SDValue visitSDIVREM(SDNode *N);
SDValue visitUDIVREM(SDNode *N);
SDValue visitAND(SDNode *N);
+ SDValue visitANDLike(SDValue N0, SDValue N1, SDNode *LocReference);
SDValue visitOR(SDNode *N);
+ SDValue visitORLike(SDValue N0, SDValue N1, SDNode *LocReference);
SDValue visitXOR(SDNode *N);
SDValue SimplifyVBinOp(SDNode *N);
- SDValue SimplifyVUnaryOp(SDNode *N);
SDValue visitSHL(SDNode *N);
SDValue visitSRA(SDNode *N);
SDValue visitSRL(SDNode *N);
SDValue visitCONCAT_VECTORS(SDNode *N);
SDValue visitEXTRACT_SUBVECTOR(SDNode *N);
SDValue visitVECTOR_SHUFFLE(SDNode *N);
+ SDValue visitSCALAR_TO_VECTOR(SDNode *N);
SDValue visitINSERT_SUBVECTOR(SDNode *N);
SDValue visitMLOAD(SDNode *N);
SDValue visitMSTORE(SDNode *N);
+ SDValue visitFP_TO_FP16(SDNode *N);
+
+ SDValue visitFADDForFMACombine(SDNode *N);
+ SDValue visitFSUBForFMACombine(SDNode *N);
SDValue XformToShuffleWithZero(SDNode *N);
SDValue ReassociateOps(unsigned Opc, SDLoc DL, SDValue LHS, SDValue RHS);
bool MergeStoresOfConstantsOrVecElts(SmallVectorImpl<MemOpLink> &StoreNodes,
EVT MemVT, unsigned NumElem,
bool IsConstantSrc, bool UseVector);
-
+
/// Merge consecutive store operations into a wide store.
/// This optimization uses wide integers or vectors when possible.
/// \return True if some memory operations were changed.
EltVT.getSizeInBits() >= SplatBitSize);
}
-// \brief Returns the SDNode if it is a constant BuildVector or constant.
-static SDNode *isConstantBuildVectorOrConstantInt(SDValue N) {
+// \brief Returns the SDNode if it is a constant integer BuildVector
+// or constant integer.
+static SDNode *isConstantIntBuildVectorOrConstantInt(SDValue N) {
if (isa<ConstantSDNode>(N))
return N.getNode();
- BuildVectorSDNode *BV = dyn_cast<BuildVectorSDNode>(N);
- if (BV && BV->isConstant())
- return BV;
+ if (ISD::isBuildVectorOfConstantSDNodes(N.getNode()))
+ return N.getNode();
+ return nullptr;
+}
+
+// \brief Returns the SDNode if it is a constant float BuildVector
+// or constant float.
+static SDNode *isConstantFPBuildVectorOrConstantFP(SDValue N) {
+ if (isa<ConstantFPSDNode>(N))
+ return N.getNode();
+ if (ISD::isBuildVectorOfConstantFPSDNodes(N.getNode()))
+ return N.getNode();
return nullptr;
}
SDValue N0, SDValue N1) {
EVT VT = N0.getValueType();
if (N0.getOpcode() == Opc) {
- if (SDNode *L = isConstantBuildVectorOrConstantInt(N0.getOperand(1))) {
- if (SDNode *R = isConstantBuildVectorOrConstantInt(N1)) {
+ if (SDNode *L = isConstantIntBuildVectorOrConstantInt(N0.getOperand(1))) {
+ if (SDNode *R = isConstantIntBuildVectorOrConstantInt(N1)) {
// reassoc. (op (op x, c1), c2) -> (op x, (op c1, c2))
if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, VT, L, R))
return DAG.getNode(Opc, DL, VT, N0.getOperand(0), OpNode);
}
if (N1.getOpcode() == Opc) {
- if (SDNode *R = isConstantBuildVectorOrConstantInt(N1.getOperand(1))) {
- if (SDNode *L = isConstantBuildVectorOrConstantInt(N0)) {
+ if (SDNode *R = isConstantIntBuildVectorOrConstantInt(N1.getOperand(1))) {
+ if (SDNode *L = isConstantIntBuildVectorOrConstantInt(N0)) {
// reassoc. (op c2, (op x, c1)) -> (op x, (op c1, c2))
if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, VT, R, L))
return DAG.getNode(Opc, DL, VT, N1.getOperand(0), OpNode);
LegalOperations = Level >= AfterLegalizeVectorOps;
LegalTypes = Level >= AfterLegalizeTypes;
- // Early exit if this basic block is in an optnone function.
- if (DAG.getMachineFunction().getFunction()->hasFnAttribute(
- Attribute::OptimizeNone))
- return;
-
// Add all the dag nodes to the worklist.
for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
E = DAG.allnodes_end(); I != E; ++I)
case ISD::CONCAT_VECTORS: return visitCONCAT_VECTORS(N);
case ISD::EXTRACT_SUBVECTOR: return visitEXTRACT_SUBVECTOR(N);
case ISD::VECTOR_SHUFFLE: return visitVECTOR_SHUFFLE(N);
+ case ISD::SCALAR_TO_VECTOR: return visitSCALAR_TO_VECTOR(N);
case ISD::INSERT_SUBVECTOR: return visitINSERT_SUBVECTOR(N);
case ISD::MLOAD: return visitMLOAD(N);
case ISD::MSTORE: return visitMSTORE(N);
+ case ISD::FP_TO_FP16: return visitFP_TO_FP16(N);
}
return SDValue();
}
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (add x, 0) -> x, vector edition
if (ISD::isBuildVectorAllZeros(N1.getNode()))
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::ADD, VT, N0C, N1C);
// canonicalize constant to RHS
- if (N0C && !N1C)
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::ADD, SDLoc(N), VT, N1, N0);
// fold (add x, 0) -> x
if (N1C && N1C->isNullValue())
N0C->getAPIntValue(), VT),
N0.getOperand(1));
// reassociate add
- SDValue RADD = ReassociateOps(ISD::ADD, SDLoc(N), N0, N1);
- if (RADD.getNode())
+ if (SDValue RADD = ReassociateOps(ISD::ADD, SDLoc(N), N0, N1))
return RADD;
// fold ((0-A) + B) -> B-A
if (N0.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N0.getOperand(0)) &&
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (sub x, 0) -> x, vector edition
if (ISD::isBuildVectorAllZeros(N1.getNode()))
APInt ConstValue0, ConstValue1;
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
N0IsConst = isConstantSplatVector(N0.getNode(), ConstValue0);
N1IsConst = isConstantSplatVector(N1.getNode(), ConstValue1);
} else {
- N0IsConst = dyn_cast<ConstantSDNode>(N0) != nullptr;
- ConstValue0 = N0IsConst ? (dyn_cast<ConstantSDNode>(N0))->getAPIntValue()
- : APInt();
- N1IsConst = dyn_cast<ConstantSDNode>(N1) != nullptr;
- ConstValue1 = N1IsConst ? (dyn_cast<ConstantSDNode>(N1))->getAPIntValue()
- : APInt();
+ N0IsConst = isa<ConstantSDNode>(N0);
+ if (N0IsConst)
+ ConstValue0 = cast<ConstantSDNode>(N0)->getAPIntValue();
+ N1IsConst = isa<ConstantSDNode>(N1);
+ if (N1IsConst)
+ ConstValue1 = cast<ConstantSDNode>(N1)->getAPIntValue();
}
// fold (mul c1, c2) -> c1*c2
if (N0IsConst && N1IsConst)
return DAG.FoldConstantArithmetic(ISD::MUL, VT, N0.getNode(), N1.getNode());
- // canonicalize constant to RHS
- if (N0IsConst && !N1IsConst)
+ // canonicalize constant to RHS (vector doesn't have to splat)
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::MUL, SDLoc(N), VT, N1, N0);
// fold (mul x, 0) -> 0
if (N1IsConst && ConstValue1 == 0)
N0.getOperand(1), N1));
// reassociate mul
- SDValue RMUL = ReassociateOps(ISD::MUL, SDLoc(N), N0, N1);
- if (RMUL.getNode())
+ if (SDValue RMUL = ReassociateOps(ISD::MUL, SDLoc(N), N0, N1))
return RMUL;
return SDValue();
EVT VT = N->getValueType(0);
// fold vector ops
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (sdiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
return DAG.getNode(ISD::SUB, SDLoc(N), VT, DAG.getConstant(0, VT), SRA);
}
- // if integer divide is expensive and we satisfy the requirements, emit an
+ // If integer divide is expensive and we satisfy the requirements, emit an
// alternate sequence.
if (N1C && !TLI.isIntDivCheap()) {
SDValue Op = BuildSDIV(N);
EVT VT = N->getValueType(0);
// fold vector ops
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (udiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
EVT VT = N->getValueType(0);
SDLoc DL(N);
- // If the type twice as wide is legal, transform the mulhu to a wider multiply
- // plus a shift.
+ // If the type is twice as wide is legal, transform the mulhu to a wider
+ // multiply plus a shift.
if (VT.isSimple() && !VT.isVector()) {
MVT Simple = VT.getSimpleVT();
unsigned SimpleSize = Simple.getSizeInBits();
EVT VT = N->getValueType(0);
SDLoc DL(N);
- // If the type twice as wide is legal, transform the mulhu to a wider multiply
- // plus a shift.
+ // If the type is twice as wide is legal, transform the mulhu to a wider
+ // multiply plus a shift.
if (VT.isSimple() && !VT.isVector()) {
MVT Simple = VT.getSimpleVT();
unsigned SimpleSize = Simple.getSizeInBits();
return SDValue();
}
+/// This contains all DAGCombine rules which reduce two values combined by
+/// an And operation to a single value. This makes them reusable in the context
+/// of visitSELECT(). Rules involving constants are not included as
+/// visitSELECT() already handles those cases.
+SDValue DAGCombiner::visitANDLike(SDValue N0, SDValue N1,
+ SDNode *LocReference) {
+ EVT VT = N1.getValueType();
+
+ // fold (and x, undef) -> 0
+ if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
+ return DAG.getConstant(0, VT);
+ // fold (and (setcc x), (setcc y)) -> (setcc (and x, y))
+ SDValue LL, LR, RL, RR, CC0, CC1;
+ if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
+ ISD::CondCode Op0 = cast<CondCodeSDNode>(CC0)->get();
+ ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
+
+ if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
+ LL.getValueType().isInteger()) {
+ // fold (and (seteq X, 0), (seteq Y, 0)) -> (seteq (or X, Y), 0)
+ if (cast<ConstantSDNode>(LR)->isNullValue() && Op1 == ISD::SETEQ) {
+ SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ORNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
+ }
+ // fold (and (seteq X, -1), (seteq Y, -1)) -> (seteq (and X, Y), -1)
+ if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETEQ) {
+ SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(N0),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ANDNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ANDNode, LR, Op1);
+ }
+ // fold (and (setgt X, -1), (setgt Y, -1)) -> (setgt (or X, Y), -1)
+ if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETGT) {
+ SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ORNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
+ }
+ }
+ // Simplify (and (setne X, 0), (setne X, -1)) -> (setuge (add X, 1), 2)
+ if (LL == RL && isa<ConstantSDNode>(LR) && isa<ConstantSDNode>(RR) &&
+ Op0 == Op1 && LL.getValueType().isInteger() &&
+ Op0 == ISD::SETNE && ((cast<ConstantSDNode>(LR)->isNullValue() &&
+ cast<ConstantSDNode>(RR)->isAllOnesValue()) ||
+ (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
+ cast<ConstantSDNode>(RR)->isNullValue()))) {
+ SDValue ADDNode = DAG.getNode(ISD::ADD, SDLoc(N0), LL.getValueType(),
+ LL, DAG.getConstant(1, LL.getValueType()));
+ AddToWorklist(ADDNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ADDNode,
+ DAG.getConstant(2, LL.getValueType()), ISD::SETUGE);
+ }
+ // canonicalize equivalent to ll == rl
+ if (LL == RR && LR == RL) {
+ Op1 = ISD::getSetCCSwappedOperands(Op1);
+ std::swap(RL, RR);
+ }
+ if (LL == RL && LR == RR) {
+ bool isInteger = LL.getValueType().isInteger();
+ ISD::CondCode Result = ISD::getSetCCAndOperation(Op0, Op1, isInteger);
+ if (Result != ISD::SETCC_INVALID &&
+ (!LegalOperations ||
+ (TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
+ TLI.isOperationLegal(ISD::SETCC,
+ getSetCCResultType(N0.getSimpleValueType())))))
+ return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
+ LL, LR, Result);
+ }
+ }
+
+ if (N0.getOpcode() == ISD::ADD && N1.getOpcode() == ISD::SRL &&
+ VT.getSizeInBits() <= 64) {
+ if (ConstantSDNode *ADDI = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
+ APInt ADDC = ADDI->getAPIntValue();
+ if (!TLI.isLegalAddImmediate(ADDC.getSExtValue())) {
+ // Look for (and (add x, c1), (lshr y, c2)). If C1 wasn't a legal
+ // immediate for an add, but it is legal if its top c2 bits are set,
+ // transform the ADD so the immediate doesn't need to be materialized
+ // in a register.
+ if (ConstantSDNode *SRLI = dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
+ APInt Mask = APInt::getHighBitsSet(VT.getSizeInBits(),
+ SRLI->getZExtValue());
+ if (DAG.MaskedValueIsZero(N0.getOperand(1), Mask)) {
+ ADDC |= Mask;
+ if (TLI.isLegalAddImmediate(ADDC.getSExtValue())) {
+ SDValue NewAdd =
+ DAG.getNode(ISD::ADD, SDLoc(N0), VT,
+ N0.getOperand(0), DAG.getConstant(ADDC, VT));
+ CombineTo(N0.getNode(), NewAdd);
+ // Return N so it doesn't get rechecked!
+ return SDValue(LocReference, 0);
+ }
+ }
+ }
+ }
+ }
+ }
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitAND(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (and x, 0) -> 0, vector edition
if (ISD::isBuildVectorAllZeros(N0.getNode()))
return N0;
}
- // fold (and x, undef) -> 0
- if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
// fold (and c1, c2) -> c1&c2
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::AND, VT, N0C, N1C);
// canonicalize constant to RHS
- if (N0C && !N1C)
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::AND, SDLoc(N), VT, N1, N0);
// fold (and x, -1) -> x
if (N1C && N1C->isAllOnesValue())
APInt::getAllOnesValue(BitWidth)))
return DAG.getConstant(0, VT);
// reassociate and
- SDValue RAND = ReassociateOps(ISD::AND, SDLoc(N), N0, N1);
- if (RAND.getNode())
+ if (SDValue RAND = ReassociateOps(ISD::AND, SDLoc(N), N0, N1))
return RAND;
// fold (and (or x, C), D) -> D if (C & D) == D
if (N1C && N0.getOpcode() == ISD::OR)
SplatBitSize = SplatBitSize * 2)
SplatValue |= SplatValue.shl(SplatBitSize);
- Constant = APInt::getAllOnesValue(BitWidth);
- for (unsigned i = 0, n = SplatBitSize/BitWidth; i < n; ++i)
- Constant &= SplatValue.lshr(i*BitWidth).zextOrTrunc(BitWidth);
+ // Make sure that variable 'Constant' is only set if 'SplatBitSize' is a
+ // multiple of 'BitWidth'. Otherwise, we could propagate a wrong value.
+ if (SplatBitSize % BitWidth == 0) {
+ Constant = APInt::getAllOnesValue(BitWidth);
+ for (unsigned i = 0, n = SplatBitSize/BitWidth; i < n; ++i)
+ Constant &= SplatValue.lshr(i*BitWidth).zextOrTrunc(BitWidth);
+ }
}
}
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
}
- // fold (and (setcc x), (setcc y)) -> (setcc (and x, y))
- SDValue LL, LR, RL, RR, CC0, CC1;
- if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
- ISD::CondCode Op0 = cast<CondCodeSDNode>(CC0)->get();
- ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
-
- if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
- LL.getValueType().isInteger()) {
- // fold (and (seteq X, 0), (seteq Y, 0)) -> (seteq (or X, Y), 0)
- if (cast<ConstantSDNode>(LR)->isNullValue() && Op1 == ISD::SETEQ) {
- SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
- LR.getValueType(), LL, RL);
- AddToWorklist(ORNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ORNode, LR, Op1);
- }
- // fold (and (seteq X, -1), (seteq Y, -1)) -> (seteq (and X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETEQ) {
- SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(N0),
- LR.getValueType(), LL, RL);
- AddToWorklist(ANDNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ANDNode, LR, Op1);
- }
- // fold (and (setgt X, -1), (setgt Y, -1)) -> (setgt (or X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETGT) {
- SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
- LR.getValueType(), LL, RL);
- AddToWorklist(ORNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ORNode, LR, Op1);
- }
- }
- // Simplify (and (setne X, 0), (setne X, -1)) -> (setuge (add X, 1), 2)
- if (LL == RL && isa<ConstantSDNode>(LR) && isa<ConstantSDNode>(RR) &&
- Op0 == Op1 && LL.getValueType().isInteger() &&
- Op0 == ISD::SETNE && ((cast<ConstantSDNode>(LR)->isNullValue() &&
- cast<ConstantSDNode>(RR)->isAllOnesValue()) ||
- (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
- cast<ConstantSDNode>(RR)->isNullValue()))) {
- SDValue ADDNode = DAG.getNode(ISD::ADD, SDLoc(N0), LL.getValueType(),
- LL, DAG.getConstant(1, LL.getValueType()));
- AddToWorklist(ADDNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ADDNode,
- DAG.getConstant(2, LL.getValueType()), ISD::SETUGE);
- }
- // canonicalize equivalent to ll == rl
- if (LL == RR && LR == RL) {
- Op1 = ISD::getSetCCSwappedOperands(Op1);
- std::swap(RL, RR);
- }
- if (LL == RL && LR == RR) {
- bool isInteger = LL.getValueType().isInteger();
- ISD::CondCode Result = ISD::getSetCCAndOperation(Op0, Op1, isInteger);
- if (Result != ISD::SETCC_INVALID &&
- (!LegalOperations ||
- (TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
- TLI.isOperationLegal(ISD::SETCC,
- getSetCCResultType(N0.getSimpleValueType())))))
- return DAG.getSetCC(SDLoc(N), N0.getValueType(),
- LL, LR, Result);
- }
- }
-
- // Simplify: (and (op x...), (op y...)) -> (op (and x, y))
- if (N0.getOpcode() == N1.getOpcode()) {
- SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
- if (Tmp.getNode()) return Tmp;
- }
-
- // fold (and (sign_extend_inreg x, i16 to i32), 1) -> (and x, 1)
- // fold (and (sra)) -> (and (srl)) when possible.
- if (!VT.isVector() &&
- SimplifyDemandedBits(SDValue(N, 0)))
- return SDValue(N, 0);
-
- // fold (zext_inreg (extload x)) -> (zextload x)
- if (ISD::isEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode())) {
- LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- EVT MemVT = LN0->getMemoryVT();
- // If we zero all the possible extended bits, then we can turn this into
- // a zextload if we are running before legalize or the operation is legal.
- unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
- if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
- BitWidth - MemVT.getScalarType().getSizeInBits())) &&
- ((!LegalOperations && !LN0->isVolatile()) ||
- TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
- SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
- LN0->getChain(), LN0->getBasePtr(),
- MemVT, LN0->getMemOperand());
- AddToWorklist(N);
- CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
- return SDValue(N, 0); // Return N so it doesn't get rechecked!
- }
- }
- // fold (zext_inreg (sextload x)) -> (zextload x) iff load has one use
- if (ISD::isSEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) &&
- N0.hasOneUse()) {
- LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- EVT MemVT = LN0->getMemoryVT();
- // If we zero all the possible extended bits, then we can turn this into
- // a zextload if we are running before legalize or the operation is legal.
- unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
- if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
- BitWidth - MemVT.getScalarType().getSizeInBits())) &&
- ((!LegalOperations && !LN0->isVolatile()) ||
- TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
- SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
- LN0->getChain(), LN0->getBasePtr(),
- MemVT, LN0->getMemOperand());
- AddToWorklist(N);
- CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
- return SDValue(N, 0); // Return N so it doesn't get rechecked!
- }
- }
// fold (and (load x), 255) -> (zextload x, i8)
// fold (and (extload x, i16), 255) -> (zextload x, i8)
}
}
- if (N0.getOpcode() == ISD::ADD && N1.getOpcode() == ISD::SRL &&
- VT.getSizeInBits() <= 64) {
- if (ConstantSDNode *ADDI = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
- APInt ADDC = ADDI->getAPIntValue();
- if (!TLI.isLegalAddImmediate(ADDC.getSExtValue())) {
- // Look for (and (add x, c1), (lshr y, c2)). If C1 wasn't a legal
- // immediate for an add, but it is legal if its top c2 bits are set,
- // transform the ADD so the immediate doesn't need to be materialized
- // in a register.
- if (ConstantSDNode *SRLI = dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
- APInt Mask = APInt::getHighBitsSet(VT.getSizeInBits(),
- SRLI->getZExtValue());
- if (DAG.MaskedValueIsZero(N0.getOperand(1), Mask)) {
- ADDC |= Mask;
- if (TLI.isLegalAddImmediate(ADDC.getSExtValue())) {
- SDValue NewAdd =
- DAG.getNode(ISD::ADD, SDLoc(N0), VT,
- N0.getOperand(0), DAG.getConstant(ADDC, VT));
- CombineTo(N0.getNode(), NewAdd);
- return SDValue(N, 0); // Return N so it doesn't get rechecked!
- }
- }
- }
- }
+ if (SDValue Combined = visitANDLike(N0, N1, N))
+ return Combined;
+
+ // Simplify: (and (op x...), (op y...)) -> (op (and x, y))
+ if (N0.getOpcode() == N1.getOpcode()) {
+ SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
+ if (Tmp.getNode()) return Tmp;
+ }
+
+ // fold (and (sign_extend_inreg x, i16 to i32), 1) -> (and x, 1)
+ // fold (and (sra)) -> (and (srl)) when possible.
+ if (!VT.isVector() &&
+ SimplifyDemandedBits(SDValue(N, 0)))
+ return SDValue(N, 0);
+
+ // fold (zext_inreg (extload x)) -> (zextload x)
+ if (ISD::isEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode())) {
+ LoadSDNode *LN0 = cast<LoadSDNode>(N0);
+ EVT MemVT = LN0->getMemoryVT();
+ // If we zero all the possible extended bits, then we can turn this into
+ // a zextload if we are running before legalize or the operation is legal.
+ unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
+ if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
+ BitWidth - MemVT.getScalarType().getSizeInBits())) &&
+ ((!LegalOperations && !LN0->isVolatile()) ||
+ TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
+ SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
+ LN0->getChain(), LN0->getBasePtr(),
+ MemVT, LN0->getMemOperand());
+ AddToWorklist(N);
+ CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
+ return SDValue(N, 0); // Return N so it doesn't get rechecked!
+ }
+ }
+ // fold (zext_inreg (sextload x)) -> (zextload x) iff load has one use
+ if (ISD::isSEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) &&
+ N0.hasOneUse()) {
+ LoadSDNode *LN0 = cast<LoadSDNode>(N0);
+ EVT MemVT = LN0->getMemoryVT();
+ // If we zero all the possible extended bits, then we can turn this into
+ // a zextload if we are running before legalize or the operation is legal.
+ unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
+ if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
+ BitWidth - MemVT.getScalarType().getSizeInBits())) &&
+ ((!LegalOperations && !LN0->isVolatile()) ||
+ TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
+ SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
+ LN0->getChain(), LN0->getBasePtr(),
+ MemVT, LN0->getMemOperand());
+ AddToWorklist(N);
+ CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
+ return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
}
-
// fold (and (or (srl N, 8), (shl N, 8)), 0xffff) -> (srl (bswap N), const)
if (N1C && N1C->getAPIntValue() == 0xffff && N0.getOpcode() == ISD::OR) {
SDValue BSwap = MatchBSwapHWordLow(N0.getNode(), N0.getOperand(0),
DAG.getNode(ISD::SRL, SDLoc(N), VT, BSwap, ShAmt));
}
+/// This contains all DAGCombine rules which reduce two values combined by
+/// an Or operation to a single value \see visitANDLike().
+SDValue DAGCombiner::visitORLike(SDValue N0, SDValue N1, SDNode *LocReference) {
+ EVT VT = N1.getValueType();
+ // fold (or x, undef) -> -1
+ if (!LegalOperations &&
+ (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)) {
+ EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT;
+ return DAG.getConstant(APInt::getAllOnesValue(EltVT.getSizeInBits()), VT);
+ }
+ // fold (or (setcc x), (setcc y)) -> (setcc (or x, y))
+ SDValue LL, LR, RL, RR, CC0, CC1;
+ if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
+ ISD::CondCode Op0 = cast<CondCodeSDNode>(CC0)->get();
+ ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
+
+ if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
+ LL.getValueType().isInteger()) {
+ // fold (or (setne X, 0), (setne Y, 0)) -> (setne (or X, Y), 0)
+ // fold (or (setlt X, 0), (setlt Y, 0)) -> (setne (or X, Y), 0)
+ if (cast<ConstantSDNode>(LR)->isNullValue() &&
+ (Op1 == ISD::SETNE || Op1 == ISD::SETLT)) {
+ SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(LR),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ORNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
+ }
+ // fold (or (setne X, -1), (setne Y, -1)) -> (setne (and X, Y), -1)
+ // fold (or (setgt X, -1), (setgt Y -1)) -> (setgt (and X, Y), -1)
+ if (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
+ (Op1 == ISD::SETNE || Op1 == ISD::SETGT)) {
+ SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(LR),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ANDNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ANDNode, LR, Op1);
+ }
+ }
+ // canonicalize equivalent to ll == rl
+ if (LL == RR && LR == RL) {
+ Op1 = ISD::getSetCCSwappedOperands(Op1);
+ std::swap(RL, RR);
+ }
+ if (LL == RL && LR == RR) {
+ bool isInteger = LL.getValueType().isInteger();
+ ISD::CondCode Result = ISD::getSetCCOrOperation(Op0, Op1, isInteger);
+ if (Result != ISD::SETCC_INVALID &&
+ (!LegalOperations ||
+ (TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
+ TLI.isOperationLegal(ISD::SETCC,
+ getSetCCResultType(N0.getValueType())))))
+ return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
+ LL, LR, Result);
+ }
+ }
+
+ // (or (and X, C1), (and Y, C2)) -> (and (or X, Y), C3) if possible.
+ if (N0.getOpcode() == ISD::AND &&
+ N1.getOpcode() == ISD::AND &&
+ N0.getOperand(1).getOpcode() == ISD::Constant &&
+ N1.getOperand(1).getOpcode() == ISD::Constant &&
+ // Don't increase # computations.
+ (N0.getNode()->hasOneUse() || N1.getNode()->hasOneUse())) {
+ // We can only do this xform if we know that bits from X that are set in C2
+ // but not in C1 are already zero. Likewise for Y.
+ const APInt &LHSMask =
+ cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
+ const APInt &RHSMask =
+ cast<ConstantSDNode>(N1.getOperand(1))->getAPIntValue();
+
+ if (DAG.MaskedValueIsZero(N0.getOperand(0), RHSMask&~LHSMask) &&
+ DAG.MaskedValueIsZero(N1.getOperand(0), LHSMask&~RHSMask)) {
+ SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
+ N0.getOperand(0), N1.getOperand(0));
+ return DAG.getNode(ISD::AND, SDLoc(LocReference), VT, X,
+ DAG.getConstant(LHSMask | RHSMask, VT));
+ }
+ }
+
+ // (or (and X, M), (and X, N)) -> (and X, (or M, N))
+ if (N0.getOpcode() == ISD::AND &&
+ N1.getOpcode() == ISD::AND &&
+ N0.getOperand(0) == N1.getOperand(0) &&
+ // Don't increase # computations.
+ (N0.getNode()->hasOneUse() || N1.getNode()->hasOneUse())) {
+ SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
+ N0.getOperand(1), N1.getOperand(1));
+ return DAG.getNode(ISD::AND, SDLoc(LocReference), VT, N0.getOperand(0), X);
+ }
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitOR(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (or x, 0) -> x, vector edition
if (ISD::isBuildVectorAllZeros(N0.getNode()))
}
}
- // fold (or x, undef) -> -1
- if (!LegalOperations &&
- (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)) {
- EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT;
- return DAG.getConstant(APInt::getAllOnesValue(EltVT.getSizeInBits()), VT);
- }
// fold (or c1, c2) -> c1|c2
ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::OR, VT, N0C, N1C);
// canonicalize constant to RHS
- if (N0C && !N1C)
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::OR, SDLoc(N), VT, N1, N0);
// fold (or x, 0) -> x
if (N1C && N1C->isNullValue())
if (N1C && DAG.MaskedValueIsZero(N0, ~N1C->getAPIntValue()))
return N1;
+ if (SDValue Combined = visitORLike(N0, N1, N))
+ return Combined;
+
// Recognize halfword bswaps as (bswap + rotl 16) or (bswap + shl 16)
SDValue BSwap = MatchBSwapHWord(N, N0, N1);
if (BSwap.getNode())
return BSwap;
// reassociate or
- SDValue ROR = ReassociateOps(ISD::OR, SDLoc(N), N0, N1);
- if (ROR.getNode())
+ if (SDValue ROR = ReassociateOps(ISD::OR, SDLoc(N), N0, N1))
return ROR;
// Canonicalize (or (and X, c1), c2) -> (and (or X, c2), c1|c2)
// iff (c1 & c2) == 0.
return SDValue();
}
}
- // fold (or (setcc x), (setcc y)) -> (setcc (or x, y))
- SDValue LL, LR, RL, RR, CC0, CC1;
- if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
- ISD::CondCode Op0 = cast<CondCodeSDNode>(CC0)->get();
- ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
-
- if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
- LL.getValueType().isInteger()) {
- // fold (or (setne X, 0), (setne Y, 0)) -> (setne (or X, Y), 0)
- // fold (or (setlt X, 0), (setlt Y, 0)) -> (setne (or X, Y), 0)
- if (cast<ConstantSDNode>(LR)->isNullValue() &&
- (Op1 == ISD::SETNE || Op1 == ISD::SETLT)) {
- SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(LR),
- LR.getValueType(), LL, RL);
- AddToWorklist(ORNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ORNode, LR, Op1);
- }
- // fold (or (setne X, -1), (setne Y, -1)) -> (setne (and X, Y), -1)
- // fold (or (setgt X, -1), (setgt Y -1)) -> (setgt (and X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
- (Op1 == ISD::SETNE || Op1 == ISD::SETGT)) {
- SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(LR),
- LR.getValueType(), LL, RL);
- AddToWorklist(ANDNode.getNode());
- return DAG.getSetCC(SDLoc(N), VT, ANDNode, LR, Op1);
- }
- }
- // canonicalize equivalent to ll == rl
- if (LL == RR && LR == RL) {
- Op1 = ISD::getSetCCSwappedOperands(Op1);
- std::swap(RL, RR);
- }
- if (LL == RL && LR == RR) {
- bool isInteger = LL.getValueType().isInteger();
- ISD::CondCode Result = ISD::getSetCCOrOperation(Op0, Op1, isInteger);
- if (Result != ISD::SETCC_INVALID &&
- (!LegalOperations ||
- (TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
- TLI.isOperationLegal(ISD::SETCC,
- getSetCCResultType(N0.getValueType())))))
- return DAG.getSetCC(SDLoc(N), N0.getValueType(),
- LL, LR, Result);
- }
- }
-
// Simplify: (or (op x...), (op y...)) -> (op (or x, y))
if (N0.getOpcode() == N1.getOpcode()) {
SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
if (Tmp.getNode()) return Tmp;
}
- // (or (and X, C1), (and Y, C2)) -> (and (or X, Y), C3) if possible.
- if (N0.getOpcode() == ISD::AND &&
- N1.getOpcode() == ISD::AND &&
- N0.getOperand(1).getOpcode() == ISD::Constant &&
- N1.getOperand(1).getOpcode() == ISD::Constant &&
- // Don't increase # computations.
- (N0.getNode()->hasOneUse() || N1.getNode()->hasOneUse())) {
- // We can only do this xform if we know that bits from X that are set in C2
- // but not in C1 are already zero. Likewise for Y.
- const APInt &LHSMask =
- cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
- const APInt &RHSMask =
- cast<ConstantSDNode>(N1.getOperand(1))->getAPIntValue();
-
- if (DAG.MaskedValueIsZero(N0.getOperand(0), RHSMask&~LHSMask) &&
- DAG.MaskedValueIsZero(N1.getOperand(0), LHSMask&~RHSMask)) {
- SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
- N0.getOperand(0), N1.getOperand(0));
- return DAG.getNode(ISD::AND, SDLoc(N), VT, X,
- DAG.getConstant(LHSMask | RHSMask, VT));
- }
- }
-
- // (or (and X, M), (and X, N)) -> (and X, (or M, N))
- if (N0.getOpcode() == ISD::AND &&
- N1.getOpcode() == ISD::AND &&
- N0.getOperand(0) == N1.getOperand(0) &&
- // Don't increase # computations.
- (N0.getNode()->hasOneUse() || N1.getNode()->hasOneUse())) {
- SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
- N0.getOperand(1), N1.getOperand(1));
- return DAG.getNode(ISD::AND, SDLoc(N), VT, N0.getOperand(0), X);
- }
-
// See if this is some rotate idiom.
if (SDNode *Rot = MatchRotate(N0, N1, SDLoc(N)))
return SDValue(Rot, 0);
// fold vector ops
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (xor x, 0) -> x, vector edition
if (ISD::isBuildVectorAllZeros(N0.getNode()))
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::XOR, VT, N0C, N1C);
// canonicalize constant to RHS
- if (N0C && !N1C)
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0);
// fold (xor x, 0) -> x
if (N1C && N1C->isNullValue())
return N0;
// reassociate xor
- SDValue RXOR = ReassociateOps(ISD::XOR, SDLoc(N), N0, N1);
- if (RXOR.getNode())
+ if (SDValue RXOR = ReassociateOps(ISD::XOR, SDLoc(N), N0, N1))
return RXOR;
// fold !(x cc y) -> (x !cc y)
if (N0 == N1)
return tryFoldToZero(SDLoc(N), TLI, VT, DAG, LegalOperations, LegalTypes);
+ // fold (xor (shl 1, x), -1) -> (rotl ~1, x)
+ // Here is a concrete example of this equivalence:
+ // i16 x == 14
+ // i16 shl == 1 << 14 == 16384 == 0b0100000000000000
+ // i16 xor == ~(1 << 14) == 49151 == 0b1011111111111111
+ //
+ // =>
+ //
+ // i16 ~1 == 0b1111111111111110
+ // i16 rol(~1, 14) == 0b1011111111111111
+ //
+ // Some additional tips to help conceptualize this transform:
+ // - Try to see the operation as placing a single zero in a value of all ones.
+ // - There exists no value for x which would allow the result to contain zero.
+ // - Values of x larger than the bitwidth are undefined and do not require a
+ // consistent result.
+ // - Pushing the zero left requires shifting one bits in from the right.
+ // A rotate left of ~1 is a nice way of achieving the desired result.
+ if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT))
+ if (auto *N1C = dyn_cast<ConstantSDNode>(N1.getNode()))
+ if (N0.getOpcode() == ISD::SHL)
+ if (auto *ShlLHS = dyn_cast<ConstantSDNode>(N0.getOperand(0)))
+ if (N1C->isAllOnesValue() && ShlLHS->isOne())
+ return DAG.getNode(ISD::ROTL, SDLoc(N), VT, DAG.getConstant(~1, VT),
+ N0.getOperand(1));
+
// Simplify: xor (op x...), (op y...) -> (op (xor x, y))
if (N0.getOpcode() == N1.getOpcode()) {
SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
// fold vector ops
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
BuildVectorSDNode *N1CV = dyn_cast<BuildVectorSDNode>(N1);
// If setcc produces all-one true value then:
// fold vector ops
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
N1C = isConstOrConstSplat(N1);
}
// fold vector ops
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
N1C = isConstOrConstSplat(N1);
}
return SimplifySelect(SDLoc(N), N0, N1, N2);
}
+ if (VT0 == MVT::i1) {
+ if (TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
+ // select (and Cond0, Cond1), X, Y
+ // -> select Cond0, (select Cond1, X, Y), Y
+ if (N0->getOpcode() == ISD::AND && N0->hasOneUse()) {
+ SDValue Cond0 = N0->getOperand(0);
+ SDValue Cond1 = N0->getOperand(1);
+ SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
+ N1.getValueType(), Cond1, N1, N2);
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Cond0,
+ InnerSelect, N2);
+ }
+ // select (or Cond0, Cond1), X, Y -> select Cond0, X, (select Cond1, X, Y)
+ if (N0->getOpcode() == ISD::OR && N0->hasOneUse()) {
+ SDValue Cond0 = N0->getOperand(0);
+ SDValue Cond1 = N0->getOperand(1);
+ SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
+ N1.getValueType(), Cond1, N1, N2);
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Cond0, N1,
+ InnerSelect);
+ }
+ }
+
+ // select Cond0, (select Cond1, X, Y), Y -> select (and Cond0, Cond1), X, Y
+ if (N1->getOpcode() == ISD::SELECT) {
+ SDValue N1_0 = N1->getOperand(0);
+ SDValue N1_1 = N1->getOperand(1);
+ SDValue N1_2 = N1->getOperand(2);
+ if (N1_2 == N2 && N0.getValueType() == N1_0.getValueType()) {
+ // Create the actual and node if we can generate good code for it.
+ if (!TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
+ SDValue And = DAG.getNode(ISD::AND, SDLoc(N), N0.getValueType(),
+ N0, N1_0);
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), And,
+ N1_1, N2);
+ }
+ // Otherwise see if we can optimize the "and" to a better pattern.
+ if (SDValue Combined = visitANDLike(N0, N1_0, N))
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Combined,
+ N1_1, N2);
+ }
+ }
+ // select Cond0, X, (select Cond1, X, Y) -> select (or Cond0, Cond1), X, Y
+ if (N2->getOpcode() == ISD::SELECT) {
+ SDValue N2_0 = N2->getOperand(0);
+ SDValue N2_1 = N2->getOperand(1);
+ SDValue N2_2 = N2->getOperand(2);
+ if (N2_1 == N1 && N0.getValueType() == N2_0.getValueType()) {
+ // Create the actual or node if we can generate good code for it.
+ if (!TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
+ SDValue Or = DAG.getNode(ISD::OR, SDLoc(N), N0.getValueType(),
+ N0, N2_0);
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Or,
+ N1, N2_2);
+ }
+ // Otherwise see if we can optimize to a better pattern.
+ if (SDValue Combined = visitORLike(N0, N2_0, N))
+ return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Combined,
+ N1, N2_2);
+ }
+ }
+ }
+
return SDValue();
}
std::tie(DataLo, DataHi) = DAG.SplitVector(Data, DL);
MachineMemOperand *MMO = DAG.getMachineFunction().
- getMachineMemOperand(MST->getPointerInfo(),
+ getMachineMemOperand(MST->getPointerInfo(),
MachineMemOperand::MOStore, LoMemVT.getStoreSize(),
Alignment, MST->getAAInfo(), MST->getRanges());
DAG.getConstant(IncrementSize, Ptr.getValueType()));
MMO = DAG.getMachineFunction().
- getMachineMemOperand(MST->getPointerInfo(),
+ getMachineMemOperand(MST->getPointerInfo(),
MachineMemOperand::MOStore, HiMemVT.getStoreSize(),
SecondHalfAlignment, MST->getAAInfo(),
MST->getRanges());
std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
MachineMemOperand *MMO = DAG.getMachineFunction().
- getMachineMemOperand(MLD->getPointerInfo(),
+ getMachineMemOperand(MLD->getPointerInfo(),
MachineMemOperand::MOLoad, LoMemVT.getStoreSize(),
Alignment, MLD->getAAInfo(), MLD->getRanges());
DAG.getConstant(IncrementSize, Ptr.getValueType()));
MMO = DAG.getMachineFunction().
- getMachineMemOperand(MLD->getPointerInfo(),
+ getMachineMemOperand(MLD->getPointerInfo(),
MachineMemOperand::MOLoad, HiMemVT.getStoreSize(),
SecondHalfAlignment, MLD->getAAInfo(), MLD->getRanges());
}
}
+ if (SimplifySelectOps(N, N1, N2))
+ return SDValue(N, 0); // Don't revisit N.
+
// If the VSELECT result requires splitting and the mask is provided by a
// SETCC, then split both nodes and its operands before legalization. This
// prevents the type legalizer from unrolling SETCC into scalar comparisons
if (N0.getValueType() == N->getValueType(0))
return N0;
// fold (truncate c1) -> c1
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, N0);
// fold (truncate (truncate x)) -> (truncate x)
if (N0.getOpcode() == ISD::TRUNCATE)
return CombineLD;
}
+ // Remove double bitcasts from shuffles - this is often a legacy of
+ // XformToShuffleWithZero being used to combine bitmaskings (of
+ // float vectors bitcast to integer vectors) into shuffles.
+ // bitcast(shuffle(bitcast(s0),bitcast(s1))) -> shuffle(s0,s1)
+ if (Level < AfterLegalizeDAG && TLI.isTypeLegal(VT) && VT.isVector() &&
+ N0->getOpcode() == ISD::VECTOR_SHUFFLE &&
+ VT.getVectorNumElements() >= N0.getValueType().getVectorNumElements() &&
+ !(VT.getVectorNumElements() % N0.getValueType().getVectorNumElements())) {
+ ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N0);
+
+ // If operands are a bitcast, peek through if it casts the original VT.
+ // If operands are a UNDEF or constant, just bitcast back to original VT.
+ auto PeekThroughBitcast = [&](SDValue Op) {
+ if (Op.getOpcode() == ISD::BITCAST &&
+ Op.getOperand(0)->getValueType(0) == VT)
+ return SDValue(Op.getOperand(0));
+ if (ISD::isBuildVectorOfConstantSDNodes(Op.getNode()) ||
+ ISD::isBuildVectorOfConstantFPSDNodes(Op.getNode()))
+ return DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Op);
+ return SDValue();
+ };
+
+ SDValue SV0 = PeekThroughBitcast(N0->getOperand(0));
+ SDValue SV1 = PeekThroughBitcast(N0->getOperand(1));
+ if (!(SV0 && SV1))
+ return SDValue();
+
+ int MaskScale =
+ VT.getVectorNumElements() / N0.getValueType().getVectorNumElements();
+ SmallVector<int, 8> NewMask;
+ for (int M : SVN->getMask())
+ for (int i = 0; i != MaskScale; ++i)
+ NewMask.push_back(M < 0 ? -1 : M * MaskScale + i);
+
+ bool LegalMask = TLI.isShuffleMaskLegal(NewMask, VT);
+ if (!LegalMask) {
+ std::swap(SV0, SV1);
+ ShuffleVectorSDNode::commuteMask(NewMask);
+ LegalMask = TLI.isShuffleMaskLegal(NewMask, VT);
+ }
+
+ if (LegalMask)
+ return DAG.getVectorShuffle(VT, SDLoc(N), SV0, SV1, NewMask);
+ }
+
return SDValue();
}
// Finally, this must be the case where we are shrinking elements: each input
// turns into multiple outputs.
- bool isS2V = ISD::isScalarToVector(BV);
unsigned NumOutputsPerInput = SrcBitSize/DstBitSize;
EVT VT = EVT::getVectorVT(*DAG.getContext(), DstEltVT,
NumOutputsPerInput*BV->getNumOperands());
for (unsigned j = 0; j != NumOutputsPerInput; ++j) {
APInt ThisVal = OpVal.trunc(DstBitSize);
Ops.push_back(DAG.getConstant(ThisVal, DstEltVT));
- if (isS2V && i == 0 && j == 0 && ThisVal.zext(SrcBitSize) == OpVal)
- // Simply turn this into a SCALAR_TO_VECTOR of the new type.
- return DAG.getNode(ISD::SCALAR_TO_VECTOR, SDLoc(BV), VT,
- Ops[0]);
OpVal = OpVal.lshr(DstBitSize);
}
return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(BV), VT, Ops);
}
-// Attempt different variants of (fadd (fmul a, b), c) -> fma or fmad
-static SDValue performFaddFmulCombines(unsigned FusedOpcode,
- bool Aggressive,
- SDNode *N,
- const TargetLowering &TLI,
- SelectionDAG &DAG) {
+/// Try to perform FMA combining on a given FADD node.
+SDValue DAGCombiner::visitFADDForFMACombine(SDNode *N) {
+
+
+
+
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
+ SDLoc SL(N);
+
+ const TargetOptions &Options = DAG.getTarget().Options;
+ bool UnsafeFPMath = (Options.AllowFPOpFusion == FPOpFusion::Fast ||
+ Options.UnsafeFPMath);
+
+ // Floating-point multiply-add with intermediate rounding.
+ bool HasFMAD = (LegalOperations &&
+ TLI.isOperationLegal(ISD::FMAD, VT));
+
+ // Floating-point multiply-add without intermediate rounding.
+ bool HasFMA = ((!LegalOperations ||
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) &&
+ TLI.isFMAFasterThanFMulAndFAdd(VT) &&
+ UnsafeFPMath);
+
+ // No valid opcode, do not combine.
+ if (!HasFMAD && !HasFMA)
+ return SDValue();
+
+ // Always prefer FMAD to FMA for precision.
+ unsigned int PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
+ bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
+ bool LookThroughFPExt = TLI.isFPExtFree(VT);
// fold (fadd (fmul x, y), z) -> (fma x, y, z)
if (N0.getOpcode() == ISD::FMUL &&
(Aggressive || N0->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1), N1);
}
// Note: Commutes FADD operands.
if (N1.getOpcode() == ISD::FMUL &&
(Aggressive || N1->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(0), N1.getOperand(1), N0);
}
+ // Look through FP_EXTEND nodes to do more combining.
+ if (UnsafeFPMath && LookThroughFPExt) {
+ // fold (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z)
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)), N1);
+ }
+
+ // fold (fadd x, (fpext (fmul y, z))) -> (fma (fpext y), (fpext z), x)
+ // Note: Commutes FADD operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(1)), N0);
+ }
+ }
+
// More folding opportunities when target permits.
- if (Aggressive) {
+ if ((UnsafeFPMath || HasFMAD) && Aggressive) {
// fold (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y (fma u, v, z))
- if (N0.getOpcode() == ISD::FMA &&
+ if (N0.getOpcode() == PreferredFusedOpcode &&
N0.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(2).getOperand(0),
N0.getOperand(2).getOperand(1),
N1));
}
// fold (fadd x, (fma y, z, (fmul u, v)) -> (fma y, z (fma u, v, x))
- if (N1->getOpcode() == ISD::FMA &&
+ if (N1->getOpcode() == PreferredFusedOpcode &&
N1.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(0), N1.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(2).getOperand(0),
N1.getOperand(2).getOperand(1),
N0));
}
+
+ if (UnsafeFPMath && LookThroughFPExt) {
+ // fold (fadd (fma x, y, (fpext (fmul u, v))), z)
+ // -> (fma x, y, (fma (fpext u), (fpext v), z))
+ auto FoldFAddFMAFPExtFMul = [&] (
+ SDValue X, SDValue Y, SDValue U, SDValue V, SDValue Z) {
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X, Y,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, U),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, V),
+ Z));
+ };
+ if (N0.getOpcode() == PreferredFusedOpcode) {
+ SDValue N02 = N0.getOperand(2);
+ if (N02.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N020 = N02.getOperand(0);
+ if (N020.getOpcode() == ISD::FMUL)
+ return FoldFAddFMAFPExtFMul(N0.getOperand(0), N0.getOperand(1),
+ N020.getOperand(0), N020.getOperand(1),
+ N1);
+ }
+ }
+
+ // fold (fadd (fpext (fma x, y, (fmul u, v))), z)
+ // -> (fma (fpext x), (fpext y), (fma (fpext u), (fpext v), z))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ auto FoldFAddFPExtFMAFMul = [&] (
+ SDValue X, SDValue Y, SDValue U, SDValue V, SDValue Z) {
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, X),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, Y),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, U),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, V),
+ Z));
+ };
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == PreferredFusedOpcode) {
+ SDValue N002 = N00.getOperand(2);
+ if (N002.getOpcode() == ISD::FMUL)
+ return FoldFAddFPExtFMAFMul(N00.getOperand(0), N00.getOperand(1),
+ N002.getOperand(0), N002.getOperand(1),
+ N1);
+ }
+ }
+
+ // fold (fadd x, (fma y, z, (fpext (fmul u, v)))
+ // -> (fma y, z, (fma (fpext u), (fpext v), x))
+ if (N1.getOpcode() == PreferredFusedOpcode) {
+ SDValue N12 = N1.getOperand(2);
+ if (N12.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N120 = N12.getOperand(0);
+ if (N120.getOpcode() == ISD::FMUL)
+ return FoldFAddFMAFPExtFMul(N1.getOperand(0), N1.getOperand(1),
+ N120.getOperand(0), N120.getOperand(1),
+ N0);
+ }
+ }
+
+ // fold (fadd x, (fpext (fma y, z, (fmul u, v)))
+ // -> (fma (fpext y), (fpext z), (fma (fpext u), (fpext v), x))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == PreferredFusedOpcode) {
+ SDValue N102 = N10.getOperand(2);
+ if (N102.getOpcode() == ISD::FMUL)
+ return FoldFAddFPExtFMAFMul(N10.getOperand(0), N10.getOperand(1),
+ N102.getOperand(0), N102.getOperand(1),
+ N0);
+ }
+ }
+ }
}
return SDValue();
}
-static SDValue performFsubFmulCombines(unsigned FusedOpcode,
- bool Aggressive,
- SDNode *N,
- const TargetLowering &TLI,
- SelectionDAG &DAG) {
+/// Try to perform FMA combining on a given FSUB node.
+SDValue DAGCombiner::visitFSUBForFMACombine(SDNode *N) {
+
+
+
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
SDLoc SL(N);
+ const TargetOptions &Options = DAG.getTarget().Options;
+ bool UnsafeFPMath = (Options.AllowFPOpFusion == FPOpFusion::Fast ||
+ Options.UnsafeFPMath);
+
+ // Floating-point multiply-add with intermediate rounding.
+ bool HasFMAD = (LegalOperations &&
+ TLI.isOperationLegal(ISD::FMAD, VT));
+
+ // Floating-point multiply-add without intermediate rounding.
+ bool HasFMA = ((!LegalOperations ||
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) &&
+ TLI.isFMAFasterThanFMulAndFAdd(VT) &&
+ UnsafeFPMath);
+
+ // No valid opcode, do not combine.
+ if (!HasFMAD && !HasFMA)
+ return SDValue();
+
+ // Always prefer FMAD to FMA for precision.
+ unsigned int PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
+ bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
+ bool LookThroughFPExt = TLI.isFPExtFree(VT);
+
// fold (fsub (fmul x, y), z) -> (fma x, y, (fneg z))
if (N0.getOpcode() == ISD::FMUL &&
(Aggressive || N0->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
DAG.getNode(ISD::FNEG, SL, VT, N1));
}
// Note: Commutes FSUB operands.
if (N1.getOpcode() == ISD::FMUL &&
(Aggressive || N1->hasOneUse()))
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
DAG.getNode(ISD::FNEG, SL, VT,
N1.getOperand(0)),
N1.getOperand(1), N0);
(Aggressive || (N0->hasOneUse() && N0.getOperand(0).hasOneUse()))) {
SDValue N00 = N0.getOperand(0).getOperand(0);
SDValue N01 = N0.getOperand(0).getOperand(1);
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
DAG.getNode(ISD::FNEG, SL, VT, N00), N01,
DAG.getNode(ISD::FNEG, SL, VT, N1));
}
+ // Look through FP_EXTEND nodes to do more combining.
+ if (UnsafeFPMath && LookThroughFPExt) {
+ // fold (fsub (fpext (fmul x, y)), z)
+ // -> (fma (fpext x), (fpext y), (fneg z))
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT, N1));
+ }
+
+ // fold (fsub x, (fpext (fmul y, z)))
+ // -> (fma (fneg (fpext y)), (fpext z), x)
+ // Note: Commutes FSUB operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(0))),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(1)),
+ N0);
+ }
+
+ // fold (fsub (fpext (fneg (fmul, x, y))), z)
+ // -> (fneg (fma (fpext x), (fpext y), z))
+ // Note: This could be removed with appropriate canonicalization of the
+ // input expression into (fneg (fadd (fpext (fmul, x, y)), z). However, the
+ // orthogonal flags -fp-contract=fast and -enable-unsafe-fp-math prevent
+ // from implementing the canonicalization in visitFSUB.
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FNEG) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(1)),
+ N1));
+ }
+ }
+ }
+
+ // fold (fsub (fneg (fpext (fmul, x, y))), z)
+ // -> (fneg (fma (fpext x)), (fpext y), z)
+ // Note: This could be removed with appropriate canonicalization of the
+ // input expression into (fneg (fadd (fpext (fmul, x, y)), z). However, the
+ // orthogonal flags -fp-contract=fast and -enable-unsafe-fp-math prevent
+ // from implementing the canonicalization in visitFSUB.
+ if (N0.getOpcode() == ISD::FNEG) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(1)),
+ N1));
+ }
+ }
+ }
+
+ }
+
// More folding opportunities when target permits.
- if (Aggressive) {
+ if ((UnsafeFPMath || HasFMAD) && Aggressive) {
// fold (fsub (fma x, y, (fmul u, v)), z)
// -> (fma x, y (fma u, v, (fneg z)))
- if (N0.getOpcode() == FusedOpcode &&
+ if (N0.getOpcode() == PreferredFusedOpcode &&
N0.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(2).getOperand(0),
N0.getOperand(2).getOperand(1),
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
N1)));
}
// fold (fsub x, (fma y, z, (fmul u, v)))
// -> (fma (fneg y), z, (fma (fneg u), v, x))
- if (N1.getOpcode() == FusedOpcode &&
+ if (N1.getOpcode() == PreferredFusedOpcode &&
N1.getOperand(2).getOpcode() == ISD::FMUL) {
SDValue N20 = N1.getOperand(2).getOperand(0);
SDValue N21 = N1.getOperand(2).getOperand(1);
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
N1.getOperand(0)),
N1.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
- N20),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, N20),
+
N21, N0));
}
+
+ if (UnsafeFPMath && LookThroughFPExt) {
+ // fold (fsub (fma x, y, (fpext (fmul u, v))), z)
+ // -> (fma x, y (fma (fpext u), (fpext v), (fneg z)))
+ if (N0.getOpcode() == PreferredFusedOpcode) {
+ SDValue N02 = N0.getOperand(2);
+ if (N02.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N020 = N02.getOperand(0);
+ if (N020.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ N0.getOperand(0), N0.getOperand(1),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N020.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N020.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT,
+ N1)));
+ }
+ }
+
+ // fold (fsub (fpext (fma x, y, (fmul u, v))), z)
+ // -> (fma (fpext x), (fpext y),
+ // (fma (fpext u), (fpext v), (fneg z)))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == PreferredFusedOpcode) {
+ SDValue N002 = N00.getOperand(2);
+ if (N002.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N002.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N002.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT,
+ N1)));
+ }
+ }
+
+ // fold (fsub x, (fma y, z, (fpext (fmul u, v))))
+ // -> (fma (fneg y), z, (fma (fneg (fpext u)), (fpext v), x))
+ if (N1.getOpcode() == PreferredFusedOpcode &&
+ N1.getOperand(2).getOpcode() == ISD::FP_EXTEND) {
+ SDValue N120 = N1.getOperand(2).getOperand(0);
+ if (N120.getOpcode() == ISD::FMUL) {
+ SDValue N1200 = N120.getOperand(0);
+ SDValue N1201 = N120.getOperand(1);
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, N1.getOperand(0)),
+ N1.getOperand(1),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL,
+ VT, N1200)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N1201),
+ N0));
+ }
+ }
+
+ // fold (fsub x, (fpext (fma y, z, (fmul u, v))))
+ // -> (fma (fneg (fpext y)), (fpext z),
+ // (fma (fneg (fpext u)), (fpext v), x))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N1.getOpcode() == ISD::FP_EXTEND &&
+ N1.getOperand(0).getOpcode() == PreferredFusedOpcode) {
+ SDValue N100 = N1.getOperand(0).getOperand(0);
+ SDValue N101 = N1.getOperand(0).getOperand(1);
+ SDValue N102 = N1.getOperand(0).getOperand(2);
+ if (N102.getOpcode() == ISD::FMUL) {
+ SDValue N1020 = N102.getOperand(0);
+ SDValue N1021 = N102.getOperand(1);
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N100)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, N101),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL,
+ VT, N1020)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N1021),
+ N0));
+ }
+ }
+ }
}
return SDValue();
const TargetOptions &Options = DAG.getTarget().Options;
// fold vector ops
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (fadd c1, c2) -> c1 + c2
if (N0CFP && N1CFP)
}
} // enable-unsafe-fp-math
- if (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Assume if there is an fmad instruction that it should be aggressively
- // used.
- if (SDValue Fused = performFaddFmulCombines(ISD::FMAD, true, N, TLI, DAG))
- return Fused;
- }
-
// FADD -> FMA combines:
- if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) &&
- TLI.isFMAFasterThanFMulAndFAdd(VT) &&
- (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) {
-
- if (!TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Don't form FMA if we are preferring FMAD.
- if (SDValue Fused
- = performFaddFmulCombines(ISD::FMA,
- TLI.enableAggressiveFMAFusion(VT),
- N, TLI, DAG)) {
- return Fused;
- }
- }
-
- // When FP_EXTEND nodes are free on the target, and there is an opportunity
- // to combine into FMA, arrange such nodes accordingly.
- if (TLI.isFPExtFree(VT)) {
-
- // fold (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z)
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(1)), N1);
- }
-
- // fold (fadd x, (fpext (fmul y, z)), z) -> (fma (fpext y), (fpext z), x)
- // Note: Commutes FADD operands.
- if (N1.getOpcode() == ISD::FP_EXTEND) {
- SDValue N10 = N1.getOperand(0);
- if (N10.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(1)), N0);
- }
- }
+ SDValue Fused = visitFADDForFMACombine(N);
+ if (Fused) {
+ AddToWorklist(Fused.getNode());
+ return Fused;
}
return SDValue();
const TargetOptions &Options = DAG.getTarget().Options;
// fold vector ops
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (fsub c1, c2) -> c1-c2
if (N0CFP && N1CFP)
}
}
- if (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Assume if there is an fmad instruction that it should be aggressively
- // used.
- if (SDValue Fused = performFsubFmulCombines(ISD::FMAD, true, N, TLI, DAG))
- return Fused;
- }
-
// FSUB -> FMA combines:
- if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) &&
- TLI.isFMAFasterThanFMulAndFAdd(VT) &&
- (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) {
-
- if (!TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Don't form FMA if we are preferring FMAD.
-
- if (SDValue Fused
- = performFsubFmulCombines(ISD::FMA,
- TLI.enableAggressiveFMAFusion(VT),
- N, TLI, DAG)) {
- return Fused;
- }
- }
-
- // When FP_EXTEND nodes are free on the target, and there is an opportunity
- // to combine into FMA, arrange such nodes accordingly.
- if (TLI.isFPExtFree(VT)) {
- // fold (fsub (fpext (fmul x, y)), z)
- // -> (fma (fpext x), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(1)),
- DAG.getNode(ISD::FNEG, SDLoc(N), VT, N1));
- }
-
- // fold (fsub x, (fpext (fmul y, z)))
- // -> (fma (fneg (fpext y)), (fpext z), x)
- // Note: Commutes FSUB operands.
- if (N1.getOpcode() == ISD::FP_EXTEND) {
- SDValue N10 = N1.getOperand(0);
- if (N10.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N10.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(1)),
- N0);
- }
-
- // fold (fsub (fpext (fneg (fmul, x, y))), z)
- // -> (fma (fneg (fpext x)), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FNEG) {
- SDValue N000 = N00.getOperand(0);
- if (N000.getOpcode() == ISD::FMUL) {
- return DAG.getNode(ISD::FMA, dl, VT,
- DAG.getNode(ISD::FNEG, dl, VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N000.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N000.getOperand(1)),
- DAG.getNode(ISD::FNEG, dl, VT, N1));
- }
- }
- }
-
- // fold (fsub (fneg (fpext (fmul, x, y))), z)
- // -> (fma (fneg (fpext x)), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FNEG) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FP_EXTEND) {
- SDValue N000 = N00.getOperand(0);
- if (N000.getOpcode() == ISD::FMUL) {
- return DAG.getNode(ISD::FMA, dl, VT,
- DAG.getNode(ISD::FNEG, dl, VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N000.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N000.getOperand(1)),
- DAG.getNode(ISD::FNEG, dl, VT, N1));
- }
- }
- }
- }
+ SDValue Fused = visitFSUBForFMACombine(N);
+ if (Fused) {
+ AddToWorklist(Fused.getNode());
+ return Fused;
}
return SDValue();
// fold vector ops
if (VT.isVector()) {
// This just handles C1 * C2 for vectors. Other vector folds are below.
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
return FoldedVOp;
- // Canonicalize vector constant to RHS.
- if (N0.getOpcode() == ISD::BUILD_VECTOR &&
- N1.getOpcode() != ISD::BUILD_VECTOR)
- if (auto *BV0 = dyn_cast<BuildVectorSDNode>(N0))
- if (BV0->isConstant())
- return DAG.getNode(N->getOpcode(), SDLoc(N), VT, N1, N0);
}
// fold (fmul c1, c2) -> c1*c2
return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, N1);
// canonicalize constant to RHS
- if (N0CFP && !N1CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0) &&
+ !isConstantFPBuildVectorOrConstantFP(N1))
return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1, N0);
// fold (fmul A, 1.0) -> A
// Fold scalars or any vector constants (not just splats).
// This fold is done in general by InstCombine, but extra fmul insts
// may have been generated during lowering.
+ SDValue N00 = N0.getOperand(0);
SDValue N01 = N0.getOperand(1);
auto *BV1 = dyn_cast<BuildVectorSDNode>(N1);
+ auto *BV00 = dyn_cast<BuildVectorSDNode>(N00);
auto *BV01 = dyn_cast<BuildVectorSDNode>(N01);
- if ((N1CFP && isConstOrConstSplatFP(N01)) ||
- (BV1 && BV01 && BV1->isConstant() && BV01->isConstant())) {
- SDLoc SL(N);
- SDValue MulConsts = DAG.getNode(ISD::FMUL, SL, VT, N01, N1);
- return DAG.getNode(ISD::FMUL, SL, VT, N0.getOperand(0), MulConsts);
+
+ // Check 1: Make sure that the first operand of the inner multiply is NOT
+ // a constant. Otherwise, we may induce infinite looping.
+ if (!(isConstOrConstSplatFP(N00) || (BV00 && BV00->isConstant()))) {
+ // Check 2: Make sure that the second operand of the inner multiply and
+ // the second operand of the outer multiply are constants.
+ if ((N1CFP && isConstOrConstSplatFP(N01)) ||
+ (BV1 && BV01 && BV1->isConstant() && BV01->isConstant())) {
+ SDLoc SL(N);
+ SDValue MulConsts = DAG.getNode(ISD::FMUL, SL, VT, N01, N1);
+ return DAG.getNode(ISD::FMUL, SL, VT, N00, MulConsts);
+ }
}
}
const TargetOptions &Options = DAG.getTarget().Options;
// fold vector ops
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVBinOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
// fold (fdiv c1, c2) -> c1/c2
if (N0CFP && N1CFP)
EVT OpVT = N0.getValueType();
// fold (sint_to_fp c1) -> c1fp
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C &&
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
// ...but only if the target supports immediate floating-point values
(!LegalOperations ||
TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT)))
EVT OpVT = N0.getValueType();
// fold (uint_to_fp c1) -> c1fp
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C &&
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
// ...but only if the target supports immediate floating-point values
(!LegalOperations ||
TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT)))
SDValue DAGCombiner::visitFP_EXTEND(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// If this is fp_round(fpextend), don't fold it, allow ourselves to be folded.
return SDValue();
// fold (fp_extend c1fp) -> c1fp
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT, N0);
+ // fold (fp_extend (fp16_to_fp op)) -> (fp16_to_fp op)
+ if (N0.getOpcode() == ISD::FP16_TO_FP &&
+ TLI.getOperationAction(ISD::FP16_TO_FP, VT) == TargetLowering::Legal)
+ return DAG.getNode(ISD::FP16_TO_FP, SDLoc(N), VT, N0.getOperand(0));
+
// Turn fp_extend(fp_round(X, 1)) -> x since the fp_round doesn't affect the
// value of X.
if (N0.getOpcode() == ISD::FP_ROUND
SDValue DAGCombiner::visitFCEIL(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// fold (fceil c1) -> fceil(c1)
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FCEIL, SDLoc(N), VT, N0);
return SDValue();
SDValue DAGCombiner::visitFTRUNC(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// fold (ftrunc c1) -> ftrunc(c1)
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FTRUNC, SDLoc(N), VT, N0);
return SDValue();
SDValue DAGCombiner::visitFFLOOR(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// fold (ffloor c1) -> ffloor(c1)
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FFLOOR, SDLoc(N), VT, N0);
return SDValue();
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVUnaryOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
-
// Constant fold FNEG.
- if (isa<ConstantFPSDNode>(N0))
- return DAG.getNode(ISD::FNEG, SDLoc(N), VT, N->getOperand(0));
+ if (isConstantFPBuildVectorOrConstantFP(N0))
+ return DAG.getNode(ISD::FNEG, SDLoc(N), VT, N0);
if (isNegatibleForFree(N0, LegalOperations, DAG.getTargetLoweringInfo(),
&DAG.getTarget().Options))
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
- if (VT.isVector()) {
- SDValue FoldedVOp = SimplifyVUnaryOp(N);
- if (FoldedVOp.getNode()) return FoldedVOp;
- }
-
// fold (fabs c1) -> fabs(c1)
- if (isa<ConstantFPSDNode>(N0))
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FABS, SDLoc(N), VT, N0);
// fold (fabs (fabs x)) -> (fabs x)
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Use)) {
if (LD->isIndexed() || LD->getBasePtr().getNode() != N)
return false;
- VT = Use->getValueType(0);
+ VT = LD->getMemoryVT();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(Use)) {
if (ST->isIndexed() || ST->getBasePtr().getNode() != N)
return false;
- VT = ST->getValue().getValueType();
+ VT = ST->getMemoryVT();
} else
return false;
LD->getMemoryVT(),
LD->isVolatile(), LD->isNonTemporal(),
LD->isInvariant(), Align, LD->getAAInfo());
- return CombineTo(N, NewLoad, SDValue(NewLoad.getNode(), 1), true);
+ if (NewLoad.getNode() != N)
+ return CombineTo(N, NewLoad, SDValue(NewLoad.getNode(), 1), true);
}
}
}
unsigned Shift = 0, SelectionDAG *DAG = nullptr)
: Inst(Inst), Origin(Origin), Shift(Shift), DAG(DAG) {}
- LoadedSlice(const LoadedSlice &LS)
- : Inst(LS.Inst), Origin(LS.Origin), Shift(LS.Shift), DAG(LS.DAG) {}
-
/// \brief Get the bits used in a chunk of bits \p BitWidth large.
/// \return Result is \p BitWidth and has used bits set to 1 and
/// not used bits set to 0.
return SDValue();
}
+namespace {
/// Helper struct to parse and store a memory address as base + index + offset.
/// We ignore sign extensions when it is safe to do so.
/// The following two expressions are not equivalent. To differentiate we need
return BaseIndexOffset(Base, Index, Off, IsIndexSignExt);
}
};
+} // namespace
bool DAGCombiner::MergeStoresOfConstantsOrVecElts(
SmallVectorImpl<MemOpLink> &StoreNodes, EVT MemVT,
// Make sure we have something to merge.
if (NumElem < 2)
return false;
-
+
int64_t ElementSizeBytes = MemVT.getSizeInBits() / 8;
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
- unsigned EarliestNodeUsed = 0;
-
+ unsigned LatestNodeUsed = 0;
+
for (unsigned i=0; i < NumElem; ++i) {
// Find a chain for the new wide-store operand. Notice that some
// of the store nodes that we found may not be selected for inclusion
// in the wide store. The chain we use needs to be the chain of the
- // earliest store node which is *used* and replaced by the wide store.
- if (StoreNodes[i].SequenceNum > StoreNodes[EarliestNodeUsed].SequenceNum)
- EarliestNodeUsed = i;
+ // latest store node which is *used* and replaced by the wide store.
+ if (StoreNodes[i].SequenceNum < StoreNodes[LatestNodeUsed].SequenceNum)
+ LatestNodeUsed = i;
}
-
- // The earliest Node in the DAG.
- LSBaseSDNode *EarliestOp = StoreNodes[EarliestNodeUsed].MemNode;
+
+ // The latest Node in the DAG.
+ LSBaseSDNode *LatestOp = StoreNodes[LatestNodeUsed].MemNode;
SDLoc DL(StoreNodes[0].MemNode);
-
+
SDValue StoredVal;
if (UseVector) {
// Find a legal type for the vector store.
return false;
Ops.push_back(Val);
}
-
+
// Build the extracted vector elements back into a vector.
StoredVal = DAG.getNode(ISD::BUILD_VECTOR, DL, Ty, Ops);
}
unsigned StoreBW = NumElem * ElementSizeBytes * 8;
APInt StoreInt(StoreBW, 0);
-
+
// Construct a single integer constant which is made of the smaller
// constant inputs.
bool IsLE = TLI.isLittleEndian();
llvm_unreachable("Invalid constant element type");
}
}
-
+
// Create the new Load and Store operations.
EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
StoredVal = DAG.getConstant(StoreInt, StoreTy);
}
-
- SDValue NewStore = DAG.getStore(EarliestOp->getChain(), DL, StoredVal,
+
+ SDValue NewStore = DAG.getStore(LatestOp->getChain(), DL, StoredVal,
FirstInChain->getBasePtr(),
FirstInChain->getPointerInfo(),
false, false,
FirstInChain->getAlignment());
-
- // Replace the first store with the new store
- CombineTo(EarliestOp, NewStore);
+
+ // Replace the last store with the new store
+ CombineTo(LatestOp, NewStore);
// Erase all other stores.
for (unsigned i = 0; i < NumElem ; ++i) {
- if (StoreNodes[i].MemNode == EarliestOp)
+ if (StoreNodes[i].MemNode == LatestOp)
continue;
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
// ReplaceAllUsesWith will replace all uses that existed when it was
DAG.ReplaceAllUsesWith(SDValue(St, 0), St->getChain());
deleteAndRecombine(St);
}
-
+
return true;
}
bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) {
+ if (OptLevel == CodeGenOpt::None)
+ return false;
+
EVT MemVT = St->getMemoryVT();
int64_t ElementSizeBytes = MemVT.getSizeInBits()/8;
bool NoVectors = DAG.getMachineFunction().getFunction()->hasFnAttribute(
bool IsConstantSrc = isa<ConstantSDNode>(StoredVal) ||
isa<ConstantFPSDNode>(StoredVal);
bool IsExtractVecEltSrc = (StoredVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT);
-
+
if (!IsConstantSrc && !IsLoadSrc && !IsExtractVecEltSrc)
return false;
// consecutive loads).
if (StoredVal.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return false;
-
+
// Find a legal type for the vector store.
EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
if (TLI.isTypeLegal(Ty))
if (NumElem < 2)
return false;
- // The earliest Node in the DAG.
- unsigned EarliestNodeUsed = 0;
- LSBaseSDNode *EarliestOp = StoreNodes[EarliestNodeUsed].MemNode;
+ // The latest Node in the DAG.
+ unsigned LatestNodeUsed = 0;
for (unsigned i=1; i<NumElem; ++i) {
// Find a chain for the new wide-store operand. Notice that some
// of the store nodes that we found may not be selected for inclusion
// in the wide store. The chain we use needs to be the chain of the
- // earliest store node which is *used* and replaced by the wide store.
- if (StoreNodes[i].SequenceNum > StoreNodes[EarliestNodeUsed].SequenceNum)
- EarliestNodeUsed = i;
+ // latest store node which is *used* and replaced by the wide store.
+ if (StoreNodes[i].SequenceNum < StoreNodes[LatestNodeUsed].SequenceNum)
+ LatestNodeUsed = i;
}
+ LSBaseSDNode *LatestOp = StoreNodes[LatestNodeUsed].MemNode;
+
// Find if it is better to use vectors or integers to load and store
// to memory.
EVT JointMemOpVT;
false, false, false,
FirstLoad->getAlignment());
- SDValue NewStore = DAG.getStore(EarliestOp->getChain(), StoreDL, NewLoad,
+ SDValue NewStore = DAG.getStore(LatestOp->getChain(), StoreDL, NewLoad,
FirstInChain->getBasePtr(),
FirstInChain->getPointerInfo(), false, false,
FirstInChain->getAlignment());
DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1), Ld->getChain());
}
- // Replace the first store with the new store.
- CombineTo(EarliestOp, NewStore);
+ // Replace the last store with the new store.
+ CombineTo(LatestOp, NewStore);
// Erase all other stores.
for (unsigned i = 0; i < NumElem ; ++i) {
// Remove all Store nodes.
- if (StoreNodes[i].MemNode == EarliestOp)
+ if (StoreNodes[i].MemNode == LatestOp)
continue;
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
DAG.ReplaceAllUsesOfValueWith(SDValue(St, 0), St->getChain());
// Try to infer better alignment information than the store already has.
if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
- if (Align > ST->getAlignment())
- return DAG.getTruncStore(Chain, SDLoc(N), Value,
+ if (Align > ST->getAlignment()) {
+ SDValue NewStore =
+ DAG.getTruncStore(Chain, SDLoc(N), Value,
Ptr, ST->getPointerInfo(), ST->getMemoryVT(),
ST->isVolatile(), ST->isNonTemporal(), Align,
ST->getAAInfo());
+ if (NewStore.getNode() != N)
+ return CombineTo(ST, NewStore, true);
+ }
}
}
// Just because the floating-point vector type is legal does not necessarily
// mean that the corresponding integer vector type is.
if (!isTypeLegal(NVT))
- return SDValue();
+ return SDValue();
SmallVector<SDValue, 8> Opnds;
for (unsigned i = 0; i != NumInScalars; ++i) {
if (ISD::allOperandsUndef(N))
return DAG.getUNDEF(VT);
- SDValue V = reduceBuildVecExtToExtBuildVec(N);
- if (V.getNode())
+ if (SDValue V = reduceBuildVecExtToExtBuildVec(N))
return V;
- V = reduceBuildVecConvertToConvertBuildVec(N);
- if (V.getNode())
+ if (SDValue V = reduceBuildVecConvertToConvertBuildVec(N))
return V;
// Check to see if this is a BUILD_VECTOR of a bunch of EXTRACT_VECTOR_ELT
} else if (VecInT.getSizeInBits() == VT.getSizeInBits() * 2) {
// If the input vector is too large, try to split it.
// We don't support having two input vectors that are too large.
- if (VecIn2.getNode())
+ // If the zero vector was used, we can not split the vector,
+ // since we'd need 3 inputs.
+ if (UsesZeroVector || VecIn2.getNode())
return SDValue();
if (!TLI.isExtractSubvectorCheap(VT, VT.getVectorNumElements()))
return SDValue();
-
+
// Try to replace VecIn1 with two extract_subvectors
// No need to update the masks, they should still be correct.
- VecIn2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
+ VecIn2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
DAG.getConstant(VT.getVectorNumElements(), TLI.getVectorIdxTy()));
VecIn1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
DAG.getConstant(0, TLI.getVectorIdxTy()));
- UsesZeroVector = false;
} else
return SDValue();
}
return SDValue();
}
+static SDValue combineConcatVectorOfScalars(SDNode *N, SelectionDAG &DAG) {
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ EVT OpVT = N->getOperand(0).getValueType();
+
+ // If the operands are legal vectors, leave them alone.
+ if (TLI.isTypeLegal(OpVT))
+ return SDValue();
+
+ SDLoc DL(N);
+ EVT VT = N->getValueType(0);
+ SmallVector<SDValue, 8> Ops;
+
+ EVT SVT = EVT::getIntegerVT(*DAG.getContext(), OpVT.getSizeInBits());
+ SDValue ScalarUndef = DAG.getNode(ISD::UNDEF, DL, SVT);
+
+ // Keep track of what we encounter.
+ bool AnyInteger = false;
+ bool AnyFP = false;
+ for (const SDValue &Op : N->ops()) {
+ if (ISD::BITCAST == Op.getOpcode() &&
+ !Op.getOperand(0).getValueType().isVector())
+ Ops.push_back(Op.getOperand(0));
+ else if (ISD::UNDEF == Op.getOpcode())
+ Ops.push_back(ScalarUndef);
+ else
+ return SDValue();
+
+ // Note whether we encounter an integer or floating point scalar.
+ // If it's neither, bail out, it could be something weird like x86mmx.
+ EVT LastOpVT = Ops.back().getValueType();
+ if (LastOpVT.isFloatingPoint())
+ AnyFP = true;
+ else if (LastOpVT.isInteger())
+ AnyInteger = true;
+ else
+ return SDValue();
+ }
+
+ // If any of the operands is a floating point scalar bitcast to a vector,
+ // use floating point types throughout, and bitcast everything.
+ // Replace UNDEFs by another scalar UNDEF node, of the final desired type.
+ if (AnyFP) {
+ SVT = EVT::getFloatingPointVT(OpVT.getSizeInBits());
+ ScalarUndef = DAG.getNode(ISD::UNDEF, DL, SVT);
+ if (AnyInteger) {
+ for (SDValue &Op : Ops) {
+ if (Op.getValueType() == SVT)
+ continue;
+ if (Op.getOpcode() == ISD::UNDEF)
+ Op = ScalarUndef;
+ else
+ Op = DAG.getNode(ISD::BITCAST, DL, SVT, Op);
+ }
+ }
+ }
+
+ EVT VecVT = EVT::getVectorVT(*DAG.getContext(), SVT,
+ VT.getSizeInBits() / SVT.getSizeInBits());
+ return DAG.getNode(ISD::BITCAST, DL, VT,
+ DAG.getNode(ISD::BUILD_VECTOR, DL, VecVT, Ops));
+}
+
SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) {
// TODO: Check to see if this is a CONCAT_VECTORS of a bunch of
// EXTRACT_SUBVECTOR operations. If so, and if the EXTRACT_SUBVECTOR vector
if (ISD::allOperandsUndef(N))
return DAG.getUNDEF(VT);
- // Optimize concat_vectors where one of the vectors is undef.
- if (N->getNumOperands() == 2 &&
- N->getOperand(1)->getOpcode() == ISD::UNDEF) {
+ // Optimize concat_vectors where all but the first of the vectors are undef.
+ if (std::all_of(std::next(N->op_begin()), N->op_end(), [](const SDValue &Op) {
+ return Op.getOpcode() == ISD::UNDEF;
+ })) {
SDValue In = N->getOperand(0);
assert(In.getValueType().isVector() && "Must concat vectors");
if (In->getOpcode() == ISD::BITCAST &&
!In->getOperand(0)->getValueType(0).isVector()) {
SDValue Scalar = In->getOperand(0);
+
+ // If the bitcast type isn't legal, it might be a trunc of a legal type;
+ // look through the trunc so we can still do the transform:
+ // concat_vectors(trunc(scalar), undef) -> scalar_to_vector(scalar)
+ if (Scalar->getOpcode() == ISD::TRUNCATE &&
+ !TLI.isTypeLegal(Scalar.getValueType()) &&
+ TLI.isTypeLegal(Scalar->getOperand(0).getValueType()))
+ Scalar = Scalar->getOperand(0);
+
EVT SclTy = Scalar->getValueType(0);
if (!SclTy.isFloatingPoint() && !SclTy.isInteger())
}
}
+ // Fold any combination of BUILD_VECTOR or UNDEF nodes into one BUILD_VECTOR.
+ // We have already tested above for an UNDEF only concatenation.
// fold (concat_vectors (BUILD_VECTOR A, B, ...), (BUILD_VECTOR C, D, ...))
// -> (BUILD_VECTOR A, B, ..., C, D, ...)
- if (N->getNumOperands() == 2 &&
- N->getOperand(0).getOpcode() == ISD::BUILD_VECTOR &&
- N->getOperand(1).getOpcode() == ISD::BUILD_VECTOR) {
- EVT VT = N->getValueType(0);
- SDValue N0 = N->getOperand(0);
- SDValue N1 = N->getOperand(1);
+ auto IsBuildVectorOrUndef = [](const SDValue &Op) {
+ return ISD::UNDEF == Op.getOpcode() || ISD::BUILD_VECTOR == Op.getOpcode();
+ };
+ bool AllBuildVectorsOrUndefs =
+ std::all_of(N->op_begin(), N->op_end(), IsBuildVectorOrUndef);
+ if (AllBuildVectorsOrUndefs) {
SmallVector<SDValue, 8> Opnds;
- unsigned BuildVecNumElts = N0.getNumOperands();
-
- EVT SclTy0 = N0.getOperand(0)->getValueType(0);
- EVT SclTy1 = N1.getOperand(0)->getValueType(0);
- if (SclTy0.isFloatingPoint()) {
- for (unsigned i = 0; i != BuildVecNumElts; ++i)
- Opnds.push_back(N0.getOperand(i));
- for (unsigned i = 0; i != BuildVecNumElts; ++i)
- Opnds.push_back(N1.getOperand(i));
- } else {
+ EVT SVT = VT.getScalarType();
+
+ EVT MinVT = SVT;
+ if (!SVT.isFloatingPoint()) {
// If BUILD_VECTOR are from built from integer, they may have different
- // operand types. Get the smaller type and truncate all operands to it.
- EVT MinTy = SclTy0.bitsLE(SclTy1) ? SclTy0 : SclTy1;
- for (unsigned i = 0; i != BuildVecNumElts; ++i)
- Opnds.push_back(DAG.getNode(ISD::TRUNCATE, SDLoc(N), MinTy,
- N0.getOperand(i)));
- for (unsigned i = 0; i != BuildVecNumElts; ++i)
- Opnds.push_back(DAG.getNode(ISD::TRUNCATE, SDLoc(N), MinTy,
- N1.getOperand(i)));
+ // operand types. Get the smallest type and truncate all operands to it.
+ bool FoundMinVT = false;
+ for (const SDValue &Op : N->ops())
+ if (ISD::BUILD_VECTOR == Op.getOpcode()) {
+ EVT OpSVT = Op.getOperand(0)->getValueType(0);
+ MinVT = (!FoundMinVT || OpSVT.bitsLE(MinVT)) ? OpSVT : MinVT;
+ FoundMinVT = true;
+ }
+ assert(FoundMinVT && "Concat vector type mismatch");
+ }
+
+ for (const SDValue &Op : N->ops()) {
+ EVT OpVT = Op.getValueType();
+ unsigned NumElts = OpVT.getVectorNumElements();
+
+ if (ISD::UNDEF == Op.getOpcode())
+ Opnds.append(NumElts, DAG.getUNDEF(MinVT));
+
+ if (ISD::BUILD_VECTOR == Op.getOpcode()) {
+ if (SVT.isFloatingPoint()) {
+ assert(SVT == OpVT.getScalarType() && "Concat vector type mismatch");
+ Opnds.append(Op->op_begin(), Op->op_begin() + NumElts);
+ } else {
+ for (unsigned i = 0; i != NumElts; ++i)
+ Opnds.push_back(
+ DAG.getNode(ISD::TRUNCATE, SDLoc(N), MinVT, Op.getOperand(i)));
+ }
+ }
}
+ assert(VT.getVectorNumElements() == Opnds.size() &&
+ "Concat vector type mismatch");
return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, Opnds);
}
+ // Fold CONCAT_VECTORS of only bitcast scalars (or undef) to BUILD_VECTOR.
+ if (SDValue V = combineConcatVectorOfScalars(N, DAG))
+ return V;
+
// Type legalization of vectors and DAG canonicalization of SHUFFLE_VECTOR
// nodes often generate nop CONCAT_VECTOR nodes.
// Scan the CONCAT_VECTOR operands and look for a CONCAT operations that
// type.
if (V->getOperand(0).getValueType() != NVT)
return SDValue();
- unsigned Idx = dyn_cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
+ unsigned Idx = N->getConstantOperandVal(1);
unsigned NumElems = NVT.getVectorNumElements();
assert((Idx % NumElems) == 0 &&
"IDX in concat is not a multiple of the result vector length.");
// We may have jumped through bitcasts, so the type of the
// BUILD_VECTOR may not match the type of the shuffle.
if (V->getValueType(0) != VT)
- NewBV = DAG.getNode(ISD::BITCAST, SDLoc(N), VT, NewBV);
+ NewBV = DAG.getNode(ISD::BITCAST, SDLoc(N), VT, NewBV);
return NewBV;
}
}
return V;
}
+ // Attempt to combine a shuffle of 2 inputs of 'scalar sources' -
+ // BUILD_VECTOR or SCALAR_TO_VECTOR into a single BUILD_VECTOR.
+ if (Level < AfterLegalizeVectorOps && TLI.isTypeLegal(VT)) {
+ SmallVector<SDValue, 8> Ops;
+ for (int M : SVN->getMask()) {
+ SDValue Op = DAG.getUNDEF(VT.getScalarType());
+ if (M >= 0) {
+ int Idx = M % NumElts;
+ SDValue &S = (M < (int)NumElts ? N0 : N1);
+ if (S.getOpcode() == ISD::BUILD_VECTOR && S.hasOneUse()) {
+ Op = S.getOperand(Idx);
+ } else if (S.getOpcode() == ISD::SCALAR_TO_VECTOR && S.hasOneUse()) {
+ if (Idx == 0)
+ Op = S.getOperand(0);
+ } else {
+ // Operand can't be combined - bail out.
+ break;
+ }
+ }
+ Ops.push_back(Op);
+ }
+ if (Ops.size() == VT.getVectorNumElements()) {
+ // BUILD_VECTOR requires all inputs to be of the same type, find the
+ // maximum type and extend them all.
+ EVT SVT = VT.getScalarType();
+ if (SVT.isInteger())
+ for (SDValue &Op : Ops)
+ SVT = (SVT.bitsLT(Op.getValueType()) ? Op.getValueType() : SVT);
+ if (SVT != VT.getScalarType())
+ for (SDValue &Op : Ops)
+ Op = TLI.isZExtFree(Op.getValueType(), SVT)
+ ? DAG.getZExtOrTrunc(Op, SDLoc(N), SVT)
+ : DAG.getSExtOrTrunc(Op, SDLoc(N), SVT);
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, Ops);
+ }
+ }
+
+ // If this shuffle only has a single input that is a bitcasted shuffle,
+ // attempt to merge the 2 shuffles and suitably bitcast the inputs/output
+ // back to their original types.
+ if (N0.getOpcode() == ISD::BITCAST && N0.hasOneUse() &&
+ N1.getOpcode() == ISD::UNDEF && Level < AfterLegalizeVectorOps &&
+ TLI.isTypeLegal(VT)) {
+
+ // Peek through the bitcast only if there is one user.
+ SDValue BC0 = N0;
+ while (BC0.getOpcode() == ISD::BITCAST) {
+ if (!BC0.hasOneUse())
+ break;
+ BC0 = BC0.getOperand(0);
+ }
+
+ auto ScaleShuffleMask = [](ArrayRef<int> Mask, int Scale) {
+ if (Scale == 1)
+ return SmallVector<int, 8>(Mask.begin(), Mask.end());
+
+ SmallVector<int, 8> NewMask;
+ for (int M : Mask)
+ for (int s = 0; s != Scale; ++s)
+ NewMask.push_back(M < 0 ? -1 : Scale * M + s);
+ return NewMask;
+ };
+
+ if (BC0.getOpcode() == ISD::VECTOR_SHUFFLE && BC0.hasOneUse()) {
+ EVT SVT = VT.getScalarType();
+ EVT InnerVT = BC0->getValueType(0);
+ EVT InnerSVT = InnerVT.getScalarType();
+
+ // Determine which shuffle works with the smaller scalar type.
+ EVT ScaleVT = SVT.bitsLT(InnerSVT) ? VT : InnerVT;
+ EVT ScaleSVT = ScaleVT.getScalarType();
+
+ if (TLI.isTypeLegal(ScaleVT) &&
+ 0 == (InnerSVT.getSizeInBits() % ScaleSVT.getSizeInBits()) &&
+ 0 == (SVT.getSizeInBits() % ScaleSVT.getSizeInBits())) {
+
+ int InnerScale = InnerSVT.getSizeInBits() / ScaleSVT.getSizeInBits();
+ int OuterScale = SVT.getSizeInBits() / ScaleSVT.getSizeInBits();
+
+ // Scale the shuffle masks to the smaller scalar type.
+ ShuffleVectorSDNode *InnerSVN = cast<ShuffleVectorSDNode>(BC0);
+ SmallVector<int, 8> InnerMask =
+ ScaleShuffleMask(InnerSVN->getMask(), InnerScale);
+ SmallVector<int, 8> OuterMask =
+ ScaleShuffleMask(SVN->getMask(), OuterScale);
+
+ // Merge the shuffle masks.
+ SmallVector<int, 8> NewMask;
+ for (int M : OuterMask)
+ NewMask.push_back(M < 0 ? -1 : InnerMask[M]);
+
+ // Test for shuffle mask legality over both commutations.
+ SDValue SV0 = BC0->getOperand(0);
+ SDValue SV1 = BC0->getOperand(1);
+ bool LegalMask = TLI.isShuffleMaskLegal(NewMask, ScaleVT);
+ if (!LegalMask) {
+ std::swap(SV0, SV1);
+ ShuffleVectorSDNode::commuteMask(NewMask);
+ LegalMask = TLI.isShuffleMaskLegal(NewMask, ScaleVT);
+ }
+
+ if (LegalMask) {
+ SV0 = DAG.getNode(ISD::BITCAST, SDLoc(N), ScaleVT, SV0);
+ SV1 = DAG.getNode(ISD::BITCAST, SDLoc(N), ScaleVT, SV1);
+ return DAG.getNode(
+ ISD::BITCAST, SDLoc(N), VT,
+ DAG.getVectorShuffle(ScaleVT, SDLoc(N), SV0, SV1, NewMask));
+ }
+ }
+ }
+ }
+
// Canonicalize shuffles according to rules:
// shuffle(A, shuffle(A, B)) -> shuffle(shuffle(A,B), A)
// shuffle(B, shuffle(A, B)) -> shuffle(shuffle(A,B), B)
// Avoid introducing shuffles with illegal mask.
if (!TLI.isShuffleMaskLegal(Mask, VT)) {
- // Compute the commuted shuffle mask and test again.
- for (unsigned i = 0; i != NumElts; ++i) {
- int idx = Mask[i];
- if (idx < 0)
- continue;
- else if (idx < (int)NumElts)
- Mask[i] = idx + NumElts;
- else
- Mask[i] = idx - NumElts;
- }
+ ShuffleVectorSDNode::commuteMask(Mask);
if (!TLI.isShuffleMaskLegal(Mask, VT))
return SDValue();
-
+
// shuffle(shuffle(A, B, M0), C, M1) -> shuffle(B, A, M2)
// shuffle(shuffle(A, B, M0), C, M1) -> shuffle(C, A, M2)
// shuffle(shuffle(A, B, M0), C, M1) -> shuffle(C, B, M2)
return SDValue();
}
+SDValue DAGCombiner::visitSCALAR_TO_VECTOR(SDNode *N) {
+ SDValue InVal = N->getOperand(0);
+ EVT VT = N->getValueType(0);
+
+ // Replace a SCALAR_TO_VECTOR(EXTRACT_VECTOR_ELT(V,C0)) pattern
+ // with a VECTOR_SHUFFLE.
+ if (InVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+ SDValue InVec = InVal->getOperand(0);
+ SDValue EltNo = InVal->getOperand(1);
+
+ // FIXME: We could support implicit truncation if the shuffle can be
+ // scaled to a smaller vector scalar type.
+ ConstantSDNode *C0 = dyn_cast<ConstantSDNode>(EltNo);
+ if (C0 && VT == InVec.getValueType() &&
+ VT.getScalarType() == InVal.getValueType()) {
+ SmallVector<int, 8> NewMask(VT.getVectorNumElements(), -1);
+ int Elt = C0->getZExtValue();
+ NewMask[0] = Elt;
+
+ if (TLI.isShuffleMaskLegal(NewMask, VT))
+ return DAG.getVectorShuffle(VT, SDLoc(N), InVec, DAG.getUNDEF(VT),
+ NewMask);
+ }
+ }
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitINSERT_SUBVECTOR(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N2 = N->getOperand(2);
return SDValue();
}
+SDValue DAGCombiner::visitFP_TO_FP16(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+
+ // fold (fp_to_fp16 (fp16_to_fp op)) -> op
+ if (N0->getOpcode() == ISD::FP16_TO_FP)
+ return N0->getOperand(0);
+
+ return SDValue();
+}
+
/// Returns a vector_shuffle if it able to transform an AND to a vector_shuffle
/// with the destination vector and a zero vector.
/// e.g. AND V, <0xffffffff, 0, 0xffffffff, 0>. ==>
/// vector_shuffle V, Zero, <0, 4, 2, 4>
SDValue DAGCombiner::XformToShuffleWithZero(SDNode *N) {
EVT VT = N->getValueType(0);
- SDLoc dl(N);
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
- if (N->getOpcode() == ISD::AND) {
- if (RHS.getOpcode() == ISD::BITCAST)
- RHS = RHS.getOperand(0);
- if (RHS.getOpcode() == ISD::BUILD_VECTOR) {
- SmallVector<int, 8> Indices;
- unsigned NumElts = RHS.getNumOperands();
- for (unsigned i = 0; i != NumElts; ++i) {
- SDValue Elt = RHS.getOperand(i);
- if (!isa<ConstantSDNode>(Elt))
- return SDValue();
+ SDLoc dl(N);
- if (cast<ConstantSDNode>(Elt)->isAllOnesValue())
- Indices.push_back(i);
- else if (cast<ConstantSDNode>(Elt)->isNullValue())
- Indices.push_back(NumElts+i);
- else
- return SDValue();
- }
+ // Make sure we're not running after operation legalization where it
+ // may have custom lowered the vector shuffles.
+ if (LegalOperations)
+ return SDValue();
+
+ if (N->getOpcode() != ISD::AND)
+ return SDValue();
- // Let's see if the target supports this vector_shuffle and make sure
- // we're not running after operation legalization where it may have
- // custom lowered the vector shuffles.
- EVT RVT = RHS.getValueType();
- if (LegalOperations || !TLI.isVectorClearMaskLegal(Indices, RVT))
+ if (RHS.getOpcode() == ISD::BITCAST)
+ RHS = RHS.getOperand(0);
+
+ if (RHS.getOpcode() == ISD::BUILD_VECTOR) {
+ SmallVector<int, 8> Indices;
+ unsigned NumElts = RHS.getNumOperands();
+
+ for (unsigned i = 0; i != NumElts; ++i) {
+ SDValue Elt = RHS.getOperand(i);
+ if (!isa<ConstantSDNode>(Elt))
return SDValue();
- // Return the new VECTOR_SHUFFLE node.
- EVT EltVT = RVT.getVectorElementType();
- SmallVector<SDValue,8> ZeroOps(RVT.getVectorNumElements(),
- DAG.getConstant(0, EltVT));
- SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), RVT, ZeroOps);
- LHS = DAG.getNode(ISD::BITCAST, dl, RVT, LHS);
- SDValue Shuf = DAG.getVectorShuffle(RVT, dl, LHS, Zero, &Indices[0]);
- return DAG.getNode(ISD::BITCAST, dl, VT, Shuf);
+ if (cast<ConstantSDNode>(Elt)->isAllOnesValue())
+ Indices.push_back(i);
+ else if (cast<ConstantSDNode>(Elt)->isNullValue())
+ Indices.push_back(NumElts+i);
+ else
+ return SDValue();
}
+
+ // Let's see if the target supports this vector_shuffle.
+ EVT RVT = RHS.getValueType();
+ if (!TLI.isVectorClearMaskLegal(Indices, RVT))
+ return SDValue();
+
+ // Return the new VECTOR_SHUFFLE node.
+ EVT EltVT = RVT.getVectorElementType();
+ SmallVector<SDValue,8> ZeroOps(RVT.getVectorNumElements(),
+ DAG.getConstant(0, EltVT));
+ SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), RVT, ZeroOps);
+ LHS = DAG.getNode(ISD::BITCAST, dl, RVT, LHS);
+ SDValue Shuf = DAG.getVectorShuffle(RVT, dl, LHS, Zero, &Indices[0]);
+ return DAG.getNode(ISD::BITCAST, dl, VT, Shuf);
}
return SDValue();
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
- SDValue Shuffle = XformToShuffleWithZero(N);
- if (Shuffle.getNode()) return Shuffle;
+
+ if (SDValue Shuffle = XformToShuffleWithZero(N))
+ return Shuffle;
// If the LHS and RHS are BUILD_VECTOR nodes, see if we can constant fold
// this operation.
return SDValue();
}
-/// Visit a binary vector operation, like FABS/FNEG.
-SDValue DAGCombiner::SimplifyVUnaryOp(SDNode *N) {
- assert(N->getValueType(0).isVector() &&
- "SimplifyVUnaryOp only works on vectors!");
-
- SDValue N0 = N->getOperand(0);
-
- if (N0.getOpcode() != ISD::BUILD_VECTOR)
- return SDValue();
-
- // Operand is a BUILD_VECTOR node, see if we can constant fold it.
- SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0, e = N0.getNumOperands(); i != e; ++i) {
- SDValue Op = N0.getOperand(i);
- if (Op.getOpcode() != ISD::UNDEF &&
- Op.getOpcode() != ISD::ConstantFP)
- break;
- EVT EltVT = Op.getValueType();
- SDValue FoldOp = DAG.getNode(N->getOpcode(), SDLoc(N0), EltVT, Op);
- if (FoldOp.getOpcode() != ISD::UNDEF &&
- FoldOp.getOpcode() != ISD::ConstantFP)
- break;
- Ops.push_back(FoldOp);
- AddToWorklist(FoldOp.getNode());
- }
-
- if (Ops.size() != N0.getNumOperands())
- return SDValue();
-
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), N0.getValueType(), Ops);
-}
-
SDValue DAGCombiner::SimplifySelect(SDLoc DL, SDValue N0,
SDValue N1, SDValue N2){
assert(N0.getOpcode() ==ISD::SETCC && "First argument must be a SetCC node!");
bool DAGCombiner::SimplifySelectOps(SDNode *TheSelect, SDValue LHS,
SDValue RHS) {
+ // fold (select (setcc x, -0.0, *lt), NaN, (fsqrt x))
+ // The select + setcc is redundant, because fsqrt returns NaN for X < -0.
+ if (const ConstantFPSDNode *NaN = isConstOrConstSplatFP(LHS)) {
+ if (NaN->isNaN() && RHS.getOpcode() == ISD::FSQRT) {
+ // We have: (select (setcc ?, ?, ?), NaN, (fsqrt ?))
+ SDValue Sqrt = RHS;
+ ISD::CondCode CC;
+ SDValue CmpLHS;
+ const ConstantFPSDNode *NegZero = nullptr;
+
+ if (TheSelect->getOpcode() == ISD::SELECT_CC) {
+ CC = dyn_cast<CondCodeSDNode>(TheSelect->getOperand(4))->get();
+ CmpLHS = TheSelect->getOperand(0);
+ NegZero = isConstOrConstSplatFP(TheSelect->getOperand(1));
+ } else {
+ // SELECT or VSELECT
+ SDValue Cmp = TheSelect->getOperand(0);
+ if (Cmp.getOpcode() == ISD::SETCC) {
+ CC = dyn_cast<CondCodeSDNode>(Cmp.getOperand(2))->get();
+ CmpLHS = Cmp.getOperand(0);
+ NegZero = isConstOrConstSplatFP(Cmp.getOperand(1));
+ }
+ }
+ if (NegZero && NegZero->isNegative() && NegZero->isZero() &&
+ Sqrt.getOperand(0) == CmpLHS && (CC == ISD::SETOLT ||
+ CC == ISD::SETULT || CC == ISD::SETLT)) {
+ // We have: (select (setcc x, -0.0, *lt), NaN, (fsqrt x))
+ CombineTo(TheSelect, Sqrt);
+ return true;
+ }
+ }
+ }
// Cannot simplify select with vector condition
if (TheSelect->getOperand(0).getValueType().isVector()) return false;
if (LHS.getOperand(0) != RHS.getOperand(0) ||
// Do not let this transformation reduce the number of volatile loads.
LLD->isVolatile() || RLD->isVolatile() ||
+ // FIXME: If either is a pre/post inc/dec load,
+ // we'd need to split out the address adjustment.
+ LLD->isIndexed() || RLD->isIndexed() ||
// If this is an EXTLOAD, the VT's must match.
LLD->getMemoryVT() != RLD->getMemoryVT() ||
// If this is an EXTLOAD, the kind of extension must match.