| /* |
| * Copyright (C) 2011, 2012, 2013 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" |
| #include "DFGSpeculativeJIT.h" |
| |
| #if ENABLE(DFG_JIT) |
| |
| #include "Arguments.h" |
| #include "ArrayPrototype.h" |
| #include "DFGAbstractInterpreterInlines.h" |
| #include "DFGCallArrayAllocatorSlowPathGenerator.h" |
| #include "DFGOperations.h" |
| #include "DFGSlowPathGenerator.h" |
| #include "JSCJSValueInlines.h" |
| #include "ObjectPrototype.h" |
| |
| namespace JSC { namespace DFG { |
| |
| #if USE(JSVALUE64) |
| |
| void SpeculativeJIT::boxInt52(GPRReg sourceGPR, GPRReg targetGPR, DataFormat format) |
| { |
| GPRReg tempGPR; |
| if (sourceGPR == targetGPR) |
| tempGPR = allocate(); |
| else |
| tempGPR = targetGPR; |
| |
| FPRReg fpr = fprAllocate(); |
| |
| if (format == DataFormatInt52) |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), sourceGPR); |
| else |
| ASSERT(format == DataFormatStrictInt52); |
| |
| m_jit.boxInt52(sourceGPR, targetGPR, tempGPR, fpr); |
| |
| if (tempGPR != targetGPR) |
| unlock(tempGPR); |
| |
| unlock(fpr); |
| } |
| |
| GPRReg SpeculativeJIT::fillJSValue(Edge edge) |
| { |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: { |
| GPRReg gpr = allocate(); |
| |
| if (edge->hasConstant()) { |
| if (isInt32Constant(edge.node())) { |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| JSValue jsValue = jsNumber(valueOfInt32Constant(edge.node())); |
| m_jit.move(MacroAssembler::Imm64(JSValue::encode(jsValue)), gpr); |
| } else if (isNumberConstant(edge.node())) { |
| info.fillJSValue(*m_stream, gpr, DataFormatJSDouble); |
| JSValue jsValue(JSValue::EncodeAsDouble, valueOfNumberConstant(edge.node())); |
| m_jit.move(MacroAssembler::Imm64(JSValue::encode(jsValue)), gpr); |
| } else { |
| ASSERT(isJSConstant(edge.node())); |
| JSValue jsValue = valueOfJSConstant(edge.node()); |
| m_jit.move(MacroAssembler::TrustedImm64(JSValue::encode(jsValue)), gpr); |
| info.fillJSValue(*m_stream, gpr, DataFormatJS); |
| } |
| |
| m_gprs.retain(gpr, virtualRegister, SpillOrderConstant); |
| } else { |
| DataFormat spillFormat = info.spillFormat(); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| switch (spillFormat) { |
| case DataFormatInt32: { |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, gpr); |
| spillFormat = DataFormatJSInt32; |
| break; |
| } |
| |
| case DataFormatInt52: |
| case DataFormatStrictInt52: { |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| boxInt52(gpr, gpr, spillFormat); |
| return gpr; |
| } |
| |
| default: |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| if (spillFormat == DataFormatDouble) { |
| // Need to box the double, since we want a JSValue. |
| m_jit.sub64(GPRInfo::tagTypeNumberRegister, gpr); |
| spillFormat = DataFormatJSDouble; |
| } else |
| RELEASE_ASSERT(spillFormat & DataFormatJS); |
| break; |
| } |
| info.fillJSValue(*m_stream, gpr, spillFormat); |
| } |
| return gpr; |
| } |
| |
| case DataFormatInt32: { |
| GPRReg gpr = info.gpr(); |
| // If the register has already been locked we need to take a copy. |
| // If not, we'll zero extend in place, so mark on the info that this is now type DataFormatInt32, not DataFormatJSInt32. |
| if (m_gprs.isLocked(gpr)) { |
| GPRReg result = allocate(); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, gpr, result); |
| return result; |
| } |
| m_gprs.lock(gpr); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, gpr); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| return gpr; |
| } |
| |
| case DataFormatDouble: { |
| FPRReg fpr = info.fpr(); |
| GPRReg gpr = boxDouble(fpr); |
| |
| // Update all info |
| info.fillJSValue(*m_stream, gpr, DataFormatJSDouble); |
| m_fprs.release(fpr); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderJS); |
| |
| return gpr; |
| } |
| |
| case DataFormatInt52: |
| case DataFormatStrictInt52: { |
| GPRReg gpr = info.gpr(); |
| lock(gpr); |
| GPRReg resultGPR = allocate(); |
| boxInt52(gpr, resultGPR, info.registerFormat()); |
| unlock(gpr); |
| return resultGPR; |
| } |
| |
| case DataFormatCell: |
| // No retag required on JSVALUE64! |
| case DataFormatJS: |
| case DataFormatJSInt32: |
| case DataFormatJSDouble: |
| case DataFormatJSCell: |
| case DataFormatJSBoolean: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| return gpr; |
| } |
| |
| case DataFormatBoolean: |
| case DataFormatStorage: |
| // this type currently never occurs |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidGPRReg; |
| } |
| } |
| |
| void SpeculativeJIT::nonSpeculativeUInt32ToNumber(Node* node) |
| { |
| SpeculateInt32Operand op1(this, node->child1()); |
| FPRTemporary boxer(this); |
| GPRTemporary result(this, Reuse, op1); |
| |
| JITCompiler::Jump positive = m_jit.branch32(MacroAssembler::GreaterThanOrEqual, op1.gpr(), TrustedImm32(0)); |
| |
| m_jit.convertInt32ToDouble(op1.gpr(), boxer.fpr()); |
| m_jit.addDouble(JITCompiler::AbsoluteAddress(&AssemblyHelpers::twoToThe32), boxer.fpr()); |
| |
| boxDouble(boxer.fpr(), result.gpr()); |
| |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| positive.link(&m_jit); |
| |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, op1.gpr(), result.gpr()); |
| |
| done.link(&m_jit); |
| |
| jsValueResult(result.gpr(), m_currentNode); |
| } |
| |
| void SpeculativeJIT::cachedGetById(CodeOrigin codeOrigin, GPRReg baseGPR, GPRReg resultGPR, unsigned identifierNumber, JITCompiler::Jump slowPathTarget, SpillRegistersMode spillMode) |
| { |
| JITGetByIdGenerator gen( |
| m_jit.codeBlock(), codeOrigin, usedRegisters(), JSValueRegs(baseGPR), |
| JSValueRegs(resultGPR), spillMode != NeedToSpill); |
| gen.generateFastPath(m_jit); |
| |
| JITCompiler::JumpList slowCases; |
| if (slowPathTarget.isSet()) |
| slowCases.append(slowPathTarget); |
| slowCases.append(gen.slowPathJump()); |
| |
| OwnPtr<SlowPathGenerator> slowPath = slowPathCall( |
| slowCases, this, operationGetByIdOptimize, resultGPR, gen.stubInfo(), baseGPR, |
| identifierUID(identifierNumber), spillMode); |
| |
| m_jit.addGetById(gen, slowPath.get()); |
| addSlowPathGenerator(slowPath.release()); |
| } |
| |
| void SpeculativeJIT::cachedPutById(CodeOrigin codeOrigin, GPRReg baseGPR, GPRReg valueGPR, Edge valueUse, GPRReg scratchGPR, unsigned identifierNumber, PutKind putKind, JITCompiler::Jump slowPathTarget) |
| { |
| writeBarrier(baseGPR, valueGPR, valueUse, WriteBarrierForPropertyAccess, scratchGPR); |
| |
| JITPutByIdGenerator gen( |
| m_jit.codeBlock(), codeOrigin, usedRegisters(), JSValueRegs(baseGPR), |
| JSValueRegs(valueGPR), scratchGPR, false, m_jit.ecmaModeFor(codeOrigin), putKind); |
| gen.generateFastPath(m_jit); |
| |
| JITCompiler::JumpList slowCases; |
| if (slowPathTarget.isSet()) |
| slowCases.append(slowPathTarget); |
| slowCases.append(gen.slowPathJump()); |
| |
| OwnPtr<SlowPathGenerator> slowPath = slowPathCall( |
| slowCases, this, gen.slowPathFunction(), NoResult, gen.stubInfo(), valueGPR, baseGPR, |
| identifierUID(identifierNumber)); |
| |
| m_jit.addPutById(gen, slowPath.get()); |
| addSlowPathGenerator(slowPath.release()); |
| } |
| |
| void SpeculativeJIT::nonSpeculativeNonPeepholeCompareNull(Edge operand, bool invert) |
| { |
| JSValueOperand arg(this, operand); |
| GPRReg argGPR = arg.gpr(); |
| |
| GPRTemporary result(this, Reuse, arg); |
| GPRReg resultGPR = result.gpr(); |
| |
| JITCompiler::Jump notCell; |
| |
| JITCompiler::Jump notMasqueradesAsUndefined; |
| if (masqueradesAsUndefinedWatchpointIsStillValid()) { |
| if (!isKnownCell(operand.node())) |
| notCell = m_jit.branchTest64(MacroAssembler::NonZero, argGPR, GPRInfo::tagMaskRegister); |
| |
| m_jit.move(invert ? TrustedImm32(1) : TrustedImm32(0), resultGPR); |
| notMasqueradesAsUndefined = m_jit.jump(); |
| } else { |
| GPRTemporary localGlobalObject(this); |
| GPRTemporary remoteGlobalObject(this); |
| |
| if (!isKnownCell(operand.node())) |
| notCell = m_jit.branchTest64(MacroAssembler::NonZero, argGPR, GPRInfo::tagMaskRegister); |
| |
| m_jit.loadPtr(JITCompiler::Address(argGPR, JSCell::structureOffset()), resultGPR); |
| JITCompiler::Jump isMasqueradesAsUndefined = m_jit.branchTest8(JITCompiler::NonZero, JITCompiler::Address(resultGPR, Structure::typeInfoFlagsOffset()), JITCompiler::TrustedImm32(MasqueradesAsUndefined)); |
| |
| m_jit.move(invert ? TrustedImm32(1) : TrustedImm32(0), resultGPR); |
| notMasqueradesAsUndefined = m_jit.jump(); |
| |
| isMasqueradesAsUndefined.link(&m_jit); |
| GPRReg localGlobalObjectGPR = localGlobalObject.gpr(); |
| GPRReg remoteGlobalObjectGPR = remoteGlobalObject.gpr(); |
| m_jit.move(JITCompiler::TrustedImmPtr(m_jit.graph().globalObjectFor(operand->codeOrigin)), localGlobalObjectGPR); |
| m_jit.loadPtr(JITCompiler::Address(resultGPR, Structure::globalObjectOffset()), remoteGlobalObjectGPR); |
| m_jit.comparePtr(invert ? JITCompiler::NotEqual : JITCompiler::Equal, localGlobalObjectGPR, remoteGlobalObjectGPR, resultGPR); |
| } |
| |
| if (!isKnownCell(operand.node())) { |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| notCell.link(&m_jit); |
| |
| m_jit.move(argGPR, resultGPR); |
| m_jit.and64(JITCompiler::TrustedImm32(~TagBitUndefined), resultGPR); |
| m_jit.compare64(invert ? JITCompiler::NotEqual : JITCompiler::Equal, resultGPR, JITCompiler::TrustedImm32(ValueNull), resultGPR); |
| |
| done.link(&m_jit); |
| } |
| |
| notMasqueradesAsUndefined.link(&m_jit); |
| |
| m_jit.or32(TrustedImm32(ValueFalse), resultGPR); |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::nonSpeculativePeepholeBranchNull(Edge operand, Node* branchNode, bool invert) |
| { |
| BasicBlock* taken = branchNode->takenBlock(); |
| BasicBlock* notTaken = branchNode->notTakenBlock(); |
| |
| if (taken == nextBlock()) { |
| invert = !invert; |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| JSValueOperand arg(this, operand); |
| GPRReg argGPR = arg.gpr(); |
| |
| GPRTemporary result(this, Reuse, arg); |
| GPRReg resultGPR = result.gpr(); |
| |
| JITCompiler::Jump notCell; |
| |
| if (masqueradesAsUndefinedWatchpointIsStillValid()) { |
| if (!isKnownCell(operand.node())) |
| notCell = m_jit.branchTest64(MacroAssembler::NonZero, argGPR, GPRInfo::tagMaskRegister); |
| |
| jump(invert ? taken : notTaken, ForceJump); |
| } else { |
| GPRTemporary localGlobalObject(this); |
| GPRTemporary remoteGlobalObject(this); |
| |
| if (!isKnownCell(operand.node())) |
| notCell = m_jit.branchTest64(MacroAssembler::NonZero, argGPR, GPRInfo::tagMaskRegister); |
| |
| m_jit.loadPtr(JITCompiler::Address(argGPR, JSCell::structureOffset()), resultGPR); |
| branchTest8(JITCompiler::Zero, JITCompiler::Address(resultGPR, Structure::typeInfoFlagsOffset()), JITCompiler::TrustedImm32(MasqueradesAsUndefined), invert ? taken : notTaken); |
| |
| GPRReg localGlobalObjectGPR = localGlobalObject.gpr(); |
| GPRReg remoteGlobalObjectGPR = remoteGlobalObject.gpr(); |
| m_jit.move(TrustedImmPtr(m_jit.graph().globalObjectFor(operand->codeOrigin)), localGlobalObjectGPR); |
| m_jit.loadPtr(JITCompiler::Address(resultGPR, Structure::globalObjectOffset()), remoteGlobalObjectGPR); |
| branchPtr(JITCompiler::Equal, localGlobalObjectGPR, remoteGlobalObjectGPR, invert ? notTaken : taken); |
| } |
| |
| if (!isKnownCell(operand.node())) { |
| jump(notTaken, ForceJump); |
| |
| notCell.link(&m_jit); |
| |
| m_jit.move(argGPR, resultGPR); |
| m_jit.and64(JITCompiler::TrustedImm32(~TagBitUndefined), resultGPR); |
| branch64(invert ? JITCompiler::NotEqual : JITCompiler::Equal, resultGPR, JITCompiler::TrustedImm64(ValueNull), taken); |
| } |
| |
| jump(notTaken); |
| } |
| |
| bool SpeculativeJIT::nonSpeculativeCompareNull(Node* node, Edge operand, bool invert) |
| { |
| unsigned branchIndexInBlock = detectPeepHoleBranch(); |
| if (branchIndexInBlock != UINT_MAX) { |
| Node* branchNode = m_block->at(branchIndexInBlock); |
| |
| RELEASE_ASSERT(node->adjustedRefCount() == 1); |
| |
| nonSpeculativePeepholeBranchNull(operand, branchNode, invert); |
| |
| use(node->child1()); |
| use(node->child2()); |
| m_indexInBlock = branchIndexInBlock; |
| m_currentNode = branchNode; |
| |
| return true; |
| } |
| |
| nonSpeculativeNonPeepholeCompareNull(operand, invert); |
| |
| return false; |
| } |
| |
| void SpeculativeJIT::nonSpeculativePeepholeBranch(Node* node, Node* branchNode, MacroAssembler::RelationalCondition cond, S_JITOperation_EJJ helperFunction) |
| { |
| BasicBlock* taken = branchNode->takenBlock(); |
| BasicBlock* notTaken = branchNode->notTakenBlock(); |
| |
| JITCompiler::ResultCondition callResultCondition = JITCompiler::NonZero; |
| |
| // The branch instruction will branch to the taken block. |
| // If taken is next, switch taken with notTaken & invert the branch condition so we can fall through. |
| if (taken == nextBlock()) { |
| cond = JITCompiler::invert(cond); |
| callResultCondition = JITCompiler::Zero; |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| JSValueOperand arg1(this, node->child1()); |
| JSValueOperand arg2(this, node->child2()); |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg arg2GPR = arg2.gpr(); |
| |
| JITCompiler::JumpList slowPath; |
| |
| if (isKnownNotInteger(node->child1().node()) || isKnownNotInteger(node->child2().node())) { |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| flushRegisters(); |
| callOperation(helperFunction, resultGPR, arg1GPR, arg2GPR); |
| |
| branchTest32(callResultCondition, resultGPR, taken); |
| } else { |
| GPRTemporary result(this, Reuse, arg2); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| if (!isKnownInteger(node->child1().node())) |
| slowPath.append(m_jit.branch64(MacroAssembler::Below, arg1GPR, GPRInfo::tagTypeNumberRegister)); |
| if (!isKnownInteger(node->child2().node())) |
| slowPath.append(m_jit.branch64(MacroAssembler::Below, arg2GPR, GPRInfo::tagTypeNumberRegister)); |
| |
| branch32(cond, arg1GPR, arg2GPR, taken); |
| |
| if (!isKnownInteger(node->child1().node()) || !isKnownInteger(node->child2().node())) { |
| jump(notTaken, ForceJump); |
| |
| slowPath.link(&m_jit); |
| |
| silentSpillAllRegisters(resultGPR); |
| callOperation(helperFunction, resultGPR, arg1GPR, arg2GPR); |
| silentFillAllRegisters(resultGPR); |
| |
| branchTest32(callResultCondition, resultGPR, taken); |
| } |
| } |
| |
| jump(notTaken); |
| |
| m_indexInBlock = m_block->size() - 1; |
| m_currentNode = branchNode; |
| } |
| |
| template<typename JumpType> |
| class CompareAndBoxBooleanSlowPathGenerator |
| : public CallSlowPathGenerator<JumpType, S_JITOperation_EJJ, GPRReg> { |
| public: |
| CompareAndBoxBooleanSlowPathGenerator( |
| JumpType from, SpeculativeJIT* jit, |
| S_JITOperation_EJJ function, GPRReg result, GPRReg arg1, GPRReg arg2) |
| : CallSlowPathGenerator<JumpType, S_JITOperation_EJJ, GPRReg>( |
| from, jit, function, NeedToSpill, result) |
| , m_arg1(arg1) |
| , m_arg2(arg2) |
| { |
| } |
| |
| protected: |
| virtual void generateInternal(SpeculativeJIT* jit) OVERRIDE |
| { |
| this->setUp(jit); |
| this->recordCall(jit->callOperation(this->m_function, this->m_result, m_arg1, m_arg2)); |
| jit->m_jit.and32(JITCompiler::TrustedImm32(1), this->m_result); |
| jit->m_jit.or32(JITCompiler::TrustedImm32(ValueFalse), this->m_result); |
| this->tearDown(jit); |
| } |
| |
| private: |
| GPRReg m_arg1; |
| GPRReg m_arg2; |
| }; |
| |
| void SpeculativeJIT::nonSpeculativeNonPeepholeCompare(Node* node, MacroAssembler::RelationalCondition cond, S_JITOperation_EJJ helperFunction) |
| { |
| JSValueOperand arg1(this, node->child1()); |
| JSValueOperand arg2(this, node->child2()); |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg arg2GPR = arg2.gpr(); |
| |
| JITCompiler::JumpList slowPath; |
| |
| if (isKnownNotInteger(node->child1().node()) || isKnownNotInteger(node->child2().node())) { |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| flushRegisters(); |
| callOperation(helperFunction, resultGPR, arg1GPR, arg2GPR); |
| |
| m_jit.or32(TrustedImm32(ValueFalse), resultGPR); |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean, UseChildrenCalledExplicitly); |
| } else { |
| GPRTemporary result(this, Reuse, arg2); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| if (!isKnownInteger(node->child1().node())) |
| slowPath.append(m_jit.branch64(MacroAssembler::Below, arg1GPR, GPRInfo::tagTypeNumberRegister)); |
| if (!isKnownInteger(node->child2().node())) |
| slowPath.append(m_jit.branch64(MacroAssembler::Below, arg2GPR, GPRInfo::tagTypeNumberRegister)); |
| |
| m_jit.compare32(cond, arg1GPR, arg2GPR, resultGPR); |
| m_jit.or32(TrustedImm32(ValueFalse), resultGPR); |
| |
| if (!isKnownInteger(node->child1().node()) || !isKnownInteger(node->child2().node())) { |
| addSlowPathGenerator(adoptPtr( |
| new CompareAndBoxBooleanSlowPathGenerator<JITCompiler::JumpList>( |
| slowPath, this, helperFunction, resultGPR, arg1GPR, arg2GPR))); |
| } |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean, UseChildrenCalledExplicitly); |
| } |
| } |
| |
| void SpeculativeJIT::nonSpeculativePeepholeStrictEq(Node* node, Node* branchNode, bool invert) |
| { |
| BasicBlock* taken = branchNode->takenBlock(); |
| BasicBlock* notTaken = branchNode->notTakenBlock(); |
| |
| // The branch instruction will branch to the taken block. |
| // If taken is next, switch taken with notTaken & invert the branch condition so we can fall through. |
| if (taken == nextBlock()) { |
| invert = !invert; |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| JSValueOperand arg1(this, node->child1()); |
| JSValueOperand arg2(this, node->child2()); |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg arg2GPR = arg2.gpr(); |
| |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| if (isKnownCell(node->child1().node()) && isKnownCell(node->child2().node())) { |
| // see if we get lucky: if the arguments are cells and they reference the same |
| // cell, then they must be strictly equal. |
| branch64(JITCompiler::Equal, arg1GPR, arg2GPR, invert ? notTaken : taken); |
| |
| silentSpillAllRegisters(resultGPR); |
| callOperation(operationCompareStrictEqCell, resultGPR, arg1GPR, arg2GPR); |
| silentFillAllRegisters(resultGPR); |
| |
| branchTest32(invert ? JITCompiler::Zero : JITCompiler::NonZero, resultGPR, taken); |
| } else { |
| m_jit.or64(arg1GPR, arg2GPR, resultGPR); |
| |
| JITCompiler::Jump twoCellsCase = m_jit.branchTest64(JITCompiler::Zero, resultGPR, GPRInfo::tagMaskRegister); |
| |
| JITCompiler::Jump leftOK = m_jit.branch64(JITCompiler::AboveOrEqual, arg1GPR, GPRInfo::tagTypeNumberRegister); |
| JITCompiler::Jump leftDouble = m_jit.branchTest64(JITCompiler::NonZero, arg1GPR, GPRInfo::tagTypeNumberRegister); |
| leftOK.link(&m_jit); |
| JITCompiler::Jump rightOK = m_jit.branch64(JITCompiler::AboveOrEqual, arg2GPR, GPRInfo::tagTypeNumberRegister); |
| JITCompiler::Jump rightDouble = m_jit.branchTest64(JITCompiler::NonZero, arg2GPR, GPRInfo::tagTypeNumberRegister); |
| rightOK.link(&m_jit); |
| |
| branch64(invert ? JITCompiler::NotEqual : JITCompiler::Equal, arg1GPR, arg2GPR, taken); |
| jump(notTaken, ForceJump); |
| |
| twoCellsCase.link(&m_jit); |
| branch64(JITCompiler::Equal, arg1GPR, arg2GPR, invert ? notTaken : taken); |
| |
| leftDouble.link(&m_jit); |
| rightDouble.link(&m_jit); |
| |
| silentSpillAllRegisters(resultGPR); |
| callOperation(operationCompareStrictEq, resultGPR, arg1GPR, arg2GPR); |
| silentFillAllRegisters(resultGPR); |
| |
| branchTest32(invert ? JITCompiler::Zero : JITCompiler::NonZero, resultGPR, taken); |
| } |
| |
| jump(notTaken); |
| } |
| |
| void SpeculativeJIT::nonSpeculativeNonPeepholeStrictEq(Node* node, bool invert) |
| { |
| JSValueOperand arg1(this, node->child1()); |
| JSValueOperand arg2(this, node->child2()); |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg arg2GPR = arg2.gpr(); |
| |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| arg2.use(); |
| |
| if (isKnownCell(node->child1().node()) && isKnownCell(node->child2().node())) { |
| // see if we get lucky: if the arguments are cells and they reference the same |
| // cell, then they must be strictly equal. |
| // FIXME: this should flush registers instead of silent spill/fill. |
| JITCompiler::Jump notEqualCase = m_jit.branch64(JITCompiler::NotEqual, arg1GPR, arg2GPR); |
| |
| m_jit.move(JITCompiler::TrustedImm64(JSValue::encode(jsBoolean(!invert))), resultGPR); |
| |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| notEqualCase.link(&m_jit); |
| |
| silentSpillAllRegisters(resultGPR); |
| callOperation(operationCompareStrictEqCell, resultGPR, arg1GPR, arg2GPR); |
| silentFillAllRegisters(resultGPR); |
| |
| m_jit.and64(JITCompiler::TrustedImm32(1), resultGPR); |
| m_jit.or32(JITCompiler::TrustedImm32(ValueFalse), resultGPR); |
| |
| done.link(&m_jit); |
| } else { |
| m_jit.or64(arg1GPR, arg2GPR, resultGPR); |
| |
| JITCompiler::JumpList slowPathCases; |
| |
| JITCompiler::Jump twoCellsCase = m_jit.branchTest64(JITCompiler::Zero, resultGPR, GPRInfo::tagMaskRegister); |
| |
| JITCompiler::Jump leftOK = m_jit.branch64(JITCompiler::AboveOrEqual, arg1GPR, GPRInfo::tagTypeNumberRegister); |
| slowPathCases.append(m_jit.branchTest64(JITCompiler::NonZero, arg1GPR, GPRInfo::tagTypeNumberRegister)); |
| leftOK.link(&m_jit); |
| JITCompiler::Jump rightOK = m_jit.branch64(JITCompiler::AboveOrEqual, arg2GPR, GPRInfo::tagTypeNumberRegister); |
| slowPathCases.append(m_jit.branchTest64(JITCompiler::NonZero, arg2GPR, GPRInfo::tagTypeNumberRegister)); |
| rightOK.link(&m_jit); |
| |
| m_jit.compare64(invert ? JITCompiler::NotEqual : JITCompiler::Equal, arg1GPR, arg2GPR, resultGPR); |
| m_jit.or32(JITCompiler::TrustedImm32(ValueFalse), resultGPR); |
| |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| twoCellsCase.link(&m_jit); |
| slowPathCases.append(m_jit.branch64(JITCompiler::NotEqual, arg1GPR, arg2GPR)); |
| |
| m_jit.move(JITCompiler::TrustedImm64(JSValue::encode(jsBoolean(!invert))), resultGPR); |
| |
| addSlowPathGenerator( |
| adoptPtr( |
| new CompareAndBoxBooleanSlowPathGenerator<MacroAssembler::JumpList>( |
| slowPathCases, this, operationCompareStrictEq, resultGPR, arg1GPR, |
| arg2GPR))); |
| |
| done.link(&m_jit); |
| } |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean, UseChildrenCalledExplicitly); |
| } |
| |
| void SpeculativeJIT::emitCall(Node* node) |
| { |
| if (node->op() != Call) |
| RELEASE_ASSERT(node->op() == Construct); |
| |
| // For constructors, the this argument is not passed but we have to make space |
| // for it. |
| int dummyThisArgument = node->op() == Call ? 0 : 1; |
| |
| CallLinkInfo::CallType callType = node->op() == Call ? CallLinkInfo::Call : CallLinkInfo::Construct; |
| |
| Edge calleeEdge = m_jit.graph().m_varArgChildren[node->firstChild()]; |
| JSValueOperand callee(this, calleeEdge); |
| GPRReg calleeGPR = callee.gpr(); |
| use(calleeEdge); |
| |
| // The call instruction's first child is the function; the subsequent children are the |
| // arguments. |
| int numPassedArgs = node->numChildren() - 1; |
| |
| int numArgs = numPassedArgs + dummyThisArgument; |
| |
| m_jit.store32(MacroAssembler::TrustedImm32(numArgs), calleeFramePayloadSlot(numArgs, JSStack::ArgumentCount)); |
| m_jit.store64(GPRInfo::callFrameRegister, calleeFrameCallerFrame(numArgs)); |
| m_jit.store64(calleeGPR, calleeFrameSlot(numArgs, JSStack::Callee)); |
| |
| for (int i = 0; i < numPassedArgs; i++) { |
| Edge argEdge = m_jit.graph().m_varArgChildren[node->firstChild() + 1 + i]; |
| JSValueOperand arg(this, argEdge); |
| GPRReg argGPR = arg.gpr(); |
| use(argEdge); |
| |
| m_jit.store64(argGPR, calleeArgumentSlot(numArgs, i + dummyThisArgument)); |
| } |
| |
| flushRegisters(); |
| |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| JITCompiler::DataLabelPtr targetToCheck; |
| JITCompiler::JumpList slowPath; |
| |
| m_jit.emitStoreCodeOrigin(node->codeOrigin); |
| |
| m_jit.addPtr(TrustedImm32(calleeFrameOffset(numArgs)), GPRInfo::callFrameRegister); |
| |
| slowPath.append(m_jit.branchPtrWithPatch(MacroAssembler::NotEqual, calleeGPR, targetToCheck, MacroAssembler::TrustedImmPtr(0))); |
| |
| m_jit.loadPtr(MacroAssembler::Address(calleeGPR, OBJECT_OFFSETOF(JSFunction, m_scope)), resultGPR); |
| m_jit.store64(resultGPR, MacroAssembler::Address(GPRInfo::callFrameRegister, static_cast<ptrdiff_t>(sizeof(Register)) * JSStack::ScopeChain)); |
| |
| JITCompiler::Call fastCall = m_jit.nearCall(); |
| |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| slowPath.link(&m_jit); |
| |
| m_jit.move(calleeGPR, GPRInfo::nonArgGPR0); |
| JITCompiler::Call slowCall = m_jit.nearCall(); |
| |
| done.link(&m_jit); |
| |
| m_jit.move(GPRInfo::returnValueGPR, resultGPR); |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJS, UseChildrenCalledExplicitly); |
| |
| m_jit.addJSCall(fastCall, slowCall, targetToCheck, callType, calleeGPR, m_currentNode->codeOrigin); |
| } |
| |
| template<bool strict> |
| GPRReg SpeculativeJIT::fillSpeculateInt32Internal(Edge edge, DataFormat& returnFormat) |
| { |
| #if DFG_ENABLE(DEBUG_VERBOSE) |
| dataLogF("SpecInt@%d ", edge->index()); |
| #endif |
| AbstractValue& value = m_state.forNode(edge); |
| SpeculatedType type = value.m_type; |
| ASSERT(edge.useKind() != KnownInt32Use || !(value.m_type & ~SpecInt32)); |
| m_interpreter.filter(value, SpecInt32); |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: { |
| if ((edge->hasConstant() && !isInt32Constant(edge.node())) || info.spillFormat() == DataFormatDouble) { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| returnFormat = DataFormatInt32; |
| return allocate(); |
| } |
| |
| GPRReg gpr = allocate(); |
| |
| if (edge->hasConstant()) { |
| m_gprs.retain(gpr, virtualRegister, SpillOrderConstant); |
| ASSERT(isInt32Constant(edge.node())); |
| m_jit.move(MacroAssembler::Imm32(valueOfInt32Constant(edge.node())), gpr); |
| info.fillInt32(*m_stream, gpr); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| |
| DataFormat spillFormat = info.spillFormat(); |
| |
| RELEASE_ASSERT((spillFormat & DataFormatJS) || spillFormat == DataFormatInt32 || spillFormat == DataFormatInt52 || spillFormat == DataFormatStrictInt52); |
| |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| |
| if (spillFormat == DataFormatJSInt32 || spillFormat == DataFormatInt32) { |
| // If we know this was spilled as an integer we can fill without checking. |
| if (strict) { |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillInt32(*m_stream, gpr); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| if (spillFormat == DataFormatInt32) { |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, gpr); |
| } else |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| returnFormat = DataFormatJSInt32; |
| return gpr; |
| } |
| if (spillFormat == DataFormatInt52 || spillFormat == DataFormatStrictInt52) { |
| // Generally, this can only happen if we've already proved that the |
| // value is an int32. That's because if a value originated as a JSValue |
| // then we would speculate that it's an int32 before representing it as |
| // an int52. Otherwise, if we knowingly produced an int52, then we would |
| // be boxing it into a value using Int52ToValue. This assertion is valid |
| // only because Int52 is something that we introduce at prediction time. |
| // However: we may have an int32-producing node replaced by an |
| // int52-producing node due to CSE. So we must do a check. |
| RELEASE_ASSERT(!(type & ~SpecMachineInt)); |
| if (type & SpecInt52) { |
| GPRReg temp = allocate(); |
| m_jit.signExtend32ToPtr(gpr, temp); |
| // Currently, we can't supply value profiling information here. :-/ |
| speculationCheck( |
| BadType, JSValueRegs(), 0, |
| m_jit.branch64(MacroAssembler::NotEqual, gpr, temp)); |
| unlock(temp); |
| } |
| if (spillFormat == DataFormatStrictInt52) |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| else { |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| m_jit.zeroExtend32ToPtr(gpr, gpr); |
| } |
| info.fillInt32(*m_stream, gpr); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| |
| // Fill as JSValue, and fall through. |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| m_gprs.unlock(gpr); |
| } |
| |
| case DataFormatJS: { |
| RELEASE_ASSERT(!(type & SpecInt52)); |
| // Check the value is an integer. |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| if (type & ~SpecInt32) |
| speculationCheck(BadType, JSValueRegs(gpr), edge, m_jit.branch64(MacroAssembler::Below, gpr, GPRInfo::tagTypeNumberRegister)); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| // If !strict we're done, return. |
| if (!strict) { |
| returnFormat = DataFormatJSInt32; |
| return gpr; |
| } |
| // else fall through & handle as DataFormatJSInt32. |
| m_gprs.unlock(gpr); |
| } |
| |
| case DataFormatJSInt32: { |
| // In a strict fill we need to strip off the value tag. |
| if (strict) { |
| GPRReg gpr = info.gpr(); |
| GPRReg result; |
| // If the register has already been locked we need to take a copy. |
| // If not, we'll zero extend in place, so mark on the info that this is now type DataFormatInt32, not DataFormatJSInt32. |
| if (m_gprs.isLocked(gpr)) |
| result = allocate(); |
| else { |
| m_gprs.lock(gpr); |
| info.fillInt32(*m_stream, gpr); |
| result = gpr; |
| } |
| m_jit.zeroExtend32ToPtr(gpr, result); |
| returnFormat = DataFormatInt32; |
| return result; |
| } |
| |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| returnFormat = DataFormatJSInt32; |
| return gpr; |
| } |
| |
| case DataFormatInt32: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| |
| case DataFormatStrictInt52: |
| case DataFormatInt52: { |
| GPRReg gpr = info.gpr(); |
| GPRReg result; |
| DataFormat oldFormat = info.registerFormat(); |
| if (m_gprs.isLocked(gpr)) { |
| result = allocate(); |
| m_jit.move(gpr, result); |
| } else { |
| lock(gpr); |
| info.fillInt32(*m_stream, gpr); |
| result = gpr; |
| } |
| RELEASE_ASSERT(!(type & ~SpecMachineInt)); |
| if (oldFormat == DataFormatInt52) |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), result); |
| if (type & SpecInt52) { |
| GPRReg temp = allocate(); |
| m_jit.signExtend32ToPtr(result, temp); |
| // Currently, we can't supply value profiling information here. :-/ |
| speculationCheck( |
| BadType, JSValueRegs(), 0, |
| m_jit.branch64(MacroAssembler::NotEqual, result, temp)); |
| unlock(temp); |
| } |
| m_jit.zeroExtend32ToPtr(result, result); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| |
| case DataFormatDouble: |
| case DataFormatJSDouble: { |
| if (edge->hasConstant() && isInt32Constant(edge.node())) { |
| GPRReg gpr = allocate(); |
| ASSERT(isInt32Constant(edge.node())); |
| m_jit.move(MacroAssembler::Imm32(valueOfInt32Constant(edge.node())), gpr); |
| returnFormat = DataFormatInt32; |
| return gpr; |
| } |
| } |
| case DataFormatCell: |
| case DataFormatBoolean: |
| case DataFormatJSCell: |
| case DataFormatJSBoolean: { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| returnFormat = DataFormatInt32; |
| return allocate(); |
| } |
| |
| case DataFormatStorage: |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidGPRReg; |
| } |
| } |
| |
| GPRReg SpeculativeJIT::fillSpeculateInt32(Edge edge, DataFormat& returnFormat) |
| { |
| return fillSpeculateInt32Internal<false>(edge, returnFormat); |
| } |
| |
| GPRReg SpeculativeJIT::fillSpeculateInt32Strict(Edge edge) |
| { |
| DataFormat mustBeDataFormatInt32; |
| GPRReg result = fillSpeculateInt32Internal<true>(edge, mustBeDataFormatInt32); |
| RELEASE_ASSERT(mustBeDataFormatInt32 == DataFormatInt32); |
| return result; |
| } |
| |
| GPRReg SpeculativeJIT::fillSpeculateInt52(Edge edge, DataFormat desiredFormat) |
| { |
| ASSERT(desiredFormat == DataFormatInt52 || desiredFormat == DataFormatStrictInt52); |
| AbstractValue& value = m_state.forNode(edge); |
| SpeculatedType type = value.m_type; |
| m_interpreter.filter(value, SpecMachineInt); |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: { |
| if ((edge->hasConstant() && !valueOfJSConstant(edge.node()).isMachineInt()) || info.spillFormat() == DataFormatDouble) { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| GPRReg gpr = allocate(); |
| |
| if (edge->hasConstant()) { |
| JSValue jsValue = valueOfJSConstant(edge.node()); |
| ASSERT(jsValue.isMachineInt()); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderConstant); |
| int64_t value = jsValue.asMachineInt(); |
| if (desiredFormat == DataFormatInt52) |
| value = value << JSValue::int52ShiftAmount; |
| m_jit.move(MacroAssembler::Imm64(value), gpr); |
| info.fillGPR(*m_stream, gpr, desiredFormat); |
| return gpr; |
| } |
| |
| DataFormat spillFormat = info.spillFormat(); |
| |
| RELEASE_ASSERT((spillFormat & DataFormatJS) || spillFormat == DataFormatInt32 || spillFormat == DataFormatInt52 || spillFormat == DataFormatStrictInt52); |
| |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| |
| if (spillFormat == DataFormatJSInt32 || spillFormat == DataFormatInt32) { |
| // If we know this was spilled as an integer we can fill without checking. |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| m_jit.signExtend32ToPtr(gpr, gpr); |
| if (desiredFormat == DataFormatStrictInt52) { |
| info.fillStrictInt52(*m_stream, gpr); |
| return gpr; |
| } |
| m_jit.lshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| info.fillInt52(*m_stream, gpr); |
| return gpr; |
| } |
| if (spillFormat == DataFormatInt52 || spillFormat == DataFormatStrictInt52) { |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| if (desiredFormat == DataFormatStrictInt52) { |
| if (spillFormat == DataFormatInt52) |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| info.fillStrictInt52(*m_stream, gpr); |
| return gpr; |
| } |
| if (spillFormat == DataFormatStrictInt52) |
| m_jit.lshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| info.fillInt52(*m_stream, gpr); |
| return gpr; |
| } |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| |
| // Fill as JSValue, and fall through. |
| info.fillJSValue(*m_stream, gpr, DataFormatJSInt32); |
| m_gprs.unlock(gpr); |
| } |
| |
| case DataFormatJS: { |
| // Check the value is an integer. Note that we would *like* to unbox an Int52 |
| // at this point but this is too costly. We only *prove* that this is an Int52 |
| // even though we check if it's an int32. |
| GPRReg gpr = info.gpr(); |
| GPRReg result; |
| if (m_gprs.isLocked(gpr)) { |
| result = allocate(); |
| m_jit.move(gpr, result); |
| } else { |
| m_gprs.lock(gpr); |
| result = gpr; |
| } |
| if (type & ~SpecInt32) |
| speculationCheck(BadType, JSValueRegs(result), edge, m_jit.branch64(MacroAssembler::Below, result, GPRInfo::tagTypeNumberRegister)); |
| if (result == gpr) // The not-already-locked, so fill in-place, case. |
| info.fillInt52(*m_stream, gpr, desiredFormat); |
| m_jit.signExtend32ToPtr(result, result); |
| if (desiredFormat == DataFormatInt52) |
| m_jit.lshift64(TrustedImm32(JSValue::int52ShiftAmount), result); |
| return result; |
| } |
| |
| case DataFormatInt32: |
| case DataFormatJSInt32: { |
| GPRReg gpr = info.gpr(); |
| GPRReg result; |
| if (m_gprs.isLocked(gpr)) { |
| result = allocate(); |
| m_jit.move(gpr, result); |
| } else { |
| m_gprs.lock(gpr); |
| info.fillInt52(*m_stream, gpr, desiredFormat); |
| result = gpr; |
| } |
| m_jit.signExtend32ToPtr(result, result); |
| if (desiredFormat == DataFormatInt52) |
| m_jit.lshift64(TrustedImm32(JSValue::int52ShiftAmount), result); |
| return result; |
| } |
| |
| case DataFormatStrictInt52: { |
| GPRReg gpr = info.gpr(); |
| bool wasLocked = m_gprs.isLocked(gpr); |
| lock(gpr); |
| if (desiredFormat == DataFormatStrictInt52) |
| return gpr; |
| if (wasLocked) { |
| GPRReg result = allocate(); |
| m_jit.move(gpr, result); |
| unlock(gpr); |
| gpr = result; |
| } else |
| info.fillInt52(*m_stream, gpr); |
| m_jit.lshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| return gpr; |
| } |
| |
| case DataFormatInt52: { |
| GPRReg gpr = info.gpr(); |
| bool wasLocked = m_gprs.isLocked(gpr); |
| lock(gpr); |
| if (desiredFormat == DataFormatInt52) |
| return gpr; |
| if (wasLocked) { |
| GPRReg result = allocate(); |
| m_jit.move(gpr, result); |
| unlock(gpr); |
| gpr = result; |
| } else |
| info.fillStrictInt52(*m_stream, gpr); |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), gpr); |
| return gpr; |
| } |
| |
| case DataFormatDouble: |
| case DataFormatJSDouble: |
| if (edge->hasConstant()) { |
| JSValue jsValue = valueOfJSConstant(edge.node()); |
| if (jsValue.isMachineInt()) { |
| int64_t value = jsValue.asMachineInt(); |
| if (desiredFormat == DataFormatInt52) |
| value = value << JSValue::int52ShiftAmount; |
| GPRReg gpr = allocate(); |
| m_jit.move(MacroAssembler::Imm64(value), gpr); |
| return gpr; |
| } |
| } |
| |
| case DataFormatCell: |
| case DataFormatBoolean: |
| case DataFormatJSCell: |
| case DataFormatJSBoolean: { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| case DataFormatStorage: |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidGPRReg; |
| } |
| } |
| |
| FPRReg SpeculativeJIT::fillSpeculateDouble(Edge edge) |
| { |
| #if DFG_ENABLE(DEBUG_VERBOSE) |
| dataLogF("SpecDouble@%d ", edge->index()); |
| #endif |
| AbstractValue& value = m_state.forNode(edge); |
| SpeculatedType type = value.m_type; |
| ASSERT(edge.useKind() != KnownNumberUse || !(value.m_type & ~SpecFullNumber)); |
| m_interpreter.filter(value, SpecFullNumber); |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| if (info.registerFormat() == DataFormatNone) { |
| if (edge->hasConstant()) { |
| GPRReg gpr = allocate(); |
| |
| if (isInt32Constant(edge.node())) { |
| FPRReg fpr = fprAllocate(); |
| m_jit.move(MacroAssembler::Imm64(reinterpretDoubleToInt64(static_cast<double>(valueOfInt32Constant(edge.node())))), gpr); |
| m_jit.move64ToDouble(gpr, fpr); |
| unlock(gpr); |
| |
| // Don't fill double here since that will lead to confusion: the |
| // register allocator will now think that this is a double while |
| // everyone else thinks it's an integer. |
| return fpr; |
| } |
| if (isNumberConstant(edge.node())) { |
| FPRReg fpr = fprAllocate(); |
| m_jit.move(MacroAssembler::Imm64(reinterpretDoubleToInt64(valueOfNumberConstant(edge.node()))), gpr); |
| m_jit.move64ToDouble(gpr, fpr); |
| unlock(gpr); |
| |
| m_fprs.retain(fpr, virtualRegister, SpillOrderDouble); |
| info.fillDouble(*m_stream, fpr); |
| return fpr; |
| } |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return fprAllocate(); |
| } |
| |
| DataFormat spillFormat = info.spillFormat(); |
| switch (spillFormat) { |
| case DataFormatDouble: { |
| FPRReg fpr = fprAllocate(); |
| m_jit.loadDouble(JITCompiler::addressFor(virtualRegister), fpr); |
| m_fprs.retain(fpr, virtualRegister, SpillOrderDouble); |
| info.fillDouble(*m_stream, fpr); |
| return fpr; |
| } |
| |
| case DataFormatInt32: { |
| GPRReg gpr = allocate(); |
| |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load32(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillInt32(*m_stream, gpr); |
| unlock(gpr); |
| break; |
| } |
| |
| case DataFormatInt52: { |
| GPRReg gpr = allocate(); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillInt52(*m_stream, gpr); |
| unlock(gpr); |
| break; |
| } |
| |
| case DataFormatStrictInt52: { |
| GPRReg gpr = allocate(); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillStrictInt52(*m_stream, gpr); |
| unlock(gpr); |
| break; |
| } |
| |
| default: |
| GPRReg gpr = allocate(); |
| |
| RELEASE_ASSERT(spillFormat & DataFormatJS); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| info.fillJSValue(*m_stream, gpr, spillFormat); |
| unlock(gpr); |
| break; |
| } |
| } |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: // Should have filled, above. |
| case DataFormatBoolean: // This type never occurs. |
| case DataFormatStorage: |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| case DataFormatCell: |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return fprAllocate(); |
| |
| case DataFormatJSCell: |
| case DataFormatJS: |
| case DataFormatJSBoolean: { |
| GPRReg jsValueGpr = info.gpr(); |
| m_gprs.lock(jsValueGpr); |
| FPRReg fpr = fprAllocate(); |
| GPRReg tempGpr = allocate(); |
| |
| JITCompiler::Jump isInteger = m_jit.branch64(MacroAssembler::AboveOrEqual, jsValueGpr, GPRInfo::tagTypeNumberRegister); |
| |
| if (type & ~SpecFullNumber) |
| speculationCheck(BadType, JSValueRegs(jsValueGpr), edge, m_jit.branchTest64(MacroAssembler::Zero, jsValueGpr, GPRInfo::tagTypeNumberRegister)); |
| |
| // First, if we get here we have a double encoded as a JSValue |
| m_jit.move(jsValueGpr, tempGpr); |
| unboxDouble(tempGpr, fpr); |
| JITCompiler::Jump hasUnboxedDouble = m_jit.jump(); |
| |
| // Finally, handle integers. |
| isInteger.link(&m_jit); |
| m_jit.convertInt32ToDouble(jsValueGpr, fpr); |
| hasUnboxedDouble.link(&m_jit); |
| |
| m_gprs.release(jsValueGpr); |
| m_gprs.unlock(jsValueGpr); |
| m_gprs.unlock(tempGpr); |
| m_fprs.retain(fpr, virtualRegister, SpillOrderDouble); |
| info.fillDouble(*m_stream, fpr); |
| info.killSpilled(); |
| return fpr; |
| } |
| |
| case DataFormatJSInt32: |
| case DataFormatInt32: { |
| FPRReg fpr = fprAllocate(); |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| m_jit.convertInt32ToDouble(gpr, fpr); |
| m_gprs.unlock(gpr); |
| return fpr; |
| } |
| |
| case DataFormatInt52: { |
| FPRReg fpr = fprAllocate(); |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| GPRReg temp = allocate(); |
| m_jit.move(gpr, temp); |
| m_jit.rshift64(TrustedImm32(JSValue::int52ShiftAmount), temp); |
| m_jit.convertInt64ToDouble(temp, fpr); |
| unlock(temp); |
| m_gprs.unlock(gpr); |
| return fpr; |
| } |
| |
| case DataFormatStrictInt52: { |
| FPRReg fpr = fprAllocate(); |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| m_jit.convertInt64ToDouble(gpr, fpr); |
| m_gprs.unlock(gpr); |
| return fpr; |
| } |
| |
| // Unbox the double |
| case DataFormatJSDouble: { |
| GPRReg gpr = info.gpr(); |
| FPRReg fpr = fprAllocate(); |
| if (m_gprs.isLocked(gpr)) { |
| // Make sure we don't trample gpr if it is in use. |
| GPRReg temp = allocate(); |
| m_jit.move(gpr, temp); |
| unboxDouble(temp, fpr); |
| unlock(temp); |
| } else |
| unboxDouble(gpr, fpr); |
| |
| m_gprs.release(gpr); |
| m_fprs.retain(fpr, virtualRegister, SpillOrderDouble); |
| |
| info.fillDouble(*m_stream, fpr); |
| return fpr; |
| } |
| |
| case DataFormatDouble: { |
| FPRReg fpr = info.fpr(); |
| m_fprs.lock(fpr); |
| return fpr; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidFPRReg; |
| } |
| } |
| |
| GPRReg SpeculativeJIT::fillSpeculateCell(Edge edge) |
| { |
| #if DFG_ENABLE(DEBUG_VERBOSE) |
| dataLogF("SpecCell@%d ", edge->index()); |
| #endif |
| AbstractValue& value = m_state.forNode(edge); |
| SpeculatedType type = value.m_type; |
| ASSERT((edge.useKind() != KnownCellUse && edge.useKind() != KnownStringUse) || !(value.m_type & ~SpecCell)); |
| m_interpreter.filter(value, SpecCell); |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: { |
| if (info.spillFormat() == DataFormatInt32 || info.spillFormat() == DataFormatDouble) { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| GPRReg gpr = allocate(); |
| |
| if (edge->hasConstant()) { |
| JSValue jsValue = valueOfJSConstant(edge.node()); |
| if (jsValue.isCell()) { |
| m_gprs.retain(gpr, virtualRegister, SpillOrderConstant); |
| m_jit.move(MacroAssembler::TrustedImm64(JSValue::encode(jsValue)), gpr); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSCell); |
| return gpr; |
| } |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return gpr; |
| } |
| RELEASE_ASSERT(info.spillFormat() & DataFormatJS); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| |
| info.fillJSValue(*m_stream, gpr, DataFormatJS); |
| if (type & ~SpecCell) |
| speculationCheck(BadType, JSValueRegs(gpr), edge, m_jit.branchTest64(MacroAssembler::NonZero, gpr, GPRInfo::tagMaskRegister)); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSCell); |
| return gpr; |
| } |
| |
| case DataFormatCell: |
| case DataFormatJSCell: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| #if DFG_ENABLE(JIT_ASSERT) |
| MacroAssembler::Jump checkCell = m_jit.branchTest64(MacroAssembler::Zero, gpr, GPRInfo::tagMaskRegister); |
| m_jit.breakpoint(); |
| checkCell.link(&m_jit); |
| #endif |
| return gpr; |
| } |
| |
| case DataFormatJS: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| if (type & ~SpecCell) |
| speculationCheck(BadType, JSValueRegs(gpr), edge, m_jit.branchTest64(MacroAssembler::NonZero, gpr, GPRInfo::tagMaskRegister)); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSCell); |
| return gpr; |
| } |
| |
| case DataFormatJSInt32: |
| case DataFormatInt32: |
| case DataFormatJSDouble: |
| case DataFormatDouble: |
| case DataFormatJSBoolean: |
| case DataFormatBoolean: |
| case DataFormatInt52: |
| case DataFormatStrictInt52: { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| case DataFormatStorage: |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidGPRReg; |
| } |
| } |
| |
| GPRReg SpeculativeJIT::fillSpeculateBoolean(Edge edge) |
| { |
| #if DFG_ENABLE(DEBUG_VERBOSE) |
| dataLogF("SpecBool@%d ", edge->index()); |
| #endif |
| AbstractValue& value = m_state.forNode(edge); |
| SpeculatedType type = value.m_type; |
| m_interpreter.filter(value, SpecBoolean); |
| VirtualRegister virtualRegister = edge->virtualRegister(); |
| GenerationInfo& info = generationInfoFromVirtualRegister(virtualRegister); |
| |
| switch (info.registerFormat()) { |
| case DataFormatNone: { |
| if (info.spillFormat() == DataFormatInt32 || info.spillFormat() == DataFormatDouble) { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| GPRReg gpr = allocate(); |
| |
| if (edge->hasConstant()) { |
| JSValue jsValue = valueOfJSConstant(edge.node()); |
| if (jsValue.isBoolean()) { |
| m_gprs.retain(gpr, virtualRegister, SpillOrderConstant); |
| m_jit.move(MacroAssembler::TrustedImm64(JSValue::encode(jsValue)), gpr); |
| info.fillJSValue(*m_stream, gpr, DataFormatJSBoolean); |
| return gpr; |
| } |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return gpr; |
| } |
| RELEASE_ASSERT(info.spillFormat() & DataFormatJS); |
| m_gprs.retain(gpr, virtualRegister, SpillOrderSpilled); |
| m_jit.load64(JITCompiler::addressFor(virtualRegister), gpr); |
| |
| info.fillJSValue(*m_stream, gpr, DataFormatJS); |
| if (type & ~SpecBoolean) { |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), gpr); |
| speculationCheck(BadType, JSValueRegs(gpr), edge, m_jit.branchTest64(MacroAssembler::NonZero, gpr, TrustedImm32(static_cast<int32_t>(~1))), SpeculationRecovery(BooleanSpeculationCheck, gpr, InvalidGPRReg)); |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), gpr); |
| } |
| info.fillJSValue(*m_stream, gpr, DataFormatJSBoolean); |
| return gpr; |
| } |
| |
| case DataFormatBoolean: |
| case DataFormatJSBoolean: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| return gpr; |
| } |
| |
| case DataFormatJS: { |
| GPRReg gpr = info.gpr(); |
| m_gprs.lock(gpr); |
| if (type & ~SpecBoolean) { |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), gpr); |
| speculationCheck(BadType, JSValueRegs(gpr), edge, m_jit.branchTest64(MacroAssembler::NonZero, gpr, TrustedImm32(static_cast<int32_t>(~1))), SpeculationRecovery(BooleanSpeculationCheck, gpr, InvalidGPRReg)); |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), gpr); |
| } |
| info.fillJSValue(*m_stream, gpr, DataFormatJSBoolean); |
| return gpr; |
| } |
| |
| case DataFormatJSInt32: |
| case DataFormatInt32: |
| case DataFormatJSDouble: |
| case DataFormatDouble: |
| case DataFormatJSCell: |
| case DataFormatCell: |
| case DataFormatInt52: |
| case DataFormatStrictInt52: { |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| return allocate(); |
| } |
| |
| case DataFormatStorage: |
| RELEASE_ASSERT_NOT_REACHED(); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| return InvalidGPRReg; |
| } |
| } |
| |
| JITCompiler::Jump SpeculativeJIT::convertToDouble(GPRReg value, FPRReg result, GPRReg tmp) |
| { |
| JITCompiler::Jump isInteger = m_jit.branch64(MacroAssembler::AboveOrEqual, value, GPRInfo::tagTypeNumberRegister); |
| |
| JITCompiler::Jump notNumber = m_jit.branchTest64(MacroAssembler::Zero, value, GPRInfo::tagTypeNumberRegister); |
| |
| m_jit.move(value, tmp); |
| unboxDouble(tmp, result); |
| |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| isInteger.link(&m_jit); |
| |
| m_jit.convertInt32ToDouble(value, result); |
| |
| done.link(&m_jit); |
| |
| return notNumber; |
| } |
| |
| void SpeculativeJIT::compileObjectEquality(Node* node) |
| { |
| SpeculateCellOperand op1(this, node->child1()); |
| SpeculateCellOperand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| GPRReg op1GPR = op1.gpr(); |
| GPRReg op2GPR = op2.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| if (masqueradesAsUndefinedWatchpointIsStillValid()) { |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), node->child1(), SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op1GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op2GPR), node->child2(), SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op2GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| GPRTemporary structure(this); |
| GPRReg structureGPR = structure.gpr(); |
| |
| m_jit.loadPtr(MacroAssembler::Address(op1GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), node->child1(), SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueSource::unboxedCell(op1GPR), node->child1(), |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| |
| m_jit.loadPtr(MacroAssembler::Address(op2GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op2GPR), node->child2(), SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueSource::unboxedCell(op2GPR), node->child2(), |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| } |
| |
| MacroAssembler::Jump falseCase = m_jit.branch64(MacroAssembler::NotEqual, op1GPR, op2GPR); |
| m_jit.move(TrustedImm32(ValueTrue), resultGPR); |
| MacroAssembler::Jump done = m_jit.jump(); |
| falseCase.link(&m_jit); |
| m_jit.move(TrustedImm32(ValueFalse), resultGPR); |
| done.link(&m_jit); |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compileObjectToObjectOrOtherEquality(Edge leftChild, Edge rightChild) |
| { |
| SpeculateCellOperand op1(this, leftChild); |
| JSValueOperand op2(this, rightChild, ManualOperandSpeculation); |
| GPRTemporary result(this); |
| |
| GPRReg op1GPR = op1.gpr(); |
| GPRReg op2GPR = op2.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| GPRTemporary structure; |
| GPRReg structureGPR = InvalidGPRReg; |
| |
| bool masqueradesAsUndefinedWatchpointValid = |
| masqueradesAsUndefinedWatchpointIsStillValid(); |
| |
| if (!masqueradesAsUndefinedWatchpointValid) { |
| // The masquerades as undefined case will use the structure register, so allocate it here. |
| // Do this at the top of the function to avoid branching around a register allocation. |
| GPRTemporary realStructure(this); |
| structure.adopt(realStructure); |
| structureGPR = structure.gpr(); |
| } |
| |
| if (masqueradesAsUndefinedWatchpointValid) { |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), leftChild, SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op1GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(op1GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), leftChild, SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueSource::unboxedCell(op1GPR), leftChild, |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| } |
| |
| // It seems that most of the time when programs do a == b where b may be either null/undefined |
| // or an object, b is usually an object. Balance the branches to make that case fast. |
| MacroAssembler::Jump rightNotCell = |
| m_jit.branchTest64(MacroAssembler::NonZero, op2GPR, GPRInfo::tagMaskRegister); |
| |
| // We know that within this branch, rightChild must be a cell. |
| if (masqueradesAsUndefinedWatchpointValid) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(op2GPR), rightChild, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op2GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(op2GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueRegs(op2GPR), rightChild, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueRegs(op2GPR), rightChild, |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| } |
| |
| // At this point we know that we can perform a straight-forward equality comparison on pointer |
| // values because both left and right are pointers to objects that have no special equality |
| // protocols. |
| MacroAssembler::Jump falseCase = m_jit.branch64(MacroAssembler::NotEqual, op1GPR, op2GPR); |
| MacroAssembler::Jump trueCase = m_jit.jump(); |
| |
| rightNotCell.link(&m_jit); |
| |
| // We know that within this branch, rightChild must not be a cell. Check if that is enough to |
| // prove that it is either null or undefined. |
| if (needsTypeCheck(rightChild, SpecCell | SpecOther)) { |
| m_jit.move(op2GPR, resultGPR); |
| m_jit.and64(MacroAssembler::TrustedImm32(~TagBitUndefined), resultGPR); |
| |
| typeCheck( |
| JSValueRegs(op2GPR), rightChild, SpecCell | SpecOther, |
| m_jit.branch64( |
| MacroAssembler::NotEqual, resultGPR, |
| MacroAssembler::TrustedImm64(ValueNull))); |
| } |
| |
| falseCase.link(&m_jit); |
| m_jit.move(TrustedImm32(ValueFalse), resultGPR); |
| MacroAssembler::Jump done = m_jit.jump(); |
| trueCase.link(&m_jit); |
| m_jit.move(TrustedImm32(ValueTrue), resultGPR); |
| done.link(&m_jit); |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compilePeepHoleObjectToObjectOrOtherEquality(Edge leftChild, Edge rightChild, Node* branchNode) |
| { |
| BasicBlock* taken = branchNode->takenBlock(); |
| BasicBlock* notTaken = branchNode->notTakenBlock(); |
| |
| SpeculateCellOperand op1(this, leftChild); |
| JSValueOperand op2(this, rightChild, ManualOperandSpeculation); |
| GPRTemporary result(this); |
| |
| GPRReg op1GPR = op1.gpr(); |
| GPRReg op2GPR = op2.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| GPRTemporary structure; |
| GPRReg structureGPR = InvalidGPRReg; |
| |
| bool masqueradesAsUndefinedWatchpointValid = |
| masqueradesAsUndefinedWatchpointIsStillValid(); |
| |
| if (!masqueradesAsUndefinedWatchpointValid) { |
| // The masquerades as undefined case will use the structure register, so allocate it here. |
| // Do this at the top of the function to avoid branching around a register allocation. |
| GPRTemporary realStructure(this); |
| structure.adopt(realStructure); |
| structureGPR = structure.gpr(); |
| } |
| |
| if (masqueradesAsUndefinedWatchpointValid) { |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), leftChild, SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op1GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(op1GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueSource::unboxedCell(op1GPR), leftChild, SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueSource::unboxedCell(op1GPR), leftChild, |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| } |
| |
| // It seems that most of the time when programs do a == b where b may be either null/undefined |
| // or an object, b is usually an object. Balance the branches to make that case fast. |
| MacroAssembler::Jump rightNotCell = |
| m_jit.branchTest64(MacroAssembler::NonZero, op2GPR, GPRInfo::tagMaskRegister); |
| |
| // We know that within this branch, rightChild must be a cell. |
| if (masqueradesAsUndefinedWatchpointValid) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(op2GPR), rightChild, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(op2GPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(op2GPR, JSCell::structureOffset()), structureGPR); |
| DFG_TYPE_CHECK( |
| JSValueRegs(op2GPR), rightChild, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| speculationCheck(BadType, JSValueRegs(op2GPR), rightChild, |
| m_jit.branchTest8( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined))); |
| } |
| |
| // At this point we know that we can perform a straight-forward equality comparison on pointer |
| // values because both left and right are pointers to objects that have no special equality |
| // protocols. |
| branch64(MacroAssembler::Equal, op1GPR, op2GPR, taken); |
| |
| // We know that within this branch, rightChild must not be a cell. Check if that is enough to |
| // prove that it is either null or undefined. |
| if (!needsTypeCheck(rightChild, SpecCell | SpecOther)) |
| rightNotCell.link(&m_jit); |
| else { |
| jump(notTaken, ForceJump); |
| |
| rightNotCell.link(&m_jit); |
| m_jit.move(op2GPR, resultGPR); |
| m_jit.and64(MacroAssembler::TrustedImm32(~TagBitUndefined), resultGPR); |
| |
| typeCheck( |
| JSValueRegs(op2GPR), rightChild, SpecCell | SpecOther, m_jit.branch64( |
| MacroAssembler::NotEqual, resultGPR, |
| MacroAssembler::TrustedImm64(ValueNull))); |
| } |
| |
| jump(notTaken); |
| } |
| |
| void SpeculativeJIT::compileInt32Compare(Node* node, MacroAssembler::RelationalCondition condition) |
| { |
| SpeculateInt32Operand op1(this, node->child1()); |
| SpeculateInt32Operand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op1, op2); |
| |
| m_jit.compare32(condition, op1.gpr(), op2.gpr(), result.gpr()); |
| |
| // If we add a DataFormatBool, we should use it here. |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compileInt52Compare(Node* node, MacroAssembler::RelationalCondition condition) |
| { |
| SpeculateWhicheverInt52Operand op1(this, node->child1()); |
| SpeculateWhicheverInt52Operand op2(this, node->child2(), op1); |
| GPRTemporary result(this, Reuse, op1, op2); |
| |
| m_jit.compare64(condition, op1.gpr(), op2.gpr(), result.gpr()); |
| |
| // If we add a DataFormatBool, we should use it here. |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compilePeepHoleInt52Branch(Node* node, Node* branchNode, JITCompiler::RelationalCondition condition) |
| { |
| BasicBlock* taken = branchNode->takenBlock(); |
| BasicBlock* notTaken = branchNode->notTakenBlock(); |
| |
| // The branch instruction will branch to the taken block. |
| // If taken is next, switch taken with notTaken & invert the branch condition so we can fall through. |
| if (taken == nextBlock()) { |
| condition = JITCompiler::invert(condition); |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| SpeculateWhicheverInt52Operand op1(this, node->child1()); |
| SpeculateWhicheverInt52Operand op2(this, node->child2(), op1); |
| |
| branch64(condition, op1.gpr(), op2.gpr(), taken); |
| jump(notTaken); |
| } |
| |
| void SpeculativeJIT::compileDoubleCompare(Node* node, MacroAssembler::DoubleCondition condition) |
| { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| SpeculateDoubleOperand op2(this, node->child2()); |
| GPRTemporary result(this); |
| |
| m_jit.move(TrustedImm32(ValueTrue), result.gpr()); |
| MacroAssembler::Jump trueCase = m_jit.branchDouble(condition, op1.fpr(), op2.fpr()); |
| m_jit.xor64(TrustedImm32(true), result.gpr()); |
| trueCase.link(&m_jit); |
| |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compileValueAdd(Node* node) |
| { |
| JSValueOperand op1(this, node->child1()); |
| JSValueOperand op2(this, node->child2()); |
| |
| GPRReg op1GPR = op1.gpr(); |
| GPRReg op2GPR = op2.gpr(); |
| |
| flushRegisters(); |
| |
| GPRResult result(this); |
| if (isKnownNotNumber(node->child1().node()) || isKnownNotNumber(node->child2().node())) |
| callOperation(operationValueAddNotNumber, result.gpr(), op1GPR, op2GPR); |
| else |
| callOperation(operationValueAdd, result.gpr(), op1GPR, op2GPR); |
| |
| jsValueResult(result.gpr(), node); |
| } |
| |
| void SpeculativeJIT::compileObjectOrOtherLogicalNot(Edge nodeUse) |
| { |
| JSValueOperand value(this, nodeUse, ManualOperandSpeculation); |
| GPRTemporary result(this); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| GPRTemporary structure; |
| GPRReg structureGPR = InvalidGPRReg; |
| |
| bool masqueradesAsUndefinedWatchpointValid = |
| masqueradesAsUndefinedWatchpointIsStillValid(); |
| |
| if (!masqueradesAsUndefinedWatchpointValid) { |
| // The masquerades as undefined case will use the structure register, so allocate it here. |
| // Do this at the top of the function to avoid branching around a register allocation. |
| GPRTemporary realStructure(this); |
| structure.adopt(realStructure); |
| structureGPR = structure.gpr(); |
| } |
| |
| MacroAssembler::Jump notCell = m_jit.branchTest64(MacroAssembler::NonZero, valueGPR, GPRInfo::tagMaskRegister); |
| if (masqueradesAsUndefinedWatchpointValid) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueGPR), nodeUse, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(valueGPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(valueGPR, JSCell::structureOffset()), structureGPR); |
| |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueGPR), nodeUse, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| structureGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| |
| MacroAssembler::Jump isNotMasqueradesAsUndefined = |
| m_jit.branchTest8( |
| MacroAssembler::Zero, |
| MacroAssembler::Address(structureGPR, Structure::typeInfoFlagsOffset()), |
| MacroAssembler::TrustedImm32(MasqueradesAsUndefined)); |
| |
| speculationCheck(BadType, JSValueRegs(valueGPR), nodeUse, |
| m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(structureGPR, Structure::globalObjectOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.graph().globalObjectFor(m_currentNode->codeOrigin)))); |
| |
| isNotMasqueradesAsUndefined.link(&m_jit); |
| } |
| m_jit.move(TrustedImm32(ValueFalse), resultGPR); |
| MacroAssembler::Jump done = m_jit.jump(); |
| |
| notCell.link(&m_jit); |
| |
| if (needsTypeCheck(nodeUse, SpecCell | SpecOther)) { |
| m_jit.move(valueGPR, resultGPR); |
| m_jit.and64(MacroAssembler::TrustedImm32(~TagBitUndefined), resultGPR); |
| typeCheck( |
| JSValueRegs(valueGPR), nodeUse, SpecCell | SpecOther, m_jit.branch64( |
| MacroAssembler::NotEqual, |
| resultGPR, |
| MacroAssembler::TrustedImm64(ValueNull))); |
| } |
| m_jit.move(TrustedImm32(ValueTrue), resultGPR); |
| |
| done.link(&m_jit); |
| |
| jsValueResult(resultGPR, m_currentNode, DataFormatJSBoolean); |
| } |
| |
| void SpeculativeJIT::compileLogicalNot(Node* node) |
| { |
| switch (node->child1().useKind()) { |
| case ObjectOrOtherUse: { |
| compileObjectOrOtherLogicalNot(node->child1()); |
| return; |
| } |
| |
| case Int32Use: { |
| SpeculateInt32Operand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| m_jit.compare32(MacroAssembler::Equal, value.gpr(), MacroAssembler::TrustedImm32(0), result.gpr()); |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| return; |
| } |
| |
| case NumberUse: { |
| SpeculateDoubleOperand value(this, node->child1()); |
| FPRTemporary scratch(this); |
| GPRTemporary result(this); |
| m_jit.move(TrustedImm32(ValueFalse), result.gpr()); |
| MacroAssembler::Jump nonZero = m_jit.branchDoubleNonZero(value.fpr(), scratch.fpr()); |
| m_jit.xor32(TrustedImm32(true), result.gpr()); |
| nonZero.link(&m_jit); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| return; |
| } |
| |
| case BooleanUse: { |
| if (!needsTypeCheck(node->child1(), SpecBoolean)) { |
| SpeculateBooleanOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| m_jit.move(value.gpr(), result.gpr()); |
| m_jit.xor64(TrustedImm32(true), result.gpr()); |
| |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| return; |
| } |
| |
| JSValueOperand value(this, node->child1(), ManualOperandSpeculation); |
| GPRTemporary result(this); // FIXME: We could reuse, but on speculation fail would need recovery to restore tag (akin to add). |
| |
| m_jit.move(value.gpr(), result.gpr()); |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), result.gpr()); |
| typeCheck( |
| JSValueRegs(value.gpr()), node->child1(), SpecBoolean, m_jit.branchTest64( |
| JITCompiler::NonZero, result.gpr(), TrustedImm32(static_cast<int32_t>(~1)))); |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueTrue)), result.gpr()); |
| |
| // If we add a DataFormatBool, we should use it here. |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| return; |
| } |
| |
| case UntypedUse: { |
| JSValueOperand arg1(this, node->child1()); |
| GPRTemporary result(this); |
| |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| arg1.use(); |
| |
| m_jit.move(arg1GPR, resultGPR); |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueFalse)), resultGPR); |
| JITCompiler::Jump slowCase = m_jit.branchTest64(JITCompiler::NonZero, resultGPR, TrustedImm32(static_cast<int32_t>(~1))); |
| |
| addSlowPathGenerator( |
| slowPathCall(slowCase, this, operationConvertJSValueToBoolean, resultGPR, arg1GPR)); |
| |
| m_jit.xor64(TrustedImm32(static_cast<int32_t>(ValueTrue)), resultGPR); |
| jsValueResult(resultGPR, node, DataFormatJSBoolean, UseChildrenCalledExplicitly); |
| return; |
| } |
| case StringUse: |
| return compileStringZeroLength(node); |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| } |
| |
| void SpeculativeJIT::emitObjectOrOtherBranch(Edge nodeUse, BasicBlock* taken, BasicBlock* notTaken) |
| { |
| JSValueOperand value(this, nodeUse, ManualOperandSpeculation); |
| GPRTemporary scratch(this); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| MacroAssembler::Jump notCell = m_jit.branchTest64(MacroAssembler::NonZero, valueGPR, GPRInfo::tagMaskRegister); |
| if (masqueradesAsUndefinedWatchpointIsStillValid()) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueGPR), nodeUse, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(valueGPR, JSCell::structureOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| } else { |
| m_jit.loadPtr(MacroAssembler::Address(valueGPR, JSCell::structureOffset()), scratchGPR); |
| |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueGPR), nodeUse, (~SpecCell) | SpecObject, m_jit.branchPtr( |
| MacroAssembler::Equal, |
| scratchGPR, |
| MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get()))); |
| |
| JITCompiler::Jump isNotMasqueradesAsUndefined = m_jit.branchTest8(JITCompiler::Zero, MacroAssembler::Address(scratchGPR, Structure::typeInfoFlagsOffset()), TrustedImm32(MasqueradesAsUndefined)); |
| |
| speculationCheck(BadType, JSValueRegs(valueGPR), nodeUse, |
| m_jit.branchPtr( |
| MacroAssembler::Equal, |
| MacroAssembler::Address(scratchGPR, Structure::globalObjectOffset()), |
| MacroAssembler::TrustedImmPtr(m_jit.graph().globalObjectFor(m_currentNode->codeOrigin)))); |
| |
| isNotMasqueradesAsUndefined.link(&m_jit); |
| } |
| jump(taken, ForceJump); |
| |
| notCell.link(&m_jit); |
| |
| if (needsTypeCheck(nodeUse, SpecCell | SpecOther)) { |
| m_jit.move(valueGPR, scratchGPR); |
| m_jit.and64(MacroAssembler::TrustedImm32(~TagBitUndefined), scratchGPR); |
| typeCheck( |
| JSValueRegs(valueGPR), nodeUse, SpecCell | SpecOther, m_jit.branch64( |
| MacroAssembler::NotEqual, scratchGPR, MacroAssembler::TrustedImm64(ValueNull))); |
| } |
| jump(notTaken); |
| |
| noResult(m_currentNode); |
| } |
| |
| void SpeculativeJIT::emitBranch(Node* node) |
| { |
| BasicBlock* taken = node->takenBlock(); |
| BasicBlock* notTaken = node->notTakenBlock(); |
| |
| switch (node->child1().useKind()) { |
| case ObjectOrOtherUse: { |
| emitObjectOrOtherBranch(node->child1(), taken, notTaken); |
| return; |
| } |
| |
| case Int32Use: |
| case NumberUse: { |
| if (node->child1().useKind() == Int32Use) { |
| bool invert = false; |
| |
| if (taken == nextBlock()) { |
| invert = true; |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| SpeculateInt32Operand value(this, node->child1()); |
| branchTest32(invert ? MacroAssembler::Zero : MacroAssembler::NonZero, value.gpr(), taken); |
| } else { |
| SpeculateDoubleOperand value(this, node->child1()); |
| FPRTemporary scratch(this); |
| branchDoubleNonZero(value.fpr(), scratch.fpr(), taken); |
| } |
| |
| jump(notTaken); |
| |
| noResult(node); |
| return; |
| } |
| |
| case UntypedUse: |
| case BooleanUse: { |
| JSValueOperand value(this, node->child1(), ManualOperandSpeculation); |
| GPRReg valueGPR = value.gpr(); |
| |
| if (node->child1().useKind() == BooleanUse) { |
| if (!needsTypeCheck(node->child1(), SpecBoolean)) { |
| MacroAssembler::ResultCondition condition = MacroAssembler::NonZero; |
| |
| if (taken == nextBlock()) { |
| condition = MacroAssembler::Zero; |
| BasicBlock* tmp = taken; |
| taken = notTaken; |
| notTaken = tmp; |
| } |
| |
| branchTest32(condition, valueGPR, TrustedImm32(true), taken); |
| jump(notTaken); |
| } else { |
| branch64(MacroAssembler::Equal, valueGPR, MacroAssembler::TrustedImm64(JSValue::encode(jsBoolean(false))), notTaken); |
| branch64(MacroAssembler::Equal, valueGPR, MacroAssembler::TrustedImm64(JSValue::encode(jsBoolean(true))), taken); |
| |
| typeCheck(JSValueRegs(valueGPR), node->child1(), SpecBoolean, m_jit.jump()); |
| } |
| value.use(); |
| } else { |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| if (node->child1()->prediction() & SpecInt32) { |
| branch64(MacroAssembler::Equal, valueGPR, MacroAssembler::TrustedImm64(JSValue::encode(jsNumber(0))), notTaken); |
| branch64(MacroAssembler::AboveOrEqual, valueGPR, GPRInfo::tagTypeNumberRegister, taken); |
| } |
| |
| if (node->child1()->prediction() & SpecBoolean) { |
| branch64(MacroAssembler::Equal, valueGPR, MacroAssembler::TrustedImm64(JSValue::encode(jsBoolean(false))), notTaken); |
| branch64(MacroAssembler::Equal, valueGPR, MacroAssembler::TrustedImm64(JSValue::encode(jsBoolean(true))), taken); |
| } |
| |
| value.use(); |
| |
| silentSpillAllRegisters(resultGPR); |
| callOperation(operationConvertJSValueToBoolean, resultGPR, valueGPR); |
| silentFillAllRegisters(resultGPR); |
| |
| branchTest32(MacroAssembler::NonZero, resultGPR, taken); |
| jump(notTaken); |
| } |
| |
| noResult(node, UseChildrenCalledExplicitly); |
| return; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| } |
| } |
| |
| void SpeculativeJIT::compile(Node* node) |
| { |
| NodeType op = node->op(); |
| |
| #if ENABLE(DFG_REGISTER_ALLOCATION_VALIDATION) |
| m_jit.clearRegisterAllocationOffsets(); |
| #endif |
| |
| switch (op) { |
| case JSConstant: |
| initConstantInfo(node); |
| break; |
| |
| case PhantomArguments: |
| initConstantInfo(node); |
| break; |
| |
| case WeakJSConstant: |
| m_jit.addWeakReference(node->weakConstant()); |
| initConstantInfo(node); |
| break; |
| |
| case Identity: { |
| // CSE should always eliminate this. |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| |
| case GetLocal: { |
| SpeculatedType prediction = node->variableAccessData()->prediction(); |
| AbstractValue& value = m_state.variables().operand(node->local()); |
| |
| // If we have no prediction for this local, then don't attempt to compile. |
| if (prediction == SpecNone) { |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| break; |
| } |
| |
| // If the CFA is tracking this variable and it found that the variable |
| // cannot have been assigned, then don't attempt to proceed. |
| if (value.isClear()) { |
| // FIXME: We should trap instead. |
| // https://bugs.webkit.org/show_bug.cgi?id=110383 |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| break; |
| } |
| |
| switch (node->variableAccessData()->flushFormat()) { |
| case FlushedDouble: { |
| FPRTemporary result(this); |
| m_jit.loadDouble(JITCompiler::addressFor(node->machineLocal()), result.fpr()); |
| VirtualRegister virtualRegister = node->virtualRegister(); |
| m_fprs.retain(result.fpr(), virtualRegister, SpillOrderDouble); |
| generationInfoFromVirtualRegister(virtualRegister).initDouble(node, node->refCount(), result.fpr()); |
| break; |
| } |
| |
| case FlushedInt32: { |
| GPRTemporary result(this); |
| m_jit.load32(JITCompiler::payloadFor(node->machineLocal()), result.gpr()); |
| |
| // Like int32Result, but don't useChildren - our children are phi nodes, |
| // and don't represent values within this dataflow with virtual registers. |
| VirtualRegister virtualRegister = node->virtualRegister(); |
| m_gprs.retain(result.gpr(), virtualRegister, SpillOrderInteger); |
| generationInfoFromVirtualRegister(virtualRegister).initInt32(node, node->refCount(), result.gpr()); |
| break; |
| } |
| |
| case FlushedInt52: { |
| GPRTemporary result(this); |
| m_jit.load64(JITCompiler::addressFor(node->machineLocal()), result.gpr()); |
| |
| VirtualRegister virtualRegister = node->virtualRegister(); |
| m_gprs.retain(result.gpr(), virtualRegister, SpillOrderJS); |
| generationInfoFromVirtualRegister(virtualRegister).initInt52(node, node->refCount(), result.gpr()); |
| break; |
| } |
| |
| default: |
| GPRTemporary result(this); |
| m_jit.load64(JITCompiler::addressFor(node->machineLocal()), result.gpr()); |
| |
| // Like jsValueResult, but don't useChildren - our children are phi nodes, |
| // and don't represent values within this dataflow with virtual registers. |
| VirtualRegister virtualRegister = node->virtualRegister(); |
| m_gprs.retain(result.gpr(), virtualRegister, SpillOrderJS); |
| |
| DataFormat format; |
| if (isCellSpeculation(value.m_type)) |
| format = DataFormatJSCell; |
| else if (isBooleanSpeculation(value.m_type)) |
| format = DataFormatJSBoolean; |
| else |
| format = DataFormatJS; |
| |
| generationInfoFromVirtualRegister(virtualRegister).initJSValue(node, node->refCount(), result.gpr(), format); |
| break; |
| } |
| break; |
| } |
| |
| case GetLocalUnlinked: { |
| GPRTemporary result(this); |
| |
| m_jit.load64(JITCompiler::addressFor(node->unlinkedMachineLocal()), result.gpr()); |
| |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| case MovHintAndCheck: { |
| compileMovHintAndCheck(node); |
| break; |
| } |
| |
| case MovHint: |
| case ZombieHint: { |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| |
| case SetLocal: { |
| // SetLocal doubles as a hint as to where a node will be stored and |
| // as a speculation point. So before we speculate make sure that we |
| // know where the child of this node needs to go in the virtual |
| // stack. |
| compileMovHint(node); |
| |
| switch (node->variableAccessData()->flushFormat()) { |
| case FlushedDouble: { |
| SpeculateDoubleOperand value(this, node->child1()); |
| m_jit.storeDouble(value.fpr(), JITCompiler::addressFor(node->machineLocal())); |
| noResult(node); |
| // Indicate that it's no longer necessary to retrieve the value of |
| // this bytecode variable from registers or other locations in the stack, |
| // but that it is stored as a double. |
| recordSetLocal(DataFormatDouble); |
| break; |
| } |
| |
| case FlushedInt32: { |
| SpeculateInt32Operand value(this, node->child1()); |
| m_jit.store32(value.gpr(), JITCompiler::payloadFor(node->machineLocal())); |
| noResult(node); |
| recordSetLocal(DataFormatInt32); |
| break; |
| } |
| |
| case FlushedInt52: { |
| SpeculateInt52Operand value(this, node->child1()); |
| m_jit.store64(value.gpr(), JITCompiler::addressFor(node->machineLocal())); |
| noResult(node); |
| recordSetLocal(DataFormatInt52); |
| break; |
| } |
| |
| case FlushedCell: { |
| SpeculateCellOperand cell(this, node->child1()); |
| GPRReg cellGPR = cell.gpr(); |
| m_jit.store64(cellGPR, JITCompiler::addressFor(node->machineLocal())); |
| noResult(node); |
| recordSetLocal(DataFormatCell); |
| break; |
| } |
| |
| case FlushedBoolean: { |
| SpeculateBooleanOperand boolean(this, node->child1()); |
| m_jit.store64(boolean.gpr(), JITCompiler::addressFor(node->machineLocal())); |
| noResult(node); |
| recordSetLocal(DataFormatBoolean); |
| break; |
| } |
| |
| case FlushedJSValue: { |
| JSValueOperand value(this, node->child1()); |
| m_jit.store64(value.gpr(), JITCompiler::addressFor(node->machineLocal())); |
| noResult(node); |
| |
| recordSetLocal(DataFormatJS); |
| |
| // If we're storing an arguments object that has been optimized away, |
| // our variable event stream for OSR exit now reflects the optimized |
| // value (JSValue()). On the slow path, we want an arguments object |
| // instead. We add an additional move hint to show OSR exit that it |
| // needs to reconstruct the arguments object. |
| if (node->child1()->op() == PhantomArguments) |
| compileMovHint(node); |
| break; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| |
| break; |
| } |
| |
| case SetArgument: |
| // This is a no-op; it just marks the fact that the argument is being used. |
| // But it may be profitable to use this as a hook to run speculation checks |
| // on arguments, thereby allowing us to trivially eliminate such checks if |
| // the argument is not used. |
| break; |
| |
| case BitAnd: |
| case BitOr: |
| case BitXor: |
| if (isInt32Constant(node->child1().node())) { |
| SpeculateInt32Operand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op2); |
| |
| bitOp(op, valueOfInt32Constant(node->child1().node()), op2.gpr(), result.gpr()); |
| |
| int32Result(result.gpr(), node); |
| } else if (isInt32Constant(node->child2().node())) { |
| SpeculateInt32Operand op1(this, node->child1()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| bitOp(op, valueOfInt32Constant(node->child2().node()), op1.gpr(), result.gpr()); |
| |
| int32Result(result.gpr(), node); |
| } else { |
| SpeculateInt32Operand op1(this, node->child1()); |
| SpeculateInt32Operand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op1, op2); |
| |
| GPRReg reg1 = op1.gpr(); |
| GPRReg reg2 = op2.gpr(); |
| bitOp(op, reg1, reg2, result.gpr()); |
| |
| int32Result(result.gpr(), node); |
| } |
| break; |
| |
| case BitRShift: |
| case BitLShift: |
| case BitURShift: |
| if (isInt32Constant(node->child2().node())) { |
| SpeculateInt32Operand op1(this, node->child1()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| shiftOp(op, op1.gpr(), valueOfInt32Constant(node->child2().node()) & 0x1f, result.gpr()); |
| |
| int32Result(result.gpr(), node); |
| } else { |
| // Do not allow shift amount to be used as the result, MacroAssembler does not permit this. |
| SpeculateInt32Operand op1(this, node->child1()); |
| SpeculateInt32Operand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| GPRReg reg1 = op1.gpr(); |
| GPRReg reg2 = op2.gpr(); |
| shiftOp(op, reg1, reg2, result.gpr()); |
| |
| int32Result(result.gpr(), node); |
| } |
| break; |
| |
| case UInt32ToNumber: { |
| compileUInt32ToNumber(node); |
| break; |
| } |
| |
| case DoubleAsInt32: { |
| compileDoubleAsInt32(node); |
| break; |
| } |
| |
| case ValueToInt32: { |
| compileValueToInt32(node); |
| break; |
| } |
| |
| case Int32ToDouble: { |
| compileInt32ToDouble(node); |
| break; |
| } |
| |
| case Int52ToValue: { |
| JSValueOperand operand(this, node->child1()); |
| GPRTemporary result(this, Reuse, operand); |
| m_jit.move(operand.gpr(), result.gpr()); |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| case Int52ToDouble: { |
| SpeculateDoubleOperand operand(this, node->child1()); |
| FPRTemporary result(this, operand); |
| m_jit.moveDouble(operand.fpr(), result.fpr()); |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| case ValueAdd: |
| case ArithAdd: |
| compileAdd(node); |
| break; |
| |
| case MakeRope: |
| compileMakeRope(node); |
| break; |
| |
| case ArithSub: |
| compileArithSub(node); |
| break; |
| |
| case ArithNegate: |
| compileArithNegate(node); |
| break; |
| |
| case ArithMul: |
| compileArithMul(node); |
| break; |
| |
| case ArithIMul: |
| compileArithIMul(node); |
| break; |
| |
| case ArithDiv: { |
| compileArithDiv(node); |
| break; |
| } |
| |
| case ArithMod: { |
| compileArithMod(node); |
| break; |
| } |
| |
| case ArithAbs: { |
| switch (node->child1().useKind()) { |
| case Int32Use: { |
| SpeculateStrictInt32Operand op1(this, node->child1()); |
| GPRTemporary result(this); |
| GPRTemporary scratch(this); |
| |
| m_jit.move(op1.gpr(), result.gpr()); |
| m_jit.rshift32(result.gpr(), MacroAssembler::TrustedImm32(31), scratch.gpr()); |
| m_jit.add32(scratch.gpr(), result.gpr()); |
| m_jit.xor32(scratch.gpr(), result.gpr()); |
| speculationCheck(Overflow, JSValueRegs(), 0, m_jit.branch32(MacroAssembler::Equal, result.gpr(), MacroAssembler::TrustedImm32(1 << 31))); |
| int32Result(result.gpr(), node); |
| break; |
| } |
| |
| case NumberUse: { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| FPRTemporary result(this); |
| |
| m_jit.absDouble(op1.fpr(), result.fpr()); |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| break; |
| } |
| |
| case ArithMin: |
| case ArithMax: { |
| switch (node->binaryUseKind()) { |
| case Int32Use: { |
| SpeculateStrictInt32Operand op1(this, node->child1()); |
| SpeculateStrictInt32Operand op2(this, node->child2()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| MacroAssembler::Jump op1Less = m_jit.branch32(op == ArithMin ? MacroAssembler::LessThan : MacroAssembler::GreaterThan, op1.gpr(), op2.gpr()); |
| m_jit.move(op2.gpr(), result.gpr()); |
| if (op1.gpr() != result.gpr()) { |
| MacroAssembler::Jump done = m_jit.jump(); |
| op1Less.link(&m_jit); |
| m_jit.move(op1.gpr(), result.gpr()); |
| done.link(&m_jit); |
| } else |
| op1Less.link(&m_jit); |
| |
| int32Result(result.gpr(), node); |
| break; |
| } |
| |
| case NumberUse: { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| SpeculateDoubleOperand op2(this, node->child2()); |
| FPRTemporary result(this, op1); |
| |
| FPRReg op1FPR = op1.fpr(); |
| FPRReg op2FPR = op2.fpr(); |
| FPRReg resultFPR = result.fpr(); |
| |
| MacroAssembler::JumpList done; |
| |
| MacroAssembler::Jump op1Less = m_jit.branchDouble(op == ArithMin ? MacroAssembler::DoubleLessThan : MacroAssembler::DoubleGreaterThan, op1FPR, op2FPR); |
| |
| // op2 is eather the lesser one or one of then is NaN |
| MacroAssembler::Jump op2Less = m_jit.branchDouble(op == ArithMin ? MacroAssembler::DoubleGreaterThanOrEqual : MacroAssembler::DoubleLessThanOrEqual, op1FPR, op2FPR); |
| |
| // Unordered case. We don't know which of op1, op2 is NaN. Manufacture NaN by adding |
| // op1 + op2 and putting it into result. |
| m_jit.addDouble(op1FPR, op2FPR, resultFPR); |
| done.append(m_jit.jump()); |
| |
| op2Less.link(&m_jit); |
| m_jit.moveDouble(op2FPR, resultFPR); |
| |
| if (op1FPR != resultFPR) { |
| done.append(m_jit.jump()); |
| |
| op1Less.link(&m_jit); |
| m_jit.moveDouble(op1FPR, resultFPR); |
| } else |
| op1Less.link(&m_jit); |
| |
| done.link(&m_jit); |
| |
| doubleResult(resultFPR, node); |
| break; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| break; |
| } |
| |
| case ArithSqrt: { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| FPRTemporary result(this, op1); |
| |
| m_jit.sqrtDouble(op1.fpr(), result.fpr()); |
| |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| case ArithSin: { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| FPRReg op1FPR = op1.fpr(); |
| |
| flushRegisters(); |
| |
| FPRResult result(this); |
| callOperation(sin, result.fpr(), op1FPR); |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| case ArithCos: { |
| SpeculateDoubleOperand op1(this, node->child1()); |
| FPRReg op1FPR = op1.fpr(); |
| |
| flushRegisters(); |
| |
| FPRResult result(this); |
| callOperation(cos, result.fpr(), op1FPR); |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| case LogicalNot: |
| compileLogicalNot(node); |
| break; |
| |
| case CompareLess: |
| if (compare(node, JITCompiler::LessThan, JITCompiler::DoubleLessThan, operationCompareLess)) |
| return; |
| break; |
| |
| case CompareLessEq: |
| if (compare(node, JITCompiler::LessThanOrEqual, JITCompiler::DoubleLessThanOrEqual, operationCompareLessEq)) |
| return; |
| break; |
| |
| case CompareGreater: |
| if (compare(node, JITCompiler::GreaterThan, JITCompiler::DoubleGreaterThan, operationCompareGreater)) |
| return; |
| break; |
| |
| case CompareGreaterEq: |
| if (compare(node, JITCompiler::GreaterThanOrEqual, JITCompiler::DoubleGreaterThanOrEqual, operationCompareGreaterEq)) |
| return; |
| break; |
| |
| case CompareEqConstant: |
| ASSERT(isNullConstant(node->child2().node())); |
| if (nonSpeculativeCompareNull(node, node->child1())) |
| return; |
| break; |
| |
| case CompareEq: |
| if (compare(node, JITCompiler::Equal, JITCompiler::DoubleEqual, operationCompareEq)) |
| return; |
| break; |
| |
| case CompareStrictEqConstant: |
| if (compileStrictEqForConstant(node, node->child1(), valueOfJSConstant(node->child2().node()))) |
| return; |
| break; |
| |
| case CompareStrictEq: |
| if (compileStrictEq(node)) |
| return; |
| break; |
| |
| case StringCharCodeAt: { |
| compileGetCharCodeAt(node); |
| break; |
| } |
| |
| case StringCharAt: { |
| // Relies on StringCharAt node having same basic layout as GetByVal |
| compileGetByValOnString(node); |
| break; |
| } |
| |
| case StringFromCharCode: { |
| compileFromCharCode(node); |
| break; |
| } |
| |
| case CheckArray: { |
| checkArray(node); |
| break; |
| } |
| |
| case Arrayify: |
| case ArrayifyToStructure: { |
| arrayify(node); |
| break; |
| } |
| |
| case GetByVal: { |
| switch (node->arrayMode().type()) { |
| case Array::SelectUsingPredictions: |
| case Array::ForceExit: |
| RELEASE_ASSERT_NOT_REACHED(); |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| break; |
| case Array::Generic: { |
| JSValueOperand base(this, node->child1()); |
| JSValueOperand property(this, node->child2()); |
| GPRReg baseGPR = base.gpr(); |
| GPRReg propertyGPR = property.gpr(); |
| |
| flushRegisters(); |
| GPRResult result(this); |
| callOperation(operationGetByVal, result.gpr(), baseGPR, propertyGPR); |
| |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| case Array::Int32: |
| case Array::Contiguous: { |
| if (node->arrayMode().isInBounds()) { |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| speculationCheck(OutOfBounds, JSValueRegs(), 0, m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength()))); |
| |
| GPRTemporary result(this); |
| m_jit.load64(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight), result.gpr()); |
| speculationCheck(LoadFromHole, JSValueRegs(), 0, m_jit.branchTest64(MacroAssembler::Zero, result.gpr())); |
| jsValueResult(result.gpr(), node, node->arrayMode().type() == Array::Int32 ? DataFormatJSInt32 : DataFormatJS); |
| break; |
| } |
| |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg baseReg = base.gpr(); |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| GPRTemporary result(this); |
| GPRReg resultReg = result.gpr(); |
| |
| MacroAssembler::JumpList slowCases; |
| |
| slowCases.append(m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength()))); |
| |
| m_jit.load64(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight), resultReg); |
| slowCases.append(m_jit.branchTest64(MacroAssembler::Zero, resultReg)); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowCases, this, operationGetByValArrayInt, |
| result.gpr(), baseReg, propertyReg)); |
| |
| jsValueResult(resultReg, node); |
| break; |
| } |
| |
| case Array::Double: { |
| if (node->arrayMode().isInBounds()) { |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| speculationCheck(OutOfBounds, JSValueRegs(), 0, m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength()))); |
| |
| FPRTemporary result(this); |
| m_jit.loadDouble(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight), result.fpr()); |
| if (!node->arrayMode().isSaneChain()) |
| speculationCheck(LoadFromHole, JSValueRegs(), 0, m_jit.branchDouble(MacroAssembler::DoubleNotEqualOrUnordered, result.fpr(), result.fpr())); |
| doubleResult(result.fpr(), node); |
| break; |
| } |
| |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg baseReg = base.gpr(); |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| GPRTemporary result(this); |
| FPRTemporary temp(this); |
| GPRReg resultReg = result.gpr(); |
| FPRReg tempReg = temp.fpr(); |
| |
| MacroAssembler::JumpList slowCases; |
| |
| slowCases.append(m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength()))); |
| |
| m_jit.loadDouble(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight), tempReg); |
| slowCases.append(m_jit.branchDouble(MacroAssembler::DoubleNotEqualOrUnordered, tempReg, tempReg)); |
| boxDouble(tempReg, resultReg); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowCases, this, operationGetByValArrayInt, |
| result.gpr(), baseReg, propertyReg)); |
| |
| jsValueResult(resultReg, node); |
| break; |
| } |
| |
| case Array::ArrayStorage: |
| case Array::SlowPutArrayStorage: { |
| if (node->arrayMode().isInBounds()) { |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| speculationCheck(OutOfBounds, JSValueRegs(), 0, m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, ArrayStorage::vectorLengthOffset()))); |
| |
| GPRTemporary result(this); |
| m_jit.load64(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0])), result.gpr()); |
| speculationCheck(LoadFromHole, JSValueRegs(), 0, m_jit.branchTest64(MacroAssembler::Zero, result.gpr())); |
| |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateStrictInt32Operand property(this, node->child2()); |
| StorageOperand storage(this, node->child3()); |
| |
| GPRReg baseReg = base.gpr(); |
| GPRReg propertyReg = property.gpr(); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| GPRTemporary result(this); |
| GPRReg resultReg = result.gpr(); |
| |
| MacroAssembler::JumpList slowCases; |
| |
| slowCases.append(m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, ArrayStorage::vectorLengthOffset()))); |
| |
| m_jit.load64(MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0])), resultReg); |
| slowCases.append(m_jit.branchTest64(MacroAssembler::Zero, resultReg)); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowCases, this, operationGetByValArrayInt, |
| result.gpr(), baseReg, propertyReg)); |
| |
| jsValueResult(resultReg, node); |
| break; |
| } |
| case Array::String: |
| compileGetByValOnString(node); |
| break; |
| case Array::Arguments: |
| compileGetByValOnArguments(node); |
| break; |
| default: { |
| TypedArrayType type = node->arrayMode().typedArrayType(); |
| if (isInt(type)) |
| compileGetByValOnIntTypedArray(node, type); |
| else |
| compileGetByValOnFloatTypedArray(node, type); |
| } } |
| break; |
| } |
| |
| case PutByValDirect: |
| case PutByVal: |
| case PutByValAlias: { |
| Edge child1 = m_jit.graph().varArgChild(node, 0); |
| Edge child2 = m_jit.graph().varArgChild(node, 1); |
| Edge child3 = m_jit.graph().varArgChild(node, 2); |
| Edge child4 = m_jit.graph().varArgChild(node, 3); |
| |
| ArrayMode arrayMode = node->arrayMode().modeForPut(); |
| bool alreadyHandled = false; |
| |
| switch (arrayMode.type()) { |
| case Array::SelectUsingPredictions: |
| case Array::ForceExit: |
| RELEASE_ASSERT_NOT_REACHED(); |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| alreadyHandled = true; |
| break; |
| case Array::Generic: { |
| RELEASE_ASSERT(node->op() == PutByVal); |
| |
| JSValueOperand arg1(this, child1); |
| JSValueOperand arg2(this, child2); |
| JSValueOperand arg3(this, child3); |
| GPRReg arg1GPR = arg1.gpr(); |
| GPRReg arg2GPR = arg2.gpr(); |
| GPRReg arg3GPR = arg3.gpr(); |
| flushRegisters(); |
| if (node->op() == PutByValDirect) |
| callOperation(m_jit.isStrictModeFor(node->codeOrigin) ? operationPutByValDirectStrict : operationPutByValDirectNonStrict, arg1GPR, arg2GPR, arg3GPR); |
| else |
| callOperation(m_jit.isStrictModeFor(node->codeOrigin) ? operationPutByValStrict : operationPutByValNonStrict, arg1GPR, arg2GPR, arg3GPR); |
| |
| noResult(node); |
| alreadyHandled = true; |
| break; |
| } |
| default: |
| break; |
| } |
| |
| if (alreadyHandled) |
| break; |
| |
| // FIXME: the base may not be necessary for some array access modes. But we have to |
| // keep it alive to this point, so it's likely to be in a register anyway. Likely |
| // no harm in locking it here. |
| SpeculateCellOperand base(this, child1); |
| SpeculateStrictInt32Operand property(this, child2); |
| |
| GPRReg baseReg = base.gpr(); |
| GPRReg propertyReg = property.gpr(); |
| |
| switch (arrayMode.type()) { |
| case Array::Int32: |
| case Array::Contiguous: { |
| JSValueOperand value(this, child3, ManualOperandSpeculation); |
| |
| GPRReg valueReg = value.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| if (arrayMode.type() == Array::Int32) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueReg), child3, SpecInt32, |
| m_jit.branch64( |
| MacroAssembler::Below, valueReg, GPRInfo::tagTypeNumberRegister)); |
| } |
| |
| if (arrayMode.type() == Array::Contiguous && Heap::isWriteBarrierEnabled()) { |
| GPRTemporary scratch(this); |
| writeBarrier(baseReg, value.gpr(), child3, WriteBarrierForPropertyAccess, scratch.gpr()); |
| } |
| |
| StorageOperand storage(this, child4); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (node->op() == PutByValAlias) { |
| // Store the value to the array. |
| GPRReg propertyReg = property.gpr(); |
| GPRReg valueReg = value.gpr(); |
| m_jit.store64(valueReg, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight)); |
| |
| noResult(node); |
| break; |
| } |
| |
| GPRTemporary temporary; |
| GPRReg temporaryReg = temporaryRegisterForPutByVal(temporary, node); |
| |
| MacroAssembler::Jump slowCase; |
| |
| if (arrayMode.isInBounds()) { |
| speculationCheck( |
| StoreToHoleOrOutOfBounds, JSValueRegs(), 0, |
| m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength()))); |
| } else { |
| MacroAssembler::Jump inBounds = m_jit.branch32(MacroAssembler::Below, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength())); |
| |
| slowCase = m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfVectorLength())); |
| |
| if (!arrayMode.isOutOfBounds()) |
| speculationCheck(OutOfBounds, JSValueRegs(), 0, slowCase); |
| |
| m_jit.add32(TrustedImm32(1), propertyReg, temporaryReg); |
| m_jit.store32(temporaryReg, MacroAssembler::Address(storageReg, Butterfly::offsetOfPublicLength())); |
| |
| inBounds.link(&m_jit); |
| } |
| |
| m_jit.store64(valueReg, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight)); |
| |
| base.use(); |
| property.use(); |
| value.use(); |
| storage.use(); |
| |
| if (arrayMode.isOutOfBounds()) { |
| if (node->op() == PutByValDirect) { |
| addSlowPathGenerator(slowPathCall( |
| slowCase, this, |
| m_jit.codeBlock()->isStrictMode() ? operationPutByValDirectBeyondArrayBoundsStrict : operationPutByValDirectBeyondArrayBoundsNonStrict, |
| NoResult, baseReg, propertyReg, valueReg)); |
| } else { |
| addSlowPathGenerator(slowPathCall( |
| slowCase, this, |
| m_jit.codeBlock()->isStrictMode() ? operationPutByValBeyondArrayBoundsStrict : operationPutByValBeyondArrayBoundsNonStrict, |
| NoResult, baseReg, propertyReg, valueReg)); |
| } |
| } |
| |
| noResult(node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case Array::Double: { |
| compileDoublePutByVal(node, base, property); |
| break; |
| } |
| |
| case Array::ArrayStorage: |
| case Array::SlowPutArrayStorage: { |
| JSValueOperand value(this, child3); |
| |
| GPRReg valueReg = value.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| if (Heap::isWriteBarrierEnabled()) { |
| GPRTemporary scratch(this); |
| writeBarrier(baseReg, value.gpr(), child3, WriteBarrierForPropertyAccess, scratch.gpr()); |
| } |
| |
| StorageOperand storage(this, child4); |
| GPRReg storageReg = storage.gpr(); |
| |
| if (node->op() == PutByValAlias) { |
| // Store the value to the array. |
| GPRReg propertyReg = property.gpr(); |
| GPRReg valueReg = value.gpr(); |
| m_jit.store64(valueReg, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); |
| |
| noResult(node); |
| break; |
| } |
| |
| GPRTemporary temporary; |
| GPRReg temporaryReg = temporaryRegisterForPutByVal(temporary, node); |
| |
| MacroAssembler::JumpList slowCases; |
| |
| MacroAssembler::Jump beyondArrayBounds = m_jit.branch32(MacroAssembler::AboveOrEqual, propertyReg, MacroAssembler::Address(storageReg, ArrayStorage::vectorLengthOffset())); |
| if (!arrayMode.isOutOfBounds()) |
| speculationCheck(OutOfBounds, JSValueRegs(), 0, beyondArrayBounds); |
| else |
| slowCases.append(beyondArrayBounds); |
| |
| // Check if we're writing to a hole; if so increment m_numValuesInVector. |
| if (arrayMode.isInBounds()) { |
| speculationCheck( |
| StoreToHole, JSValueRegs(), 0, |
| m_jit.branchTest64(MacroAssembler::Zero, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0])))); |
| } else { |
| MacroAssembler::Jump notHoleValue = m_jit.branchTest64(MacroAssembler::NonZero, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); |
| if (arrayMode.isSlowPut()) { |
| // This is sort of strange. If we wanted to optimize this code path, we would invert |
| // the above branch. But it's simply not worth it since this only happens if we're |
| // already having a bad time. |
| slowCases.append(m_jit.jump()); |
| } else { |
| m_jit.add32(TrustedImm32(1), MacroAssembler::Address(storageReg, ArrayStorage::numValuesInVectorOffset())); |
| |
| // If we're writing to a hole we might be growing the array; |
| MacroAssembler::Jump lengthDoesNotNeedUpdate = m_jit.branch32(MacroAssembler::Below, propertyReg, MacroAssembler::Address(storageReg, ArrayStorage::lengthOffset())); |
| m_jit.add32(TrustedImm32(1), propertyReg, temporaryReg); |
| m_jit.store32(temporaryReg, MacroAssembler::Address(storageReg, ArrayStorage::lengthOffset())); |
| |
| lengthDoesNotNeedUpdate.link(&m_jit); |
| } |
| notHoleValue.link(&m_jit); |
| } |
| |
| // Store the value to the array. |
| m_jit.store64(valueReg, MacroAssembler::BaseIndex(storageReg, propertyReg, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); |
| |
| base.use(); |
| property.use(); |
| value.use(); |
| storage.use(); |
| |
| if (!slowCases.empty()) { |
| if (node->op() == PutByValDirect) { |
| addSlowPathGenerator(slowPathCall( |
| slowCases, this, |
| m_jit.codeBlock()->isStrictMode() ? operationPutByValDirectBeyondArrayBoundsStrict : operationPutByValDirectBeyondArrayBoundsNonStrict, |
| NoResult, baseReg, propertyReg, valueReg)); |
| } else { |
| addSlowPathGenerator(slowPathCall( |
| slowCases, this, |
| m_jit.codeBlock()->isStrictMode() ? operationPutByValBeyondArrayBoundsStrict : operationPutByValBeyondArrayBoundsNonStrict, |
| NoResult, baseReg, propertyReg, valueReg)); |
| } |
| } |
| |
| noResult(node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case Array::Arguments: { |
| JSValueOperand value(this, child3); |
| GPRTemporary scratch(this); |
| GPRTemporary scratch2(this); |
| |
| GPRReg valueReg = value.gpr(); |
| GPRReg scratchReg = scratch.gpr(); |
| GPRReg scratch2Reg = scratch2.gpr(); |
| |
| if (!m_compileOkay) |
| return; |
| |
| // Two really lame checks. |
| speculationCheck( |
| Uncountable, JSValueSource(), 0, |
| m_jit.branch32( |
| MacroAssembler::AboveOrEqual, propertyReg, |
| MacroAssembler::Address(baseReg, Arguments::offsetOfNumArguments()))); |
| speculationCheck( |
| Uncountable, JSValueSource(), 0, |
| m_jit.branchTestPtr( |
| MacroAssembler::NonZero, |
| MacroAssembler::Address( |
| baseReg, Arguments::offsetOfSlowArgumentData()))); |
| |
| m_jit.move(propertyReg, scratch2Reg); |
| m_jit.signExtend32ToPtr(scratch2Reg, scratch2Reg); |
| m_jit.loadPtr( |
| MacroAssembler::Address(baseReg, Arguments::offsetOfRegisters()), |
| scratchReg); |
| |
| m_jit.store64( |
| valueReg, |
| MacroAssembler::BaseIndex( |
| scratchReg, scratch2Reg, MacroAssembler::TimesEight, |
| CallFrame::thisArgumentOffset() * sizeof(Register) + sizeof(Register))); |
| |
| noResult(node); |
| break; |
| } |
| |
| default: { |
| TypedArrayType type = arrayMode.typedArrayType(); |
| if (isInt(type)) |
| compilePutByValForIntTypedArray(base.gpr(), property.gpr(), node, type); |
| else |
| compilePutByValForFloatTypedArray(base.gpr(), property.gpr(), node, type); |
| } } |
| |
| break; |
| } |
| |
| case RegExpExec: { |
| if (compileRegExpExec(node)) |
| return; |
| if (!node->adjustedRefCount()) { |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateCellOperand argument(this, node->child2()); |
| GPRReg baseGPR = base.gpr(); |
| GPRReg argumentGPR = argument.gpr(); |
| |
| flushRegisters(); |
| GPRResult result(this); |
| callOperation(operationRegExpTest, result.gpr(), baseGPR, argumentGPR); |
| |
| // Must use jsValueResult because otherwise we screw up register |
| // allocation, which thinks that this node has a result. |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateCellOperand argument(this, node->child2()); |
| GPRReg baseGPR = base.gpr(); |
| GPRReg argumentGPR = argument.gpr(); |
| |
| flushRegisters(); |
| GPRResult result(this); |
| callOperation(operationRegExpExec, result.gpr(), baseGPR, argumentGPR); |
| |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| case RegExpTest: { |
| SpeculateCellOperand base(this, node->child1()); |
| SpeculateCellOperand argument(this, node->child2()); |
| GPRReg baseGPR = base.gpr(); |
| GPRReg argumentGPR = argument.gpr(); |
| |
| flushRegisters(); |
| GPRResult result(this); |
| callOperation(operationRegExpTest, result.gpr(), baseGPR, argumentGPR); |
| |
| // If we add a DataFormatBool, we should use it here. |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case ArrayPush: { |
| ASSERT(node->arrayMode().isJSArray()); |
| |
| SpeculateCellOperand base(this, node->child1()); |
| GPRTemporary storageLength(this); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg storageLengthGPR = storageLength.gpr(); |
| |
| StorageOperand storage(this, node->child3()); |
| GPRReg storageGPR = storage.gpr(); |
| |
| switch (node->arrayMode().type()) { |
| case Array::Int32: |
| case Array::Contiguous: { |
| JSValueOperand value(this, node->child2(), ManualOperandSpeculation); |
| GPRReg valueGPR = value.gpr(); |
| |
| if (node->arrayMode().type() == Array::Int32) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(valueGPR), node->child2(), SpecInt32, |
| m_jit.branch64( |
| MacroAssembler::Below, valueGPR, GPRInfo::tagTypeNumberRegister)); |
| } |
| |
| if (node->arrayMode().type() != Array::Int32 && Heap::isWriteBarrierEnabled()) { |
| GPRTemporary scratch(this); |
| writeBarrier(baseGPR, valueGPR, node->child2(), WriteBarrierForPropertyAccess, scratch.gpr(), storageLengthGPR); |
| } |
| |
| m_jit.load32(MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength()), storageLengthGPR); |
| MacroAssembler::Jump slowPath = m_jit.branch32(MacroAssembler::AboveOrEqual, storageLengthGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfVectorLength())); |
| m_jit.store64(valueGPR, MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight)); |
| m_jit.add32(TrustedImm32(1), storageLengthGPR); |
| m_jit.store32(storageLengthGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength())); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, storageLengthGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowPath, this, operationArrayPush, NoResult, storageLengthGPR, |
| valueGPR, baseGPR)); |
| |
| jsValueResult(storageLengthGPR, node); |
| break; |
| } |
| |
| case Array::Double: { |
| SpeculateDoubleOperand value(this, node->child2()); |
| FPRReg valueFPR = value.fpr(); |
| |
| DFG_TYPE_CHECK( |
| JSValueRegs(), node->child2(), SpecFullRealNumber, |
| m_jit.branchDouble(MacroAssembler::DoubleNotEqualOrUnordered, valueFPR, valueFPR)); |
| |
| m_jit.load32(MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength()), storageLengthGPR); |
| MacroAssembler::Jump slowPath = m_jit.branch32(MacroAssembler::AboveOrEqual, storageLengthGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfVectorLength())); |
| m_jit.storeDouble(valueFPR, MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight)); |
| m_jit.add32(TrustedImm32(1), storageLengthGPR); |
| m_jit.store32(storageLengthGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength())); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, storageLengthGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowPath, this, operationArrayPushDouble, NoResult, storageLengthGPR, |
| valueFPR, baseGPR)); |
| |
| jsValueResult(storageLengthGPR, node); |
| break; |
| } |
| |
| case Array::ArrayStorage: { |
| JSValueOperand value(this, node->child2()); |
| GPRReg valueGPR = value.gpr(); |
| |
| if (Heap::isWriteBarrierEnabled()) { |
| GPRTemporary scratch(this); |
| writeBarrier(baseGPR, valueGPR, node->child2(), WriteBarrierForPropertyAccess, scratch.gpr(), storageLengthGPR); |
| } |
| |
| m_jit.load32(MacroAssembler::Address(storageGPR, ArrayStorage::lengthOffset()), storageLengthGPR); |
| |
| // Refuse to handle bizarre lengths. |
| speculationCheck(Uncountable, JSValueRegs(), 0, m_jit.branch32(MacroAssembler::Above, storageLengthGPR, TrustedImm32(0x7ffffffe))); |
| |
| MacroAssembler::Jump slowPath = m_jit.branch32(MacroAssembler::AboveOrEqual, storageLengthGPR, MacroAssembler::Address(storageGPR, ArrayStorage::vectorLengthOffset())); |
| |
| m_jit.store64(valueGPR, MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); |
| |
| m_jit.add32(TrustedImm32(1), storageLengthGPR); |
| m_jit.store32(storageLengthGPR, MacroAssembler::Address(storageGPR, ArrayStorage::lengthOffset())); |
| m_jit.add32(TrustedImm32(1), MacroAssembler::Address(storageGPR, OBJECT_OFFSETOF(ArrayStorage, m_numValuesInVector))); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, storageLengthGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowPath, this, operationArrayPush, NoResult, storageLengthGPR, |
| valueGPR, baseGPR)); |
| |
| jsValueResult(storageLengthGPR, node); |
| break; |
| } |
| |
| default: |
| CRASH(); |
| break; |
| } |
| break; |
| } |
| |
| case ArrayPop: { |
| ASSERT(node->arrayMode().isJSArray()); |
| |
| SpeculateCellOperand base(this, node->child1()); |
| StorageOperand storage(this, node->child2()); |
| GPRTemporary value(this); |
| GPRTemporary storageLength(this); |
| FPRTemporary temp(this); // This is kind of lame, since we don't always need it. I'm relying on the fact that we don't have FPR pressure, especially in code that uses pop(). |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg storageGPR = storage.gpr(); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg storageLengthGPR = storageLength.gpr(); |
| FPRReg tempFPR = temp.fpr(); |
| |
| switch (node->arrayMode().type()) { |
| case Array::Int32: |
| case Array::Double: |
| case Array::Contiguous: { |
| m_jit.load32( |
| MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength()), storageLengthGPR); |
| MacroAssembler::Jump undefinedCase = |
| m_jit.branchTest32(MacroAssembler::Zero, storageLengthGPR); |
| m_jit.sub32(TrustedImm32(1), storageLengthGPR); |
| m_jit.store32( |
| storageLengthGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength())); |
| MacroAssembler::Jump slowCase; |
| if (node->arrayMode().type() == Array::Double) { |
| m_jit.loadDouble( |
| MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight), |
| tempFPR); |
| // FIXME: This would not have to be here if changing the publicLength also zeroed the values between the old |
| // length and the new length. |
| m_jit.store64( |
| MacroAssembler::TrustedImm64((int64_t)0), MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight)); |
| slowCase = m_jit.branchDouble(MacroAssembler::DoubleNotEqualOrUnordered, tempFPR, tempFPR); |
| boxDouble(tempFPR, valueGPR); |
| } else { |
| m_jit.load64( |
| MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight), |
| valueGPR); |
| // FIXME: This would not have to be here if changing the publicLength also zeroed the values between the old |
| // length and the new length. |
| m_jit.store64( |
| MacroAssembler::TrustedImm64((int64_t)0), MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight)); |
| slowCase = m_jit.branchTest64(MacroAssembler::Zero, valueGPR); |
| } |
| |
| addSlowPathGenerator( |
| slowPathMove( |
| undefinedCase, this, |
| MacroAssembler::TrustedImm64(JSValue::encode(jsUndefined())), valueGPR)); |
| addSlowPathGenerator( |
| slowPathCall( |
| slowCase, this, operationArrayPopAndRecoverLength, valueGPR, baseGPR)); |
| |
| // We can't know for sure that the result is an int because of the slow paths. :-/ |
| jsValueResult(valueGPR, node); |
| break; |
| } |
| |
| case Array::ArrayStorage: { |
| m_jit.load32(MacroAssembler::Address(storageGPR, ArrayStorage::lengthOffset()), storageLengthGPR); |
| |
| JITCompiler::Jump undefinedCase = |
| m_jit.branchTest32(MacroAssembler::Zero, storageLengthGPR); |
| |
| m_jit.sub32(TrustedImm32(1), storageLengthGPR); |
| |
| JITCompiler::JumpList slowCases; |
| slowCases.append(m_jit.branch32(MacroAssembler::AboveOrEqual, storageLengthGPR, MacroAssembler::Address(storageGPR, ArrayStorage::vectorLengthOffset()))); |
| |
| m_jit.load64(MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0])), valueGPR); |
| slowCases.append(m_jit.branchTest64(MacroAssembler::Zero, valueGPR)); |
| |
| m_jit.store32(storageLengthGPR, MacroAssembler::Address(storageGPR, ArrayStorage::lengthOffset())); |
| |
| m_jit.store64(MacroAssembler::TrustedImm64((int64_t)0), MacroAssembler::BaseIndex(storageGPR, storageLengthGPR, MacroAssembler::TimesEight, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); |
| m_jit.sub32(MacroAssembler::TrustedImm32(1), MacroAssembler::Address(storageGPR, OBJECT_OFFSETOF(ArrayStorage, m_numValuesInVector))); |
| |
| addSlowPathGenerator( |
| slowPathMove( |
| undefinedCase, this, |
| MacroAssembler::TrustedImm64(JSValue::encode(jsUndefined())), valueGPR)); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| slowCases, this, operationArrayPop, valueGPR, baseGPR)); |
| |
| jsValueResult(valueGPR, node); |
| break; |
| } |
| |
| default: |
| CRASH(); |
| break; |
| } |
| break; |
| } |
| |
| case DFG::Jump: { |
| jump(node->takenBlock()); |
| noResult(node); |
| break; |
| } |
| |
| case Branch: |
| emitBranch(node); |
| break; |
| |
| case Switch: |
| emitSwitch(node); |
| break; |
| |
| case Return: { |
| ASSERT(GPRInfo::callFrameRegister != GPRInfo::regT1); |
| ASSERT(GPRInfo::regT1 != GPRInfo::returnValueGPR); |
| ASSERT(GPRInfo::returnValueGPR != GPRInfo::callFrameRegister); |
| |
| #if DFG_ENABLE(SUCCESS_STATS) |
| static SamplingCounter counter("SpeculativeJIT"); |
| m_jit.emitCount(counter); |
| #endif |
| |
| // Return the result in returnValueGPR. |
| JSValueOperand op1(this, node->child1()); |
| m_jit.move(op1.gpr(), GPRInfo::returnValueGPR); |
| |
| // Grab the return address. |
| m_jit.emitGetReturnPCFromCallFrameHeaderPtr(GPRInfo::regT1); |
| // Restore our caller's "r". |
| m_jit.emitGetCallerFrameFromCallFrameHeaderPtr(GPRInfo::callFrameRegister); |
| // Return. |
| m_jit.restoreReturnAddressBeforeReturn(GPRInfo::regT1); |
| m_jit.ret(); |
| |
| noResult(node); |
| break; |
| } |
| |
| case Throw: |
| case ThrowReferenceError: { |
| // We expect that throw statements are rare and are intended to exit the code block |
| // anyway, so we just OSR back to the old JIT for now. |
| terminateSpeculativeExecution(Uncountable, JSValueRegs(), 0); |
| break; |
| } |
| |
| case ToPrimitive: { |
| RELEASE_ASSERT(node->child1().useKind() == UntypedUse); |
| JSValueOperand op1(this, node->child1()); |
| GPRTemporary result(this, Reuse, op1); |
| |
| GPRReg op1GPR = op1.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| op1.use(); |
| |
| if (!(m_state.forNode(node->child1()).m_type & ~(SpecFullNumber | SpecBoolean))) |
| m_jit.move(op1GPR, resultGPR); |
| else { |
| MacroAssembler::Jump alreadyPrimitive = m_jit.branchTest64(MacroAssembler::NonZero, op1GPR, GPRInfo::tagMaskRegister); |
| MacroAssembler::Jump notPrimitive = m_jit.branchPtr(MacroAssembler::NotEqual, MacroAssembler::Address(op1GPR, JSCell::structureOffset()), MacroAssembler::TrustedImmPtr(m_jit.vm()->stringStructure.get())); |
| |
| alreadyPrimitive.link(&m_jit); |
| m_jit.move(op1GPR, resultGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall(notPrimitive, this, operationToPrimitive, resultGPR, op1GPR)); |
| } |
| |
| jsValueResult(resultGPR, node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case ToString: { |
| if (node->child1().useKind() == UntypedUse) { |
| JSValueOperand op1(this, node->child1()); |
| GPRReg op1GPR = op1.gpr(); |
| |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| flushRegisters(); |
| |
| JITCompiler::Jump done; |
| if (node->child1()->prediction() & SpecString) { |
| JITCompiler::Jump slowPath1 = m_jit.branchTest64( |
| JITCompiler::NonZero, op1GPR, GPRInfo::tagMaskRegister); |
| JITCompiler::Jump slowPath2 = m_jit.branchPtr( |
| JITCompiler::NotEqual, |
| JITCompiler::Address(op1GPR, JSCell::structureOffset()), |
| TrustedImmPtr(m_jit.vm()->stringStructure.get())); |
| m_jit.move(op1GPR, resultGPR); |
| done = m_jit.jump(); |
| slowPath1.link(&m_jit); |
| slowPath2.link(&m_jit); |
| } |
| callOperation(operationToString, resultGPR, op1GPR); |
| if (done.isSet()) |
| done.link(&m_jit); |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| compileToStringOnCell(node); |
| break; |
| } |
| |
| case NewStringObject: { |
| compileNewStringObject(node); |
| break; |
| } |
| |
| case NewArray: { |
| JSGlobalObject* globalObject = m_jit.graph().globalObjectFor(node->codeOrigin); |
| if (!globalObject->isHavingABadTime() && !hasArrayStorage(node->indexingType())) { |
| Structure* structure = globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType()); |
| RELEASE_ASSERT(structure->indexingType() == node->indexingType()); |
| ASSERT( |
| hasUndecided(structure->indexingType()) |
| || hasInt32(structure->indexingType()) |
| || hasDouble(structure->indexingType()) |
| || hasContiguous(structure->indexingType())); |
| |
| unsigned numElements = node->numChildren(); |
| |
| GPRTemporary result(this); |
| GPRTemporary storage(this); |
| |
| GPRReg resultGPR = result.gpr(); |
| GPRReg storageGPR = storage.gpr(); |
| |
| emitAllocateJSArray(resultGPR, structure, storageGPR, numElements); |
| |
| // At this point, one way or another, resultGPR and storageGPR have pointers to |
| // the JSArray and the Butterfly, respectively. |
| |
| ASSERT(!hasUndecided(structure->indexingType()) || !node->numChildren()); |
| |
| for (unsigned operandIdx = 0; operandIdx < node->numChildren(); ++operandIdx) { |
| Edge use = m_jit.graph().m_varArgChildren[node->firstChild() + operandIdx]; |
| switch (node->indexingType()) { |
| case ALL_BLANK_INDEXING_TYPES: |
| case ALL_UNDECIDED_INDEXING_TYPES: |
| CRASH(); |
| break; |
| case ALL_DOUBLE_INDEXING_TYPES: { |
| SpeculateDoubleOperand operand(this, use); |
| FPRReg opFPR = operand.fpr(); |
| DFG_TYPE_CHECK( |
| JSValueRegs(), use, SpecFullRealNumber, |
| m_jit.branchDouble( |
| MacroAssembler::DoubleNotEqualOrUnordered, opFPR, opFPR)); |
| m_jit.storeDouble(opFPR, MacroAssembler::Address(storageGPR, sizeof(double) * operandIdx)); |
| break; |
| } |
| case ALL_INT32_INDEXING_TYPES: |
| case ALL_CONTIGUOUS_INDEXING_TYPES: { |
| JSValueOperand operand(this, use, ManualOperandSpeculation); |
| GPRReg opGPR = operand.gpr(); |
| if (hasInt32(node->indexingType())) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(opGPR), use, SpecInt32, |
| m_jit.branch64( |
| MacroAssembler::Below, opGPR, GPRInfo::tagTypeNumberRegister)); |
| } |
| m_jit.store64(opGPR, MacroAssembler::Address(storageGPR, sizeof(JSValue) * operandIdx)); |
| break; |
| } |
| default: |
| CRASH(); |
| break; |
| } |
| } |
| |
| // Yuck, we should *really* have a way of also returning the storageGPR. But |
| // that's the least of what's wrong with this code. We really shouldn't be |
| // allocating the array after having computed - and probably spilled to the |
| // stack - all of the things that will go into the array. The solution to that |
| // bigger problem will also likely fix the redundancy in reloading the storage |
| // pointer that we currently have. |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| if (!node->numChildren()) { |
| flushRegisters(); |
| GPRResult result(this); |
| callOperation(operationNewEmptyArray, result.gpr(), globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType())); |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| size_t scratchSize = sizeof(EncodedJSValue) * node->numChildren(); |
| ScratchBuffer* scratchBuffer = m_jit.vm()->scratchBufferForSize(scratchSize); |
| EncodedJSValue* buffer = scratchBuffer ? static_cast<EncodedJSValue*>(scratchBuffer->dataBuffer()) : 0; |
| |
| for (unsigned operandIdx = 0; operandIdx < node->numChildren(); ++operandIdx) { |
| // Need to perform the speculations that this node promises to perform. If we're |
| // emitting code here and the indexing type is not array storage then there is |
| // probably something hilarious going on and we're already failing at all the |
| // things, but at least we're going to be sound. |
| Edge use = m_jit.graph().m_varArgChildren[node->firstChild() + operandIdx]; |
| switch (node->indexingType()) { |
| case ALL_BLANK_INDEXING_TYPES: |
| case ALL_UNDECIDED_INDEXING_TYPES: |
| CRASH(); |
| break; |
| case ALL_DOUBLE_INDEXING_TYPES: { |
| SpeculateDoubleOperand operand(this, use); |
| GPRTemporary scratch(this); |
| FPRReg opFPR = operand.fpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| DFG_TYPE_CHECK( |
| JSValueRegs(), use, SpecFullRealNumber, |
| m_jit.branchDouble( |
| MacroAssembler::DoubleNotEqualOrUnordered, opFPR, opFPR)); |
| m_jit.boxDouble(opFPR, scratchGPR); |
| m_jit.store64(scratchGPR, buffer + operandIdx); |
| break; |
| } |
| case ALL_INT32_INDEXING_TYPES: { |
| JSValueOperand operand(this, use, ManualOperandSpeculation); |
| GPRReg opGPR = operand.gpr(); |
| if (hasInt32(node->indexingType())) { |
| DFG_TYPE_CHECK( |
| JSValueRegs(opGPR), use, SpecInt32, |
| m_jit.branch64( |
| MacroAssembler::Below, opGPR, GPRInfo::tagTypeNumberRegister)); |
| } |
| m_jit.store64(opGPR, buffer + operandIdx); |
| break; |
| } |
| case ALL_CONTIGUOUS_INDEXING_TYPES: |
| case ALL_ARRAY_STORAGE_INDEXING_TYPES: { |
| JSValueOperand operand(this, use); |
| GPRReg opGPR = operand.gpr(); |
| m_jit.store64(opGPR, buffer + operandIdx); |
| operand.use(); |
| break; |
| } |
| default: |
| CRASH(); |
| break; |
| } |
| } |
| |
| switch (node->indexingType()) { |
| case ALL_DOUBLE_INDEXING_TYPES: |
| case ALL_INT32_INDEXING_TYPES: |
| useChildren(node); |
| break; |
| default: |
| break; |
| } |
| |
| flushRegisters(); |
| |
| if (scratchSize) { |
| GPRTemporary scratch(this); |
| |
| // Tell GC mark phase how much of the scratch buffer is active during call. |
| m_jit.move(TrustedImmPtr(scratchBuffer->activeLengthPtr()), scratch.gpr()); |
| m_jit.storePtr(TrustedImmPtr(scratchSize), scratch.gpr()); |
| } |
| |
| GPRResult result(this); |
| |
| callOperation( |
| operationNewArray, result.gpr(), globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType()), |
| static_cast<void*>(buffer), node->numChildren()); |
| |
| if (scratchSize) { |
| GPRTemporary scratch(this); |
| |
| m_jit.move(TrustedImmPtr(scratchBuffer->activeLengthPtr()), scratch.gpr()); |
| m_jit.storePtr(TrustedImmPtr(0), scratch.gpr()); |
| } |
| |
| cellResult(result.gpr(), node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case NewArrayWithSize: { |
| JSGlobalObject* globalObject = m_jit.graph().globalObjectFor(node->codeOrigin); |
| if (!globalObject->isHavingABadTime() && !hasArrayStorage(node->indexingType())) { |
| SpeculateStrictInt32Operand size(this, node->child1()); |
| GPRTemporary result(this); |
| GPRTemporary storage(this); |
| GPRTemporary scratch(this); |
| GPRTemporary scratch2(this); |
| |
| GPRReg sizeGPR = size.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| GPRReg storageGPR = storage.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| GPRReg scratch2GPR = scratch2.gpr(); |
| |
| MacroAssembler::JumpList slowCases; |
| slowCases.append(m_jit.branch32(MacroAssembler::AboveOrEqual, sizeGPR, TrustedImm32(MIN_SPARSE_ARRAY_INDEX))); |
| |
| ASSERT((1 << 3) == sizeof(JSValue)); |
| m_jit.move(sizeGPR, scratchGPR); |
| m_jit.lshift32(TrustedImm32(3), scratchGPR); |
| m_jit.add32(TrustedImm32(sizeof(IndexingHeader)), scratchGPR, resultGPR); |
| slowCases.append( |
| emitAllocateBasicStorage(resultGPR, storageGPR)); |
| m_jit.subPtr(scratchGPR, storageGPR); |
| Structure* structure = globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType()); |
| emitAllocateJSObject<JSArray>(resultGPR, ImmPtr(structure), storageGPR, scratchGPR, scratch2GPR, slowCases); |
| |
| m_jit.store32(sizeGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfPublicLength())); |
| m_jit.store32(sizeGPR, MacroAssembler::Address(storageGPR, Butterfly::offsetOfVectorLength())); |
| |
| if (hasDouble(node->indexingType())) { |
| m_jit.move(TrustedImm64(bitwise_cast<int64_t>(QNaN)), scratchGPR); |
| m_jit.move(sizeGPR, scratch2GPR); |
| MacroAssembler::Jump done = m_jit.branchTest32(MacroAssembler::Zero, scratch2GPR); |
| MacroAssembler::Label loop = m_jit.label(); |
| m_jit.sub32(TrustedImm32(1), scratch2GPR); |
| m_jit.store64(scratchGPR, MacroAssembler::BaseIndex(storageGPR, scratch2GPR, MacroAssembler::TimesEight)); |
| m_jit.branchTest32(MacroAssembler::NonZero, scratch2GPR).linkTo(loop, &m_jit); |
| done.link(&m_jit); |
| } |
| |
| addSlowPathGenerator(adoptPtr( |
| new CallArrayAllocatorWithVariableSizeSlowPathGenerator( |
| slowCases, this, operationNewArrayWithSize, resultGPR, |
| globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType()), |
| globalObject->arrayStructureForIndexingTypeDuringAllocation(ArrayWithArrayStorage), |
| sizeGPR))); |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| SpeculateStrictInt32Operand size(this, node->child1()); |
| GPRReg sizeGPR = size.gpr(); |
| flushRegisters(); |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| GPRReg structureGPR = selectScratchGPR(sizeGPR); |
| MacroAssembler::Jump bigLength = m_jit.branch32(MacroAssembler::AboveOrEqual, sizeGPR, TrustedImm32(MIN_SPARSE_ARRAY_INDEX)); |
| m_jit.move(TrustedImmPtr(globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType())), structureGPR); |
| MacroAssembler::Jump done = m_jit.jump(); |
| bigLength.link(&m_jit); |
| m_jit.move(TrustedImmPtr(globalObject->arrayStructureForIndexingTypeDuringAllocation(ArrayWithArrayStorage)), structureGPR); |
| done.link(&m_jit); |
| callOperation(operationNewArrayWithSize, resultGPR, structureGPR, sizeGPR); |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case NewArrayBuffer: { |
| JSGlobalObject* globalObject = m_jit.graph().globalObjectFor(node->codeOrigin); |
| IndexingType indexingType = node->indexingType(); |
| if (!globalObject->isHavingABadTime() && !hasArrayStorage(indexingType)) { |
| unsigned numElements = node->numConstants(); |
| |
| GPRTemporary result(this); |
| GPRTemporary storage(this); |
| |
| GPRReg resultGPR = result.gpr(); |
| GPRReg storageGPR = storage.gpr(); |
| |
| emitAllocateJSArray(resultGPR, globalObject->arrayStructureForIndexingTypeDuringAllocation(indexingType), storageGPR, numElements); |
| |
| RELEASE_ASSERT(indexingType & IsArray); |
| JSValue* data = m_jit.codeBlock()->constantBuffer(node->startConstant()); |
| if (indexingType == ArrayWithDouble) { |
| for (unsigned index = 0; index < node->numConstants(); ++index) { |
| double value = data[index].asNumber(); |
| m_jit.store64( |
| Imm64(bitwise_cast<int64_t>(value)), |
| MacroAssembler::Address(storageGPR, sizeof(double) * index)); |
| } |
| } else { |
| for (unsigned index = 0; index < node->numConstants(); ++index) { |
| m_jit.store64( |
| Imm64(JSValue::encode(data[index])), |
| MacroAssembler::Address(storageGPR, sizeof(JSValue) * index)); |
| } |
| } |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| flushRegisters(); |
| GPRResult result(this); |
| |
| callOperation(operationNewArrayBuffer, result.gpr(), globalObject->arrayStructureForIndexingTypeDuringAllocation(node->indexingType()), node->startConstant(), node->numConstants()); |
| |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| case NewTypedArray: { |
| switch (node->child1().useKind()) { |
| case Int32Use: |
| compileNewTypedArray(node); |
| break; |
| case UntypedUse: { |
| JSValueOperand argument(this, node->child1()); |
| GPRReg argumentGPR = argument.gpr(); |
| |
| flushRegisters(); |
| |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| JSGlobalObject* globalObject = m_jit.graph().globalObjectFor(node->codeOrigin); |
| callOperation( |
| operationNewTypedArrayWithOneArgumentForType(node->typedArrayType()), |
| resultGPR, globalObject->typedArrayStructure(node->typedArrayType()), |
| argumentGPR); |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| break; |
| } |
| |
| case NewRegexp: { |
| flushRegisters(); |
| GPRResult result(this); |
| |
| callOperation(operationNewRegexp, result.gpr(), m_jit.codeBlock()->regexp(node->regexpIndex())); |
| |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| case ToThis: { |
| ASSERT(node->child1().useKind() == UntypedUse); |
| JSValueOperand thisValue(this, node->child1()); |
| GPRTemporary temp(this); |
| GPRReg thisValueGPR = thisValue.gpr(); |
| GPRReg tempGPR = temp.gpr(); |
| |
| MacroAssembler::JumpList slowCases; |
| slowCases.append(m_jit.branchTest64( |
| MacroAssembler::NonZero, thisValueGPR, GPRInfo::tagMaskRegister)); |
| m_jit.loadPtr( |
| MacroAssembler::Address(thisValueGPR, JSCell::structureOffset()), tempGPR); |
| slowCases.append(m_jit.branch8( |
| MacroAssembler::NotEqual, |
| MacroAssembler::Address(tempGPR, Structure::typeInfoTypeOffset()), |
| TrustedImm32(FinalObjectType))); |
| m_jit.move(thisValueGPR, tempGPR); |
| J_JITOperation_EJ function; |
| if (m_jit.graph().executableFor(node->codeOrigin)->isStrictMode()) |
| function = operationToThisStrict; |
| else |
| function = operationToThis; |
| addSlowPathGenerator( |
| slowPathCall(slowCases, this, function, tempGPR, thisValueGPR)); |
| |
| jsValueResult(tempGPR, node); |
| break; |
| } |
| |
| case CreateThis: { |
| // Note that there is not so much profit to speculate here. The only things we |
| // speculate on are (1) that it's a cell, since that eliminates cell checks |
| // later if the proto is reused, and (2) if we have a FinalObject prediction |
| // then we speculate because we want to get recompiled if it isn't (since |
| // otherwise we'd start taking slow path a lot). |
| |
| SpeculateCellOperand callee(this, node->child1()); |
| GPRTemporary result(this); |
| GPRTemporary allocator(this); |
| GPRTemporary structure(this); |
| GPRTemporary scratch(this); |
| |
| GPRReg calleeGPR = callee.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| GPRReg allocatorGPR = allocator.gpr(); |
| GPRReg structureGPR = structure.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| MacroAssembler::JumpList slowPath; |
| |
| m_jit.loadPtr(JITCompiler::Address(calleeGPR, JSFunction::offsetOfAllocationProfile() + ObjectAllocationProfile::offsetOfAllocator()), allocatorGPR); |
| m_jit.loadPtr(JITCompiler::Address(calleeGPR, JSFunction::offsetOfAllocationProfile() + ObjectAllocationProfile::offsetOfStructure()), structureGPR); |
| slowPath.append(m_jit.branchTestPtr(MacroAssembler::Zero, allocatorGPR)); |
| emitAllocateJSObject(resultGPR, allocatorGPR, structureGPR, TrustedImmPtr(0), scratchGPR, slowPath); |
| |
| addSlowPathGenerator(slowPathCall(slowPath, this, operationCreateThis, resultGPR, calleeGPR, node->inlineCapacity())); |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case AllocationProfileWatchpoint: { |
| noResult(node); |
| break; |
| } |
| |
| case NewObject: { |
| GPRTemporary result(this); |
| GPRTemporary allocator(this); |
| GPRTemporary scratch(this); |
| |
| GPRReg resultGPR = result.gpr(); |
| GPRReg allocatorGPR = allocator.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| MacroAssembler::JumpList slowPath; |
| |
| Structure* structure = node->structure(); |
| size_t allocationSize = JSFinalObject::allocationSize(structure->inlineCapacity()); |
| MarkedAllocator* allocatorPtr = &m_jit.vm()->heap.allocatorForObjectWithoutDestructor(allocationSize); |
| |
| m_jit.move(TrustedImmPtr(allocatorPtr), allocatorGPR); |
| emitAllocateJSObject(resultGPR, allocatorGPR, TrustedImmPtr(structure), TrustedImmPtr(0), scratchGPR, slowPath); |
| |
| addSlowPathGenerator(slowPathCall(slowPath, this, operationNewObject, resultGPR, structure)); |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case GetCallee: { |
| GPRTemporary result(this); |
| m_jit.loadPtr(JITCompiler::addressFor(JSStack::Callee), result.gpr()); |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| case GetScope: { |
| SpeculateCellOperand function(this, node->child1()); |
| GPRTemporary result(this, Reuse, function); |
| m_jit.loadPtr(JITCompiler::Address(function.gpr(), JSFunction::offsetOfScopeChain()), result.gpr()); |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| case GetMyScope: { |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.loadPtr(JITCompiler::addressFor(JSStack::ScopeChain), resultGPR); |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case SkipTopScope: { |
| SpeculateCellOperand scope(this, node->child1()); |
| GPRTemporary result(this, Reuse, scope); |
| GPRReg resultGPR = result.gpr(); |
| m_jit.move(scope.gpr(), resultGPR); |
| JITCompiler::Jump activationNotCreated = |
| m_jit.branchTest64( |
| JITCompiler::Zero, |
| JITCompiler::addressFor( |
| static_cast<VirtualRegister>(m_jit.graph().machineActivationRegister()))); |
| m_jit.loadPtr(JITCompiler::Address(resultGPR, JSScope::offsetOfNext()), resultGPR); |
| activationNotCreated.link(&m_jit); |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case SkipScope: { |
| SpeculateCellOperand scope(this, node->child1()); |
| GPRTemporary result(this, Reuse, scope); |
| m_jit.loadPtr(JITCompiler::Address(scope.gpr(), JSScope::offsetOfNext()), result.gpr()); |
| cellResult(result.gpr(), node); |
| break; |
| } |
| |
| case GetClosureRegisters: { |
| SpeculateCellOperand scope(this, node->child1()); |
| GPRTemporary result(this); |
| GPRReg scopeGPR = scope.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.loadPtr(JITCompiler::Address(scopeGPR, JSVariableObject::offsetOfRegisters()), resultGPR); |
| storageResult(resultGPR, node); |
| break; |
| } |
| case GetClosureVar: { |
| StorageOperand registers(this, node->child1()); |
| GPRTemporary result(this); |
| GPRReg registersGPR = registers.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.load64(JITCompiler::Address(registersGPR, node->varNumber() * sizeof(Register)), resultGPR); |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| case PutClosureVar: { |
| SpeculateCellOperand scope(this, node->child1()); |
| StorageOperand registers(this, node->child2()); |
| JSValueOperand value(this, node->child3()); |
| GPRTemporary scratchRegister(this); |
| |
| GPRReg scopeGPR = scope.gpr(); |
| GPRReg registersGPR = registers.gpr(); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg scratchGPR = scratchRegister.gpr(); |
| |
| m_jit.store64(valueGPR, JITCompiler::Address(registersGPR, node->varNumber() * sizeof(Register))); |
| writeBarrier(scopeGPR, valueGPR, node->child3(), WriteBarrierForVariableAccess, scratchGPR); |
| noResult(node); |
| break; |
| } |
| case GetById: { |
| ASSERT(node->prediction()); |
| |
| switch (node->child1().useKind()) { |
| case CellUse: { |
| SpeculateCellOperand base(this, node->child1()); |
| GPRTemporary result(this, Reuse, base); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| base.use(); |
| |
| cachedGetById(node->codeOrigin, baseGPR, resultGPR, node->identifierNumber()); |
| |
| jsValueResult(resultGPR, node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case UntypedUse: { |
| JSValueOperand base(this, node->child1()); |
| GPRTemporary result(this, Reuse, base); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| base.use(); |
| |
| JITCompiler::Jump notCell = m_jit.branchTest64(JITCompiler::NonZero, baseGPR, GPRInfo::tagMaskRegister); |
| |
| cachedGetById(node->codeOrigin, baseGPR, resultGPR, node->identifierNumber(), notCell); |
| |
| jsValueResult(resultGPR, node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| break; |
| } |
| |
| case GetByIdFlush: { |
| if (!node->prediction()) { |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| break; |
| } |
| |
| switch (node->child1().useKind()) { |
| case CellUse: { |
| SpeculateCellOperand base(this, node->child1()); |
| GPRReg baseGPR = base.gpr(); |
| |
| GPRResult result(this); |
| |
| GPRReg resultGPR = result.gpr(); |
| |
| base.use(); |
| |
| flushRegisters(); |
| |
| cachedGetById(node->codeOrigin, baseGPR, resultGPR, node->identifierNumber(), JITCompiler::Jump(), DontSpill); |
| |
| jsValueResult(resultGPR, node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case UntypedUse: { |
| JSValueOperand base(this, node->child1()); |
| GPRReg baseGPR = base.gpr(); |
| |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| base.use(); |
| flushRegisters(); |
| |
| JITCompiler::Jump notCell = m_jit.branchTest64(JITCompiler::NonZero, baseGPR, GPRInfo::tagMaskRegister); |
| |
| cachedGetById(node->codeOrigin, baseGPR, resultGPR, node->identifierNumber(), notCell, DontSpill); |
| |
| jsValueResult(resultGPR, node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| default: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| break; |
| } |
| |
| case GetArrayLength: |
| compileGetArrayLength(node); |
| break; |
| |
| case CheckFunction: { |
| SpeculateCellOperand function(this, node->child1()); |
| speculationCheck(BadFunction, JSValueSource::unboxedCell(function.gpr()), node->child1(), m_jit.branchWeakPtr(JITCompiler::NotEqual, function.gpr(), node->function())); |
| noResult(node); |
| break; |
| } |
| |
| case CheckExecutable: { |
| SpeculateCellOperand function(this, node->child1()); |
| speculationCheck(BadExecutable, JSValueSource::unboxedCell(function.gpr()), node->child1(), m_jit.branchWeakPtr(JITCompiler::NotEqual, JITCompiler::Address(function.gpr(), JSFunction::offsetOfExecutable()), node->executable())); |
| noResult(node); |
| break; |
| } |
| |
| case CheckStructure: { |
| SpeculateCellOperand base(this, node->child1()); |
| |
| ASSERT(node->structureSet().size()); |
| |
| ExitKind exitKind; |
| if (node->child1()->op() == WeakJSConstant) |
| exitKind = BadWeakConstantCache; |
| else |
| exitKind = BadCache; |
| |
| if (node->structureSet().size() == 1) { |
| speculationCheck( |
| exitKind, JSValueSource::unboxedCell(base.gpr()), 0, |
| m_jit.branchWeakPtr( |
| JITCompiler::NotEqual, |
| JITCompiler::Address(base.gpr(), JSCell::structureOffset()), |
| node->structureSet()[0])); |
| } else { |
| GPRTemporary structure(this); |
| |
| m_jit.loadPtr(JITCompiler::Address(base.gpr(), JSCell::structureOffset()), structure.gpr()); |
| |
| JITCompiler::JumpList done; |
| |
| for (size_t i = 0; i < node->structureSet().size() - 1; ++i) |
| done.append(m_jit.branchWeakPtr(JITCompiler::Equal, structure.gpr(), node->structureSet()[i])); |
| |
| speculationCheck( |
| exitKind, JSValueSource::unboxedCell(base.gpr()), 0, |
| m_jit.branchWeakPtr( |
| JITCompiler::NotEqual, structure.gpr(), node->structureSet().last())); |
| |
| done.link(&m_jit); |
| } |
| |
| noResult(node); |
| break; |
| } |
| |
| case StructureTransitionWatchpoint: { |
| // There is a fascinating question here of what to do about array profiling. |
| // We *could* try to tell the OSR exit about where the base of the access is. |
| // The DFG will have kept it alive, though it may not be in a register, and |
| // we shouldn't really load it since that could be a waste. For now though, |
| // we'll just rely on the fact that when a watchpoint fires then that's |
| // quite a hint already. |
| |
| m_jit.addWeakReference(node->structure()); |
| |
| #if !ASSERT_DISABLED |
| SpeculateCellOperand op1(this, node->child1()); |
| JITCompiler::Jump isOK = m_jit.branchPtr(JITCompiler::Equal, JITCompiler::Address(op1.gpr(), JSCell::structureOffset()), TrustedImmPtr(node->structure())); |
| m_jit.breakpoint(); |
| isOK.link(&m_jit); |
| #else |
| speculateCell(node->child1()); |
| #endif |
| |
| noResult(node); |
| break; |
| } |
| |
| case PhantomPutStructure: { |
| ASSERT(isKnownCell(node->child1().node())); |
| m_jit.jitCode()->common.notifyCompilingStructureTransition(m_jit.graph().m_plan, m_jit.codeBlock(), node); |
| noResult(node); |
| break; |
| } |
| |
| case PutStructure: { |
| m_jit.jitCode()->common.notifyCompilingStructureTransition(m_jit.graph().m_plan, m_jit.codeBlock(), node); |
| |
| SpeculateCellOperand base(this, node->child1()); |
| GPRReg baseGPR = base.gpr(); |
| |
| |
| #if ENABLE(WRITE_BARRIER_PROFILING) |
| // Must always emit this write barrier as the structure transition itself requires it |
| writeBarrier(baseGPR, node->structureTransitionData().newStructure, WriteBarrierForGenericAccess); |
| #endif |
| |
| m_jit.storePtr(MacroAssembler::TrustedImmPtr(node->structureTransitionData().newStructure), MacroAssembler::Address(baseGPR, JSCell::structureOffset())); |
| |
| noResult(node); |
| break; |
| } |
| |
| case AllocatePropertyStorage: |
| compileAllocatePropertyStorage(node); |
| break; |
| |
| case ReallocatePropertyStorage: |
| compileReallocatePropertyStorage(node); |
| break; |
| |
| case GetButterfly: { |
| SpeculateCellOperand base(this, node->child1()); |
| GPRTemporary result(this, Reuse, base); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.loadPtr(JITCompiler::Address(baseGPR, JSObject::butterflyOffset()), resultGPR); |
| |
| storageResult(resultGPR, node); |
| break; |
| } |
| |
| case GetIndexedPropertyStorage: { |
| compileGetIndexedPropertyStorage(node); |
| break; |
| } |
| |
| case GetTypedArrayByteOffset: { |
| compileGetTypedArrayByteOffset(node); |
| break; |
| } |
| |
| case GetByOffset: { |
| StorageOperand storage(this, node->child1()); |
| GPRTemporary result(this, Reuse, storage); |
| |
| GPRReg storageGPR = storage.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| StorageAccessData& storageAccessData = m_jit.graph().m_storageAccessData[node->storageAccessDataIndex()]; |
| |
| m_jit.load64(JITCompiler::Address(storageGPR, offsetRelativeToBase(storageAccessData.offset)), resultGPR); |
| |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| |
| case PutByOffset: { |
| #if ENABLE(WRITE_BARRIER_PROFILING) |
| SpeculateCellOperand base(this, node->child2()); |
| #endif |
| StorageOperand storage(this, node->child1()); |
| JSValueOperand value(this, node->child3()); |
| |
| GPRReg storageGPR = storage.gpr(); |
| GPRReg valueGPR = value.gpr(); |
| |
| #if ENABLE(WRITE_BARRIER_PROFILING) |
| writeBarrier(base.gpr(), value.gpr(), node->child3(), WriteBarrierForPropertyAccess); |
| #endif |
| |
| StorageAccessData& storageAccessData = m_jit.graph().m_storageAccessData[node->storageAccessDataIndex()]; |
| |
| m_jit.store64(valueGPR, JITCompiler::Address(storageGPR, offsetRelativeToBase(storageAccessData.offset))); |
| |
| noResult(node); |
| break; |
| } |
| |
| case PutById: { |
| SpeculateCellOperand base(this, node->child1()); |
| JSValueOperand value(this, node->child2()); |
| GPRTemporary scratch(this); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| base.use(); |
| value.use(); |
| |
| cachedPutById(node->codeOrigin, baseGPR, valueGPR, node->child2(), scratchGPR, node->identifierNumber(), NotDirect); |
| |
| noResult(node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case PutByIdDirect: { |
| SpeculateCellOperand base(this, node->child1()); |
| JSValueOperand value(this, node->child2()); |
| GPRTemporary scratch(this); |
| |
| GPRReg baseGPR = base.gpr(); |
| GPRReg valueGPR = value.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| base.use(); |
| value.use(); |
| |
| cachedPutById(node->codeOrigin, baseGPR, valueGPR, node->child2(), scratchGPR, node->identifierNumber(), Direct); |
| |
| noResult(node, UseChildrenCalledExplicitly); |
| break; |
| } |
| |
| case GetGlobalVar: { |
| GPRTemporary result(this); |
| |
| m_jit.load64(node->registerPointer(), result.gpr()); |
| |
| jsValueResult(result.gpr(), node); |
| break; |
| } |
| |
| case PutGlobalVar: { |
| JSValueOperand value(this, node->child1()); |
| |
| if (Heap::isWriteBarrierEnabled()) { |
| GPRTemporary scratch(this); |
| GPRReg scratchReg = scratch.gpr(); |
| |
| writeBarrier(m_jit.globalObjectFor(node->codeOrigin), value.gpr(), node->child1(), WriteBarrierForVariableAccess, scratchReg); |
| } |
| |
| m_jit.store64(value.gpr(), node->registerPointer()); |
| |
| noResult(node); |
| break; |
| } |
| |
| case GlobalVarWatchpoint: { |
| #if DFG_ENABLE(JIT_ASSERT) |
| GPRTemporary scratch(this); |
| GPRReg scratchGPR = scratch.gpr(); |
| m_jit.load64(node->registerPointer(), scratchGPR); |
| JITCompiler::Jump ok = m_jit.branch64( |
| JITCompiler::Equal, scratchGPR, |
| TrustedImm64(JSValue::encode(node->registerPointer()->get()))); |
| m_jit.breakpoint(); |
| ok.link(&m_jit); |
| #endif |
| |
| noResult(node); |
| break; |
| } |
| |
| case VarInjectionWatchpoint: { |
| noResult(node); |
| break; |
| } |
| |
| case CheckHasInstance: { |
| SpeculateCellOperand base(this, node->child1()); |
| GPRTemporary structure(this); |
| |
| // Speculate that base 'ImplementsDefaultHasInstance'. |
| m_jit.loadPtr(MacroAssembler::Address(base.gpr(), JSCell::structureOffset()), structure.gpr()); |
| speculationCheck(Uncountable, JSValueRegs(), 0, m_jit.branchTest8(MacroAssembler::Zero, MacroAssembler::Address(structure.gpr(), Structure::typeInfoFlagsOffset()), MacroAssembler::TrustedImm32(ImplementsDefaultHasInstance))); |
| |
| noResult(node); |
| break; |
| } |
| |
| case InstanceOf: { |
| compileInstanceOf(node); |
| break; |
| } |
| |
| case IsUndefined: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this); |
| GPRTemporary localGlobalObject(this); |
| GPRTemporary remoteGlobalObject(this); |
| |
| JITCompiler::Jump isCell = m_jit.branchTest64(JITCompiler::Zero, value.gpr(), GPRInfo::tagMaskRegister); |
| |
| m_jit.compare64(JITCompiler::Equal, value.gpr(), TrustedImm32(ValueUndefined), result.gpr()); |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| isCell.link(&m_jit); |
| JITCompiler::Jump notMasqueradesAsUndefined; |
| if (masqueradesAsUndefinedWatchpointIsStillValid()) { |
| m_jit.move(TrustedImm32(0), result.gpr()); |
| notMasqueradesAsUndefined = m_jit.jump(); |
| } else { |
| m_jit.loadPtr(JITCompiler::Address(value.gpr(), JSCell::structureOffset()), result.gpr()); |
| JITCompiler::Jump isMasqueradesAsUndefined = m_jit.branchTest8(JITCompiler::NonZero, JITCompiler::Address(result.gpr(), Structure::typeInfoFlagsOffset()), TrustedImm32(MasqueradesAsUndefined)); |
| m_jit.move(TrustedImm32(0), result.gpr()); |
| notMasqueradesAsUndefined = m_jit.jump(); |
| |
| isMasqueradesAsUndefined.link(&m_jit); |
| GPRReg localGlobalObjectGPR = localGlobalObject.gpr(); |
| GPRReg remoteGlobalObjectGPR = remoteGlobalObject.gpr(); |
| m_jit.move(TrustedImmPtr(m_jit.globalObjectFor(node->codeOrigin)), localGlobalObjectGPR); |
| m_jit.loadPtr(JITCompiler::Address(result.gpr(), Structure::globalObjectOffset()), remoteGlobalObjectGPR); |
| m_jit.comparePtr(JITCompiler::Equal, localGlobalObjectGPR, remoteGlobalObjectGPR, result.gpr()); |
| } |
| |
| notMasqueradesAsUndefined.link(&m_jit); |
| done.link(&m_jit); |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case IsBoolean: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| m_jit.move(value.gpr(), result.gpr()); |
| m_jit.xor64(JITCompiler::TrustedImm32(ValueFalse), result.gpr()); |
| m_jit.test64(JITCompiler::Zero, result.gpr(), JITCompiler::TrustedImm32(static_cast<int32_t>(~1)), result.gpr()); |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case IsNumber: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| m_jit.test64(JITCompiler::NonZero, value.gpr(), GPRInfo::tagTypeNumberRegister, result.gpr()); |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case IsString: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| JITCompiler::Jump isNotCell = m_jit.branchTest64(JITCompiler::NonZero, value.gpr(), GPRInfo::tagMaskRegister); |
| |
| m_jit.loadPtr(JITCompiler::Address(value.gpr(), JSCell::structureOffset()), result.gpr()); |
| m_jit.compare8(JITCompiler::Equal, JITCompiler::Address(result.gpr(), Structure::typeInfoTypeOffset()), TrustedImm32(StringType), result.gpr()); |
| m_jit.or32(TrustedImm32(ValueFalse), result.gpr()); |
| JITCompiler::Jump done = m_jit.jump(); |
| |
| isNotCell.link(&m_jit); |
| m_jit.move(TrustedImm32(ValueFalse), result.gpr()); |
| |
| done.link(&m_jit); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case IsObject: { |
| JSValueOperand value(this, node->child1()); |
| GPRReg valueGPR = value.gpr(); |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| flushRegisters(); |
| callOperation(operationIsObject, resultGPR, valueGPR); |
| m_jit.or32(TrustedImm32(ValueFalse), resultGPR); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case IsFunction: { |
| JSValueOperand value(this, node->child1()); |
| GPRReg valueGPR = value.gpr(); |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| flushRegisters(); |
| callOperation(operationIsFunction, resultGPR, valueGPR); |
| m_jit.or32(TrustedImm32(ValueFalse), resultGPR); |
| jsValueResult(result.gpr(), node, DataFormatJSBoolean); |
| break; |
| } |
| |
| case TypeOf: { |
| JSValueOperand value(this, node->child1(), ManualOperandSpeculation); |
| GPRReg valueGPR = value.gpr(); |
| GPRTemporary temp(this); |
| GPRReg tempGPR = temp.gpr(); |
| GPRResult result(this); |
| GPRReg resultGPR = result.gpr(); |
| JITCompiler::JumpList doneJumps; |
| |
| flushRegisters(); |
| |
| ASSERT(node->child1().useKind() == UntypedUse || node->child1().useKind() == CellUse || node->child1().useKind() == StringUse); |
| |
| JITCompiler::Jump isNotCell = m_jit.branchTest64(JITCompiler::NonZero, valueGPR, GPRInfo::tagMaskRegister); |
| if (node->child1().useKind() != UntypedUse) |
| DFG_TYPE_CHECK(JSValueSource(valueGPR), node->child1(), SpecCell, isNotCell); |
| |
| if (!node->child1()->shouldSpeculateObject() || node->child1().useKind() == StringUse) { |
| m_jit.loadPtr(JITCompiler::Address(valueGPR, JSCell::structureOffset()), tempGPR); |
| JITCompiler::Jump notString = m_jit.branch8(JITCompiler::NotEqual, JITCompiler::Address(tempGPR, Structure::typeInfoTypeOffset()), TrustedImm32(StringType)); |
| if (node->child1().useKind() == StringUse) |
| DFG_TYPE_CHECK(JSValueSource(valueGPR), node->child1(), SpecString, notString); |
| m_jit.move(TrustedImmPtr(m_jit.vm()->smallStrings.stringString()), resultGPR); |
| doneJumps.append(m_jit.jump()); |
| if (node->child1().useKind() != StringUse) { |
| notString.link(&m_jit); |
| callOperation(operationTypeOf, resultGPR, valueGPR); |
| doneJumps.append(m_jit.jump()); |
| } |
| } else { |
| callOperation(operationTypeOf, resultGPR, valueGPR); |
| doneJumps.append(m_jit.jump()); |
| } |
| |
| if (node->child1().useKind() == UntypedUse) { |
| isNotCell.link(&m_jit); |
| JITCompiler::Jump notNumber = m_jit.branchTest64(JITCompiler::Zero, valueGPR, GPRInfo::tagTypeNumberRegister); |
| m_jit.move(TrustedImmPtr(m_jit.vm()->smallStrings.numberString()), resultGPR); |
| doneJumps.append(m_jit.jump()); |
| notNumber.link(&m_jit); |
| |
| JITCompiler::Jump notUndefined = m_jit.branch64(JITCompiler::NotEqual, valueGPR, JITCompiler::TrustedImm64(ValueUndefined)); |
| m_jit.move(TrustedImmPtr(m_jit.vm()->smallStrings.undefinedString()), resultGPR); |
| doneJumps.append(m_jit.jump()); |
| notUndefined.link(&m_jit); |
| |
| JITCompiler::Jump notNull = m_jit.branch64(JITCompiler::NotEqual, valueGPR, JITCompiler::TrustedImm64(ValueNull)); |
| m_jit.move(TrustedImmPtr(m_jit.vm()->smallStrings.objectString()), resultGPR); |
| doneJumps.append(m_jit.jump()); |
| notNull.link(&m_jit); |
| |
| // Only boolean left |
| m_jit.move(TrustedImmPtr(m_jit.vm()->smallStrings.booleanString()), resultGPR); |
| } |
| doneJumps.link(&m_jit); |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case Flush: |
| break; |
| |
| case Breakpoint: |
| #if ENABLE(DEBUG_WITH_BREAKPOINT) |
| m_jit.breakpoint(); |
| #else |
| RELEASE_ASSERT_NOT_REACHED(); |
| #endif |
| break; |
| |
| case Call: |
| case Construct: |
| emitCall(node); |
| break; |
| |
| case CreateActivation: { |
| RELEASE_ASSERT(!node->codeOrigin.inlineCallFrame); |
| |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| GPRReg valueGPR = value.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.move(valueGPR, resultGPR); |
| |
| JITCompiler::Jump notCreated = m_jit.branchTest64(JITCompiler::Zero, resultGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| notCreated, this, operationCreateActivation, resultGPR, |
| framePointerOffsetToGetActivationRegisters())); |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case CreateArguments: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| GPRReg valueGPR = value.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.move(valueGPR, resultGPR); |
| |
| JITCompiler::Jump notCreated = m_jit.branchTest64(JITCompiler::Zero, resultGPR); |
| |
| if (node->codeOrigin.inlineCallFrame) { |
| addSlowPathGenerator( |
| slowPathCall( |
| notCreated, this, operationCreateInlinedArguments, resultGPR, |
| node->codeOrigin.inlineCallFrame)); |
| } else { |
| addSlowPathGenerator( |
| slowPathCall(notCreated, this, operationCreateArguments, resultGPR)); |
| } |
| |
| cellResult(resultGPR, node); |
| break; |
| } |
| |
| case TearOffActivation: { |
| RELEASE_ASSERT(!node->codeOrigin.inlineCallFrame); |
| |
| JSValueOperand activationValue(this, node->child1()); |
| GPRTemporary scratch(this); |
| GPRReg activationValueGPR = activationValue.gpr(); |
| GPRReg scratchGPR = scratch.gpr(); |
| |
| JITCompiler::Jump notCreated = m_jit.branchTest64(JITCompiler::Zero, activationValueGPR); |
| |
| SharedSymbolTable* symbolTable = m_jit.symbolTableFor(node->codeOrigin); |
| int registersOffset = JSActivation::registersOffset(symbolTable); |
| |
| int bytecodeCaptureStart = symbolTable->captureStart(); |
| int machineCaptureStart = m_jit.graph().m_machineCaptureStart; |
| for (int i = symbolTable->captureCount(); i--;) { |
| m_jit.load64( |
| JITCompiler::Address( |
| GPRInfo::callFrameRegister, |
| (machineCaptureStart - i) * sizeof(Register)), |
| scratchGPR); |
| m_jit.store64( |
| scratchGPR, |
| JITCompiler::Address( |
| activationValueGPR, |
| registersOffset + (bytecodeCaptureStart - i) * sizeof(Register))); |
| } |
| m_jit.addPtr(TrustedImm32(registersOffset), activationValueGPR, scratchGPR); |
| m_jit.storePtr(scratchGPR, JITCompiler::Address(activationValueGPR, JSActivation::offsetOfRegisters())); |
| |
| notCreated.link(&m_jit); |
| noResult(node); |
| break; |
| } |
| |
| case TearOffArguments: { |
| JSValueOperand unmodifiedArgumentsValue(this, node->child1()); |
| JSValueOperand activationValue(this, node->child2()); |
| GPRReg unmodifiedArgumentsValueGPR = unmodifiedArgumentsValue.gpr(); |
| GPRReg activationValueGPR = activationValue.gpr(); |
| |
| JITCompiler::Jump created = m_jit.branchTest64(JITCompiler::NonZero, unmodifiedArgumentsValueGPR); |
| |
| if (node->codeOrigin.inlineCallFrame) { |
| addSlowPathGenerator( |
| slowPathCall( |
| created, this, operationTearOffInlinedArguments, NoResult, |
| unmodifiedArgumentsValueGPR, activationValueGPR, node->codeOrigin.inlineCallFrame)); |
| } else { |
| addSlowPathGenerator( |
| slowPathCall( |
| created, this, operationTearOffArguments, NoResult, unmodifiedArgumentsValueGPR, activationValueGPR)); |
| } |
| |
| noResult(node); |
| break; |
| } |
| |
| case GetMyArgumentsLength: { |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| if (!isEmptySpeculation( |
| m_state.variables().operand( |
| m_jit.graph().argumentsRegisterFor(node->codeOrigin)).m_type)) { |
| speculationCheck( |
| ArgumentsEscaped, JSValueRegs(), 0, |
| m_jit.branchTest64( |
| JITCompiler::NonZero, |
| JITCompiler::addressFor( |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin)))); |
| } |
| |
| RELEASE_ASSERT(!node->codeOrigin.inlineCallFrame); |
| m_jit.load32(JITCompiler::payloadFor(JSStack::ArgumentCount), resultGPR); |
| m_jit.sub32(TrustedImm32(1), resultGPR); |
| int32Result(resultGPR, node); |
| break; |
| } |
| |
| case GetMyArgumentsLengthSafe: { |
| GPRTemporary result(this); |
| GPRReg resultGPR = result.gpr(); |
| |
| JITCompiler::Jump created = m_jit.branchTest64( |
| JITCompiler::NonZero, |
| JITCompiler::addressFor( |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin))); |
| |
| if (node->codeOrigin.inlineCallFrame) { |
| m_jit.move( |
| Imm64(JSValue::encode(jsNumber(node->codeOrigin.inlineCallFrame->arguments.size() - 1))), |
| resultGPR); |
| } else { |
| m_jit.load32(JITCompiler::payloadFor(JSStack::ArgumentCount), resultGPR); |
| m_jit.sub32(TrustedImm32(1), resultGPR); |
| m_jit.or64(GPRInfo::tagTypeNumberRegister, resultGPR); |
| } |
| |
| // FIXME: the slow path generator should perform a forward speculation that the |
| // result is an integer. For now we postpone the speculation by having this return |
| // a JSValue. |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| created, this, operationGetArgumentsLength, resultGPR, |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin).offset())); |
| |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| |
| case GetMyArgumentByVal: { |
| SpeculateStrictInt32Operand index(this, node->child1()); |
| GPRTemporary result(this); |
| GPRReg indexGPR = index.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| if (!isEmptySpeculation( |
| m_state.variables().operand( |
| m_jit.graph().argumentsRegisterFor(node->codeOrigin)).m_type)) { |
| speculationCheck( |
| ArgumentsEscaped, JSValueRegs(), 0, |
| m_jit.branchTest64( |
| JITCompiler::NonZero, |
| JITCompiler::addressFor( |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin)))); |
| } |
| |
| m_jit.add32(TrustedImm32(1), indexGPR, resultGPR); |
| if (node->codeOrigin.inlineCallFrame) { |
| speculationCheck( |
| Uncountable, JSValueRegs(), 0, |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, |
| resultGPR, |
| Imm32(node->codeOrigin.inlineCallFrame->arguments.size()))); |
| } else { |
| speculationCheck( |
| Uncountable, JSValueRegs(), 0, |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, |
| resultGPR, |
| JITCompiler::payloadFor(JSStack::ArgumentCount))); |
| } |
| |
| JITCompiler::JumpList slowArgument; |
| JITCompiler::JumpList slowArgumentOutOfBounds; |
| if (m_jit.symbolTableFor(node->codeOrigin)->slowArguments()) { |
| RELEASE_ASSERT(!node->codeOrigin.inlineCallFrame); |
| const SlowArgument* slowArguments = m_jit.graph().m_slowArguments.get(); |
| |
| slowArgumentOutOfBounds.append( |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, indexGPR, |
| Imm32(m_jit.symbolTableFor(node->codeOrigin)->parameterCount()))); |
| |
| COMPILE_ASSERT(sizeof(SlowArgument) == 8, SlowArgument_size_is_eight_bytes); |
| m_jit.move(ImmPtr(slowArguments), resultGPR); |
| m_jit.load32( |
| JITCompiler::BaseIndex( |
| resultGPR, indexGPR, JITCompiler::TimesEight, |
| OBJECT_OFFSETOF(SlowArgument, index)), |
| resultGPR); |
| m_jit.signExtend32ToPtr(resultGPR, resultGPR); |
| m_jit.load64( |
| JITCompiler::BaseIndex( |
| GPRInfo::callFrameRegister, resultGPR, JITCompiler::TimesEight), |
| resultGPR); |
| slowArgument.append(m_jit.jump()); |
| } |
| slowArgumentOutOfBounds.link(&m_jit); |
| |
| m_jit.signExtend32ToPtr(resultGPR, resultGPR); |
| |
| m_jit.load64( |
| JITCompiler::BaseIndex( |
| GPRInfo::callFrameRegister, resultGPR, JITCompiler::TimesEight, m_jit.offsetOfArgumentsIncludingThis(node->codeOrigin)), |
| resultGPR); |
| |
| slowArgument.link(&m_jit); |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| |
| case GetMyArgumentByValSafe: { |
| SpeculateStrictInt32Operand index(this, node->child1()); |
| GPRTemporary result(this); |
| GPRReg indexGPR = index.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| JITCompiler::JumpList slowPath; |
| slowPath.append( |
| m_jit.branchTest64( |
| JITCompiler::NonZero, |
| JITCompiler::addressFor( |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin)))); |
| |
| m_jit.add32(TrustedImm32(1), indexGPR, resultGPR); |
| if (node->codeOrigin.inlineCallFrame) { |
| slowPath.append( |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, |
| resultGPR, |
| Imm32(node->codeOrigin.inlineCallFrame->arguments.size()))); |
| } else { |
| slowPath.append( |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, |
| resultGPR, |
| JITCompiler::payloadFor(JSStack::ArgumentCount))); |
| } |
| |
| JITCompiler::JumpList slowArgument; |
| JITCompiler::JumpList slowArgumentOutOfBounds; |
| if (m_jit.symbolTableFor(node->codeOrigin)->slowArguments()) { |
| RELEASE_ASSERT(!node->codeOrigin.inlineCallFrame); |
| const SlowArgument* slowArguments = m_jit.graph().m_slowArguments.get(); |
| |
| slowArgumentOutOfBounds.append( |
| m_jit.branch32( |
| JITCompiler::AboveOrEqual, indexGPR, |
| Imm32(m_jit.symbolTableFor(node->codeOrigin)->parameterCount()))); |
| |
| COMPILE_ASSERT(sizeof(SlowArgument) == 8, SlowArgument_size_is_eight_bytes); |
| m_jit.move(ImmPtr(slowArguments), resultGPR); |
| m_jit.load32( |
| JITCompiler::BaseIndex( |
| resultGPR, indexGPR, JITCompiler::TimesEight, |
| OBJECT_OFFSETOF(SlowArgument, index)), |
| resultGPR); |
| m_jit.signExtend32ToPtr(resultGPR, resultGPR); |
| m_jit.load64( |
| JITCompiler::BaseIndex( |
| GPRInfo::callFrameRegister, resultGPR, JITCompiler::TimesEight), |
| resultGPR); |
| slowArgument.append(m_jit.jump()); |
| } |
| slowArgumentOutOfBounds.link(&m_jit); |
| |
| m_jit.signExtend32ToPtr(resultGPR, resultGPR); |
| |
| m_jit.load64( |
| JITCompiler::BaseIndex( |
| GPRInfo::callFrameRegister, resultGPR, JITCompiler::TimesEight, m_jit.offsetOfArgumentsIncludingThis(node->codeOrigin)), |
| resultGPR); |
| |
| if (node->codeOrigin.inlineCallFrame) { |
| addSlowPathGenerator( |
| slowPathCall( |
| slowPath, this, operationGetInlinedArgumentByVal, resultGPR, |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin).offset(), |
| node->codeOrigin.inlineCallFrame, |
| indexGPR)); |
| } else { |
| addSlowPathGenerator( |
| slowPathCall( |
| slowPath, this, operationGetArgumentByVal, resultGPR, |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin).offset(), |
| indexGPR)); |
| } |
| |
| slowArgument.link(&m_jit); |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| |
| case CheckArgumentsNotCreated: { |
| ASSERT(!isEmptySpeculation( |
| m_state.variables().operand( |
| m_jit.graph().argumentsRegisterFor(node->codeOrigin)).m_type)); |
| speculationCheck( |
| ArgumentsEscaped, JSValueRegs(), 0, |
| m_jit.branchTest64( |
| JITCompiler::NonZero, |
| JITCompiler::addressFor( |
| m_jit.graph().machineArgumentsRegisterFor(node->codeOrigin)))); |
| noResult(node); |
| break; |
| } |
| |
| case NewFunctionNoCheck: |
| compileNewFunctionNoCheck(node); |
| break; |
| |
| case NewFunction: { |
| JSValueOperand value(this, node->child1()); |
| GPRTemporary result(this, Reuse, value); |
| |
| GPRReg valueGPR = value.gpr(); |
| GPRReg resultGPR = result.gpr(); |
| |
| m_jit.move(valueGPR, resultGPR); |
| |
| JITCompiler::Jump notCreated = m_jit.branchTest64(JITCompiler::Zero, resultGPR); |
| |
| addSlowPathGenerator( |
| slowPathCall( |
| notCreated, this, operationNewFunction, |
| resultGPR, m_jit.codeBlock()->functionDecl(node->functionDeclIndex()))); |
| |
| jsValueResult(resultGPR, node); |
| break; |
| } |
| |
| case NewFunctionExpression: |
| compileNewFunctionExpression(node); |
| break; |
| |
| case In: |
| compileIn(node); |
| break; |
| |
| case CountExecution: |
| m_jit.add64(TrustedImm32(1), MacroAssembler::AbsoluteAddress(node->executionCounter()->address())); |
| break; |
| |
| case ForceOSRExit: { |
| terminateSpeculativeExecution(InadequateCoverage, JSValueRegs(), 0); |
| break; |
| } |
| |
| case InvalidationPoint: |
| emitInvalidationPoint(node); |
| break; |
| |
| case CheckWatchdogTimer: |
| speculationCheck( |
| WatchdogTimerFired, JSValueRegs(), 0, |
| m_jit.branchTest8( |
| JITCompiler::NonZero, |
| JITCompiler::AbsoluteAddress(m_jit.vm()->watchdog.timerDidFireAddress()))); |
| break; |
| |
| case Phantom: |
| DFG_NODE_DO_TO_CHILDREN(m_jit.graph(), node, speculate); |
| noResult(node); |
| break; |
| |
| case PhantomLocal: |
| case LoopHint: |
| // This is a no-op. |
| noResult(node); |
| break; |
| |
| case Unreachable: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| |
| #if ENABLE(FTL_JIT) |
| case CheckTierUpInLoop: { |
| MacroAssembler::Jump done = m_jit.branchAdd32( |
| MacroAssembler::Signed, |
| TrustedImm32(Options::ftlTierUpCounterIncrementForLoop()), |
| MacroAssembler::AbsoluteAddress(&m_jit.jitCode()->tierUpCounter.m_counter)); |
| |
| silentSpillAllRegisters(InvalidGPRReg); |
| m_jit.setupArgumentsExecState(); |
| appendCall(triggerTierUpNow); |
| silentFillAllRegisters(InvalidGPRReg); |
| |
| done.link(&m_jit); |
| break; |
| } |
| |
| case CheckTierUpAtReturn: { |
| MacroAssembler::Jump done = m_jit.branchAdd32( |
| MacroAssembler::Signed, |
| TrustedImm32(Options::ftlTierUpCounterIncrementForReturn()), |
| MacroAssembler::AbsoluteAddress(&m_jit.jitCode()->tierUpCounter.m_counter)); |
| |
| silentSpillAllRegisters(InvalidGPRReg); |
| m_jit.setupArgumentsExecState(); |
| appendCall(triggerTierUpNow); |
| silentFillAllRegisters(InvalidGPRReg); |
| |
| done.link(&m_jit); |
| break; |
| } |
| |
| case CheckTierUpAndOSREnter: { |
| ASSERT(!node->codeOrigin.inlineCallFrame); |
| |
| GPRTemporary temp(this); |
| GPRReg tempGPR = temp.gpr(); |
| |
| MacroAssembler::Jump done = m_jit.branchAdd32( |
| MacroAssembler::Signed, |
| TrustedImm32(Options::ftlTierUpCounterIncrementForLoop()), |
| MacroAssembler::AbsoluteAddress(&m_jit.jitCode()->tierUpCounter.m_counter)); |
| |
| silentSpillAllRegisters(tempGPR); |
| m_jit.setupArgumentsWithExecState( |
| TrustedImm32(node->codeOrigin.bytecodeIndex), |
| TrustedImm32(m_stream->size())); |
| appendCallSetResult(triggerOSREntryNow, tempGPR); |
| MacroAssembler::Jump dontEnter = m_jit.branchTestPtr(MacroAssembler::Zero, tempGPR); |
| m_jit.jump(tempGPR); |
| dontEnter.link(&m_jit); |
| silentFillAllRegisters(tempGPR); |
| |
| done.link(&m_jit); |
| break; |
| } |
| #else // ENABLE(FTL_JIT) |
| case CheckTierUpInLoop: |
| case CheckTierUpAtReturn: |
| case CheckTierUpAndOSREnter: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| #endif // ENABLE(FTL_JIT) |
| |
| case LastNodeType: |
| case Phi: |
| case Upsilon: |
| case GetArgument: |
| case ExtractOSREntryLocal: |
| RELEASE_ASSERT_NOT_REACHED(); |
| break; |
| } |
| |
| if (!m_compileOkay) |
| return; |
| |
| if (node->hasResult() && node->mustGenerate()) |
| use(node); |
| } |
| |
| #endif |
| |
| } } // namespace JSC::DFG |
| |
| #endif |