2012-10-02 Filip Pizlo <fpizlo@apple.com>

REGRESSION(r128400): ASSERT (crash in release) @ app.asana.com

https://bugs.webkit.org/show_bug.cgi?id=98175

Reviewed by Oliver Hunt.

It's bad karma to create a new structure while stealing another structure's property table

while leaving the m_offset unset. Eventually someone will then steal your property table, and

then you won't know how many properties you have anymore.

* runtime/Structure.cpp:

(JSC::Structure::nonPropertyTransition):

2012-10-02 Michael Saboff <msaboff@apple.com>

Comment additions after r130109

Rubber stamped by Geoffrey Garen.

Updated comments to how array storage works.

* runtime/ArrayStorage.h:

* runtime/JSArray.cpp:

2012-10-01 Mark Hahnenberg <mhahnenberg@apple.com>

Block freeing thread should sleep indefinitely when there's no work to do

https://bugs.webkit.org/show_bug.cgi?id=98084

Reviewed by Geoffrey Garen.

Currently the block freeing thread wakes up once a second to check if there are any blocks

for it to release back to the OS. This is wasteful. We should change it to sleep when it

realizes there are no more blocks to free. Any thread that returns a block to the BlockAllocator

should then notify the block freeing thread that there is more work to do now.

* heap/BlockAllocator.cpp:

(JSC::BlockAllocator::BlockAllocator):

(JSC::BlockAllocator::blockFreeingThreadMain):

* heap/BlockAllocator.h:

(BlockAllocator):

(JSC::BlockAllocator::deallocate):

2012-10-01 Michael Saboff <msaboff@apple.com>

JSArray::unshiftCountSlowCase needs to clear array slots when adding space to end of array

https://bugs.webkit.org/show_bug.cgi?id=98101

Reviewed by Filip Pizlo.

Cleared new array entries when adding to end due to shifting contents to lower memory. Also

checnaged the order of moving array contents and metadata in the shift left case to avoid

clobbering the metadata with array contents. Optimized the to only make a memmove if the

count is non-zero.

* runtime/JSArray.cpp:

(JSC::JSArray::unshiftCountSlowCase):

(JSC::JSArray::unshiftCount):

2012-10-01 Filip Pizlo <fpizlo@apple.com>

Address a FIXME in JSArray::sort

https://bugs.webkit.org/show_bug.cgi?id=98080

<rdar://problem/12407844>

Reviewed by Oliver Hunt.

Get rid of fast sorting of sparse maps. I don't know that it's broken but I do know that we don't

have coverage for it. Then also address the FIXME in JSArray::sort regarding side-effecting

compare functions.

* runtime/ArrayPrototype.cpp:

(JSC::arrayProtoFuncSort):

* runtime/JSArray.cpp:

(JSC::JSArray::sortNumeric):

(JSC::JSArray::sort):

(JSC::JSArray::compactForSorting):

* runtime/JSArray.h:

(JSArray):

* runtime/JSObject.h:

(JSC::JSObject::hasSparseMap):

(JSObject):

2012-10-01 Jonathan Liu <net147@gmail.com>

Remove unused sys/mman.h include

https://bugs.webkit.org/show_bug.cgi?id=97995

Reviewed by Kentaro Hara.

The sys/mman.h is not used and removing it improves portability as not

all systems have sys/mman.h.

* jit/ExecutableAllocatorFixedVMPool.cpp:

2012-09-28 Filip Pizlo <fpizlo@apple.com>

ASSERTION in m_graph[tailNodeIndex].op() == Flush || m_graph[tailNodeIndex].op() == SetLocal on plus.google.com

https://bugs.webkit.org/show_bug.cgi?id=97656

Reviewed by Mark Hahnenberg.

There were two bugs here:

1) In case of multiple GetLocals to the same captured variable, the bytecode parser would linke the first,

rather than the last, of the GetLocals into the vars-at-tail table.

2) The constant folding phase was asserting that any GetLocal it eliminates must be linked into the

vars-at-tail table, when for captured variables only the last of those should be.

* dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::getLocal):

* dfg/DFGConstantFoldingPhase.cpp:

(JSC::DFG::ConstantFoldingPhase::foldConstants):

2012-09-28 Filip Pizlo <fpizlo@apple.com>

DFGStructureHoistingPhase SetLocal assumes StructureTransitionWatchpoint has a structure set

https://bugs.webkit.org/show_bug.cgi?id=97810

Reviewed by Mark Hahnenberg.

No tests because this can't happen in ToT: the structure check hoisting phase runs before any

CFA or folding, so the only StructureTransitionWatchpoints it will see are the ones inserted

by the parser. But the parser will only insert StructureTransitinWatchpoints on constants, which

will not be subject to SetLocals.

Still, it would be good to fix this in case things changed.

* dfg/DFGStructureCheckHoistingPhase.cpp:

(JSC::DFG::StructureCheckHoistingPhase::run):

2012-09-27 Mark Lam <mark.lam@apple.com>

Put initializeHostCallReturnValue() behind #if ENABLE(JIT).

Fixes non JIT builds.

https://bugs.webkit.org/show_bug.cgi?id=97838.

Reviewed by John Sullivan.

* runtime/JSGlobalData.cpp:

(JSC::JSGlobalData::JSGlobalData):

2012-09-27 Mark Lam <mark.lam@apple.com>

Fixed CallFrameClosure::resetCallFrame() to use the valid

range of argument index values.

https://bugs.webkit.org/show_bug.cgi?id=97836.

Reviewed by Gavin Barraclough.

* interpreter/CallFrame.h:

(ExecState):

* interpreter/CallFrameClosure.h:

(JSC::CallFrameClosure::resetCallFrame):

2012-09-27 Patrick Gansterer <paroga@webkit.org>

Fix usage of COMPILER() macros

https://bugs.webkit.org/show_bug.cgi?id=97642

Reviewed by Geoffrey Garen.

Add COMPILER(GCC) around compiler specific code and remove it from generic code.

This allows us to implement the DFG code for other compilers to in a next step.

* dfg/DFGOperations.cpp:

* jit/HostCallReturnValue.h:

2012-09-27 Andreas Kling <kling@webkit.org>

3.20MB below FunctionParameters::create() on Membuster3.

<http://webkit.org/b/97730>

Reviewed by Anders Carlsson.

Figure out the exact space needed for parameter identifiers and use reserveInitialCapacity().

Reduces memory consumption on Membuster3 by ~1.60 MB.

* parser/Nodes.cpp:

(JSC::FunctionParameters::FunctionParameters):

2012-09-27 Csaba Osztrogonác <ossy@webkit.org>, Tor Arne Vestbø <vestbo@webkit.org>

[Qt] Enable the LLInt on Linux

https://bugs.webkit.org/show_bug.cgi?id=95749

Reviewed by Simon Hausmann.

* DerivedSources.pri:

* JavaScriptCore.pro:

* LLIntOffsetsExtractor.pro: Added.

* Target.pri:

2012-09-27 Patrick Gansterer <paroga@webkit.org>

[CMake] Fix build with static JavaScriptCore library

* shell/CMakeLists.txt: Define STATICALLY_LINKED_WITH_JavaScriptCore if

JavaScriptCore_LIBRARY_TYPE is set to STATIC.

2012-09-26 Gavin Barraclough <barraclough@apple.com>

Proxy the global this in JSC

https://bugs.webkit.org/show_bug.cgi?id=97734

Reviewed by Filip Pizlo.

Eeep – fix a bug - was leaving the global this proxy's structure's globalObject as 0,

and setting the proxy's prototype as the global object, rather than its prototype.

* jsc.cpp:

(GlobalObject::create):

* runtime/JSProxy.h:

(JSC::JSProxy::createStructure):

2012-09-26 Gavin Barraclough <barraclough@apple.com>

Speculative Windows build fix.

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:

2012-09-26 Filip Pizlo <fpizlo@apple.com>

Unreviewed, 32-bit build fix.

* llint/LowLevelInterpreter32_64.asm:

2012-09-26 Filip Pizlo <fpizlo@apple.com>

jneq_ptr shouldn't have a pointer

https://bugs.webkit.org/show_bug.cgi?id=97739

Reviewed by Oliver Hunt.

Slamming pointers directly into bytecode is sometimes cool, but in this case it's

unwieldy and confusing. Switched the instruction to use an enum instead. This has

zero effect on code gen behavior in the JITs. In the LLInt, there is now more

indirection, but that doesn't affect benchmarks.

* CMakeLists.txt:

* GNUmakefile.list.am:

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:

* JavaScriptCore.xcodeproj/project.pbxproj:

* Target.pri:

* bytecode/Instruction.h:

(JSC::Instruction::Instruction):

(Instruction):

* bytecode/SpecialPointer.cpp: Added.

(JSC):

(JSC::actualPointerFor):

* bytecode/SpecialPointer.h: Added.

(JSC):

(JSC::pointerIsFunction):

(JSC::pointerIsCell):

* bytecompiler/BytecodeGenerator.cpp:

(JSC::BytecodeGenerator::emitJumpIfNotFunctionCall):

(JSC::BytecodeGenerator::emitJumpIfNotFunctionApply):

* dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::parseBlock):

* jit/JITOpcodes.cpp:

(JSC::JIT::emit_op_jneq_ptr):

* jit/JITOpcodes32_64.cpp:

(JSC::JIT::emit_op_jneq_ptr):

* llint/LowLevelInterpreter32_64.asm:

* llint/LowLevelInterpreter64.asm:

* runtime/JSGlobalObject.cpp:

(JSC::JSGlobalObject::reset):

(JSC):

* runtime/JSGlobalObject.h:

(JSGlobalObject):

(JSC::JSGlobalObject::actualPointerFor):

2012-09-26 Gavin Barraclough <barraclough@apple.com>

REGRESSION (r129456): http/tests/security/xss-eval.html is failing on JSC platforms

https://bugs.webkit.org/show_bug.cgi?id=97529

Reviewed by Filip Pizlo.

A recent patch changed JSC's EvalError behaviour; bring this more into line with other browsers.

JSC currently throws an EvalError if you try to call eval with a this object that doesn't

match the given eval function. This does not match other browsers, which generally just

ignore the this value that was passed, and eval the string in the eval function's environment.

* runtime/JSGlobalObjectFunctions.cpp:

(JSC::globalFuncEval):

- Remove EvalError, ignore passed this value.

2012-09-26 Gavin Barraclough <barraclough@apple.com>

Proxy the global this in JSC

https://bugs.webkit.org/show_bug.cgi?id=97734

Reviewed by Oliver Hunt.

Having jsc diverge from WebCore here is not beneficial; it potentially masks bugs and/or performance

problems from command line testing.

* jsc.cpp:

(GlobalObject::create):

- Create a this value proxy for the global object.

* runtime/JSGlobalObject.h:

(JSGlobalObject):

- Make setGlobalThis protected.

* runtime/JSProxy.h:

(JSC::JSProxy::create):

(JSC::JSProxy::target):

(JSC::JSProxy::finishCreation):

(JSProxy):

- Allow proxy target to be a JSObject, add target to create method.

2012-09-26 Gavin Barraclough <barraclough@apple.com>

Speculative Windows build fix.

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:

2012-09-26 Filip Pizlo <fpizlo@apple.com>

JSObject::ensureArrayStorage() ignores the possibility that extensions have been prevented

https://bugs.webkit.org/show_bug.cgi?id=97719

Reviewed by Gavin Barraclough.

* runtime/JSObject.cpp:

(JSC::JSObject::ensureArrayStorageSlow):

(JSC):

* runtime/JSObject.h:

(JSC::JSObject::ensureArrayStorage):

(JSObject):

2012-09-26 Gavin Barraclough <barraclough@apple.com>

Generalize JSGlobalThis as JSProxy

https://bugs.webkit.org/show_bug.cgi?id=97716

Reviewed by Oliver Hunt.

Generalize JSGlobalThis as JSProxy and move proxying functionality up from the window shell into JSProxy.

* CMakeLists.txt:

* GNUmakefile.list.am:

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:

* JavaScriptCore.xcodeproj/project.pbxproj:

* Target.pri:

* runtime/JSGlobalObject.cpp:

(JSC::JSGlobalObject::toThisObject):

- Hoist toThisObject from WebCore.

(JSC):

* runtime/JSGlobalObject.h:

- removed include.

(JSC::JSGlobalObject::finishCreation):

- JSGlobalThis -> JSObject

(JSGlobalObject):

- Hoist toThisObject from WebCore.

* runtime/JSGlobalThis.cpp: Removed.

* runtime/JSGlobalThis.h: Removed.

* runtime/JSObject.cpp:

- removed include.

* runtime/JSObject.h:

(JSObject):

(JSC::JSObject::isProxy):

- isGlobalThis -> isProxy

- GlobalThisType -> ProxyType

* runtime/JSProxy.cpp: Copied from Source/JavaScriptCore/runtime/JSGlobalThis.cpp.

(JSC):

(JSC::JSProxy::visitChildren):

(JSC::JSProxy::setTarget):

(JSC::JSProxy::className):

(JSC::JSProxy::getOwnPropertySlot):

(JSC::JSProxy::getOwnPropertySlotByIndex):

(JSC::JSProxy::getOwnPropertyDescriptor):

(JSC::JSProxy::put):

(JSC::JSProxy::putByIndex):

(JSC::JSProxy::putDirectVirtual):

(JSC::JSProxy::defineOwnProperty):

(JSC::JSProxy::deleteProperty):

(JSC::JSProxy::deletePropertyByIndex):

(JSC::JSProxy::getPropertyNames):

(JSC::JSProxy::getOwnPropertyNames):

- Class cretaed from JSGlobalThis, JSDOMWindowShell.

* runtime/JSProxy.h: Copied from Source/JavaScriptCore/runtime/JSGlobalThis.h.

(JSC::JSProxy::create):

(JSC::JSProxy::createStructure):

(JSProxy):

(JSC::JSProxy::target):

(JSC::JSProxy::JSProxy):

- Class cretaed from JSGlobalThis, JSDOMWindowShell.

* runtime/JSType.h:

- GlobalThisType -> ProxyType

2012-09-26 Michael Saboff <msaboff@apple.com>

Add ability for JSArray::unshiftCount to unshift in middle of an array

https://bugs.webkit.org/show_bug.cgi?id=97691

Reviewed by Filip Pizlo.

Changed JSArray::unshiftCount and unshiftCountSlowCase to handle unshifting from the middle of an

array. Depending on where the unshift point is, either the front part of the array will be moved

"left" or the back part will be moved right. Given that unshiftCount only works on contiguous

arrays it is safe to use memmove for the moves.

This change is worth 25% performance improvement on pdfjs. It doesn't seem to have any impact on

any other benchmarks.

* runtime/ArrayPrototype.cpp:

(JSC::unshift):

* runtime/JSArray.cpp:

(JSC::JSArray::unshiftCountSlowCase):

(JSC::JSArray::unshiftCount):

* runtime/JSArray.h:

(JSArray):

2012-09-26 Sheriff Bot <webkit.review.bot@gmail.com>

Unreviewed, rolling out r129592.

http://trac.webkit.org/changeset/129592

https://bugs.webkit.org/show_bug.cgi?id=97670

Failures in Chromium security tests (Requested by schenney on

#webkit).

* runtime/JSGlobalObjectFunctions.cpp:

(JSC::globalFuncEval):

2012-09-25 Gavin Barraclough <barraclough@apple.com>

REGRESSION (r129456): http/tests/security/xss-eval.html is failing on JSC platforms

https://bugs.webkit.org/show_bug.cgi?id=97529

Reviewed by Filip Pizlo.

A recent patch changed JSC's EvalError behaviour; bring this more into line with other browsers.

JSC currently throws an EvalError if you try to call eval with a this object that doesn't

match the given eval function. This does not match other browsers, which generally just

ignore the this value that was passed, and eval the string in the eval function's environment.

* runtime/JSGlobalObjectFunctions.cpp:

(JSC::globalFuncEval):

- Remove EvalError, ignore passed this value.

2012-09-25 Filip Pizlo <fpizlo@apple.com>

DFG ArrayPush, ArrayPop don't handle clobbering or having a bad time correctly

https://bugs.webkit.org/show_bug.cgi?id=97535

Reviewed by Oliver Hunt.

* dfg/DFGAbstractState.cpp:

(JSC::DFG::AbstractState::execute):

* dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::handleIntrinsic):

* dfg/DFGStructureCheckHoistingPhase.cpp:

(JSC::DFG::StructureCheckHoistingPhase::run):

2012-09-25 Geoffrey Garen <ggaren@apple.com>

JSC should dump object size inference statistics

https://bugs.webkit.org/show_bug.cgi?id=97618

Reviewed by Filip Pizlo.

Added an option to dump object size inference statistics.

To see statistics on live objects:

jsc --showHeapStatistics=1

To see cumulative statistics on all objects ever allocated:

jsc --showHeapStatistics=1 --objectsAreImmortal=1

(This is useful for showing GC churn caused by over-allocation.)

To support this second mode, I refactored Zombies to separate out their

immortality feature so I could reuse it.

* heap/Heap.cpp:

(JSC::MarkObject): Helper for making things immortal. We have to checked

for being zapped because blocks start out in this state.

(JSC::StorageStatistics): Gather statistics by walking the heap. Ignore

arrays and hash tables for now because they're not our focus. (We'll

remove these exceptions in future.)

(JSC::Heap::collect): Moved zombify to the end so it wouldn't interfere

with statistics gathering.

(JSC::Heap::showStatistics):

(JSC::Heap::markAllObjects): Factored out helper, so statistics could

take advantage of immortal objects.

(Zombify): Don't mark immortal objects -- that's another class's job now.

(JSC::Zombify::operator()):

(JSC::Heap::zombifyDeadObjects): Take advantage of forEachDeadCell instead

of rolling our own.

* heap/Heap.h:

(Heap):

* heap/MarkedSpace.h:

(MarkedSpace):

(JSC::MarkedSpace::forEachDeadCell): Added, so clients don't have to do

the iteration logic themselves.

* runtime/Options.cpp:

(JSC::Options::initialize):

* runtime/Options.h: New options, listed above. Make sure to initialize

based on environment variable first, so we can override with specific settings.

2012-09-25 Filip Pizlo <fpizlo@apple.com>

We shouldn't use the optimized versions of shift/unshift if the user is doing crazy things to the array

https://bugs.webkit.org/show_bug.cgi?id=97603

<rdar://problem/12370864>

Reviewed by Gavin Barraclough.

You changed the length behind our backs? No optimizations for you then!

* runtime/ArrayPrototype.cpp:

(JSC::shift):

(JSC::unshift):

* runtime/JSArray.cpp:

(JSC::JSArray::shiftCount):

2012-09-25 Filip Pizlo <fpizlo@apple.com>

JSC bindings appear to sometimes ignore the possibility of arrays being in sparse mode

https://bugs.webkit.org/show_bug.cgi?id=95610

Reviewed by Oliver Hunt.

Add better support for quickly accessing the indexed storage from bindings.

* runtime/JSObject.h:

(JSC::JSObject::tryGetIndexQuickly):

(JSObject):

(JSC::JSObject::getDirectIndex):

(JSC::JSObject::getIndex):

2012-09-25 Filip Pizlo <fpizlo@apple.com>

Structure check hoisting phase doesn't know about the side-effecting nature of Arrayify

https://bugs.webkit.org/show_bug.cgi?id=97537

Reviewed by Mark Hahnenberg.

No tests because if we use Arrayify then we also use PutByVal(BlankToXYZ), and the latter is

already known to be side-effecting. So this bug shouldn't have had any symptoms, as far as I

can tell.

* dfg/DFGStructureCheckHoistingPhase.cpp:

(JSC::DFG::StructureCheckHoistingPhase::run):

2012-09-25 Gavin Barraclough <barraclough@apple.com>

Regression: put beyond vector length prefers prototype setters to sparse properties

https://bugs.webkit.org/show_bug.cgi?id=97593

Reviewed by Geoff Garen & Filip Pizlo.

* runtime/JSObject.cpp:

(JSC::JSObject::putByIndexBeyondVectorLength):

- Check for self properties in the sparse map - if present, don't examine the protochain.

2012-09-24 Gavin Barraclough <barraclough@apple.com>

https://bugs.webkit.org/show_bug.cgi?id=97530

Regression, freeze applied to numeric properties of non-array objects

Reviewed by Filip Pizlo.

Object.freeze has a fast implementation in JSObject, but this hasn't been updated to take into account numeric properties in butterflies.

For now, just fall back to the generic implementation if the object has numeric properties.

* runtime/ObjectConstructor.cpp:

(JSC::objectConstructorFreeze):

- fallback if the object has a non-zero indexed property vector length.

2012-09-24 Gavin Barraclough <barraclough@apple.com>

Bug in numeric accessors on global environment

https://bugs.webkit.org/show_bug.cgi?id=97526

Reviewed by Geoff Garen.

I've hit this assert in test262 in browser, but haven't yet worked out how to repro in a test case :-/

The sparsemap is failing to map back from the global object to the window shell.

A test case would need to resolve a numeric property name against the global environment.

(JSC::SparseArrayEntry::get):

(JSC::SparseArrayEntry::put):

- Add missing toThisObject calls.

2012-09-24 Filip Pizlo <fpizlo@apple.com>

SerializedScriptValue isn't aware of indexed storage, but should be

https://bugs.webkit.org/show_bug.cgi?id=97515

<rdar://problem/12361874>

Reviewed by Sam Weinig.

Export a method that WebCore now uses.

* JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.def:

* runtime/JSObject.h:

(JSObject):

2012-09-24 Gavin Barraclough <barraclough@apple.com>

Remove JSObject::unwrappedGlobalObject(), JSObject::unwrappedObject()

https://bugs.webkit.org/show_bug.cgi?id=97519

Reviewed by Geoff Garen.

unwrappedGlobalObject() was only needed because globalObject() doesn't always return a helpful result -

specifically for WebCore's window shell the structure's globalObject is set to null. We can fix this by

simply keeping the structure up to date as the window navigates, obviating the need for this function.

The only other use of unwrappedObject() came from globalFuncEval(), and this can be trivially removed

by flipping the way we perform this globalObject check (which we may also be able to remove!) - instead

of getting the globalObject from the provided this value & comparing to the expected globalObject, we

can get the this value from the expected globalObject, and compare to that provided.

* runtime/JSGlobalObject.cpp:

- Call globalObject() instead of unwrappedGlobalObject().

* runtime/JSGlobalObjectFunctions.cpp:

(JSC::globalFuncEval):

- Changed to compare this object values, instead of globalObjects -

this means we only need to be able to map globalObject -> this,

and not vice versa.

* runtime/JSObject.cpp:

(JSC::JSObject::allowsAccessFrom):

(JSC::JSObject::createInheritorID):

- Call globalObject() instead of unwrappedGlobalObject().

* runtime/JSObject.h:

(JSObject):

- Removed unwrappedGlobalObject(), unwrappedObject().

2012-09-24 Mark Lam <mark.lam@apple.com>

Deleting the classic interpreter and cleaning up some build options.

https://bugs.webkit.org/show_bug.cgi?id=96969.

Reviewed by Geoffrey Garen.

* bytecode/CodeBlock.cpp:

(JSC::CodeBlock::dump):

(JSC::CodeBlock::finalizeUnconditionally):

(JSC::CodeBlock::stronglyVisitStrongReferences):

(JSC):

* bytecode/Instruction.h:

(JSC::Instruction::Instruction):

* interpreter/AbstractPC.cpp:

(JSC::AbstractPC::AbstractPC):

* interpreter/AbstractPC.h:

(AbstractPC):

* interpreter/CallFrame.h:

(ExecState):

* interpreter/Interpreter.cpp:

(JSC):

(JSC::Interpreter::Interpreter):

(JSC::Interpreter::~Interpreter):

(JSC::Interpreter::initialize):

(JSC::Interpreter::isOpcode):

(JSC::Interpreter::unwindCallFrame):

(JSC::getLineNumberForCallFrame):

(JSC::getCallerInfo):

(JSC::getSourceURLFromCallFrame):

(JSC::Interpreter::execute):

(JSC::Interpreter::executeCall):

(JSC::Interpreter::executeConstruct):

(JSC::Interpreter::retrieveArgumentsFromVMCode):

(JSC::Interpreter::retrieveCallerFromVMCode):

(JSC::Interpreter::retrieveLastCaller):

* interpreter/Interpreter.h:

(JSC::Interpreter::getOpcodeID):

(Interpreter):

* jit/ExecutableAllocatorFixedVMPool.cpp:

(JSC::FixedVMPoolExecutableAllocator::FixedVMPoolExecutableAllocator):

* offlineasm/asm.rb:

* offlineasm/offsets.rb:

* runtime/Executable.cpp:

(JSC::EvalExecutable::compileInternal):

(JSC::ProgramExecutable::compileInternal):

(JSC::FunctionExecutable::compileForCallInternal):

(JSC::FunctionExecutable::compileForConstructInternal):

* runtime/Executable.h:

(JSC::NativeExecutable::create):

(NativeExecutable):

(JSC::NativeExecutable::finishCreation):

* runtime/JSGlobalData.cpp:

(JSC):

(JSC::JSGlobalData::JSGlobalData):

(JSC::JSGlobalData::getHostFunction):

* runtime/JSGlobalData.h:

(JSGlobalData):

(JSC::JSGlobalData::canUseJIT):

(JSC::JSGlobalData::canUseRegExpJIT):

* runtime/Options.cpp:

(JSC::Options::initialize):

2012-09-24 Filip Pizlo <fpizlo@apple.com>

Nested try/finally should not confuse the finally unpopper in BytecodeGenerator::emitComplexJumpScopes

https://bugs.webkit.org/show_bug.cgi?id=97508

<rdar://problem/12361132>

Reviewed by Sam Weinig.

We're reusing some vector for multiple iterations of a loop, but we were forgetting to clear its

contents from one iteration to the next. Hence if you did multiple iterations of finally unpopping

(like in a nested try/finally and a jump out of both of them) then you'd get a corrupted try

context stack afterwards.

* bytecompiler/BytecodeGenerator.cpp:

(JSC::BytecodeGenerator::emitComplexJumpScopes):

2012-09-24 Filip Pizlo <fpizlo@apple.com>

ValueToInt32 bool case does bad things to registers

https://bugs.webkit.org/show_bug.cgi?id=97505

<rdar://problem/12356331>

Reviewed by Mark Hahnenberg.

* dfg/DFGSpeculativeJIT.cpp:

(JSC::DFG::SpeculativeJIT::compileValueToInt32):

2012-09-24 Mark Lam <mark.lam@apple.com>

Add cloopDo instruction for debugging the llint C++ backend.

https://bugs.webkit.org/show_bug.cgi?id=97502.

Reviewed by Geoffrey Garen.

* offlineasm/cloop.rb:

* offlineasm/instructions.rb:

* offlineasm/parser.rb:

2012-09-24 Filip Pizlo <fpizlo@apple.com>

JSArray::putByIndex asserts with readonly property on prototype

https://bugs.webkit.org/show_bug.cgi?id=97435

<rdar://problem/12357084>

Reviewed by Geoffrey Garen.

Boy, there were some problems:

- putDirectIndex() should know that it can set the index quickly even if it's a hole and we're

in SlowPut mode, since that's the whole point of PutDirect.

- We should have a fast path for putByIndex().

- The LiteralParser should not use push(), since that may throw if we're having a bad time.

* interpreter/Interpreter.cpp:

(JSC::eval):

* runtime/JSObject.h:

(JSC::JSObject::putByIndexInline):

(JSObject):

(JSC::JSObject::putDirectIndex):

* runtime/LiteralParser.cpp:

(JSC::::parse):

2012-09-24 Mark Lam <mark.lam@apple.com>

Added a missing "if VALUE_PROFILER" around an access to ArrayProfile record.

https://bugs.webkit.org/show_bug.cgi?id=97496.

Reviewed by Filip Pizlo.

* llint/LowLevelInterpreter32_64.asm:

* llint/LowLevelInterpreter64.asm:

2012-09-24 Geoffrey Garen <ggaren@apple.com>

Inlined activation tear-off in the DFG

https://bugs.webkit.org/show_bug.cgi?id=97487

Reviewed by Filip Pizlo.

* dfg/DFGOperations.cpp:

* dfg/DFGOperations.h: Activation tear-off is always inlined now, so I

removed its out-of-line implementation.

* dfg/DFGSpeculativeJIT32_64.cpp:

(JSC::DFG::SpeculativeJIT::compile):

* dfg/DFGSpeculativeJIT64.cpp:

(JSC::DFG::SpeculativeJIT::compile): Inlined the variable copy and update

of JSVariableObject::m_registers. This usually turns into < 10 instructions,

which is close to pure win as compared to the operation function call.

* runtime/JSActivation.h:

(JSActivation):

(JSC::JSActivation::registersOffset):

(JSC::JSActivation::tearOff):

(JSC::JSActivation::isTornOff):

(JSC):

(JSC::JSActivation::storageOffset):

(JSC::JSActivation::storage): Tiny bit of refactoring so the JIT can

share the pointer math helper functions we use internally.

2012-09-24 Balazs Kilvady <kilvadyb@homejinni.com>

MIPS: store8 functions added to MacroAssembler.

MIPS store8 functions

https://bugs.webkit.org/show_bug.cgi?id=97243

Reviewed by Oliver Hunt.

Add MIPS store8 functions.

* assembler/MIPSAssembler.h:

(JSC::MIPSAssembler::lhu): New function.

(MIPSAssembler):

(JSC::MIPSAssembler::sb): New function.

(JSC::MIPSAssembler::sh): New function.

* assembler/MacroAssemblerMIPS.h:

(JSC::MacroAssemblerMIPS::store8): New function.

(MacroAssemblerMIPS):

(JSC::MacroAssemblerMIPS::store16): New function.

2012-09-23 Geoffrey Garen <ggaren@apple.com>

PutScopedVar should not be marked as clobbering the world

https://bugs.webkit.org/show_bug.cgi?id=97416

Reviewed by Filip Pizlo.

No performance change.

PutScopedVar doesn't have arbitrary side-effects, so it shouldn't be marked

as such.

* dfg/DFGNodeType.h:

(DFG):

2012-09-23 Geoffrey Garen <ggaren@apple.com>

I accidentally the whole 32-bit :(.

Unbreak the DFG in 32-bit with the 32-bit path I forgot in my last patch.

* dfg/DFGSpeculativeJIT32_64.cpp:

(JSC::DFG::SpeculativeJIT::compile):

2012-09-23 Byungwoo Lee <bw80.lee@gmail.com>

Fix build warnings : -Wunused-parameter, -Wparentheses, -Wuninitialized.

https://bugs.webkit.org/show_bug.cgi?id=97306

Reviewed by Benjamin Poulain.

Fix build warning about -Wunused-parameter on MachineStackMarker.cpp,

LLIntSlowPaths.cpp, DatePrototype.cpp, Options.cpp by using

UNUSED_PARAM() macro or remove parameter name.

* heap/MachineStackMarker.cpp:

(JSC::pthreadSignalHandlerSuspendResume):

* llint/LLIntSlowPaths.cpp:

(JSC::LLInt::entryOSR):

* runtime/DatePrototype.cpp:

(JSC::formatLocaleDate):

* runtime/Options.cpp:

(JSC::computeNumberOfGCMarkers):

2012-09-23 Gavin Barraclough <barraclough@apple.com>

Sorting a non-array creates propreties (spec-violation)

https://bugs.webkit.org/show_bug.cgi?id=25477

Reviewed by Oliver Hunt.

We're just calling get() to get properties, which is converting missing properties to

undefined. Hole values should be retained, and moved to the end of the array.

* runtime/ArrayPrototype.cpp:

(JSC::getOrHole):

- Helper function, returns JSValue() instead of undefined for missing properties.

(JSC::arrayProtoFuncSort):

- Implemented per 15.4.4.11, see comments above.

2012-09-23 Geoffrey Garen <ggaren@apple.com>

CSE for access to closure variables (get_/put_scoped_var)

https://bugs.webkit.org/show_bug.cgi?id=97414

Reviewed by Oliver Hunt.

I separated loading a scope from loading its storage pointer, so we can

CSE the storage pointer load. Then, I copied the global var CSE and adjusted

it for closure vars.

* dfg/DFGAbstractState.cpp:

(JSC::DFG::AbstractState::execute): Renamed GetScopeChain => GetScope to

reflect renames from a few weeks ago.

Added a case for the storage pointer load, similar to object storage pointer load.

* dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::parseBlock): Added an independent node for

the storage pointer.

* dfg/DFGCSEPhase.cpp:

(JSC::DFG::CSEPhase::scopedVarLoadElimination):

(CSEPhase):

(JSC::DFG::CSEPhase::scopedVarStoreElimination):

(JSC::DFG::CSEPhase::getScopeLoadElimination):

(JSC::DFG::CSEPhase::getScopeRegistersLoadElimination):

(JSC::DFG::CSEPhase::setLocalStoreElimination):

(JSC::DFG::CSEPhase::performNodeCSE): Copied globalVarLoad/StoreElimination

and adapted the same logic to closure vars.

* dfg/DFGNode.h:

(JSC::DFG::Node::hasScopeChainDepth):

(JSC::DFG::Node::scope):

(Node):

* dfg/DFGNodeType.h:

(DFG): GetScopedVar and GetGlobalVar are no longer MustGenerate. I'm not

sure why they ever were. But these are simple load operations so, if they're

unused, they're truly dead.

* dfg/DFGPredictionPropagationPhase.cpp:

(JSC::DFG::PredictionPropagationPhase::propagate):

* dfg/DFGSpeculativeJIT32_64.cpp:

(JSC::DFG::SpeculativeJIT::compile):

* dfg/DFGSpeculativeJIT64.cpp:

(JSC::DFG::SpeculativeJIT::compile): Updated for renames and split-out

node for getting the storage pointer.

2012-09-21 Geoffrey Garen <ggaren@apple.com>

Unreviewed, rolled out a line I committed by accident.

* interpreter/Interpreter.cpp:

(JSC::Interpreter::execute):

2012-09-21 Geoffrey Garen <ggaren@apple.com>

Optimized closures that capture arguments

https://bugs.webkit.org/show_bug.cgi?id=97358

Reviewed by Oliver Hunt.

Previously, the activation object was responsible for capturing all

arguments in a way that was convenient for the arguments object. Now,

we move all captured variables into a contiguous region in the stack,

allocate an activation for exactly that size, and make the arguments

object responsible for knowing all the places to which arguments could

have moved.

This seems like the right tradeoff because

(a) Closures are common and long-lived, so we want them to be small.

(b) Our primary strategy for optimizing the arguments object is to make

it go away. If you're allocating arguments objects, you're already having

a bad time.

(c) It's common to use either the arguments object or named argument

closure, but not both.

* bytecode/CodeBlock.cpp:

(JSC::CodeBlock::dump):

(JSC::CodeBlock::CodeBlock):

* bytecode/CodeBlock.h:

(JSC::CodeBlock::argumentsRegister):

(JSC::CodeBlock::activationRegister):

(JSC::CodeBlock::isCaptured):

(JSC::CodeBlock::argumentIndexAfterCapture): m_numCapturedVars is gone

now -- we have an explicit range instead.

* bytecompiler/BytecodeGenerator.cpp:

(JSC::BytecodeGenerator::BytecodeGenerator): Move captured arguments

into the captured region of local variables for space efficiency. Record

precise data about where they moved for the sake of the arguments object.

Some of this data was previously wrong, but it didn't cause any problems

because the arguments weren't actually moving.

* dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::flushArgumentsAndCapturedVariables): Don't

assume that captured vars are in any particular location -- always ask

the CodeBlock. This is better encapsulation.

(JSC::DFG::ByteCodeParser::parseCodeBlock):

* dfg/DFGSpeculativeJIT32_64.cpp:

(JSC::DFG::SpeculativeJIT::compile):

* dfg/DFGSpeculativeJIT64.cpp:

(JSC::DFG::SpeculativeJIT::compile): I rename things sometimes.