet):
(JSC::StructureSet::singletonStructure):
* bytecode/StructureStubInfo.h:
(JSC::StructureStubInfo::initGetByIdProto):
(JSC::StructureStubInfo::initGetByIdChain):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleGetById):
* dfg/DFGRepatch.cpp:
(JSC::DFG::tryCacheGetByID):
* jit/JITStubs.cpp:
(JSC::JITThunks::tryCacheGetByID):
* runtime/JSGlobalObject.h:
(JSC::Structure::prototypeForLookup):
(JSC):
* runtime/Structure.h:
(Structure):
2012-06-07 Gavin Barraclough <barraclough@apple.com>
Remove JSObject::m_inheritorID
https://bugs.webkit.org/show_bug.cgi?id=88378
Reviewed by Geoff Garen.
This is rarely used, and not performance critical (the commonly accessed copy is cached on JSFunction),
and most objects don't need an inheritorID (this value is only used if the object is used as a prototype).
Instead use a private named value in the object's property storage.
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::emitAllocateBasicJSObject):
- No need m_inheritorID to initialize!
* jit/JITInlineMethods.h:
(JSC::JIT::emitAllocateBasicJSObject):
- No need m_inheritorID to initialize!
* llint/LowLevelInterpreter.asm:
- No need m_inheritorID to initialize!
* runtime/JSGlobalData.h:
(JSGlobalData):
- Added private name 'm_inheritorIDKey'.
* runtime/JSGlobalThis.cpp:
(JSC::JSGlobalThis::setUnwrappedObject):
- resetInheritorID is now passed a JSGlobalData&.
* runtime/JSObject.cpp:
(JSC::JSObject::visitChildren):
- No m_inheritorID to be marked.
(JSC::JSObject::createInheritorID):
- Store the newly created inheritorID in the property map.
* runtime/JSObject.h:
(JSC::JSObject::resetInheritorID):
- Remove the inheritorID from property storage.
(JSC::JSObject::inheritorID):
- Read the inheritorID from property storage.
2012-06-07 Gavin Barraclough <barraclough@apple.com>
Math.pow on iOS does not support denormal numbers.
https://bugs.webkit.org/show_bug.cgi?id=88592
Reviewed by Filip Pizlo.
Import an implementation from fdlibm, detect cases where it is safe to use the system
implementation & where we should fall back to fdlibm.
* runtime/MathObject.cpp:
(JSC::isDenormal):
(JSC::isEdgeCase):
(JSC::mathPow):
- On iOS, detect cases where denormal support may be required & use fdlibm in these cases.
(JSC::mathProtoFuncPow):
- Changed to use mathPow.
(JSC::fdlibmScalbn):
(JSC::fdlibmPow):
- These functions imported from fdlibm; original style retained to ease future merging.
2012-06-07 Patrick Gansterer <paroga@webkit.org>
Unreviewed. Build fix for !ENABLE(JIT) after r119441.
* interpreter/Interpreter.cpp:
(JSC::Interpreter::privateExecute):
2012-06-07 Andy Wingo <wingo@igalia.com>
Unreviewed build fix after r119593.
* llint/LLIntOfflineAsmConfig.h (OFFLINE_ASM_GLOBAL_LABEL): Fix
uses of "name" to be "label", the macro's parameter. Otherwise we
serialize mentions of the literal symbol "name" into the objcode.
Causes a build error using GNU ld (not gold).
2012-06-06 Ryosuke Niwa <rniwa@webkit.org>
Chromium build fix attempt. Why do we need to list these files in gyp!?
* JavaScriptCore.gypi:
2012-06-06 Filip Pizlo <fpizlo@apple.com>
PredictedType should be called SpeculatedType
https://bugs.webkit.org/show_bug.cgi?id=88477
Rubber stamped by Gavin Barraclough.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::shouldOptimizeNow):
(JSC::CodeBlock::dumpValueProfiles):
* bytecode/CodeBlock.h:
(JSC::CodeBlock::valueProfilePredictionForBytecodeOffset):
* bytecode/LazyOperandValueProfile.cpp:
(JSC::LazyOperandValueProfileParser::prediction):
* bytecode/LazyOperandValueProfile.h:
(LazyOperandValueProfileParser):
* bytecode/PredictedType.cpp: Removed.
* bytecode/PredictedType.h: Removed.
* bytecode/SpeculatedType.cpp: Copied from Source/JavaScriptCore/bytecode/PredictedType.cpp.
(JSC::speculationToString):
(JSC::speculationToAbbreviatedString):
(JSC::speculationFromClassInfo):
(JSC::speculationFromStructure):
(JSC::speculationFromCell):
(JSC::speculationFromValue):
* bytecode/SpeculatedType.h: Copied from Source/JavaScriptCore/bytecode/PredictedType.h.
(JSC):
(JSC::isAnySpeculation):
(JSC::isCellSpeculation):
(JSC::isObjectSpeculation):
(JSC::isFinalObjectSpeculation):
(JSC::isFinalObjectOrOtherSpeculation):
(JSC::isFixedIndexedStorageObjectSpeculation):
(JSC::isStringSpeculation):
(JSC::isArraySpeculation):
(JSC::isFunctionSpeculation):
(JSC::isInt8ArraySpeculation):
(JSC::isInt16ArraySpeculation):
(JSC::isInt32ArraySpeculation):
(JSC::isUint8ArraySpeculation):
(JSC::isUint8ClampedArraySpeculation):
(JSC::isUint16ArraySpeculation):
(JSC::isUint32ArraySpeculation):
(JSC::isFloat32ArraySpeculation):
(JSC::isFloat64ArraySpeculation):
(JSC::isArgumentsSpeculation):
(JSC::isActionableIntMutableArraySpeculation):
(JSC::isActionableFloatMutableArraySpeculation):
(JSC::isActionableTypedMutableArraySpeculation):
(JSC::isActionableMutableArraySpeculation):
(JSC::isActionableArraySpeculation):
(JSC::isArrayOrOtherSpeculation):
(JSC::isMyArgumentsSpeculation):
(JSC::isInt32Speculation):
(JSC::isDoubleRealSpeculation):
(JSC::isDoubleSpeculation):
(JSC::isNumberSpeculation):
(JSC::isBooleanSpeculation):
(JSC::isOtherSpeculation):
(JSC::isEmptySpeculation):
(JSC::mergeSpeculations):
(JSC::mergeSpeculation):
* bytecode/StructureSet.h:
(JSC::StructureSet::speculationFromStructures):
* bytecode/ValueProfile.h:
(JSC::ValueProfileBase::ValueProfileBase):
(JSC::ValueProfileBase::dump):
(JSC::ValueProfileBase::computeUpdatedPrediction):
(ValueProfileBase):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::initialize):
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::mergeStateAtTail):
* dfg/DFGAbstractState.h:
(JSC::DFG::AbstractState::speculateInt32Unary):
(JSC::DFG::AbstractState::speculateNumberUnary):
(JSC::DFG::AbstractState::speculateBooleanUnary):
(JSC::DFG::AbstractState::speculateInt32Binary):
(JSC::DFG::AbstractState::speculateNumberBinary):
* dfg/DFGAbstractValue.h:
(JSC::DFG::StructureAbstractValue::filter):
(JSC::DFG::StructureAbstractValue::speculationFromStructures):
(JSC::DFG::AbstractValue::AbstractValue):
(JSC::DFG::AbstractValue::clear):
(JSC::DFG::AbstractValue::isClear):
(JSC::DFG::AbstractValue::makeTop):
(JSC::DFG::AbstractValue::clobberStructures):
(JSC::DFG::AbstractValue::isTop):
(JSC::DFG::AbstractValue::set):
(JSC::DFG::AbstractValue::merge):
(JSC::DFG::AbstractValue::filter):
(JSC::DFG::AbstractValue::validateIgnoringValue):
(JSC::DFG::AbstractValue::validate):
(JSC::DFG::AbstractValue::checkConsistency):
(JSC::DFG::AbstractValue::dump):
(AbstractValue):
* dfg/DFGArgumentPosition.h:
(JSC::DFG::ArgumentPosition::ArgumentPosition):
(JSC::DFG::ArgumentPosition::mergeArgumentAwareness):
(JSC::DFG::ArgumentPosition::prediction):
(ArgumentPosition):
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
* dfg/DFGByteCodeParser.cpp:
(ByteCodeParser):
(JSC::DFG::ByteCodeParser::injectLazyOperandSpeculation):
(JSC::DFG::ByteCodeParser::getLocal):
(JSC::DFG::ByteCodeParser::getArgument):
(JSC::DFG::ByteCodeParser::addCall):
(JSC::DFG::ByteCodeParser::getSpeculationWithoutOSRExit):
(JSC::DFG::ByteCodeParser::getSpeculation):
(InlineStackEntry):
(JSC::DFG::ByteCodeParser::handleCall):
(JSC::DFG::ByteCodeParser::handleIntrinsic):
(JSC::DFG::ByteCodeParser::handleGetById):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::fixVariableAccessSpeculations):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::getIndexedPropertyStorageLoadElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixDoubleEdge):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::nameOfVariableAccessData):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::predictArgumentTypes):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::getJSConstantSpeculation):
(JSC::DFG::Graph::isPredictedNumerical):
(JSC::DFG::Graph::byValIsPure):
* dfg/DFGJITCompiler.h:
(JSC::DFG::JITCompiler::getSpeculation):
* dfg/DFGNode.h:
(JSC::DFG::Node::Node):
(JSC::DFG::Node::getHeapPrediction):
(JSC::DFG::Node::predictHeap):
(JSC::DFG::Node::prediction):
(JSC::DFG::Node::predict):
(JSC::DFG::Node::shouldSpeculateInteger):
(JSC::DFG::Node::shouldSpeculateDouble):
(JSC::DFG::Node::shouldSpeculateNumber):
(JSC::DFG::Node::shouldSpeculateBoolean):
(JSC::DFG::Node::shouldSpeculateFinalObject):
(JSC::DFG::Node::shouldSpeculateFinalObjectOrOther):
(JSC::DFG::Node::shouldSpeculateArray):
(JSC::DFG::Node::shouldSpeculateArguments):
(JSC::DFG::Node::shouldSpeculateInt8Array):
(JSC::DFG::Node::shouldSpeculateInt16Array):
(JSC::DFG::Node::shouldSpeculateInt32Array):
(JSC::DFG::Node::shouldSpeculateUint8Array):
(JSC::DFG::Node::shouldSpeculateUint8ClampedArray):
(JSC::DFG::Node::shouldSpeculateUint16Array):
(JSC::DFG::Node::shouldSpeculateUint32Array):
(JSC::DFG::Node::shouldSpeculateFloat32Array):
(JSC::DFG::Node::shouldSpeculateFloat64Array):
(JSC::DFG::Node::shouldSpeculateArrayOrOther):
(JSC::DFG::Node::shouldSpeculateObject):
(JSC::DFG::Node::shouldSpeculateCell):
(Node):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::setPrediction):
(JSC::DFG::PredictionPropagationPhase::mergePrediction):
(JSC::DFG::PredictionPropagationPhase::propagate):
(JSC::DFG::PredictionPropagationPhase::doRoundOfDoubleVoting):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::fillStorage):
(JSC::DFG::SpeculativeJIT::writeBarrier):
(JSC::DFG::GPRTemporary::GPRTemporary):
(JSC::DFG::FPRTemporary::FPRTemporary):
(JSC::DFG::SpeculativeJIT::compilePeepHoleDoubleBranch):
(JSC::DFG::SpeculativeJIT::compilePeepHoleObjectEquality):
(JSC::DFG::SpeculativeJIT::compilePeepHoleBranch):
(JSC::DFG::SpeculativeJIT::compile):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::compileGetCharCodeAt):
(JSC::DFG::SpeculativeJIT::compileGetByValOnString):
(JSC::DFG::SpeculativeJIT::compileValueToInt32):
(JSC::DFG::SpeculativeJIT::compileDoubleAsInt32):
(JSC::DFG::SpeculativeJIT::compileInt32ToDouble):
(JSC::DFG::SpeculativeJIT::compileGetTypedArrayLength):
(JSC::DFG::SpeculativeJIT::compileGetByValOnIntTypedArray):
(JSC::DFG::SpeculativeJIT::compilePutByValForIntTypedArray):
(JSC::DFG::SpeculativeJIT::compileGetByValOnFloatTypedArray):
(JSC::DFG::SpeculativeJIT::compilePutByValForFloatTypedArray):
(JSC::DFG::SpeculativeJIT::compileInstanceOf):
(JSC::DFG::SpeculativeJIT::compileAdd):
(JSC::DFG::SpeculativeJIT::compileArithSub):
(JSC::DFG::SpeculativeJIT::compileArithNegate):
(JSC::DFG::SpeculativeJIT::compileArithMul):
(JSC::DFG::SpeculativeJIT::compileArithMod):
(JSC::DFG::SpeculativeJIT::compare):
(JSC::DFG::SpeculativeJIT::compileStrictEq):
(JSC::DFG::SpeculativeJIT::compileGetIndexedPropertyStorage):
(JSC::DFG::SpeculativeJIT::compileGetByValOnArguments):
(JSC::DFG::SpeculativeJIT::compileGetArgumentsLength):
(JSC::DFG::SpeculativeJIT::compileRegExpExec):
* dfg/DFGSpeculativeJIT.h:
(DFG):
(JSC::DFG::ValueSource::forSpeculation):
(SpeculativeJIT):
(GPRTemporary):
(FPRTemporary):
(JSC::DFG::SpecDoubleOperand::SpecDoubleOperand):
(JSC::DFG::SpecDoubleOperand::~SpecDoubleOperand):
(JSC::DFG::SpecDoubleOperand::fpr):
(JSC::DFG::SpecCellOperand::SpecCellOperand):
(JSC::DFG::SpecCellOperand::~SpecCellOperand):
(JSC::DFG::SpecCellOperand::gpr):
(JSC::DFG::SpecBooleanOperand::SpecBooleanOperand):
(JSC::DFG::SpecBooleanOperand::~SpecBooleanOperand):
(JSC::DFG::SpecBooleanOperand::gpr):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::fillSpeculateIntInternal):
(JSC::DFG::SpeculativeJIT::fillSpecDouble):
(JSC::DFG::SpeculativeJIT::fillSpecCell):
(JSC::DFG::SpeculativeJIT::fillSpecBoolean):
(JSC::DFG::SpeculativeJIT::compileObjectEquality):
(JSC::DFG::SpeculativeJIT::compileObjectToObjectOrOtherEquality):
(JSC::DFG::SpeculativeJIT::compilePeepHoleObjectToObjectOrOtherEquality):
(JSC::DFG::SpeculativeJIT::compileDoubleCompare):
(JSC::DFG::SpeculativeJIT::compileLogicalNot):
(JSC::DFG::SpeculativeJIT::emitBranch):
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::fillSpeculateIntInternal):
(JSC::DFG::SpeculativeJIT::fillSpecDouble):
(JSC::DFG::SpeculativeJIT::fillSpecCell):
(JSC::DFG::SpeculativeJIT::fillSpecBoolean):
(JSC::DFG::SpeculativeJIT::compileObjectEquality):
(JSC::DFG::SpeculativeJIT::compileObjectToObjectOrOtherEquality):
(JSC::DFG::SpeculativeJIT::compilePeepHoleObjectToObjectOrOtherEquality):
(JSC::DFG::SpeculativeJIT::compileDoubleCompare):
(JSC::DFG::SpeculativeJIT::compileLogicalNot):
(JSC::DFG::SpeculativeJIT::emitBranch):
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGVariableAccessData.h:
(JSC::DFG::VariableAccessData::VariableAccessData):
(JSC::DFG::VariableAccessData::predict):
(JSC::DFG::VariableAccessData::nonUnifiedPrediction):
(JSC::DFG::VariableAccessData::prediction):
(JSC::DFG::VariableAccessData::argumentAwarePrediction):
(JSC::DFG::VariableAccessData::mergeArgumentAwarePrediction):
(JSC::DFG::VariableAccessData::shouldUseDoubleFormatAccordingToVote):
(JSC::DFG::VariableAccessData::makePredictionForDoubleFormat):
(VariableAccessData):
2012-06-06 Filip Pizlo <fpizlo@apple.com>
Global object variable accesses should not require an extra load
https://bugs.webkit.org/show_bug.cgi?id=88385
Reviewed by Gavin Barraclough and Geoffrey Garen.
Previously, if you wanted to access a global variable, you'd first have
to load the register array from the appropriate global object and then
either load or store at an offset to the register array. This is because
JSGlobalObject inherited from JSVariableObject, and JSVariableObject is
designed with the pessimistic assumption that its register array may
point into the call stack. This is never the case for global objects.
Hence, even though the global object may add more registers at any time,
it does not need to store them in a contiguous array. It can use a
SegmentedVector or similar.
This patch refactors global objects and variable objects as follows:
- The functionality to track variables in an indexable array using a
SymbolTable to map names to indices is moved into JSSymbolTableObject,
which is now a supertype of JSVariableObject. JSVariableObject is now
just a holder for a registers array and implements the registerAt()
method that is left abstract in JSSymbolTableObject. Because all users
of JSVariableObject know whether they are a JSStaticScopeObject,
JSActivation, or JSGlobalObject, this "abstract" method is not virtual;
instead the utility methods that would call registerAt() are now
template functions that require you to know statically what subtype of
JSSymbolTableObject you're using (JSVariableObject or something else),
so that registerAt() can be statically bound.
- A new class is added called JSSegmentedVariableObject, which only
differs from JSVariableObject in how it allocates registers. It uses a
SegmentedVector instead of manually managing a pointer to a contiguous
slab of registers. This changes the interface somewhat; for example
with JSVariableObject if you wanted to add a register you had to do
it yourself since the JSVariableObject didn't know how the registers
array ought to be allocated. With JSSegmentedVariableObject you can
just call addRegisters(). JSSegmentedVariableObject preserves the
invariant that once you get a pointer into a register, that pointer
will continue to be valid so long as the JSSegmentedVariableObject is
alive. This allows the JITs and interpreters to skip the extra load.
- JSGlobalObject now inherits from JSSegmentedVariableObject. For now
(and possibly forever) it is the only subtype of this new class.
- The bytecode format is changed so that get_global_var and
put_global_var have a pointer to the register directly rather than
having an index. A convenience method is provided in
JSSegmentedVariableObject to get the index given a a pointer, which is
used for assertions and debug dumps.
This appears to be a 1% across the board win.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
* bytecode/Instruction.h:
(Instruction):
(JSC::Instruction::Instruction):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::ResolveResult::registerPointer):
(JSC):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::retrieveLastUnaryOp):
(JSC::BytecodeGenerator::resolve):
(JSC::BytecodeGenerator::resolveConstDecl):
(JSC::BytecodeGenerator::emitGetStaticVar):
(JSC::BytecodeGenerator::emitPutStaticVar):
* bytecompiler/BytecodeGenerator.h:
(ResolveResult):
(BytecodeGenerator):
* dfg/DFGAssemblyHelpers.h:
(AssemblyHelpers):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::globalVarLoadElimination):
(JSC::DFG::CSEPhase::globalVarStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::dump):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::globalObjectFor):
(Graph):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasVarNumber):
(Node):
(JSC::DFG::Node::hasRegisterPointer):
(JSC::DFG::Node::registerPointer):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* heap/Heap.h:
(Heap):
(JSC::Heap::isWriteBarrierEnabled):
(JSC):
* interpreter/Interpreter.cpp:
(JSC::Interpreter::execute):
(JSC::Interpreter::privateExecute):
* jit/JITPropertyAccess.cpp:
(JSC::JIT::emit_op_get_global_var):
(JSC::JIT::emit_op_put_global_var):
* jit/JITPropertyAccess32_64.cpp:
(JSC::JIT::emit_op_get_global_var):
(JSC::JIT::emit_op_put_global_var):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
* runtime/JSGlobalObject.cpp:
(JSC):
(JSC::JSGlobalObject::put):
(JSC::JSGlobalObject::putDirectVirtual):
(JSC::JSGlobalObject::defineOwnProperty):
(JSC::JSGlobalObject::visitChildren):
(JSC::JSGlobalObject::addStaticGlobals):
(JSC::JSGlobalObject::getOwnPropertySlot):
(JSC::JSGlobalObject::getOwnPropertyDescriptor):
* runtime/JSGlobalObject.h:
(JSGlobalObject):
(JSC::JSGlobalObject::JSGlobalObject):
(JSC):
(JSC::JSGlobalObject::hasOwnPropertyForWrite):
* runtime/JSSegmentedVariableObject.cpp: Added.
(JSC):
(JSC::JSSegmentedVariableObject::findRegisterIndex):
(JSC::JSSegmentedVariableObject::addRegisters):
(JSC::JSSegmentedVariableObject::visitChildren):
* runtime/JSSegmentedVariableObject.h: Added.
(JSC):
(JSSegmentedVariableObject):
(JSC::JSSegmentedVariableObject::registerAt):
(JSC::JSSegmentedVariableObject::assertRegisterIsInThisObject):
(JSC::JSSegmentedVariableObject::JSSegmentedVariableObject):
(JSC::JSSegmentedVariableObject::finishCreation):
* runtime/JSStaticScopeObject.cpp:
(JSC::JSStaticScopeObject::put):
(JSC::JSStaticScopeObject::putDirectVirtual):
(JSC::JSStaticScopeObject::getOwnPropertySlot):
* runtime/JSSymbolTableObject.cpp: Added.
(JSC):
(JSC::JSSymbolTableObject::destroy):
(JSC::JSSymbolTableObject::deleteProperty):
(JSC::JSSymbolTableObject::getOwnPropertyNames):
(JSC::JSSymbolTableObject::putDirectVirtual):
(JSC::JSSymbolTableObject::isDynamicScope):
* runtime/JSSymbolTableObject.h: Added.
(JSC):
(JSSymbolTableObject):
(JSC::JSSymbolTableObject::symbolTable):
(JSC::JSSymbolTableObject::JSSymbolTableObject):
(JSC::JSSymbolTableObject::finishCreation):
(JSC::symbolTableGet):
(JSC::symbolTablePut):
(JSC::symbolTablePutWithAttributes):
* runtime/JSVariableObject.cpp:
(JSC):
* runtime/JSVariableObject.h:
(JSVariableObject):
(JSC::JSVariableObject::JSVariableObject):
(JSC::JSVariableObject::finishCreation):
(JSC):
* runtime/WriteBarrier.h:
2012-06-06 Filip Pizlo <fpizlo@apple.com>
DFG arguments access slow path should not crash if the arguments haven't been created
https://bugs.webkit.org/show_bug.cgi?id=88471
Reviewed by Gavin Barraclough.
* dfg/DFGCCallHelpers.h:
(JSC::DFG::CCallHelpers::setupArgumentsWithExecState):
(CCallHelpers):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-06-06 Michael Saboff <msaboff@apple.com>
ENH: Add Logging to GC Marking Phase
https://bugs.webkit.org/show_bug.cgi?id=88364
Reviewed by Filip Pizlo.
Log GC marking to stderr or a file. The logging in controlled
with the define ENABLE_OBJECT_MARK_LOGGING in wtf/Platform.h.
If DATA_LOG_TO_FILE in wtf/DataLog.cpp is set to 1, output is
logged to a file otherwise it is logged to stderr.
When logging is enabled, the GC is built single threaded since the
log output from the various threads isn't buffered and output in a
thread safe manner.
* heap/Heap.cpp:
(JSC::Heap::markRoots):
* heap/MarkStack.cpp:
(JSC::MarkStackThreadSharedData::resetChildren):
(JSC::MarkStackThreadSharedData::childVisitCount):
(JSC::MarkStackThreadSharedData::markingThreadMain):
(JSC::MarkStackThreadSharedData::markingThreadStartFunc):
(JSC::MarkStackThreadSharedData::MarkStackThreadSharedData):
(JSC::MarkStackThreadSharedData::reset):
* heap/MarkStack.h:
(MarkStackThreadSharedData):
(MarkStack):
(JSC::MarkStack::sharedData):
(JSC::MarkStack::resetChildCount):
(JSC::MarkStack::childCount):
(JSC::MarkStack::incrementChildCount):
* runtime/JSArray.cpp:
(JSC::JSArray::visitChildren):
* runtime/JSCell.cpp:
(JSC::JSCell::className):
* runtime/JSCell.h:
(JSCell):
(JSC::JSCell::visitChildren):
* runtime/JSString.cpp:
(JSC::JSString::visitChildren):
* runtime/JSString.h:
(JSString):
* runtime/Structure.h:
(JSC::MarkStack::internalAppend):
2012-06-06 Gavin Barraclough <barraclough@apple.com>
Assigning to a static property should not change iteration order
https://bugs.webkit.org/show_bug.cgi?id=88401
Reviewed by Geoff Garen.
A specific iteration order is not defined by the spec, but test-262 somewhat tenuously
requires that it is at least stable, e.g. ch10/10.4/10.4.2/S10.4.2_A1.1_T1.js
Whilst it is not clear that this behavior really arises from the specification, it
would seem like common sense to conform to this.
The problem here is that we allow properties in the structure to shadow those in the
static table, and we iterate the properties in the structure first - which means that
as values of existing properties are modified, their iteration order changes too.
The easy fix is to iterate the properties from the static table first. This has a
further benefit, since it will mean that user added properties will come after those
present in the static table (respected the expected insertion-order).
* runtime/JSObject.cpp:
(JSC::JSObject::getOwnPropertyNames):
- Iterate static properties first.
2012-06-06 Andy Wingo <wingo@igalia.com>
Ensure consistent order of evaluation in LLInt slow paths
https://bugs.webkit.org/show_bug.cgi?id=88409
Reviewed by Geoffrey Garen.
* llint/LLIntSlowPaths.cpp:
(slow_path_mul)
(slow_path_sub)
(slow_path_div)
(slow_path_mod)
(slow_path_lshift)
(slow_path_rshift)
(slow_path_urshift)
(slow_path_bitand)
(slow_path_bitor)
(slow_path_bitxor): Avoid calling toNumber, toInt32, or toUInt32
multiple times without intervening sequence points. Fixes
fast/js/exception-sequencing-binops.html with GCC 4.7 on x86-64
Linux, which reordered evaluation of the arguments to fmod.
2012-06-06 Andy Wingo <wingo@igalia.com>
[GTK] Enable the LLInt
https://bugs.webkit.org/show_bug.cgi?id=88315
Reviewed by Filip Pizlo.
* GNUmakefile.am: Add rules to generate LLIntDesiredOffsets.h and
LLIntAssembly.h.
* GNUmakefile.list.am: Add offlineasm and llint files to the
dist. Add LLInt source files to the build.
* llint/LowLevelInterpreter.asm (crash): Generate a store of
0xbbadbeef to a register, not to a constant. Otherwise, gas was
failing to assemble result.
* offlineasm/asm.rb (labelReference): Generate a
SYMBOL_STRING_RELOCATION instead of a SYMBOL_STRING, so that we go
through the PLT on ELF systems.
2012-06-06 Andy Wingo <wingo@igalia.com>
REGRESSION (r106478): None of the Paper.js JavaScript examples work
https://bugs.webkit.org/show_bug.cgi?id=87158
Reviewed by Michael Saboff.
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::resolve): If we have to bail out to
dynamicResolve(), only skip static scopes from the head of the
scope chain. Before, we were also skipping activations with
direct eval as well, which was incorrect.
2012-06-06 Dan Bernstein <mitz@apple.com>
Reverted r119567, the fix for <http://webkit.org/b/88378>, because it broke the 32-bit build.
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::emitAllocateBasicJSObject):
* jit/JITInlineMethods.h:
(JSC::JIT::emitAllocateBasicJSObject):
* llint/LowLevelInterpreter.asm:
* runtime/JSGlobalData.h:
(JSGlobalData):
* runtime/JSGlobalThis.cpp:
(JSC::JSGlobalThis::setUnwrappedObject):
* runtime/JSObject.cpp:
(JSC::JSObject::visitChildren):
(JSC::JSObject::createInheritorID):
* runtime/JSObject.h:
(JSObject):
(JSC::JSObject::resetInheritorID):
(JSC):
(JSC::JSObject::offsetOfInheritorID):
(JSC::JSObject::inheritorID):
2012-06-05 Yuqiang Xian <yuqiang.xian@intel.com>
Improve Math.round and Math.floor intrinsic
https://bugs.webkit.org/show_bug.cgi?id=88314
Reviewed by Filip Pizlo.
Currently we call a native function from the JIT code to complete the
"round" and "floor" operations. We could inline some fast paths
especially for those positive values on the platforms where floating
point truncation is supported.
This brings 3% gain on Kraken, especially 32% on audio-oscillator,
and slight win on SunSpider, measured on IA32.
* jit/ThunkGenerators.cpp:
(JSC::floorThunkGenerator):
(JSC):
(JSC::roundThunkGenerator):
2012-06-05 Gavin Barraclough <barraclough@apple.com>
Remove JSObject::m_inheritorID
https://bugs.webkit.org/show_bug.cgi?id=88378
Reviewed by Geoff Garen.
This is rarely used, and not performance critical (the commonly accessed copy is cached on JSFunction),
and most objects don't need an inheritorID (this value is only used if the object is used as a prototype).
Instead use a private named value in the object's property storage.
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::emitAllocateBasicJSObject):
- No need m_inheritorID to initialize!
* jit/JITInlineMethods.h:
(JSC::JIT::emitAllocateBasicJSObject):
- No need m_inheritorID to initialize!
* llint/LowLevelInterpreter.asm:
- No need m_inheritorID to initialize!
* runtime/JSGlobalData.h:
(JSGlobalData):
- Added private name 'm_inheritorIDKey'.
* runtime/JSGlobalThis.cpp:
(JSC::JSGlobalThis::setUnwrappedObject):
- resetInheritorID is now passed a JSGlobalData&.
* runtime/JSObject.cpp:
(JSC::JSObject::visitChildren):
- No m_inheritorID to be marked.
(JSC::JSObject::createInheritorID):
- Store the newly created inheritorID in the property map.
* runtime/JSObject.h:
(JSC::JSObject::resetInheritorID):
- Remove the inheritorID from property storage.
(JSC::JSObject::inheritorID):
- Read the inheritorID from property storage.
2012-06-05 Filip Pizlo <fpizlo@apple.com>
DFG CFG simplification should not attempt to deref nodes inside of an unreachable subgraph
https://bugs.webkit.org/show_bug.cgi?id=88362
Reviewed by Gavin Barraclough.
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::fixPhis):
(JSC::DFG::CFGSimplificationPhase::removePotentiallyDeadPhiReference):
2012-06-05 Mark Hahnenberg <mhahnenberg@apple.com>
Entry into JSC should CRASH() if the Heap is busy
https://bugs.webkit.org/show_bug.cgi?id=88355
Reviewed by Geoffrey Garen.
Interpreter::execute() returns jsNull() right now if we try to enter it while
the Heap is busy (e.g. with a collection), which is okay, but some code paths
that call Interpreter::execute() allocate objects before checking if the Heap
is busy. Attempting to execute JS code while the Heap is busy should not be
allowed and should be enforced by a release-mode CRASH() to prevent vague,
unhelpful backtraces later on if somebody makes a mistake. Normally, recursively
executing JS code is okay, e.g. for evals, but it should not occur during a
Heap allocation or collection because the Heap is not guaranteed to be in a
consistent state (especially during collections). We are protected from
executing JS on the same Heap concurrently on two separate threads because
they must each take a JSLock first. However, we are not protected from reentrant
execution of JS on the same thread because JSLock allows reentrancy. Therefore,
we should fail early if we detect an entrance into JS code while the Heap is busy.
* heap/Heap.cpp: Changed Heap::collect so that it sets the m_operationInProgress field
at the beginning of collection and then unsets it at the end so that it is set at all
times throughout the duration of a collection rather than sporadically during various
phases. There is no reason to unset during a collection because our collector does
not currently support running additional JS between the phases of a collection.
(JSC::Heap::getConservativeRegisterRoots):
(JSC::Heap::markRoots):
(JSC::Heap::collect):
* interpreter/Interpreter.cpp:
(JSC::Interpreter::execute): Crash if the Heap is busy.
* runtime/Completion.cpp: Crash if the Heap is busy. We do it here before we call
Interpreter::execute() because we do some allocation prior to calling execute() which
could cause Heap corruption if, for example, that allocation caused a collection.
(JSC::evaluate):
2012-06-05 Dongwoo Im <dw.im@samsung.com>
Add 'isProtocolHandlerRegistered' and 'unregisterProtocolHandler'.
https://bugs.webkit.org/show_bug.cgi?id=73176
Reviewed by Adam Barth.
Two more APIs are added in Custom Scheme Handler specification.
http://dev.w3.org/html5/spec/Overview.html#custom-handlers
One is 'isProtocolHandlerRegistered' to query whether the specific URL
is registered or not.
The other is 'unregisterProtocolHandler' to remove the registered URL.
* Configurations/FeatureDefines.xcconfig: Add a macro 'ENABLE_CUSTOM_SCHEME_HANDLER'.
2012-06-04 Filip Pizlo <fpizlo@apple.com>
DFG CFG simplification should correct the variables at the head of the predecessor block
https://bugs.webkit.org/show_bug.cgi?id=88284
Reviewed by Geoffrey Garen.
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::mergeBlocks):
2012-06-04 Geoffrey Garen <ggaren@apple.com>
Unreviewed.
Rolled out r119364 because it's still causing crashes (when running
v8-earley in release builds of DRT)
This time for sure!
* heap/Heap.cpp:
(JSC::Heap::collect):
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep):
* heap/MarkedBlock.h:
(JSC::MarkedBlock::resetAllocator):
(JSC):
* heap/MarkedSpace.cpp:
(JSC::ResetAllocator::operator()):
(JSC):
(JSC::MarkedSpace::resetAllocators):
(JSC::MarkedSpace::sweepWeakSets):
* heap/MarkedSpace.h:
(MarkedSpace):
* heap/WeakBlock.cpp:
(JSC::WeakBlock::sweep):
* heap/WeakSet.cpp:
(JSC::WeakSet::sweep):
(JSC::WeakSet::tryFindAllocator):
* heap/WeakSet.h:
(JSC::WeakSet::shrink):
2012-06-04 Filip Pizlo <fpizlo@apple.com>
DFG arguments simplification should have rationalized handling of TearOffArguments
https://bugs.webkit.org/show_bug.cgi?id=88206
Reviewed by Geoffrey Garen.
- Accesses to the unmodified arguments register ought to have the same effect on
alias/escape analysis of arguments as accesses to the mutable arguments register.
- The existence of TearOffArguments should not get in the way of arguments aliasing.
- TearOffArguments should be eliminated if CreateArguments is eliminated.
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUse):
2012-06-04 Gavin Barraclough <barraclough@apple.com>
Remove enabledProfilerReference
https://bugs.webkit.org/show_bug.cgi?id=88258
Reviewed by Michael Saboff.
Make the enabled profiler a member of JSGlobalData, and switch code that accesses it to do so directly
via the JSGlobalData, rather than holding a Profiler** reference to it. Do not pass the Profiler**
reference to JIT code. This patch does not change the stack layout on entry into JIT code (passing an
unused void* instead), since this is an intrusive change better handled in a separate patch.
* interpreter/Interpreter.cpp:
(JSC::Interpreter::throwException):
(JSC::Interpreter::execute):
(JSC::Interpreter::executeCall):
(JSC::Interpreter::executeConstruct):
(JSC::Interpreter::privateExecute):
* jit/JITCode.h:
(JSC::JITCode::execute):
- Don't pass Profiler** to JIT code.
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_profile_will_call):
(JSC::JIT::emit_op_profile_did_call):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_profile_will_call):
(JSC::JIT::emit_op_profile_did_call):
* jit/JITStubs.cpp:
(JSC):
(JSC::ctiTrampoline):
(JSC::ctiVMThrowTrampoline):
(JSC::ctiOpThrowNotCaught):
(JSC::JITThunks::JITThunks):
(JSC::DEFINE_STUB_FUNCTION):
- For ARM_THUMB2, rename ENABLE_PROFILER_REFERENCE_OFFSET to FIRST_STACK_ARGUMENT (which is how it is being used).
- For MIPS, remove ENABLE_PROFILER_REFERENCE_OFFSET.
* jit/JITStubs.h:
(JITStackFrame):
(JSC):
- Renamed enabledProfilerReference to unusedX.
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LowLevelInterpreter.asm:
* profiler/Profiler.cpp:
(JSC):
(JSC::Profiler::startProfiling):
(JSC::Profiler::stopProfiling):
* profiler/Profiler.h:
(Profiler):
- Removed s_sharedEnabledProfilerReference, enabledProfilerReference().
* runtime/JSGlobalData.cpp:
(JSC::JSGlobalData::JSGlobalData):
* runtime/JSGlobalData.h:
(JSC):
(JSC::JSGlobalData::enabledProfiler):
(JSGlobalData):
- Added m_enabledProfiler, enabledProfiler().
* runtime/JSGlobalObject.cpp:
(JSC::JSGlobalObject::~JSGlobalObject):
2012-06-04 Filip Pizlo <fpizlo@apple.com>
get_argument_by_val should be profiled everywhere
https://bugs.webkit.org/show_bug.cgi?id=88205
Reviewed by Geoffrey Garen.
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emitSlow_op_get_argument_by_val):
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
2012-06-04 Filip Pizlo <fpizlo@apple.com>
DFG arguments simplification takes unkindly to direct accesses to the arguments register
https://bugs.webkit.org/show_bug.cgi?id=88261
Reviewed by Geoffrey Garen.
Fixed arguments simplification for direct accesses to the arguments register, which may
arise if CSE had not run. Fixed CSE so that it does run prior to arguments simplification,
by making it a full-fledged member of the fixpoint. Fixed other issues in arguments
simplification, like realizing that it needs to bail if there is a direct assignment to
the arguments register, and failing to turn CreateArguments into PhantomArguments. Also
fixed CSE's handling of store elimination of captured locals in the presence of a
GetMyArgumentByVal (or one of its friends), and fixed CSE to correctly fixup variables at
tail if the Flush it removes is the last operation on a local in a basic block.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
(JSC::DFG::ArgumentsSimplificationPhase::isOKToOptimize):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::run):
(JSC::DFG::CSEPhase::setLocalStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
(CSEPhase):
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
2012-06-04 Anders Carlsson <andersca@apple.com>
Fix a struct/class mismatch.
* heap/Handle.h:
(Handle):
2012-06-04 David Kilzer <ddkilzer@apple.com>
BUILD FIX: FeatureDefines.xcconfig should match across projects
* Configurations/FeatureDefines.xcconfig:
- Add missing ENABLE_LEGACY_CSS_VENDOR_PREFIXES.
2012-06-02 Geoffrey Garen <ggaren@apple.com>
Weak pointer finalization should be lazy
https://bugs.webkit.org/show_bug.cgi?id=87599
Reviewed by Sam Weinig.
This time for sure!
* heap/Heap.cpp:
(JSC::Heap::collect): Don't sweep eagerly -- we'll sweep lazily instead.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep): Sweep our weak set before we sweep our other
destructors -- this is our last chance to run weak set finalizers before
we recycle our memory.
* heap/MarkedBlock.h:
(JSC::MarkedBlock::resetAllocator):
* heap/MarkedSpace.cpp:
(JSC::MarkedSpace::resetAllocators):
* heap/MarkedSpace.h:
(JSC::MarkedSpace::resetAllocators): Don't force allocator reset anymore.
It will happen automatically when a weak set is swept. It's simpler to
have only one canonical way for this to happen, and it wasn't buying
us anything to do it eagerly.
* heap/WeakBlock.cpp:
(JSC::WeakBlock::sweep): Don't short-circuit a sweep unless we know
the sweep would be a no-op. If even one finalizer is pending, we need to
run it, since we won't get another chance.
* heap/WeakSet.cpp:
(JSC::WeakSet::sweep): This loop can be simpler now that
WeakBlock::sweep() does what we mean.
Reset our allocator after a sweep because this is the optimal time to
start trying to recycle old weak pointers.
(JSC::WeakSet::tryFindAllocator): Don't sweep when searching for an
allocator because we've swept already, and forcing a new sweep would be
wasteful.
* heap/WeakSet.h:
(JSC::WeakSet::shrink): Be sure to reset our allocator after a shrink
because the shrink may have removed the block the allocator was going to
allocate out of.
2012-06-02 Filip Pizlo <fpizlo@apple.com>
If the DFG bytecode parser detects that op_method_check has gone polymorphic, it
shouldn't revert all the way to GetById/GetByIdFlush
https://bugs.webkit.org/show_bug.cgi?id=88176
Reviewed by Geoffrey Garen.
Refactored the code so that the op_method_check case of the parser gracefully falls
through to all of the goodness of the normal op_get_by_id case.
* dfg/DFGByteCodeParser.cpp:
(ByteCodeParser):
(JSC::DFG::ByteCodeParser::handleGetById):
(DFG):
(JSC::DFG::ByteCodeParser::parseBlock):
2012-06-02 Filip Pizlo <fpizlo@apple.com>
DFG CSE should be able to eliminate unnecessary flushes of arguments and captured variables
https://bugs.webkit.org/show_bug.cgi?id=87929
Reviewed by Geoffrey Garen.
Slight speed-up on V8. Big win (up to 50%) on programs that inline very small functions.
This required a bunch of changes:
- The obvious change is making CSE essentially ignore whether or not the set of
operations between the Flush and the SetLocal can exit, and instead focus on whether or
not that set of operations can clobber the world or access local variables. This code
is now refactored to return a set of flags indicating any of these events, and the CSE
decides what to do based on those flags. If the set of operations is non-clobbering
and non-accessing, then the Flush is turned into a Phantom on the child of the
SetLocal. This expands the liveness of the relevant variable but virtually guarantees
that it will be register allocated and not flushed to the stack. So, yeah, this patch
is a lot of work to save a few stores to the stack.
- Previously, CheckArgumentsNotCreated was optimized "lazily" in that you only knew if
it was a no-op if you were holding onto a CFA abstract state. But this would make the
CSE act pessimistically, since it doesn't use the CFA. Hence, this patch changes the
constant folding phase into something more broad; it now fixes up
CheckArgumentsNotCreated nodes by turning them into phantoms if it knows that they are
no-ops.
- Arguments simplification was previously relying on this very strange PhantomArguments
node, which had two different meanings: for normal execution it meant the empty value
but for OSR exit it meant that the arguments should be reified. This produces problems
when set SetLocals to the captured arguments registers are CSE'd away, since we'd be
triggering reification of arguments without having initialized the arguments registers
to empty. The cleanest solution was to fix PhantomArguments to have one meaning:
namely, arguments reification on OSR exit. Hence, this patch changes arguments
simplification to change SetLocal of CreateArguments on the arguments registers to be
a SetLocal of Empty.
- Argument value recoveries were previously derived from the value source of the
arguments at the InlineStart. But that relies on all SetLocals to arguments having
been flushed. It's possible that we could have elided the SetLocal to the arguments
at the callsite because there were subsequent SetLocals to the arguments inside of the
callee, in which case the InlineStart would get the wrong information. Hence, this
patch changes argument value recovery computation to operate over the ArgumentPositions
directly.
- But that doesn't actually work, because previously, there was no way to link an
InlineStart back to the corresponding ArgumentPositions, at least not without some
ugliness. So this patch instates the rule that the m_argumentPositions vector consists
of disjoint subsequences such that each subsequence corresponds to an inline callsite
and can be identified by its first index, and within each subsequence are the
ArgumentPositions of all of the arguments ordered by argument index. This required
flipping the order in which ArgumentPositions are added to the vector, and giving
InlineStart an operand that indicates the start of that inline callsite's
ArgumentPosition subsequence.
- This patch also revealed a nasty bug in the reification of arguments in inline call
frames on OSR exit. Since the reification was happening after the values of virtual
registers were recovered, the value recoveries of the inline arguments were wrong.
Hence using operationCreateInlinedArguments is wrong. For example a value recovery
might say that you have to box a double, but if we had already boxed it then boxing
it a second time will result in garbage. The specific case of this bug was this patch
uncovered was that now it is possible for an inline call frame to not have any valid
value recoveries for any inline arguments, if the optimization elides all argument
flushes, while at the same time optimizing away arguments creation. Then OSR exit
would try to recover the arguments using the inline call frame, which had bogus
information, and humorous crashes would ensue. This patch fixes this issue by moving
arguments reification to after call frame reification, so that arguments reification
can always use operationCreateArguments instead of operationCreateInlinedArguments.
- This patch may turn a Flush into a Phantom. That's kind of the whole point. But that
broke forward speculation checks, which knew to look for a Flush prior to a SetLocal
but didn't know that there could alternatively be a Phantom in place of the Flush.
This patch fixes that by augmenting the forward speculation check logic.
- Finally, in the process of having fun with all of the above, I realized that my DFG
validation was not actually running on every phase like I had originally designed it
to. In fact it was only running just after bytecode parsing. I initially tried to
make it run in every phase but found that this causes some tests to timeout
(specifically the evil fuzzing ones), so I decided on a compromise where: (i) in
release mode validation never runs, (ii) in debug mode validation will run just
after parsing and just before the backend, and (iii) it's possible with a simple
switch to enable validation to run on every phase.
Luckily all of the above issues were already covered by the 77 or so DFG-specific
layout tests. Hence, this patch does not introduce any new tests despite being so
meaty.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGArgumentPosition.h:
(JSC::DFG::ArgumentPosition::prediction):
(JSC::DFG::ArgumentPosition::doubleFormatState):
(JSC::DFG::ArgumentPosition::shouldUseDoubleFormat):
(ArgumentPosition):
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::SetLocalStoreEliminationResult::SetLocalStoreEliminationResult):
(SetLocalStoreEliminationResult):
(JSC::DFG::CSEPhase::setLocalStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCommon.h:
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
* dfg/DFGNode.h:
(Node):
(JSC::DFG::Node::hasArgumentPositionStart):
(JSC::DFG::Node::argumentPositionStart):
* dfg/DFGOSRExitCompiler32_64.cpp:
(JSC::DFG::OSRExitCompiler::compileExit):
* dfg/DFGOSRExitCompiler64.cpp:
(JSC::DFG::OSRExitCompiler::compileExit):
* dfg/DFGPhase.cpp:
(DFG):
* dfg/DFGPhase.h:
(Phase):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::forwardSpeculationCheck):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-06-02 Geoffrey Garen <ggaren@apple.com>
DOM string cache should hash pointers, not characters
https://bugs.webkit.org/show_bug.cgi?id=88175
Reviewed by Phil Pizlo and Sam Weinig.
* heap/Weak.h:
(JSC::weakAdd):
(JSC::weakRemove): Made these function templates slightly more generic
to accommodate new client types.
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG CFA should know that PutByVal can clobber the world
https://bugs.webkit.org/show_bug.cgi?id=88155
Reviewed by Gavin Barraclough.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG CFA should mark basic blocks as having constants if local accesses yield constants
https://bugs.webkit.org/show_bug.cgi?id=88153
Reviewed by Gavin Barraclough.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG arguments simplification phase uses a node.codeOrigin after appending a node
https://bugs.webkit.org/show_bug.cgi?id=88151
Reviewed by Geoffrey Garen.
The right thing to do is to save the CodeOrigin before appending to the graph.
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG should not emit unnecessary speculation checks when performing an int32 to double conversion on
a value that is proved to be a number, predicted to be an int32, but not proved to be an int32
https://bugs.webkit.org/show_bug.cgi?id=88146
Reviewed by Gavin Barraclough.
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileInt32ToDouble):
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG constant folding search for the last local access skips the immediately previous local access
https://bugs.webkit.org/show_bug.cgi?id=88141
Reviewed by Michael Saboff.
If you use a loop in the style of:
for (i = start; i--;)
then you need to remember that the first value of 'i' that the loop body will see is 'start - 1'.
Hence the following is probably wrong:
for (i = start - 1; i--;)
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
2012-06-01 Filip Pizlo <fpizlo@apple.com>
DFG constant folding should be OK with GetLocal of captured variables having a constant
https://bugs.webkit.org/show_bug.cgi?id=88137
Reviewed by Gavin Barraclough.
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
2012-05-31 Mark Hahnenberg <mhahnenberg@apple.com>
JSGlobalObject does not mark m_privateNameStructure
https://bugs.webkit.org/show_bug.cgi?id=88023
Rubber stamped by Gavin Barraclough.
* runtime/JSGlobalObject.cpp:
(JSC::JSGlobalObject::visitChildren): We need to mark this so it doesn't get
inadvertently garbage collected.
2012-05-31 Erik Arvidsson <arv@chromium.org>
Make DOM Exceptions Errors
https://bugs.webkit.org/show_bug.cgi?id=85078
Reviewed by Oliver Hunt.
WebIDL mandates that exceptions should have Error.prototype on its prototype chain.
For JSC we have access to the Error.prototype from the binding code.
For V8 we set a field in the WrapperTypeInfo and when the constructor function is created we
set the prototype as needed.
Updated test: fast/dom/DOMException/prototype-object.html
* JavaScriptCore.xcodeproj/project.pbxproj:
* runtime/JSGlobalObject.cpp:
(JSC::JSGlobalObject::reset):
* runtime/JSGlobalObject.h:
(JSC):
(JSGlobalObject):
(JSC::JSGlobalObject::errorPrototype):
2012-05-31 Andy Wingo <wingo@igalia.com>
Fix reference to unset variable in debug mode
https://bugs.webkit.org/show_bug.cgi?id=87981
Reviewed by Geoffrey Garen.
* runtime/JSONObject.cpp (Stringifier::Holder::Holder):
Initialize m_size in debug mode, as we check it later in an assert.
2012-05-30 Mark Hahnenberg <mhahnenberg@apple.com>
Heap should sweep incrementally
https://bugs.webkit.org/show_bug.cgi?id=85429
We shouldn't have to wait for the opportunistic GC timer to fire in order
to call object destructors. Instead, we should incrementally sweep some
subset of the blocks requiring sweeping periodically. We tie this sweeping
to a timer rather than to collections because we want to reclaim this memory
even if we stop allocating. This way, our memory usage scales smoothly with
actual use, regardless of whether we've recently done an opportunistic GC or not.
Reviewed by Geoffrey Garen.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.gypi:
* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* heap/Heap.cpp:
(JSC::Heap::Heap):
(JSC::Heap::collect): We no longer sweep during a full sweep. We only shrink now,
which we will switch over to being done during incremental sweeping too as soon as
all finalizers can be run lazily (and, by extension, incrementally).
(JSC::Heap::sweeper):
(JSC):
* heap/Heap.h:
(JSC):
(Heap):
* heap/IncrementalSweeper.cpp: Added.
(JSC):
(JSC::IncrementalSweeper::timerDidFire): The IncrementalSweeper works very similarly to
GCActivityCallback. It is tied to a run-loop based timer that fires periodically based
on how long the previous sweep increment took to run. The IncrementalSweeper doesn't do
anything if the platform doesn't support CoreFoundation.
(JSC::IncrementalSweeper::IncrementalSweeper):
(JSC::IncrementalSweeper::~IncrementalSweeper):
(JSC::IncrementalSweeper::create):
(JSC::IncrementalSweeper::scheduleTimer):
(JSC::IncrementalSweeper::cancelTimer):
(JSC::IncrementalSweeper::doSweep): Iterates over the snapshot of the MarkedSpace taken
during the last collection, checking to see which blocks need sweeping. If it successfully
gets to the end of the blocks that need sweeping then it cancels the timer.
(JSC::IncrementalSweeper::startSweeping): We take a snapshot of the Heap and store it in
a Vector that the incremental sweep will iterate over. We also reset our index into this Vector.
* heap/IncrementalSweeper.h: Added.
(JSC):
(IncrementalSweeper):
* heap/MarkedBlock.h:
(JSC::MarkedBlock::needsSweeping): If a block is in the Marked state it needs sweeping
to be usable and to run any destructors that need to be run.
2012-05-30 Patrick Gansterer <paroga@webkit.org>
[WINCE] Fix JSString after r115516.
https://bugs.webkit.org/show_bug.cgi?id=87892
Reviewed by Geoffrey Garen.
r115516 splitted JSString into two classes, with addition nested classes.
Add a workaround for the WinCE compiler since it can't resolve the friend class
declerations corretly and denies the access to protected members of JSString.
* runtime/JSString.h:
(JSC::JSRopeString::RopeBuilder::append):
(JSC::JSRopeString::append):
(JSRopeString):
2012-05-30 Oliver Hunt <oliver@apple.com>
Really provide error information with the inspector disabled
https://bugs.webkit.org/show_bug.cgi?id=87910
Reviewed by Filip Pizlo.
Don't bother checking for anything other than pre-existing error info.
In the absence of complete line number information you'll only get the
line a function starts on, but at least it's something.
* interpreter/Interpreter.cpp:
(JSC::Interpreter::throwException):
2012-05-30 Filip Pizlo <fpizlo@apple.com>
LLInt broken on x86-32 with JIT turned off
https://bugs.webkit.org/show_bug.cgi?id=87906
Reviewed by Geoffrey Garen.
Fixed the code to not clobber registers that contain important things, like the call frame.
* llint/LowLevelInterpreter32_64.asm:
2012-05-30 Filip Pizlo <fpizlo@apple.com>
ScriptDebugServer wants sourceIDs that are non-zero because that's what HashMaps want, so JSC should placate it
https://bugs.webkit.org/show_bug.cgi?id=87887
Reviewed by Darin Adler.
Better fix - we now never call SourceProvider::asID() if SourceProvider* is 0.
* parser/Nodes.h:
(JSC::ScopeNode::sourceID):
* parser/SourceCode.h:
(JSC::SourceCode::providerID):
(SourceCode):
* parser/SourceProvider.h:
(SourceProvider):
(JSC::SourceProvider::asID):
* runtime/Executable.h:
(JSC::ScriptExecutable::sourceID):
2012-05-30 Filip Pizlo <fpizlo@apple.com>
ScriptDebugServer wants sourceIDs that are non-zero because that's what HashMaps want, so JSC should placate it
https://bugs.webkit.org/show_bug.cgi?id=87887
Reviewed by Geoffrey Garen.
* parser/SourceProvider.h:
(JSC::SourceProvider::asID):
2012-05-30 Oliver Hunt <oliver@apple.com>
DFG does not correctly handle exceptions caught in the LLInt
https://bugs.webkit.org/show_bug.cgi?id=87885
Reviewed by Filip Pizlo.
Make the DFG use genericThrow, rather than reimplementing a small portion of it.
Also make the LLInt slow paths validate that their PC is correct.
* dfg/DFGOperations.cpp:
* llint/LLIntSlowPaths.cpp:
(LLInt):
2012-05-29 Filip Pizlo <fpizlo@apple.com>
DFG CFA should infer types and values of captured variables
https://bugs.webkit.org/show_bug.cgi?id=87813
Reviewed by Gavin Barraclough.
Slight speed-up in V8/earley-boyer (~1%).
* bytecode/CodeBlock.h:
(JSC::CodeBlock::argumentsAreCaptured):
(JSC::CodeBlock::argumentIsCaptured):
(CodeBlock):
* dfg/DFGAbstractState.cpp:
(DFG):
(JSC::DFG::AbstractState::beginBasicBlock):
(JSC::DFG::AbstractState::initialize):
(JSC::DFG::AbstractState::endBasicBlock):
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::clobberWorld):
(JSC::DFG::AbstractState::clobberStructures):
(JSC::DFG::AbstractState::mergeStateAtTail):
(JSC::DFG::AbstractState::merge):
(JSC::DFG::AbstractState::mergeToSuccessors):
* dfg/DFGAbstractState.h:
(JSC::DFG::AbstractState::variables):
(AbstractState):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-30 Patrick Gansterer <paroga@webkit.org>
Unreviewed. Build fix for !ENABLE(JIT) after r117823.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
2012-05-30 Sheriff Bot <webkit.review.bot@gmail.com>
Unreviewed, rolling out r118868.
http://trac.webkit.org/changeset/118868
https://bugs.webkit.org/show_bug.cgi?id=87828
introduced ~20 crashes on Mac and Qt bots (Requested by pizlo_
on #webkit).
* heap/Heap.cpp:
(JSC::Heap::collect):
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep):
* heap/MarkedBlock.h:
(JSC::MarkedBlock::sweepWeakSet):
(JSC):
* heap/MarkedSpace.cpp:
(JSC::SweepWeakSet::operator()):
(JSC):
(JSC::MarkedSpace::sweepWeakSets):
* heap/MarkedSpace.h:
(MarkedSpace):
2012-05-29 Geoffrey Garen <ggaren@apple.com>
Rolled back in r118646, now that
https://bugs.webkit.org/show_bug.cgi?id=87784 is fixed.
http://trac.webkit.org/changeset/118646
https://bugs.webkit.org/show_bug.cgi?id=87599
* heap/Heap.cpp:
(JSC::Heap::collect):
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep):
* heap/MarkedBlock.h:
(JSC):
* heap/MarkedSpace.cpp:
(JSC):
* heap/MarkedSpace.h:
(MarkedSpace):
2012-05-29 Filip Pizlo <fpizlo@apple.com>
DFG should keep captured variables alive until the (inline) return.
https://bugs.webkit.org/show_bug.cgi?id=87205
Reviewed by Gavin Barraclough.
Changes the way we do flushing for captured variables and arguments. Instead of flushing
each SetLocal immediately, we flush at kill points. So a SetLocal will cause a Flush of
whatever was live in the variable previously, and a return will cause a Flush of all
captured variables and all arguments.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::set):
(JSC::DFG::ByteCodeParser::setLocal):
(JSC::DFG::ByteCodeParser::getArgument):
(JSC::DFG::ByteCodeParser::setArgument):
(JSC::DFG::ByteCodeParser::findArgumentPositionForArgument):
(ByteCodeParser):
(JSC::DFG::ByteCodeParser::findArgumentPositionForLocal):
(JSC::DFG::ByteCodeParser::findArgumentPosition):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::flushDirect):
(JSC::DFG::ByteCodeParser::flushArgumentsAndCapturedVariables):
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::setLocalStoreElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::forwardSpeculationCheck):
2012-05-29 Geoffrey Garen <ggaren@apple.com>
WeakGCMap should be lazy-finalization-safe
https://bugs.webkit.org/show_bug.cgi?id=87784
Reviewed by Darin Adler.
* runtime/WeakGCMap.h:
(JSC::WeakGCMap::get): Since this is a map of raw WeakImpl pointers, and
not Weak<T>, we need to verify manually that the WeakImpl is live before
we return its payload.
2012-05-29 Mark Hahnenberg <mhahnenberg@apple.com>
CopiedSpace::doneCopying could start another collection
https://bugs.webkit.org/show_bug.cgi?id=86538
Reviewed by Geoffrey Garen.
It's possible that if we don't have anything at the head of to-space
after a collection and the BlockAllocator doesn't have any fresh blocks
to give us right now we could start another collection while still in
the middle of the first collection when we call CopiedSpace::addNewBlock().
One way to resolve this would be to have Heap::shouldCollect() check that
m_operationInProgress is NoOperation. This would prevent the path in
getFreshBlock() that starts the collection if we're already in the middle of one.
I could not come up with a test case to reproduce this crash on ToT.
* heap/Heap.h:
(JSC::Heap::shouldCollect): We shouldn't collect if we're already in the middle
of a collection, i.e. the current operation should be NoOperation.
2012-05-29 David Barr <davidbarr@chromium.org>
Introduce ENABLE_CSS_IMAGE_RESOLUTION compile flag
https://bugs.webkit.org/show_bug.cgi?id=87685
Reviewed by Eric Seidel.
Add a configuration option for CSS image-resolution support, disabling it by default.
* Configurations/FeatureDefines.xcconfig:
2012-05-28 Sheriff Bot <webkit.review.bot@gmail.com>
Unreviewed, rolling out r118646.
http://trac.webkit.org/changeset/118646
https://bugs.webkit.org/show_bug.cgi?id=87691
broke V8 raytrace benchmark (Requested by pizlo_ on #webkit).
* heap/Heap.cpp:
(JSC::Heap::collect):
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep):
* heap/MarkedBlock.h:
(JSC::MarkedBlock::sweepWeakSet):
(JSC):
* heap/MarkedSpace.cpp:
(JSC::SweepWeakSet::operator()):
(JSC):
(JSC::MarkedSpace::sweepWeakSets):
* heap/MarkedSpace.h:
(MarkedSpace):
2012-05-28 Filip Pizlo <fpizlo@apple.com>
DFG should not generate code for code that the CFA proves to be unreachable
https://bugs.webkit.org/show_bug.cgi?id=87682
Reviewed by Sam Weinig.
This also fixes a small performance bug where CFA was not marking blocks
as having constants (and hence not triggering constant folding) if the only
constants were on GetLocals.
And fixing that bug revealed another bug: constant folding was assuming that
a GetLocal must be the first access to a local in a basic block. This isn't
true. The first access may be a Flush. This patch fixes that issue using the
safest approach possible, since we don't need to be clever for something that
only happens in one of our benchmarks.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGConstantFoldingPhase.cpp:
(JSC::DFG::ConstantFoldingPhase::run):
* dfg/DFGJITCompiler.h:
(JSC::DFG::JITCompiler::noticeOSREntry):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-28 Carlos Garcia Campos <cgarcia@igalia.com>
Unreviewed. Fix make distcheck.
* GNUmakefile.list.am: Add missing header file.
2012-05-27 Geoffrey Garen <ggaren@apple.com>
Weak pointer finalization should be lazy
https://bugs.webkit.org/show_bug.cgi?id=87599
Reviewed by Darin Adler.
* heap/Heap.cpp:
(JSC::Heap::collect): Don't force immediate finalization -- it will
happen lazily.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::sweep): Sweep a block's weak set when sweeping the
block. The weak set may not have been swept yet, and this is our last
chance to run weak finalizers before we recycle the memory they reference.
* heap/MarkedBlock.h:
* heap/MarkedSpace.cpp:
(JSC::MarkedBlock::sweepWeakSets):
* heap/MarkedSpace.h:
(JSC::MarkedSpace::sweepWeakSets): Nixed sweepWeakSets because it's unused
now.
2012-05-26 Geoffrey Garen <ggaren@apple.com>
WebKit should be lazy-finalization-safe (esp. the DOM) v2
https://bugs.webkit.org/show_bug.cgi?id=87581
Reviewed by Oliver Hunt.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::callDestructor):
* heap/WeakBlock.h:
* heap/WeakSetInlines.h:
(JSC::WeakBlock::finalize): Since we don't guarantee destruction order,
it's not valid to access GC pointers like the Structure pointer during
finalization. We NULL out the structure pointer in debug builds to try
to make this programming mistake more obvious.
* API/JSCallbackConstructor.cpp:
(JSC::JSCallbackConstructor::destroy):
* API/JSCallbackObject.cpp:
(JSC::::destroy):
(JSC::JSCallbackObjectData::finalize):
* runtime/Arguments.cpp:
(JSC::Arguments::destroy):
* runtime/DateInstance.cpp:
(JSC::DateInstance::destroy):
* runtime/Error.cpp:
(JSC::StrictModeTypeErrorFunction::destroy):
* runtime/Executable.cpp:
(JSC::ExecutableBase::destroy):
(JSC::NativeExecutable::destroy):
(JSC::ScriptExecutable::destroy):
(JSC::EvalExecutable::destroy):
(JSC::ProgramExecutable::destroy):
(JSC::FunctionExecutable::destroy):
* runtime/JSGlobalObject.cpp:
(JSC::JSGlobalObject::destroy):
* runtime/JSPropertyNameIterator.cpp:
(JSC::JSPropertyNameIterator::destroy):
* runtime/JSStaticScopeObject.cpp:
(JSC::JSStaticScopeObject::destroy):
* runtime/JSString.cpp:
(JSC::JSString::destroy):
* runtime/JSVariableObject.cpp:
(JSC::JSVariableObject::destroy):
* runtime/NameInstance.cpp:
(JSC::NameInstance::destroy):
* runtime/RegExp.cpp:
(JSC::RegExp::destroy):
* runtime/RegExpConstructor.cpp:
(JSC::RegExpConstructor::destroy):
* runtime/Structure.cpp:
(JSC::Structure::destroy):
* runtime/StructureChain.cpp:
(JSC::StructureChain::destroy): Use static_cast instead of jsCast because
jsCast does Structure-based validation, and our Structure is not guaranteed
to be alive when we get finalized.
2012-05-22 Filip Pizlo <fpizlo@apple.com>
DFG CSE should eliminate redundant WeakJSConstants
https://bugs.webkit.org/show_bug.cgi?id=87179
Reviewed by Gavin Barraclough.
Merged r118141 from dfgopt.
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::weakConstantCSE):
(CSEPhase):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGNode.h:
(JSC::DFG::Node::weakConstant):
2012-05-22 Filip Pizlo <fpizlo@apple.com>
DFG CSE should do redundant store elimination
https://bugs.webkit.org/show_bug.cgi?id=87161
Reviewed by Oliver Hunt.
Merge r118138 from dfgopt.
This patch adds redundant store elimination. For example, consider this
code:
o.x = 42;
o.x = 84;
If o.x is speculated to be a well-behaved field, the first assignment is
unnecessary, since the second just overwrites it. We would like to
eliminate the first assignment in these cases. The need for this
optimization arises mostly from stores that our runtime requires. For
example:
o = {f:1, g:2, h:3};
This will have four assignments to the structure for the newly created
object - one assignment for the empty structure, one for {f}, one for
{f, g}, and one for {f, g, h}. We would like to only have the last of
those assigments in this case.
Intriguingly, doing so for captured variables breaks the way arguments
simplification used to work. Consider that prior to either arguments
simplification or store elimination we will have IR that looks like:
a: SetLocal(r0, Empty)
b: SetLocal(r1, Empty)
c: GetLocal(r0)
d: CreateArguments(@c)
e: SetLocal(r0, @d)
f: SetLocal(r1, @d)
Then redundant store elimination will eliminate the stores that
initialize the arguments registers to Empty, but then arguments
simplification eliminates the stores that initialize the arguments to
the newly created arguments - and at this point we no longer have any
stores to the arguments register, leading to hilarious crashes. This
patch therefore changes arguments simplification to replace
CreateArguments with JSConstant(Empty) rather than eliminating the
SetLocals. But this revealed bugs where arguments simplification was
being overzealous, so I fixed those bugs.
This is a minor speed-up on V8/early and a handful of other tests.
* bytecode/CodeBlock.h:
(JSC::CodeBlock::uncheckedActivationRegister):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUse):
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUses):
(JSC::DFG::ArgumentsSimplificationPhase::observeProperArgumentsUse):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::globalVarStoreElimination):
(CSEPhase):
(JSC::DFG::CSEPhase::putStructureStoreElimination):
(JSC::DFG::CSEPhase::putByOffsetStoreElimination):
(JSC::DFG::CSEPhase::setLocalStoreElimination):
(JSC::DFG::CSEPhase::setReplacement):
(JSC::DFG::CSEPhase::eliminate):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::uncheckedActivationRegisterFor):
(Graph):
* dfg/DFGNode.h:
(JSC::DFG::Node::isPhantomArguments):
(Node):
(JSC::DFG::Node::hasConstant):
(JSC::DFG::Node::valueOfJSConstant):
(JSC::DFG::Node::hasStructureTransitionData):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::computeValueRecoveryFor):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-21 Filip Pizlo <fpizlo@apple.com>
DFG ConvertThis should just be a CheckStructure if the structure is known
https://bugs.webkit.org/show_bug.cgi?id=87057
Reviewed by Gavin Barraclough.
Merged r118021 from dfgopt.
This gives ValueProfile the ability to track singleton values - i.e. profiling
sites that always see the same value.
That is then used to profile the structure in op_convert_this.
This is then used to optimize op_convert_this into a CheckStructure if the
structure is always the same.
That then results in better CSE in inlined code that uses 'this', since
previously we couldn't CSE accesses on 'this' from different inline call frames.
Also fixed a bug where we were unnecessarily flushing 'this'.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
(JSC::CodeBlock::stronglyVisitStrongReferences):
* bytecode/LazyOperandValueProfile.cpp:
(JSC::CompressedLazyOperandValueProfileHolder::computeUpdatedPredictions):
* bytecode/LazyOperandValueProfile.h:
(CompressedLazyOperandValueProfileHolder):
* bytecode/Opcode.h:
(JSC):
(JSC::padOpcodeName):
* bytecode/ValueProfile.h:
(JSC::ValueProfileBase::ValueProfileBase):
(JSC::ValueProfileBase::dump):
(JSC::ValueProfileBase::computeUpdatedPrediction):
(ValueProfileBase):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::BytecodeGenerator):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setArgument):
(JSC::DFG::ByteCodeParser::parseBlock):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_convert_this):
(JSC::JIT::emitSlow_op_convert_this):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_convert_this):
(JSC::JIT::emitSlow_op_convert_this):
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
* runtime/JSValue.h:
(JSValue):
* runtime/Structure.h:
(JSC::JSValue::structureOrUndefined):
(JSC):
2012-05-24 Tim Horton <timothy_horton@apple.com>
Add feature defines for web-facing parts of CSS Regions and Exclusions
https://bugs.webkit.org/show_bug.cgi?id=87442
<rdar://problem/10887709>
Reviewed by Dan Bernstein.
* Configurations/FeatureDefines.xcconfig:
2012-05-24 Geoffrey Garen <ggaren@apple.com>
WebKit should be lazy-finalization-safe (esp. the DOM)
https://bugs.webkit.org/show_bug.cgi?id=87456
Reviewed by Filip Pizlo.
Lazy finalization adds one twist to weak pointer use:
A HashMap of weak pointers may contain logically null entries.
(Weak pointers behave as-if null once their payloads die.)
Insertion must not assume that a pre-existing entry is
necessarily valid, and iteration must not assume that all
entries can be dereferenced.
(Previously, I thought that it also added a second twist:
A demand-allocated weak pointer may replace a dead payload
before the payload's finalizer runs. In that case, when the
payload's finalizer runs, the payload has already been
overwritten, and the finalizer should not clear the payload,
which now points to something new.
But that's not the case here, since we cancel the old payload's
finalizer when we over-write it. I've added ASSERTs to verify this
assumption, in case it ever changes.)
* API/JSClassRef.cpp:
(OpaqueJSClass::prototype): No need to specify null; that's the default.
* API/JSWeakObjectMapRefPrivate.cpp: Use remove, since take() is gone.
* heap/PassWeak.h:
(WeakImplAccessor::was): This is no longer a debug-only function, since
it's required to reason about lazily finalized pointers.
* heap/Weak.h:
(JSC::weakAdd):
(JSC::weakRemove):
(JSC::weakClear): Added these helper functions for the common idioms of
what clients want to do in their weak pointer finalizers.
* jit/JITStubs.cpp:
(JSC::JITThunks::hostFunctionStub): Use the new idioms. Otherwise, we
would return NULL for a "zombie" executable weak pointer that was waiting
for finalization (item (2)), and finalizing a dead executable weak pointer
would potentially destroy a new, live one (item (1)).
* runtime/RegExpCache.cpp:
(JSC::RegExpCache::lookupOrCreate):
(JSC::RegExpCache::finalize): Ditto.
(JSC::RegExpCache::invalidateCode): Check for null while iterating. (See
item (2).)
* runtime/Structure.cpp:
(JSC::StructureTransitionTable::contains):
(JSC::StructureTransitionTable::add): Use get and set instead of add and
contains, since add and contains are not compatible with lazy finalization.
* runtime/WeakGCMap.h:
(WeakGCMap):
(JSC::WeakGCMap::clear):
(JSC::WeakGCMap::remove): Removed a bunch of code that was incompatible with
lazy finalization because I didn't feel like making it compatible, and I had
no way to test it.
2012-05-24 Filip Pizlo <fpizlo@apple.com>
REGRESSION (r118013-r118031): Loops/Reloads under www.yahoo.com, quits after three tries with error
https://bugs.webkit.org/show_bug.cgi?id=87327
Reviewed by Geoffrey Garen.
If you use AbstractValue::filter(StructureSet) to test subset relationships between TOP and a
set containing >=2 elements, you're going to have a bad time.
That's because AbstractValue considers a set with >=2 elements to be equivalent to TOP, in order
to save space and speed up convergence. So filtering has no effect in this case, which made
the code think that the abstract value was proving that the structure check was unnecessary.
The correct thing to do is to use isSubsetOf() on the StructureAbstractValue, which does the
right thingies for TOP and >=2 elements.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-24 Filip Pizlo <fpizlo@apple.com>
new test fast/js/dfg-arguments-mixed-alias.html fails on JSVALUE32_64
https://bugs.webkit.org/show_bug.cgi?id=87378
Reviewed by Gavin Barraclough.
- Captured variable tracking forgot did not consistently handle arguments, leading to OSR
badness.
- Nodes capable of exiting were tracked in a non-monotonic way, leading to compiler errors.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::CSEPhase):
(CSEPhase):
(JSC::DFG::performCSE):
* dfg/DFGCSEPhase.h:
(DFG):
* dfg/DFGCommon.h:
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::resetExitStates):
(DFG):
* dfg/DFGGraph.h:
(Graph):
* dfg/DFGPhase.h:
(DFG):
(JSC::DFG::runPhase):
2012-05-24 Geoffrey Garen <ggaren@apple.com>
Made WeakSet per-block instead of per-heap
https://bugs.webkit.org/show_bug.cgi?id=87401
Reviewed by Oliver Hunt.
This allows us fast access to the set of all weak pointers for a block,
which is a step toward lazy finalization.
No performance change.
* heap/Heap.cpp:
(JSC::Heap::Heap):
(JSC::Heap::lastChanceToFinalize): Removed the per-heap weak set, since
it's per-block now.
(JSC::Heap::markRoots): Delegate weak set visiting to the marked space,
since it knows how to iterate all blocks.
(JSC::Heap::collect): Moved the reaping outside of markRoots, since it
doesn't mark anything.
Make sure to reset allocators after shrinking, since shrinking may
deallocate the current allocator.
* heap/Heap.h:
(Heap): No more per-heap weak set, since it's per-block now.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::MarkedBlock):
* heap/MarkedBlock.h:
(MarkedBlock):
(JSC::MarkedBlock::lastChanceToFinalize): Migrated finalization logic
here from the heap, so the heap doesn't need to know about our internal
data structures like our weak set.
(JSC::MarkedBlock::heap):
(JSC::MarkedBlock::weakSet):
(JSC::MarkedBlock::shrink):
(JSC::MarkedBlock::resetAllocator):
(JSC::MarkedBlock::visitWeakSet):
(JSC::MarkedBlock::reapWeakSet):
(JSC::MarkedBlock::sweepWeakSet):
* heap/MarkedSpace.cpp:
(JSC::VisitWeakSet::VisitWeakSet):
(JSC::VisitWeakSet::operator()):
(VisitWeakSet):
(JSC):
(JSC::ReapWeakSet::operator()):
(JSC::SweepWeakSet::operator()):
(JSC::LastChanceToFinalize::operator()):
(JSC::MarkedSpace::lastChanceToFinalize):
(JSC::ResetAllocator::operator()):
(JSC::MarkedSpace::resetAllocators):
(JSC::MarkedSpace::visitWeakSets):
(JSC::MarkedSpace::reapWeakSets):
(JSC::MarkedSpace::sweepWeakSets):
(JSC::Shrink::operator()):
(JSC::MarkedSpace::shrink):
* heap/MarkedSpace.h:
(MarkedSpace): Make sure to account for our weak sets when sweeping,
shrinking, etc.
* heap/WeakSet.cpp:
(JSC):
* heap/WeakSet.h:
(WeakSet):
(JSC::WeakSet::heap):
(JSC):
(JSC::WeakSet::lastChanceToFinalize):
(JSC::WeakSet::visit):
(JSC::WeakSet::reap):
(JSC::WeakSet::shrink):
(JSC::WeakSet::resetAllocator): Inlined some things since they're called
once per block now instead of once per heap.
* heap/WeakSetInlines.h:
(JSC::WeakSet::allocate): Use the per-block weak set since there is no
per-heap weak set anymore.
2012-05-24 Gavin Barraclough <barraclough@apple.com>
Fix arm build
Rubber stamped by Geoff Garen
* dfg/DFGGPRInfo.h:
(GPRInfo):
2012-05-24 Gavin Barraclough <barraclough@apple.com>
Move cacheFlush from ExecutableAllocator to Assembler classes
https://bugs.webkit.org/show_bug.cgi?id=87420
Reviewed by Oliver Hunt.
Makes more sense there, & remove a pile of #ifdefs.
* assembler/ARMAssembler.cpp:
(JSC):
(JSC::ARMAssembler::cacheFlush):
* assembler/ARMAssembler.h:
(ARMAssembler):
(JSC::ARMAssembler::cacheFlush):
* assembler/ARMv7Assembler.h:
(JSC::ARMv7Assembler::relinkJump):
(JSC::ARMv7Assembler::cacheFlush):
(ARMv7Assembler):
(JSC::ARMv7Assembler::setInt32):
(JSC::ARMv7Assembler::setUInt7ForLoad):
* assembler/AbstractMacroAssembler.h:
(JSC::AbstractMacroAssembler::cacheFlush):
* assembler/LinkBuffer.h:
(JSC::LinkBuffer::performFinalization):
* assembler/MIPSAssembler.h:
(JSC::MIPSAssembler::relinkJump):
(JSC::MIPSAssembler::relinkCall):
(JSC::MIPSAssembler::repatchInt32):
(JSC::MIPSAssembler::cacheFlush):
(MIPSAssembler):
* assembler/SH4Assembler.h:
(JSC::SH4Assembler::repatchCompact):
(JSC::SH4Assembler::cacheFlush):
(SH4Assembler):
* assembler/X86Assembler.h:
(X86Assembler):
(JSC::X86Assembler::cacheFlush):
* jit/ExecutableAllocator.cpp:
(JSC):
* jit/ExecutableAllocator.h:
(ExecutableAllocator):
2012-05-24 John Mellor <johnme@chromium.org>
Font Boosting: Add compile flag and runtime setting
https://bugs.webkit.org/show_bug.cgi?id=87394
Reviewed by Adam Barth.
Add ENABLE_FONT_BOOSTING.
* Configurations/FeatureDefines.xcconfig:
2012-05-24 Allan Sandfeld Jensen <allan.jensen@nokia.com>
cti_vm_throw gets kicked out by gcc 4.6 -flto
https://bugs.webkit.org/show_bug.cgi?id=56088
Reviewed by Darin Adler.
Add REFERENCED_FROM_ASM to functions only referenced from assembler.
* dfg/DFGOperations.cpp:
* jit/HostCallReturnValue.h:
* jit/JITStubs.h:
* jit/ThunkGenerators.cpp:
2012-05-24 Filip Pizlo <fpizlo@apple.com>
Incorrect merge of r117542 from dfg opt branch in r118323 is leading to fast/js/dfg-arguments-osr-exit.html failing
https://bugs.webkit.org/show_bug.cgi?id=87350
Reviewed by Maciej Stachowiak.
The dfgopt branch introduced the notion of a local variable being killed because it was aliased
to the Arguments object as in cases like:
var a = arguments;
return a.length;
This required changes to OSR exit handling - if the variable is dead but aliased to arguments, then
OSR exit should reify the arguments. But meanwhile, in tip of tree we introduced special handling for
dead variables on OSR exit. When the two were merged in r118323, the structure of the if/else branches
ended up being such that we would treat dead arguments variables as totally dead as opposed to treating
them as variables that need arguments reification.
This fixes the structure of the relevant if/else block so that variables that are dead-but-arguments
end up being treated as reified arguments objects, while variables that are dead but not aliased to
arguments are treated as tip of tree would have treated them (initialize to Undefined).
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-24 Csaba Osztrogonác <ossy@webkit.org>
Unreviewed 32 bit buildfix after r118325.
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::fillSpeculateIntInternal): Use ASSERT_UNUSED instead ASSERT.
2012-05-23 Filip Pizlo <fpizlo@apple.com>
DFG operationTearOffActivation should return after handling the null activation case
https://bugs.webkit.org/show_bug.cgi?id=87348
<rdar://problem/11522295>
Reviewed by Oliver Hunt.
* dfg/DFGOperations.cpp:
2012-05-23 Filip Pizlo <fpizlo@apple.com>
Unreviewed, merge the arguments fix in r118138 to get bots green.
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUse):
2012-05-20 Filip Pizlo <fpizlo@apple.com>
DFG CFA should record if a node can OSR exit
https://bugs.webkit.org/show_bug.cgi?id=86905
Reviewed by Oliver Hunt.
Merged r117931 from dfgopt.
Adds a NodeFlag that denotes nodes that are known to not have OSR exits.
This ought to aid any backwards analyses that need to know when a
backward flow merge might happen due to a side exit.
Also added assertions into speculationCheck() that ensure that we did not
mark a node as non-exiting and then promptly compile in an exit. This
helped catch some minor bugs where we were doing unnecessary speculation
checks.
This is a perf-neutral change. The speculation checks that this removes
were not on hot paths of major benchmarks.
* bytecode/PredictedType.h:
(JSC):
(JSC::isAnyPrediction):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGAbstractState.h:
(JSC::DFG::AbstractState::speculateInt32Unary):
(AbstractState):
(JSC::DFG::AbstractState::speculateNumberUnary):
(JSC::DFG::AbstractState::speculateBooleanUnary):
(JSC::DFG::AbstractState::speculateInt32Binary):
(JSC::DFG::AbstractState::speculateNumberBinary):
* dfg/DFGNode.h:
(JSC::DFG::Node::mergeFlags):
(JSC::DFG::Node::filterFlags):
(Node):
(JSC::DFG::Node::setCanExit):
(JSC::DFG::Node::canExit):
* dfg/DFGNodeFlags.cpp:
(JSC::DFG::nodeFlagsAsString):
* dfg/DFGNodeFlags.h:
(DFG):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::SpeculativeJIT):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::compileValueToInt32):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::speculationCheck):
(JSC::DFG::SpeculativeJIT::forwardSpeculationCheck):
(JSC::DFG::SpeculativeJIT::terminateSpeculativeExecution):
(SpeculativeJIT):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::fillSpeculateIntInternal):
(JSC::DFG::SpeculativeJIT::fillSpeculateDouble):
(JSC::DFG::SpeculativeJIT::fillSpeculateCell):
(JSC::DFG::SpeculativeJIT::fillSpeculateBoolean):
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::fillSpeculateIntInternal):
(JSC::DFG::SpeculativeJIT::fillSpeculateDouble):
(JSC::DFG::SpeculativeJIT::fillSpeculateCell):
(JSC::DFG::SpeculativeJIT::fillSpeculateBoolean):
(JSC::DFG::SpeculativeJIT::compile):
2012-05-20 Filip Pizlo <fpizlo@apple.com>
DFG should not do unnecessary indirections when storing to objects
https://bugs.webkit.org/show_bug.cgi?id=86959
Reviewed by Oliver Hunt.
Merged r117819 from dfgopt.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::getByOffsetLoadElimination):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-17 Filip Pizlo <fpizlo@apple.com>
DFG should optimize aliased uses of the Arguments object of the current call frame
https://bugs.webkit.org/show_bug.cgi?id=86552
Reviewed by Geoff Garen.
Merged r117542 and r117543 from dfgopt.
Performs must-alias and escape analysis on uses of CreateArguments, and if
a variable is must-aliased to CreateArguments and does not escape, then we
turn all uses of that variable into direct arguments accesses.
36% speed-up on V8/earley leading to a 2.3% speed-up overall in V8.
* bytecode/CodeBlock.h:
(JSC::CodeBlock::uncheckedArgumentsRegister):
* bytecode/ValueRecovery.h:
(JSC::ValueRecovery::argumentsThatWereNotCreated):
(ValueRecovery):
(JSC::ValueRecovery::dump):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGAdjacencyList.h:
(AdjacencyList):
(JSC::DFG::AdjacencyList::removeEdgeFromBag):
* dfg/DFGArgumentsSimplificationPhase.cpp:
(JSC::DFG::ArgumentsSimplificationPhase::run):
(ArgumentsSimplificationPhase):
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUse):
(JSC::DFG::ArgumentsSimplificationPhase::observeBadArgumentsUses):
(JSC::DFG::ArgumentsSimplificationPhase::observeProperArgumentsUse):
(JSC::DFG::ArgumentsSimplificationPhase::isOKToOptimize):
(JSC::DFG::ArgumentsSimplificationPhase::removeArgumentsReferencingPhantomChild):
* dfg/DFGAssemblyHelpers.h:
(JSC::DFG::AssemblyHelpers::argumentsRegisterFor):
(AssemblyHelpers):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::removePotentiallyDeadPhiReference):
* dfg/DFGGPRInfo.h:
(GPRInfo):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::collectGarbage):
(DFG):
* dfg/DFGGraph.h:
(Graph):
(JSC::DFG::Graph::executableFor):
(JSC::DFG::Graph::argumentsRegisterFor):
(JSC::DFG::Graph::uncheckedArgumentsRegisterFor):
(JSC::DFG::Graph::clobbersWorld):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasHeapPrediction):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOSRExitCompiler.cpp:
* dfg/DFGOSRExitCompiler.h:
(JSC::DFG::OSRExitCompiler::OSRExitCompiler):
(OSRExitCompiler):
* dfg/DFGOSRExitCompiler32_64.cpp:
(JSC::DFG::OSRExitCompiler::compileExit):
* dfg/DFGOSRExitCompiler64.cpp:
(JSC::DFG::OSRExitCompiler::compileExit):
* dfg/DFGOperations.cpp:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::ValueSource::dump):
(JSC::DFG::SpeculativeJIT::compile):
(JSC::DFG::SpeculativeJIT::computeValueRecoveryFor):
* dfg/DFGSpeculativeJIT.h:
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGVariableAccessData.h:
(JSC::DFG::VariableAccessData::VariableAccessData):
(JSC::DFG::VariableAccessData::mergeIsArgumentsAlias):
(VariableAccessData):
(JSC::DFG::VariableAccessData::isArgumentsAlias):
* jit/JITOpcodes.cpp:
(JSC::JIT::emitSlow_op_get_argument_by_val):
2012-05-23 Filip Pizlo <fpizlo@apple.com>
DFGCapabilities should not try to get an arguments register from code blocks that don't have one
https://bugs.webkit.org/show_bug.cgi?id=87332
Reviewed by Andy Estes.
* dfg/DFGCapabilities.h:
(JSC::DFG::canInlineOpcode):
2012-05-23 Filip Pizlo <fpizlo@apple.com>
DFG should have sparse conditional constant propagation
https://bugs.webkit.org/show_bug.cgi?id=86580
Reviewed by Oliver Hunt.
Merged r117370 from dfgopt.
This enhances CFA so that if it suspects at any point during the fixpoint that a
branch will only go one way, then it only propagates in that one way.
This vastly increases the opportunities for CFG simplification. For example, it
enables us to evaporate this loop:
for (var i = 0; i < 1; ++i) doThings(i);
As a result, it uncovered loads of bugs in the CFG simplifier. In particular:
- Phi fixup was assuming that all Phis worth fixing up are shouldGenerate().
That's not true; we also fixup Phis that are dead.
- GetLocal fixup was assuming that it's only necessary to rewire links to a
GetLocal, and that the GetLocal's own links don't need to be rewired. Untrue,
because the GetLocal may not be rewirable (first block has no GetLocal for r42
but second block does have a GetLocal), in which case it will refer to a Phi
in the second block. We need it to refer to a Phi from the first block to
ensure that subsequent transformations work.
- Tail operand fixup was ignoring the fact that Phis in successors may contain
references to the children of our tail variables. Hence, successor Phi child
substitution needs to use the original second block variable table as its
prior, rather than trying to reconstruct the prior later (since by that point
the children of the second block's tail variables will have been fixed up, so
we will not know what the prior would have been).
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::beginBasicBlock):
(JSC::DFG::AbstractState::endBasicBlock):
(JSC::DFG::AbstractState::reset):
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::mergeToSuccessors):
* dfg/DFGAbstractState.h:
(JSC::DFG::AbstractState::branchDirectionToString):
(AbstractState):
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::run):
(JSC::DFG::CFGSimplificationPhase::removePotentiallyDeadPhiReference):
(JSC::DFG::CFGSimplificationPhase::OperandSubstitution::OperandSubstitution):
(OperandSubstitution):
(JSC::DFG::CFGSimplificationPhase::skipGetLocal):
(JSC::DFG::CFGSimplificationPhase::recordPossibleIncomingReference):
(CFGSimplificationPhase):
(JSC::DFG::CFGSimplificationPhase::fixTailOperand):
(JSC::DFG::CFGSimplificationPhase::mergeBlocks):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::changeEdge):
2012-05-23 Ojan Vafai <ojan@chromium.org>
add back the ability to disable flexbox
https://bugs.webkit.org/show_bug.cgi?id=87147
Reviewed by Tony Chang.
* Configurations/FeatureDefines.xcconfig:
2012-05-23 Filip Pizlo <fpizlo@apple.com>
Unreviewed, fix Windows build.
* bytecode/CodeBlock.h:
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
(JSC::DFG::canCompileOpcodes):
* dfg/DFGCommon.h:
(DFG):
2012-05-23 Filip Pizlo <fpizlo@apple.com>
DFG should optimize inlined uses of arguments.length and arguments[i]
https://bugs.webkit.org/show_bug.cgi?id=86327
Reviewed by Gavin Barraclough.
Merged r117017 from dfgopt.
Turns inlined uses of arguments.length into a constant.
Turns inlined uses of arguments[constant] into a direct reference to the
argument.
Big win on micro-benchmarks. Not yet a win on V8 because the hot uses of
arguments.length and arguments[i] are aliased. I'll leave the aliasing
optimizations to a later patch.
* CMakeLists.txt:
* GNUmakefile.list.am:
* JavaScriptCore.xcodeproj/project.pbxproj:
* Target.pri:
* bytecode/DFGExitProfile.h:
(FrequentExitSite):
(JSC::DFG::FrequentExitSite::FrequentExitSite):
(JSC::DFG::QueryableExitProfile::hasExitSite):
(QueryableExitProfile):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGArgumentsSimplificationPhase.cpp: Added.
(DFG):
(ArgumentsSimplificationPhase):
(JSC::DFG::ArgumentsSimplificationPhase::ArgumentsSimplificationPhase):
(JSC::DFG::ArgumentsSimplificationPhase::run):
(JSC::DFG::performArgumentsSimplification):
* dfg/DFGArgumentsSimplificationPhase.h: Added.
(DFG):
* dfg/DFGAssemblyHelpers.cpp:
(JSC::DFG::AssemblyHelpers::executableFor):
(DFG):
* dfg/DFGAssemblyHelpers.h:
(AssemblyHelpers):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::getLocalLoadElimination):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::executableFor):
(Graph):
(JSC::DFG::Graph::clobbersWorld):
* dfg/DFGNode.h:
(JSC::DFG::Node::convertToConstant):
(JSC::DFG::Node::convertToGetLocalUnlinked):
(Node):
(JSC::DFG::Node::unlinkedLocal):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOSRExit.cpp:
(JSC::DFG::OSRExit::considerAddingAsFrequentExitSiteSlow):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-13 Filip Pizlo <fpizlo@apple.com>
DFG should be able to optimize foo.apply(bar, arguments)
https://bugs.webkit.org/show_bug.cgi?id=86306
Reviewed by Gavin Barraclough.
Merge r116912 from dfgopt.
Enables compilation of op_jneq_ptr and some forms of op_call_varargs.
Also includes a bunch of bug fixes that were made necessary by the increased
pressure on the CFG simplifier.
This is a 1-2% win on V8.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::printCallOp):
(JSC::CodeBlock::CodeBlock):
(JSC::ProgramCodeBlock::canCompileWithDFGInternal):
(JSC::EvalCodeBlock::canCompileWithDFGInternal):
(JSC::FunctionCodeBlock::canCompileWithDFGInternal):
* bytecode/CodeBlock.h:
(CodeBlock):
(JSC::CodeBlock::canCompileWithDFG):
(JSC::CodeBlock::canCompileWithDFGState):
(ProgramCodeBlock):
(EvalCodeBlock):
(FunctionCodeBlock):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::processPhiStack):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFGSimplificationPhase.cpp:
(JSC::DFG::CFGSimplificationPhase::run):
(JSC::DFG::CFGSimplificationPhase::fixPossibleGetLocal):
(JSC::DFG::CFGSimplificationPhase::fixTailOperand):
(JSC::DFG::CFGSimplificationPhase::mergeBlocks):
* dfg/DFGCSEPhase.cpp:
(JSC::DFG::CSEPhase::getLocalLoadElimination):
(CSEPhase):
(JSC::DFG::CSEPhase::setReplacement):
(JSC::DFG::CSEPhase::performNodeCSE):
* dfg/DFGCapabilities.cpp:
(JSC::DFG::debugFail):
(DFG):
(JSC::DFG::canHandleOpcodes):
(JSC::DFG::canCompileOpcodes):
(JSC::DFG::canInlineOpcodes):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
(JSC::DFG::canInlineOpcode):
(DFG):
(JSC::DFG::canCompileOpcodes):
(JSC::DFG::canCompileEval):
(JSC::DFG::canCompileProgram):
(JSC::DFG::canCompileFunctionForCall):
(JSC::DFG::canCompileFunctionForConstruct):
* dfg/DFGCommon.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::dump):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::emitCall):
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGValidate.cpp:
(Validate):
(JSC::DFG::Validate::validate):
(JSC::DFG::Validate::checkOperand):
(JSC::DFG::Validate::reportValidationContext):
* jit/JIT.cpp:
(JSC::JIT::emitOptimizationCheck):
(JSC::JIT::privateCompileSlowCases):
(JSC::JIT::privateCompile):
* jit/JIT.h:
* jit/JITArithmetic.cpp:
(JSC::JIT::compileBinaryArithOp):
* jit/JITPropertyAccess.cpp:
(JSC::JIT::privateCompilePutByIdTransition):
* jit/JITPropertyAccess32_64.cpp:
(JSC::JIT::privateCompilePutByIdTransition):
* tools/CodeProfile.cpp:
(JSC::CodeProfile::sample):
2012-05-23 Geoffrey Garen <ggaren@apple.com>
Refactored WeakBlock to use malloc, clarify behavior
https://bugs.webkit.org/show_bug.cgi?id=87318
Reviewed by Filip Pizlo.
We want to use malloc so we can make these smaller than 4KB,
since an individual MarkedBlock will usually have fewer than
4KB worth of weak pointers.
* heap/Heap.cpp:
(JSC::Heap::markRoots): Renamed visitLiveWeakImpls to visit, since
we no longer need to distinguish from "visitDeadWeakImpls".
Renamed "visitDeadWeakImpls" to "reap" because we're not actually
doing any visiting -- we're just tagging things as dead.
* heap/WeakBlock.cpp:
(JSC::WeakBlock::create):
(JSC::WeakBlock::destroy):
(JSC::WeakBlock::WeakBlock): Malloc!
(JSC::WeakBlock::visit):
(JSC::WeakBlock::reap): Renamed as above.
* heap/WeakBlock.h:
(WeakBlock): Reduced to 3KB, as explained above.
* heap/WeakSet.cpp:
(JSC::WeakSet::visit):
(JSC::WeakSet::reap):
* heap/WeakSet.h:
(WeakSet): Updated for renames, and to match WebKit style.
2012-05-23 Filip Pizlo <fpizlo@apple.com>
Use after free in JSC::DFG::ByteCodeParser::processPhiStack
https://bugs.webkit.org/show_bug.cgi?id=87312
<rdar://problem/11518848>
Reviewed by Oliver Hunt.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::processPhiStack):
(JSC::DFG::ByteCodeParser::parse):
2012-05-23 Filip Pizlo <fpizlo@apple.com>
It should be possible to make C function calls from DFG code on ARM in debug mode
https://bugs.webkit.org/show_bug.cgi?id=87313
Reviewed by Gavin Barraclough.
* dfg/DFGSpeculativeJIT.h:
(SpeculativeJIT):
2012-05-11 Filip Pizlo <fpizlo@apple.com>
DFG should be able to inline functions that use arguments reflectively
https://bugs.webkit.org/show_bug.cgi?id=86132
Reviewed by Oliver Hunt.
Merged r116838 from dfgopt.
This turns on inlining of functions that use arguments reflectively, but it
does not do any of the obvious optimizations that this exposes. I'll save that
for another patch - the important thing for now is that this contains all of
the plumbing necessary to make this kind of inlining sound even in bizarro
cases like an inline callee escaping the arguments object to parts of the
inline caller where the arguments are otherwise dead. Or even more fun cases
like where you've inlined to an inline stack that is three-deep, and the
function on top of the inline stack reflectively accesses the arguments of a
function that is in the middle of the inline stack. Any subsequent
optimizations that we do for the obvious cases of arguments usage in inline
functions will have to take care not to break the baseline functionality that
this patch plumbs together.
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::printCallOp):
(JSC::CodeBlock::dump):
* bytecode/CodeBlock.h:
* dfg/DFGAssemblyHelpers.h:
(JSC::DFG::AssemblyHelpers::argumentsRegisterFor):
(AssemblyHelpers):
* dfg/DFGByteCodeParser.cpp:
(InlineStackEntry):
(JSC::DFG::ByteCodeParser::handleCall):
(JSC::DFG::ByteCodeParser::handleInlining):
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCCallHelpers.h:
(JSC::DFG::CCallHelpers::setupArgumentsWithExecState):
(CCallHelpers):
* dfg/DFGCapabilities.h:
(JSC::DFG::canInlineOpcode):
* dfg/DFGDriver.cpp:
(JSC::DFG::compile):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* interpreter/CallFrame.cpp:
(JSC):
(JSC::CallFrame::someCodeBlockForPossiblyInlinedCode):
* interpreter/CallFrame.h:
(ExecState):
(JSC::ExecState::someCodeBlockForPossiblyInlinedCode):
* interpreter/Interpreter.cpp:
(JSC::Interpreter::retrieveArgumentsFromVMCode):
* runtime/Arguments.cpp:
(JSC::Arguments::tearOff):
(JSC):
(JSC::Arguments::tearOffForInlineCallFrame):
* runtime/Arguments.h:
(Arguments):
(JSC::Arguments::create):
(JSC::Arguments::finishCreation):
(JSC):
2012-05-23 Filip Pizlo <fpizlo@apple.com>
Every OSR exit on ARM results in a crash
https://bugs.webkit.org/show_bug.cgi?id=87307
Reviewed by Geoffrey Garen.
* dfg/DFGThunks.cpp:
(JSC::DFG::osrExitGenerationThunkGenerator):
2012-05-23 Geoffrey Garen <ggaren@apple.com>
Refactored heap tear-down to use normal value semantics (i.e., destructors)
https://bugs.webkit.org/show_bug.cgi?id=87302
Reviewed by Oliver Hunt.
This is a step toward incremental DOM finalization.
* heap/CopiedSpace.cpp:
(JSC::CopiedSpace::~CopiedSpace):
* heap/CopiedSpace.h:
(CopiedSpace): Just use our destructor, instead of relying on the heap
to send us a special message at a special time.
* heap/Heap.cpp:
(JSC::Heap::Heap): Use OwnPtr for m_markListSet because this is not Sparta.
(JSC::Heap::~Heap): No need for delete or freeAllBlocks because normal
destructors do this work automatically now.
(JSC::Heap::lastChanceToFinalize): Just call lastChanceToFinalize on our
sub-objects, and assume it does the right thing. This improves encapsulation,
so we can add items requiring finalization to our sub-objects.
* heap/Heap.h: Moved m_blockAllocator to get the right destruction order.
* heap/MarkedSpace.cpp:
(Take):
(JSC):
(JSC::Take::Take):
(JSC::Take::operator()):
(JSC::Take::returnValue): Moved to the top of the file so it can be used
in another function.
(JSC::MarkedSpace::~MarkedSpace): Delete all outstanding memory, like a good
destructor should.
(JSC::MarkedSpace::lastChanceToFinalize): Moved some code here from the heap,
since it pertains to our internal implementation details.
* heap/MarkedSpace.h:
(MarkedSpace):
* heap/WeakBlock.cpp:
(JSC::WeakBlock::lastChanceToFinalize):
* heap/WeakBlock.h:
(WeakBlock):
* heap/WeakSet.cpp:
(JSC::WeakSet::lastChanceToFinalize):
* heap/WeakSet.h:
(WeakSet): Stop using a special freeAllBlocks() callback and just implement
lastChanceToFinalize.
2011-05-22 Geoffrey Garen <ggaren@apple.com>
Encapsulated some calculations for whether portions of the heap are empty
https://bugs.webkit.org/show_bug.cgi?id=87210
Reviewed by Gavin Barraclough.
This is a step toward incremental DOM finalization.
* heap/Heap.cpp:
(JSC::Heap::~Heap): Explicitly call freeAllBlocks() instead of relying
implicitly on all blocks thinking they're empty. In future, we may
choose to tear down the heap without first setting all data structures
to "empty".
* heap/MarkedBlock.h:
(JSC::MarkedBlock::isEmpty):
(JSC::MarkedBlock::gatherDirtyCells): Renamed markCountIsZero to isEmpty,
in preparation for making it check for outstanding finalizers in addition
to marked cells.
* heap/MarkedSpace.cpp:
(Take):
(JSC::Take::Take):
(JSC::Take::operator()):
(JSC::Take::returnValue):
(JSC::MarkedSpace::shrink):
(JSC::MarkedSpace::freeAllBlocks): Refactored the "Take" functor to support
a conditional isEmpty check, so it dould be shared by shrink() and freeAllBlocks().
* heap/WeakBlock.cpp:
(JSC::WeakBlock::WeakBlock):
(JSC::WeakBlock::visitLiveWeakImpls):
(JSC::WeakBlock::visitDeadWeakImpls):
* heap/WeakBlock.h:
(WeakBlock):
(JSC::WeakBlock::isEmpty):
* heap/WeakSet.cpp:
(JSC::WeakSet::sweep):
(JSC::WeakSet::shrink): Use isEmpty(), in preparation for changes in
its implementation.
2012-05-23 Oswald Buddenhagen <oswald.buddenhagen@nokia.com>
[Qt] Remove references to $$QT_SOURCE_TREE
With a modularized Qt, it's ambigious. What we really want is qtbase,
which qtcore is a proxy for (we assume it will always live in qtbase).
Reviewed by Tor Arne Vestbø.
* JavaScriptCore.pri:
* Target.pri:
2012-05-09 Filip Pizlo <fpizlo@apple.com>
DFG should allow inlining in case of certain arity mismatches
https://bugs.webkit.org/show_bug.cgi?id=86059
Reviewed by Geoff Garen.
Merge r116620 from dfgopt.
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::handleInlining):
2012-05-08 Filip Pizlo <fpizlo@apple.com>
DFG variable capture analysis should work even if the variables arose through inlining
https://bugs.webkit.org/show_bug.cgi?id=85945
Reviewed by Oliver Hunt.
Merged r116555 from dfgopt.
This just changes how the DFG queries whether a variable is captured. It does not
change any user-visible behavior.
As part of this change, I further solidified the policy that the CFA behaves in an
undefined way for captured locals and queries about their values will not yield
reliable results. This will likely be changed in the future, but for now it makes
sense.
One fun part about this change is that it recognizes that the same variable may
be both captured and not, at the same time, because their live interval spans
inlining boundaries. This only happens in the case of arguments to functions that
capture their arguments, and this change treats them with just the right touch of
conservatism: they will be treated as if captured by the caller as well as the
callee.
Finally, this also adds captured variable reasoning to the InlineCallFrame, which
I thought might be useful for later tooling.
This is perf-neutral, since it does it does not make the DFG take advantage of this
new functionality in any way. In particular, it is still the case that the DFG will
not inline functions that use arguments reflectively or that create activations.
* bytecode/CodeBlock.h:
(CodeBlock):
(JSC::CodeBlock::needsActivation):
(JSC::CodeBlock::argumentIsCaptured):
(JSC::CodeBlock::localIsCaptured):
(JSC::CodeBlock::isCaptured):
* bytecode/CodeOrigin.h:
(InlineCallFrame):
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::initialize):
(JSC::DFG::AbstractState::endBasicBlock):
(JSC::DFG::AbstractState::execute):
(JSC::DFG::AbstractState::merge):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::newVariableAccessData):
(JSC::DFG::ByteCodeParser::getLocal):
(JSC::DFG::ByteCodeParser::setLocal):
(JSC::DFG::ByteCodeParser::getArgument):
(JSC::DFG::ByteCodeParser::setArgument):
(JSC::DFG::ByteCodeParser::flushArgument):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::processPhiStack):
(JSC::DFG::ByteCodeParser::fixVariableAccessPredictions):
(JSC::DFG::ByteCodeParser::InlineStackEntry::InlineStackEntry):
* dfg/DFGCFGSimplificationPhase.cpp:
(CFGSimplificationPhase):
(JSC::DFG::CFGSimplificationPhase::keepOperandAlive):
(JSC::DFG::CFGSimplificationPhase::fixPossibleGetLocal):
(JSC::DFG::CFGSimplificationPhase::fixTailOperand):
* dfg/DFGCommon.h:
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::nameOfVariableAccessData):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::needsActivation):
(JSC::DFG::Graph::usesArguments):
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::doRoundOfDoubleVoting):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGVariableAccessData.h:
(JSC::DFG::VariableAccessData::VariableAccessData):
(JSC::DFG::VariableAccessData::mergeIsCaptured):
(VariableAccessData):
(JSC::DFG::VariableAccessData::isCaptured):
2012-05-08 Filip Pizlo <fpizlo@apple.com>
DFG should support op_get_argument_by_val and op_get_arguments_length
https://bugs.webkit.org/show_bug.cgi?id=85911
Reviewed by Oliver Hunt.
Merged r116467 from dfgopt.
This adds a simple and relatively conservative implementation of op_get_argument_by_val
and op_get_arguments_length. We can optimize these later. For now it's great to have
the additional coverage.
This patch appears to be perf-neutral.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGAssemblyHelpers.h:
(JSC::DFG::AssemblyHelpers::addressFor):
(JSC::DFG::AssemblyHelpers::tagFor):
(JSC::DFG::AssemblyHelpers::payloadFor):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
(JSC::DFG::canInlineOpcode):
* dfg/DFGNode.h:
(JSC::DFG::Node::hasHeapPrediction):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.h:
(JSC::DFG::SpeculativeJIT::callOperation):
(SpeculativeJIT):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_get_argument_by_val):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_get_argument_by_val):
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
2012-05-07 Filip Pizlo <fpizlo@apple.com>
DFG should support op_tear_off_arguments
https://bugs.webkit.org/show_bug.cgi?id=85847
Reviewed by Michael Saboff.
Merged r116378 from dfgopt.
* dfg/DFGAbstractState.cpp:
(JSC::DFG::AbstractState::execute):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::parseBlock):
* dfg/DFGCapabilities.h:
(JSC::DFG::canCompileOpcode):
(JSC::DFG::canInlineOpcode):
* dfg/DFGNodeType.h:
(DFG):
* dfg/DFGOperations.cpp:
* dfg/DFGOperations.h:
* dfg/DFGPredictionPropagationPhase.cpp:
(JSC::DFG::PredictionPropagationPhase::propagate):
* dfg/DFGSpeculativeJIT.h:
(SpeculativeJIT):
(JSC::DFG::SpeculativeJIT::callOperation):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
2012-05-22 Mark Hahnenberg <mhahnenberg@apple.com>
CopiedSpace::contains doesn't check for oversize blocks
https://bugs.webkit.org/show_bug.cgi?id=87180
Reviewed by Geoffrey Garen.
When doing a conservative scan we use CopiedSpace::contains to determine if a particular
address points into the CopiedSpace. Currently contains() only checks if the address
points to a block in to-space, which means that pointers to oversize blocks may not get scanned.
* heap/CopiedSpace.cpp:
(JSC::CopiedSpace::tryAllocateOversize):
(JSC::CopiedSpace::tryReallocateOversize):
(JSC::CopiedSpace::doneFillingBlock):
(JSC::CopiedSpace::doneCopying):
* heap/CopiedSpace.h: Refactored CopiedSpace so that all blocks (oversize and to-space) are
in a single hash set and bloom filter for membership testing.
(CopiedSpace):
* heap/CopiedSpaceInlineMethods.h:
(JSC::CopiedSpace::contains): We check for the normal block first. Since the oversize blocks are
only page aligned, rather than block aligned, we have to re-mask the ptr to check if it's in
CopiedSpace. Also added a helper function of the same name that takes a CopiedBlock* and checks
if it's in CopiedSpace so that check isn't typed out twice.
(JSC):
(JSC::CopiedSpace::startedCopying):
(JSC::CopiedSpace::addNewBlock):
2012-05-22 Geoffrey Garen <ggaren@apple.com>
CopiedBlock and MarkedBlock should have proper value semantics (i.e., destructors)
https://bugs.webkit.org/show_bug.cgi?id=87172
Reviewed by Oliver Hunt and Phil Pizlo.
This enables MarkedBlock to own non-trivial sub-objects that require
destruction. It also fixes a FIXME about casting a CopiedBlock to a
MarkedBlock at destroy time.
CopiedBlock and MarkedBlock now accept an allocation chunk at create
time and return it at destroy time. Their client is expected to
allocate, recycle, and destroy these chunks.
* heap/BlockAllocator.cpp:
(JSC::BlockAllocator::releaseFreeBlocks):
(JSC::BlockAllocator::blockFreeingThreadMain): Don't call MarkedBlock::destroy
because we expect that to be called before a block is put on our free
list now. Do manually deallocate our allocation chunk because that's
our job now.
* heap/BlockAllocator.h:
(BlockAllocator):
(JSC::BlockAllocator::allocate): Allocate never fails now. This is a
cleaner abstraction because only one object does all the VM allocation
and deallocation. Caching is an implementation detail.
(JSC::BlockAllocator::deallocate): We take an allocation chunk argument
instead of a block because we now expect the block to have been destroyed
before we recycle its memory. For convenience, we still use the HeapBlock
class as our linked list node. This is OK because HeapBlock is a POD type.
* heap/CopiedBlock.h:
(CopiedBlock):
(JSC::CopiedBlock::create):
(JSC::CopiedBlock::destroy):
(JSC::CopiedBlock::CopiedBlock): Added proper create and destroy functions,
to match MarkedBlock.
* heap/CopiedSpace.cpp:
(JSC::CopiedSpace::tryAllocateOversize):
(JSC::CopiedSpace::tryReallocateOversize):
(JSC::CopiedSpace::doneCopying):
(JSC::CopiedSpace::getFreshBlock):
(JSC::CopiedSpace::freeAllBlocks):
* heap/CopiedSpaceInlineMethods.h:
(JSC::CopiedSpace::recycleBlock): Make sure to call destroy before
returning a block to the BlockAllocator. Otherwise, our destructors
won't run. (If we get this wrong now, we'll get a compile error.)
* heap/HeapBlock.h:
(JSC::HeapBlock::HeapBlock): const!
* heap/MarkedAllocator.cpp:
(JSC::MarkedAllocator::allocateBlock): No need to distinguish between
create and recycle -- MarkedBlock always accepts memory allocated by
its client now.
* heap/MarkedBlock.cpp:
(JSC::MarkedBlock::create): Don't allocate memory -- we assume that we're
passed already-allocated memory, to clarify the responsibility for VM
recycling.
(JSC::MarkedBlock::destroy): Do run our destructor before giving back
our VM -- that is the whole point of this patch.
(JSC::MarkedBlock::MarkedBlock):
* heap/MarkedBlock.h:
(MarkedBlock):
* heap/MarkedSpace.cpp: const!
(JSC::MarkedSpace::freeBlocks): Make sure to call destroy before
returning a block to the BlockAllocator. Otherwise, our destructors
won't run. (If we get this wrong now, we'll get a compile error.)
== Rolled over to ChangeLog-2012-05-22 ==