static
SDValue Compact8x32ShuffleNode(ShuffleVectorSDNode *SVOp,
SelectionDAG &DAG) {
- EVT VT = SVOp->getValueType(0);
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
DebugLoc dl = SVOp->getDebugLoc();
if (VT != MVT::v8i32 && VT != MVT::v8f32)
/// their permute mask.
static SDValue CommuteVectorShuffle(ShuffleVectorSDNode *SVOp,
SelectionDAG &DAG) {
- EVT VT = SVOp->getValueType(0);
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
unsigned NumElems = VT.getVectorNumElements();
SmallVector<int, 8> MaskVec;
/// Always build ones vectors as <4 x i32> or <8 x i32>. For 256-bit types with
/// no AVX2 supprt, use two <4 x i32> inserted in a <8 x i32> appropriately.
/// Then bitcast to their original type, ensuring they get CSE'd.
-static SDValue getOnesVector(EVT VT, bool HasInt256, SelectionDAG &DAG,
+static SDValue getOnesVector(MVT VT, bool HasInt256, SelectionDAG &DAG,
DebugLoc dl) {
assert(VT.isVector() && "Expected a vector type");
if (!Subtarget->hasFp256())
return SDValue();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
DebugLoc dl = Op.getDebugLoc();
assert((VT.is128BitVector() || VT.is256BitVector()) &&
X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
DebugLoc dl = Op.getDebugLoc();
- EVT VT = Op.getValueType();
- EVT ExtVT = VT.getVectorElementType();
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT ExtVT = VT.getVectorElementType();
unsigned NumElems = Op.getNumOperands();
// Vectors containing all zeros can be matched by pxor and xorps later
// to create 256-bit vectors from two other 128-bit ones.
static SDValue LowerAVXCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- EVT ResVT = Op.getValueType();
+ MVT ResVT = Op.getValueType().getSimpleVT();
assert(ResVT.is256BitVector() && "Value type must be 256-bit wide");
SDValue V1 = SVOp->getOperand(0);
SDValue V2 = SVOp->getOperand(1);
DebugLoc dl = SVOp->getDebugLoc();
- EVT VT = SVOp->getValueType(0);
- EVT EltVT = VT.getVectorElementType();
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
+ MVT EltVT = VT.getVectorElementType();
unsigned NumElems = VT.getVectorNumElements();
if (!Subtarget->hasSSE41() || EltVT == MVT::i8)
// Check the mask for BLEND and build the value.
unsigned MaskValue = 0;
// There are 2 lanes if (NumElems > 8), and 1 lane otherwise.
- unsigned NumLanes = (NumElems-1)/8 + 1;
+ unsigned NumLanes = (NumElems-1)/8 + 1;
unsigned NumElemsInLane = NumElems / NumLanes;
// Blend for v16i16 should be symetric for the both lanes.
for (unsigned i = 0; i < NumElemsInLane; ++i) {
- int SndLaneEltIdx = (NumLanes == 2) ?
+ int SndLaneEltIdx = (NumLanes == 2) ?
SVOp->getMaskElt(i + NumElemsInLane) : -1;
int EltIdx = SVOp->getMaskElt(i);
- if ((EltIdx == -1 || EltIdx == (int)i) &&
- (SndLaneEltIdx == -1 || SndLaneEltIdx == (int)(i + NumElemsInLane)))
+ if ((EltIdx < 0 || EltIdx == (int)i) &&
+ (SndLaneEltIdx < 0 || SndLaneEltIdx == (int)(i + NumElemsInLane)))
continue;
- if (((unsigned)EltIdx == (i + NumElems)) &&
- (SndLaneEltIdx == -1 ||
+ if (((unsigned)EltIdx == (i + NumElems)) &&
+ (SndLaneEltIdx < 0 ||
(unsigned)SndLaneEltIdx == i + NumElems + NumElemsInLane))
MaskValue |= (1<<i);
- else
+ else
return SDValue();
}
// Convert i32 vectors to floating point if it is not AVX2.
// AVX2 introduced VPBLENDD instruction for 128 and 256-bit vectors.
- EVT BlendVT = VT;
+ MVT BlendVT = VT;
if (EltVT == MVT::i64 || (EltVT == MVT::i32 && !Subtarget->hasInt256())) {
- BlendVT = EVT::getVectorVT(*DAG.getContext(),
- EVT::getFloatingPointVT(EltVT.getSizeInBits()),
- NumElems);
+ BlendVT = MVT::getVectorVT(MVT::getFloatingPointVT(EltVT.getSizeInBits()),
+ NumElems);
V1 = DAG.getNode(ISD::BITCAST, dl, VT, V1);
V2 = DAG.getNode(ISD::BITCAST, dl, VT, V2);
}
-
- SDValue Ret = DAG.getNode(X86ISD::BLENDI, dl, BlendVT, V1, V2,
- DAG.getConstant(MaskValue, MVT::i32));
+
+ SDValue Ret = DAG.getNode(X86ISD::BLENDI, dl, BlendVT, V1, V2,
+ DAG.getConstant(MaskValue, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, VT, Ret);
}
SDValue LowerVECTOR_SHUFFLEv32i8(ShuffleVectorSDNode *SVOp,
const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
- EVT VT = SVOp->getValueType(0);
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
SDValue V1 = SVOp->getOperand(0);
SDValue V2 = SVOp->getOperand(1);
DebugLoc dl = SVOp->getDebugLoc();
/// vector_shuffle X, Y, <2, 3, | 10, 11, | 0, 1, | 14, 15>
static
SDValue RewriteAsNarrowerShuffle(ShuffleVectorSDNode *SVOp,
- SelectionDAG &DAG, DebugLoc dl) {
+ SelectionDAG &DAG) {
MVT VT = SVOp->getValueType(0).getSimpleVT();
+ DebugLoc dl = SVOp->getDebugLoc();
unsigned NumElems = VT.getVectorNumElements();
MVT NewVT;
unsigned Scale;
/// getVZextMovL - Return a zero-extending vector move low node.
///
-static SDValue getVZextMovL(EVT VT, EVT OpVT,
+static SDValue getVZextMovL(MVT VT, EVT OpVT,
SDValue SrcOp, SelectionDAG &DAG,
const X86Subtarget *Subtarget, DebugLoc dl) {
if (VT == MVT::v2f64 || VT == MVT::v4f32) {
if (NewOp.getNode())
return NewOp;
- EVT VT = SVOp->getValueType(0);
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
unsigned NumElems = VT.getVectorNumElements();
unsigned NumLaneElems = NumElems / 2;
DebugLoc dl = SVOp->getDebugLoc();
- MVT EltVT = VT.getVectorElementType().getSimpleVT();
- EVT NVT = MVT::getVectorVT(EltVT, NumLaneElems);
+ MVT EltVT = VT.getVectorElementType();
+ MVT NVT = MVT::getVectorVT(EltVT, NumLaneElems);
SDValue Output[2];
SmallVector<int, 16> Mask;
SDValue V1 = SVOp->getOperand(0);
SDValue V2 = SVOp->getOperand(1);
DebugLoc dl = SVOp->getDebugLoc();
- EVT VT = SVOp->getValueType(0);
+ MVT VT = SVOp->getValueType(0).getSimpleVT();
assert(VT.is128BitVector() && "Unsupported vector size");
SDValue
X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
DebugLoc dl = Op.getDebugLoc();
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
// do it!
if (VT == MVT::v8i16 || VT == MVT::v16i8 ||
VT == MVT::v16i16 || VT == MVT::v32i8) {
- SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+ SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
if (NewOp.getNode())
return DAG.getNode(ISD::BITCAST, dl, VT, NewOp);
} else if ((VT == MVT::v4i32 ||
// FIXME: Figure out a cleaner way to do this.
// Try to make use of movq to zero out the top part.
if (ISD::isBuildVectorAllZeros(V2.getNode())) {
- SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+ SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
if (NewOp.getNode()) {
- EVT NewVT = NewOp.getValueType();
+ MVT NewVT = NewOp.getValueType().getSimpleVT();
if (isCommutedMOVLMask(cast<ShuffleVectorSDNode>(NewOp)->getMask(),
NewVT, true, false))
return getVZextMovL(VT, NewVT, NewOp.getOperand(0),
DAG, Subtarget, dl);
}
} else if (ISD::isBuildVectorAllZeros(V1.getNode())) {
- SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+ SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
if (NewOp.getNode()) {
- EVT NewVT = NewOp.getValueType();
+ MVT NewVT = NewOp.getValueType().getSimpleVT();
if (isMOVLMask(cast<ShuffleVectorSDNode>(NewOp)->getMask(), NewVT))
return getVZextMovL(VT, NewVT, NewOp.getOperand(1),
DAG, Subtarget, dl);
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
DebugLoc dl = Op.getDebugLoc();
unsigned NumElems = VT.getVectorNumElements();
bool V1IsUndef = V1.getOpcode() == ISD::UNDEF;
if (isShift && ShVal.hasOneUse()) {
// If the shifted value has multiple uses, it may be cheaper to use
// v_set0 + movlhps or movhlps, etc.
- EVT EltVT = VT.getVectorElementType();
+ MVT EltVT = VT.getVectorElementType();
ShAmt *= EltVT.getSizeInBits();
return getVShift(isLeft, VT, ShVal, ShAmt, DAG, *this, dl);
}
if (isShift) {
// No better options. Use a vshldq / vsrldq.
- EVT EltVT = VT.getVectorElementType();
+ MVT EltVT = VT.getVectorElementType();
ShAmt *= EltVT.getSizeInBits();
return getVShift(isLeft, VT, ShVal, ShAmt, DAG, *this, dl);
}
return SDValue();
}
-SDValue
-X86TargetLowering::LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op,
- SelectionDAG &DAG) const {
- EVT VT = Op.getValueType();
+static SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.getValueType().getSimpleVT();
DebugLoc dl = Op.getDebugLoc();
- if (!Op.getOperand(0).getValueType().is128BitVector())
+ if (!Op.getOperand(0).getValueType().getSimpleVT().is128BitVector())
return SDValue();
if (VT.getSizeInBits() == 8) {
return SDValue();
SDValue Vec = Op.getOperand(0);
- EVT VecVT = Vec.getValueType();
+ MVT VecVT = Vec.getValueType().getSimpleVT();
// If this is a 256-bit vector result, first extract the 128-bit vector and
// then extract the element from the 128-bit vector.
return Res;
}
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
DebugLoc dl = Op.getDebugLoc();
// TODO: handle v16i8.
if (VT.getSizeInBits() == 16) {
MVT::v4i32, Vec),
Op.getOperand(1)));
// Transform it so it match pextrw which produces a 32-bit result.
- EVT EltVT = MVT::i32;
+ MVT EltVT = MVT::i32;
SDValue Extract = DAG.getNode(X86ISD::PEXTRW, dl, EltVT,
Op.getOperand(0), Op.getOperand(1));
SDValue Assert = DAG.getNode(ISD::AssertZext, dl, EltVT, Extract,
// SHUFPS the element to the lowest double word, then movss.
int Mask[4] = { static_cast<int>(Idx), -1, -1, -1 };
- EVT VVT = Op.getOperand(0).getValueType();
+ MVT VVT = Op.getOperand(0).getValueType().getSimpleVT();
SDValue Vec = DAG.getVectorShuffle(VVT, dl, Op.getOperand(0),
DAG.getUNDEF(VVT), Mask);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, Vec,
// Note if the lower 64 bits of the result of the UNPCKHPD is then stored
// to a f64mem, the whole operation is folded into a single MOVHPDmr.
int Mask[2] = { 1, -1 };
- EVT VVT = Op.getOperand(0).getValueType();
+ MVT VVT = Op.getOperand(0).getValueType().getSimpleVT();
SDValue Vec = DAG.getVectorShuffle(VVT, dl, Op.getOperand(0),
DAG.getUNDEF(VVT), Mask);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, Vec,
return SDValue();
}
-SDValue
-X86TargetLowering::LowerINSERT_VECTOR_ELT_SSE4(SDValue Op,
- SelectionDAG &DAG) const {
- EVT VT = Op.getValueType();
- EVT EltVT = VT.getVectorElementType();
+static SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT EltVT = VT.getVectorElementType();
DebugLoc dl = Op.getDebugLoc();
SDValue N0 = Op.getOperand(0);
SDValue
X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const {
- EVT VT = Op.getValueType();
- EVT EltVT = VT.getVectorElementType();
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT EltVT = VT.getVectorElementType();
DebugLoc dl = Op.getDebugLoc();
SDValue N0 = Op.getOperand(0);
static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
LLVMContext *Context = DAG.getContext();
DebugLoc dl = Op.getDebugLoc();
- EVT OpVT = Op.getValueType();
+ MVT OpVT = Op.getValueType().getSimpleVT();
// If this is a 256-bit vector result, first insert into a 128-bit
// vector and then insert into the 256-bit vector.
SDValue
X86TargetLowering::LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
- int64_t Offset,
- SelectionDAG &DAG) const {
+ int64_t Offset, SelectionDAG &DAG) const {
// Create the TargetGlobalAddress node, folding in the constant
// offset if it is legal.
unsigned char OpFlags =
case TLSModel::LocalExec:
return LowerToTLSExecModel(GA, DAG, getPointerTy(), model,
Subtarget->is64Bit(),
- getTargetMachine().getRelocationModel() == Reloc::PIC_);
+ getTargetMachine().getRelocationModel() == Reloc::PIC_);
}
llvm_unreachable("Unknown TLS model.");
}
SVT == MVT::v8i8 || SVT == MVT::v8i16) &&
"Custom UINT_TO_FP is not supported!");
- EVT NVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, SVT.getVectorNumElements());
+ EVT NVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32,
+ SVT.getVectorNumElements());
return DAG.getNode(ISD::SINT_TO_FP, dl, Op.getValueType(),
DAG.getNode(ISD::ZERO_EXTEND, dl, NVT, N0));
}
return DAG.getNode(ISD::FP_ROUND, dl, DstVT, Add, DAG.getIntPtrConstant(0));
}
-std::pair<SDValue,SDValue> X86TargetLowering::
-FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, bool IsSigned, bool IsReplace) const {
+std::pair<SDValue,SDValue>
+X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
+ bool IsSigned, bool IsReplace) const {
DebugLoc DL = Op.getDebugLoc();
EVT DstTy = Op.getValueType();
static SDValue LowerAVXExtend(SDValue Op, SelectionDAG &DAG,
const X86Subtarget *Subtarget) {
- EVT VT = Op->getValueType(0);
+ MVT VT = Op->getValueType(0).getSimpleVT();
SDValue In = Op->getOperand(0);
- EVT InVT = In.getValueType();
+ MVT InVT = In.getValueType().getSimpleVT();
DebugLoc dl = Op->getDebugLoc();
// Optimize vectors in AVX mode:
SDValue OpLo = getUnpackl(DAG, dl, InVT, In, NeedZero ? ZeroVec : Undef);
SDValue OpHi = getUnpackh(DAG, dl, InVT, In, NeedZero ? ZeroVec : Undef);
- EVT HVT = EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(),
+ MVT HVT = MVT::getVectorVT(VT.getVectorElementType(),
VT.getVectorNumElements()/2);
OpLo = DAG.getNode(ISD::BITCAST, dl, HVT, OpLo);
SDValue X86TargetLowering::LowerZERO_EXTEND(SDValue Op,
SelectionDAG &DAG) const {
DebugLoc DL = Op.getDebugLoc();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
SDValue In = Op.getOperand(0);
- EVT SVT = In.getValueType();
+ MVT SVT = In.getValueType().getSimpleVT();
if (Subtarget->hasFp256()) {
SDValue Res = LowerAVXExtend(Op, DAG, Subtarget);
return DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i32, Lo, Hi);
}
-SDValue X86TargetLowering::lowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
DebugLoc DL = Op.getDebugLoc();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
SDValue In = Op.getOperand(0);
- EVT SVT = In.getValueType();
+ MVT SVT = In.getValueType().getSimpleVT();
if ((VT == MVT::v4i32) && (SVT == MVT::v4i64)) {
// On AVX2, v4i64 -> v4i32 becomes VPERMD.
SDValue X86TargetLowering::LowerFP_TO_SINT(SDValue Op,
SelectionDAG &DAG) const {
- if (Op.getValueType().isVector()) {
- if (Op.getValueType() == MVT::v8i16)
- return DAG.getNode(ISD::TRUNCATE, Op.getDebugLoc(), Op.getValueType(),
+ MVT VT = Op.getValueType().getSimpleVT();
+ if (VT.isVector()) {
+ if (VT == MVT::v8i16)
+ return DAG.getNode(ISD::TRUNCATE, Op.getDebugLoc(), VT,
DAG.getNode(ISD::FP_TO_SINT, Op.getDebugLoc(),
MVT::v8i32, Op.getOperand(0)));
return SDValue();
return FIST;
}
-SDValue X86TargetLowering::lowerFP_EXTEND(SDValue Op,
- SelectionDAG &DAG) const {
+static SDValue LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) {
DebugLoc DL = Op.getDebugLoc();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
SDValue In = Op.getOperand(0);
- EVT SVT = In.getValueType();
+ MVT SVT = In.getValueType().getSimpleVT();
assert(SVT == MVT::v2f32 && "Only customize MVT::v2f32 type legalization!");
SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
LLVMContext *Context = DAG.getContext();
DebugLoc dl = Op.getDebugLoc();
- EVT VT = Op.getValueType();
- EVT EltVT = VT;
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT EltVT = VT;
unsigned NumElts = VT == MVT::f64 ? 2 : 4;
if (VT.isVector()) {
EltVT = VT.getVectorElementType();
SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) const {
LLVMContext *Context = DAG.getContext();
DebugLoc dl = Op.getDebugLoc();
- EVT VT = Op.getValueType();
- EVT EltVT = VT;
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT EltVT = VT;
unsigned NumElts = VT == MVT::f64 ? 2 : 4;
if (VT.isVector()) {
EltVT = VT.getVectorElementType();
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
DebugLoc dl = Op.getDebugLoc();
- EVT VT = Op.getValueType();
- EVT SrcVT = Op1.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
+ MVT SrcVT = Op1.getValueType().getSimpleVT();
// If second operand is smaller, extend it first.
if (SrcVT.bitsLT(VT)) {
static SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) {
SDValue N0 = Op.getOperand(0);
DebugLoc dl = Op.getDebugLoc();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
// Lower ISD::FGETSIGN to (AND (X86ISD::FGETSIGNx86 ...) 1).
SDValue xFGETSIGN = DAG.getNode(X86ISD::FGETSIGNx86, dl, VT, N0,
// LowerVectorAllZeroTest - Check whether an OR'd tree is PTEST-able.
//
-SDValue X86TargetLowering::LowerVectorAllZeroTest(SDValue Op, SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerVectorAllZeroTest(SDValue Op,
+ SelectionDAG &DAG) const {
assert(Op.getOpcode() == ISD::OR && "Only check OR'd tree.");
if (!Subtarget->hasSSE41())
return SDValue();
}
-SDValue X86TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
-
- if (Op.getValueType().isVector()) return LowerVSETCC(Op, DAG);
-
- assert(Op.getValueType() == MVT::i8 && "SetCC type must be 8-bit integer");
- SDValue Op0 = Op.getOperand(0);
- SDValue Op1 = Op.getOperand(1);
- DebugLoc dl = Op.getDebugLoc();
- ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
-
- // Optimize to BT if possible.
- // Lower (X & (1 << N)) == 0 to BT(X, N).
- // Lower ((X >>u N) & 1) != 0 to BT(X, N).
- // Lower ((X >>s N) & 1) != 0 to BT(X, N).
- if (Op0.getOpcode() == ISD::AND && Op0.hasOneUse() &&
- Op1.getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(Op1)->isNullValue() &&
- (CC == ISD::SETEQ || CC == ISD::SETNE)) {
- SDValue NewSetCC = LowerToBT(Op0, CC, dl, DAG);
- if (NewSetCC.getNode())
- return NewSetCC;
- }
-
- // Look for X == 0, X == 1, X != 0, or X != 1. We can simplify some forms of
- // these.
- if (Op1.getOpcode() == ISD::Constant &&
- (cast<ConstantSDNode>(Op1)->getZExtValue() == 1 ||
- cast<ConstantSDNode>(Op1)->isNullValue()) &&
- (CC == ISD::SETEQ || CC == ISD::SETNE)) {
-
- // If the input is a setcc, then reuse the input setcc or use a new one with
- // the inverted condition.
- if (Op0.getOpcode() == X86ISD::SETCC) {
- X86::CondCode CCode = (X86::CondCode)Op0.getConstantOperandVal(0);
- bool Invert = (CC == ISD::SETNE) ^
- cast<ConstantSDNode>(Op1)->isNullValue();
- if (!Invert) return Op0;
-
- CCode = X86::GetOppositeBranchCondition(CCode);
- return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
- DAG.getConstant(CCode, MVT::i8), Op0.getOperand(1));
- }
- }
-
- bool isFP = Op1.getValueType().isFloatingPoint();
- unsigned X86CC = TranslateX86CC(CC, isFP, Op0, Op1, DAG);
- if (X86CC == X86::COND_INVALID)
- return SDValue();
-
- SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, DAG);
- EFLAGS = ConvertCmpIfNecessary(EFLAGS, DAG);
- return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
- DAG.getConstant(X86CC, MVT::i8), EFLAGS);
-}
-
// Lower256IntVSETCC - Break a VSETCC 256-bit integer VSETCC into two new 128
// ones, and then concatenate the result back.
static SDValue Lower256IntVSETCC(SDValue Op, SelectionDAG &DAG) {
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
assert(VT.is256BitVector() && Op.getOpcode() == ISD::SETCC &&
"Unsupported value type for operation");
SDValue RHS2 = Extract128BitVector(RHS, NumElems/2, DAG, dl);
// Issue the operation on the smaller types and concatenate the result back
- MVT EltVT = VT.getVectorElementType().getSimpleVT();
- EVT NewVT = MVT::getVectorVT(EltVT, NumElems/2);
+ MVT EltVT = VT.getVectorElementType();
+ MVT NewVT = MVT::getVectorVT(EltVT, NumElems/2);
return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT,
DAG.getNode(Op.getOpcode(), dl, NewVT, LHS1, RHS1, CC),
DAG.getNode(Op.getOpcode(), dl, NewVT, LHS2, RHS2, CC));
}
-SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
+static SDValue LowerVSETCC(SDValue Op, const X86Subtarget *Subtarget,
+ SelectionDAG &DAG) {
SDValue Cond;
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue CC = Op.getOperand(2);
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
- bool isFP = Op.getOperand(1).getValueType().isFloatingPoint();
+ bool isFP = Op.getOperand(1).getValueType().getSimpleVT().isFloatingPoint();
DebugLoc dl = Op.getDebugLoc();
if (isFP) {
#ifndef NDEBUG
- EVT EltVT = Op0.getValueType().getVectorElementType();
+ MVT EltVT = Op0.getValueType().getVectorElementType().getSimpleVT();
assert(EltVT == MVT::f32 || EltVT == MVT::f64);
#endif
return Result;
}
+SDValue X86TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
+
+ MVT VT = Op.getValueType().getSimpleVT();
+
+ if (VT.isVector()) return LowerVSETCC(Op, Subtarget, DAG);
+
+ assert(VT == MVT::i8 && "SetCC type must be 8-bit integer");
+ SDValue Op0 = Op.getOperand(0);
+ SDValue Op1 = Op.getOperand(1);
+ DebugLoc dl = Op.getDebugLoc();
+ ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
+
+ // Optimize to BT if possible.
+ // Lower (X & (1 << N)) == 0 to BT(X, N).
+ // Lower ((X >>u N) & 1) != 0 to BT(X, N).
+ // Lower ((X >>s N) & 1) != 0 to BT(X, N).
+ if (Op0.getOpcode() == ISD::AND && Op0.hasOneUse() &&
+ Op1.getOpcode() == ISD::Constant &&
+ cast<ConstantSDNode>(Op1)->isNullValue() &&
+ (CC == ISD::SETEQ || CC == ISD::SETNE)) {
+ SDValue NewSetCC = LowerToBT(Op0, CC, dl, DAG);
+ if (NewSetCC.getNode())
+ return NewSetCC;
+ }
+
+ // Look for X == 0, X == 1, X != 0, or X != 1. We can simplify some forms of
+ // these.
+ if (Op1.getOpcode() == ISD::Constant &&
+ (cast<ConstantSDNode>(Op1)->getZExtValue() == 1 ||
+ cast<ConstantSDNode>(Op1)->isNullValue()) &&
+ (CC == ISD::SETEQ || CC == ISD::SETNE)) {
+
+ // If the input is a setcc, then reuse the input setcc or use a new one with
+ // the inverted condition.
+ if (Op0.getOpcode() == X86ISD::SETCC) {
+ X86::CondCode CCode = (X86::CondCode)Op0.getConstantOperandVal(0);
+ bool Invert = (CC == ISD::SETNE) ^
+ cast<ConstantSDNode>(Op1)->isNullValue();
+ if (!Invert) return Op0;
+
+ CCode = X86::GetOppositeBranchCondition(CCode);
+ return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+ DAG.getConstant(CCode, MVT::i8), Op0.getOperand(1));
+ }
+ }
+
+ bool isFP = Op1.getValueType().getSimpleVT().isFloatingPoint();
+ unsigned X86CC = TranslateX86CC(CC, isFP, Op0, Op1, DAG);
+ if (X86CC == X86::COND_INVALID)
+ return SDValue();
+
+ SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, DAG);
+ EFLAGS = ConvertCmpIfNecessary(EFLAGS, DAG);
+ return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+ DAG.getConstant(X86CC, MVT::i8), EFLAGS);
+}
+
// isX86LogicalCmp - Return true if opcode is a X86 logical comparison.
static bool isX86LogicalCmp(SDValue Op) {
unsigned Opc = Op.getNode()->getOpcode();
SDValue Cmp = Cond.getOperand(1);
unsigned Opc = Cmp.getOpcode();
- EVT VT = Op.getValueType();
+ MVT VT = Op.getValueType().getSimpleVT();
bool IllegalFPCMov = false;
if (VT.isFloatingPoint() && !VT.isVector() &&
SDValue X86TargetLowering::LowerSIGN_EXTEND(SDValue Op,
SelectionDAG &DAG) const {
- EVT VT = Op->getValueType(0);
+ MVT VT = Op->getValueType(0).getSimpleVT();
SDValue In = Op->getOperand(0);
- EVT InVT = In.getValueType();
+ MVT InVT = In.getValueType().getSimpleVT();
DebugLoc dl = Op->getDebugLoc();
if ((VT != MVT::v4i64 || InVT != MVT::v4i32) &&
SDValue OpHi = DAG.getVectorShuffle(InVT, dl, In, Undef, &ShufMask2[0]);
- EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), VT.getScalarType(),
+ MVT HalfVT = MVT::getVectorVT(VT.getScalarType(),
VT.getVectorNumElements()/2);
OpLo = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpLo);
case ISD::SRL_PARTS: return LowerShiftParts(Op, DAG);
case ISD::SINT_TO_FP: return LowerSINT_TO_FP(Op, DAG);
case ISD::UINT_TO_FP: return LowerUINT_TO_FP(Op, DAG);
- case ISD::TRUNCATE: return lowerTRUNCATE(Op, DAG);
+ case ISD::TRUNCATE: return LowerTRUNCATE(Op, DAG);
case ISD::ZERO_EXTEND: return LowerZERO_EXTEND(Op, DAG);
case ISD::SIGN_EXTEND: return LowerSIGN_EXTEND(Op, DAG);
case ISD::ANY_EXTEND: return LowerANY_EXTEND(Op, DAG);
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
case ISD::FP_TO_UINT: return LowerFP_TO_UINT(Op, DAG);
- case ISD::FP_EXTEND: return lowerFP_EXTEND(Op, DAG);
+ case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
case ISD::FABS: return LowerFABS(Op, DAG);
case ISD::FNEG: return LowerFNEG(Op, DAG);
case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG);
static SDValue PerformSExtCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const X86Subtarget *Subtarget) {
- EVT VT = N->getValueType(0);
-
- if (!VT.isVector())
- return SDValue();
-
- SDValue In = N->getOperand(0);
- EVT InVT = In.getValueType();
- DebugLoc dl = N->getDebugLoc();
- unsigned ExtendedEltSize = VT.getVectorElementType().getSizeInBits();
-
- // Split SIGN_EXTEND operation to use vmovsx instruction when possible
- if (InVT == MVT::v8i8) {
- if (ExtendedEltSize > 16 && !Subtarget->hasInt256())
- In = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i16, In);
- if (ExtendedEltSize > 32)
- In = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i32, In);
- return DAG.getNode(ISD::SIGN_EXTEND, dl, VT, In);
- }
-
- if ((InVT == MVT::v4i8 || InVT == MVT::v4i16) &&
- ExtendedEltSize > 32 && !Subtarget->hasInt256()) {
- In = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, In);
- return DAG.getNode(ISD::SIGN_EXTEND, dl, VT, In);
- }
-
if (!DCI.isBeforeLegalizeOps())
return SDValue();
if (!Subtarget->hasFp256())
return SDValue();
- if (VT.is256BitVector()) {
+ EVT VT = N->getValueType(0);
+ if (VT.isVector() && VT.getSizeInBits() == 256) {
SDValue R = WidenMaskArithmetic(N, DAG, DCI, Subtarget);
if (R.getNode())
return R;