| /* | 
 |  * Copyright (C) 2008-2021 Apple Inc. All rights reserved. | 
 |  * | 
 |  * Redistribution and use in source and binary forms, with or without | 
 |  * modification, are permitted provided that the following conditions | 
 |  * are met: | 
 |  * 1. Redistributions of source code must retain the above copyright | 
 |  *    notice, this list of conditions and the following disclaimer. | 
 |  * 2. Redistributions in binary form must reproduce the above copyright | 
 |  *    notice, this list of conditions and the following disclaimer in the | 
 |  *    documentation and/or other materials provided with the distribution. | 
 |  * | 
 |  * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY | 
 |  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | 
 |  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | 
 |  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL APPLE INC. OR | 
 |  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, | 
 |  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | 
 |  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR | 
 |  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY | 
 |  * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | 
 |  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | 
 |  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.  | 
 |  */ | 
 |  | 
 | #pragma once | 
 |  | 
 | #if ENABLE(JIT) | 
 |  | 
 | // We've run into some problems where changing the size of the class JIT leads to | 
 | // performance fluctuations. Try forcing alignment in an attempt to stabilize this. | 
 | #if COMPILER(GCC_COMPATIBLE) | 
 | #define JIT_CLASS_ALIGNMENT alignas(32) | 
 | #else | 
 | #define JIT_CLASS_ALIGNMENT | 
 | #endif | 
 |  | 
 | #define ASSERT_JIT_OFFSET(actual, expected) ASSERT_WITH_MESSAGE(actual == expected, "JIT Offset \"%s\" should be %d, not %d.\n", #expected, static_cast<int>(expected), static_cast<int>(actual)); | 
 |  | 
 | #include "ByValInfo.h" | 
 | #include "CodeBlock.h" | 
 | #include "CommonSlowPaths.h" | 
 | #include "JITDisassembler.h" | 
 | #include "JITInlineCacheGenerator.h" | 
 | #include "JITMathIC.h" | 
 | #include "JITRightShiftGenerator.h" | 
 | #include "JSInterfaceJIT.h" | 
 | #include "PCToCodeOriginMap.h" | 
 | #include "UnusedPointer.h" | 
 | #include <wtf/UniqueRef.h> | 
 |  | 
 | namespace JSC { | 
 |  | 
 |     enum OpcodeID : unsigned; | 
 |  | 
 |     class ArrayAllocationProfile; | 
 |     class CallLinkInfo; | 
 |     class CodeBlock; | 
 |     class FunctionExecutable; | 
 |     class JIT; | 
 |     class Identifier; | 
 |     class Interpreter; | 
 |     class BlockDirectory; | 
 |     class Register; | 
 |     class StructureChain; | 
 |     class StructureStubInfo; | 
 |  | 
 |     struct Instruction; | 
 |     struct OperandTypes; | 
 |     struct SimpleJumpTable; | 
 |     struct StringJumpTable; | 
 |  | 
 |     struct OpPutByVal; | 
 |     struct OpPutByValDirect; | 
 |     struct OpPutPrivateName; | 
 |     struct OpPutToScope; | 
 |  | 
 |     template<PtrTag tag> | 
 |     struct CallRecord { | 
 |         MacroAssembler::Call from; | 
 |         FunctionPtr<tag> callee; | 
 |  | 
 |         CallRecord() | 
 |         { | 
 |         } | 
 |  | 
 |         CallRecord(MacroAssembler::Call from, FunctionPtr<tag> callee) | 
 |             : from(from) | 
 |             , callee(callee) | 
 |         { | 
 |         } | 
 |     }; | 
 |  | 
 |     using FarCallRecord = CallRecord<OperationPtrTag>; | 
 |     using NearCallRecord = CallRecord<JSInternalPtrTag>; | 
 |  | 
 |     struct NearJumpRecord { | 
 |         MacroAssembler::Jump from; | 
 |         CodeLocationLabel<JITThunkPtrTag> target; | 
 |  | 
 |         NearJumpRecord() = default; | 
 |         NearJumpRecord(MacroAssembler::Jump from, CodeLocationLabel<JITThunkPtrTag> target) | 
 |             : from(from) | 
 |             , target(target) | 
 |         { } | 
 |     }; | 
 |  | 
 |     struct JumpTable { | 
 |         MacroAssembler::Jump from; | 
 |         unsigned toBytecodeOffset; | 
 |  | 
 |         JumpTable(MacroAssembler::Jump f, unsigned t) | 
 |             : from(f) | 
 |             , toBytecodeOffset(t) | 
 |         { | 
 |         } | 
 |     }; | 
 |  | 
 |     struct SlowCaseEntry { | 
 |         MacroAssembler::Jump from; | 
 |         BytecodeIndex to; | 
 |          | 
 |         SlowCaseEntry(MacroAssembler::Jump f, BytecodeIndex t) | 
 |             : from(f) | 
 |             , to(t) | 
 |         { | 
 |         } | 
 |     }; | 
 |  | 
 |     struct SwitchRecord { | 
 |         enum Type { | 
 |             Immediate, | 
 |             Character, | 
 |             String | 
 |         }; | 
 |  | 
 |         Type type; | 
 |  | 
 |         BytecodeIndex bytecodeIndex; | 
 |         unsigned defaultOffset; | 
 |         unsigned tableIndex; | 
 |  | 
 |         SwitchRecord(unsigned tableIndex, BytecodeIndex bytecodeIndex, unsigned defaultOffset, Type type) | 
 |             : type(type) | 
 |             , bytecodeIndex(bytecodeIndex) | 
 |             , defaultOffset(defaultOffset) | 
 |             , tableIndex(tableIndex) | 
 |         { | 
 |         } | 
 |     }; | 
 |  | 
 |     struct ByValCompilationInfo { | 
 |         ByValCompilationInfo() { } | 
 |          | 
 |         ByValCompilationInfo(ByValInfo* byValInfo, BytecodeIndex bytecodeIndex, MacroAssembler::PatchableJump notIndexJump, MacroAssembler::PatchableJump badTypeJump, JITArrayMode arrayMode, ArrayProfile* arrayProfile, MacroAssembler::Label doneTarget, MacroAssembler::Label nextHotPathTarget) | 
 |             : byValInfo(byValInfo) | 
 |             , bytecodeIndex(bytecodeIndex) | 
 |             , notIndexJump(notIndexJump) | 
 |             , badTypeJump(badTypeJump) | 
 |             , arrayMode(arrayMode) | 
 |             , arrayProfile(arrayProfile) | 
 |             , doneTarget(doneTarget) | 
 |             , nextHotPathTarget(nextHotPathTarget) | 
 |         { | 
 |         } | 
 |  | 
 |         ByValCompilationInfo(ByValInfo* byValInfo, BytecodeIndex bytecodeIndex, MacroAssembler::PatchableJump notIndexJump, MacroAssembler::Label doneTarget, MacroAssembler::Label nextHotPathTarget) | 
 |             : byValInfo(byValInfo) | 
 |             , bytecodeIndex(bytecodeIndex) | 
 |             , notIndexJump(notIndexJump) | 
 |             , doneTarget(doneTarget) | 
 |             , nextHotPathTarget(nextHotPathTarget) | 
 |         { | 
 |         } | 
 |  | 
 |         ByValInfo* byValInfo; | 
 |         BytecodeIndex bytecodeIndex; | 
 |         MacroAssembler::PatchableJump notIndexJump; | 
 |         MacroAssembler::PatchableJump badTypeJump; | 
 |         JITArrayMode arrayMode; | 
 |         ArrayProfile* arrayProfile; | 
 |         MacroAssembler::Label doneTarget; | 
 |         MacroAssembler::Label nextHotPathTarget; | 
 |         MacroAssembler::Label slowPathTarget; | 
 |         MacroAssembler::Call returnAddress; | 
 |     }; | 
 |  | 
 |     struct CallCompilationInfo { | 
 |         MacroAssembler::Label slowPathStart; | 
 |         MacroAssembler::Label doneLocation; | 
 |         CallLinkInfo* callLinkInfo; | 
 |     }; | 
 |  | 
 |     void ctiPatchCallByReturnAddress(ReturnAddressPtr, FunctionPtr<CFunctionPtrTag> newCalleeFunction); | 
 |  | 
 |     class JIT_CLASS_ALIGNMENT JIT : private JSInterfaceJIT { | 
 |         friend class JITSlowPathCall; | 
 |         friend class JITStubCall; | 
 |         friend class JITThunks; | 
 |  | 
 |         using MacroAssembler::Jump; | 
 |         using MacroAssembler::JumpList; | 
 |         using MacroAssembler::Label; | 
 |  | 
 |         static constexpr uintptr_t patchGetByIdDefaultStructure = unusedPointer; | 
 |         static constexpr int patchGetByIdDefaultOffset = 0; | 
 |         // Magic number - initial offset cannot be representable as a signed 8bit value, or the X86Assembler | 
 |         // will compress the displacement, and we may not be able to fit a patched offset. | 
 |         static constexpr int patchPutByIdDefaultOffset = 256; | 
 |  | 
 |     public: | 
 |         JIT(VM&, CodeBlock* = nullptr, BytecodeIndex loopOSREntryBytecodeOffset = BytecodeIndex(0)); | 
 |         ~JIT(); | 
 |  | 
 |         VM& vm() { return *JSInterfaceJIT::vm(); } | 
 |  | 
 |         void compileAndLinkWithoutFinalizing(JITCompilationEffort); | 
 |         CompilationResult finalizeOnMainThread(); | 
 |         size_t codeSize() const; | 
 |  | 
 |         void doMainThreadPreparationBeforeCompile(); | 
 |          | 
 |         static CompilationResult compile(VM& vm, CodeBlock* codeBlock, JITCompilationEffort effort, BytecodeIndex bytecodeOffset = BytecodeIndex(0)) | 
 |         { | 
 |             return JIT(vm, codeBlock, bytecodeOffset).privateCompile(effort); | 
 |         } | 
 |          | 
 |         static void compilePutByVal(const ConcurrentJSLocker& locker, VM& vm, CodeBlock* codeBlock, ByValInfo* byValInfo, ReturnAddressPtr returnAddress, JITArrayMode arrayMode) | 
 |         { | 
 |             JIT jit(vm, codeBlock); | 
 |             jit.m_bytecodeIndex = byValInfo->bytecodeIndex; | 
 |             jit.privateCompilePutByVal<OpPutByVal>(locker, byValInfo, returnAddress, arrayMode); | 
 |         } | 
 |          | 
 |         static void compileDirectPutByVal(const ConcurrentJSLocker& locker, VM& vm, CodeBlock* codeBlock, ByValInfo* byValInfo, ReturnAddressPtr returnAddress, JITArrayMode arrayMode) | 
 |         { | 
 |             JIT jit(vm, codeBlock); | 
 |             jit.m_bytecodeIndex = byValInfo->bytecodeIndex; | 
 |             jit.privateCompilePutByVal<OpPutByValDirect>(locker, byValInfo, returnAddress, arrayMode); | 
 |         } | 
 |  | 
 |         template<typename Op> | 
 |         static void compilePutByValWithCachedId(VM& vm, CodeBlock* codeBlock, ByValInfo* byValInfo, ReturnAddressPtr returnAddress, PutKind putKind, CacheableIdentifier propertyName) | 
 |         { | 
 |             JIT jit(vm, codeBlock); | 
 |             jit.m_bytecodeIndex = byValInfo->bytecodeIndex; | 
 |             jit.privateCompilePutByValWithCachedId<Op>(byValInfo, returnAddress, putKind, propertyName); | 
 |         } | 
 |  | 
 |         static void compilePutPrivateNameWithCachedId(VM& vm, CodeBlock* codeBlock, ByValInfo* byValInfo, ReturnAddressPtr returnAddress, CacheableIdentifier propertyName) | 
 |         { | 
 |             JIT jit(vm, codeBlock); | 
 |             jit.m_bytecodeIndex = byValInfo->bytecodeIndex; | 
 |             jit.privateCompilePutPrivateNameWithCachedId(byValInfo, returnAddress, propertyName); | 
 |         } | 
 |  | 
 |         static void compileHasIndexedProperty(VM& vm, CodeBlock* codeBlock, ByValInfo* byValInfo, ReturnAddressPtr returnAddress, JITArrayMode arrayMode) | 
 |         { | 
 |             JIT jit(vm, codeBlock); | 
 |             jit.m_bytecodeIndex = byValInfo->bytecodeIndex; | 
 |             jit.privateCompileHasIndexedProperty(byValInfo, returnAddress, arrayMode); | 
 |         } | 
 |  | 
 |         static unsigned frameRegisterCountFor(CodeBlock*); | 
 |         static int stackPointerOffsetFor(CodeBlock*); | 
 |  | 
 |         JS_EXPORT_PRIVATE static HashMap<CString, Seconds> compileTimeStats(); | 
 |         JS_EXPORT_PRIVATE static Seconds totalCompileTime(); | 
 |  | 
 |     private: | 
 |         void privateCompileMainPass(); | 
 |         void privateCompileLinkPass(); | 
 |         void privateCompileSlowCases(); | 
 |         void link(); | 
 |         CompilationResult privateCompile(JITCompilationEffort); | 
 |          | 
 |         void privateCompileGetByVal(const ConcurrentJSLocker&, ByValInfo*, ReturnAddressPtr, JITArrayMode); | 
 |         template<typename Op> | 
 |         void privateCompilePutByVal(const ConcurrentJSLocker&, ByValInfo*, ReturnAddressPtr, JITArrayMode); | 
 |         template<typename Op> | 
 |         void privateCompilePutByValWithCachedId(ByValInfo*, ReturnAddressPtr, PutKind, CacheableIdentifier); | 
 |  | 
 |         void privateCompilePutPrivateNameWithCachedId(ByValInfo*, ReturnAddressPtr, CacheableIdentifier); | 
 |  | 
 |         void privateCompileHasIndexedProperty(ByValInfo*, ReturnAddressPtr, JITArrayMode); | 
 |  | 
 |         void privateCompilePatchGetArrayLength(ReturnAddressPtr returnAddress); | 
 |  | 
 |         // Add a call out from JIT code, without an exception check. | 
 |         Call appendCall(const FunctionPtr<CFunctionPtrTag> function) | 
 |         { | 
 |             Call functionCall = call(OperationPtrTag); | 
 |             m_farCalls.append(FarCallRecord(functionCall, function.retagged<OperationPtrTag>())); | 
 |             return functionCall; | 
 |         } | 
 |  | 
 |         void appendCall(Address function) | 
 |         { | 
 |             call(function, OperationPtrTag); | 
 |         } | 
 |  | 
 | #if OS(WINDOWS) && CPU(X86_64) | 
 |         Call appendCallWithSlowPathReturnType(const FunctionPtr<CFunctionPtrTag> function) | 
 |         { | 
 |             Call functionCall = callWithSlowPathReturnType(OperationPtrTag); | 
 |             m_farCalls.append(FarCallRecord(functionCall, function.retagged<OperationPtrTag>())); | 
 |             return functionCall; | 
 |         } | 
 | #endif | 
 |  | 
 |         void exceptionCheck(Jump jumpToHandler) | 
 |         { | 
 |             m_exceptionChecks.append(jumpToHandler); | 
 |         } | 
 |  | 
 |         void exceptionCheck() | 
 |         { | 
 |             m_exceptionChecks.append(emitExceptionCheck(vm())); | 
 |         } | 
 |  | 
 |         void exceptionCheckWithCallFrameRollback() | 
 |         { | 
 |             m_exceptionChecksWithCallFrameRollback.append(emitExceptionCheck(vm())); | 
 |         } | 
 |  | 
 |         void privateCompileExceptionHandlers(); | 
 |  | 
 |         void advanceToNextCheckpoint(); | 
 |         void emitJumpSlowToHotForCheckpoint(Jump); | 
 |  | 
 |         void addSlowCase(Jump); | 
 |         void addSlowCase(const JumpList&); | 
 |         void addSlowCase(); | 
 |         void addJump(Jump, int); | 
 |         void addJump(const JumpList&, int); | 
 |         void emitJumpSlowToHot(Jump, int); | 
 |  | 
 |         template<typename Op> | 
 |         void compileOpCall(const Instruction*, unsigned callLinkInfoIndex); | 
 |         template<typename Op> | 
 |         void compileOpCallSlowCase(const Instruction*, Vector<SlowCaseEntry>::iterator&, unsigned callLinkInfoIndex); | 
 |         template<typename Op> | 
 |         std::enable_if_t< | 
 |             Op::opcodeID != op_call_varargs && Op::opcodeID != op_construct_varargs | 
 |             && Op::opcodeID != op_tail_call_varargs && Op::opcodeID != op_tail_call_forward_arguments | 
 |         , void> compileSetupFrame(const Op&, CallLinkInfo*); | 
 |  | 
 |         template<typename Op> | 
 |         std::enable_if_t< | 
 |             Op::opcodeID == op_call_varargs || Op::opcodeID == op_construct_varargs | 
 |             || Op::opcodeID == op_tail_call_varargs || Op::opcodeID == op_tail_call_forward_arguments | 
 |         , void> compileSetupFrame(const Op&, CallLinkInfo*); | 
 |  | 
 |         template<typename Op> | 
 |         bool compileTailCall(const Op&, CallLinkInfo*, unsigned callLinkInfoIndex); | 
 |         template<typename Op> | 
 |         bool compileCallEval(const Op&); | 
 |         void compileCallEvalSlowCase(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         template<typename Op> | 
 |         void emitPutCallResult(const Op&); | 
 |  | 
 |         template<typename Op> void compileOpStrictEq(const Instruction*); | 
 |         template<typename Op> void compileOpStrictEqJump(const Instruction*); | 
 |         enum class CompileOpEqType { Eq, NEq }; | 
 |         void compileOpEqJumpSlow(Vector<SlowCaseEntry>::iterator&, CompileOpEqType, int jumpTarget); | 
 |         bool isOperandConstantDouble(VirtualRegister); | 
 |          | 
 |         enum WriteBarrierMode { UnconditionalWriteBarrier, ShouldFilterBase, ShouldFilterValue, ShouldFilterBaseAndValue }; | 
 |         // value register in write barrier is used before any scratch registers | 
 |         // so may safely be the same as either of the scratch registers. | 
 |         void emitWriteBarrier(VirtualRegister owner, WriteBarrierMode); | 
 |         void emitWriteBarrier(VirtualRegister owner, VirtualRegister value, WriteBarrierMode); | 
 |         void emitWriteBarrier(JSCell* owner, VirtualRegister value, WriteBarrierMode); | 
 |         void emitWriteBarrier(JSCell* owner); | 
 |  | 
 |         // This assumes that the value to profile is in regT0 and that regT3 is available for | 
 |         // scratch. | 
 | #if USE(JSVALUE64) | 
 |         void emitValueProfilingSite(ValueProfile&, GPRReg); | 
 |         void emitValueProfilingSite(ValueProfile&, JSValueRegs); | 
 |         template<typename Metadata> void emitValueProfilingSite(Metadata&, GPRReg); | 
 |         template<typename Metadata> void emitValueProfilingSite(Metadata&, JSValueRegs); | 
 | #else | 
 |         void emitValueProfilingSite(ValueProfile&, JSValueRegs); | 
 |         template<typename Metadata> void emitValueProfilingSite(Metadata&, JSValueRegs); | 
 | #endif | 
 |  | 
 |         void emitValueProfilingSiteIfProfiledOpcode(...); | 
 |         template<typename Op> | 
 |         std::enable_if_t<std::is_same<decltype(Op::Metadata::m_profile), ValueProfile>::value, void> | 
 |         emitValueProfilingSiteIfProfiledOpcode(Op bytecode); | 
 |  | 
 |         void emitArrayProfilingSiteWithCell(RegisterID cell, RegisterID indexingType, ArrayProfile*); | 
 |         void emitArrayProfileStoreToHoleSpecialCase(ArrayProfile*); | 
 |         void emitArrayProfileOutOfBoundsSpecialCase(ArrayProfile*); | 
 |          | 
 |         JITArrayMode chooseArrayMode(ArrayProfile*); | 
 |          | 
 |         // Property is in regT1, base is in regT0. regT2 contains indexing type. | 
 |         // Property is int-checked and zero extended. Base is cell checked. | 
 |         // Structure is already profiled. Returns the slow cases. Fall-through | 
 |         // case contains result in regT0, and it is not yet profiled. | 
 |         JumpList emitInt32Load(const Instruction* instruction, PatchableJump& badType, ByValInfo* byValInfo) { return emitContiguousLoad(instruction, badType, byValInfo, Int32Shape); } | 
 |         JumpList emitDoubleLoad(const Instruction*, PatchableJump& badType, ByValInfo*); | 
 |         JumpList emitContiguousLoad(const Instruction*, PatchableJump& badType, ByValInfo*, IndexingType expectedShape = ContiguousShape); | 
 |         JumpList emitArrayStorageLoad(const Instruction*, PatchableJump& badType, ByValInfo*); | 
 |         JumpList emitLoadForArrayMode(const Instruction*, JITArrayMode, PatchableJump& badType, ByValInfo*); | 
 |  | 
 |         // Property is in regT1, base is in regT0. regT2 contains indecing type. | 
 |         // The value to store is not yet loaded. Property is int-checked and | 
 |         // zero-extended. Base is cell checked. Structure is already profiled. | 
 |         // returns the slow cases. | 
 |         template<typename Op> | 
 |         JumpList emitInt32PutByVal(Op bytecode, PatchableJump& badType, ByValInfo* byValInfo) | 
 |         { | 
 |             return emitGenericContiguousPutByVal(bytecode, badType, byValInfo, Int32Shape); | 
 |         } | 
 |         template<typename Op> | 
 |         JumpList emitDoublePutByVal(Op bytecode, PatchableJump& badType, ByValInfo* byValInfo) | 
 |         { | 
 |             return emitGenericContiguousPutByVal(bytecode, badType, byValInfo, DoubleShape); | 
 |         } | 
 |         template<typename Op> | 
 |         JumpList emitContiguousPutByVal(Op bytecode, PatchableJump& badType, ByValInfo* byValInfo) | 
 |         { | 
 |             return emitGenericContiguousPutByVal(bytecode, badType, byValInfo); | 
 |         } | 
 |         template<typename Op> | 
 |         JumpList emitGenericContiguousPutByVal(Op, PatchableJump& badType, ByValInfo*, IndexingType indexingShape = ContiguousShape); | 
 |         template<typename Op> | 
 |         JumpList emitArrayStoragePutByVal(Op, PatchableJump& badType, ByValInfo*); | 
 |         template<typename Op> | 
 |         JumpList emitIntTypedArrayPutByVal(Op, PatchableJump& badType, ByValInfo*, TypedArrayType); | 
 |         template<typename Op> | 
 |         JumpList emitFloatTypedArrayPutByVal(Op, PatchableJump& badType, ByValInfo*, TypedArrayType); | 
 |  | 
 |         template<typename Op> | 
 |         ECMAMode ecmaMode(Op); | 
 |  | 
 |         // Determines the type of private field access for a bytecode. | 
 |         template<typename Op> | 
 |         PrivateFieldPutKind privateFieldPutKind(Op); | 
 |  | 
 |         // Identifier check helper for GetByVal and PutByVal. | 
 |         void emitByValIdentifierCheck(RegisterID cell, RegisterID scratch, CacheableIdentifier, JumpList& slowCases); | 
 |  | 
 |         JITPutByIdGenerator emitPutPrivateNameWithCachedId(OpPutPrivateName, CacheableIdentifier, JumpList& doneCases, JumpList& slowCases); | 
 |  | 
 |         template<typename Op> | 
 |         JITPutByIdGenerator emitPutByValWithCachedId(Op, PutKind, CacheableIdentifier, JumpList& doneCases, JumpList& slowCases); | 
 |  | 
 |         enum FinalObjectMode { MayBeFinal, KnownNotFinal }; | 
 |  | 
 |         void emitGetVirtualRegister(VirtualRegister src, JSValueRegs dst); | 
 |         void emitPutVirtualRegister(VirtualRegister dst, JSValueRegs src); | 
 |  | 
 |         int32_t getOperandConstantInt(VirtualRegister src); | 
 |         double getOperandConstantDouble(VirtualRegister src); | 
 |  | 
 | #if USE(JSVALUE32_64) | 
 |         bool getOperandConstantInt(VirtualRegister op1, VirtualRegister op2, VirtualRegister& op, int32_t& constant); | 
 |  | 
 |         void emitLoadDouble(VirtualRegister, FPRegisterID value); | 
 |         void emitLoadTag(VirtualRegister, RegisterID tag); | 
 |         void emitLoadPayload(VirtualRegister, RegisterID payload); | 
 |  | 
 |         void emitLoad(const JSValue& v, RegisterID tag, RegisterID payload); | 
 |         void emitLoad(VirtualRegister, RegisterID tag, RegisterID payload, RegisterID base = callFrameRegister); | 
 |         void emitLoad2(VirtualRegister, RegisterID tag1, RegisterID payload1, VirtualRegister, RegisterID tag2, RegisterID payload2); | 
 |  | 
 |         void emitStore(VirtualRegister, RegisterID tag, RegisterID payload, RegisterID base = callFrameRegister); | 
 |         void emitStore(VirtualRegister, const JSValue constant, RegisterID base = callFrameRegister); | 
 |         void emitStoreInt32(VirtualRegister, RegisterID payload, bool indexIsInt32 = false); | 
 |         void emitStoreInt32(VirtualRegister, TrustedImm32 payload, bool indexIsInt32 = false); | 
 |         void emitStoreCell(VirtualRegister, RegisterID payload, bool indexIsCell = false); | 
 |         void emitStoreBool(VirtualRegister, RegisterID payload, bool indexIsBool = false); | 
 |         void emitStoreDouble(VirtualRegister, FPRegisterID value); | 
 |  | 
 |         void emitJumpSlowCaseIfNotJSCell(VirtualRegister); | 
 |         void emitJumpSlowCaseIfNotJSCell(VirtualRegister, RegisterID tag); | 
 |         void emitJumpSlowCaseIfNotJSCell(RegisterID); | 
 |  | 
 |         void compileGetByIdHotPath(const Identifier*); | 
 |  | 
 |         // Arithmetic opcode helpers | 
 |         template <typename Op> | 
 |         void emitBinaryDoubleOp(const Instruction *, OperandTypes, JumpList& notInt32Op1, JumpList& notInt32Op2, bool op1IsInRegisters = true, bool op2IsInRegisters = true); | 
 |  | 
 | #else // USE(JSVALUE32_64) | 
 |         void emitGetVirtualRegister(VirtualRegister src, RegisterID dst); | 
 |         void emitGetVirtualRegisters(VirtualRegister src1, RegisterID dst1, VirtualRegister src2, RegisterID dst2); | 
 |         void emitPutVirtualRegister(VirtualRegister dst, RegisterID from = regT0); | 
 |         void emitStoreCell(VirtualRegister dst, RegisterID payload, bool /* only used in JSValue32_64 */ = false) | 
 |         { | 
 |             emitPutVirtualRegister(dst, payload); | 
 |         } | 
 |  | 
 |         Jump emitJumpIfBothJSCells(RegisterID, RegisterID, RegisterID); | 
 |         void emitJumpSlowCaseIfJSCell(RegisterID); | 
 |         void emitJumpSlowCaseIfNotJSCell(RegisterID); | 
 |         void emitJumpSlowCaseIfNotJSCell(RegisterID, VirtualRegister); | 
 |         Jump emitJumpIfNotInt(RegisterID, RegisterID, RegisterID scratch); | 
 |         PatchableJump emitPatchableJumpIfNotInt(RegisterID); | 
 |         void emitJumpSlowCaseIfNotInt(RegisterID); | 
 |         void emitJumpSlowCaseIfNotNumber(RegisterID); | 
 |         void emitJumpSlowCaseIfNotInt(RegisterID, RegisterID, RegisterID scratch); | 
 |  | 
 |         void compileGetByIdHotPath(VirtualRegister baseReg, const Identifier*); | 
 |  | 
 | #endif // USE(JSVALUE32_64) | 
 |  | 
 |         template<typename Op> | 
 |         void emit_compareAndJump(const Instruction*, RelationalCondition); | 
 |         void emit_compareAndJumpImpl(VirtualRegister op1, VirtualRegister op2, unsigned target, RelationalCondition); | 
 |         template<typename Op> | 
 |         void emit_compareUnsigned(const Instruction*, RelationalCondition); | 
 |         void emit_compareUnsignedImpl(VirtualRegister dst, VirtualRegister op1, VirtualRegister op2, RelationalCondition); | 
 |         template<typename Op> | 
 |         void emit_compareUnsignedAndJump(const Instruction*, RelationalCondition); | 
 |         void emit_compareUnsignedAndJumpImpl(VirtualRegister op1, VirtualRegister op2, unsigned target, RelationalCondition); | 
 |         template<typename Op, typename SlowOperation> | 
 |         void emit_compareAndJumpSlow(const Instruction*, DoubleCondition, SlowOperation, bool invert, Vector<SlowCaseEntry>::iterator&); | 
 |         template<typename SlowOperation> | 
 |         void emit_compareAndJumpSlowImpl(VirtualRegister op1, VirtualRegister op2, unsigned target, size_t instructionSize, DoubleCondition, SlowOperation, bool invert, Vector<SlowCaseEntry>::iterator&); | 
 |          | 
 |         void assertStackPointerOffset(); | 
 |  | 
 |         void emit_op_add(const Instruction*); | 
 |         void emit_op_bitand(const Instruction*); | 
 |         void emit_op_bitor(const Instruction*); | 
 |         void emit_op_bitxor(const Instruction*); | 
 |         void emit_op_bitnot(const Instruction*); | 
 |         void emit_op_call(const Instruction*); | 
 |         void emit_op_tail_call(const Instruction*); | 
 |         void emit_op_call_eval(const Instruction*); | 
 |         void emit_op_call_varargs(const Instruction*); | 
 |         void emit_op_tail_call_varargs(const Instruction*); | 
 |         void emit_op_tail_call_forward_arguments(const Instruction*); | 
 |         void emit_op_construct_varargs(const Instruction*); | 
 |         void emit_op_catch(const Instruction*); | 
 |         void emit_op_construct(const Instruction*); | 
 |         void emit_op_create_this(const Instruction*); | 
 |         void emit_op_to_this(const Instruction*); | 
 |         void emit_op_get_argument(const Instruction*); | 
 |         void emit_op_argument_count(const Instruction*); | 
 |         void emit_op_get_rest_length(const Instruction*); | 
 |         void emit_op_check_tdz(const Instruction*); | 
 |         void emit_op_identity_with_profile(const Instruction*); | 
 |         void emit_op_debug(const Instruction*); | 
 |         void emit_op_del_by_id(const Instruction*); | 
 |         void emitSlow_op_del_by_id(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emit_op_del_by_val(const Instruction*); | 
 |         void emitSlow_op_del_by_val(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emit_op_div(const Instruction*); | 
 |         void emit_op_end(const Instruction*); | 
 |         void emit_op_enter(const Instruction*); | 
 |         void emit_op_get_scope(const Instruction*); | 
 |         void emit_op_eq(const Instruction*); | 
 |         void emit_op_eq_null(const Instruction*); | 
 |         void emit_op_below(const Instruction*); | 
 |         void emit_op_beloweq(const Instruction*); | 
 |         void emit_op_try_get_by_id(const Instruction*); | 
 |         void emit_op_get_by_id(const Instruction*); | 
 |         void emit_op_get_by_id_with_this(const Instruction*); | 
 |         void emit_op_get_by_id_direct(const Instruction*); | 
 |         void emit_op_get_by_val(const Instruction*); | 
 |         void emit_op_get_private_name(const Instruction*); | 
 |         void emit_op_set_private_brand(const Instruction*); | 
 |         void emit_op_check_private_brand(const Instruction*); | 
 |         void emit_op_get_argument_by_val(const Instruction*); | 
 |         void emit_op_get_prototype_of(const Instruction*); | 
 |         void emit_op_in_by_id(const Instruction*); | 
 |         void emit_op_in_by_val(const Instruction*); | 
 |         void emit_op_has_private_name(const Instruction*); | 
 |         void emit_op_has_private_brand(const Instruction*); | 
 |         void emit_op_init_lazy_reg(const Instruction*); | 
 |         void emit_op_overrides_has_instance(const Instruction*); | 
 |         void emit_op_instanceof(const Instruction*); | 
 |         void emit_op_instanceof_custom(const Instruction*); | 
 |         void emit_op_is_empty(const Instruction*); | 
 |         void emit_op_typeof_is_undefined(const Instruction*); | 
 |         void emit_op_is_undefined_or_null(const Instruction*); | 
 |         void emit_op_is_boolean(const Instruction*); | 
 |         void emit_op_is_number(const Instruction*); | 
 | #if USE(BIGINT32) | 
 |         void emit_op_is_big_int(const Instruction*); | 
 | #else | 
 |         NO_RETURN void emit_op_is_big_int(const Instruction*); | 
 | #endif | 
 |         void emit_op_is_object(const Instruction*); | 
 |         void emit_op_is_cell_with_type(const Instruction*); | 
 |         void emit_op_jeq_null(const Instruction*); | 
 |         void emit_op_jfalse(const Instruction*); | 
 |         void emit_op_jmp(const Instruction*); | 
 |         void emit_op_jneq_null(const Instruction*); | 
 |         void emit_op_jundefined_or_null(const Instruction*); | 
 |         void emit_op_jnundefined_or_null(const Instruction*); | 
 |         void emit_op_jneq_ptr(const Instruction*); | 
 |         void emit_op_jless(const Instruction*); | 
 |         void emit_op_jlesseq(const Instruction*); | 
 |         void emit_op_jgreater(const Instruction*); | 
 |         void emit_op_jgreatereq(const Instruction*); | 
 |         void emit_op_jnless(const Instruction*); | 
 |         void emit_op_jnlesseq(const Instruction*); | 
 |         void emit_op_jngreater(const Instruction*); | 
 |         void emit_op_jngreatereq(const Instruction*); | 
 |         void emit_op_jeq(const Instruction*); | 
 |         void emit_op_jneq(const Instruction*); | 
 |         void emit_op_jstricteq(const Instruction*); | 
 |         void emit_op_jnstricteq(const Instruction*); | 
 |         void emit_op_jbelow(const Instruction*); | 
 |         void emit_op_jbeloweq(const Instruction*); | 
 |         void emit_op_jtrue(const Instruction*); | 
 |         void emit_op_loop_hint(const Instruction*); | 
 |         void emit_op_check_traps(const Instruction*); | 
 |         void emit_op_nop(const Instruction*); | 
 |         void emit_op_super_sampler_begin(const Instruction*); | 
 |         void emit_op_super_sampler_end(const Instruction*); | 
 |         void emit_op_lshift(const Instruction*); | 
 |         void emit_op_mod(const Instruction*); | 
 |         void emit_op_mov(const Instruction*); | 
 |         void emit_op_mul(const Instruction*); | 
 |         void emit_op_negate(const Instruction*); | 
 |         void emit_op_neq(const Instruction*); | 
 |         void emit_op_neq_null(const Instruction*); | 
 |         void emit_op_new_array(const Instruction*); | 
 |         void emit_op_new_array_with_size(const Instruction*); | 
 |         void emit_op_new_func(const Instruction*); | 
 |         void emit_op_new_func_exp(const Instruction*); | 
 |         void emit_op_new_generator_func(const Instruction*); | 
 |         void emit_op_new_generator_func_exp(const Instruction*); | 
 |         void emit_op_new_async_func(const Instruction*); | 
 |         void emit_op_new_async_func_exp(const Instruction*); | 
 |         void emit_op_new_async_generator_func(const Instruction*); | 
 |         void emit_op_new_async_generator_func_exp(const Instruction*); | 
 |         void emit_op_new_object(const Instruction*); | 
 |         void emit_op_new_regexp(const Instruction*); | 
 |         void emit_op_not(const Instruction*); | 
 |         void emit_op_nstricteq(const Instruction*); | 
 |         void emit_op_dec(const Instruction*); | 
 |         void emit_op_inc(const Instruction*); | 
 |         void emit_op_profile_type(const Instruction*); | 
 |         void emit_op_profile_control_flow(const Instruction*); | 
 |         void emit_op_get_parent_scope(const Instruction*); | 
 |         void emit_op_put_by_id(const Instruction*); | 
 |         template<typename Op = OpPutByVal> | 
 |         void emit_op_put_by_val(const Instruction*); | 
 |         void emit_op_put_by_val_direct(const Instruction*); | 
 |         void emit_op_put_private_name(const Instruction*); | 
 |         void emit_op_put_getter_by_id(const Instruction*); | 
 |         void emit_op_put_setter_by_id(const Instruction*); | 
 |         void emit_op_put_getter_setter_by_id(const Instruction*); | 
 |         void emit_op_put_getter_by_val(const Instruction*); | 
 |         void emit_op_put_setter_by_val(const Instruction*); | 
 |         void emit_op_ret(const Instruction*); | 
 |         void emit_op_rshift(const Instruction*); | 
 |         void emit_op_set_function_name(const Instruction*); | 
 |         void emit_op_stricteq(const Instruction*); | 
 |         void emit_op_sub(const Instruction*); | 
 |         void emit_op_switch_char(const Instruction*); | 
 |         void emit_op_switch_imm(const Instruction*); | 
 |         void emit_op_switch_string(const Instruction*); | 
 |         void emit_op_tear_off_arguments(const Instruction*); | 
 |         void emit_op_throw(const Instruction*); | 
 |         void emit_op_to_number(const Instruction*); | 
 |         void emit_op_to_numeric(const Instruction*); | 
 |         void emit_op_to_string(const Instruction*); | 
 |         void emit_op_to_object(const Instruction*); | 
 |         void emit_op_to_primitive(const Instruction*); | 
 |         void emit_op_unexpected_load(const Instruction*); | 
 |         void emit_op_unsigned(const Instruction*); | 
 |         void emit_op_urshift(const Instruction*); | 
 |         template <typename OpCodeType> | 
 |         void emit_op_has_structure_propertyImpl(const Instruction*); | 
 |         void emit_op_has_enumerable_indexed_property(const Instruction*); | 
 |         void emit_op_has_enumerable_structure_property(const Instruction*); | 
 |         void emit_op_has_own_structure_property(const Instruction*); | 
 |         void emit_op_in_structure_property(const Instruction*); | 
 |         void emit_op_get_direct_pname(const Instruction*); | 
 |         void emit_op_enumerator_structure_pname(const Instruction*); | 
 |         void emit_op_enumerator_generic_pname(const Instruction*); | 
 |         void emit_op_get_internal_field(const Instruction*); | 
 |         void emit_op_put_internal_field(const Instruction*); | 
 |         void emit_op_log_shadow_chicken_prologue(const Instruction*); | 
 |         void emit_op_log_shadow_chicken_tail(const Instruction*); | 
 |         void emit_op_to_property_key(const Instruction*); | 
 |  | 
 |         void emitSlow_op_add(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_call(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_tail_call(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_call_eval(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_call_varargs(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_tail_call_varargs(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_tail_call_forward_arguments(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_construct_varargs(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_construct(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_eq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_callee(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_try_get_by_id(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_by_id(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_by_id_with_this(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_by_id_direct(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_by_val(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_private_name(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_set_private_brand(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_check_private_brand(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_get_argument_by_val(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_in_by_id(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_in_by_val(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_has_private_name(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_has_private_brand(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_instanceof(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_instanceof_custom(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jless(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jlesseq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jgreater(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jgreatereq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jnless(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jnlesseq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jngreater(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jngreatereq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jeq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jneq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jstricteq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jnstricteq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_jtrue(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_loop_hint(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_check_traps(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_mod(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_mul(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_negate(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_neq(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_new_object(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_put_by_id(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_put_by_val(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_put_private_name(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_sub(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emitSlow_op_has_enumerable_indexed_property(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |  | 
 |         void emit_op_resolve_scope(const Instruction*); | 
 |         void emit_op_get_from_scope(const Instruction*); | 
 |         void emit_op_put_to_scope(const Instruction*); | 
 |         void emit_op_get_from_arguments(const Instruction*); | 
 |         void emit_op_put_to_arguments(const Instruction*); | 
 | #if !ENABLE(EXTRA_CTI_THUNKS) | 
 |         void emitSlow_op_get_from_scope(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 | #endif | 
 |         void emitSlow_op_put_to_scope(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |  | 
 |         void emitSlowCaseCall(const Instruction*, Vector<SlowCaseEntry>::iterator&, SlowPathFunction); | 
 |  | 
 |         void emit_op_iterator_open(const Instruction*); | 
 |         void emitSlow_op_iterator_open(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |         void emit_op_iterator_next(const Instruction*); | 
 |         void emitSlow_op_iterator_next(const Instruction*, Vector<SlowCaseEntry>::iterator&); | 
 |  | 
 |         void emitRightShift(const Instruction*, bool isUnsigned); | 
 |         void emitRightShiftSlowCase(const Instruction*, Vector<SlowCaseEntry>::iterator&, bool isUnsigned); | 
 |  | 
 |         void emitHasPrivate(VirtualRegister dst, VirtualRegister base, VirtualRegister propertyOrBrand, AccessType); | 
 |         void emitHasPrivateSlow(VirtualRegister dst, AccessType); | 
 |  | 
 |         template<typename Op> | 
 |         void emitNewFuncCommon(const Instruction*); | 
 |         template<typename Op> | 
 |         void emitNewFuncExprCommon(const Instruction*); | 
 |         void emitVarInjectionCheck(bool needsVarInjectionChecks); | 
 |         void emitVarReadOnlyCheck(ResolveType); | 
 |         void emitResolveClosure(VirtualRegister dst, VirtualRegister scope, bool needsVarInjectionChecks, unsigned depth); | 
 |         void emitLoadWithStructureCheck(VirtualRegister scope, Structure** structureSlot); | 
 | #if USE(JSVALUE64) | 
 |         void emitGetVarFromPointer(JSValue* operand, GPRReg); | 
 |         void emitGetVarFromIndirectPointer(JSValue** operand, GPRReg); | 
 | #else | 
 |         void emitGetVarFromIndirectPointer(JSValue** operand, GPRReg tag, GPRReg payload); | 
 |         void emitGetVarFromPointer(JSValue* operand, GPRReg tag, GPRReg payload); | 
 | #endif | 
 |         void emitGetClosureVar(VirtualRegister scope, uintptr_t operand); | 
 |         void emitNotifyWrite(WatchpointSet*); | 
 |         void emitNotifyWrite(GPRReg pointerToSet); | 
 |         void emitPutGlobalVariable(JSValue* operand, VirtualRegister value, WatchpointSet*); | 
 |         void emitPutGlobalVariableIndirect(JSValue** addressOfOperand, VirtualRegister value, WatchpointSet**); | 
 |         void emitPutClosureVar(VirtualRegister scope, uintptr_t operand, VirtualRegister value, WatchpointSet*); | 
 |  | 
 |         void emitInitRegister(VirtualRegister); | 
 |  | 
 |         void emitPutIntToCallFrameHeader(RegisterID from, VirtualRegister); | 
 |  | 
 |         JSValue getConstantOperand(VirtualRegister); | 
 |         bool isOperandConstantInt(VirtualRegister); | 
 |         bool isOperandConstantChar(VirtualRegister); | 
 |  | 
 |         template <typename Op, typename Generator, typename ProfiledFunction, typename NonProfiledFunction> | 
 |         void emitMathICFast(JITUnaryMathIC<Generator>*, const Instruction*, ProfiledFunction, NonProfiledFunction); | 
 |         template <typename Op, typename Generator, typename ProfiledFunction, typename NonProfiledFunction> | 
 |         void emitMathICFast(JITBinaryMathIC<Generator>*, const Instruction*, ProfiledFunction, NonProfiledFunction); | 
 |  | 
 |         template <typename Op, typename Generator, typename ProfiledRepatchFunction, typename ProfiledFunction, typename RepatchFunction> | 
 |         void emitMathICSlow(JITBinaryMathIC<Generator>*, const Instruction*, ProfiledRepatchFunction, ProfiledFunction, RepatchFunction); | 
 |         template <typename Op, typename Generator, typename ProfiledRepatchFunction, typename ProfiledFunction, typename RepatchFunction> | 
 |         void emitMathICSlow(JITUnaryMathIC<Generator>*, const Instruction*, ProfiledRepatchFunction, ProfiledFunction, RepatchFunction); | 
 |  | 
 | #if ENABLE(EXTRA_CTI_THUNKS) | 
 |         // Thunk generators. | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_del_by_id_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_del_by_val_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_get_by_id_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_get_by_id_with_this_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_get_by_val_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_get_from_scopeGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_get_private_name_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_put_by_id_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_put_by_val_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_put_private_name_prepareCallGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_put_to_scopeGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> slow_op_resolve_scopeGenerator(VM&); | 
 |  | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_check_traps_handlerGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_enter_handlerGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_ret_handlerGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_throw_handlerGenerator(VM&); | 
 |  | 
 |         static constexpr bool thunkIsUsedForOpGetFromScope(ResolveType resolveType) | 
 |         { | 
 |             // GlobalVar because it is more efficient to emit inline than use a thunk. | 
 |             // ResolvedClosureVar and ModuleVar because we don't use these types with op_get_from_scope. | 
 |             return !(resolveType == GlobalVar || resolveType == ResolvedClosureVar || resolveType == ModuleVar); | 
 |         } | 
 |  | 
 | #define DECLARE_GET_FROM_SCOPE_GENERATOR(resolveType) \ | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_get_from_scope_##resolveType##Generator(VM&); | 
 |         FOR_EACH_RESOLVE_TYPE(DECLARE_GET_FROM_SCOPE_GENERATOR) | 
 | #undef DECLARE_GET_FROM_SCOPE_GENERATOR | 
 |  | 
 |         MacroAssemblerCodeRef<JITThunkPtrTag> generateOpGetFromScopeThunk(ResolveType, const char* thunkName); | 
 |  | 
 |         static constexpr bool thunkIsUsedForOpResolveScope(ResolveType resolveType) | 
 |         { | 
 |             // ModuleVar because it is more efficient to emit inline than use a thunk. | 
 |             // ResolvedClosureVar because we don't use these types with op_resolve_scope. | 
 |             return !(resolveType == ResolvedClosureVar || resolveType == ModuleVar); | 
 |         } | 
 |  | 
 | #define DECLARE_RESOLVE_SCOPE_GENERATOR(resolveType) \ | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> op_resolve_scope_##resolveType##Generator(VM&); | 
 |         FOR_EACH_RESOLVE_TYPE(DECLARE_RESOLVE_SCOPE_GENERATOR) | 
 | #undef DECLARE_RESOLVE_SCOPE_GENERATOR | 
 |  | 
 |         MacroAssemblerCodeRef<JITThunkPtrTag> generateOpResolveScopeThunk(ResolveType, const char* thunkName); | 
 |  | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> valueIsFalseyGenerator(VM&); | 
 |         static MacroAssemblerCodeRef<JITThunkPtrTag> valueIsTruthyGenerator(VM&); | 
 |  | 
 | #endif // ENABLE(EXTRA_CTI_THUNKS) | 
 |  | 
 |         Jump getSlowCase(Vector<SlowCaseEntry>::iterator& iter) | 
 |         { | 
 |             return iter++->from; | 
 |         } | 
 |         void linkSlowCase(Vector<SlowCaseEntry>::iterator& iter) | 
 |         { | 
 |             if (iter->from.isSet()) | 
 |                 iter->from.link(this); | 
 |             ++iter; | 
 |         } | 
 |         void linkDummySlowCase(Vector<SlowCaseEntry>::iterator& iter) | 
 |         { | 
 |             ASSERT(!iter->from.isSet()); | 
 |             ++iter; | 
 |         } | 
 |         void linkSlowCaseIfNotJSCell(Vector<SlowCaseEntry>::iterator&, VirtualRegister); | 
 |         void linkAllSlowCasesForBytecodeIndex(Vector<SlowCaseEntry>& slowCases, | 
 |             Vector<SlowCaseEntry>::iterator&, BytecodeIndex bytecodeOffset); | 
 |  | 
 |         void linkAllSlowCases(Vector<SlowCaseEntry>::iterator& iter) | 
 |         { | 
 |             linkAllSlowCasesForBytecodeIndex(m_slowCases, iter, m_bytecodeIndex); | 
 |         } | 
 |  | 
 |         bool hasAnySlowCases(Vector<SlowCaseEntry>& slowCases, Vector<SlowCaseEntry>::iterator&, BytecodeIndex bytecodeOffset); | 
 |         bool hasAnySlowCases(Vector<SlowCaseEntry>::iterator& iter) | 
 |         { | 
 |             return hasAnySlowCases(m_slowCases, iter, m_bytecodeIndex); | 
 |         } | 
 |  | 
 |         MacroAssembler::Call appendCallWithExceptionCheck(const FunctionPtr<CFunctionPtrTag>); | 
 |         void appendCallWithExceptionCheck(Address); | 
 | #if OS(WINDOWS) && CPU(X86_64) | 
 |         MacroAssembler::Call appendCallWithExceptionCheckAndSlowPathReturnType(const FunctionPtr<CFunctionPtrTag>); | 
 | #endif | 
 |         MacroAssembler::Call appendCallWithCallFrameRollbackOnException(const FunctionPtr<CFunctionPtrTag>); | 
 |         MacroAssembler::Call appendCallWithExceptionCheckSetJSValueResult(const FunctionPtr<CFunctionPtrTag>, VirtualRegister result); | 
 |         void appendCallWithExceptionCheckSetJSValueResult(Address, VirtualRegister result); | 
 |         template<typename Metadata> | 
 |         MacroAssembler::Call appendCallWithExceptionCheckSetJSValueResultWithProfile(Metadata&, const FunctionPtr<CFunctionPtrTag>, VirtualRegister result); | 
 |         template<typename Metadata> | 
 |         void appendCallWithExceptionCheckSetJSValueResultWithProfile(Metadata&, Address, VirtualRegister result); | 
 |          | 
 |         template<typename OperationType, typename... Args> | 
 |         std::enable_if_t<FunctionTraits<OperationType>::hasResult, MacroAssembler::Call> | 
 |         callOperation(OperationType operation, VirtualRegister result, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             return appendCallWithExceptionCheckSetJSValueResult(operation, result); | 
 |         } | 
 |  | 
 |         template<typename OperationType, typename... Args> | 
 |         std::enable_if_t<FunctionTraits<OperationType>::hasResult, void> | 
 |         callOperation(Address target, VirtualRegister result, Args... args) | 
 |         { | 
 |             setupArgumentsForIndirectCall<OperationType>(target, args...); | 
 |             return appendCallWithExceptionCheckSetJSValueResult(Address(GPRInfo::nonArgGPR0, target.offset), result); | 
 |         } | 
 |  | 
 | #if OS(WINDOWS) && CPU(X86_64) | 
 |         template<typename Type> | 
 |         struct is64BitType { | 
 |             static constexpr bool value = sizeof(Type) <= 8; | 
 |         }; | 
 |  | 
 |         template<> | 
 |         struct is64BitType<void> { | 
 |             static constexpr bool value = true; | 
 |         }; | 
 |  | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperation(OperationType operation, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             // x64 Windows cannot use standard call when the return type is larger than 64 bits. | 
 |             if constexpr (is64BitType<typename FunctionTraits<OperationType>::ResultType>::value) | 
 |                 return appendCallWithExceptionCheck(operation); | 
 |             return appendCallWithExceptionCheckAndSlowPathReturnType(operation); | 
 |         } | 
 | #else // OS(WINDOWS) && CPU(X86_64) | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperation(OperationType operation, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             return appendCallWithExceptionCheck(operation); | 
 |         } | 
 | #endif // OS(WINDOWS) && CPU(X86_64) | 
 |  | 
 |         template<typename OperationType, typename... Args> | 
 |         void callOperation(Address target, Args... args) | 
 |         { | 
 | #if OS(WINDOWS) && CPU(X86_64) | 
 |             // x64 Windows cannot use standard call when the return type is larger than 64 bits. | 
 |             static_assert(is64BitType<typename FunctionTraits<OperationType>::ResultType>::value); | 
 | #endif | 
 |             setupArgumentsForIndirectCall<OperationType>(target, args...); | 
 |             appendCallWithExceptionCheck(Address(GPRInfo::nonArgGPR0, target.offset)); | 
 |         } | 
 |  | 
 |         template<typename Metadata, typename OperationType, typename... Args> | 
 |         std::enable_if_t<FunctionTraits<OperationType>::hasResult, MacroAssembler::Call> | 
 |         callOperationWithProfile(Metadata& metadata, OperationType operation, VirtualRegister result, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             return appendCallWithExceptionCheckSetJSValueResultWithProfile(metadata, operation, result); | 
 |         } | 
 |  | 
 |         template<typename OperationType, typename Metadata, typename... Args> | 
 |         std::enable_if_t<FunctionTraits<OperationType>::hasResult, void> | 
 |         callOperationWithProfile(Metadata& metadata, Address target, VirtualRegister result, Args... args) | 
 |         { | 
 |             setupArgumentsForIndirectCall<OperationType>(target, args...); | 
 |             return appendCallWithExceptionCheckSetJSValueResultWithProfile(metadata, Address(GPRInfo::nonArgGPR0, target.offset), result); | 
 |         } | 
 |  | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperationWithResult(OperationType operation, JSValueRegs resultRegs, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             auto result = appendCallWithExceptionCheck(operation); | 
 |             setupResults(resultRegs); | 
 |             return result; | 
 |         } | 
 |  | 
 | #if OS(WINDOWS) && CPU(X86_64) | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperationNoExceptionCheck(OperationType operation, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             updateTopCallFrame(); | 
 |             // x64 Windows cannot use standard call when the return type is larger than 64 bits. | 
 |             if constexpr (is64BitType<typename FunctionTraits<OperationType>::ResultType>::value) | 
 |                 return appendCall(operation); | 
 |             return appendCallWithSlowPathReturnType(operation); | 
 |         } | 
 | #else // OS(WINDOWS) && CPU(X86_64) | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperationNoExceptionCheck(OperationType operation, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             updateTopCallFrame(); | 
 |             return appendCall(operation); | 
 |         } | 
 | #endif // OS(WINDOWS) && CPU(X86_64) | 
 |  | 
 |         template<typename OperationType, typename... Args> | 
 |         MacroAssembler::Call callOperationWithCallFrameRollbackOnException(OperationType operation, Args... args) | 
 |         { | 
 |             setupArguments<OperationType>(args...); | 
 |             return appendCallWithCallFrameRollbackOnException(operation); | 
 |         } | 
 |  | 
 |         enum class ProfilingPolicy { | 
 |             ShouldEmitProfiling, | 
 |             NoProfiling | 
 |         }; | 
 |  | 
 |         template<typename Op, typename SnippetGenerator> | 
 |         void emitBitBinaryOpFastPath(const Instruction* currentInstruction, ProfilingPolicy shouldEmitProfiling = ProfilingPolicy::NoProfiling); | 
 |  | 
 |         void emitRightShiftFastPath(const Instruction* currentInstruction, OpcodeID); | 
 |  | 
 |         template<typename Op> | 
 |         void emitRightShiftFastPath(const Instruction* currentInstruction, JITRightShiftGenerator::ShiftType); | 
 |  | 
 |         void updateTopCallFrame(); | 
 |  | 
 |         Call emitNakedNearCall(CodePtr<NoPtrTag> function = { }); | 
 |         Call emitNakedNearTailCall(CodePtr<NoPtrTag> function = { }); | 
 |         Jump emitNakedNearJump(CodePtr<JITThunkPtrTag> function = { }); | 
 |  | 
 |         // Loads the character value of a single character string into dst. | 
 |         void emitLoadCharacterString(RegisterID src, RegisterID dst, JumpList& failures); | 
 |  | 
 |         int jumpTarget(const Instruction*, int target); | 
 |          | 
 | #if ENABLE(DFG_JIT) | 
 |         void emitEnterOptimizationCheck(); | 
 | #else | 
 |         void emitEnterOptimizationCheck() { } | 
 | #endif | 
 |  | 
 | #ifndef NDEBUG | 
 |         void printBytecodeOperandTypes(VirtualRegister src1, VirtualRegister src2); | 
 | #endif | 
 |  | 
 | #if ENABLE(SAMPLING_FLAGS) | 
 |         void setSamplingFlag(int32_t); | 
 |         void clearSamplingFlag(int32_t); | 
 | #endif | 
 |  | 
 | #if ENABLE(SAMPLING_COUNTERS) | 
 |         void emitCount(AbstractSamplingCounter&, int32_t = 1); | 
 | #endif | 
 |  | 
 | #if ENABLE(DFG_JIT) | 
 |         bool canBeOptimized() { return m_canBeOptimized; } | 
 |         bool canBeOptimizedOrInlined() { return m_canBeOptimizedOrInlined; } | 
 |         bool shouldEmitProfiling() { return m_shouldEmitProfiling; } | 
 | #else | 
 |         bool canBeOptimized() { return false; } | 
 |         bool canBeOptimizedOrInlined() { return false; } | 
 |         // Enables use of value profiler with tiered compilation turned off, | 
 |         // in which case all code gets profiled. | 
 |         bool shouldEmitProfiling() { return false; } | 
 | #endif | 
 |  | 
 |         static bool reportCompileTimes(); | 
 |         static bool computeCompileTimes(); | 
 |          | 
 |         // If you need to check a value from the metadata table and you need it to | 
 |         // be consistent across the fast and slow path, then you want to use this. | 
 |         // It will give the slow path the same value read by the fast path. | 
 |         GetPutInfo copiedGetPutInfo(OpPutToScope); | 
 |         template<typename BinaryOp> | 
 |         BinaryArithProfile copiedArithProfile(BinaryOp); | 
 |  | 
 |         Interpreter* m_interpreter; | 
 |  | 
 |         Vector<FarCallRecord> m_farCalls; | 
 |         Vector<NearCallRecord> m_nearCalls; | 
 |         Vector<NearJumpRecord> m_nearJumps; | 
 |         Vector<Label> m_labels; | 
 |         HashMap<BytecodeIndex, Label> m_checkpointLabels; | 
 |         Vector<JITGetByIdGenerator> m_getByIds; | 
 |         Vector<JITGetByValGenerator> m_getByVals; | 
 |         Vector<JITGetByIdWithThisGenerator> m_getByIdsWithThis; | 
 |         Vector<JITPutByIdGenerator> m_putByIds; | 
 |         Vector<JITInByIdGenerator> m_inByIds; | 
 |         Vector<JITInByValGenerator> m_inByVals; | 
 |         Vector<JITDelByIdGenerator> m_delByIds; | 
 |         Vector<JITDelByValGenerator> m_delByVals; | 
 |         Vector<JITInstanceOfGenerator> m_instanceOfs; | 
 |         Vector<JITPrivateBrandAccessGenerator> m_privateBrandAccesses; | 
 |         Vector<ByValCompilationInfo> m_byValCompilationInfo; | 
 |         Vector<CallCompilationInfo> m_callCompilationInfo; | 
 |         Vector<JumpTable> m_jmpTable; | 
 |  | 
 |         BytecodeIndex m_bytecodeIndex; | 
 |         Vector<SlowCaseEntry> m_slowCases; | 
 |         Vector<SwitchRecord> m_switches; | 
 |  | 
 |         HashMap<unsigned, unsigned> m_copiedGetPutInfos; | 
 |         HashMap<uint64_t, BinaryArithProfile> m_copiedArithProfiles; | 
 |  | 
 |         JumpList m_exceptionChecks; | 
 |         JumpList m_exceptionChecksWithCallFrameRollback; | 
 | #if !ENABLE(EXTRA_CTI_THUNKS) | 
 |         Label m_exceptionHandler; | 
 | #endif | 
 |  | 
 |         unsigned m_getByIdIndex { UINT_MAX }; | 
 |         unsigned m_getByValIndex { UINT_MAX }; | 
 |         unsigned m_getByIdWithThisIndex { UINT_MAX }; | 
 |         unsigned m_putByIdIndex { UINT_MAX }; | 
 |         unsigned m_inByIdIndex { UINT_MAX }; | 
 |         unsigned m_inByValIndex { UINT_MAX }; | 
 |         unsigned m_delByValIndex { UINT_MAX }; | 
 |         unsigned m_delByIdIndex { UINT_MAX }; | 
 |         unsigned m_instanceOfIndex { UINT_MAX }; | 
 |         unsigned m_privateBrandAccessIndex { UINT_MAX }; | 
 |         unsigned m_byValInstructionIndex { UINT_MAX }; | 
 |         unsigned m_callLinkInfoIndex { UINT_MAX }; | 
 |         unsigned m_bytecodeCountHavingSlowCase { 0 }; | 
 |          | 
 |         Label m_arityCheck; | 
 |         std::unique_ptr<LinkBuffer> m_linkBuffer; | 
 |  | 
 |         std::unique_ptr<JITDisassembler> m_disassembler; | 
 |         RefPtr<Profiler::Compilation> m_compilation; | 
 |  | 
 |         PCToCodeOriginMapBuilder m_pcToCodeOriginMapBuilder; | 
 |         std::unique_ptr<PCToCodeOriginMap> m_pcToCodeOriginMap; | 
 |  | 
 |         HashMap<const Instruction*, void*> m_instructionToMathIC; | 
 |         HashMap<const Instruction*, UniqueRef<MathICGenerationState>> m_instructionToMathICGenerationState; | 
 |  | 
 |         bool m_canBeOptimized; | 
 |         bool m_canBeOptimizedOrInlined; | 
 |         bool m_shouldEmitProfiling; | 
 |         BytecodeIndex m_loopOSREntryBytecodeIndex; | 
 |  | 
 |         RefPtr<DirectJITCode> m_jitCode; | 
 |     }; | 
 |  | 
 | } // namespace JSC | 
 |  | 
 |  | 
 | #endif // ENABLE(JIT) |