// Implements [SU]INT_TO_FP vector promotion; this is a [zs]ext of the input
// operand to the next size up.
SDValue PromoteVectorOpINT_TO_FP(SDValue Op);
+ // Implements FP_TO_[SU]INT vector promotion of the result type; it is
+ // promoted to the next size up integer type. The result is then truncated
+ // back to the original type.
+ SDValue PromoteVectorOpFP_TO_INT(SDValue Op, bool isSigned);
public:
bool Run();
};
bool VectorLegalizer::Run() {
+ // Before we start legalizing vector nodes, check if there are any vectors.
+ bool HasVectors = false;
+ for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
+ E = std::prev(DAG.allnodes_end()); I != std::next(E); ++I) {
+ // Check if the values of the nodes contain vectors. We don't need to check
+ // the operands because we are going to check their values at some point.
+ for (SDNode::value_iterator J = I->value_begin(), E = I->value_end();
+ J != E; ++J)
+ HasVectors |= J->isVector();
+
+ // If we found a vector node we can start the legalization.
+ if (HasVectors)
+ break;
+ }
+
+ // If this basic block has no vectors then no need to legalize vectors.
+ if (!HasVectors)
+ return false;
+
// The legalize process is inherently a bottom-up recursive process (users
// legalize their uses before themselves). Given infinite stack space, we
// could just start legalizing on the root and traverse the whole graph. In
// node is only legalized after all of its operands are legalized.
DAG.AssignTopologicalOrder();
for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
- E = prior(DAG.allnodes_end()); I != llvm::next(E); ++I)
+ E = std::prev(DAG.allnodes_end()); I != std::next(E); ++I)
LegalizeOp(SDValue(I, 0));
// Finally, it's possible the root changed. Get the new root.
return TranslateLegalizeResults(Op, Result);
case TargetLowering::Custom:
Changed = true;
- return LegalizeOp(TLI.LowerOperation(Result, DAG));
+ return TranslateLegalizeResults(Op, TLI.LowerOperation(Result, DAG));
case TargetLowering::Expand:
Changed = true;
return LegalizeOp(ExpandStore(Op));
case ISD::SRL:
case ISD::ROTL:
case ISD::ROTR:
+ case ISD::BSWAP:
case ISD::CTLZ:
case ISD::CTTZ:
case ISD::CTLZ_ZERO_UNDEF:
case ISD::FP_TO_UINT:
case ISD::FNEG:
case ISD::FABS:
+ case ISD::FCOPYSIGN:
case ISD::FSQRT:
case ISD::FSIN:
case ISD::FCOS:
case ISD::FTRUNC:
case ISD::FRINT:
case ISD::FNEARBYINT:
+ case ISD::FROUND:
case ISD::FFLOOR:
case ISD::FP_ROUND:
case ISD::FP_EXTEND:
Result = PromoteVectorOpINT_TO_FP(Op);
Changed = true;
break;
+ case ISD::FP_TO_UINT:
+ case ISD::FP_TO_SINT:
+ // Promote the operation by extending the operand.
+ Result = PromoteVectorOpFP_TO_INT(Op, Op->getOpcode() == ISD::FP_TO_SINT);
+ Changed = true;
+ break;
}
break;
case TargetLowering::Legal: break;
assert(Op.getNode()->getNumValues() == 1 &&
"Can't promote a vector with multiple results!");
MVT NVT = TLI.getTypeToPromoteTo(Op.getOpcode(), VT);
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
SmallVector<SDValue, 4> Operands(Op.getNumOperands());
for (unsigned j = 0; j != Op.getNumOperands(); ++j) {
//
// Increase the bitwidth of the element to the next pow-of-two
// (which is greater than 8 bits).
- unsigned NumElts = VT.getVectorNumElements();
- EVT EltVT = VT.getVectorElementType();
- EltVT = EVT::getIntegerVT(*DAG.getContext(), 2 * EltVT.getSizeInBits());
- assert(EltVT.isSimple() && "Promoting to a non-simple vector type!");
-
- // Build a new vector type and check if it is legal.
- MVT NVT = MVT::getVectorVT(EltVT.getSimpleVT(), NumElts);
- DebugLoc dl = Op.getDebugLoc();
+ EVT NVT = VT.widenIntegerVectorElementType(*DAG.getContext());
+ assert(NVT.isSimple() && "Promoting to a non-simple vector type!");
+ SDLoc dl(Op);
SmallVector<SDValue, 4> Operands(Op.getNumOperands());
unsigned Opc = Op.getOpcode() == ISD::UINT_TO_FP ? ISD::ZERO_EXTEND :
Operands.size());
}
+// For FP_TO_INT we promote the result type to a vector type with wider
+// elements and then truncate the result. This is different from the default
+// PromoteVector which uses bitcast to promote thus assumning that the
+// promoted vector type has the same overall size.
+SDValue VectorLegalizer::PromoteVectorOpFP_TO_INT(SDValue Op, bool isSigned) {
+ assert(Op.getNode()->getNumValues() == 1 &&
+ "Can't promote a vector with multiple results!");
+ EVT VT = Op.getValueType();
+
+ EVT NewVT;
+ unsigned NewOpc;
+ while (1) {
+ NewVT = VT.widenIntegerVectorElementType(*DAG.getContext());
+ assert(NewVT.isSimple() && "Promoting to a non-simple vector type!");
+ if (TLI.isOperationLegalOrCustom(ISD::FP_TO_SINT, NewVT)) {
+ NewOpc = ISD::FP_TO_SINT;
+ break;
+ }
+ if (!isSigned && TLI.isOperationLegalOrCustom(ISD::FP_TO_UINT, NewVT)) {
+ NewOpc = ISD::FP_TO_UINT;
+ break;
+ }
+ }
+
+ SDLoc loc(Op);
+ SDValue promoted = DAG.getNode(NewOpc, SDLoc(Op), NewVT, Op.getOperand(0));
+ return DAG.getNode(ISD::TRUNCATE, SDLoc(Op), VT, promoted);
+}
+
SDValue VectorLegalizer::ExpandLoad(SDValue Op) {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
LoadSDNode *LD = cast<LoadSDNode>(Op.getNode());
SDValue Chain = LD->getChain();
SDValue BasePTR = LD->getBasePtr();
ScalarLoad = DAG.getLoad(WideVT, dl, Chain, BasePTR,
LD->getPointerInfo().getWithOffset(Offset),
LD->isVolatile(), LD->isNonTemporal(),
- LD->isInvariant(), LD->getAlignment());
+ LD->isInvariant(), LD->getAlignment(),
+ LD->getTBAAInfo());
} else {
EVT LoadVT = WideVT;
while (RemainingBytes < LoadBytes) {
ScalarLoad = DAG.getExtLoad(ISD::EXTLOAD, dl, WideVT, Chain, BasePTR,
LD->getPointerInfo().getWithOffset(Offset),
LoadVT, LD->isVolatile(),
- LD->isNonTemporal(), LD->getAlignment());
+ LD->isNonTemporal(), LD->getAlignment(),
+ LD->getTBAAInfo());
}
RemainingBytes -= LoadBytes;
Offset += LoadBytes;
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getIntPtrConstant(LoadBytes));
+ DAG.getConstant(LoadBytes, BasePTR.getValueType()));
LoadVals.push_back(ScalarLoad.getValue(0));
LoadChains.push_back(ScalarLoad.getValue(1));
Chain, BasePTR, LD->getPointerInfo().getWithOffset(Idx * Stride),
SrcVT.getScalarType(),
LD->isVolatile(), LD->isNonTemporal(),
- LD->getAlignment());
+ LD->getAlignment(), LD->getTBAAInfo());
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getIntPtrConstant(Stride));
+ DAG.getConstant(Stride, BasePTR.getValueType()));
Vals.push_back(ScalarLoad.getValue(0));
LoadChains.push_back(ScalarLoad.getValue(1));
}
SDValue VectorLegalizer::ExpandStore(SDValue Op) {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
StoreSDNode *ST = cast<StoreSDNode>(Op.getNode());
SDValue Chain = ST->getChain();
SDValue BasePTR = ST->getBasePtr();
unsigned Alignment = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
+ const MDNode *TBAAInfo = ST->getTBAAInfo();
unsigned NumElem = StVT.getVectorNumElements();
// The type of the data we want to save
SmallVector<SDValue, 8> Stores;
for (unsigned Idx = 0; Idx < NumElem; Idx++) {
SDValue Ex = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
- RegSclVT, Value, DAG.getIntPtrConstant(Idx));
+ RegSclVT, Value, DAG.getConstant(Idx, TLI.getVectorIdxTy()));
// 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);
+ isVolatile, isNonTemporal, Alignment, TBAAInfo);
BasePTR = DAG.getNode(ISD::ADD, dl, BasePTR.getValueType(), BasePTR,
- DAG.getIntPtrConstant(Stride));
+ DAG.getConstant(Stride, BasePTR.getValueType()));
Stores.push_back(Store);
}
SDValue VectorLegalizer::ExpandSELECT(SDValue Op) {
// Lower a select instruction where the condition is a scalar and the
// operands are vectors. Lower this select to VSELECT and implement it
- // using XOR AND OR. The selector bit is broadcasted.
+ // using XOR AND OR. The selector bit is broadcasted.
EVT VT = Op.getValueType();
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
SDValue Mask = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
return DAG.UnrollVectorOp(Op.getNode());
// Generate a mask operand.
- EVT MaskTy = TLI.getSetCCResultType(VT);
- assert(MaskTy.isVector() && "Invalid CC type");
- assert(MaskTy.getSizeInBits() == Op1.getValueType().getSizeInBits()
- && "Invalid mask size");
+ EVT MaskTy = VT.changeVectorElementTypeToInteger();
// What is the size of each element in the vector mask.
EVT BitTy = MaskTy.getScalarType();
- Mask = DAG.getNode(ISD::SELECT, DL, BitTy, Mask,
+ Mask = DAG.getSelect(DL, BitTy, Mask,
DAG.getConstant(APInt::getAllOnesValue(BitTy.getSizeInBits()), BitTy),
DAG.getConstant(0, BitTy));
TLI.getOperationAction(ISD::SHL, VT) == TargetLowering::Expand)
return DAG.UnrollVectorOp(Op.getNode());
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
EVT OrigTy = cast<VTSDNode>(Op->getOperand(1))->getVT();
unsigned BW = VT.getScalarType().getSizeInBits();
SDValue VectorLegalizer::ExpandVSELECT(SDValue Op) {
// Implement VSELECT in terms of XOR, AND, OR
// on platforms which do not support blend natively.
- EVT VT = Op.getOperand(0).getValueType();
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
SDValue Mask = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue Op2 = Op.getOperand(2);
+ EVT VT = Mask.getValueType();
+
// If we can't even use the basic vector operations of
// AND,OR,XOR, we will have to scalarize the op.
// Notice that the operation may be 'promoted' which means that it is
TargetLowering::ZeroOrNegativeOneBooleanContent)
return DAG.UnrollVectorOp(Op.getNode());
- assert(VT.getSizeInBits() == Op1.getValueType().getSizeInBits()
- && "Invalid mask size");
+ // If the mask and the type are different sizes, unroll the vector op. This
+ // can occur when getSetCCResultType returns something that is different in
+ // size from the operand types. For example, v4i8 = select v4i32, v4i8, v4i8.
+ if (VT.getSizeInBits() != Op1.getValueType().getSizeInBits())
+ return DAG.UnrollVectorOp(Op.getNode());
+
// Bitcast the operands to be the same type as the mask.
// This is needed when we select between FP types because
// the mask is a vector of integers.
SDValue VectorLegalizer::ExpandUINT_TO_FLOAT(SDValue Op) {
EVT VT = Op.getOperand(0).getValueType();
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
// Make sure that the SINT_TO_FP and SRL instructions are available.
if (TLI.getOperationAction(ISD::SINT_TO_FP, VT) == TargetLowering::Expand ||
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, Op.getDebugLoc(), Op.getValueType(),
+ return DAG.getNode(ISD::FSUB, SDLoc(Op), Op.getValueType(),
Zero, Op.getOperand(0));
}
return DAG.UnrollVectorOp(Op.getNode());
EVT EltVT = VT.getVectorElementType();
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1), CC = Op.getOperand(2);
EVT TmpEltVT = LHS.getValueType().getVectorElementType();
- DebugLoc dl = Op.getDebugLoc();
+ 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.getIntPtrConstant(i));
+ DAG.getConstant(i, TLI.getVectorIdxTy()));
SDValue RHSElem = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, TmpEltVT, RHS,
- DAG.getIntPtrConstant(i));
- Ops[i] = DAG.getNode(ISD::SETCC, dl, TLI.getSetCCResultType(TmpEltVT),
+ DAG.getConstant(i, TLI.getVectorIdxTy()));
+ Ops[i] = DAG.getNode(ISD::SETCC, dl,
+ TLI.getSetCCResultType(*DAG.getContext(), TmpEltVT),
LHSElem, RHSElem, CC);
- Ops[i] = DAG.getNode(ISD::SELECT, dl, EltVT, Ops[i],
- DAG.getConstant(APInt::getAllOnesValue
- (EltVT.getSizeInBits()), EltVT),
- DAG.getConstant(0, EltVT));
+ Ops[i] = DAG.getSelect(dl, EltVT, Ops[i],
+ DAG.getConstant(APInt::getAllOnesValue
+ (EltVT.getSizeInBits()), EltVT),
+ DAG.getConstant(0, EltVT));
}
return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], NumElems);
}
projects / oota-llvm.git / blobdiff