SDValue ExpandLoad(SDValue Op);
SDValue ExpandStore(SDValue Op);
SDValue ExpandFNEG(SDValue Op);
+ SDValue ExpandBITREVERSE(SDValue Op);
/// \brief Implements vector promotion.
///
DAG.AssignTopologicalOrder();
for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
E = std::prev(DAG.allnodes_end()); I != std::next(E); ++I)
- LegalizeOp(SDValue(I, 0));
+ LegalizeOp(SDValue(&*I, 0));
// Finally, it's possible the root changed. Get the new root.
SDValue OldRoot = DAG.getRoot();
// Legalize the operands
SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0, e = Node->getNumOperands(); i != e; ++i)
- Ops.push_back(LegalizeOp(Node->getOperand(i)));
+ for (const SDValue &Op : Node->op_values())
+ Ops.push_back(LegalizeOp(Op));
SDValue Result = SDValue(DAG.UpdateNodeOperands(Op.getNode(), Ops), 0);
+ bool HasVectorValue = false;
if (Op.getOpcode() == ISD::LOAD) {
LoadSDNode *LD = cast<LoadSDNode>(Op.getNode());
ISD::LoadExtType ExtType = LD->getExtensionType();
return TranslateLegalizeResults(Op, Result);
case TargetLowering::Custom:
if (SDValue Lowered = TLI.LowerOperation(Result, DAG)) {
+ if (Lowered == Result)
+ return TranslateLegalizeResults(Op, Lowered);
Changed = true;
if (Lowered->getNumValues() != Op->getNumValues()) {
// This expanded to something other than the load. Assume the
assert(Result.getValue(1).use_empty() &&
"There are still live users of the old chain!");
return LegalizeOp(Lowered);
- } else {
- return TranslateLegalizeResults(Op, Lowered);
}
+ return TranslateLegalizeResults(Op, Lowered);
}
case TargetLowering::Expand:
Changed = true;
EVT StVT = ST->getMemoryVT();
MVT ValVT = ST->getValue().getSimpleValueType();
if (StVT.isVector() && ST->isTruncatingStore())
- switch (TLI.getTruncStoreAction(ValVT, StVT.getSimpleVT())) {
+ switch (TLI.getTruncStoreAction(ValVT, StVT)) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal:
return TranslateLegalizeResults(Op, Result);
- case TargetLowering::Custom:
- Changed = true;
- return TranslateLegalizeResults(Op, TLI.LowerOperation(Result, DAG));
+ case TargetLowering::Custom: {
+ SDValue Lowered = TLI.LowerOperation(Result, DAG);
+ Changed = Lowered != Result;
+ return TranslateLegalizeResults(Op, Lowered);
+ }
case TargetLowering::Expand:
Changed = true;
return LegalizeOp(ExpandStore(Op));
}
- }
+ } else if (Op.getOpcode() == ISD::MSCATTER || Op.getOpcode() == ISD::MSTORE)
+ HasVectorValue = true;
- bool HasVectorValue = false;
for (SDNode::value_iterator J = Node->value_begin(), E = Node->value_end();
J != E;
++J)
case ISD::UDIV:
case ISD::SREM:
case ISD::UREM:
+ case ISD::SDIVREM:
+ case ISD::UDIVREM:
case ISD::FADD:
case ISD::FSUB:
case ISD::FMUL:
case ISD::ROTL:
case ISD::ROTR:
case ISD::BSWAP:
+ case ISD::BITREVERSE:
case ISD::CTLZ:
case ISD::CTTZ:
case ISD::CTLZ_ZERO_UNDEF:
case ISD::FABS:
case ISD::FMINNUM:
case ISD::FMAXNUM:
+ case ISD::FMINNAN:
+ case ISD::FMAXNAN:
case ISD::FCOPYSIGN:
case ISD::FSQRT:
case ISD::FSIN:
case ISD::ANY_EXTEND_VECTOR_INREG:
case ISD::SIGN_EXTEND_VECTOR_INREG:
case ISD::ZERO_EXTEND_VECTOR_INREG:
+ case ISD::SMIN:
+ case ISD::SMAX:
+ case ISD::UMIN:
+ case ISD::UMAX:
QueryType = Node->getValueType(0);
break;
case ISD::FP_ROUND_INREG:
case ISD::UINT_TO_FP:
QueryType = Node->getOperand(0).getValueType();
break;
+ case ISD::MSCATTER:
+ QueryType = cast<MaskedScatterSDNode>(Node)->getValue().getValueType();
+ break;
+ case ISD::MSTORE:
+ QueryType = cast<MaskedStoreSDNode>(Node)->getValue().getValueType();
+ break;
}
switch (TLI.getOperationAction(Node->getOpcode(), QueryType)) {
+ default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Promote:
Result = Promote(Op);
Changed = true;
// There are currently two cases of vector promotion:
// 1) Bitcasting a vector of integers to a different type to a vector of the
- // same overall length. For example, x86 promotes ISD::AND on v2i32 to v1i64.
- // 2) Extending a vector of floats to a vector of the same number oflarger
+ // same overall length. For example, x86 promotes ISD::AND v2i32 to v1i64.
+ // 2) Extending a vector of floats to a vector of the same number of larger
// floats. For example, AArch64 promotes ISD::FADD on v4f16 to v4f32.
MVT VT = Op.getSimpleValueType();
assert(Op.getNode()->getNumValues() == 1 &&
if (Op.getOperand(j)
.getValueType()
.getVectorElementType()
- .isFloatingPoint())
+ .isFloatingPoint() &&
+ NVT.isVector() && NVT.getVectorElementType().isFloatingPoint())
Operands[j] = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Op.getOperand(j));
else
Operands[j] = DAG.getNode(ISD::BITCAST, dl, NVT, Op.getOperand(j));
Operands[j] = Op.getOperand(j);
}
- Op = DAG.getNode(Op.getOpcode(), dl, NVT, Operands);
- if (VT.isFloatingPoint() ||
- (VT.isVector() && VT.getVectorElementType().isFloatingPoint()))
- return DAG.getNode(ISD::FP_ROUND, dl, VT, Op, DAG.getIntPtrConstant(0));
+ Op = DAG.getNode(Op.getOpcode(), dl, NVT, Operands, Op.getNode()->getFlags());
+ if ((VT.isFloatingPoint() && NVT.isFloatingPoint()) ||
+ (VT.isVector() && VT.getVectorElementType().isFloatingPoint() &&
+ NVT.isVector() && NVT.getVectorElementType().isFloatingPoint()))
+ return DAG.getNode(ISD::FP_ROUND, dl, VT, Op, DAG.getIntPtrConstant(0, dl));
else
return DAG.getNode(ISD::BITCAST, dl, VT, Op);
}
// Instead, we load all significant words, mask bits off, and concatenate
// them to form each element. Finally, they are extended to destination
// scalar type to build the destination vector.
- EVT WideVT = TLI.getPointerTy();
+ EVT WideVT = TLI.getPointerTy(DAG.getDataLayout());
assert(WideVT.isRound() &&
"Could not handle the sophisticated case when the widest integer is"
ScalarLoad = DAG.getLoad(WideVT, dl, Chain, BasePTR,
LD->getPointerInfo().getWithOffset(Offset),
LD->isVolatile(), LD->isNonTemporal(),
- LD->isInvariant(), LD->getAlignment(),
+ LD->isInvariant(),
+ MinAlign(LD->getAlignment(), Offset),
LD->getAAInfo());
} else {
EVT LoadVT = WideVT;
LD->getPointerInfo().getWithOffset(Offset),
LoadVT, LD->isVolatile(),
LD->isNonTemporal(), LD->isInvariant(),
- LD->getAlignment(), LD->getAAInfo());
+ MinAlign(LD->getAlignment(), Offset),
+ LD->getAAInfo());
}
RemainingBytes -= LoadBytes;
Offset += LoadBytes;
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getConstant(LoadBytes, BasePTR.getValueType()));
+ DAG.getConstant(LoadBytes, dl,
+ BasePTR.getValueType()));
LoadVals.push_back(ScalarLoad.getValue(0));
LoadChains.push_back(ScalarLoad.getValue(1));
// Extract bits, pack and extend/trunc them into destination type.
unsigned SrcEltBits = SrcEltVT.getSizeInBits();
- SDValue SrcEltBitMask = DAG.getConstant((1U << SrcEltBits) - 1, WideVT);
+ SDValue SrcEltBitMask = DAG.getConstant((1U << SrcEltBits) - 1, dl, WideVT);
unsigned BitOffset = 0;
unsigned WideIdx = 0;
SDValue Lo, Hi, ShAmt;
if (BitOffset < WideBits) {
- ShAmt = DAG.getConstant(BitOffset, TLI.getShiftAmountTy(WideVT));
+ ShAmt = DAG.getConstant(
+ BitOffset, dl, TLI.getShiftAmountTy(WideVT, DAG.getDataLayout()));
Lo = DAG.getNode(ISD::SRL, dl, WideVT, LoadVals[WideIdx], ShAmt);
Lo = DAG.getNode(ISD::AND, dl, WideVT, Lo, SrcEltBitMask);
}
WideIdx++;
BitOffset -= WideBits;
if (BitOffset > 0) {
- ShAmt = DAG.getConstant(SrcEltBits - BitOffset,
- TLI.getShiftAmountTy(WideVT));
+ ShAmt = DAG.getConstant(
+ SrcEltBits - BitOffset, dl,
+ TLI.getShiftAmountTy(WideVT, DAG.getDataLayout()));
Hi = DAG.getNode(ISD::SHL, dl, WideVT, LoadVals[WideIdx], ShAmt);
Hi = DAG.getNode(ISD::AND, dl, WideVT, Hi, SrcEltBitMask);
}
Lo = DAG.getZExtOrTrunc(Lo, dl, DstEltVT);
break;
case ISD::SEXTLOAD:
- ShAmt = DAG.getConstant(WideBits - SrcEltBits,
- TLI.getShiftAmountTy(WideVT));
+ ShAmt =
+ DAG.getConstant(WideBits - SrcEltBits, dl,
+ TLI.getShiftAmountTy(WideVT, DAG.getDataLayout()));
Lo = DAG.getNode(ISD::SHL, dl, WideVT, Lo, ShAmt);
Lo = DAG.getNode(ISD::SRA, dl, WideVT, Lo, ShAmt);
Lo = DAG.getSExtOrTrunc(Lo, dl, DstEltVT);
Chain, BasePTR, LD->getPointerInfo().getWithOffset(Idx * Stride),
SrcVT.getScalarType(),
LD->isVolatile(), LD->isNonTemporal(), LD->isInvariant(),
- LD->getAlignment(), LD->getAAInfo());
+ MinAlign(LD->getAlignment(), Idx * Stride), LD->getAAInfo());
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getConstant(Stride, BasePTR.getValueType()));
+ DAG.getConstant(Stride, dl, BasePTR.getValueType()));
Vals.push_back(ScalarLoad.getValue(0));
LoadChains.push_back(ScalarLoad.getValue(1));
// and save them into memory individually.
SmallVector<SDValue, 8> Stores;
for (unsigned Idx = 0; Idx < NumElem; Idx++) {
- SDValue Ex = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
- RegSclVT, Value, DAG.getConstant(Idx, TLI.getVectorIdxTy()));
+ SDValue Ex = DAG.getNode(
+ ISD::EXTRACT_VECTOR_ELT, dl, RegSclVT, Value,
+ DAG.getConstant(Idx, dl, TLI.getVectorIdxTy(DAG.getDataLayout())));
// This scalar TruncStore may be illegal, but we legalize it later.
SDValue Store = DAG.getTruncStore(Chain, dl, Ex, BasePTR,
ST->getPointerInfo().getWithOffset(Idx*Stride), MemSclVT,
- isVolatile, isNonTemporal, Alignment, AAInfo);
+ isVolatile, isNonTemporal, MinAlign(Alignment, Idx*Stride),
+ AAInfo);
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getConstant(Stride, BasePTR.getValueType()));
+ DAG.getConstant(Stride, dl, BasePTR.getValueType()));
Stores.push_back(Store);
}
return ExpandFNEG(Op);
case ISD::SETCC:
return UnrollVSETCC(Op);
+ case ISD::BITREVERSE:
+ return ExpandBITREVERSE(Op);
default:
return DAG.UnrollVectorOp(Op.getNode());
}
EVT BitTy = MaskTy.getScalarType();
Mask = DAG.getSelect(DL, BitTy, Mask,
- DAG.getConstant(APInt::getAllOnesValue(BitTy.getSizeInBits()), BitTy),
- DAG.getConstant(0, BitTy));
+ DAG.getConstant(APInt::getAllOnesValue(BitTy.getSizeInBits()), DL,
+ BitTy),
+ DAG.getConstant(0, DL, BitTy));
// Broadcast the mask so that the entire vector is all-one or all zero.
SmallVector<SDValue, 8> Ops(NumElem, Mask);
Op2 = DAG.getNode(ISD::BITCAST, DL, MaskTy, Op2);
SDValue AllOnes = DAG.getConstant(
- APInt::getAllOnesValue(BitTy.getSizeInBits()), MaskTy);
+ APInt::getAllOnesValue(BitTy.getSizeInBits()), DL, MaskTy);
SDValue NotMask = DAG.getNode(ISD::XOR, DL, MaskTy, Mask, AllOnes);
Op1 = DAG.getNode(ISD::AND, DL, MaskTy, Op1, Mask);
unsigned BW = VT.getScalarType().getSizeInBits();
unsigned OrigBW = OrigTy.getScalarType().getSizeInBits();
- SDValue ShiftSz = DAG.getConstant(BW - OrigBW, VT);
+ SDValue ShiftSz = DAG.getConstant(BW - OrigBW, DL, VT);
Op = Op.getOperand(0);
Op = DAG.getNode(ISD::SHL, DL, VT, Op, ShiftSz);
// Place the extended lanes into the correct locations.
int ExtLaneScale = NumSrcElements / NumElements;
- int EndianOffset = TLI.isBigEndian() ? ExtLaneScale - 1 : 0;
+ int EndianOffset = DAG.getDataLayout().isBigEndian() ? ExtLaneScale - 1 : 0;
for (int i = 0; i < NumElements; ++i)
ShuffleMask[i * ExtLaneScale + EndianOffset] = i;
// without full scalarization than the sign extension does.
unsigned EltWidth = VT.getVectorElementType().getSizeInBits();
unsigned SrcEltWidth = SrcVT.getVectorElementType().getSizeInBits();
- SDValue ShiftAmount = DAG.getConstant(EltWidth - SrcEltWidth, VT);
+ SDValue ShiftAmount = DAG.getConstant(EltWidth - SrcEltWidth, DL, VT);
return DAG.getNode(ISD::SRA, DL, VT,
DAG.getNode(ISD::SHL, DL, VT, Op, ShiftAmount),
ShiftAmount);
// Build up a zero vector to blend into this one.
EVT SrcScalarVT = SrcVT.getScalarType();
- SDValue ScalarZero = DAG.getTargetConstant(0, SrcScalarVT);
+ SDValue ScalarZero = DAG.getTargetConstant(0, DL, SrcScalarVT);
SmallVector<SDValue, 4> BuildVectorOperands(NumSrcElements, ScalarZero);
SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, DL, SrcVT, BuildVectorOperands);
ShuffleMask.push_back(i);
int ExtLaneScale = NumSrcElements / NumElements;
- int EndianOffset = TLI.isBigEndian() ? ExtLaneScale - 1 : 0;
+ int EndianOffset = DAG.getDataLayout().isBigEndian() ? ExtLaneScale - 1 : 0;
for (int i = 0; i < NumElements; ++i)
ShuffleMask[i * ExtLaneScale + EndianOffset] = NumSrcElements + i;
return DAG.getNode(ISD::BITCAST, DL, VT, Op);
}
+SDValue VectorLegalizer::ExpandBITREVERSE(SDValue Op) {
+ EVT VT = Op.getValueType();
+
+ // If we have the scalar operation, it's probably cheaper to unroll it.
+ if (TLI.isOperationLegalOrCustom(ISD::BITREVERSE, VT.getScalarType()))
+ return DAG.UnrollVectorOp(Op.getNode());
+
+ // If we have the appropriate vector bit operations, it is better to use them
+ // than unrolling and expanding each component.
+ if (!TLI.isOperationLegalOrCustom(ISD::SHL, VT) ||
+ !TLI.isOperationLegalOrCustom(ISD::SRL, VT) ||
+ !TLI.isOperationLegalOrCustom(ISD::AND, VT) ||
+ !TLI.isOperationLegalOrCustom(ISD::OR, VT))
+ return DAG.UnrollVectorOp(Op.getNode());
+
+ // Let LegalizeDAG handle this later.
+ return Op;
+}
+
SDValue VectorLegalizer::ExpandVSELECT(SDValue Op) {
// Implement VSELECT in terms of XOR, AND, OR
// on platforms which do not support blend natively.
Op2 = DAG.getNode(ISD::BITCAST, DL, VT, Op2);
SDValue AllOnes = DAG.getConstant(
- APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()), VT);
+ APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()), DL, VT);
SDValue NotMask = DAG.getNode(ISD::XOR, DL, VT, Mask, AllOnes);
Op1 = DAG.getNode(ISD::AND, DL, VT, Op1, Mask);
"Elements in vector-UINT_TO_FP must be 32 or 64 bits wide");
unsigned BW = SVT.getSizeInBits();
- SDValue HalfWord = DAG.getConstant(BW/2, VT);
+ SDValue HalfWord = DAG.getConstant(BW/2, DL, VT);
// Constants to clear the upper part of the word.
// Notice that we can also use SHL+SHR, but using a constant is slightly
// faster on x86.
uint64_t HWMask = (SVT.getSizeInBits()==64)?0x00000000FFFFFFFF:0x0000FFFF;
- SDValue HalfWordMask = DAG.getConstant(HWMask, VT);
+ SDValue HalfWordMask = DAG.getConstant(HWMask, DL, VT);
// Two to the power of half-word-size.
- SDValue TWOHW = DAG.getConstantFP((1<<(BW/2)), Op.getValueType());
+ SDValue TWOHW = DAG.getConstantFP(1 << (BW/2), DL, Op.getValueType());
// Clear upper part of LO, lower HI
SDValue HI = DAG.getNode(ISD::SRL, DL, VT, Op.getOperand(0), HalfWord);
// Convert hi and lo to floats
// Convert the hi part back to the upper values
+ // TODO: Can any fast-math-flags be set on these nodes?
SDValue fHI = DAG.getNode(ISD::SINT_TO_FP, DL, Op.getValueType(), HI);
fHI = DAG.getNode(ISD::FMUL, DL, Op.getValueType(), fHI, TWOHW);
SDValue fLO = DAG.getNode(ISD::SINT_TO_FP, DL, Op.getValueType(), LO);
SDValue VectorLegalizer::ExpandFNEG(SDValue Op) {
if (TLI.isOperationLegalOrCustom(ISD::FSUB, Op.getValueType())) {
- SDValue Zero = DAG.getConstantFP(-0.0, Op.getValueType());
- return DAG.getNode(ISD::FSUB, SDLoc(Op), Op.getValueType(),
+ SDLoc DL(Op);
+ SDValue Zero = DAG.getConstantFP(-0.0, DL, Op.getValueType());
+ // TODO: If FNEG had fast-math-flags, they'd get propagated to this FSUB.
+ return DAG.getNode(ISD::FSUB, DL, Op.getValueType(),
Zero, Op.getOperand(0));
}
return DAG.UnrollVectorOp(Op.getNode());
SDLoc dl(Op);
SmallVector<SDValue, 8> Ops(NumElems);
for (unsigned i = 0; i < NumElems; ++i) {
- SDValue LHSElem = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, TmpEltVT, LHS,
- DAG.getConstant(i, TLI.getVectorIdxTy()));
- SDValue RHSElem = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, TmpEltVT, RHS,
- DAG.getConstant(i, TLI.getVectorIdxTy()));
+ SDValue LHSElem = DAG.getNode(
+ ISD::EXTRACT_VECTOR_ELT, dl, TmpEltVT, LHS,
+ DAG.getConstant(i, dl, TLI.getVectorIdxTy(DAG.getDataLayout())));
+ SDValue RHSElem = DAG.getNode(
+ ISD::EXTRACT_VECTOR_ELT, dl, TmpEltVT, RHS,
+ DAG.getConstant(i, dl, TLI.getVectorIdxTy(DAG.getDataLayout())));
Ops[i] = DAG.getNode(ISD::SETCC, dl,
- TLI.getSetCCResultType(*DAG.getContext(), TmpEltVT),
+ TLI.getSetCCResultType(DAG.getDataLayout(),
+ *DAG.getContext(), TmpEltVT),
LHSElem, RHSElem, CC);
Ops[i] = DAG.getSelect(dl, EltVT, Ops[i],
DAG.getConstant(APInt::getAllOnesValue
- (EltVT.getSizeInBits()), EltVT),
- DAG.getConstant(0, EltVT));
+ (EltVT.getSizeInBits()), dl, EltVT),
+ DAG.getConstant(0, dl, EltVT));
}
return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}