#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include <cctype>
using namespace llvm;
-/// We are in the process of implementing a new TypeLegalization action
-/// - the promotion of vector elements. This feature is disabled by default
-/// and only enabled using this flag.
-static cl::opt<bool>
-AllowPromoteIntElem("promote-elements", cl::Hidden,
- cl::desc("Allow promotion of integer vector element types"));
-
-namespace llvm {
-TLSModel::Model getTLSModel(const GlobalValue *GV, Reloc::Model reloc) {
- bool isLocal = GV->hasLocalLinkage();
- bool isDeclaration = GV->isDeclaration();
- // FIXME: what should we do for protected and internal visibility?
- // For variables, is internal different from hidden?
- bool isHidden = GV->hasHiddenVisibility();
-
- if (reloc == Reloc::PIC_) {
- if (isLocal || isHidden)
- return TLSModel::LocalDynamic;
- else
- return TLSModel::GeneralDynamic;
- } else {
- if (!isDeclaration || isHidden)
- return TLSModel::LocalExec;
- else
- return TLSModel::InitialExec;
- }
-}
-}
-
/// InitLibcallNames - Set default libcall names.
///
static void InitLibcallNames(const char **Names) {
Names[RTLIB::MUL_I32] = "__mulsi3";
Names[RTLIB::MUL_I64] = "__muldi3";
Names[RTLIB::MUL_I128] = "__multi3";
+ Names[RTLIB::MULO_I32] = "__mulosi4";
+ Names[RTLIB::MULO_I64] = "__mulodi4";
+ Names[RTLIB::MULO_I128] = "__muloti4";
Names[RTLIB::SDIV_I8] = "__divqi3";
Names[RTLIB::SDIV_I16] = "__divhi3";
Names[RTLIB::SDIV_I32] = "__divsi3";
Names[RTLIB::REM_F64] = "fmod";
Names[RTLIB::REM_F80] = "fmodl";
Names[RTLIB::REM_PPCF128] = "fmodl";
+ Names[RTLIB::FMA_F32] = "fmaf";
+ Names[RTLIB::FMA_F64] = "fma";
+ Names[RTLIB::FMA_F80] = "fmal";
+ Names[RTLIB::FMA_PPCF128] = "fmal";
Names[RTLIB::POWI_F32] = "__powisf2";
Names[RTLIB::POWI_F64] = "__powidf2";
Names[RTLIB::POWI_F80] = "__powixf2";
Names[RTLIB::SYNC_FETCH_AND_OR_8] = "__sync_fetch_and_or_8";
Names[RTLIB::SYNC_FETCH_AND_XOR_1] = "__sync_fetch_and_xor_1";
Names[RTLIB::SYNC_FETCH_AND_XOR_2] = "__sync_fetch_and_xor_2";
- Names[RTLIB::SYNC_FETCH_AND_XOR_4] = "__sync_fetch_and-xor_4";
+ Names[RTLIB::SYNC_FETCH_AND_XOR_4] = "__sync_fetch_and_xor_4";
Names[RTLIB::SYNC_FETCH_AND_XOR_8] = "__sync_fetch_and_xor_8";
Names[RTLIB::SYNC_FETCH_AND_NAND_1] = "__sync_fetch_and_nand_1";
Names[RTLIB::SYNC_FETCH_AND_NAND_2] = "__sync_fetch_and_nand_2";
/// NOTE: The constructor takes ownership of TLOF.
TargetLowering::TargetLowering(const TargetMachine &tm,
const TargetLoweringObjectFile *tlof)
- : TM(tm), TD(TM.getTargetData()), TLOF(*tlof),
- mayPromoteElements(AllowPromoteIntElem) {
+ : TM(tm), TD(TM.getTargetData()), TLOF(*tlof) {
// All operations default to being supported.
memset(OpActions, 0, sizeof(OpActions));
memset(LoadExtActions, 0, sizeof(LoadExtActions));
// ConstantFP nodes default to expand. Targets can either change this to
// Legal, in which case all fp constants are legal, or use isFPImmLegal()
// to optimize expansions for certain constants.
+ setOperationAction(ISD::ConstantFP, MVT::f16, Expand);
setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
setOperationAction(ISD::ConstantFP, MVT::f80, Expand);
// These library functions default to expand.
- setOperationAction(ISD::FLOG , MVT::f64, Expand);
- setOperationAction(ISD::FLOG2, MVT::f64, Expand);
- setOperationAction(ISD::FLOG10,MVT::f64, Expand);
- setOperationAction(ISD::FEXP , MVT::f64, Expand);
- setOperationAction(ISD::FEXP2, MVT::f64, Expand);
- setOperationAction(ISD::FLOG , MVT::f32, Expand);
- setOperationAction(ISD::FLOG2, MVT::f32, Expand);
- setOperationAction(ISD::FLOG10,MVT::f32, Expand);
- setOperationAction(ISD::FEXP , MVT::f32, Expand);
- setOperationAction(ISD::FEXP2, MVT::f32, Expand);
+ setOperationAction(ISD::FLOG , MVT::f16, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f16, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f16, Expand);
+ setOperationAction(ISD::FEXP , MVT::f16, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f16, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f16, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f16, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f16, Expand);
+ setOperationAction(ISD::FRINT, MVT::f16, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f16, Expand);
+ setOperationAction(ISD::FLOG , MVT::f32, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f32, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f32, Expand);
+ setOperationAction(ISD::FEXP , MVT::f32, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f32, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f32, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f32, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f32, Expand);
+ setOperationAction(ISD::FRINT, MVT::f32, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f32, Expand);
+ setOperationAction(ISD::FLOG , MVT::f64, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f64, Expand);
+ setOperationAction(ISD::FEXP , MVT::f64, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f64, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f64, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f64, Expand);
+ setOperationAction(ISD::FRINT, MVT::f64, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f64, Expand);
// Default ISD::TRAP to expand (which turns it into abort).
setOperationAction(ISD::TRAP, MVT::Other, Expand);
IntDivIsCheap = false;
Pow2DivIsCheap = false;
JumpIsExpensive = false;
+ predictableSelectIsExpensive = false;
StackPointerRegisterToSaveRestore = 0;
ExceptionPointerRegister = 0;
ExceptionSelectorRegister = 0;
BooleanContents = UndefinedBooleanContent;
- SchedPreferenceInfo = Sched::Latency;
+ BooleanVectorContents = UndefinedBooleanContent;
+ SchedPreferenceInfo = Sched::ILP;
JumpBufSize = 0;
JumpBufAlignment = 0;
MinFunctionAlignment = 0;
PrefLoopAlignment = 0;
MinStackArgumentAlignment = 1;
ShouldFoldAtomicFences = false;
+ InsertFencesForAtomic = false;
+ SupportJumpTables = true;
+ MinimumJumpTableEntries = 4;
InitLibcallNames(LibcallRoutineNames);
InitCmpLibcallCCs(CmpLibcallCCs);
NewVT = EltTy;
IntermediateVT = NewVT;
+ unsigned NewVTSize = NewVT.getSizeInBits();
+
+ // Convert sizes such as i33 to i64.
+ if (!isPowerOf2_32(NewVTSize))
+ NewVTSize = NextPowerOf2(NewVTSize);
+
EVT DestVT = TLI->getRegisterType(NewVT);
RegisterVT = DestVT;
if (EVT(DestVT).bitsLT(NewVT)) // Value is expanded, e.g. i64 -> i16.
- return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits());
+ return NumVectorRegs*(NewVTSize/DestVT.getSizeInBits());
// Otherwise, promotion or legal types use the same number of registers as
// the vector decimated to the appropriate level.
return false;
}
-/// hasLegalSuperRegRegClasses - Return true if the specified register class
-/// has one or more super-reg register classes that are legal.
-bool
-TargetLowering::hasLegalSuperRegRegClasses(const TargetRegisterClass *RC) const{
- if (*RC->superregclasses_begin() == 0)
- return false;
- for (TargetRegisterInfo::regclass_iterator I = RC->superregclasses_begin(),
- E = RC->superregclasses_end(); I != E; ++I) {
- const TargetRegisterClass *RRC = *I;
- if (isLegalRC(RRC))
- return true;
- }
- return false;
-}
-
/// findRepresentativeClass - Return the largest legal super-reg register class
/// of the register class for the specified type and its associated "cost".
std::pair<const TargetRegisterClass*, uint8_t>
TargetLowering::findRepresentativeClass(EVT VT) const {
+ const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
const TargetRegisterClass *RC = RegClassForVT[VT.getSimpleVT().SimpleTy];
if (!RC)
return std::make_pair(RC, 0);
+
+ // Compute the set of all super-register classes.
+ BitVector SuperRegRC(TRI->getNumRegClasses());
+ for (SuperRegClassIterator RCI(RC, TRI); RCI.isValid(); ++RCI)
+ SuperRegRC.setBitsInMask(RCI.getMask());
+
+ // Find the first legal register class with the largest spill size.
const TargetRegisterClass *BestRC = RC;
- for (TargetRegisterInfo::regclass_iterator I = RC->superregclasses_begin(),
- E = RC->superregclasses_end(); I != E; ++I) {
- const TargetRegisterClass *RRC = *I;
- if (RRC->isASubClass() || !isLegalRC(RRC))
+ for (int i = SuperRegRC.find_first(); i >= 0; i = SuperRegRC.find_next(i)) {
+ const TargetRegisterClass *SuperRC = TRI->getRegClass(i);
+ // We want the largest possible spill size.
+ if (SuperRC->getSize() <= BestRC->getSize())
+ continue;
+ if (!isLegalRC(SuperRC))
continue;
- if (!hasLegalSuperRegRegClasses(RRC))
- return std::make_pair(RRC, 1);
- BestRC = RRC;
+ BestRC = SuperRC;
}
return std::make_pair(BestRC, 1);
}
-
/// computeRegisterProperties - Once all of the register classes are added,
/// this allows us to compute derived properties we expose.
void TargetLowering::computeRegisterProperties() {
LegalIntReg = IntReg;
} else {
RegisterTypeForVT[IntReg] = TransformToType[IntReg] =
- (MVT::SimpleValueType)LegalIntReg;
+ (const MVT::SimpleValueType)LegalIntReg;
ValueTypeActions.setTypeAction(IVT, TypePromoteInteger);
}
}
unsigned NElts = VT.getVectorNumElements();
if (NElts != 1) {
bool IsLegalWiderType = false;
+ // First try to promote the elements of integer vectors. If no legal
+ // promotion was found, fallback to the widen-vector method.
for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
EVT SVT = (MVT::SimpleValueType)nVT;
-
- // If we allow the promotion of vector elements using a flag,
- // then return TypePromoteInteger on vector elements.
- if (mayPromoteElements) {
- // Promote vectors of integers to vectors with the same number
- // of elements, with a wider element type.
- if (SVT.getVectorElementType().getSizeInBits() > EltVT.getSizeInBits()
- && SVT.getVectorNumElements() == NElts &&
- isTypeLegal(SVT) && SVT.getScalarType().isInteger()) {
- TransformToType[i] = SVT;
- RegisterTypeForVT[i] = SVT;
- NumRegistersForVT[i] = 1;
- ValueTypeActions.setTypeAction(VT, TypePromoteInteger);
- IsLegalWiderType = true;
- break;
- }
+ // Promote vectors of integers to vectors with the same number
+ // of elements, with a wider element type.
+ if (SVT.getVectorElementType().getSizeInBits() > EltVT.getSizeInBits()
+ && SVT.getVectorNumElements() == NElts &&
+ isTypeLegal(SVT) && SVT.getScalarType().isInteger()) {
+ TransformToType[i] = SVT;
+ RegisterTypeForVT[i] = SVT;
+ NumRegistersForVT[i] = 1;
+ ValueTypeActions.setTypeAction(VT, TypePromoteInteger);
+ IsLegalWiderType = true;
+ break;
}
+ }
+
+ if (IsLegalWiderType) continue;
+ // Try to widen the vector.
+ for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
+ EVT SVT = (MVT::SimpleValueType)nVT;
if (SVT.getVectorElementType() == EltVT &&
SVT.getVectorNumElements() > NElts &&
isTypeLegal(SVT)) {
return NULL;
}
-
-MVT::SimpleValueType TargetLowering::getSetCCResultType(EVT VT) const {
+EVT TargetLowering::getSetCCResultType(EVT VT) const {
+ assert(!VT.isVector() && "No default SetCC type for vectors!");
return PointerTy.SimpleTy;
}
unsigned NumElts = VT.getVectorNumElements();
// If there is a wider vector type with the same element type as this one,
- // we should widen to that legal vector type. This handles things like
- // <2 x float> -> <4 x float>.
- if (NumElts != 1 && getTypeAction(Context, VT) == TypeWidenVector) {
+ // or a promoted vector type that has the same number of elements which
+ // are wider, then we should convert to that legal vector type.
+ // This handles things like <2 x float> -> <4 x float> and
+ // <4 x i1> -> <4 x i32>.
+ LegalizeTypeAction TA = getTypeAction(Context, VT);
+ if (NumElts != 1 && (TA == TypeWidenVector || TA == TypePromoteInteger)) {
RegisterVT = getTypeToTransformTo(Context, VT);
if (isTypeLegal(RegisterVT)) {
IntermediateVT = RegisterVT;
EVT DestVT = getRegisterType(Context, NewVT);
RegisterVT = DestVT;
+ unsigned NewVTSize = NewVT.getSizeInBits();
+
+ // Convert sizes such as i33 to i64.
+ if (!isPowerOf2_32(NewVTSize))
+ NewVTSize = NextPowerOf2(NewVTSize);
+
if (DestVT.bitsLT(NewVT)) // Value is expanded, e.g. i64 -> i16.
- return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits());
+ return NumVectorRegs*(NewVTSize/DestVT.getSizeInBits());
// Otherwise, promotion or legal types use the same number of registers as
// the vector decimated to the appropriate level.
/// type of the given function. This does not require a DAG or a return value,
/// and is suitable for use before any DAGs for the function are constructed.
/// TODO: Move this out of TargetLowering.cpp.
-void llvm::GetReturnInfo(const Type* ReturnType, Attributes attr,
+void llvm::GetReturnInfo(Type* ReturnType, Attributes attr,
SmallVectorImpl<ISD::OutputArg> &Outs,
- const TargetLowering &TLI,
- SmallVectorImpl<uint64_t> *Offsets) {
+ const TargetLowering &TLI) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, ReturnType, ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) return;
- unsigned Offset = 0;
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
- if (attr & Attribute::SExt)
+ if (attr.hasSExtAttr())
ExtendKind = ISD::SIGN_EXTEND;
- else if (attr & Attribute::ZExt)
+ else if (attr.hasZExtAttr())
ExtendKind = ISD::ZERO_EXTEND;
// FIXME: C calling convention requires the return type to be promoted to
unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
- unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
- PartVT.getTypeForEVT(ReturnType->getContext()));
// 'inreg' on function refers to return value
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
- if (attr & Attribute::InReg)
+ if (attr.hasInRegAttr())
Flags.setInReg();
// Propagate extension type if any
- if (attr & Attribute::SExt)
+ if (attr.hasSExtAttr())
Flags.setSExt();
- else if (attr & Attribute::ZExt)
+ else if (attr.hasZExtAttr())
Flags.setZExt();
- for (unsigned i = 0; i < NumParts; ++i) {
+ for (unsigned i = 0; i < NumParts; ++i)
Outs.push_back(ISD::OutputArg(Flags, PartVT, /*isFixed=*/true));
- if (Offsets) {
- Offsets->push_back(Offset);
- Offset += PartSize;
- }
- }
}
}
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area. This is the actual
/// alignment, not its logarithm.
-unsigned TargetLowering::getByValTypeAlignment(const Type *Ty) const {
+unsigned TargetLowering::getByValTypeAlignment(Type *Ty) const {
return TD->getCallFrameTypeAlignment(Ty);
}
SDValue TargetLowering::getPICJumpTableRelocBase(SDValue Table,
SelectionDAG &DAG) const {
// If our PIC model is GP relative, use the global offset table as the base.
- if (getJumpTableEncoding() == MachineJumpTableInfo::EK_GPRel32BlockAddress)
+ unsigned JTEncoding = getJumpTableEncoding();
+
+ if ((JTEncoding == MachineJumpTableInfo::EK_GPRel64BlockAddress) ||
+ (JTEncoding == MachineJumpTableInfo::EK_GPRel32BlockAddress))
return DAG.getGLOBAL_OFFSET_TABLE(getPointerTy());
+
return Table;
}
if (Depth != 0) {
// If not at the root, Just compute the KnownZero/KnownOne bits to
// simplify things downstream.
- TLO.DAG.ComputeMaskedBits(Op, DemandedMask, KnownZero, KnownOne, Depth);
+ TLO.DAG.ComputeMaskedBits(Op, KnownZero, KnownOne, Depth);
return false;
}
// If this is the root being simplified, allow it to have multiple uses,
switch (Op.getOpcode()) {
case ISD::Constant:
// We know all of the bits for a constant!
- KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue() & NewMask;
- KnownZero = ~KnownOne & NewMask;
+ KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue();
+ KnownZero = ~KnownOne;
return false; // Don't fall through, will infinitely loop.
case ISD::AND:
// If the RHS is a constant, check to see if the LHS would be zero without
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
APInt LHSZero, LHSOne;
// Do not increment Depth here; that can cause an infinite loop.
- TLO.DAG.ComputeMaskedBits(Op.getOperand(0), NewMask,
- LHSZero, LHSOne, Depth);
+ TLO.DAG.ComputeMaskedBits(Op.getOperand(0), LHSZero, LHSOne, Depth);
// If the LHS already has zeros where RHSC does, this and is dead.
if ((LHSZero & NewMask) == (~RHSC->getAPIntValue() & NewMask))
return TLO.CombineTo(Op, Op.getOperand(0));
// If all of the unknown bits are known to be zero on one side or the other
// (but not both) turn this into an *inclusive* or.
- // e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0
+ // e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) if C1&C2 == 0
if ((NewMask & ~KnownZero & ~KnownZero2) == 0)
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::OR, dl, Op.getValueType(),
Op.getOperand(0),
// If all of the demanded bits on one side are known, and all of the set
// bits on that side are also known to be set on the other side, turn this
// into an AND, as we know the bits will be cleared.
- // e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
- if ((NewMask & (KnownZero|KnownOne)) == NewMask) { // all known
- if ((KnownOne & KnownOne2) == KnownOne) {
+ // e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 if (C1&C2) == C2
+ // NB: it is okay if more bits are known than are requested
+ if ((NewMask & (KnownZero|KnownOne)) == NewMask) { // all known on one side
+ if (KnownOne == KnownOne2) { // set bits are the same on both sides
EVT VT = Op.getValueType();
SDValue ANDC = TLO.DAG.getConstant(~KnownOne & NewMask, VT);
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::AND, dl, VT,
if (InOp.getNode()->getOpcode() == ISD::ANY_EXTEND) {
SDValue InnerOp = InOp.getNode()->getOperand(0);
EVT InnerVT = InnerOp.getValueType();
- if ((APInt::getHighBitsSet(BitWidth,
- BitWidth - InnerVT.getSizeInBits()) &
- DemandedMask) == 0 &&
+ unsigned InnerBits = InnerVT.getSizeInBits();
+ if (ShAmt < InnerBits && NewMask.lshr(InnerBits) == 0 &&
isTypeDesirableForOp(ISD::SHL, InnerVT)) {
EVT ShTy = getShiftAmountTy(InnerVT);
if (!APInt(BitWidth, ShAmt).isIntN(ShTy.getSizeInBits()))
// always convert this into a logical shr, even if the shift amount is
// variable. The low bit of the shift cannot be an input sign bit unless
// the shift amount is >= the size of the datatype, which is undefined.
- if (DemandedMask == 1)
+ if (NewMask == 1)
return TLO.CombineTo(Op,
TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
Op.getOperand(0), Op.getOperand(1)));
}
break;
case ISD::SIGN_EXTEND_INREG: {
- EVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+ EVT ExVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+
+ APInt MsbMask = APInt::getHighBitsSet(BitWidth, 1);
+ // If we only care about the highest bit, don't bother shifting right.
+ if (MsbMask == DemandedMask) {
+ unsigned ShAmt = ExVT.getScalarType().getSizeInBits();
+ SDValue InOp = Op.getOperand(0);
+
+ // Compute the correct shift amount type, which must be getShiftAmountTy
+ // for scalar types after legalization.
+ EVT ShiftAmtTy = Op.getValueType();
+ if (TLO.LegalTypes() && !ShiftAmtTy.isVector())
+ ShiftAmtTy = getShiftAmountTy(ShiftAmtTy);
+
+ SDValue ShiftAmt = TLO.DAG.getConstant(BitWidth - ShAmt, ShiftAmtTy);
+ return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SHL, dl,
+ Op.getValueType(), InOp, ShiftAmt));
+ }
// Sign extension. Compute the demanded bits in the result that are not
// present in the input.
APInt NewBits =
APInt::getHighBitsSet(BitWidth,
- BitWidth - EVT.getScalarType().getSizeInBits());
+ BitWidth - ExVT.getScalarType().getSizeInBits());
// If none of the extended bits are demanded, eliminate the sextinreg.
if ((NewBits & NewMask) == 0)
return TLO.CombineTo(Op, Op.getOperand(0));
APInt InSignBit =
- APInt::getSignBit(EVT.getScalarType().getSizeInBits()).zext(BitWidth);
+ APInt::getSignBit(ExVT.getScalarType().getSizeInBits()).zext(BitWidth);
APInt InputDemandedBits =
APInt::getLowBitsSet(BitWidth,
- EVT.getScalarType().getSizeInBits()) &
+ ExVT.getScalarType().getSizeInBits()) &
NewMask;
// Since the sign extended bits are demanded, we know that the sign
// If the input sign bit is known zero, convert this into a zero extension.
if (KnownZero.intersects(InSignBit))
return TLO.CombineTo(Op,
- TLO.DAG.getZeroExtendInReg(Op.getOperand(0),dl,EVT));
+ TLO.DAG.getZeroExtendInReg(Op.getOperand(0),dl,ExVT));
if (KnownOne.intersects(InSignBit)) { // Input sign bit known set
KnownOne |= NewBits;
// If the sign bit is known one, the top bits match.
if (KnownOne.intersects(InSignBit)) {
- KnownOne |= NewBits;
- KnownZero &= ~NewBits;
+ KnownOne |= NewBits;
+ assert((KnownZero & NewBits) == 0);
} else { // Otherwise, top bits aren't known.
- KnownOne &= ~NewBits;
- KnownZero &= ~NewBits;
+ assert((KnownOne & NewBits) == 0);
+ assert((KnownZero & NewBits) == 0);
}
break;
}
break;
}
case ISD::AssertZext: {
- // Demand all the bits of the input that are demanded in the output.
- // The low bits are obvious; the high bits are demanded because we're
- // asserting that they're zero here.
- if (SimplifyDemandedBits(Op.getOperand(0), NewMask,
+ // AssertZext demands all of the high bits, plus any of the low bits
+ // demanded by its users.
+ EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+ APInt InMask = APInt::getLowBitsSet(BitWidth,
+ VT.getSizeInBits());
+ if (SimplifyDemandedBits(Op.getOperand(0), ~InMask | NewMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
- APInt InMask = APInt::getLowBitsSet(BitWidth,
- VT.getSizeInBits());
KnownZero |= ~InMask & NewMask;
break;
}
case ISD::BITCAST:
- // If this is an FP->Int bitcast and if the sign bit is the only thing that
- // is demanded, turn this into a FGETSIGN.
- if (NewMask == APInt::getSignBit(Op.getValueType().getSizeInBits()) &&
- Op.getOperand(0).getValueType().isFloatingPoint() &&
- !Op.getOperand(0).getValueType().isVector()) {
- if (isOperationLegalOrCustom(ISD::FGETSIGN, Op.getValueType())) {
+ // If this is an FP->Int bitcast and if the sign bit is the only
+ // thing demanded, turn this into a FGETSIGN.
+ if (!TLO.LegalOperations() &&
+ !Op.getValueType().isVector() &&
+ !Op.getOperand(0).getValueType().isVector() &&
+ NewMask == APInt::getSignBit(Op.getValueType().getSizeInBits()) &&
+ Op.getOperand(0).getValueType().isFloatingPoint()) {
+ bool OpVTLegal = isOperationLegalOrCustom(ISD::FGETSIGN, Op.getValueType());
+ bool i32Legal = isOperationLegalOrCustom(ISD::FGETSIGN, MVT::i32);
+ if ((OpVTLegal || i32Legal) && Op.getValueType().isSimple()) {
+ EVT Ty = OpVTLegal ? Op.getValueType() : MVT::i32;
// Make a FGETSIGN + SHL to move the sign bit into the appropriate
// place. We expect the SHL to be eliminated by other optimizations.
- SDValue Sign = TLO.DAG.getNode(ISD::FGETSIGN, dl, Op.getValueType(),
- Op.getOperand(0));
+ SDValue Sign = TLO.DAG.getNode(ISD::FGETSIGN, dl, Ty, Op.getOperand(0));
+ unsigned OpVTSizeInBits = Op.getValueType().getSizeInBits();
+ if (!OpVTLegal && OpVTSizeInBits > 32)
+ Sign = TLO.DAG.getNode(ISD::ZERO_EXTEND, dl, Op.getValueType(), Sign);
unsigned ShVal = Op.getValueType().getSizeInBits()-1;
SDValue ShAmt = TLO.DAG.getConstant(ShVal, Op.getValueType());
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SHL, dl,
// FALL THROUGH
default:
// Just use ComputeMaskedBits to compute output bits.
- TLO.DAG.ComputeMaskedBits(Op, NewMask, KnownZero, KnownOne, Depth);
+ TLO.DAG.ComputeMaskedBits(Op, KnownZero, KnownOne, Depth);
break;
}
/// in Mask are known to be either zero or one and return them in the
/// KnownZero/KnownOne bitsets.
void TargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
- const APInt &Mask,
APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
Op.getOpcode() == ISD::INTRINSIC_VOID) &&
"Should use MaskedValueIsZero if you don't know whether Op"
" is a target node!");
- KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
+ KnownZero = KnownOne = APInt(KnownOne.getBitWidth(), 0);
}
/// ComputeNumSignBitsForTargetNode - This method can be implemented by
// Fall back to ComputeMaskedBits to catch other known cases.
EVT OpVT = Val.getValueType();
unsigned BitWidth = OpVT.getScalarType().getSizeInBits();
- APInt Mask = APInt::getAllOnesValue(BitWidth);
APInt KnownZero, KnownOne;
- DAG.ComputeMaskedBits(Val, Mask, KnownZero, KnownOne);
+ DAG.ComputeMaskedBits(Val, KnownZero, KnownOne);
return (KnownZero.countPopulation() == BitWidth - 1) &&
(KnownOne.countPopulation() == 1);
}
// comparisons.
if (isa<ConstantSDNode>(N0.getNode()))
return DAG.getSetCC(dl, VT, N1, N0, ISD::getSetCCSwappedOperands(Cond));
-
+
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode())) {
const APInt &C1 = N1C->getAPIntValue();
return DAG.getSetCC(dl, VT, And, DAG.getConstant(0, CTVT), CC);
}
- // TODO: (ctpop x) == 1 -> x && (x & x-1) == 0 iff ctpop is illegal.
+ // TODO: (ctpop x) == 1 -> x && (x & x-1) == 0 if ctpop is illegal.
}
// (zext x) == C --> x == (trunc C)
}
}
- // Make sure we're not loosing bits from the constant.
+ // Make sure we're not losing bits from the constant.
if (MinBits < C1.getBitWidth() && MinBits > C1.getActiveBits()) {
EVT MinVT = EVT::getIntegerVT(*DAG.getContext(), MinBits);
if (isTypeDesirableForOp(ISD::SETCC, MinVT)) {
unsigned NewAlign = MinAlign(Lod->getAlignment(), bestOffset);
SDValue NewLoad = DAG.getLoad(newVT, dl, Lod->getChain(), Ptr,
Lod->getPointerInfo().getWithOffset(bestOffset),
- false, false, NewAlign);
+ false, false, false, NewAlign);
return DAG.getSetCC(dl, VT,
DAG.getNode(ISD::AND, dl, newVT, NewLoad,
DAG.getConstant(bestMask.trunc(bestWidth),
}
} else if (N1C->getAPIntValue() == 1 &&
(VT == MVT::i1 ||
- getBooleanContents() == ZeroOrOneBooleanContent)) {
+ getBooleanContents(false) == ZeroOrOneBooleanContent)) {
SDValue Op0 = N0;
if (Op0.getOpcode() == ISD::TRUNCATE)
Op0 = Op0.getOperand(0);
N0.getOpcode() == ISD::AND)
if (ConstantSDNode *AndRHS =
dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
- EVT ShiftTy = DCI.isBeforeLegalize() ?
+ EVT ShiftTy = DCI.isBeforeLegalizeOps() ?
getPointerTy() : getShiftAmountTy(N0.getValueType());
if (Cond == ISD::SETNE && C1 == 0) {// (X & 8) != 0 --> (X & 8) >> 3
// Perform the xform if the AND RHS is a single bit.
}
}
}
+
+ if (C1.getMinSignedBits() <= 64 &&
+ !isLegalICmpImmediate(C1.getSExtValue())) {
+ // (X & -256) == 256 -> (X >> 8) == 1
+ if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) &&
+ N0.getOpcode() == ISD::AND && N0.hasOneUse()) {
+ if (ConstantSDNode *AndRHS =
+ dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
+ const APInt &AndRHSC = AndRHS->getAPIntValue();
+ if ((-AndRHSC).isPowerOf2() && (AndRHSC & C1) == C1) {
+ unsigned ShiftBits = AndRHSC.countTrailingZeros();
+ EVT ShiftTy = DCI.isBeforeLegalizeOps() ?
+ getPointerTy() : getShiftAmountTy(N0.getValueType());
+ EVT CmpTy = N0.getValueType();
+ SDValue Shift = DAG.getNode(ISD::SRL, dl, CmpTy, N0.getOperand(0),
+ DAG.getConstant(ShiftBits, ShiftTy));
+ SDValue CmpRHS = DAG.getConstant(C1.lshr(ShiftBits), CmpTy);
+ return DAG.getSetCC(dl, VT, Shift, CmpRHS, Cond);
+ }
+ }
+ } else if (Cond == ISD::SETULT || Cond == ISD::SETUGE ||
+ Cond == ISD::SETULE || Cond == ISD::SETUGT) {
+ bool AdjOne = (Cond == ISD::SETULE || Cond == ISD::SETUGT);
+ // X < 0x100000000 -> (X >> 32) < 1
+ // X >= 0x100000000 -> (X >> 32) >= 1
+ // X <= 0x0ffffffff -> (X >> 32) < 1
+ // X > 0x0ffffffff -> (X >> 32) >= 1
+ unsigned ShiftBits;
+ APInt NewC = C1;
+ ISD::CondCode NewCond = Cond;
+ if (AdjOne) {
+ ShiftBits = C1.countTrailingOnes();
+ NewC = NewC + 1;
+ NewCond = (Cond == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
+ } else {
+ ShiftBits = C1.countTrailingZeros();
+ }
+ NewC = NewC.lshr(ShiftBits);
+ if (ShiftBits && isLegalICmpImmediate(NewC.getSExtValue())) {
+ EVT ShiftTy = DCI.isBeforeLegalizeOps() ?
+ getPointerTy() : getShiftAmountTy(N0.getValueType());
+ EVT CmpTy = N0.getValueType();
+ SDValue Shift = DAG.getNode(ISD::SRL, dl, CmpTy, N0,
+ DAG.getConstant(ShiftBits, ShiftTy));
+ SDValue CmpRHS = DAG.getConstant(NewC, CmpTy);
+ return DAG.getSetCC(dl, VT, Shift, CmpRHS, NewCond);
+ }
+ }
+ }
}
if (isa<ConstantFPSDNode>(N0.getNode())) {
}
if (N0 == N1) {
+ // The sext(setcc()) => setcc() optimization relies on the appropriate
+ // constant being emitted.
+ uint64_t EqVal = 0;
+ switch (getBooleanContents(N0.getValueType().isVector())) {
+ case UndefinedBooleanContent:
+ case ZeroOrOneBooleanContent:
+ EqVal = ISD::isTrueWhenEqual(Cond);
+ break;
+ case ZeroOrNegativeOneBooleanContent:
+ EqVal = ISD::isTrueWhenEqual(Cond) ? -1 : 0;
+ break;
+ }
+
// We can always fold X == X for integer setcc's.
- if (N0.getValueType().isInteger())
- return DAG.getConstant(ISD::isTrueWhenEqual(Cond), VT);
+ if (N0.getValueType().isInteger()) {
+ return DAG.getConstant(EqVal, VT);
+ }
unsigned UOF = ISD::getUnorderedFlavor(Cond);
if (UOF == 2) // FP operators that are undefined on NaNs.
- return DAG.getConstant(ISD::isTrueWhenEqual(Cond), VT);
+ return DAG.getConstant(EqVal, VT);
if (UOF == unsigned(ISD::isTrueWhenEqual(Cond)))
- return DAG.getConstant(UOF, VT);
+ return DAG.getConstant(EqVal, VT);
// Otherwise, we can't fold it. However, we can simplify it to SETUO/SETO
// if it is not already.
ISD::CondCode NewCond = UOF == 0 ? ISD::SETO : ISD::SETUO;
- if (NewCond != Cond)
+ if (NewCond != Cond && (DCI.isBeforeLegalizeOps() ||
+ getCondCodeAction(NewCond, N0.getValueType()) == Legal))
return DAG.getSetCC(dl, VT, N0, N1, NewCond);
}
}
}
+ // If RHS is a legal immediate value for a compare instruction, we need
+ // to be careful about increasing register pressure needlessly.
+ bool LegalRHSImm = false;
+
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(N1)) {
if (ConstantSDNode *LHSR = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
// Turn (X+C1) == C2 --> X == C2-C1
N0.getValueType()), Cond);
}
- // Turn (X^C1) == C2 into X == C1^C2 iff X&~C1 = 0.
+ // Turn (X^C1) == C2 into X == C1^C2 if X&~C1 = 0.
if (N0.getOpcode() == ISD::XOR)
// If we know that all of the inverted bits are zero, don't bother
// performing the inversion.
Cond);
}
}
+
+ // Could RHSC fold directly into a compare?
+ if (RHSC->getValueType(0).getSizeInBits() <= 64)
+ LegalRHSImm = isLegalICmpImmediate(RHSC->getSExtValue());
}
// Simplify (X+Z) == X --> Z == 0
- if (N0.getOperand(0) == N1)
- return DAG.getSetCC(dl, VT, N0.getOperand(1),
- DAG.getConstant(0, N0.getValueType()), Cond);
- if (N0.getOperand(1) == N1) {
- if (DAG.isCommutativeBinOp(N0.getOpcode()))
- return DAG.getSetCC(dl, VT, N0.getOperand(0),
- DAG.getConstant(0, N0.getValueType()), Cond);
- else if (N0.getNode()->hasOneUse()) {
- assert(N0.getOpcode() == ISD::SUB && "Unexpected operation!");
- // (Z-X) == X --> Z == X<<1
- SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(),
- N1,
+ // Don't do this if X is an immediate that can fold into a cmp
+ // instruction and X+Z has other uses. It could be an induction variable
+ // chain, and the transform would increase register pressure.
+ if (!LegalRHSImm || N0.getNode()->hasOneUse()) {
+ if (N0.getOperand(0) == N1)
+ return DAG.getSetCC(dl, VT, N0.getOperand(1),
+ DAG.getConstant(0, N0.getValueType()), Cond);
+ if (N0.getOperand(1) == N1) {
+ if (DAG.isCommutativeBinOp(N0.getOpcode()))
+ return DAG.getSetCC(dl, VT, N0.getOperand(0),
+ DAG.getConstant(0, N0.getValueType()), Cond);
+ else if (N0.getNode()->hasOneUse()) {
+ assert(N0.getOpcode() == ISD::SUB && "Unexpected operation!");
+ // (Z-X) == X --> Z == X<<1
+ SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(), N1,
DAG.getConstant(1, getShiftAmountTy(N1.getValueType())));
- if (!DCI.isCalledByLegalizer())
- DCI.AddToWorklist(SH.getNode());
- return DAG.getSetCC(dl, VT, N0.getOperand(0), SH, Cond);
+ if (!DCI.isCalledByLegalizer())
+ DCI.AddToWorklist(SH.getNode());
+ return DAG.getSetCC(dl, VT, N0.getOperand(0), SH, Cond);
+ }
}
}
}
TargetLowering::ConstraintType
TargetLowering::getConstraintType(const std::string &Constraint) const {
- // FIXME: lots more standard ones to handle.
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default: break;
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector. If it is invalid, don't add anything to Ops.
void TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
- char ConstraintLetter,
+ std::string &Constraint,
std::vector<SDValue> &Ops,
SelectionDAG &DAG) const {
+
+ if (Constraint.length() > 1) return;
+
+ char ConstraintLetter = Constraint[0];
switch (ConstraintLetter) {
default: break;
case 'X': // Allows any operand; labels (basic block) use this.
}
}
-std::vector<unsigned> TargetLowering::
-getRegClassForInlineAsmConstraint(const std::string &Constraint,
- EVT VT) const {
- return std::vector<unsigned>();
-}
-
-
std::pair<unsigned, const TargetRegisterClass*> TargetLowering::
getRegForInlineAsmConstraint(const std::string &Constraint,
EVT VT) const {
// If none of the value types for this register class are valid, we
// can't use it. For example, 64-bit reg classes on 32-bit targets.
- bool isLegal = false;
- for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
- I != E; ++I) {
- if (isTypeLegal(*I)) {
- isLegal = true;
- break;
- }
- }
-
- if (!isLegal) continue;
+ if (!isLegalRC(RC))
+ continue;
for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
I != E; ++I) {
// corresponding argument.
assert(!CS.getType()->isVoidTy() &&
"Bad inline asm!");
- if (const StructType *STy = dyn_cast<StructType>(CS.getType())) {
+ if (StructType *STy = dyn_cast<StructType>(CS.getType())) {
OpInfo.ConstraintVT = getValueType(STy->getElementType(ResNo));
} else {
assert(ResNo == 0 && "Asm only has one result!");
}
if (OpInfo.CallOperandVal) {
- const llvm::Type *OpTy = OpInfo.CallOperandVal->getType();
+ llvm::Type *OpTy = OpInfo.CallOperandVal->getType();
if (OpInfo.isIndirect) {
- const llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
+ llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
if (!PtrTy)
report_fatal_error("Indirect operand for inline asm not a pointer!");
OpTy = PtrTy->getElementType();
}
-
+
// Look for vector wrapped in a struct. e.g. { <16 x i8> }.
- if (const StructType *STy = dyn_cast<StructType>(OpTy))
+ if (StructType *STy = dyn_cast<StructType>(OpTy))
if (STy->getNumElements() == 1)
OpTy = STy->getElementType(0);
AsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+ std::pair<unsigned, const TargetRegisterClass*> MatchRC =
+ getRegForInlineAsmConstraint(OpInfo.ConstraintCode,
+ OpInfo.ConstraintVT);
+ std::pair<unsigned, const TargetRegisterClass*> InputRC =
+ getRegForInlineAsmConstraint(Input.ConstraintCode,
+ Input.ConstraintVT);
if ((OpInfo.ConstraintVT.isInteger() !=
Input.ConstraintVT.isInteger()) ||
- (OpInfo.ConstraintVT.getSizeInBits() !=
- Input.ConstraintVT.getSizeInBits())) {
+ (MatchRC.second != InputRC.second)) {
report_fatal_error("Unsupported asm: input constraint"
" with a matching output constraint of"
" incompatible type!");
/// is.
static unsigned getConstraintGenerality(TargetLowering::ConstraintType CT) {
switch (CT) {
- default: llvm_unreachable("Unknown constraint type!");
case TargetLowering::C_Other:
case TargetLowering::C_Unknown:
return 0;
case TargetLowering::C_Memory:
return 3;
}
+ llvm_unreachable("Invalid constraint type");
}
/// Examine constraint type and operand type and determine a weight value.
assert(OpInfo.Codes[i].size() == 1 &&
"Unhandled multi-letter 'other' constraint");
std::vector<SDValue> ResultOps;
- TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i][0],
+ TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i],
ResultOps, *DAG);
if (!ResultOps.empty()) {
BestType = CType;
/// isLegalAddressingMode - Return true if the addressing mode represented
/// by AM is legal for this target, for a load/store of the specified type.
bool TargetLowering::isLegalAddressingMode(const AddrMode &AM,
- const Type *Ty) const {
+ Type *Ty) const {
// The default implementation of this implements a conservative RISCy, r+r and
// r+i addr mode.
return true;
}
+/// BuildExactDiv - Given an exact SDIV by a constant, create a multiplication
+/// with the multiplicative inverse of the constant.
+SDValue TargetLowering::BuildExactSDIV(SDValue Op1, SDValue Op2, DebugLoc dl,
+ SelectionDAG &DAG) const {
+ ConstantSDNode *C = cast<ConstantSDNode>(Op2);
+ APInt d = C->getAPIntValue();
+ assert(d != 0 && "Division by zero!");
+
+ // Shift the value upfront if it is even, so the LSB is one.
+ unsigned ShAmt = d.countTrailingZeros();
+ if (ShAmt) {
+ // TODO: For UDIV use SRL instead of SRA.
+ SDValue Amt = DAG.getConstant(ShAmt, getShiftAmountTy(Op1.getValueType()));
+ Op1 = DAG.getNode(ISD::SRA, dl, Op1.getValueType(), Op1, Amt);
+ d = d.ashr(ShAmt);
+ }
+
+ // Calculate the multiplicative inverse, using Newton's method.
+ APInt t, xn = d;
+ while ((t = d*xn) != 1)
+ xn *= APInt(d.getBitWidth(), 2) - t;
+
+ Op2 = DAG.getConstant(xn, Op1.getValueType());
+ return DAG.getNode(ISD::MUL, dl, Op1.getValueType(), Op1, Op2);
+}
+
/// BuildSDIVSequence - Given an ISD::SDIV node expressing a divide by constant,
/// return a DAG expression to select that will generate the same value by
/// multiplying by a magic number. See:
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
-SDValue TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG,
- std::vector<SDNode*>* Created) const {
+SDValue TargetLowering::
+BuildSDIV(SDNode *N, SelectionDAG &DAG, bool IsAfterLegalization,
+ std::vector<SDNode*>* Created) const {
EVT VT = N->getValueType(0);
DebugLoc dl= N->getDebugLoc();
// Multiply the numerator (operand 0) by the magic value
// FIXME: We should support doing a MUL in a wider type
SDValue Q;
- if (isOperationLegalOrCustom(ISD::MULHS, VT))
+ if (IsAfterLegalization ? isOperationLegal(ISD::MULHS, VT) :
+ isOperationLegalOrCustom(ISD::MULHS, VT))
Q = DAG.getNode(ISD::MULHS, dl, VT, N->getOperand(0),
DAG.getConstant(magics.m, VT));
- else if (isOperationLegalOrCustom(ISD::SMUL_LOHI, VT))
+ else if (IsAfterLegalization ? isOperationLegal(ISD::SMUL_LOHI, VT) :
+ isOperationLegalOrCustom(ISD::SMUL_LOHI, VT))
Q = SDValue(DAG.getNode(ISD::SMUL_LOHI, dl, DAG.getVTList(VT, VT),
N->getOperand(0),
DAG.getConstant(magics.m, VT)).getNode(), 1);
/// return a DAG expression to select that will generate the same value by
/// multiplying by a magic number. See:
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
-SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG,
- std::vector<SDNode*>* Created) const {
+SDValue TargetLowering::
+BuildUDIV(SDNode *N, SelectionDAG &DAG, bool IsAfterLegalization,
+ std::vector<SDNode*>* Created) const {
EVT VT = N->getValueType(0);
DebugLoc dl = N->getDebugLoc();
// Multiply the numerator (operand 0) by the magic value
// FIXME: We should support doing a MUL in a wider type
- if (isOperationLegalOrCustom(ISD::MULHU, VT))
+ if (IsAfterLegalization ? isOperationLegal(ISD::MULHU, VT) :
+ isOperationLegalOrCustom(ISD::MULHU, VT))
Q = DAG.getNode(ISD::MULHU, dl, VT, Q, DAG.getConstant(magics.m, VT));
- else if (isOperationLegalOrCustom(ISD::UMUL_LOHI, VT))
+ else if (IsAfterLegalization ? isOperationLegal(ISD::UMUL_LOHI, VT) :
+ isOperationLegalOrCustom(ISD::UMUL_LOHI, VT))
Q = SDValue(DAG.getNode(ISD::UMUL_LOHI, dl, DAG.getVTList(VT, VT), Q,
DAG.getConstant(magics.m, VT)).getNode(), 1);
else