blob: 08c48fd3132e700954b9c08613f9dd0907ce2d24 [file] [log] [blame]
/*
* Copyright (C) 2008-2022 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.
*/
#include "config.h"
#if ENABLE(JIT)
#include "JIT.h"
#include "BaselineJITRegisters.h"
#include "BytecodeOperandsForCheckpoint.h"
#include "CacheableIdentifierInlines.h"
#include "CallFrameShuffler.h"
#include "CodeBlock.h"
#include "JITInlines.h"
#include "JITThunks.h"
#include "ScratchRegisterAllocator.h"
#include "SetupVarargsFrame.h"
#include "SlowPathCall.h"
#include "StackAlignment.h"
#include "ThunkGenerators.h"
namespace JSC {
void JIT::emit_op_ret(const JSInstruction* currentInstruction)
{
static_assert(noOverlap(returnValueJSR, callFrameRegister));
// Return the result in returnValueGPR (returnValueGPR2/returnValueGPR on 32-bit).
auto bytecode = currentInstruction->as<OpRet>();
emitGetVirtualRegister(bytecode.m_value, returnValueJSR);
jumpThunk(CodeLocationLabel { vm().getCTIStub(CommonJITThunkID::ReturnFromBaseline).retaggedCode<NoPtrTag>() });
}
template<typename Op>
void JIT::emitPutCallResult(const Op& bytecode)
{
emitValueProfilingSite(bytecode, returnValueJSR);
emitPutVirtualRegister(destinationFor(bytecode, m_bytecodeIndex.checkpoint()).virtualRegister(), returnValueJSR);
}
template<typename Op>
void JIT::compileSetupFrame(const Op& bytecode)
{
constexpr auto opcodeID = Op::opcodeID;
if constexpr (opcodeID == op_call_varargs || opcodeID == op_construct_varargs || opcodeID == op_tail_call_varargs || opcodeID == op_tail_call_forward_arguments) {
VirtualRegister thisValue = bytecode.m_thisValue;
VirtualRegister arguments = bytecode.m_arguments;
int firstFreeRegister = bytecode.m_firstFree.offset(); // FIXME: Why is this a virtual register if we never use it as one...
int firstVarArgOffset = bytecode.m_firstVarArg;
{
constexpr GPRReg globalObjectGPR = preferredArgumentGPR<S_JITOperation_GJZZ, 0>();
constexpr JSValueRegs argumentsJSR = preferredArgumentJSR<S_JITOperation_GJZZ, 1>();
S_JITOperation_GJZZ sizeOperation;
if constexpr (opcodeID == op_tail_call_forward_arguments)
sizeOperation = operationSizeFrameForForwardArguments;
else
sizeOperation = operationSizeFrameForVarargs;
loadGlobalObject(globalObjectGPR);
emitGetVirtualRegister(arguments, argumentsJSR);
callOperation(sizeOperation, globalObjectGPR, argumentsJSR, -firstFreeRegister, firstVarArgOffset);
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetVarargsFrame(*this, returnValueGPR, false, regT1, regT1);
}
#if USE(JSVALUE64)
addPtr(TrustedImm32(-static_cast<int32_t>(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf<stackAlignmentBytes()>(5 * sizeof(void*)))), regT1, stackPointerRegister);
#elif USE(JSVALUE32_64)
addPtr(TrustedImm32(-(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf<stackAlignmentBytes()>(6 * sizeof(void*)))), regT1, stackPointerRegister);
#endif
{
emitGetVirtualRegister(arguments, jsRegT32);
F_JITOperation_GFJZZ setupOperation;
if constexpr (opcodeID == op_tail_call_forward_arguments)
setupOperation = operationSetupForwardArgumentsFrame;
else
setupOperation = operationSetupVarargsFrame;
loadGlobalObject(regT4);
callOperation(setupOperation, regT4, regT1, jsRegT32, firstVarArgOffset, regT0);
move(returnValueGPR, regT5);
}
// Profile the argument count.
load32(Address(regT5, CallFrameSlot::argumentCountIncludingThis * static_cast<int>(sizeof(Register)) + PayloadOffset), regT2);
move(TrustedImm32(CallLinkInfo::maxProfiledArgumentCountIncludingThisForVarargs), regT0);
#if CPU(ARM64) || CPU(X86_64)
moveConditionally32(Above, regT2, regT0, regT0, regT2, regT2);
#else
auto lower = branch32(BelowOrEqual, regT2, regT0);
move(regT0, regT2);
lower.link(this);
#endif
materializePointerIntoMetadata(bytecode, Op::Metadata::offsetOfCallLinkInfo(), regT0);
Jump notBiggest = branch32(Above, Address(regT0, CallLinkInfo::offsetOfMaxArgumentCountIncludingThisForVarargs()), regT2);
store8(regT2, Address(regT0, CallLinkInfo::offsetOfMaxArgumentCountIncludingThisForVarargs()));
notBiggest.link(this);
// Initialize 'this'.
constexpr JSValueRegs thisJSR = jsRegT10;
emitGetVirtualRegister(thisValue, thisJSR);
storeValue(thisJSR, Address(regT5, CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register))));
addPtr(TrustedImm32(sizeof(CallerFrameAndPC)), regT5, stackPointerRegister);
} else {
unsigned checkpoint = m_bytecodeIndex.checkpoint();
int argCountIncludingThis = argumentCountIncludingThisFor(bytecode, checkpoint);
int registerOffset = -static_cast<int>(stackOffsetInRegistersForCall(bytecode, checkpoint));
if constexpr (opcodeID == op_call || opcodeID == op_tail_call || opcodeID == op_iterator_open || opcodeID == op_call_ignore_result) {
if (shouldEmitProfiling()) {
constexpr JSValueRegs tmpJSR = returnValueJSR;
constexpr GPRReg tmpGPR = tmpJSR.payloadGPR();
emitGetVirtualRegister(VirtualRegister(registerOffset + CallFrame::argumentOffsetIncludingThis(0)), tmpJSR);
Jump done = branchIfNotCell(tmpJSR);
load32(Address(tmpJSR.payloadGPR(), JSCell::structureIDOffset()), tmpGPR);
store32ToMetadata(tmpGPR, bytecode, Op::Metadata::offsetOfArrayProfile() + ArrayProfile::offsetOfLastSeenStructureID());
done.link(this);
}
}
addPtr(TrustedImm32(registerOffset * sizeof(Register) + sizeof(CallerFrameAndPC)), callFrameRegister, stackPointerRegister);
store32(TrustedImm32(argCountIncludingThis), Address(stackPointerRegister, CallFrameSlot::argumentCountIncludingThis * static_cast<int>(sizeof(Register)) + PayloadOffset - sizeof(CallerFrameAndPC)));
}
}
template<typename Op>
void JIT::compileCallDirectEval(const Op&)
{
}
template<>
void JIT::compileCallDirectEval(const OpCallDirectEval& bytecode)
{
using BaselineJITRegisters::CallDirectEval::SlowPath::calleeFrameGPR;
using BaselineJITRegisters::CallDirectEval::SlowPath::thisValueJSR;
using BaselineJITRegisters::CallDirectEval::SlowPath::scopeGPR;
addPtr(TrustedImm32(-static_cast<ptrdiff_t>(sizeof(CallerFrameAndPC))), stackPointerRegister, calleeFrameGPR);
storePtr(callFrameRegister, Address(calleeFrameGPR, CallFrame::callerFrameOffset()));
resetSP();
emitGetVirtualRegister(bytecode.m_thisValue, thisValueJSR);
emitGetVirtualRegisterPayload(bytecode.m_scope, scopeGPR);
callOperation(selectCallDirectEvalOperation(bytecode.m_lexicallyScopedFeatures), calleeFrameGPR, scopeGPR, thisValueJSR);
addSlowCase(branchIfEmpty(returnValueJSR));
setFastPathResumePoint();
emitPutCallResult(bytecode);
}
void JIT::compileCallDirectEvalSlowCase(const JSInstruction* instruction, Vector<SlowCaseEntry>::iterator& iter)
{
linkAllSlowCases(iter);
auto bytecode = instruction->as<OpCallDirectEval>();
int registerOffset = -bytecode.m_argv;
addPtr(TrustedImm32(registerOffset * sizeof(Register) + sizeof(CallerFrameAndPC)), callFrameRegister, stackPointerRegister);
static_assert(noOverlap(BaselineJITRegisters::Call::calleeJSR, BaselineJITRegisters::Call::callLinkInfoGPR, regT3));
loadValue(Address(stackPointerRegister, sizeof(Register) * CallFrameSlot::callee - sizeof(CallerFrameAndPC)), BaselineJITRegisters::Call::calleeJSR);
materializePointerIntoMetadata(bytecode, OpCallDirectEval::Metadata::offsetOfCallLinkInfo(), BaselineJITRegisters::Call::callLinkInfoGPR);
emitVirtualCallWithoutMovingGlobalObject(*m_vm, BaselineJITRegisters::Call::callLinkInfoGPR, CallMode::Regular);
resetSP();
}
template<typename Op>
bool JIT::compileTailCall(const Op&, BaselineUnlinkedCallLinkInfo*, unsigned)
{
return false;
}
template<>
bool JIT::compileTailCall(const OpTailCall& bytecode, BaselineUnlinkedCallLinkInfo* callLinkInfo, unsigned callLinkInfoIndex)
{
CallLinkInfo::emitTailCallFastPath(*this, callLinkInfo, scopedLambda<void()>([&] {
CallFrameShuffleData shuffleData = CallFrameShuffleData::createForBaselineOrLLIntTailCall(bytecode, m_unlinkedCodeBlock->numParameters());
CallFrameShuffler shuffler { *this, shuffleData };
shuffler.setCalleeJSValueRegs(BaselineJITRegisters::Call::calleeJSR);
shuffler.lockGPR(BaselineJITRegisters::Call::callLinkInfoGPR);
shuffler.lockGPR(BaselineJITRegisters::Call::callTargetGPR);
shuffler.prepareForTailCall();
}));
auto doneLocation = label();
m_callCompilationInfo[callLinkInfoIndex].doneLocation = doneLocation;
return true;
}
template<typename Op>
void JIT::compileOpCall(const JSInstruction* instruction)
{
auto bytecode = instruction->as<Op>();
VirtualRegister callee = calleeFor(bytecode, m_bytecodeIndex.checkpoint());
/* Caller always:
- Updates callFrameRegister to callee callFrame.
- Initializes ArgumentCount; CallerFrame; Callee.
For a JS call:
- Callee initializes ReturnPC; CodeBlock.
- Callee restores callFrameRegister before return.
For a non-JS call:
- Caller initializes ReturnPC; CodeBlock.
- Caller restores callFrameRegister after return.
*/
unsigned callLinkInfoIndex = 0;
BaselineUnlinkedCallLinkInfo* callLinkInfo = nullptr;
if constexpr (Op::opcodeID != op_call_direct_eval) {
callLinkInfo = addUnlinkedCallLinkInfo();
callLinkInfo->bytecodeIndex = m_bytecodeIndex;
callLinkInfoIndex = m_callCompilationInfo.size();
m_callCompilationInfo.append(CallCompilationInfo());
m_callCompilationInfo[callLinkInfoIndex].unlinkedCallLinkInfo = callLinkInfo;
}
compileSetupFrame(bytecode);
// SP holds newCallFrame + sizeof(CallerFrameAndPC), with ArgumentCount initialized.
uint32_t locationBits = CallSiteIndex(m_bytecodeIndex).bits();
store32(TrustedImm32(locationBits), Address(callFrameRegister, CallFrameSlot::argumentCountIncludingThis * static_cast<int>(sizeof(Register)) + TagOffset));
emitGetVirtualRegister(callee, BaselineJITRegisters::Call::calleeJSR);
storeValue(BaselineJITRegisters::Call::calleeJSR, Address(stackPointerRegister, CallFrameSlot::callee * static_cast<int>(sizeof(Register)) - sizeof(CallerFrameAndPC)));
if constexpr (Op::opcodeID == op_call_direct_eval) {
compileCallDirectEval(bytecode);
return;
}
materializePointerIntoMetadata(bytecode, Op::Metadata::offsetOfCallLinkInfo(), BaselineJITRegisters::Call::callLinkInfoGPR);
if constexpr (Op::opcodeID == op_tail_call)
compileTailCall(bytecode, callLinkInfo, callLinkInfoIndex);
else {
if constexpr (Op::opcodeID == op_tail_call_varargs || Op::opcodeID == op_tail_call_forward_arguments) {
CallLinkInfo::emitTailCallFastPath(*this, callLinkInfo, scopedLambda<void()>([&] {
emitRestoreCalleeSaves();
prepareForTailCallSlow(RegisterSet {
BaselineJITRegisters::Call::calleeJSR.payloadGPR(),
#if USE(JSVALUE32_64)
BaselineJITRegisters::Call::calleeJSR.tagGPR(),
#endif
BaselineJITRegisters::Call::callLinkInfoGPR,
BaselineJITRegisters::Call::callTargetGPR,
});
}));
auto doneLocation = label();
m_callCompilationInfo[callLinkInfoIndex].doneLocation = doneLocation;
} else {
CallLinkInfo::emitFastPath(*this, callLinkInfo);
auto doneLocation = label();
m_callCompilationInfo[callLinkInfoIndex].doneLocation = doneLocation;
if constexpr (Op::opcodeID != op_iterator_open && Op::opcodeID != op_iterator_next)
setFastPathResumePoint();
resetSP();
if constexpr (Op::opcodeID != op_call_ignore_result)
emitPutCallResult(bytecode);
}
}
}
void JIT::emit_op_call(const JSInstruction* currentInstruction)
{
compileOpCall<OpCall>(currentInstruction);
}
void JIT::emit_op_call_ignore_result(const JSInstruction* currentInstruction)
{
compileOpCall<OpCallIgnoreResult>(currentInstruction);
}
void JIT::emit_op_tail_call(const JSInstruction* currentInstruction)
{
compileOpCall<OpTailCall>(currentInstruction);
}
void JIT::emit_op_call_direct_eval(const JSInstruction* currentInstruction)
{
compileOpCall<OpCallDirectEval>(currentInstruction);
}
void JIT::emit_op_call_varargs(const JSInstruction* currentInstruction)
{
compileOpCall<OpCallVarargs>(currentInstruction);
}
void JIT::emit_op_tail_call_varargs(const JSInstruction* currentInstruction)
{
compileOpCall<OpTailCallVarargs>(currentInstruction);
}
void JIT::emit_op_tail_call_forward_arguments(const JSInstruction* currentInstruction)
{
compileOpCall<OpTailCallForwardArguments>(currentInstruction);
}
void JIT::emit_op_construct_varargs(const JSInstruction* currentInstruction)
{
compileOpCall<OpConstructVarargs>(currentInstruction);
}
void JIT::emit_op_construct(const JSInstruction* currentInstruction)
{
compileOpCall<OpConstruct>(currentInstruction);
}
void JIT::emitSlow_op_call_direct_eval(const JSInstruction* currentInstruction, Vector<SlowCaseEntry>::iterator& iter)
{
compileCallDirectEvalSlowCase(currentInstruction, iter);
}
void JIT::emit_op_iterator_open(const JSInstruction* instruction)
{
auto bytecode = instruction->as<OpIteratorOpen>();
auto* tryFastFunction = ([&] () {
switch (instruction->width()) {
case Narrow: return iterator_open_try_fast_narrow;
case Wide16: return iterator_open_try_fast_wide16;
case Wide32: return iterator_open_try_fast_wide32;
default: RELEASE_ASSERT_NOT_REACHED();
}
})();
GetByIdModeMetadata modeMetadata = bytecode.metadata(m_profiledCodeBlock).m_modeMetadata;
CacheType cacheType = CacheType::GetByIdSelf;
if (modeMetadata.mode == GetByIdMode::ProtoLoad)
cacheType = CacheType::GetByIdPrototype;
JITSlowPathCall slowPathCall(this, tryFastFunction);
slowPathCall.call();
Jump fastCase = branch32(NotEqual, GPRInfo::returnValueGPR2, TrustedImm32(static_cast<uint32_t>(IterationMode::Generic)));
compileOpCall<OpIteratorOpen>(instruction);
advanceToNextCheckpoint();
// call result (iterator) is in returnValueJSR
using BaselineJITRegisters::GetById::baseJSR;
using BaselineJITRegisters::GetById::resultJSR;
using BaselineJITRegisters::GetById::stubInfoGPR;
moveValueRegs(returnValueJSR, baseJSR);
auto [ stubInfo, stubInfoIndex ] = addUnlinkedStructureStubInfo();
loadStructureStubInfo(stubInfoIndex, stubInfoGPR);
emitJumpSlowCaseIfNotJSCell(baseJSR);
static_assert(noOverlap(returnValueJSR, stubInfoGPR));
const Identifier* ident = &vm().propertyNames->next;
JITGetByIdGenerator gen(
nullptr, stubInfo, JITType::BaselineJIT, CodeOrigin(m_bytecodeIndex), CallSiteIndex(BytecodeIndex(m_bytecodeIndex.offset())), RegisterSetBuilder::stubUnavailableRegisters(),
CacheableIdentifier::createFromImmortalIdentifier(ident->impl()), baseJSR, resultJSR, stubInfoGPR, AccessType::GetById, cacheType);
gen.generateDataICFastPath(*this);
resetSP(); // We might OSR exit here, so we need to conservatively reset SP
addSlowCase();
m_getByIds.append(gen);
setFastPathResumePoint();
emitValueProfilingSite(bytecode, resultJSR);
emitPutVirtualRegister(bytecode.m_next, resultJSR);
fastCase.link(this);
}
void JIT::emitSlow_op_iterator_open(const JSInstruction*, Vector<SlowCaseEntry>::iterator& iter)
{
linkAllSlowCases(iter);
using BaselineJITRegisters::GetById::baseJSR;
using BaselineJITRegisters::GetById::stubInfoGPR;
JumpList notObject;
notObject.append(branchIfNotCell(baseJSR));
notObject.append(branchIfNotObject(baseJSR.payloadGPR()));
JITGetByIdGenerator& gen = m_getByIds[m_getByIdIndex++];
gen.reportBaselineDataICSlowPathBegin(label());
nearCallThunk(CodeLocationLabel { InlineCacheCompiler::generateSlowPathCode(vm(), gen.accessType()).retaggedCode<NoPtrTag>() });
static_assert(BaselineJITRegisters::GetById::resultJSR == returnValueJSR);
jump().linkTo(fastPathResumePoint(), this);
notObject.link(this);
loadGlobalObject(argumentGPR0);
callOperation(operationThrowIteratorResultIsNotObject, argumentGPR0);
}
void JIT::emit_op_iterator_next(const JSInstruction* instruction)
{
auto bytecode = instruction->as<OpIteratorNext>();
using BaselineJITRegisters::GetById::baseJSR;
using BaselineJITRegisters::GetById::resultJSR;
using BaselineJITRegisters::GetById::stubInfoGPR;
constexpr JSValueRegs nextJSR = baseJSR; // Used as temporary register
emitGetVirtualRegister(bytecode.m_next, nextJSR);
Jump genericCase = branchIfNotEmpty(nextJSR);
JumpList doneCases;
#if CPU(ARM64) || CPU(X86_64)
loadGlobalObject(argumentGPR0);
emitGetVirtualRegister(bytecode.m_iterator, argumentGPR1);
emitGetVirtualRegister(bytecode.m_iterable, argumentGPR2);
materializePointerIntoMetadata(bytecode, 0, argumentGPR3);
callOperation(operationIteratorNextTryFast, argumentGPR0, argumentGPR1, argumentGPR2, argumentGPR3);
emitPutVirtualRegister(bytecode.m_done, returnValueGPR);
emitPutVirtualRegister(bytecode.m_value, returnValueGPR2);
doneCases.append(branchIfEmpty(JSValueRegs { returnValueGPR2 }));
emitValueProfilingSite(bytecode, JSValueRegs { returnValueGPR2 });
#else
auto* tryFastFunction = ([&] () {
switch (instruction->width()) {
case Narrow: return iterator_next_try_fast_narrow;
case Wide16: return iterator_next_try_fast_wide16;
case Wide32: return iterator_next_try_fast_wide32;
default: RELEASE_ASSERT_NOT_REACHED();
}
})();
JITSlowPathCall slowPathCall(this, tryFastFunction);
slowPathCall.call();
#endif
doneCases.append(jump());
genericCase.link(this);
load8FromMetadata(bytecode, OpIteratorNext::Metadata::offsetOfIterationMetadata() + IterationModeMetadata::offsetOfSeenModes(), regT0);
or32(TrustedImm32(static_cast<uint8_t>(IterationMode::Generic)), regT0);
store8ToMetadata(regT0, bytecode, OpIteratorNext::Metadata::offsetOfIterationMetadata() + IterationModeMetadata::offsetOfSeenModes());
compileOpCall<OpIteratorNext>(instruction);
advanceToNextCheckpoint();
// call result ({ done, value } JSObject) in regT0 (regT1/regT0 or 32-bit)
static_assert(noOverlap(resultJSR, stubInfoGPR));
moveValueRegs(returnValueJSR, baseJSR);
addSlowCase(branchIfNotCell(baseJSR));
addSlowCase(branchIfNotObject(baseJSR.payloadGPR()));
{
auto [ stubInfo, stubInfoIndex ] = addUnlinkedStructureStubInfo();
loadStructureStubInfo(stubInfoIndex, stubInfoGPR);
JITGetByIdGenerator gen(
nullptr, stubInfo, JITType::BaselineJIT, CodeOrigin(m_bytecodeIndex), CallSiteIndex(BytecodeIndex(m_bytecodeIndex.offset())), RegisterSetBuilder::stubUnavailableRegisters(),
CacheableIdentifier::createFromImmortalIdentifier(vm().propertyNames->done.impl()), baseJSR, resultJSR, stubInfoGPR, AccessType::GetById, CacheType::GetByIdSelf);
gen.generateDataICFastPath(*this);
resetSP(); // We might OSR exit here, so we need to conservatively reset SP
addSlowCase();
m_getByIds.append(gen);
BytecodeIndex bytecodeIndex = m_bytecodeIndex;
advanceToNextCheckpoint();
emitValueProfilingSite(bytecode, bytecodeIndex, resultJSR);
emitPutVirtualRegister(bytecode.m_done, resultJSR);
}
{
auto usedRegisters = RegisterSetBuilder(resultJSR).buildAndValidate();
ScratchRegisterAllocator scratchAllocator(usedRegisters);
GPRReg scratch1 = scratchAllocator.allocateScratchGPR();
GPRReg scratch2 = scratchAllocator.allocateScratchGPR();
const bool shouldCheckMasqueradesAsUndefined = false;
JumpList iterationDone = branchIfTruthy(vm(), resultJSR, scratch1, scratch2, fpRegT0, fpRegT1, shouldCheckMasqueradesAsUndefined, CCallHelpers::LazyBaselineGlobalObject);
emitGetVirtualRegister(bytecode.m_value, baseJSR);
auto [ stubInfo, stubInfoIndex ] = addUnlinkedStructureStubInfo();
loadStructureStubInfo(stubInfoIndex, stubInfoGPR);
JITGetByIdGenerator gen(
nullptr, stubInfo, JITType::BaselineJIT, CodeOrigin(m_bytecodeIndex), CallSiteIndex(BytecodeIndex(m_bytecodeIndex.offset())), RegisterSetBuilder::stubUnavailableRegisters(),
CacheableIdentifier::createFromImmortalIdentifier(vm().propertyNames->value.impl()), baseJSR, resultJSR, stubInfoGPR, AccessType::GetById, CacheType::GetByIdSelf);
gen.generateDataICFastPath(*this);
resetSP(); // We might OSR exit here, so we need to conservatively reset SP
addSlowCase();
m_getByIds.append(gen);
setFastPathResumePoint();
emitValueProfilingSite(bytecode, resultJSR);
emitPutVirtualRegister(bytecode.m_value, resultJSR);
iterationDone.link(this);
}
doneCases.link(this);
}
void JIT::emitSlow_op_iterator_next(const JSInstruction*, Vector<SlowCaseEntry>::iterator& iter)
{
using BaselineJITRegisters::GetById::baseJSR;
using BaselineJITRegisters::GetById::resultJSR;
using BaselineJITRegisters::GetById::stubInfoGPR;
linkAllSlowCases(iter);
loadGlobalObject(argumentGPR0);
callOperation(operationThrowIteratorResultIsNotObject, argumentGPR0);
{
JITGetByIdGenerator& gen = m_getByIds[m_getByIdIndex++];
gen.reportBaselineDataICSlowPathBegin(label());
nearCallThunk(CodeLocationLabel { InlineCacheCompiler::generateSlowPathCode(vm(), gen.accessType()).retaggedCode<NoPtrTag>() });
static_assert(BaselineJITRegisters::GetById::resultJSR == returnValueJSR);
emitJumpSlowToHotForCheckpoint(jump());
}
{
linkAllSlowCases(iter);
JITGetByIdGenerator& gen = m_getByIds[m_getByIdIndex++];
gen.reportBaselineDataICSlowPathBegin(label());
nearCallThunk(CodeLocationLabel { InlineCacheCompiler::generateSlowPathCode(vm(), gen.accessType()).retaggedCode<NoPtrTag>() });
static_assert(BaselineJITRegisters::GetById::resultJSR == returnValueJSR);
}
}
} // namespace JSC
#endif // ENABLE(JIT)