Skip to content
  • barraclough@apple.com's avatar
    2009-05-21 Gavin Barraclough <barraclough@apple.com> · 04b762d2
    barraclough@apple.com authored
            Reviewed by Darin Adler.
            Addition of MacroAssemblerCodeRef.h rubber stamped by Geoff Garen.
    
            Refactor JIT code-handle objects.  The representation of generated code is currently
            a bit of a mess.  We have a class JITCode which wraps the pointer to a block of
            generated code, but this object does not reference the executable pool meaning that
            external events (the pool being derefed) could make the pointer become invalid.
            To overcome this both the JIT and Yarr implement further (and similar) objects to
            wrap the code pointer with a RefPtr to the pool.  To add to the mire, as well as the
            CodeBlock containing a handle onto the code the FunctionBodyNode also contains a
            copy of the code pointer which is used almost (but not entirely) uniquely to access
            the JIT code for a function.
    
            Rationalization of all this:
    
                * Add a new type 'MacroAssembler::CodeRef' as a handle for a block of JIT generated code.
                * Change the JIT & Yarr to internally handle code using CodeRefs.
                * Move the CodeRef (formerly anow defunct JITCodeRef) from CodeBlock to its owner node.
                * Remove the (now) redundant code pointer from FunctionBodyNode.
    
            While tidying this up I've made the PatchBuffer return code in new allocations using a CodeRef,
            and have enforced an interface that the PatchBuffer will always be used, and 'finalizeCode()' or
            'finalizeCodeAddendum()' will always be called exactly once on the PatchBuffer to complete code generation.
    
            This gives us a potentially useful hook ('PatchBuffer::performFinalization()') at the end of generation,
            which may have a number of uses.  It may be helpful should we wish to switch our generation
            model to allow RW/RX exclusive memory, and it may be useful on non-cache-coherent platforms to
            give us an oportunity to cache flush as necessary.
    
            No performance impact.
    
            * assembler/AbstractMacroAssembler.h:
            (JSC::AbstractMacroAssembler::ProcessorReturnAddress::relinkCallerToTrampoline):
            (JSC::AbstractMacroAssembler::CodeRef::CodeRef):
            (JSC::AbstractMacroAssembler::CodeRef::trampolineAt):
            (JSC::AbstractMacroAssembler::PatchBuffer::PatchBuffer):
            (JSC::AbstractMacroAssembler::PatchBuffer::~PatchBuffer):
            (JSC::AbstractMacroAssembler::PatchBuffer::link):
            (JSC::AbstractMacroAssembler::PatchBuffer::linkTailRecursive):
            (JSC::AbstractMacroAssembler::PatchBuffer::patch):
            (JSC::AbstractMacroAssembler::PatchBuffer::complete):
            (JSC::AbstractMacroAssembler::PatchBuffer::finalize):
            (JSC::AbstractMacroAssembler::PatchBuffer::entry):
            * bytecode/CodeBlock.cpp:
            (JSC::CodeBlock::CodeBlock):
            (JSC::CodeBlock::reparseForExceptionInfoIfNecessary):
            (JSC::CodeBlock::setJITCode):
            * bytecode/CodeBlock.h:
            (JSC::CodeBlock::getBytecodeIndex):
            (JSC::CodeBlock::executablePool):
            * interpreter/CallFrameClosure.h:
            * interpreter/Interpreter.cpp:
            (JSC::Interpreter::execute):
            (JSC::Interpreter::prepareForRepeatCall):
            * jit/JIT.cpp:
            (JSC::JIT::privateCompile):
            (JSC::JIT::privateCompileCTIMachineTrampolines):
            (JSC::JIT::linkCall):
            * jit/JIT.h:
            * jit/JITCode.h:
            (JSC::JITCode::JITCode):
            (JSC::JITCode::operator bool):
            (JSC::JITCode::addressForCall):
            (JSC::JITCode::offsetOf):
            (JSC::JITCode::execute):
            (JSC::JITCode::size):
            (JSC::JITCode::executablePool):
            (JSC::JITCode::HostFunction):
            * jit/JITPropertyAccess.cpp:
            (JSC::JIT::privateCompilePutByIdTransition):
            (JSC::JIT::privateCompilePatchGetArrayLength):
            (JSC::JIT::privateCompileGetByIdProto):
            (JSC::JIT::privateCompileGetByIdSelfList):
            (JSC::JIT::privateCompileGetByIdProtoList):
            (JSC::JIT::privateCompileGetByIdChainList):
            (JSC::JIT::privateCompileGetByIdChain):
            * jit/JITStubs.cpp:
            (JSC::JITStubs::cti_vm_dontLazyLinkCall):
            (JSC::JITStubs::cti_vm_lazyLinkCall):
            * parser/Nodes.cpp:
            (JSC::ProgramNode::generateJITCode):
            (JSC::EvalNode::generateJITCode):
            (JSC::FunctionBodyNode::FunctionBodyNode):
            (JSC::FunctionBodyNode::createNativeThunk):
            (JSC::FunctionBodyNode::generateJITCode):
            * parser/Nodes.h:
            (JSC::ScopeNode::generatedJITCode):
            (JSC::ScopeNode::getExecutablePool):
            (JSC::ScopeNode::setJITCode):
            (JSC::ProgramNode::jitCode):
            (JSC::EvalNode::jitCode):
            (JSC::FunctionBodyNode::jitCode):
            * runtime/RegExp.cpp:
            (JSC::RegExp::match):
            * yarr/RegexJIT.cpp:
            (JSC::Yarr::RegexGenerator::compile):
            (JSC::Yarr::jitCompileRegex):
            (JSC::Yarr::executeRegex):
            * yarr/RegexJIT.h:
            (JSC::Yarr::RegexCodeBlock::RegexCodeBlock):
            (JSC::Yarr::RegexCodeBlock::pcreFallback):
            (JSC::Yarr::RegexCodeBlock::setFallback):
            (JSC::Yarr::RegexCodeBlock::operator bool):
            (JSC::Yarr::RegexCodeBlock::set):
            (JSC::Yarr::RegexCodeBlock::execute):
    
    
    
    git-svn-id: http://svn.webkit.org/repository/webkit/trunk@44030 268f45cc-cd09-0410-ab3c-d52691b4dbfc
    04b762d2