unsigned FastISel::getRegForGEPIndex(Value *Idx) {
unsigned IdxN = getRegForValue(Idx);
if (IdxN == 0)
// Unhandled operand. Halt "fast" selection and bail.
return 0;
// If the index is smaller or larger than intptr_t, truncate or extend it.
MVT PtrVT = TLI.getPointerTy();
EVT IdxVT = EVT::getEVT(Idx->getType(), /*HandleUnknown=*/false);
if (IdxVT.bitsLT(PtrVT))
IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::SIGN_EXTEND, IdxN);
else if (IdxVT.bitsGT(PtrVT))
IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::TRUNCATE, IdxN);
return IdxN;
}
std::pair<unsigned, bool> FastISel::getRegForGEPIndex(const Value *Idx) {
unsigned IdxN = getRegForValue(Idx);
if (IdxN == 0)
// Unhandled operand. Halt "fast" selection and bail.
return std::pair<unsigned, bool>(0, false);
bool IdxNIsKill = hasTrivialKill(Idx);
// If the index is smaller or larger than intptr_t, truncate or extend it.
MVT PtrVT = TLI.getPointerTy();
EVT IdxVT = EVT::getEVT(Idx->getType(), /*HandleUnknown=*/false);
if (IdxVT.bitsLT(PtrVT)) {
IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::SIGN_EXTEND,
IdxN, IdxNIsKill);
IdxNIsKill = true;
}
else if (IdxVT.bitsGT(PtrVT)) {
IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::TRUNCATE,
IdxN, IdxNIsKill);
IdxNIsKill = true;
}
return std::pair<unsigned, bool>(IdxN, IdxNIsKill);
}
bool
FastISel::SelectOperator(const User *I, unsigned Opcode) {
switch (Opcode) {
case Instruction::Add:
return SelectBinaryOp(I, ISD::ADD);
case Instruction::FAdd:
return SelectBinaryOp(I, ISD::FADD);
case Instruction::Sub:
return SelectBinaryOp(I, ISD::SUB);
case Instruction::FSub:
// FNeg is currently represented in LLVM IR as a special case of FSub.
if (BinaryOperator::isFNeg(I))
return SelectFNeg(I);
return SelectBinaryOp(I, ISD::FSUB);
case Instruction::Mul:
return SelectBinaryOp(I, ISD::MUL);
case Instruction::FMul:
return SelectBinaryOp(I, ISD::FMUL);
case Instruction::SDiv:
return SelectBinaryOp(I, ISD::SDIV);
case Instruction::UDiv:
return SelectBinaryOp(I, ISD::UDIV);
case Instruction::FDiv:
return SelectBinaryOp(I, ISD::FDIV);
case Instruction::SRem:
return SelectBinaryOp(I, ISD::SREM);
case Instruction::URem:
return SelectBinaryOp(I, ISD::UREM);
case Instruction::FRem:
return SelectBinaryOp(I, ISD::FREM);
case Instruction::Shl:
return SelectBinaryOp(I, ISD::SHL);
case Instruction::LShr:
return SelectBinaryOp(I, ISD::SRL);
case Instruction::AShr:
return SelectBinaryOp(I, ISD::SRA);
case Instruction::And:
return SelectBinaryOp(I, ISD::AND);
case Instruction::Or:
return SelectBinaryOp(I, ISD::OR);
case Instruction::Xor:
return SelectBinaryOp(I, ISD::XOR);
case Instruction::GetElementPtr:
return SelectGetElementPtr(I);
case Instruction::Br: {
const BranchInst *BI = cast<BranchInst>(I);
if (BI->isUnconditional()) {
const BasicBlock *LLVMSucc = BI->getSuccessor(0);
MachineBasicBlock *MSucc = FuncInfo.MBBMap[LLVMSucc];
FastEmitBranch(MSucc, BI->getDebugLoc());
return true;
}
// Conditional branches are not handed yet.
// Halt "fast" selection and bail.
return false;
}
case Instruction::Unreachable:
// Nothing to emit.
return true;
case Instruction::Alloca:
// FunctionLowering has the static-sized case covered.
if (FuncInfo.StaticAllocaMap.count(cast<AllocaInst>(I)))
return true;
// Dynamic-sized alloca is not handled yet.
return false;
case Instruction::Call:
return SelectCall(I);
case Instruction::BitCast:
return SelectBitCast(I);
case Instruction::FPToSI:
return SelectCast(I, ISD::FP_TO_SINT);
case Instruction::ZExt:
return SelectCast(I, ISD::ZERO_EXTEND);
case Instruction::SExt:
return SelectCast(I, ISD::SIGN_EXTEND);
case Instruction::Trunc:
return SelectCast(I, ISD::TRUNCATE);
case Instruction::SIToFP:
return SelectCast(I, ISD::SINT_TO_FP);
case Instruction::IntToPtr: // Deliberate fall-through.
case Instruction::PtrToInt: {
EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
EVT DstVT = TLI.getValueType(I->getType());
if (DstVT.bitsGT(SrcVT))
return SelectCast(I, ISD::ZERO_EXTEND);
if (DstVT.bitsLT(SrcVT))
return SelectCast(I, ISD::TRUNCATE);
unsigned Reg = getRegForValue(I->getOperand(0));
if (Reg == 0) return false;
UpdateValueMap(I, Reg);
return true;
}
case Instruction::PHI:
llvm_unreachable("FastISel shouldn't visit PHI nodes!");
default:
// Unhandled instruction. Halt "fast" selection and bail.
return false;
}
}
bool FastISel::SelectCall(const User *I) {
const Function *F = cast<CallInst>(I)->getCalledFunction();
if (!F) return false;
// Handle selected intrinsic function calls.
unsigned IID = F->getIntrinsicID();
switch (IID) {
default: break;
case Intrinsic::dbg_declare: {
const DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
if (!DIVariable(DI->getVariable()).Verify() ||
!FuncInfo.MF->getMMI().hasDebugInfo())
return true;
const Value *Address = DI->getAddress();
if (!Address || isa<UndefValue>(Address) || isa<AllocaInst>(Address))
return true;
unsigned Reg = 0;
unsigned Offset = 0;
if (const Argument *Arg = dyn_cast<Argument>(Address)) {
if (Arg->hasByValAttr()) {
// Byval arguments' frame index is recorded during argument lowering.
// Use this info directly.
Offset = FuncInfo.getByValArgumentFrameIndex(Arg);
if (Offset)
Reg = TRI.getFrameRegister(*FuncInfo.MF);
}
}
if (!Reg)
Reg = getRegForValue(Address);
if (Reg)
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
TII.get(TargetOpcode::DBG_VALUE))
.addReg(Reg, RegState::Debug).addImm(Offset)
.addMetadata(DI->getVariable());
return true;
}
case Intrinsic::dbg_value: {
// This form of DBG_VALUE is target-independent.
const DbgValueInst *DI = cast<DbgValueInst>(I);
const TargetInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
const Value *V = DI->getValue();
if (!V) {
// Currently the optimizer can produce this; insert an undef to
// help debugging. Probably the optimizer should not do this.
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
.addReg(0U).addImm(DI->getOffset())
.addMetadata(DI->getVariable());
} else if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
.addImm(CI->getZExtValue()).addImm(DI->getOffset())
.addMetadata(DI->getVariable());
} else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
.addFPImm(CF).addImm(DI->getOffset())
.addMetadata(DI->getVariable());
} else if (unsigned Reg = lookUpRegForValue(V)) {
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
.addReg(Reg, RegState::Debug).addImm(DI->getOffset())
.addMetadata(DI->getVariable());
} else {
// We can't yet handle anything else here because it would require
// generating code, thus altering codegen because of debug info.
DEBUG(dbgs() << "Dropping debug info for " << DI);
}
return true;
}
case Intrinsic::eh_exception: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EXCEPTIONADDR, VT)) {
default: break;
case TargetLowering::Expand: {
assert(FuncInfo.MBB->isLandingPad() &&
"Call to eh.exception not in landing pad!");
unsigned Reg = TLI.getExceptionAddressRegister();
const TargetRegisterClass *RC = TLI.getRegClassFor(VT);
unsigned ResultReg = createResultReg(RC);
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
ResultReg).addReg(Reg);
UpdateValueMap(I, ResultReg);
return true;
}
}
break;
}
case Intrinsic::eh_selector: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EHSELECTION, VT)) {
default: break;
case TargetLowering::Expand: {
if (FuncInfo.MBB->isLandingPad())
AddCatchInfo(*cast<CallInst>(I), &FuncInfo.MF->getMMI(), FuncInfo.MBB);
else {
#ifndef NDEBUG
FuncInfo.CatchInfoLost.insert(cast<CallInst>(I));
#endif
// FIXME: Mark exception selector register as live in. Hack for PR1508.
unsigned Reg = TLI.getExceptionSelectorRegister();
if (Reg) FuncInfo.MBB->addLiveIn(Reg);
}
unsigned Reg = TLI.getExceptionSelectorRegister();
EVT SrcVT = TLI.getPointerTy();
const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
unsigned ResultReg = createResultReg(RC);
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
ResultReg).addReg(Reg);
bool ResultRegIsKill = hasTrivialKill(I);
// Cast the register to the type of the selector.
if (SrcVT.bitsGT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
ResultReg, ResultRegIsKill);
else if (SrcVT.bitsLT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
ISD::SIGN_EXTEND, ResultReg, ResultRegIsKill);
if (ResultReg == 0)
// Unhandled operand. Halt "fast" selection and bail.
return false;
UpdateValueMap(I, ResultReg);
return true;
}
}
break;
}
}
// An arbitrary call. Bail.
return false;
}
SDValue
X86SelectionDAGInfo::EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
SDValue Chain,
SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256)
return SDValue();
// If not DWORD aligned or size is more than the threshold, call the library.
// The libc version is likely to be faster for these cases. It can use the
// address value and run time information about the CPU.
if ((Align & 3) != 0 ||
!ConstantSize ||
ConstantSize->getZExtValue() >
Subtarget->getMaxInlineSizeThreshold()) {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *V = dyn_cast<ConstantSDNode>(Src);
if (const char *bzeroEntry = V &&
V->isNullValue() ? Subtarget->getBZeroEntry() : nullptr) {
EVT IntPtr = TLI.getPointerTy();
Type *IntPtrTy = getDataLayout()->getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(bzeroEntry, IntPtr), &Args, 0)
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memset.
return SDValue();
}
uint64_t SizeVal = ConstantSize->getZExtValue();
SDValue InFlag;
EVT AVT;
SDValue Count;
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Src);
unsigned BytesLeft = 0;
bool TwoRepStos = false;
if (ValC) {
unsigned ValReg;
uint64_t Val = ValC->getZExtValue() & 255;
// If the value is a constant, then we can potentially use larger sets.
switch (Align & 3) {
case 2: // WORD aligned
AVT = MVT::i16;
ValReg = X86::AX;
Val = (Val << 8) | Val;
break;
case 0: // DWORD aligned
AVT = MVT::i32;
ValReg = X86::EAX;
Val = (Val << 8) | Val;
Val = (Val << 16) | Val;
if (Subtarget->is64Bit() && ((Align & 0x7) == 0)) { // QWORD aligned
AVT = MVT::i64;
ValReg = X86::RAX;
Val = (Val << 32) | Val;
}
break;
default: // Byte aligned
AVT = MVT::i8;
ValReg = X86::AL;
Count = DAG.getIntPtrConstant(SizeVal);
break;
}
if (AVT.bitsGT(MVT::i8)) {
unsigned UBytes = AVT.getSizeInBits() / 8;
Count = DAG.getIntPtrConstant(SizeVal / UBytes);
BytesLeft = SizeVal % UBytes;
}
Chain = DAG.getCopyToReg(Chain, dl, ValReg, DAG.getConstant(Val, AVT),
InFlag);
InFlag = Chain.getValue(1);
} else {
AVT = MVT::i8;
Count = DAG.getIntPtrConstant(SizeVal);
Chain = DAG.getCopyToReg(Chain, dl, X86::AL, Src, InFlag);
InFlag = Chain.getValue(1);
}
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RCX :
X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RDI :
X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
if (TwoRepStos) {
InFlag = Chain.getValue(1);
Count = Size;
EVT CVT = Count.getValueType();
SDValue Left = DAG.getNode(ISD::AND, dl, CVT, Count,
DAG.getConstant((AVT == MVT::i64) ? 7 : 3, CVT));
Chain = DAG.getCopyToReg(Chain, dl, (CVT == MVT::i64) ? X86::RCX :
X86::ECX,
Left, InFlag);
InFlag = Chain.getValue(1);
Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(MVT::i8), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
} else if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT AddrVT = Dst.getValueType();
EVT SizeVT = Size.getValueType();
Chain = DAG.getMemset(Chain, dl,
DAG.getNode(ISD::ADD, dl, AddrVT, Dst,
DAG.getConstant(Offset, AddrVT)),
Src,
DAG.getConstant(BytesLeft, SizeVT),
Align, isVolatile, DstPtrInfo.getWithOffset(Offset));
}
// TODO: Use a Tokenfactor, as in memcpy, instead of a single chain.
return Chain;
}
bool FastISel::SelectCall(User *I) {
Function *F = cast<CallInst>(I)->getCalledFunction();
if (!F) return false;
unsigned IID = F->getIntrinsicID();
switch (IID) {
default: break;
case Intrinsic::dbg_stoppoint:
case Intrinsic::dbg_region_start:
case Intrinsic::dbg_region_end:
case Intrinsic::dbg_func_start:
// FIXME - Remove this instructions once the dust settles.
return true;
case Intrinsic::dbg_declare: {
DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
if (!isValidDebugInfoIntrinsic(*DI, CodeGenOpt::None) || !DW
|| !DW->ShouldEmitDwarfDebug())
return true;
Value *Address = DI->getAddress();
if (BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
Address = BCI->getOperand(0);
AllocaInst *AI = dyn_cast<AllocaInst>(Address);
// Don't handle byval struct arguments or VLAs, for example.
if (!AI) break;
DenseMap<const AllocaInst*, int>::iterator SI =
StaticAllocaMap.find(AI);
if (SI == StaticAllocaMap.end()) break; // VLAs.
int FI = SI->second;
if (MMI) {
MetadataContext &TheMetadata =
DI->getParent()->getContext().getMetadata();
unsigned MDDbgKind = TheMetadata.getMDKind("dbg");
MDNode *Dbg = TheMetadata.getMD(MDDbgKind, DI);
MMI->setVariableDbgInfo(DI->getVariable(), FI, Dbg);
}
return true;
}
case Intrinsic::eh_exception: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EXCEPTIONADDR, VT)) {
default: break;
case TargetLowering::Expand: {
assert(MBB->isLandingPad() && "Call to eh.exception not in landing pad!");
unsigned Reg = TLI.getExceptionAddressRegister();
const TargetRegisterClass *RC = TLI.getRegClassFor(VT);
unsigned ResultReg = createResultReg(RC);
bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg,
Reg, RC, RC);
assert(InsertedCopy && "Can't copy address registers!");
InsertedCopy = InsertedCopy;
UpdateValueMap(I, ResultReg);
return true;
}
}
break;
}
case Intrinsic::eh_selector: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EHSELECTION, VT)) {
default: break;
case TargetLowering::Expand: {
if (MMI) {
if (MBB->isLandingPad())
AddCatchInfo(*cast<CallInst>(I), MMI, MBB);
else {
#ifndef NDEBUG
CatchInfoLost.insert(cast<CallInst>(I));
#endif
// FIXME: Mark exception selector register as live in. Hack for PR1508.
unsigned Reg = TLI.getExceptionSelectorRegister();
if (Reg) MBB->addLiveIn(Reg);
}
unsigned Reg = TLI.getExceptionSelectorRegister();
EVT SrcVT = TLI.getPointerTy();
const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
unsigned ResultReg = createResultReg(RC);
bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg, Reg,
RC, RC);
assert(InsertedCopy && "Can't copy address registers!");
InsertedCopy = InsertedCopy;
// Cast the register to the type of the selector.
if (SrcVT.bitsGT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
ResultReg);
else if (SrcVT.bitsLT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
ISD::SIGN_EXTEND, ResultReg);
if (ResultReg == 0)
// Unhandled operand. Halt "fast" selection and bail.
return false;
UpdateValueMap(I, ResultReg);
} else {
unsigned ResultReg =
getRegForValue(Constant::getNullValue(I->getType()));
UpdateValueMap(I, ResultReg);
}
return true;
}
}
break;
}
}
return false;
}
bool FastISel::SelectCall(const User *I) {
const Function *F = cast<CallInst>(I)->getCalledFunction();
if (!F) return false;
// Handle selected intrinsic function calls.
unsigned IID = F->getIntrinsicID();
switch (IID) {
default: break;
case Intrinsic::dbg_declare: {
const DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
if (!DIDescriptor::ValidDebugInfo(DI->getVariable(), CodeGenOpt::None) ||
!MF.getMMI().hasDebugInfo())
return true;
const Value *Address = DI->getAddress();
if (!Address)
return true;
if (isa<UndefValue>(Address))
return true;
const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
// Don't handle byval struct arguments or VLAs, for example.
if (!AI) break;
DenseMap<const AllocaInst*, int>::iterator SI =
StaticAllocaMap.find(AI);
if (SI == StaticAllocaMap.end()) break; // VLAs.
int FI = SI->second;
if (!DI->getDebugLoc().isUnknown())
MF.getMMI().setVariableDbgInfo(DI->getVariable(), FI, DI->getDebugLoc());
// Building the map above is target independent. Generating DBG_VALUE
// inline is target dependent; do this now.
(void)TargetSelectInstruction(cast<Instruction>(I));
return true;
}
case Intrinsic::dbg_value: {
// This form of DBG_VALUE is target-independent.
const DbgValueInst *DI = cast<DbgValueInst>(I);
const TargetInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
const Value *V = DI->getValue();
if (!V) {
// Currently the optimizer can produce this; insert an undef to
// help debugging. Probably the optimizer should not do this.
BuildMI(MBB, DL, II).addReg(0U).addImm(DI->getOffset()).
addMetadata(DI->getVariable());
} else if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
BuildMI(MBB, DL, II).addImm(CI->getZExtValue()).addImm(DI->getOffset()).
addMetadata(DI->getVariable());
} else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
BuildMI(MBB, DL, II).addFPImm(CF).addImm(DI->getOffset()).
addMetadata(DI->getVariable());
} else if (unsigned Reg = lookUpRegForValue(V)) {
BuildMI(MBB, DL, II).addReg(Reg, RegState::Debug).addImm(DI->getOffset()).
addMetadata(DI->getVariable());
} else {
// We can't yet handle anything else here because it would require
// generating code, thus altering codegen because of debug info.
// Insert an undef so we can see what we dropped.
BuildMI(MBB, DL, II).addReg(0U).addImm(DI->getOffset()).
addMetadata(DI->getVariable());
}
return true;
}
case Intrinsic::eh_exception: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EXCEPTIONADDR, VT)) {
default: break;
case TargetLowering::Expand: {
assert(MBB->isLandingPad() && "Call to eh.exception not in landing pad!");
unsigned Reg = TLI.getExceptionAddressRegister();
const TargetRegisterClass *RC = TLI.getRegClassFor(VT);
unsigned ResultReg = createResultReg(RC);
bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg,
Reg, RC, RC);
assert(InsertedCopy && "Can't copy address registers!");
InsertedCopy = InsertedCopy;
UpdateValueMap(I, ResultReg);
return true;
}
}
break;
}
case Intrinsic::eh_selector: {
EVT VT = TLI.getValueType(I->getType());
switch (TLI.getOperationAction(ISD::EHSELECTION, VT)) {
default: break;
case TargetLowering::Expand: {
if (MBB->isLandingPad())
AddCatchInfo(*cast<CallInst>(I), &MF.getMMI(), MBB);
else {
#ifndef NDEBUG
CatchInfoLost.insert(cast<CallInst>(I));
#endif
// FIXME: Mark exception selector register as live in. Hack for PR1508.
unsigned Reg = TLI.getExceptionSelectorRegister();
if (Reg) MBB->addLiveIn(Reg);
}
unsigned Reg = TLI.getExceptionSelectorRegister();
EVT SrcVT = TLI.getPointerTy();
const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
unsigned ResultReg = createResultReg(RC);
bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg, Reg,
RC, RC);
assert(InsertedCopy && "Can't copy address registers!");
InsertedCopy = InsertedCopy;
// Cast the register to the type of the selector.
if (SrcVT.bitsGT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
ResultReg);
else if (SrcVT.bitsLT(MVT::i32))
ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
ISD::SIGN_EXTEND, ResultReg);
if (ResultReg == 0)
// Unhandled operand. Halt "fast" selection and bail.
return false;
UpdateValueMap(I, ResultReg);
return true;
}
}
break;
}
}
// An arbitrary call. Bail.
return false;
}
SDValue X86SelectionDAGInfo::EmitTargetCodeForMemset(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Val,
SDValue Size, unsigned Align, bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
const X86Subtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<X86Subtarget>();
#ifndef NDEBUG
// If the base register might conflict with our physical registers, bail out.
const MCPhysReg ClobberSet[] = {X86::RCX, X86::RAX, X86::RDI,
X86::ECX, X86::EAX, X86::EDI};
assert(!isBaseRegConflictPossible(DAG, ClobberSet));
#endif
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256)
return SDValue();
// If not DWORD aligned or size is more than the threshold, call the library.
// The libc version is likely to be faster for these cases. It can use the
// address value and run time information about the CPU.
if ((Align & 3) != 0 || !ConstantSize ||
ConstantSize->getZExtValue() > Subtarget.getMaxInlineSizeThreshold()) {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Val);
if (const char *bzeroName = (ValC && ValC->isNullValue())
? DAG.getTargetLoweringInfo().getLibcallName(RTLIB::BZERO)
: nullptr) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
EVT IntPtr = TLI.getPointerTy(DAG.getDataLayout());
Type *IntPtrTy = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setLibCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(bzeroName, IntPtr),
std::move(Args))
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memset.
return SDValue();
}
uint64_t SizeVal = ConstantSize->getZExtValue();
SDValue InFlag;
EVT AVT;
SDValue Count;
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Val);
unsigned BytesLeft = 0;
if (ValC) {
unsigned ValReg;
uint64_t Val = ValC->getZExtValue() & 255;
// If the value is a constant, then we can potentially use larger sets.
switch (Align & 3) {
case 2: // WORD aligned
AVT = MVT::i16;
ValReg = X86::AX;
Val = (Val << 8) | Val;
break;
case 0: // DWORD aligned
AVT = MVT::i32;
ValReg = X86::EAX;
Val = (Val << 8) | Val;
Val = (Val << 16) | Val;
if (Subtarget.is64Bit() && ((Align & 0x7) == 0)) { // QWORD aligned
AVT = MVT::i64;
ValReg = X86::RAX;
Val = (Val << 32) | Val;
}
break;
default: // Byte aligned
AVT = MVT::i8;
ValReg = X86::AL;
Count = DAG.getIntPtrConstant(SizeVal, dl);
break;
}
if (AVT.bitsGT(MVT::i8)) {
unsigned UBytes = AVT.getSizeInBits() / 8;
Count = DAG.getIntPtrConstant(SizeVal / UBytes, dl);
BytesLeft = SizeVal % UBytes;
}
Chain = DAG.getCopyToReg(Chain, dl, ValReg, DAG.getConstant(Val, dl, AVT),
InFlag);
InFlag = Chain.getValue(1);
} else {
AVT = MVT::i8;
Count = DAG.getIntPtrConstant(SizeVal, dl);
Chain = DAG.getCopyToReg(Chain, dl, X86::AL, Val, InFlag);
InFlag = Chain.getValue(1);
}
bool Use64BitRegs = Subtarget.isTarget64BitLP64();
Chain = DAG.getCopyToReg(Chain, dl, Use64BitRegs ? X86::RCX : X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Use64BitRegs ? X86::RDI : X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT AddrVT = Dst.getValueType();
EVT SizeVT = Size.getValueType();
Chain = DAG.getMemset(Chain, dl,
DAG.getNode(ISD::ADD, dl, AddrVT, Dst,
DAG.getConstant(Offset, dl, AddrVT)),
Val,
DAG.getConstant(BytesLeft, dl, SizeVT),
Align, isVolatile, false,
DstPtrInfo.getWithOffset(Offset));
}
// TODO: Use a Tokenfactor, as in memcpy, instead of a single chain.
return Chain;
}