diff options
Diffstat (limited to 'contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp')
| -rw-r--r-- | contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 4064 |
1 files changed, 2653 insertions, 1411 deletions
diff --git a/contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 1df61e4..a71c676 100644 --- a/contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/contrib/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -246,10 +246,11 @@ namespace { 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); @@ -267,6 +268,7 @@ namespace { SDValue visitZERO_EXTEND(SDNode *N); SDValue visitANY_EXTEND(SDNode *N); SDValue visitSIGN_EXTEND_INREG(SDNode *N); + SDValue visitSIGN_EXTEND_VECTOR_INREG(SDNode *N); SDValue visitTRUNCATE(SDNode *N); SDValue visitBITCAST(SDNode *N); SDValue visitBUILD_PAIR(SDNode *N); @@ -302,9 +304,16 @@ namespace { 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 visitMGATHER(SDNode *N); + SDValue visitMSCATTER(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); @@ -327,6 +336,7 @@ namespace { SDValue SimplifyNodeWithTwoResults(SDNode *N, unsigned LoOp, unsigned HiOp); SDValue CombineConsecutiveLoads(SDNode *N, EVT VT); + SDValue CombineExtLoad(SDNode *N); SDValue ConstantFoldBITCASTofBUILD_VECTOR(SDNode *, EVT); SDValue BuildSDIV(SDNode *N); SDValue BuildSDIVPow2(SDNode *N); @@ -363,6 +373,28 @@ namespace { /// chain (aliasing node.) SDValue FindBetterChain(SDNode *N, SDValue Chain); + /// Holds a pointer to an LSBaseSDNode as well as information on where it + /// is located in a sequence of memory operations connected by a chain. + struct MemOpLink { + MemOpLink (LSBaseSDNode *N, int64_t Offset, unsigned Seq): + MemNode(N), OffsetFromBase(Offset), SequenceNum(Seq) { } + // Ptr to the mem node. + LSBaseSDNode *MemNode; + // Offset from the base ptr. + int64_t OffsetFromBase; + // What is the sequence number of this mem node. + // Lowest mem operand in the DAG starts at zero. + unsigned SequenceNum; + }; + + /// This is a helper function for MergeConsecutiveStores. When the source + /// elements of the consecutive stores are all constants or all extracted + /// vector elements, try to merge them into one larger store. + /// \return True if a merged store was created. + 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. @@ -380,12 +412,9 @@ namespace { DAGCombiner(SelectionDAG &D, AliasAnalysis &A, CodeGenOpt::Level OL) : DAG(D), TLI(D.getTargetLoweringInfo()), Level(BeforeLegalizeTypes), OptLevel(OL), LegalOperations(false), LegalTypes(false), AA(A) { - AttributeSet FnAttrs = - DAG.getMachineFunction().getFunction()->getAttributes(); - ForCodeSize = - FnAttrs.hasAttribute(AttributeSet::FunctionIndex, - Attribute::OptimizeForSize) || - FnAttrs.hasAttribute(AttributeSet::FunctionIndex, Attribute::MinSize); + auto *F = DAG.getMachineFunction().getFunction(); + ForCodeSize = F->hasFnAttribute(Attribute::OptimizeForSize) || + F->hasFnAttribute(Attribute::MinSize); } /// Runs the dag combiner on all nodes in the work list @@ -446,7 +475,7 @@ void TargetLowering::DAGCombinerInfo::RemoveFromWorklist(SDNode *N) { } SDValue TargetLowering::DAGCombinerInfo:: -CombineTo(SDNode *N, const std::vector<SDValue> &To, bool AddTo) { +CombineTo(SDNode *N, ArrayRef<SDValue> To, bool AddTo) { return ((DAGCombiner*)DC)->CombineTo(N, &To[0], To.size(), AddTo); } @@ -566,7 +595,7 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG, case ISD::ConstantFP: { APFloat V = cast<ConstantFPSDNode>(Op)->getValueAPF(); V.changeSign(); - return DAG.getConstantFP(V, Op.getValueType()); + return DAG.getConstantFP(V, SDLoc(Op), Op.getValueType()); } case ISD::FADD: // FIXME: determine better conditions for this xform. @@ -590,7 +619,7 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG, // fold (fneg (fsub 0, B)) -> B if (ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(Op.getOperand(0))) - if (N0CFP->getValueAPF().isZero()) + if (N0CFP->isZero()) return Op.getOperand(1); // fold (fneg (fsub A, B)) -> (fsub B, A) @@ -683,13 +712,23 @@ static bool isConstantSplatVector(SDNode *N, APInt& SplatValue) { 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; } @@ -735,10 +774,10 @@ SDValue DAGCombiner::ReassociateOps(unsigned Opc, SDLoc DL, 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)) + if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, L, R)) return DAG.getNode(Opc, DL, VT, N0.getOperand(0), OpNode); return SDValue(); } @@ -755,10 +794,10 @@ SDValue DAGCombiner::ReassociateOps(unsigned Opc, SDLoc DL, } 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)) + if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, R, L)) return DAG.getNode(Opc, DL, VT, N1.getOperand(0), OpNode); return SDValue(); } @@ -1309,6 +1348,7 @@ SDValue DAGCombiner::visit(SDNode *N) { case ISD::ZERO_EXTEND: return visitZERO_EXTEND(N); case ISD::ANY_EXTEND: return visitANY_EXTEND(N); case ISD::SIGN_EXTEND_INREG: return visitSIGN_EXTEND_INREG(N); + case ISD::SIGN_EXTEND_VECTOR_INREG: return visitSIGN_EXTEND_VECTOR_INREG(N); case ISD::TRUNCATE: return visitTRUNCATE(N); case ISD::BITCAST: return visitBITCAST(N); case ISD::BUILD_PAIR: return visitBUILD_PAIR(N); @@ -1344,9 +1384,13 @@ SDValue DAGCombiner::visit(SDNode *N) { 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::MGATHER: return visitMGATHER(N); case ISD::MLOAD: return visitMLOAD(N); + case ISD::MSCATTER: return visitMSCATTER(N); case ISD::MSTORE: return visitMSTORE(N); + case ISD::FP_TO_FP16: return visitFP_TO_FP16(N); } return SDValue(); } @@ -1412,9 +1456,10 @@ SDValue DAGCombiner::combine(SDNode *N) { SDNode *CSENode; if (const BinaryWithFlagsSDNode *BinNode = dyn_cast<BinaryWithFlagsSDNode>(N)) { - CSENode = DAG.getNodeIfExists( - N->getOpcode(), N->getVTList(), Ops, BinNode->hasNoUnsignedWrap(), - BinNode->hasNoSignedWrap(), BinNode->isExact()); + CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops, + BinNode->Flags.hasNoUnsignedWrap(), + BinNode->Flags.hasNoSignedWrap(), + BinNode->Flags.hasExact()); } else { CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops); } @@ -1471,7 +1516,7 @@ SDValue DAGCombiner::visitTokenFactor(SDNode *N) { switch (Op.getOpcode()) { case ISD::EntryToken: // Entry tokens don't need to be added to the list. They are - // rededundant. + // redundant. Changed = true; break; @@ -1500,7 +1545,7 @@ SDValue DAGCombiner::visitTokenFactor(SDNode *N) { SDValue Result; - // If we've change things around then replace token factor. + // If we've changed things around then replace token factor. if (Changed) { if (Ops.empty()) { // The entry token is the only possible outcome. @@ -1510,8 +1555,11 @@ SDValue DAGCombiner::visitTokenFactor(SDNode *N) { Result = DAG.getNode(ISD::TokenFactor, SDLoc(N), MVT::Other, Ops); } - // Don't add users to work list. - return CombineTo(N, Result, false); + // Add users to worklist if AA is enabled, since it may introduce + // a lot of new chained token factors while removing memory deps. + bool UseAA = CombinerAA.getNumOccurrences() > 0 ? CombinerAA + : DAG.getSubtarget().useAA(); + return CombineTo(N, Result, UseAA /*add to worklist*/); } return Result; @@ -1534,17 +1582,42 @@ SDValue DAGCombiner::visitMERGE_VALUES(SDNode *N) { return SDValue(N, 0); // Return N so it doesn't get rechecked! } +static bool isNullConstant(SDValue V) { + ConstantSDNode *Const = dyn_cast<ConstantSDNode>(V); + return Const != nullptr && Const->isNullValue(); +} + +static bool isNullFPConstant(SDValue V) { + ConstantFPSDNode *Const = dyn_cast<ConstantFPSDNode>(V); + return Const != nullptr && Const->isZero() && !Const->isNegative(); +} + +static bool isAllOnesConstant(SDValue V) { + ConstantSDNode *Const = dyn_cast<ConstantSDNode>(V); + return Const != nullptr && Const->isAllOnesValue(); +} + +static bool isOneConstant(SDValue V) { + ConstantSDNode *Const = dyn_cast<ConstantSDNode>(V); + return Const != nullptr && Const->isOne(); +} + +/// If \p N is a ContantSDNode with isOpaque() == false return it casted to a +/// ContantSDNode pointer else nullptr. +static ConstantSDNode *getAsNonOpaqueConstant(SDValue N) { + ConstantSDNode *Const = dyn_cast<ConstantSDNode>(N); + return Const != nullptr && !Const->isOpaque() ? Const : nullptr; +} + SDValue DAGCombiner::visitADD(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); // 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())) @@ -1559,13 +1632,16 @@ SDValue DAGCombiner::visitADD(SDNode *N) { if (N1.getOpcode() == ISD::UNDEF) return N1; // fold (add c1, c2) -> c1+c2 + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + ConstantSDNode *N1C = getAsNonOpaqueConstant(N1); if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::ADD, VT, N0C, N1C); + return DAG.FoldConstantArithmetic(ISD::ADD, SDLoc(N), 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()) + if (isNullConstant(N1)) return N0; // fold (add Sym, c) -> Sym+c if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(N0)) @@ -1576,22 +1652,21 @@ SDValue DAGCombiner::visitADD(SDNode *N) { (uint64_t)N1C->getSExtValue()); // fold ((c1-A)+c2) -> (c1+c2)-A if (N1C && N0.getOpcode() == ISD::SUB) - if (ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.getOperand(0))) - return DAG.getNode(ISD::SUB, SDLoc(N), VT, + if (ConstantSDNode *N0C = getAsNonOpaqueConstant(N0.getOperand(0))) { + SDLoc DL(N); + return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(N1C->getAPIntValue()+ - N0C->getAPIntValue(), VT), + N0C->getAPIntValue(), DL, 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)) && - cast<ConstantSDNode>(N0.getOperand(0))->isNullValue()) + if (N0.getOpcode() == ISD::SUB && isNullConstant(N0.getOperand(0))) return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1, N0.getOperand(1)); // fold (A + (0-B)) -> A-B - if (N1.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N1.getOperand(0)) && - cast<ConstantSDNode>(N1.getOperand(0))->isNullValue()) + if (N1.getOpcode() == ISD::SUB && isNullConstant(N1.getOperand(0))) return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, N1.getOperand(1)); // fold (A+(B-A)) -> B if (N1.getOpcode() == ISD::SUB && N0 == N1.getOperand(1)) @@ -1651,34 +1726,27 @@ SDValue DAGCombiner::visitADD(SDNode *N) { } // fold (add x, shl(0 - y, n)) -> sub(x, shl(y, n)) - if (N1.getOpcode() == ISD::SHL && - N1.getOperand(0).getOpcode() == ISD::SUB) - if (ConstantSDNode *C = - dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(0))) - if (C->getAPIntValue() == 0) - return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, - DAG.getNode(ISD::SHL, SDLoc(N), VT, - N1.getOperand(0).getOperand(1), - N1.getOperand(1))); - if (N0.getOpcode() == ISD::SHL && - N0.getOperand(0).getOpcode() == ISD::SUB) - if (ConstantSDNode *C = - dyn_cast<ConstantSDNode>(N0.getOperand(0).getOperand(0))) - if (C->getAPIntValue() == 0) - return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1, - DAG.getNode(ISD::SHL, SDLoc(N), VT, - N0.getOperand(0).getOperand(1), - N0.getOperand(1))); + if (N1.getOpcode() == ISD::SHL && N1.getOperand(0).getOpcode() == ISD::SUB && + isNullConstant(N1.getOperand(0).getOperand(0))) + return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, + DAG.getNode(ISD::SHL, SDLoc(N), VT, + N1.getOperand(0).getOperand(1), + N1.getOperand(1))); + if (N0.getOpcode() == ISD::SHL && N0.getOperand(0).getOpcode() == ISD::SUB && + isNullConstant(N0.getOperand(0).getOperand(0))) + return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1, + DAG.getNode(ISD::SHL, SDLoc(N), VT, + N0.getOperand(0).getOperand(1), + N0.getOperand(1))); if (N1.getOpcode() == ISD::AND) { SDValue AndOp0 = N1.getOperand(0); - ConstantSDNode *AndOp1 = dyn_cast<ConstantSDNode>(N1->getOperand(1)); unsigned NumSignBits = DAG.ComputeNumSignBits(AndOp0); unsigned DestBits = VT.getScalarType().getSizeInBits(); // (add z, (and (sbbl x, x), 1)) -> (sub z, (sbbl x, x)) // and similar xforms where the inner op is either ~0 or 0. - if (NumSignBits == DestBits && AndOp1 && AndOp1->isOne()) { + if (NumSignBits == DestBits && isOneConstant(N1->getOperand(1))) { SDLoc DL(N); return DAG.getNode(ISD::SUB, DL, VT, N->getOperand(0), AndOp0); } @@ -1699,7 +1767,7 @@ SDValue DAGCombiner::visitADD(SDNode *N) { if (TN->getVT() == MVT::i1) { SDLoc DL(N); SDValue ZExt = DAG.getNode(ISD::AND, DL, VT, N1.getOperand(0), - DAG.getConstant(1, VT)); + DAG.getConstant(1, DL, VT)); return DAG.getNode(ISD::SUB, DL, VT, N0, ZExt); } } @@ -1710,8 +1778,6 @@ SDValue DAGCombiner::visitADD(SDNode *N) { SDValue DAGCombiner::visitADDC(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); // If the flag result is dead, turn this into an ADD. @@ -1721,11 +1787,13 @@ SDValue DAGCombiner::visitADDC(SDNode *N) { SDLoc(N), MVT::Glue)); // canonicalize constant to RHS. + ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); + ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); if (N0C && !N1C) return DAG.getNode(ISD::ADDC, SDLoc(N), N->getVTList(), N1, N0); // fold (addc x, 0) -> x + no carry out - if (N1C && N1C->isNullValue()) + if (isNullConstant(N1)) return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE, SDLoc(N), MVT::Glue)); @@ -1752,10 +1820,10 @@ SDValue DAGCombiner::visitADDE(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); SDValue CarryIn = N->getOperand(2); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); // canonicalize constant to RHS + ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); + ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); if (N0C && !N1C) return DAG.getNode(ISD::ADDE, SDLoc(N), N->getVTList(), N1, N0, CarryIn); @@ -1773,25 +1841,21 @@ static SDValue tryFoldToZero(SDLoc DL, const TargetLowering &TLI, EVT VT, SelectionDAG &DAG, bool LegalOperations, bool LegalTypes) { if (!VT.isVector()) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, DL, VT); if (!LegalOperations || TLI.isOperationLegal(ISD::BUILD_VECTOR, VT)) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, DL, VT); return SDValue(); } SDValue DAGCombiner::visitSUB(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.getNode()); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode()); - ConstantSDNode *N1C1 = N1.getOpcode() != ISD::ADD ? nullptr : - dyn_cast<ConstantSDNode>(N1.getOperand(1).getNode()); EVT VT = N0.getValueType(); // 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())) @@ -1803,14 +1867,18 @@ SDValue DAGCombiner::visitSUB(SDNode *N) { if (N0 == N1) return tryFoldToZero(SDLoc(N), TLI, VT, DAG, LegalOperations, LegalTypes); // fold (sub c1, c2) -> c1-c2 + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + ConstantSDNode *N1C = getAsNonOpaqueConstant(N1); if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::SUB, VT, N0C, N1C); + return DAG.FoldConstantArithmetic(ISD::SUB, SDLoc(N), VT, N0C, N1C); // fold (sub x, c) -> (add x, -c) - if (N1C) - return DAG.getNode(ISD::ADD, SDLoc(N), VT, N0, - DAG.getConstant(-N1C->getAPIntValue(), VT)); + if (N1C) { + SDLoc DL(N); + return DAG.getNode(ISD::ADD, DL, VT, N0, + DAG.getConstant(-N1C->getAPIntValue(), DL, VT)); + } // Canonicalize (sub -1, x) -> ~x, i.e. (xor x, -1) - if (N0C && N0C->isAllOnesValue()) + if (isAllOnesConstant(N0)) return DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0); // fold A-(A-B) -> B if (N1.getOpcode() == ISD::SUB && N0 == N1.getOperand(0)) @@ -1822,10 +1890,13 @@ SDValue DAGCombiner::visitSUB(SDNode *N) { if (N0.getOpcode() == ISD::ADD && N0.getOperand(1) == N1) return N0.getOperand(0); // fold C2-(A+C1) -> (C2-C1)-A + ConstantSDNode *N1C1 = N1.getOpcode() != ISD::ADD ? nullptr : + dyn_cast<ConstantSDNode>(N1.getOperand(1).getNode()); if (N1.getOpcode() == ISD::ADD && N0C && N1C1) { + SDLoc DL(N); SDValue NewC = DAG.getConstant(N0C->getAPIntValue() - N1C1->getAPIntValue(), - VT); - return DAG.getNode(ISD::SUB, SDLoc(N), VT, NewC, + DL, VT); + return DAG.getNode(ISD::SUB, DL, VT, NewC, N1.getOperand(0)); } // fold ((A+(B+or-C))-B) -> A+or-C @@ -1866,7 +1937,7 @@ SDValue DAGCombiner::visitSUB(SDNode *N) { if (GlobalAddressSDNode *GB = dyn_cast<GlobalAddressSDNode>(N1)) if (GA->getGlobal() == GB->getGlobal()) return DAG.getConstant((uint64_t)GA->getOffset() - GB->getOffset(), - VT); + SDLoc(N), VT); } // sub X, (sextinreg Y i1) -> add X, (and Y 1) @@ -1875,7 +1946,7 @@ SDValue DAGCombiner::visitSUB(SDNode *N) { if (TN->getVT() == MVT::i1) { SDLoc DL(N); SDValue ZExt = DAG.getNode(ISD::AND, DL, VT, N1.getOperand(0), - DAG.getConstant(1, VT)); + DAG.getConstant(1, DL, VT)); return DAG.getNode(ISD::ADD, DL, VT, N0, ZExt); } } @@ -1886,8 +1957,6 @@ SDValue DAGCombiner::visitSUB(SDNode *N) { SDValue DAGCombiner::visitSUBC(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); // If the flag result is dead, turn this into an SUB. @@ -1897,18 +1966,20 @@ SDValue DAGCombiner::visitSUBC(SDNode *N) { MVT::Glue)); // fold (subc x, x) -> 0 + no borrow - if (N0 == N1) - return CombineTo(N, DAG.getConstant(0, VT), - DAG.getNode(ISD::CARRY_FALSE, SDLoc(N), + if (N0 == N1) { + SDLoc DL(N); + return CombineTo(N, DAG.getConstant(0, DL, VT), + DAG.getNode(ISD::CARRY_FALSE, DL, MVT::Glue)); + } // fold (subc x, 0) -> x + no borrow - if (N1C && N1C->isNullValue()) + if (isNullConstant(N1)) return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE, SDLoc(N), MVT::Glue)); // Canonicalize (sub -1, x) -> ~x, i.e. (xor x, -1) + no borrow - if (N0C && N0C->isAllOnesValue()) + if (isAllOnesConstant(N0)) return CombineTo(N, DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0), DAG.getNode(ISD::CARRY_FALSE, SDLoc(N), MVT::Glue)); @@ -1935,33 +2006,41 @@ SDValue DAGCombiner::visitMUL(SDNode *N) { // fold (mul x, undef) -> 0 if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); bool N0IsConst = false; bool N1IsConst = false; + bool N1IsOpaqueConst = false; + bool N0IsOpaqueConst = false; 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(); + N0IsOpaqueConst = cast<ConstantSDNode>(N0)->isOpaque(); + } + N1IsConst = isa<ConstantSDNode>(N1); + if (N1IsConst) { + ConstValue1 = cast<ConstantSDNode>(N1)->getAPIntValue(); + N1IsOpaqueConst = cast<ConstantSDNode>(N1)->isOpaque(); + } } // fold (mul c1, c2) -> c1*c2 - if (N0IsConst && N1IsConst) - return DAG.FoldConstantArithmetic(ISD::MUL, VT, N0.getNode(), N1.getNode()); + if (N0IsConst && N1IsConst && !N0IsOpaqueConst && !N1IsOpaqueConst) + return DAG.FoldConstantArithmetic(ISD::MUL, SDLoc(N), 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) @@ -1974,23 +2053,30 @@ SDValue DAGCombiner::visitMUL(SDNode *N) { if (N1IsConst && ConstValue1 == 1 && IsFullSplat) return N0; // fold (mul x, -1) -> 0-x - if (N1IsConst && ConstValue1.isAllOnesValue()) - return DAG.getNode(ISD::SUB, SDLoc(N), VT, - DAG.getConstant(0, VT), N0); + if (N1IsConst && ConstValue1.isAllOnesValue()) { + SDLoc DL(N); + return DAG.getNode(ISD::SUB, DL, VT, + DAG.getConstant(0, DL, VT), N0); + } // fold (mul x, (1 << c)) -> x << c - if (N1IsConst && ConstValue1.isPowerOf2() && IsFullSplat) - return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0, - DAG.getConstant(ConstValue1.logBase2(), + if (N1IsConst && !N1IsOpaqueConst && ConstValue1.isPowerOf2() && + IsFullSplat) { + SDLoc DL(N); + return DAG.getNode(ISD::SHL, DL, VT, N0, + DAG.getConstant(ConstValue1.logBase2(), DL, getShiftAmountTy(N0.getValueType()))); + } // fold (mul x, -(1 << c)) -> -(x << c) or (-x) << c - if (N1IsConst && (-ConstValue1).isPowerOf2() && IsFullSplat) { + if (N1IsConst && !N1IsOpaqueConst && (-ConstValue1).isPowerOf2() && + IsFullSplat) { unsigned Log2Val = (-ConstValue1).logBase2(); + SDLoc DL(N); // FIXME: If the input is something that is easily negated (e.g. a // single-use add), we should put the negate there. - return DAG.getNode(ISD::SUB, SDLoc(N), VT, - DAG.getConstant(0, VT), - DAG.getNode(ISD::SHL, SDLoc(N), VT, N0, - DAG.getConstant(Log2Val, + return DAG.getNode(ISD::SUB, DL, VT, + DAG.getConstant(0, DL, VT), + DAG.getNode(ISD::SHL, DL, VT, N0, + DAG.getConstant(Log2Val, DL, getShiftAmountTy(N0.getValueType())))); } @@ -2041,8 +2127,7 @@ SDValue DAGCombiner::visitMUL(SDNode *N) { 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(); @@ -2051,26 +2136,27 @@ SDValue DAGCombiner::visitMUL(SDNode *N) { SDValue DAGCombiner::visitSDIV(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = isConstOrConstSplat(N0); - ConstantSDNode *N1C = isConstOrConstSplat(N1); 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 - if (N0C && N1C && !N1C->isNullValue()) - return DAG.FoldConstantArithmetic(ISD::SDIV, VT, N0C, N1C); + ConstantSDNode *N0C = isConstOrConstSplat(N0); + ConstantSDNode *N1C = isConstOrConstSplat(N1); + if (N0C && N1C && !N0C->isOpaque() && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::SDIV, SDLoc(N), VT, N0C, N1C); // fold (sdiv X, 1) -> X - if (N1C && N1C->getAPIntValue() == 1LL) + if (N1C && N1C->isOne()) return N0; // fold (sdiv X, -1) -> 0-X - if (N1C && N1C->isAllOnesValue()) - return DAG.getNode(ISD::SUB, SDLoc(N), VT, - DAG.getConstant(0, VT), N0); + if (N1C && N1C->isAllOnesValue()) { + SDLoc DL(N); + return DAG.getNode(ISD::SUB, DL, VT, + DAG.getConstant(0, DL, VT), N0); + } // If we know the sign bits of both operands are zero, strength reduce to a // udiv instead. Handles (X&15) /s 4 -> X&15 >> 2 if (!VT.isVector()) { @@ -2080,8 +2166,9 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) { } // fold (sdiv X, pow2) -> simple ops after legalize - if (N1C && !N1C->isNullValue() && (N1C->getAPIntValue().isPowerOf2() || - (-N1C->getAPIntValue()).isPowerOf2())) { + if (N1C && !N1C->isNullValue() && !N1C->isOpaque() && + (N1C->getAPIntValue().isPowerOf2() || + (-N1C->getAPIntValue()).isPowerOf2())) { // If dividing by powers of two is cheap, then don't perform the following // fold. if (TLI.isPow2SDivCheap()) @@ -2093,24 +2180,26 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) { return Res; unsigned lg2 = N1C->getAPIntValue().countTrailingZeros(); + SDLoc DL(N); // Splat the sign bit into the register SDValue SGN = - DAG.getNode(ISD::SRA, SDLoc(N), VT, N0, - DAG.getConstant(VT.getScalarSizeInBits() - 1, + DAG.getNode(ISD::SRA, DL, VT, N0, + DAG.getConstant(VT.getScalarSizeInBits() - 1, DL, getShiftAmountTy(N0.getValueType()))); AddToWorklist(SGN.getNode()); // Add (N0 < 0) ? abs2 - 1 : 0; SDValue SRL = - DAG.getNode(ISD::SRL, SDLoc(N), VT, SGN, - DAG.getConstant(VT.getScalarSizeInBits() - lg2, + DAG.getNode(ISD::SRL, DL, VT, SGN, + DAG.getConstant(VT.getScalarSizeInBits() - lg2, DL, getShiftAmountTy(SGN.getValueType()))); - SDValue ADD = DAG.getNode(ISD::ADD, SDLoc(N), VT, N0, SRL); + SDValue ADD = DAG.getNode(ISD::ADD, DL, VT, N0, SRL); AddToWorklist(SRL.getNode()); AddToWorklist(ADD.getNode()); // Divide by pow2 - SDValue SRA = DAG.getNode(ISD::SRA, SDLoc(N), VT, ADD, - DAG.getConstant(lg2, getShiftAmountTy(ADD.getValueType()))); + SDValue SRA = DAG.getNode(ISD::SRA, DL, VT, ADD, + DAG.getConstant(lg2, DL, + getShiftAmountTy(ADD.getValueType()))); // If we're dividing by a positive value, we're done. Otherwise, we must // negate the result. @@ -2118,10 +2207,10 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) { return SRA; AddToWorklist(SRA.getNode()); - return DAG.getNode(ISD::SUB, SDLoc(N), VT, DAG.getConstant(0, VT), SRA); + return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, 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); @@ -2130,7 +2219,7 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) { // undef / X -> 0 if (N0.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // X / undef -> undef if (N1.getOpcode() == ISD::UNDEF) return N1; @@ -2141,36 +2230,40 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) { SDValue DAGCombiner::visitUDIV(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = isConstOrConstSplat(N0); - ConstantSDNode *N1C = isConstOrConstSplat(N1); 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 - if (N0C && N1C && !N1C->isNullValue()) - return DAG.FoldConstantArithmetic(ISD::UDIV, VT, N0C, N1C); + ConstantSDNode *N0C = isConstOrConstSplat(N0); + ConstantSDNode *N1C = isConstOrConstSplat(N1); + if (N0C && N1C) + if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::UDIV, SDLoc(N), VT, + N0C, N1C)) + return Folded; // fold (udiv x, (1 << c)) -> x >>u c - if (N1C && N1C->getAPIntValue().isPowerOf2()) - return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, - DAG.getConstant(N1C->getAPIntValue().logBase2(), + if (N1C && !N1C->isOpaque() && N1C->getAPIntValue().isPowerOf2()) { + SDLoc DL(N); + return DAG.getNode(ISD::SRL, DL, VT, N0, + DAG.getConstant(N1C->getAPIntValue().logBase2(), DL, getShiftAmountTy(N0.getValueType()))); + } // fold (udiv x, (shl c, y)) -> x >>u (log2(c)+y) iff c is power of 2 if (N1.getOpcode() == ISD::SHL) { - if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) { + if (ConstantSDNode *SHC = getAsNonOpaqueConstant(N1.getOperand(0))) { if (SHC->getAPIntValue().isPowerOf2()) { EVT ADDVT = N1.getOperand(1).getValueType(); - SDValue Add = DAG.getNode(ISD::ADD, SDLoc(N), ADDVT, + SDLoc DL(N); + SDValue Add = DAG.getNode(ISD::ADD, DL, ADDVT, N1.getOperand(1), DAG.getConstant(SHC->getAPIntValue() .logBase2(), - ADDVT)); + DL, ADDVT)); AddToWorklist(Add.getNode()); - return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, Add); + return DAG.getNode(ISD::SRL, DL, VT, N0, Add); } } } @@ -2182,7 +2275,7 @@ SDValue DAGCombiner::visitUDIV(SDNode *N) { // undef / X -> 0 if (N0.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // X / undef -> undef if (N1.getOpcode() == ISD::UNDEF) return N1; @@ -2193,13 +2286,15 @@ SDValue DAGCombiner::visitUDIV(SDNode *N) { SDValue DAGCombiner::visitSREM(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = isConstOrConstSplat(N0); - ConstantSDNode *N1C = isConstOrConstSplat(N1); EVT VT = N->getValueType(0); // fold (srem c1, c2) -> c1%c2 - if (N0C && N1C && !N1C->isNullValue()) - return DAG.FoldConstantArithmetic(ISD::SREM, VT, N0C, N1C); + ConstantSDNode *N0C = isConstOrConstSplat(N0); + ConstantSDNode *N1C = isConstOrConstSplat(N1); + if (N0C && N1C) + if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::SREM, SDLoc(N), VT, + N0C, N1C)) + return Folded; // If we know the sign bits of both operands are zero, strength reduce to a // urem instead. Handles (X & 0x0FFFFFFF) %s 16 -> X&15 if (!VT.isVector()) { @@ -2224,7 +2319,7 @@ SDValue DAGCombiner::visitSREM(SDNode *N) { // undef % X -> 0 if (N0.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // X % undef -> undef if (N1.getOpcode() == ISD::UNDEF) return N1; @@ -2235,27 +2330,33 @@ SDValue DAGCombiner::visitSREM(SDNode *N) { SDValue DAGCombiner::visitUREM(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = isConstOrConstSplat(N0); - ConstantSDNode *N1C = isConstOrConstSplat(N1); EVT VT = N->getValueType(0); // fold (urem c1, c2) -> c1%c2 - if (N0C && N1C && !N1C->isNullValue()) - return DAG.FoldConstantArithmetic(ISD::UREM, VT, N0C, N1C); + ConstantSDNode *N0C = isConstOrConstSplat(N0); + ConstantSDNode *N1C = isConstOrConstSplat(N1); + if (N0C && N1C) + if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::UREM, SDLoc(N), VT, + N0C, N1C)) + return Folded; // fold (urem x, pow2) -> (and x, pow2-1) - if (N1C && !N1C->isNullValue() && N1C->getAPIntValue().isPowerOf2()) - return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, - DAG.getConstant(N1C->getAPIntValue()-1,VT)); + if (N1C && !N1C->isNullValue() && !N1C->isOpaque() && + N1C->getAPIntValue().isPowerOf2()) { + SDLoc DL(N); + return DAG.getNode(ISD::AND, DL, VT, N0, + DAG.getConstant(N1C->getAPIntValue() - 1, DL, VT)); + } // fold (urem x, (shl pow2, y)) -> (and x, (add (shl pow2, y), -1)) if (N1.getOpcode() == ISD::SHL) { - if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) { + if (ConstantSDNode *SHC = getAsNonOpaqueConstant(N1.getOperand(0))) { if (SHC->getAPIntValue().isPowerOf2()) { + SDLoc DL(N); SDValue Add = - DAG.getNode(ISD::ADD, SDLoc(N), VT, N1, - DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), + DAG.getNode(ISD::ADD, DL, VT, N1, + DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), DL, VT)); AddToWorklist(Add.getNode()); - return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, Add); + return DAG.getNode(ISD::AND, DL, VT, N0, Add); } } } @@ -2277,7 +2378,7 @@ SDValue DAGCombiner::visitUREM(SDNode *N) { // undef % X -> 0 if (N0.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // X % undef -> undef if (N1.getOpcode() == ISD::UNDEF) return N1; @@ -2288,21 +2389,23 @@ SDValue DAGCombiner::visitUREM(SDNode *N) { SDValue DAGCombiner::visitMULHS(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N->getValueType(0); SDLoc DL(N); // fold (mulhs x, 0) -> 0 - if (N1C && N1C->isNullValue()) + if (isNullConstant(N1)) return N1; // fold (mulhs x, 1) -> (sra x, size(x)-1) - if (N1C && N1C->getAPIntValue() == 1) - return DAG.getNode(ISD::SRA, SDLoc(N), N0.getValueType(), N0, + if (isOneConstant(N1)) { + SDLoc DL(N); + return DAG.getNode(ISD::SRA, DL, N0.getValueType(), N0, DAG.getConstant(N0.getValueType().getSizeInBits() - 1, + DL, getShiftAmountTy(N0.getValueType()))); + } // fold (mulhs x, undef) -> 0 if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // If the type twice as wide is legal, transform the mulhs to a wider multiply // plus a shift. @@ -2315,7 +2418,8 @@ SDValue DAGCombiner::visitMULHS(SDNode *N) { N1 = DAG.getNode(ISD::SIGN_EXTEND, DL, NewVT, N1); N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1); N1 = DAG.getNode(ISD::SRL, DL, NewVT, N1, - DAG.getConstant(SimpleSize, getShiftAmountTy(N1.getValueType()))); + DAG.getConstant(SimpleSize, DL, + getShiftAmountTy(N1.getValueType()))); return DAG.getNode(ISD::TRUNCATE, DL, VT, N1); } } @@ -2326,19 +2430,18 @@ SDValue DAGCombiner::visitMULHS(SDNode *N) { SDValue DAGCombiner::visitMULHU(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N->getValueType(0); SDLoc DL(N); // fold (mulhu x, 0) -> 0 - if (N1C && N1C->isNullValue()) + if (isNullConstant(N1)) return N1; // fold (mulhu x, 1) -> 0 - if (N1C && N1C->getAPIntValue() == 1) - return DAG.getConstant(0, N0.getValueType()); + if (isOneConstant(N1)) + return DAG.getConstant(0, DL, N0.getValueType()); // fold (mulhu x, undef) -> 0 if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, DL, VT); // If the type twice as wide is legal, transform the mulhu to a wider multiply // plus a shift. @@ -2351,7 +2454,8 @@ SDValue DAGCombiner::visitMULHU(SDNode *N) { N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, NewVT, N1); N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1); N1 = DAG.getNode(ISD::SRL, DL, NewVT, N1, - DAG.getConstant(SimpleSize, getShiftAmountTy(N1.getValueType()))); + DAG.getConstant(SimpleSize, DL, + getShiftAmountTy(N1.getValueType()))); return DAG.getNode(ISD::TRUNCATE, DL, VT, N1); } } @@ -2417,8 +2521,8 @@ SDValue DAGCombiner::visitSMUL_LOHI(SDNode *N) { 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(); @@ -2429,7 +2533,8 @@ SDValue DAGCombiner::visitSMUL_LOHI(SDNode *N) { Lo = DAG.getNode(ISD::MUL, DL, NewVT, Lo, Hi); // Compute the high part as N1. Hi = DAG.getNode(ISD::SRL, DL, NewVT, Lo, - DAG.getConstant(SimpleSize, getShiftAmountTy(Lo.getValueType()))); + DAG.getConstant(SimpleSize, DL, + getShiftAmountTy(Lo.getValueType()))); Hi = DAG.getNode(ISD::TRUNCATE, DL, VT, Hi); // Compute the low part as N0. Lo = DAG.getNode(ISD::TRUNCATE, DL, VT, Lo); @@ -2447,8 +2552,8 @@ SDValue DAGCombiner::visitUMUL_LOHI(SDNode *N) { 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(); @@ -2459,7 +2564,8 @@ SDValue DAGCombiner::visitUMUL_LOHI(SDNode *N) { Lo = DAG.getNode(ISD::MUL, DL, NewVT, Lo, Hi); // Compute the high part as N1. Hi = DAG.getNode(ISD::SRL, DL, NewVT, Lo, - DAG.getConstant(SimpleSize, getShiftAmountTy(Lo.getValueType()))); + DAG.getConstant(SimpleSize, DL, + getShiftAmountTy(Lo.getValueType()))); Hi = DAG.getNode(ISD::TRUNCATE, DL, VT, Hi); // Compute the low part as N0. Lo = DAG.getNode(ISD::TRUNCATE, DL, VT, Lo); @@ -2615,7 +2721,7 @@ SDValue DAGCombiner::SimplifyBinOpWithSameOpcodeHands(SDNode *N) { // build vector of all zeros that might be illegal at this stage. if (N->getOpcode() == ISD::XOR && ShOp.getOpcode() != ISD::UNDEF) { if (!LegalTypes) - ShOp = DAG.getConstant(0, VT); + ShOp = DAG.getConstant(0, SDLoc(N), VT); else ShOp = SDValue(); } @@ -2636,7 +2742,7 @@ SDValue DAGCombiner::SimplifyBinOpWithSameOpcodeHands(SDNode *N) { ShOp = N0->getOperand(0); if (N->getOpcode() == ISD::XOR && ShOp.getOpcode() != ISD::UNDEF) { if (!LegalTypes) - ShOp = DAG.getConstant(0, VT); + ShOp = DAG.getConstant(0, SDLoc(N), VT); else ShOp = SDValue(); } @@ -2657,19 +2763,122 @@ SDValue DAGCombiner::SimplifyBinOpWithSameOpcodeHands(SDNode *N) { 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, SDLoc(LocReference), 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 (isNullConstant(LR) && 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); + } + if (isAllOnesConstant(LR)) { + // fold (and (seteq X, -1), (seteq Y, -1)) -> (seteq (and X, Y), -1) + if (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 (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 && ((isNullConstant(LR) && isAllOnesConstant(RR)) || + (isAllOnesConstant(LR) && isNullConstant(RR)))) { + SDLoc DL(N0); + SDValue ADDNode = DAG.getNode(ISD::ADD, DL, LL.getValueType(), + LL, DAG.getConstant(1, DL, + LL.getValueType())); + AddToWorklist(ADDNode.getNode()); + return DAG.getSetCC(SDLoc(LocReference), VT, ADDNode, + DAG.getConstant(2, DL, 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())) { + SDLoc DL(N0); + SDValue NewAdd = + DAG.getNode(ISD::ADD, DL, VT, + N0.getOperand(0), DAG.getConstant(ADDC, DL, 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); - SDValue LL, LR, RL, RR, CC0, CC1; - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N1.getValueType(); - unsigned BitWidth = VT.getScalarType().getSizeInBits(); // 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())) @@ -2677,13 +2886,13 @@ SDValue DAGCombiner::visitAND(SDNode *N) { return DAG.getConstant( APInt::getNullValue( N0.getValueType().getScalarType().getSizeInBits()), - N0.getValueType()); + SDLoc(N), N0.getValueType()); if (ISD::isBuildVectorAllZeros(N1.getNode())) // do not return N1, because undef node may exist in N1 return DAG.getConstant( APInt::getNullValue( N1.getValueType().getScalarType().getSizeInBits()), - N1.getValueType()); + SDLoc(N), N1.getValueType()); // fold (and x, -1) -> x, vector edition if (ISD::isBuildVectorAllOnes(N0.getNode())) @@ -2692,25 +2901,25 @@ SDValue DAGCombiner::visitAND(SDNode *N) { 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 - if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::AND, VT, N0C, N1C); + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); + if (N0C && N1C && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::AND, SDLoc(N), 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()) + if (isAllOnesConstant(N1)) return N0; // if (and x, c) is known to be zero, return 0 + unsigned BitWidth = VT.getScalarType().getSizeInBits(); if (N1C && DAG.MaskedValueIsZero(SDValue(N, 0), APInt::getAllOnesValue(BitWidth))) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), 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) @@ -2840,117 +3049,6 @@ SDValue DAGCombiner::visitAND(SDNode *N) { return SDValue(N, 0); // Return N so it doesn't get rechecked! } } - // fold (and (setcc x), (setcc y)) -> (setcc (and x, y)) - 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) @@ -3001,8 +3099,9 @@ SDValue DAGCombiner::visitAND(SDNode *N) { unsigned LVTStoreBytes = LoadedVT.getStoreSize(); unsigned EVTStoreBytes = ExtVT.getStoreSize(); unsigned PtrOff = LVTStoreBytes - EVTStoreBytes; - NewPtr = DAG.getNode(ISD::ADD, SDLoc(LN0), PtrType, - NewPtr, DAG.getConstant(PtrOff, PtrType)); + SDLoc DL(LN0); + NewPtr = DAG.getNode(ISD::ADD, DL, PtrType, + NewPtr, DAG.getConstant(PtrOff, DL, PtrType)); Alignment = MinAlign(Alignment, PtrOff); } @@ -3022,33 +3121,60 @@ SDValue DAGCombiner::visitAND(SDNode *N) { } } - 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), @@ -3159,9 +3285,12 @@ SDValue DAGCombiner::MatchBSwapHWordLow(SDNode *N, SDValue N0, SDValue N1, } SDValue Res = DAG.getNode(ISD::BSWAP, SDLoc(N), VT, N00); - if (OpSizeInBits > 16) - Res = DAG.getNode(ISD::SRL, SDLoc(N), VT, Res, - DAG.getConstant(OpSizeInBits-16, getShiftAmountTy(VT))); + if (OpSizeInBits > 16) { + SDLoc DL(N); + Res = DAG.getNode(ISD::SRL, DL, VT, Res, + DAG.getConstant(OpSizeInBits - 16, DL, + getShiftAmountTy(VT))); + } return Res; } @@ -3299,33 +3428,125 @@ SDValue DAGCombiner::MatchBSwapHWord(SDNode *N, SDValue N0, SDValue N1) { if (Parts[0] != Parts[1] || Parts[0] != Parts[2] || Parts[0] != Parts[3]) return SDValue(); - SDValue BSwap = DAG.getNode(ISD::BSWAP, SDLoc(N), VT, - SDValue(Parts[0],0)); + SDLoc DL(N); + SDValue BSwap = DAG.getNode(ISD::BSWAP, DL, VT, + SDValue(Parts[0], 0)); // Result of the bswap should be rotated by 16. If it's not legal, then // do (x << 16) | (x >> 16). - SDValue ShAmt = DAG.getConstant(16, getShiftAmountTy(VT)); + SDValue ShAmt = DAG.getConstant(16, DL, getShiftAmountTy(VT)); if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT)) - return DAG.getNode(ISD::ROTL, SDLoc(N), VT, BSwap, ShAmt); + return DAG.getNode(ISD::ROTL, DL, VT, BSwap, ShAmt); if (TLI.isOperationLegalOrCustom(ISD::ROTR, VT)) - return DAG.getNode(ISD::ROTR, SDLoc(N), VT, BSwap, ShAmt); - return DAG.getNode(ISD::OR, SDLoc(N), VT, - DAG.getNode(ISD::SHL, SDLoc(N), VT, BSwap, ShAmt), - DAG.getNode(ISD::SRL, SDLoc(N), VT, BSwap, ShAmt)); + return DAG.getNode(ISD::ROTR, DL, VT, BSwap, ShAmt); + return DAG.getNode(ISD::OR, DL, VT, + DAG.getNode(ISD::SHL, DL, VT, BSwap, ShAmt), + DAG.getNode(ISD::SRL, DL, 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()), + SDLoc(LocReference), 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 && 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 (isNullConstant(LR) && (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 (isAllOnesConstant(LR) && (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 && + // 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. + if (const ConstantSDNode *N0O1C = + getAsNonOpaqueConstant(N0.getOperand(1))) { + if (const ConstantSDNode *N1O1C = + getAsNonOpaqueConstant(N1.getOperand(1))) { + // 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 = N0O1C->getAPIntValue(); + const APInt &RHSMask = N1O1C->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)); + SDLoc DL(LocReference); + return DAG.getNode(ISD::AND, DL, VT, X, + DAG.getConstant(LHSMask | RHSMask, DL, 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); - SDValue LL, LR, RL, RR, CC0, CC1; - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N1.getValueType(); // 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())) @@ -3339,13 +3560,13 @@ SDValue DAGCombiner::visitOR(SDNode *N) { return DAG.getConstant( APInt::getAllOnesValue( N0.getValueType().getScalarType().getSizeInBits()), - N0.getValueType()); + SDLoc(N), N0.getValueType()); if (ISD::isBuildVectorAllOnes(N1.getNode())) // do not return N1, because undef node may exist in N1 return DAG.getConstant( APInt::getAllOnesValue( N1.getValueType().getScalarType().getSizeInBits()), - N1.getValueType()); + SDLoc(N), N1.getValueType()); // fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf A, B, Mask1) // fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf B, A, Mask2) @@ -3404,28 +3625,28 @@ SDValue DAGCombiner::visitOR(SDNode *N) { } } - // 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 - if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::OR, VT, N0C, N1C); + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); + if (N0C && N1C && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N), 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 (isNullConstant(N1)) return N0; // fold (or x, -1) -> -1 - if (N1C && N1C->isAllOnesValue()) + if (isAllOnesConstant(N1)) return N1; // fold (or x, c) -> c iff (x & ~c) == 0 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()) @@ -3435,8 +3656,7 @@ SDValue DAGCombiner::visitOR(SDNode *N) { 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. @@ -3444,86 +3664,20 @@ SDValue DAGCombiner::visitOR(SDNode *N) { isa<ConstantSDNode>(N0.getOperand(1))) { ConstantSDNode *C1 = cast<ConstantSDNode>(N0.getOperand(1)); if ((C1->getAPIntValue() & N1C->getAPIntValue()) != 0) { - if (SDValue COR = DAG.FoldConstantArithmetic(ISD::OR, VT, N1C, C1)) + if (SDValue COR = DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N1), VT, + N1C, C1)) return DAG.getNode( ISD::AND, SDLoc(N), VT, DAG.getNode(ISD::OR, SDLoc(N0), VT, N0.getOperand(0), N1), COR); return SDValue(); } } - // fold (or (setcc x), (setcc y)) -> (setcc (or x, y)) - 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)); - } - } - // See if this is some rotate idiom. if (SDNode *Rot = MatchRotate(N0, N1, SDLoc(N))) return SDValue(Rot, 0); @@ -3751,7 +3905,7 @@ SDNode *DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, SDLoc DL) { Mask &= cast<ConstantSDNode>(RHSMask)->getAPIntValue() | LHSBits; } - Rot = DAG.getNode(ISD::AND, DL, VT, Rot, DAG.getConstant(Mask, VT)); + Rot = DAG.getNode(ISD::AND, DL, VT, Rot, DAG.getConstant(Mask, DL, VT)); } return Rot.getNode(); @@ -3793,15 +3947,12 @@ SDNode *DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, SDLoc DL) { SDValue DAGCombiner::visitXOR(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - SDValue LHS, RHS, CC; - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); // 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())) @@ -3812,27 +3963,30 @@ SDValue DAGCombiner::visitXOR(SDNode *N) { // fold (xor undef, undef) -> 0. This is a common idiom (misuse). if (N0.getOpcode() == ISD::UNDEF && N1.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // fold (xor x, undef) -> undef if (N0.getOpcode() == ISD::UNDEF) return N0; if (N1.getOpcode() == ISD::UNDEF) return N1; // fold (xor c1, c2) -> c1^c2 + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + ConstantSDNode *N1C = getAsNonOpaqueConstant(N1); if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::XOR, VT, N0C, N1C); + return DAG.FoldConstantArithmetic(ISD::XOR, SDLoc(N), 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()) + if (isNullConstant(N1)) 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) + SDValue LHS, RHS, CC; if (TLI.isConstTrueVal(N1.getNode()) && isSetCCEquivalent(N0, LHS, RHS, CC)) { bool isInt = LHS.getValueType().isInteger(); ISD::CondCode NotCC = ISD::getSetCCInverse(cast<CondCodeSDNode>(CC)->get(), @@ -3853,18 +4007,19 @@ SDValue DAGCombiner::visitXOR(SDNode *N) { } // fold (not (zext (setcc x, y))) -> (zext (not (setcc x, y))) - if (N1C && N1C->getAPIntValue() == 1 && N0.getOpcode() == ISD::ZERO_EXTEND && + if (isOneConstant(N1) && N0.getOpcode() == ISD::ZERO_EXTEND && N0.getNode()->hasOneUse() && isSetCCEquivalent(N0.getOperand(0), LHS, RHS, CC)){ SDValue V = N0.getOperand(0); - V = DAG.getNode(ISD::XOR, SDLoc(N0), V.getValueType(), V, - DAG.getConstant(1, V.getValueType())); + SDLoc DL(N0); + V = DAG.getNode(ISD::XOR, DL, V.getValueType(), V, + DAG.getConstant(1, DL, V.getValueType())); AddToWorklist(V.getNode()); return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), VT, V); } // fold (not (or x, y)) -> (and (not x), (not y)) iff x or y are setcc - if (N1C && N1C->getAPIntValue() == 1 && VT == MVT::i1 && + if (isOneConstant(N1) && VT == MVT::i1 && (N0.getOpcode() == ISD::OR || N0.getOpcode() == ISD::AND)) { SDValue LHS = N0.getOperand(0), RHS = N0.getOperand(1); if (isOneUseSetCC(RHS) || isOneUseSetCC(LHS)) { @@ -3876,7 +4031,7 @@ SDValue DAGCombiner::visitXOR(SDNode *N) { } } // fold (not (or x, y)) -> (and (not x), (not y)) iff x or y are constants - if (N1C && N1C->isAllOnesValue() && + if (isAllOnesConstant(N1) && (N0.getOpcode() == ISD::OR || N0.getOpcode() == ISD::AND)) { SDValue LHS = N0.getOperand(0), RHS = N0.getOperand(1); if (isa<ConstantSDNode>(RHS) || isa<ConstantSDNode>(LHS)) { @@ -3897,21 +4052,48 @@ SDValue DAGCombiner::visitXOR(SDNode *N) { } // fold (xor (xor x, c1), c2) -> (xor x, (xor c1, c2)) if (N1C && N0.getOpcode() == ISD::XOR) { - ConstantSDNode *N00C = dyn_cast<ConstantSDNode>(N0.getOperand(0)); - ConstantSDNode *N01C = dyn_cast<ConstantSDNode>(N0.getOperand(1)); - if (N00C) - return DAG.getNode(ISD::XOR, SDLoc(N), VT, N0.getOperand(1), + if (const ConstantSDNode *N00C = getAsNonOpaqueConstant(N0.getOperand(0))) { + SDLoc DL(N); + return DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(1), DAG.getConstant(N1C->getAPIntValue() ^ - N00C->getAPIntValue(), VT)); - if (N01C) - return DAG.getNode(ISD::XOR, SDLoc(N), VT, N0.getOperand(0), + N00C->getAPIntValue(), DL, VT)); + } + if (const ConstantSDNode *N01C = getAsNonOpaqueConstant(N0.getOperand(1))) { + SDLoc DL(N); + return DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(0), DAG.getConstant(N1C->getAPIntValue() ^ - N01C->getAPIntValue(), VT)); + N01C->getAPIntValue(), DL, VT)); + } } // fold (xor x, x) -> 0 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) && N0.getOpcode() == ISD::SHL + && isAllOnesConstant(N1) && isOneConstant(N0.getOperand(0))) { + SDLoc DL(N); + return DAG.getNode(ISD::ROTL, DL, VT, DAG.getConstant(~1, DL, VT), + N0.getOperand(1)); + } + // Simplify: xor (op x...), (op y...) -> (op (xor x, y)) if (N0.getOpcode() == N1.getOpcode()) { SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N); @@ -3929,10 +4111,6 @@ SDValue DAGCombiner::visitXOR(SDNode *N) { /// Handle transforms common to the three shifts, when the shift amount is a /// constant. SDValue DAGCombiner::visitShiftByConstant(SDNode *N, ConstantSDNode *Amt) { - // We can't and shouldn't fold opaque constants. - if (Amt->isOpaque()) - return SDValue(); - SDNode *LHS = N->getOperand(0).getNode(); if (!LHS->hasOneUse()) return SDValue(); @@ -3959,8 +4137,8 @@ SDValue DAGCombiner::visitShiftByConstant(SDNode *N, ConstantSDNode *Amt) { } // We require the RHS of the binop to be a constant and not opaque as well. - ConstantSDNode *BinOpCst = dyn_cast<ConstantSDNode>(LHS->getOperand(1)); - if (!BinOpCst || BinOpCst->isOpaque()) return SDValue(); + ConstantSDNode *BinOpCst = getAsNonOpaqueConstant(LHS->getOperand(1)); + if (!BinOpCst) return SDValue(); // FIXME: disable this unless the input to the binop is a shift by a constant. // If it is not a shift, it pessimizes some common cases like: @@ -4013,14 +4191,17 @@ SDValue DAGCombiner::distributeTruncateThroughAnd(SDNode *N) { SDValue N01 = N->getOperand(0).getOperand(1); if (ConstantSDNode *N01C = isConstOrConstSplat(N01)) { - EVT TruncVT = N->getValueType(0); - SDValue N00 = N->getOperand(0).getOperand(0); - APInt TruncC = N01C->getAPIntValue(); - TruncC = TruncC.trunc(TruncVT.getScalarSizeInBits()); + if (!N01C->isOpaque()) { + EVT TruncVT = N->getValueType(0); + SDValue N00 = N->getOperand(0).getOperand(0); + APInt TruncC = N01C->getAPIntValue(); + TruncC = TruncC.trunc(TruncVT.getScalarSizeInBits()); + SDLoc DL(N); - return DAG.getNode(ISD::AND, SDLoc(N), TruncVT, - DAG.getNode(ISD::TRUNCATE, SDLoc(N), TruncVT, N00), - DAG.getConstant(TruncC, TruncVT)); + return DAG.getNode(ISD::AND, DL, TruncVT, + DAG.getNode(ISD::TRUNCATE, DL, TruncVT, N00), + DAG.getConstant(TruncC, DL, TruncVT)); + } } } @@ -4042,15 +4223,14 @@ SDValue DAGCombiner::visitRotate(SDNode *N) { SDValue DAGCombiner::visitSHL(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); unsigned OpSizeInBits = VT.getScalarSizeInBits(); // 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: @@ -4064,7 +4244,8 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { if (N01CV && N01CV->isConstant() && N00.getOpcode() == ISD::SETCC && TLI.getBooleanContents(N00.getOperand(0).getValueType()) == TargetLowering::ZeroOrNegativeOneBooleanContent) { - if (SDValue C = DAG.FoldConstantArithmetic(ISD::SHL, VT, N01CV, N1CV)) + if (SDValue C = DAG.FoldConstantArithmetic(ISD::SHL, SDLoc(N), VT, + N01CV, N1CV)) return DAG.getNode(ISD::AND, SDLoc(N), VT, N00, C); } } else { @@ -4074,10 +4255,11 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { } // fold (shl c1, c2) -> c1<<c2 - if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::SHL, VT, N0C, N1C); + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + if (N0C && N1C && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::SHL, SDLoc(N), VT, N0C, N1C); // fold (shl 0, x) -> 0 - if (N0C && N0C->isNullValue()) + if (isNullConstant(N0)) return N0; // fold (shl x, c >= size(x)) -> undef if (N1C && N1C->getZExtValue() >= OpSizeInBits) @@ -4087,11 +4269,11 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { return N0; // fold (shl undef, x) -> 0 if (N0.getOpcode() == ISD::UNDEF) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // if (shl x, c) is known to be zero, return 0 if (DAG.MaskedValueIsZero(SDValue(N, 0), APInt::getAllOnesValue(OpSizeInBits))) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // fold (shl x, (trunc (and y, c))) -> (shl x, (and (trunc y), (trunc c))). if (N1.getOpcode() == ISD::TRUNCATE && N1.getOperand(0).getOpcode() == ISD::AND) { @@ -4108,10 +4290,11 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { if (ConstantSDNode *N0C1 = isConstOrConstSplat(N0.getOperand(1))) { uint64_t c1 = N0C1->getZExtValue(); uint64_t c2 = N1C->getZExtValue(); + SDLoc DL(N); if (c1 + c2 >= OpSizeInBits) - return DAG.getConstant(0, VT); - return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(c1 + c2, N1.getValueType())); + return DAG.getConstant(0, DL, VT); + return DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0), + DAG.getConstant(c1 + c2, DL, N1.getValueType())); } } @@ -4131,12 +4314,13 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { EVT InnerShiftVT = N0Op0.getValueType(); uint64_t InnerShiftSize = InnerShiftVT.getScalarSizeInBits(); if (c2 >= OpSizeInBits - InnerShiftSize) { + SDLoc DL(N0); if (c1 + c2 >= OpSizeInBits) - return DAG.getConstant(0, VT); - return DAG.getNode(ISD::SHL, SDLoc(N0), VT, - DAG.getNode(N0.getOpcode(), SDLoc(N0), VT, + return DAG.getConstant(0, DL, VT); + return DAG.getNode(ISD::SHL, DL, VT, + DAG.getNode(N0.getOpcode(), DL, VT, N0Op0->getOperand(0)), - DAG.getConstant(c1 + c2, N1.getValueType())); + DAG.getConstant(c1 + c2, DL, N1.getValueType())); } } } @@ -4154,8 +4338,10 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { if (c1 == c2) { SDValue NewOp0 = N0.getOperand(0); EVT CountVT = NewOp0.getOperand(1).getValueType(); - SDValue NewSHL = DAG.getNode(ISD::SHL, SDLoc(N), NewOp0.getValueType(), - NewOp0, DAG.getConstant(c2, CountVT)); + SDLoc DL(N); + SDValue NewSHL = DAG.getNode(ISD::SHL, DL, NewOp0.getValueType(), + NewOp0, + DAG.getConstant(c2, DL, CountVT)); AddToWorklist(NewSHL.getNode()); return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N0), VT, NewSHL); } @@ -4176,25 +4362,30 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { SDValue Shift; if (c2 > c1) { Mask = Mask.shl(c2 - c1); - Shift = DAG.getNode(ISD::SHL, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(c2 - c1, N1.getValueType())); + SDLoc DL(N); + Shift = DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0), + DAG.getConstant(c2 - c1, DL, N1.getValueType())); } else { Mask = Mask.lshr(c1 - c2); - Shift = DAG.getNode(ISD::SRL, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(c1 - c2, N1.getValueType())); + SDLoc DL(N); + Shift = DAG.getNode(ISD::SRL, DL, VT, N0.getOperand(0), + DAG.getConstant(c1 - c2, DL, N1.getValueType())); } - return DAG.getNode(ISD::AND, SDLoc(N0), VT, Shift, - DAG.getConstant(Mask, VT)); + SDLoc DL(N0); + return DAG.getNode(ISD::AND, DL, VT, Shift, + DAG.getConstant(Mask, DL, VT)); } } } // fold (shl (sra x, c1), c1) -> (and x, (shl -1, c1)) if (N1C && N0.getOpcode() == ISD::SRA && N1 == N0.getOperand(1)) { unsigned BitSize = VT.getScalarSizeInBits(); + SDLoc DL(N); SDValue HiBitsMask = DAG.getConstant(APInt::getHighBitsSet(BitSize, - BitSize - N1C->getZExtValue()), VT); - return DAG.getNode(ISD::AND, SDLoc(N), VT, N0.getOperand(0), + BitSize - N1C->getZExtValue()), + DL, VT); + return DAG.getNode(ISD::AND, DL, VT, N0.getOperand(0), HiBitsMask); } @@ -4210,7 +4401,7 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { return DAG.getNode(ISD::ADD, SDLoc(N), VT, Shl0, Shl1); } - if (N1C) { + if (N1C && !N1C->isOpaque()) { SDValue NewSHL = visitShiftByConstant(N, N1C); if (NewSHL.getNode()) return NewSHL; @@ -4222,27 +4413,27 @@ SDValue DAGCombiner::visitSHL(SDNode *N) { SDValue DAGCombiner::visitSRA(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); unsigned OpSizeInBits = VT.getScalarType().getSizeInBits(); // 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 (sra c1, c2) -> (sra c1, c2) - if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::SRA, VT, N0C, N1C); + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + if (N0C && N1C && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::SRA, SDLoc(N), VT, N0C, N1C); // fold (sra 0, x) -> 0 - if (N0C && N0C->isNullValue()) + if (isNullConstant(N0)) return N0; // fold (sra -1, x) -> -1 - if (N0C && N0C->isAllOnesValue()) + if (isAllOnesConstant(N0)) return N0; // fold (sra x, (setge c, size(x))) -> undef if (N1C && N1C->getZExtValue() >= OpSizeInBits) @@ -4270,8 +4461,9 @@ SDValue DAGCombiner::visitSRA(SDNode *N) { unsigned Sum = N1C->getZExtValue() + C1->getZExtValue(); if (Sum >= OpSizeInBits) Sum = OpSizeInBits - 1; - return DAG.getNode(ISD::SRA, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(Sum, N1.getValueType())); + SDLoc DL(N); + return DAG.getNode(ISD::SRA, DL, VT, N0.getOperand(0), + DAG.getConstant(Sum, DL, N1.getValueType())); } } @@ -4303,14 +4495,15 @@ SDValue DAGCombiner::visitSRA(SDNode *N) { TLI.isOperationLegalOrCustom(ISD::TRUNCATE, VT) && TLI.isTruncateFree(VT, TruncVT)) { - SDValue Amt = DAG.getConstant(ShiftAmt, - getShiftAmountTy(N0.getOperand(0).getValueType())); - SDValue Shift = DAG.getNode(ISD::SRL, SDLoc(N0), VT, - N0.getOperand(0), Amt); - SDValue Trunc = DAG.getNode(ISD::TRUNCATE, SDLoc(N0), TruncVT, - Shift); - return DAG.getNode(ISD::SIGN_EXTEND, SDLoc(N), - N->getValueType(0), Trunc); + SDLoc DL(N); + SDValue Amt = DAG.getConstant(ShiftAmt, DL, + getShiftAmountTy(N0.getOperand(0).getValueType())); + SDValue Shift = DAG.getNode(ISD::SRL, DL, VT, + N0.getOperand(0), Amt); + SDValue Trunc = DAG.getNode(ISD::TRUNCATE, DL, TruncVT, + Shift); + return DAG.getNode(ISD::SIGN_EXTEND, DL, + N->getValueType(0), Trunc); } } } @@ -4337,12 +4530,13 @@ SDValue DAGCombiner::visitSRA(SDNode *N) { EVT LargeVT = N0Op0.getValueType(); if (LargeVT.getScalarSizeInBits() - OpSizeInBits == LargeShiftVal) { + SDLoc DL(N); SDValue Amt = - DAG.getConstant(LargeShiftVal + N1C->getZExtValue(), + DAG.getConstant(LargeShiftVal + N1C->getZExtValue(), DL, getShiftAmountTy(N0Op0.getOperand(0).getValueType())); - SDValue SRA = DAG.getNode(ISD::SRA, SDLoc(N), LargeVT, + SDValue SRA = DAG.getNode(ISD::SRA, DL, LargeVT, N0Op0.getOperand(0), Amt); - return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, SRA); + return DAG.getNode(ISD::TRUNCATE, DL, VT, SRA); } } } @@ -4356,7 +4550,7 @@ SDValue DAGCombiner::visitSRA(SDNode *N) { if (DAG.SignBitIsZero(N0)) return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, N1); - if (N1C) { + if (N1C && !N1C->isOpaque()) { SDValue NewSRA = visitShiftByConstant(N, N1C); if (NewSRA.getNode()) return NewSRA; @@ -4368,24 +4562,24 @@ SDValue DAGCombiner::visitSRA(SDNode *N) { SDValue DAGCombiner::visitSRL(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); EVT VT = N0.getValueType(); unsigned OpSizeInBits = VT.getScalarType().getSizeInBits(); // 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 (srl c1, c2) -> c1 >>u c2 - if (N0C && N1C) - return DAG.FoldConstantArithmetic(ISD::SRL, VT, N0C, N1C); + ConstantSDNode *N0C = getAsNonOpaqueConstant(N0); + if (N0C && N1C && !N1C->isOpaque()) + return DAG.FoldConstantArithmetic(ISD::SRL, SDLoc(N), VT, N0C, N1C); // fold (srl 0, x) -> 0 - if (N0C && N0C->isNullValue()) + if (isNullConstant(N0)) return N0; // fold (srl x, c >= size(x)) -> undef if (N1C && N1C->getZExtValue() >= OpSizeInBits) @@ -4396,17 +4590,18 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { // if (srl x, c) is known to be zero, return 0 if (N1C && DAG.MaskedValueIsZero(SDValue(N, 0), APInt::getAllOnesValue(OpSizeInBits))) - return DAG.getConstant(0, VT); + return DAG.getConstant(0, SDLoc(N), VT); // fold (srl (srl x, c1), c2) -> 0 or (srl x, (add c1, c2)) if (N1C && N0.getOpcode() == ISD::SRL) { if (ConstantSDNode *N01C = isConstOrConstSplat(N0.getOperand(1))) { uint64_t c1 = N01C->getZExtValue(); uint64_t c2 = N1C->getZExtValue(); + SDLoc DL(N); if (c1 + c2 >= OpSizeInBits) - return DAG.getConstant(0, VT); - return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(c1 + c2, N1.getValueType())); + return DAG.getConstant(0, DL, VT); + return DAG.getNode(ISD::SRL, DL, VT, N0.getOperand(0), + DAG.getConstant(c1 + c2, DL, N1.getValueType())); } } @@ -4422,12 +4617,14 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { uint64_t InnerShiftSize = InnerShiftVT.getScalarType().getSizeInBits(); // This is only valid if the OpSizeInBits + c1 = size of inner shift. if (c1 + OpSizeInBits == InnerShiftSize) { + SDLoc DL(N0); if (c1 + c2 >= InnerShiftSize) - return DAG.getConstant(0, VT); - return DAG.getNode(ISD::TRUNCATE, SDLoc(N0), VT, - DAG.getNode(ISD::SRL, SDLoc(N0), InnerShiftVT, + return DAG.getConstant(0, DL, VT); + return DAG.getNode(ISD::TRUNCATE, DL, VT, + DAG.getNode(ISD::SRL, DL, InnerShiftVT, N0.getOperand(0)->getOperand(0), - DAG.getConstant(c1 + c2, ShiftCountVT))); + DAG.getConstant(c1 + c2, DL, + ShiftCountVT))); } } @@ -4436,8 +4633,9 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { unsigned BitSize = N0.getScalarValueSizeInBits(); if (BitSize <= 64) { uint64_t ShAmt = N1C->getZExtValue() + 64 - BitSize; - return DAG.getNode(ISD::AND, SDLoc(N), VT, N0.getOperand(0), - DAG.getConstant(~0ULL >> ShAmt, VT)); + SDLoc DL(N); + return DAG.getNode(ISD::AND, DL, VT, N0.getOperand(0), + DAG.getConstant(~0ULL >> ShAmt, DL, VT)); } } @@ -4451,14 +4649,17 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { if (!LegalTypes || TLI.isTypeDesirableForOp(ISD::SRL, SmallVT)) { uint64_t ShiftAmt = N1C->getZExtValue(); - SDValue SmallShift = DAG.getNode(ISD::SRL, SDLoc(N0), SmallVT, + SDLoc DL0(N0); + SDValue SmallShift = DAG.getNode(ISD::SRL, DL0, SmallVT, N0.getOperand(0), - DAG.getConstant(ShiftAmt, getShiftAmountTy(SmallVT))); + DAG.getConstant(ShiftAmt, DL0, + getShiftAmountTy(SmallVT))); AddToWorklist(SmallShift.getNode()); APInt Mask = APInt::getAllOnesValue(OpSizeInBits).lshr(ShiftAmt); - return DAG.getNode(ISD::AND, SDLoc(N), VT, - DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, SmallShift), - DAG.getConstant(Mask, VT)); + SDLoc DL(N); + return DAG.getNode(ISD::AND, DL, VT, + DAG.getNode(ISD::ANY_EXTEND, DL, VT, SmallShift), + DAG.getConstant(Mask, DL, VT)); } } @@ -4477,12 +4678,12 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { // If any of the input bits are KnownOne, then the input couldn't be all // zeros, thus the result of the srl will always be zero. - if (KnownOne.getBoolValue()) return DAG.getConstant(0, VT); + if (KnownOne.getBoolValue()) return DAG.getConstant(0, SDLoc(N0), VT); // If all of the bits input the to ctlz node are known to be zero, then // the result of the ctlz is "32" and the result of the shift is one. APInt UnknownBits = ~KnownZero; - if (UnknownBits == 0) return DAG.getConstant(1, VT); + if (UnknownBits == 0) return DAG.getConstant(1, SDLoc(N0), VT); // Otherwise, check to see if there is exactly one bit input to the ctlz. if ((UnknownBits & (UnknownBits - 1)) == 0) { @@ -4494,13 +4695,16 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { SDValue Op = N0.getOperand(0); if (ShAmt) { - Op = DAG.getNode(ISD::SRL, SDLoc(N0), VT, Op, - DAG.getConstant(ShAmt, getShiftAmountTy(Op.getValueType()))); + SDLoc DL(N0); + Op = DAG.getNode(ISD::SRL, DL, VT, Op, + DAG.getConstant(ShAmt, DL, + getShiftAmountTy(Op.getValueType()))); AddToWorklist(Op.getNode()); } - return DAG.getNode(ISD::XOR, SDLoc(N), VT, - Op, DAG.getConstant(1, VT)); + SDLoc DL(N); + return DAG.getNode(ISD::XOR, DL, VT, + Op, DAG.getConstant(1, DL, VT)); } } @@ -4517,7 +4721,7 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { if (N1C && SimplifyDemandedBits(SDValue(N, 0))) return SDValue(N, 0); - if (N1C) { + if (N1C && !N1C->isOpaque()) { SDValue NewSRL = visitShiftByConstant(N, N1C); if (NewSRL.getNode()) return NewSRL; @@ -4565,7 +4769,7 @@ SDValue DAGCombiner::visitCTLZ(SDNode *N) { EVT VT = N->getValueType(0); // fold (ctlz c1) -> c2 - if (isa<ConstantSDNode>(N0)) + if (isConstantIntBuildVectorOrConstantInt(N0)) return DAG.getNode(ISD::CTLZ, SDLoc(N), VT, N0); return SDValue(); } @@ -4575,7 +4779,7 @@ SDValue DAGCombiner::visitCTLZ_ZERO_UNDEF(SDNode *N) { EVT VT = N->getValueType(0); // fold (ctlz_zero_undef c1) -> c2 - if (isa<ConstantSDNode>(N0)) + if (isConstantIntBuildVectorOrConstantInt(N0)) return DAG.getNode(ISD::CTLZ_ZERO_UNDEF, SDLoc(N), VT, N0); return SDValue(); } @@ -4585,7 +4789,7 @@ SDValue DAGCombiner::visitCTTZ(SDNode *N) { EVT VT = N->getValueType(0); // fold (cttz c1) -> c2 - if (isa<ConstantSDNode>(N0)) + if (isConstantIntBuildVectorOrConstantInt(N0)) return DAG.getNode(ISD::CTTZ, SDLoc(N), VT, N0); return SDValue(); } @@ -4595,7 +4799,7 @@ SDValue DAGCombiner::visitCTTZ_ZERO_UNDEF(SDNode *N) { EVT VT = N->getValueType(0); // fold (cttz_zero_undef c1) -> c2 - if (isa<ConstantSDNode>(N0)) + if (isConstantIntBuildVectorOrConstantInt(N0)) return DAG.getNode(ISD::CTTZ_ZERO_UNDEF, SDLoc(N), VT, N0); return SDValue(); } @@ -4605,7 +4809,7 @@ SDValue DAGCombiner::visitCTPOP(SDNode *N) { EVT VT = N->getValueType(0); // fold (ctpop c1) -> c2 - if (isa<ConstantSDNode>(N0)) + if (isConstantIntBuildVectorOrConstantInt(N0)) return DAG.getNode(ISD::CTPOP, SDLoc(N), VT, N0); return SDValue(); } @@ -4651,23 +4855,19 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); SDValue N2 = N->getOperand(2); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); - ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1); - ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2); EVT VT = N->getValueType(0); EVT VT0 = N0.getValueType(); // fold (select C, X, X) -> X if (N1 == N2) return N1; - // fold (select true, X, Y) -> X - if (N0C && !N0C->isNullValue()) - return N1; - // fold (select false, X, Y) -> Y - if (N0C && N0C->isNullValue()) - return N2; + if (const ConstantSDNode *N0C = dyn_cast<const ConstantSDNode>(N0)) { + // fold (select true, X, Y) -> X + // fold (select false, X, Y) -> Y + return !N0C->isNullValue() ? N1 : N2; + } // fold (select C, 1, X) -> (or C, X) - if (VT == MVT::i1 && N1C && N1C->getAPIntValue() == 1) + if (VT == MVT::i1 && isOneConstant(N1)) return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N2); // fold (select C, 0, 1) -> (xor C, 1) // We can't do this reliably if integer based booleans have different contents @@ -4684,40 +4884,43 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) { TLI.getBooleanContents(false, true) && TLI.getBooleanContents(false, false) == TargetLowering::ZeroOrOneBooleanContent)) && - N1C && N2C && N1C->isNullValue() && N2C->getAPIntValue() == 1) { + isNullConstant(N1) && isOneConstant(N2)) { SDValue XORNode; - if (VT == VT0) - return DAG.getNode(ISD::XOR, SDLoc(N), VT0, - N0, DAG.getConstant(1, VT0)); - XORNode = DAG.getNode(ISD::XOR, SDLoc(N0), VT0, - N0, DAG.getConstant(1, VT0)); + if (VT == VT0) { + SDLoc DL(N); + return DAG.getNode(ISD::XOR, DL, VT0, + N0, DAG.getConstant(1, DL, VT0)); + } + SDLoc DL0(N0); + XORNode = DAG.getNode(ISD::XOR, DL0, VT0, + N0, DAG.getConstant(1, DL0, VT0)); AddToWorklist(XORNode.getNode()); if (VT.bitsGT(VT0)) return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), VT, XORNode); return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, XORNode); } // fold (select C, 0, X) -> (and (not C), X) - if (VT == VT0 && VT == MVT::i1 && N1C && N1C->isNullValue()) { + if (VT == VT0 && VT == MVT::i1 && isNullConstant(N1)) { SDValue NOTNode = DAG.getNOT(SDLoc(N0), N0, VT); AddToWorklist(NOTNode.getNode()); return DAG.getNode(ISD::AND, SDLoc(N), VT, NOTNode, N2); } // fold (select C, X, 1) -> (or (not C), X) - if (VT == VT0 && VT == MVT::i1 && N2C && N2C->getAPIntValue() == 1) { + if (VT == VT0 && VT == MVT::i1 && isOneConstant(N2)) { SDValue NOTNode = DAG.getNOT(SDLoc(N0), N0, VT); AddToWorklist(NOTNode.getNode()); return DAG.getNode(ISD::OR, SDLoc(N), VT, NOTNode, N1); } // fold (select C, X, 0) -> (and C, X) - if (VT == MVT::i1 && N2C && N2C->isNullValue()) + if (VT == MVT::i1 && isNullConstant(N2)) return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, N1); // fold (select X, X, Y) -> (or X, Y) // fold (select X, 1, Y) -> (or X, Y) - if (VT == MVT::i1 && (N0 == N1 || (N1C && N1C->getAPIntValue() == 1))) + if (VT == MVT::i1 && (N0 == N1 || isOneConstant(N1))) return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N2); // fold (select X, Y, X) -> (and X, Y) // fold (select X, Y, 0) -> (and X, Y) - if (VT == MVT::i1 && (N0 == N2 || (N2C && N2C->getAPIntValue() == 0))) + if (VT == MVT::i1 && (N0 == N2 || isNullConstant(N2))) return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, N1); // If we can fold this based on the true/false value, do so. @@ -4757,6 +4960,69 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) { 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(); } @@ -4832,6 +5098,67 @@ static SDValue ConvertSelectToConcatVector(SDNode *N, SelectionDAG &DAG) { TopHalf->isNullValue() ? RHS->getOperand(1) : LHS->getOperand(1)); } +SDValue DAGCombiner::visitMSCATTER(SDNode *N) { + + if (Level >= AfterLegalizeTypes) + return SDValue(); + + MaskedScatterSDNode *MSC = cast<MaskedScatterSDNode>(N); + SDValue Mask = MSC->getMask(); + SDValue Data = MSC->getValue(); + SDLoc DL(N); + + // If the MSCATTER data type 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 + // and enables future optimizations (e.g. min/max pattern matching on X86). + if (Mask.getOpcode() != ISD::SETCC) + return SDValue(); + + // Check if any splitting is required. + if (TLI.getTypeAction(*DAG.getContext(), Data.getValueType()) != + TargetLowering::TypeSplitVector) + return SDValue(); + SDValue MaskLo, MaskHi, Lo, Hi; + std::tie(MaskLo, MaskHi) = SplitVSETCC(Mask.getNode(), DAG); + + EVT LoVT, HiVT; + std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(MSC->getValueType(0)); + + SDValue Chain = MSC->getChain(); + + EVT MemoryVT = MSC->getMemoryVT(); + unsigned Alignment = MSC->getOriginalAlignment(); + + EVT LoMemVT, HiMemVT; + std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT); + + SDValue DataLo, DataHi; + std::tie(DataLo, DataHi) = DAG.SplitVector(Data, DL); + + SDValue BasePtr = MSC->getBasePtr(); + SDValue IndexLo, IndexHi; + std::tie(IndexLo, IndexHi) = DAG.SplitVector(MSC->getIndex(), DL); + + MachineMemOperand *MMO = DAG.getMachineFunction(). + getMachineMemOperand(MSC->getPointerInfo(), + MachineMemOperand::MOStore, LoMemVT.getStoreSize(), + Alignment, MSC->getAAInfo(), MSC->getRanges()); + + SDValue OpsLo[] = { Chain, DataLo, MaskLo, BasePtr, IndexLo }; + Lo = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataLo.getValueType(), + DL, OpsLo, MMO); + + SDValue OpsHi[] = {Chain, DataHi, MaskHi, BasePtr, IndexHi}; + Hi = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataHi.getValueType(), + DL, OpsHi, MMO); + + AddToWorklist(Lo.getNode()); + AddToWorklist(Hi.getNode()); + + return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi); +} + SDValue DAGCombiner::visitMSTORE(SDNode *N) { if (Level >= AfterLegalizeTypes) @@ -4878,7 +5205,7 @@ SDValue DAGCombiner::visitMSTORE(SDNode *N) { 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()); @@ -4887,10 +5214,10 @@ SDValue DAGCombiner::visitMSTORE(SDNode *N) { unsigned IncrementSize = LoMemVT.getSizeInBits()/8; Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, DL, Ptr.getValueType())); MMO = DAG.getMachineFunction(). - getMachineMemOperand(MST->getPointerInfo(), + getMachineMemOperand(MST->getPointerInfo(), MachineMemOperand::MOStore, HiMemVT.getStoreSize(), SecondHalfAlignment, MST->getAAInfo(), MST->getRanges()); @@ -4906,6 +5233,83 @@ SDValue DAGCombiner::visitMSTORE(SDNode *N) { return SDValue(); } +SDValue DAGCombiner::visitMGATHER(SDNode *N) { + + if (Level >= AfterLegalizeTypes) + return SDValue(); + + MaskedGatherSDNode *MGT = dyn_cast<MaskedGatherSDNode>(N); + SDValue Mask = MGT->getMask(); + SDLoc DL(N); + + // If the MGATHER 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 + // and enables future optimizations (e.g. min/max pattern matching on X86). + + if (Mask.getOpcode() != ISD::SETCC) + return SDValue(); + + EVT VT = N->getValueType(0); + + // Check if any splitting is required. + if (TLI.getTypeAction(*DAG.getContext(), VT) != + TargetLowering::TypeSplitVector) + return SDValue(); + + SDValue MaskLo, MaskHi, Lo, Hi; + std::tie(MaskLo, MaskHi) = SplitVSETCC(Mask.getNode(), DAG); + + SDValue Src0 = MGT->getValue(); + SDValue Src0Lo, Src0Hi; + std::tie(Src0Lo, Src0Hi) = DAG.SplitVector(Src0, DL); + + EVT LoVT, HiVT; + std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(VT); + + SDValue Chain = MGT->getChain(); + EVT MemoryVT = MGT->getMemoryVT(); + unsigned Alignment = MGT->getOriginalAlignment(); + + EVT LoMemVT, HiMemVT; + std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT); + + SDValue BasePtr = MGT->getBasePtr(); + SDValue Index = MGT->getIndex(); + SDValue IndexLo, IndexHi; + std::tie(IndexLo, IndexHi) = DAG.SplitVector(Index, DL); + + MachineMemOperand *MMO = DAG.getMachineFunction(). + getMachineMemOperand(MGT->getPointerInfo(), + MachineMemOperand::MOLoad, LoMemVT.getStoreSize(), + Alignment, MGT->getAAInfo(), MGT->getRanges()); + + SDValue OpsLo[] = { Chain, Src0Lo, MaskLo, BasePtr, IndexLo }; + Lo = DAG.getMaskedGather(DAG.getVTList(LoVT, MVT::Other), LoVT, DL, OpsLo, + MMO); + + SDValue OpsHi[] = {Chain, Src0Hi, MaskHi, BasePtr, IndexHi}; + Hi = DAG.getMaskedGather(DAG.getVTList(HiVT, MVT::Other), HiVT, DL, OpsHi, + MMO); + + AddToWorklist(Lo.getNode()); + AddToWorklist(Hi.getNode()); + + // Build a factor node to remember that this load is independent of the + // other one. + Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo.getValue(1), + Hi.getValue(1)); + + // Legalized the chain result - switch anything that used the old chain to + // use the new one. + DAG.ReplaceAllUsesOfValueWith(SDValue(MGT, 1), Chain); + + SDValue GatherRes = DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, Lo, Hi); + + SDValue RetOps[] = { GatherRes, Chain }; + return DAG.getMergeValues(RetOps, DL); +} + SDValue DAGCombiner::visitMLOAD(SDNode *N) { if (Level >= AfterLegalizeTypes) @@ -4953,7 +5357,7 @@ SDValue DAGCombiner::visitMLOAD(SDNode *N) { 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()); @@ -4962,10 +5366,10 @@ SDValue DAGCombiner::visitMLOAD(SDNode *N) { unsigned IncrementSize = LoMemVT.getSizeInBits()/8; Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, DL, Ptr.getValueType())); MMO = DAG.getMachineFunction(). - getMachineMemOperand(MLD->getPointerInfo(), + getMachineMemOperand(MLD->getPointerInfo(), MachineMemOperand::MOLoad, HiMemVT.getStoreSize(), SecondHalfAlignment, MLD->getAAInfo(), MLD->getRanges()); @@ -5021,7 +5425,7 @@ SDValue DAGCombiner::visitVSELECT(SDNode *N) { EVT VT = LHS.getValueType(); SDValue Shift = DAG.getNode( ISD::SRA, DL, VT, LHS, - DAG.getConstant(VT.getScalarType().getSizeInBits() - 1, VT)); + DAG.getConstant(VT.getScalarType().getSizeInBits() - 1, DL, VT)); SDValue Add = DAG.getNode(ISD::ADD, DL, VT, LHS, Shift); AddToWorklist(Shift.getNode()); AddToWorklist(Add.getNode()); @@ -5029,6 +5433,9 @@ SDValue DAGCombiner::visitVSELECT(SDNode *N) { } } + 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 @@ -5141,7 +5548,8 @@ static SDNode *tryToFoldExtendOfConstant(SDNode *N, const TargetLowering &TLI, EVT VT = N->getValueType(0); assert((Opcode == ISD::SIGN_EXTEND || Opcode == ISD::ZERO_EXTEND || - Opcode == ISD::ANY_EXTEND) && "Expected EXTEND dag node in input!"); + Opcode == ISD::ANY_EXTEND || Opcode == ISD::SIGN_EXTEND_VECTOR_INREG) + && "Expected EXTEND dag node in input!"); // fold (sext c1) -> c1 // fold (zext c1) -> c1 @@ -5163,7 +5571,7 @@ static SDNode *tryToFoldExtendOfConstant(SDNode *N, const TargetLowering &TLI, unsigned EVTBits = N0->getValueType(0).getScalarType().getSizeInBits(); unsigned ShAmt = VTBits - EVTBits; SmallVector<SDValue, 8> Elts; - unsigned NumElts = N0->getNumOperands(); + unsigned NumElts = VT.getVectorNumElements(); SDLoc DL(N); for (unsigned i=0; i != NumElts; ++i) { @@ -5173,14 +5581,15 @@ static SDNode *tryToFoldExtendOfConstant(SDNode *N, const TargetLowering &TLI, continue; } + SDLoc DL(Op); ConstantSDNode *CurrentND = cast<ConstantSDNode>(Op); const APInt &C = APInt(VTBits, CurrentND->getAPIntValue().getZExtValue()); - if (Opcode == ISD::SIGN_EXTEND) + if (Opcode == ISD::SIGN_EXTEND || Opcode == ISD::SIGN_EXTEND_VECTOR_INREG) Elts.push_back(DAG.getConstant(C.shl(ShAmt).ashr(ShAmt).getZExtValue(), - SVT)); + DL, SVT)); else Elts.push_back(DAG.getConstant(C.shl(ShAmt).lshr(ShAmt).getZExtValue(), - SVT)); + DL, SVT)); } return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Elts).getNode(); @@ -5271,6 +5680,102 @@ void DAGCombiner::ExtendSetCCUses(const SmallVectorImpl<SDNode *> &SetCCs, } } +// FIXME: Bring more similar combines here, common to sext/zext (maybe aext?). +SDValue DAGCombiner::CombineExtLoad(SDNode *N) { + SDValue N0 = N->getOperand(0); + EVT DstVT = N->getValueType(0); + EVT SrcVT = N0.getValueType(); + + assert((N->getOpcode() == ISD::SIGN_EXTEND || + N->getOpcode() == ISD::ZERO_EXTEND) && + "Unexpected node type (not an extend)!"); + + // fold (sext (load x)) to multiple smaller sextloads; same for zext. + // For example, on a target with legal v4i32, but illegal v8i32, turn: + // (v8i32 (sext (v8i16 (load x)))) + // into: + // (v8i32 (concat_vectors (v4i32 (sextload x)), + // (v4i32 (sextload (x + 16))))) + // Where uses of the original load, i.e.: + // (v8i16 (load x)) + // are replaced with: + // (v8i16 (truncate + // (v8i32 (concat_vectors (v4i32 (sextload x)), + // (v4i32 (sextload (x + 16))))))) + // + // This combine is only applicable to illegal, but splittable, vectors. + // All legal types, and illegal non-vector types, are handled elsewhere. + // This combine is controlled by TargetLowering::isVectorLoadExtDesirable. + // + if (N0->getOpcode() != ISD::LOAD) + return SDValue(); + + LoadSDNode *LN0 = cast<LoadSDNode>(N0); + + if (!ISD::isNON_EXTLoad(LN0) || !ISD::isUNINDEXEDLoad(LN0) || + !N0.hasOneUse() || LN0->isVolatile() || !DstVT.isVector() || + !DstVT.isPow2VectorType() || !TLI.isVectorLoadExtDesirable(SDValue(N, 0))) + return SDValue(); + + SmallVector<SDNode *, 4> SetCCs; + if (!ExtendUsesToFormExtLoad(N, N0, N->getOpcode(), SetCCs, TLI)) + return SDValue(); + + ISD::LoadExtType ExtType = + N->getOpcode() == ISD::SIGN_EXTEND ? ISD::SEXTLOAD : ISD::ZEXTLOAD; + + // Try to split the vector types to get down to legal types. + EVT SplitSrcVT = SrcVT; + EVT SplitDstVT = DstVT; + while (!TLI.isLoadExtLegalOrCustom(ExtType, SplitDstVT, SplitSrcVT) && + SplitSrcVT.getVectorNumElements() > 1) { + SplitDstVT = DAG.GetSplitDestVTs(SplitDstVT).first; + SplitSrcVT = DAG.GetSplitDestVTs(SplitSrcVT).first; + } + + if (!TLI.isLoadExtLegalOrCustom(ExtType, SplitDstVT, SplitSrcVT)) + return SDValue(); + + SDLoc DL(N); + const unsigned NumSplits = + DstVT.getVectorNumElements() / SplitDstVT.getVectorNumElements(); + const unsigned Stride = SplitSrcVT.getStoreSize(); + SmallVector<SDValue, 4> Loads; + SmallVector<SDValue, 4> Chains; + + SDValue BasePtr = LN0->getBasePtr(); + for (unsigned Idx = 0; Idx < NumSplits; Idx++) { + const unsigned Offset = Idx * Stride; + const unsigned Align = MinAlign(LN0->getAlignment(), Offset); + + SDValue SplitLoad = DAG.getExtLoad( + ExtType, DL, SplitDstVT, LN0->getChain(), BasePtr, + LN0->getPointerInfo().getWithOffset(Offset), SplitSrcVT, + LN0->isVolatile(), LN0->isNonTemporal(), LN0->isInvariant(), + Align, LN0->getAAInfo()); + + BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr, + DAG.getConstant(Stride, DL, BasePtr.getValueType())); + + Loads.push_back(SplitLoad.getValue(0)); + Chains.push_back(SplitLoad.getValue(1)); + } + + SDValue NewChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains); + SDValue NewValue = DAG.getNode(ISD::CONCAT_VECTORS, DL, DstVT, Loads); + + CombineTo(N, NewValue); + + // Replace uses of the original load (before extension) + // with a truncate of the concatenated sextloaded vectors. + SDValue Trunc = + DAG.getNode(ISD::TRUNCATE, SDLoc(N0), N0.getValueType(), NewValue); + CombineTo(N0.getNode(), Trunc, NewChain); + ExtendSetCCUses(SetCCs, Trunc, NewValue, DL, + (ISD::NodeType)N->getOpcode()); + return SDValue(N, 0); // Return N so it doesn't get rechecked! +} + SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { SDValue N0 = N->getOperand(0); EVT VT = N->getValueType(0); @@ -5337,17 +5842,18 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { } // fold (sext (load x)) -> (sext (truncate (sextload x))) - // None of the supported targets knows how to perform load and sign extend - // on vectors in one instruction. We only perform this transformation on - // scalars. - if (ISD::isNON_EXTLoad(N0.getNode()) && !VT.isVector() && - ISD::isUNINDEXEDLoad(N0.getNode()) && - ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) || + // Only generate vector extloads when 1) they're legal, and 2) they are + // deemed desirable by the target. + if (ISD::isNON_EXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) && + ((!LegalOperations && !VT.isVector() && + !cast<LoadSDNode>(N0)->isVolatile()) || TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, N0.getValueType()))) { bool DoXform = true; SmallVector<SDNode*, 4> SetCCs; if (!N0.hasOneUse()) DoXform = ExtendUsesToFormExtLoad(N, N0, ISD::SIGN_EXTEND, SetCCs, TLI); + if (VT.isVector()) + DoXform &= TLI.isVectorLoadExtDesirable(SDValue(N, 0)); if (DoXform) { LoadSDNode *LN0 = cast<LoadSDNode>(N0); SDValue ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, SDLoc(N), VT, @@ -5364,6 +5870,11 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { } } + // fold (sext (load x)) to multiple smaller sextloads. + // Only on illegal but splittable vectors. + if (SDValue ExtLoad = CombineExtLoad(N)) + return ExtLoad; + // fold (sext (sextload x)) -> (sext (truncate (sextload x))) // fold (sext ( extload x)) -> (sext (truncate (sextload x))) if ((ISD::isSEXTLoad(N0.getNode()) || ISD::isEXTLoad(N0.getNode())) && @@ -5407,14 +5918,15 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { LN0->getMemOperand()); APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); Mask = Mask.sext(VT.getSizeInBits()); - SDValue And = DAG.getNode(N0.getOpcode(), SDLoc(N), VT, - ExtLoad, DAG.getConstant(Mask, VT)); + SDLoc DL(N); + SDValue And = DAG.getNode(N0.getOpcode(), DL, VT, + ExtLoad, DAG.getConstant(Mask, DL, VT)); SDValue Trunc = DAG.getNode(ISD::TRUNCATE, SDLoc(N0.getOperand(0)), N0.getOperand(0).getValueType(), ExtLoad); CombineTo(N, And); CombineTo(N0.getOperand(0).getNode(), Trunc, ExtLoad.getValue(1)); - ExtendSetCCUses(SetCCs, Trunc, ExtLoad, SDLoc(N), + ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL, ISD::SIGN_EXTEND); return SDValue(N, 0); // Return N so it doesn't get rechecked! } @@ -5457,11 +5969,12 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { // sext(setcc x, y, cc) -> (select (setcc x, y, cc), -1, 0) unsigned ElementWidth = VT.getScalarType().getSizeInBits(); + SDLoc DL(N); SDValue NegOne = - DAG.getConstant(APInt::getAllOnesValue(ElementWidth), VT); + DAG.getConstant(APInt::getAllOnesValue(ElementWidth), DL, VT); SDValue SCC = - SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1), - NegOne, DAG.getConstant(0, VT), + SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1), + NegOne, DAG.getConstant(0, DL, VT), cast<CondCodeSDNode>(N0.getOperand(2))->get(), true); if (SCC.getNode()) return SCC; @@ -5473,7 +5986,7 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { SDValue SetCC = DAG.getSetCC(DL, SetCCVT, N0.getOperand(0), N0.getOperand(1), CC); return DAG.getSelect(DL, VT, SetCC, - NegOne, DAG.getConstant(0, VT)); + NegOne, DAG.getConstant(0, DL, VT)); } } } @@ -5507,11 +6020,9 @@ static bool isTruncateOf(SelectionDAG &DAG, SDValue N, SDValue &Op, SDValue Op1 = N->getOperand(1); assert(Op0.getValueType() == Op1.getValueType()); - ConstantSDNode *COp0 = dyn_cast<ConstantSDNode>(Op0); - ConstantSDNode *COp1 = dyn_cast<ConstantSDNode>(Op1); - if (COp0 && COp0->isNullValue()) + if (isNullConstant(Op0)) Op = Op1; - else if (COp1 && COp1->isNullValue()) + else if (isNullConstant(Op1)) Op = Op0; else return false; @@ -5622,22 +6133,24 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { } APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); Mask = Mask.zext(VT.getSizeInBits()); - return DAG.getNode(ISD::AND, SDLoc(N), VT, - X, DAG.getConstant(Mask, VT)); + SDLoc DL(N); + return DAG.getNode(ISD::AND, DL, VT, + X, DAG.getConstant(Mask, DL, VT)); } // fold (zext (load x)) -> (zext (truncate (zextload x))) - // None of the supported targets knows how to perform load and vector_zext - // on vectors in one instruction. We only perform this transformation on - // scalars. - if (ISD::isNON_EXTLoad(N0.getNode()) && !VT.isVector() && - ISD::isUNINDEXEDLoad(N0.getNode()) && - ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) || + // Only generate vector extloads when 1) they're legal, and 2) they are + // deemed desirable by the target. + if (ISD::isNON_EXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) && + ((!LegalOperations && !VT.isVector() && + !cast<LoadSDNode>(N0)->isVolatile()) || TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, N0.getValueType()))) { bool DoXform = true; SmallVector<SDNode*, 4> SetCCs; if (!N0.hasOneUse()) DoXform = ExtendUsesToFormExtLoad(N, N0, ISD::ZERO_EXTEND, SetCCs, TLI); + if (VT.isVector()) + DoXform &= TLI.isVectorLoadExtDesirable(SDValue(N, 0)); if (DoXform) { LoadSDNode *LN0 = cast<LoadSDNode>(N0); SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N), VT, @@ -5655,6 +6168,11 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { } } + // fold (zext (load x)) to multiple smaller zextloads. + // Only on illegal but splittable vectors. + if (SDValue ExtLoad = CombineExtLoad(N)) + return ExtLoad; + // fold (zext (and/or/xor (load x), cst)) -> // (and/or/xor (zextload x), (zext cst)) if ((N0.getOpcode() == ISD::AND || N0.getOpcode() == ISD::OR || @@ -5677,14 +6195,15 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { LN0->getMemOperand()); APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); Mask = Mask.zext(VT.getSizeInBits()); - SDValue And = DAG.getNode(N0.getOpcode(), SDLoc(N), VT, - ExtLoad, DAG.getConstant(Mask, VT)); + SDLoc DL(N); + SDValue And = DAG.getNode(N0.getOpcode(), DL, VT, + ExtLoad, DAG.getConstant(Mask, DL, VT)); SDValue Trunc = DAG.getNode(ISD::TRUNCATE, SDLoc(N0.getOperand(0)), N0.getOperand(0).getValueType(), ExtLoad); CombineTo(N, And); CombineTo(N0.getOperand(0).getNode(), Trunc, ExtLoad.getValue(1)); - ExtendSetCCUses(SetCCs, Trunc, ExtLoad, SDLoc(N), + ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL, ISD::ZERO_EXTEND); return SDValue(N, 0); // Return N so it doesn't get rechecked! } @@ -5722,19 +6241,20 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { // zext(setcc) -> (and (vsetcc), (1, 1, ...) for vectors. // Only do this before legalize for now. EVT EltVT = VT.getVectorElementType(); + SDLoc DL(N); SmallVector<SDValue,8> OneOps(VT.getVectorNumElements(), - DAG.getConstant(1, EltVT)); + DAG.getConstant(1, DL, EltVT)); if (VT.getSizeInBits() == N0VT.getSizeInBits()) // We know that the # elements of the results is the same as the // # elements of the compare (and the # elements of the compare result // for that matter). Check to see that they are the same size. If so, // we know that the element size of the sext'd result matches the // element size of the compare operands. - return DAG.getNode(ISD::AND, SDLoc(N), VT, - DAG.getSetCC(SDLoc(N), VT, N0.getOperand(0), + return DAG.getNode(ISD::AND, DL, VT, + DAG.getSetCC(DL, VT, N0.getOperand(0), N0.getOperand(1), cast<CondCodeSDNode>(N0.getOperand(2))->get()), - DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, + DAG.getNode(ISD::BUILD_VECTOR, DL, VT, OneOps)); // If the desired elements are smaller or larger than the source @@ -5747,18 +6267,19 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { EVT::getVectorVT(*DAG.getContext(), MatchingElementType, N0VT.getVectorNumElements()); SDValue VsetCC = - DAG.getSetCC(SDLoc(N), MatchingVectorType, N0.getOperand(0), + DAG.getSetCC(DL, MatchingVectorType, N0.getOperand(0), N0.getOperand(1), cast<CondCodeSDNode>(N0.getOperand(2))->get()); - return DAG.getNode(ISD::AND, SDLoc(N), VT, - DAG.getSExtOrTrunc(VsetCC, SDLoc(N), VT), - DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, OneOps)); + return DAG.getNode(ISD::AND, DL, VT, + DAG.getSExtOrTrunc(VsetCC, DL, VT), + DAG.getNode(ISD::BUILD_VECTOR, DL, VT, OneOps)); } // zext(setcc x,y,cc) -> select_cc x, y, 1, 0, cc + SDLoc DL(N); SDValue SCC = - SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1), - DAG.getConstant(1, VT), DAG.getConstant(0, VT), + SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1), + DAG.getConstant(1, DL, VT), DAG.getConstant(0, DL, VT), cast<CondCodeSDNode>(N0.getOperand(2))->get(), true); if (SCC.getNode()) return SCC; } @@ -5850,8 +6371,9 @@ SDValue DAGCombiner::visitANY_EXTEND(SDNode *N) { } APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); Mask = Mask.zext(VT.getSizeInBits()); - return DAG.getNode(ISD::AND, SDLoc(N), VT, - X, DAG.getConstant(Mask, VT)); + SDLoc DL(N); + return DAG.getNode(ISD::AND, DL, VT, + X, DAG.getConstant(Mask, DL, VT)); } // fold (aext (load x)) -> (aext (truncate (extload x))) @@ -5934,9 +6456,10 @@ SDValue DAGCombiner::visitANY_EXTEND(SDNode *N) { } // aext(setcc x,y,cc) -> select_cc x, y, 1, 0, cc + SDLoc DL(N); SDValue SCC = - SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1), - DAG.getConstant(1, VT), DAG.getConstant(0, VT), + SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1), + DAG.getConstant(1, DL, VT), DAG.getConstant(0, DL, VT), cast<CondCodeSDNode>(N0.getOperand(2))->get(), true); if (SCC.getNode()) return SCC; @@ -5957,7 +6480,7 @@ SDValue DAGCombiner::GetDemandedBits(SDValue V, const APInt &Mask) { const APInt &CVal = CV->getAPIntValue(); APInt NewVal = CVal & Mask; if (NewVal != CVal) - return DAG.getConstant(NewVal, V.getValueType()); + return DAG.getConstant(NewVal, SDLoc(V), V.getValueType()); break; } case ISD::OR: @@ -5972,7 +6495,7 @@ SDValue DAGCombiner::GetDemandedBits(SDValue V, const APInt &Mask) { // Only look at single-use SRLs. if (!V.getNode()->hasOneUse()) break; - if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(V.getOperand(1))) { + if (ConstantSDNode *RHSC = getAsNonOpaqueConstant(V.getOperand(1))) { // See if we can recursively simplify the LHS. unsigned Amt = RHSC->getZExtValue(); @@ -6117,9 +6640,10 @@ SDValue DAGCombiner::ReduceLoadWidth(SDNode *N) { uint64_t PtrOff = ShAmt / 8; unsigned NewAlign = MinAlign(LN0->getAlignment(), PtrOff); - SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(LN0), + SDLoc DL(LN0); + SDValue NewPtr = DAG.getNode(ISD::ADD, DL, PtrType, LN0->getBasePtr(), - DAG.getConstant(PtrOff, PtrType)); + DAG.getConstant(PtrOff, DL, PtrType)); AddToWorklist(NewPtr.getNode()); SDValue Load; @@ -6148,11 +6672,12 @@ SDValue DAGCombiner::ReduceLoadWidth(SDNode *N) { // no larger than the source) then the useful bits of the result are // zero; we can't simply return the shortened shift, because the result // of that operation is undefined. + SDLoc DL(N0); if (ShLeftAmt >= VT.getSizeInBits()) - Result = DAG.getConstant(0, VT); + Result = DAG.getConstant(0, DL, VT); else - Result = DAG.getNode(ISD::SHL, SDLoc(N0), VT, - Result, DAG.getConstant(ShLeftAmt, ShImmTy)); + Result = DAG.getNode(ISD::SHL, DL, VT, + Result, DAG.getConstant(ShLeftAmt, DL, ShImmTy)); } // Return the new loaded value. @@ -6279,7 +6804,7 @@ SDValue DAGCombiner::visitSIGN_EXTEND_INREG(SDNode *N) { ConstantSDNode *CurrentND = cast<ConstantSDNode>(Op); const APInt &C = APInt(VTBits, CurrentND->getAPIntValue().getZExtValue()); Elts.push_back(DAG.getConstant(C.shl(ShAmt).ashr(ShAmt).getZExtValue(), - Op.getValueType())); + SDLoc(Op), Op.getValueType())); } return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, Elts); @@ -6288,6 +6813,20 @@ SDValue DAGCombiner::visitSIGN_EXTEND_INREG(SDNode *N) { return SDValue(); } +SDValue DAGCombiner::visitSIGN_EXTEND_VECTOR_INREG(SDNode *N) { + SDValue N0 = N->getOperand(0); + EVT VT = N->getValueType(0); + + if (N0.getOpcode() == ISD::UNDEF) + return DAG.getUNDEF(VT); + + if (SDNode *Res = tryToFoldExtendOfConstant(N, TLI, DAG, LegalTypes, + LegalOperations)) + return SDValue(Res, 0); + + return SDValue(); +} + SDValue DAGCombiner::visitTRUNCATE(SDNode *N) { SDValue N0 = N->getOperand(0); EVT VT = N->getValueType(0); @@ -6297,7 +6836,7 @@ SDValue DAGCombiner::visitTRUNCATE(SDNode *N) { 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) @@ -6350,9 +6889,10 @@ SDValue DAGCombiner::visitTRUNCATE(SDNode *N) { SDValue V = DAG.getNode(ISD::BITCAST, SDLoc(N), NVT, N0.getOperand(0)); + SDLoc DL(N); return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, - SDLoc(N), TrTy, V, - DAG.getConstant(Index, IndexTy)); + DL, TrTy, V, + DAG.getConstant(Index, DL, IndexTy)); } } @@ -6598,13 +7138,14 @@ SDValue DAGCombiner::visitBITCAST(SDNode *N) { N0.getOperand(0)); AddToWorklist(NewConv.getNode()); + SDLoc DL(N); APInt SignBit = APInt::getSignBit(VT.getSizeInBits()); if (N0.getOpcode() == ISD::FNEG) - return DAG.getNode(ISD::XOR, SDLoc(N), VT, - NewConv, DAG.getConstant(SignBit, VT)); + return DAG.getNode(ISD::XOR, DL, VT, + NewConv, DAG.getConstant(SignBit, DL, VT)); assert(N0.getOpcode() == ISD::FABS); - return DAG.getNode(ISD::AND, SDLoc(N), VT, - NewConv, DAG.getConstant(~SignBit, VT)); + return DAG.getNode(ISD::AND, DL, VT, + NewConv, DAG.getConstant(~SignBit, DL, VT)); } // fold (bitconvert (fcopysign cst, x)) -> @@ -6629,9 +7170,11 @@ SDValue DAGCombiner::visitBITCAST(SDNode *N) { } else if (OrigXWidth > VTWidth) { // To get the sign bit in the right place, we have to shift it right // before truncating. - X = DAG.getNode(ISD::SRL, SDLoc(X), + SDLoc DL(X); + X = DAG.getNode(ISD::SRL, DL, X.getValueType(), X, - DAG.getConstant(OrigXWidth-VTWidth, X.getValueType())); + DAG.getConstant(OrigXWidth-VTWidth, DL, + X.getValueType())); AddToWorklist(X.getNode()); X = DAG.getNode(ISD::TRUNCATE, SDLoc(X), VT, X); AddToWorklist(X.getNode()); @@ -6639,13 +7182,13 @@ SDValue DAGCombiner::visitBITCAST(SDNode *N) { APInt SignBit = APInt::getSignBit(VT.getSizeInBits()); X = DAG.getNode(ISD::AND, SDLoc(X), VT, - X, DAG.getConstant(SignBit, VT)); + X, DAG.getConstant(SignBit, SDLoc(X), VT)); AddToWorklist(X.getNode()); SDValue Cst = DAG.getNode(ISD::BITCAST, SDLoc(N0), VT, N0.getOperand(0)); Cst = DAG.getNode(ISD::AND, SDLoc(Cst), VT, - Cst, DAG.getConstant(~SignBit, VT)); + Cst, DAG.getConstant(~SignBit, SDLoc(Cst), VT)); AddToWorklist(Cst.getNode()); return DAG.getNode(ISD::OR, SDLoc(N), VT, X, Cst); @@ -6659,6 +7202,51 @@ SDValue DAGCombiner::visitBITCAST(SDNode *N) { 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(); } @@ -6727,6 +7315,8 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode *BV, EVT DstEltVT) { return ConstantFoldBITCASTofBUILD_VECTOR(Tmp, DstEltVT); } + SDLoc DL(BV); + // Okay, we know the src/dst types are both integers of differing types. // Handling growing first. assert(SrcEltVT.isInteger() && DstEltVT.isInteger()); @@ -6753,16 +7343,15 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode *BV, EVT DstEltVT) { if (EltIsUndef) Ops.push_back(DAG.getUNDEF(DstEltVT)); else - Ops.push_back(DAG.getConstant(NewBits, DstEltVT)); + Ops.push_back(DAG.getConstant(NewBits, DL, DstEltVT)); } EVT VT = EVT::getVectorVT(*DAG.getContext(), DstEltVT, Ops.size()); - return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(BV), VT, Ops); + return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Ops); } // 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()); @@ -6770,8 +7359,7 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode *BV, EVT DstEltVT) { for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) { if (BV->getOperand(i).getOpcode() == ISD::UNDEF) { - for (unsigned j = 0; j != NumOutputsPerInput; ++j) - Ops.push_back(DAG.getUNDEF(DstEltVT)); + Ops.append(NumOutputsPerInput, DAG.getUNDEF(DstEltVT)); continue; } @@ -6780,11 +7368,7 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode *BV, EVT DstEltVT) { 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]); + Ops.push_back(DAG.getConstant(ThisVal, DL, DstEltVT)); OpVal = OpVal.lshr(DstBitSize); } @@ -6793,7 +7377,450 @@ ConstantFoldBITCASTofBUILD_VECTOR(SDNode *BV, EVT DstEltVT) { std::reverse(Ops.end()-NumOutputsPerInput, Ops.end()); } - return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(BV), VT, Ops); + return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Ops); +} + +/// 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(PreferredFusedOpcode, SL, VT, + N0.getOperand(0), N0.getOperand(1), N1); + } + + // fold (fadd x, (fmul y, z)) -> (fma y, z, x) + // Note: Commutes FADD operands. + if (N1.getOpcode() == ISD::FMUL && + (Aggressive || N1->hasOneUse())) { + 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 ((UnsafeFPMath || HasFMAD) && Aggressive) { + // fold (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y (fma u, v, z)) + if (N0.getOpcode() == PreferredFusedOpcode && + N0.getOperand(2).getOpcode() == ISD::FMUL) { + return DAG.getNode(PreferredFusedOpcode, SL, VT, + N0.getOperand(0), N0.getOperand(1), + 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() == PreferredFusedOpcode && + N1.getOperand(2).getOpcode() == ISD::FMUL) { + return DAG.getNode(PreferredFusedOpcode, SL, VT, + N1.getOperand(0), N1.getOperand(1), + 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(); +} + +/// 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(PreferredFusedOpcode, SL, VT, + N0.getOperand(0), N0.getOperand(1), + DAG.getNode(ISD::FNEG, SL, VT, N1)); + } + + // fold (fsub x, (fmul y, z)) -> (fma (fneg y), z, x) + // Note: Commutes FSUB operands. + if (N1.getOpcode() == ISD::FMUL && + (Aggressive || N1->hasOneUse())) + return DAG.getNode(PreferredFusedOpcode, SL, VT, + DAG.getNode(ISD::FNEG, SL, VT, + N1.getOperand(0)), + N1.getOperand(1), N0); + + // fold (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)) + if (N0.getOpcode() == ISD::FNEG && + N0.getOperand(0).getOpcode() == ISD::FMUL && + (Aggressive || (N0->hasOneUse() && N0.getOperand(0).hasOneUse()))) { + SDValue N00 = N0.getOperand(0).getOperand(0); + SDValue N01 = N0.getOperand(0).getOperand(1); + 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 ((UnsafeFPMath || HasFMAD) && Aggressive) { + // fold (fsub (fma x, y, (fmul u, v)), z) + // -> (fma x, y (fma u, v, (fneg z))) + if (N0.getOpcode() == PreferredFusedOpcode && + N0.getOperand(2).getOpcode() == ISD::FMUL) { + return DAG.getNode(PreferredFusedOpcode, SL, VT, + N0.getOperand(0), N0.getOperand(1), + DAG.getNode(PreferredFusedOpcode, SL, VT, + N0.getOperand(2).getOperand(0), + N0.getOperand(2).getOperand(1), + 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() == PreferredFusedOpcode && + N1.getOperand(2).getOpcode() == ISD::FMUL) { + SDValue N20 = N1.getOperand(2).getOperand(0); + SDValue N21 = N1.getOperand(2).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, 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(); } SDValue DAGCombiner::visitFADD(SDNode *N) { @@ -6802,32 +7829,32 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0); ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1); EVT VT = N->getValueType(0); + SDLoc DL(N); 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) - return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0, N1); + return DAG.getNode(ISD::FADD, DL, VT, N0, N1); // canonicalize constant to RHS if (N0CFP && !N1CFP) - return DAG.getNode(ISD::FADD, SDLoc(N), VT, N1, N0); + return DAG.getNode(ISD::FADD, DL, VT, N1, N0); // fold (fadd A, (fneg B)) -> (fsub A, B) if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) && isNegatibleForFree(N1, LegalOperations, TLI, &Options) == 2) - return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N0, + return DAG.getNode(ISD::FSUB, DL, VT, N0, GetNegatedExpression(N1, DAG, LegalOperations)); // fold (fadd (fneg A), B) -> (fsub B, A) if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) && isNegatibleForFree(N0, LegalOperations, TLI, &Options) == 2) - return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N1, + return DAG.getNode(ISD::FSUB, DL, VT, N1, GetNegatedExpression(N0, DAG, LegalOperations)); // If 'unsafe math' is enabled, fold lots of things. @@ -6837,23 +7864,22 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { bool AllowNewConst = (Level < AfterLegalizeDAG); // fold (fadd A, 0) -> A - if (N1CFP && N1CFP->getValueAPF().isZero()) + if (N1CFP && N1CFP->isZero()) return N0; // fold (fadd (fadd x, c1), c2) -> (fadd x, (fadd c1, c2)) if (N1CFP && N0.getOpcode() == ISD::FADD && N0.getNode()->hasOneUse() && isa<ConstantFPSDNode>(N0.getOperand(1))) - return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0.getOperand(0), - DAG.getNode(ISD::FADD, SDLoc(N), VT, - N0.getOperand(1), N1)); + return DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(0), + DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(1), N1)); // If allowed, fold (fadd (fneg x), x) -> 0.0 if (AllowNewConst && N0.getOpcode() == ISD::FNEG && N0.getOperand(0) == N1) - return DAG.getConstantFP(0.0, VT); + return DAG.getConstantFP(0.0, DL, VT); // If allowed, fold (fadd x, (fneg x)) -> 0.0 if (AllowNewConst && N1.getOpcode() == ISD::FNEG && N1.getOperand(0) == N0) - return DAG.getConstantFP(0.0, VT); + return DAG.getConstantFP(0.0, DL, VT); // We can fold chains of FADD's of the same value into multiplications. // This transform is not safe in general because we are reducing the number @@ -6865,21 +7891,18 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { // (fadd (fmul x, c), x) -> (fmul x, c+1) if (CFP01 && !CFP00 && N0.getOperand(0) == N1) { - SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT, - SDValue(CFP01, 0), - DAG.getConstantFP(1.0, VT)); - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1, NewCFP); + SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, SDValue(CFP01, 0), + DAG.getConstantFP(1.0, DL, VT)); + return DAG.getNode(ISD::FMUL, DL, VT, N1, NewCFP); } // (fadd (fmul x, c), (fadd x, x)) -> (fmul x, c+2) if (CFP01 && !CFP00 && N1.getOpcode() == ISD::FADD && N1.getOperand(0) == N1.getOperand(1) && N0.getOperand(0) == N1.getOperand(0)) { - SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT, - SDValue(CFP01, 0), - DAG.getConstantFP(2.0, VT)); - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, - N0.getOperand(0), NewCFP); + SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, SDValue(CFP01, 0), + DAG.getConstantFP(2.0, DL, VT)); + return DAG.getNode(ISD::FMUL, DL, VT, N0.getOperand(0), NewCFP); } } @@ -6889,20 +7912,18 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { // (fadd x, (fmul x, c)) -> (fmul x, c+1) if (CFP11 && !CFP10 && N1.getOperand(0) == N0) { - SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT, - SDValue(CFP11, 0), - DAG.getConstantFP(1.0, VT)); - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, NewCFP); + SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, SDValue(CFP11, 0), + DAG.getConstantFP(1.0, DL, VT)); + return DAG.getNode(ISD::FMUL, DL, VT, N0, NewCFP); } // (fadd (fadd x, x), (fmul x, c)) -> (fmul x, c+2) if (CFP11 && !CFP10 && N0.getOpcode() == ISD::FADD && N0.getOperand(0) == N0.getOperand(1) && N1.getOperand(0) == N0.getOperand(0)) { - SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT, - SDValue(CFP11, 0), - DAG.getConstantFP(2.0, VT)); - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1.getOperand(0), NewCFP); + SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, SDValue(CFP11, 0), + DAG.getConstantFP(2.0, DL, VT)); + return DAG.getNode(ISD::FMUL, DL, VT, N1.getOperand(0), NewCFP); } } @@ -6910,18 +7931,20 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N0.getOperand(0)); // (fadd (fadd x, x), x) -> (fmul x, 3.0) if (!CFP && N0.getOperand(0) == N0.getOperand(1) && - (N0.getOperand(0) == N1)) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, - N1, DAG.getConstantFP(3.0, VT)); + (N0.getOperand(0) == N1)) { + return DAG.getNode(ISD::FMUL, DL, VT, + N1, DAG.getConstantFP(3.0, DL, VT)); + } } if (N1.getOpcode() == ISD::FADD && AllowNewConst) { ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0)); // (fadd x, (fadd x, x)) -> (fmul x, 3.0) if (!CFP10 && N1.getOperand(0) == N1.getOperand(1) && - N1.getOperand(0) == N0) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, - N0, DAG.getConstantFP(3.0, VT)); + N1.getOperand(0) == N0) { + return DAG.getNode(ISD::FMUL, DL, VT, + N0, DAG.getConstantFP(3.0, DL, VT)); + } } // (fadd (fadd x, x), (fadd x, x)) -> (fmul x, 4.0) @@ -6929,81 +7952,18 @@ SDValue DAGCombiner::visitFADD(SDNode *N) { N0.getOpcode() == ISD::FADD && N1.getOpcode() == ISD::FADD && N0.getOperand(0) == N0.getOperand(1) && N1.getOperand(0) == N1.getOperand(1) && - N0.getOperand(0) == N1.getOperand(0)) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, - N0.getOperand(0), DAG.getConstantFP(4.0, VT)); + N0.getOperand(0) == N1.getOperand(0)) { + return DAG.getNode(ISD::FMUL, DL, VT, + N0.getOperand(0), DAG.getConstantFP(4.0, DL, VT)); + } } } // enable-unsafe-fp-math // FADD -> FMA combines: - if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) && - TLI.isFMAFasterThanFMulAndFAdd(VT) && - (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) { - - // fold (fadd (fmul x, y), z) -> (fma x, y, z) - if (N0.getOpcode() == ISD::FMUL && - (N0->hasOneUse() || TLI.enableAggressiveFMAFusion(VT))) - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - N0.getOperand(0), N0.getOperand(1), N1); - - // fold (fadd x, (fmul y, z)) -> (fma y, z, x) - // Note: Commutes FADD operands. - if (N1.getOpcode() == ISD::FMUL && - (N1->hasOneUse() || TLI.enableAggressiveFMAFusion(VT))) - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - N1.getOperand(0), N1.getOperand(1), N0); - - // 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); - } - } - - // More folding opportunities when target permits. - if (TLI.enableAggressiveFMAFusion(VT)) { - - // fold (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y (fma u, v, z)) - if (N0.getOpcode() == ISD::FMA && - N0.getOperand(2).getOpcode() == ISD::FMUL) - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - N0.getOperand(0), N0.getOperand(1), - DAG.getNode(ISD::FMA, SDLoc(N), 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 && - N1.getOperand(2).getOpcode() == ISD::FMUL) - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - N1.getOperand(0), N1.getOperand(1), - DAG.getNode(ISD::FMA, SDLoc(N), VT, - N1.getOperand(2).getOperand(0), - N1.getOperand(2).getOperand(1), - N0)); - } + SDValue Fused = visitFADDForFMACombine(N); + if (Fused) { + AddToWorklist(Fused.getNode()); + return Fused; } return SDValue(); @@ -7019,14 +7979,13 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) { 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) - return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N0, N1); + return DAG.getNode(ISD::FSUB, dl, VT, N0, N1); // fold (fsub A, (fneg B)) -> (fadd A, B) if (isNegatibleForFree(N1, LegalOperations, TLI, &Options)) @@ -7036,11 +7995,11 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) { // If 'unsafe math' is enabled, fold lots of things. if (Options.UnsafeFPMath) { // (fsub A, 0) -> A - if (N1CFP && N1CFP->getValueAPF().isZero()) + if (N1CFP && N1CFP->isZero()) return N0; // (fsub 0, B) -> -B - if (N0CFP && N0CFP->getValueAPF().isZero()) { + if (N0CFP && N0CFP->isZero()) { if (isNegatibleForFree(N1, LegalOperations, TLI, &Options)) return GetNegatedExpression(N1, DAG, LegalOperations); if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT)) @@ -7049,7 +8008,7 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) { // (fsub x, x) -> 0.0 if (N0 == N1) - return DAG.getConstantFP(0.0f, VT); + return DAG.getConstantFP(0.0f, dl, VT); // (fsub x, (fadd x, y)) -> (fneg y) // (fsub x, (fadd y, x)) -> (fneg y) @@ -7066,138 +8025,10 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) { } // FSUB -> FMA combines: - if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) && - TLI.isFMAFasterThanFMulAndFAdd(VT) && - (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) { - - // fold (fsub (fmul x, y), z) -> (fma x, y, (fneg z)) - if (N0.getOpcode() == ISD::FMUL && - (N0->hasOneUse() || TLI.enableAggressiveFMAFusion(VT))) - return DAG.getNode(ISD::FMA, dl, VT, - N0.getOperand(0), N0.getOperand(1), - DAG.getNode(ISD::FNEG, dl, VT, N1)); - - // fold (fsub x, (fmul y, z)) -> (fma (fneg y), z, x) - // Note: Commutes FSUB operands. - if (N1.getOpcode() == ISD::FMUL && - (N1->hasOneUse() || TLI.enableAggressiveFMAFusion(VT))) - return DAG.getNode(ISD::FMA, dl, VT, - DAG.getNode(ISD::FNEG, dl, VT, - N1.getOperand(0)), - N1.getOperand(1), N0); - - // fold (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)) - if (N0.getOpcode() == ISD::FNEG && - N0.getOperand(0).getOpcode() == ISD::FMUL && - ((N0->hasOneUse() && N0.getOperand(0).hasOneUse()) || - TLI.enableAggressiveFMAFusion(VT))) { - SDValue N00 = N0.getOperand(0).getOperand(0); - SDValue N01 = N0.getOperand(0).getOperand(1); - return DAG.getNode(ISD::FMA, dl, VT, - DAG.getNode(ISD::FNEG, dl, VT, N00), N01, - DAG.getNode(ISD::FNEG, dl, VT, N1)); - } - - // 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)); - } - } - } - } - - // More folding opportunities when target permits. - if (TLI.enableAggressiveFMAFusion(VT)) { - - // fold (fsub (fma x, y, (fmul u, v)), z) - // -> (fma x, y (fma u, v, (fneg z))) - if (N0.getOpcode() == ISD::FMA && - N0.getOperand(2).getOpcode() == ISD::FMUL) - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - N0.getOperand(0), N0.getOperand(1), - DAG.getNode(ISD::FMA, SDLoc(N), VT, - N0.getOperand(2).getOperand(0), - N0.getOperand(2).getOperand(1), - DAG.getNode(ISD::FNEG, SDLoc(N), VT, - N1))); - - // fold (fsub x, (fma y, z, (fmul u, v))) - // -> (fma (fneg y), z, (fma (fneg u), v, x)) - if (N1.getOpcode() == ISD::FMA && - N1.getOperand(2).getOpcode() == ISD::FMUL) { - SDValue N20 = N1.getOperand(2).getOperand(0); - SDValue N21 = N1.getOperand(2).getOperand(1); - return DAG.getNode(ISD::FMA, SDLoc(N), VT, - DAG.getNode(ISD::FNEG, SDLoc(N), VT, - N1.getOperand(0)), - N1.getOperand(1), - DAG.getNode(ISD::FMA, SDLoc(N), VT, - DAG.getNode(ISD::FNEG, SDLoc(N), VT, - N20), - N21, N0)); - } - } + SDValue Fused = visitFSUBForFMACombine(N); + if (Fused) { + AddToWorklist(Fused.getNode()); + return Fused; } return SDValue(); @@ -7209,29 +8040,24 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) { ConstantFPSDNode *N0CFP = isConstOrConstSplatFP(N0); ConstantFPSDNode *N1CFP = isConstOrConstSplatFP(N1); EVT VT = N->getValueType(0); + SDLoc DL(N); const TargetOptions &Options = DAG.getTarget().Options; // 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 if (N0CFP && N1CFP) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, N1); + return DAG.getNode(ISD::FMUL, DL, VT, N0, N1); // canonicalize constant to RHS - if (N0CFP && !N1CFP) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1, N0); + if (isConstantFPBuildVectorOrConstantFP(N0) && + !isConstantFPBuildVectorOrConstantFP(N1)) + return DAG.getNode(ISD::FMUL, DL, VT, N1, N0); // fold (fmul A, 1.0) -> A if (N1CFP && N1CFP->isExactlyValue(1.0)) @@ -7239,7 +8065,7 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) { if (Options.UnsafeFPMath) { // fold (fmul A, 0) -> 0 - if (N1CFP && N1CFP->getValueAPF().isZero()) + if (N1CFP && N1CFP->isZero()) return N1; // fold (fmul (fmul x, c1), c2) -> (fmul x, (fmul c1, c2)) @@ -7247,14 +8073,22 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) { // 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())) { + SDValue MulConsts = DAG.getNode(ISD::FMUL, DL, VT, N01, N1); + return DAG.getNode(ISD::FMUL, DL, VT, N00, MulConsts); + } } } @@ -7263,21 +8097,20 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) { // during an early run of DAGCombiner can prevent folding with fmuls // inserted during lowering. if (N0.getOpcode() == ISD::FADD && N0.getOperand(0) == N0.getOperand(1)) { - SDLoc SL(N); - const SDValue Two = DAG.getConstantFP(2.0, VT); - SDValue MulConsts = DAG.getNode(ISD::FMUL, SL, VT, Two, N1); - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0.getOperand(0), MulConsts); + const SDValue Two = DAG.getConstantFP(2.0, DL, VT); + SDValue MulConsts = DAG.getNode(ISD::FMUL, DL, VT, Two, N1); + return DAG.getNode(ISD::FMUL, DL, VT, N0.getOperand(0), MulConsts); } } // fold (fmul X, 2.0) -> (fadd X, X) if (N1CFP && N1CFP->isExactlyValue(+2.0)) - return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0, N0); + return DAG.getNode(ISD::FADD, DL, VT, N0, N0); // fold (fmul X, -1.0) -> (fneg X) if (N1CFP && N1CFP->isExactlyValue(-1.0)) if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT)) - return DAG.getNode(ISD::FNEG, SDLoc(N), VT, N0); + return DAG.getNode(ISD::FNEG, DL, VT, N0); // fold (fmul (fneg X), (fneg Y)) -> (fmul X, Y) if (char LHSNeg = isNegatibleForFree(N0, LegalOperations, TLI, &Options)) { @@ -7285,7 +8118,7 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) { // Both can be negated for free, check to see if at least one is cheaper // negated. if (LHSNeg == 2 || RHSNeg == 2) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, + return DAG.getNode(ISD::FMUL, DL, VT, GetNegatedExpression(N0, DAG, LegalOperations), GetNegatedExpression(N1, DAG, LegalOperations)); } @@ -7364,14 +8197,14 @@ SDValue DAGCombiner::visitFMA(SDNode *N) { if (Options.UnsafeFPMath && N1CFP && N0 == N2) return DAG.getNode(ISD::FMUL, dl, VT, N0, DAG.getNode(ISD::FADD, dl, VT, - N1, DAG.getConstantFP(1.0, VT))); + N1, DAG.getConstantFP(1.0, dl, VT))); // (fma x, c, (fneg x)) -> (fmul x, (c-1)) if (Options.UnsafeFPMath && N1CFP && N2.getOpcode() == ISD::FNEG && N2.getOperand(0) == N0) return DAG.getNode(ISD::FMUL, dl, VT, N0, DAG.getNode(ISD::FADD, dl, VT, - N1, DAG.getConstantFP(-1.0, VT))); + N1, DAG.getConstantFP(-1.0, dl, VT))); return SDValue(); @@ -7387,10 +8220,9 @@ SDValue DAGCombiner::visitFDIV(SDNode *N) { 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) @@ -7412,8 +8244,8 @@ SDValue DAGCombiner::visitFDIV(SDNode *N) { // TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT) || TLI.isOperationLegal(llvm::ISD::ConstantFP, VT) || TLI.isFPImmLegal(Recip, VT))) - return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, - DAG.getConstantFP(Recip, VT)); + return DAG.getNode(ISD::FMUL, DL, VT, N0, + DAG.getConstantFP(Recip, DL, VT)); } // If this FDIV is part of a reciprocal square root, it may be folded @@ -7492,24 +8324,22 @@ SDValue DAGCombiner::visitFDIV(SDNode *N) { SmallVector<SDNode *, 4> Users; // Find all FDIV users of the same divisor. - for (SDNode::use_iterator UI = N1.getNode()->use_begin(), - UE = N1.getNode()->use_end(); - UI != UE; ++UI) { - SDNode *User = UI.getUse().getUser(); - if (User->getOpcode() == ISD::FDIV && User->getOperand(1) == N1) - Users.push_back(User); + for (auto *U : N1->uses()) { + if (U->getOpcode() == ISD::FDIV && U->getOperand(1) == N1) + Users.push_back(U); } if (TLI.combineRepeatedFPDivisors(Users.size())) { - SDValue FPOne = DAG.getConstantFP(1.0, VT); // floating point 1.0 - SDValue Reciprocal = DAG.getNode(ISD::FDIV, SDLoc(N), VT, FPOne, N1); + SDValue FPOne = DAG.getConstantFP(1.0, DL, VT); + SDValue Reciprocal = DAG.getNode(ISD::FDIV, DL, VT, FPOne, N1); // Dividend / Divisor -> Dividend * Reciprocal - for (auto I = Users.begin(), E = Users.end(); I != E; ++I) { - if ((*I)->getOperand(0) != FPOne) { - SDValue NewNode = DAG.getNode(ISD::FMUL, SDLoc(*I), VT, - (*I)->getOperand(0), Reciprocal); - DAG.ReplaceAllUsesWith(*I, NewNode.getNode()); + for (auto *U : Users) { + SDValue Dividend = U->getOperand(0); + if (Dividend != FPOne) { + SDValue NewNode = DAG.getNode(ISD::FMUL, SDLoc(U), VT, Dividend, + Reciprocal); + DAG.ReplaceAllUsesWith(U, NewNode.getNode()); } } return SDValue(); @@ -7539,20 +8369,21 @@ SDValue DAGCombiner::visitFSQRT(SDNode *N) { // Compute this as X * (1/sqrt(X)) = X * (X ** -0.5) if (SDValue RV = BuildRsqrtEstimate(N->getOperand(0))) { EVT VT = RV.getValueType(); - RV = DAG.getNode(ISD::FMUL, SDLoc(N), VT, N->getOperand(0), RV); + SDLoc DL(N); + RV = DAG.getNode(ISD::FMUL, DL, VT, N->getOperand(0), RV); AddToWorklist(RV.getNode()); // Unfortunately, RV is now NaN if the input was exactly 0. // Select out this case and force the answer to 0. - SDValue Zero = DAG.getConstantFP(0.0, VT); + SDValue Zero = DAG.getConstantFP(0.0, DL, VT); SDValue ZeroCmp = - DAG.getSetCC(SDLoc(N), TLI.getSetCCResultType(*DAG.getContext(), VT), + DAG.getSetCC(DL, TLI.getSetCCResultType(*DAG.getContext(), VT), N->getOperand(0), Zero, ISD::SETEQ); AddToWorklist(ZeroCmp.getNode()); AddToWorklist(RV.getNode()); RV = DAG.getNode(VT.isVector() ? ISD::VSELECT : ISD::SELECT, - SDLoc(N), VT, ZeroCmp, Zero, RV); + DL, VT, ZeroCmp, Zero, RV); return RV; } } @@ -7611,12 +8442,11 @@ SDValue DAGCombiner::visitFCOPYSIGN(SDNode *N) { SDValue DAGCombiner::visitSINT_TO_FP(SDNode *N) { SDValue N0 = N->getOperand(0); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); EVT VT = N->getValueType(0); EVT OpVT = N0.getValueType(); // fold (sint_to_fp c1) -> c1fp - if (N0C && + if (isConstantIntBuildVectorOrConstantInt(N0) && // ...but only if the target supports immediate floating-point values (!LegalOperations || TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) @@ -7638,11 +8468,12 @@ SDValue DAGCombiner::visitSINT_TO_FP(SDNode *N) { !VT.isVector() && (!LegalOperations || TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) { + SDLoc DL(N); SDValue Ops[] = { N0.getOperand(0), N0.getOperand(1), - DAG.getConstantFP(-1.0, VT) , DAG.getConstantFP(0.0, VT), + DAG.getConstantFP(-1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT), N0.getOperand(2) }; - return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops); + return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops); } // fold (sint_to_fp (zext (setcc x, y, cc))) -> @@ -7651,11 +8482,12 @@ SDValue DAGCombiner::visitSINT_TO_FP(SDNode *N) { N0.getOperand(0).getOpcode() == ISD::SETCC &&!VT.isVector() && (!LegalOperations || TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) { + SDLoc DL(N); SDValue Ops[] = { N0.getOperand(0).getOperand(0), N0.getOperand(0).getOperand(1), - DAG.getConstantFP(1.0, VT) , DAG.getConstantFP(0.0, VT), + DAG.getConstantFP(1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT), N0.getOperand(0).getOperand(2) }; - return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops); + return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops); } } @@ -7664,12 +8496,11 @@ SDValue DAGCombiner::visitSINT_TO_FP(SDNode *N) { SDValue DAGCombiner::visitUINT_TO_FP(SDNode *N) { SDValue N0 = N->getOperand(0); - ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0); EVT VT = N->getValueType(0); EVT OpVT = N0.getValueType(); // fold (uint_to_fp c1) -> c1fp - if (N0C && + if (isConstantIntBuildVectorOrConstantInt(N0) && // ...but only if the target supports immediate floating-point values (!LegalOperations || TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) @@ -7691,39 +8522,82 @@ SDValue DAGCombiner::visitUINT_TO_FP(SDNode *N) { if (N0.getOpcode() == ISD::SETCC && !VT.isVector() && (!LegalOperations || TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) { + SDLoc DL(N); SDValue Ops[] = { N0.getOperand(0), N0.getOperand(1), - DAG.getConstantFP(1.0, VT), DAG.getConstantFP(0.0, VT), + DAG.getConstantFP(1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT), N0.getOperand(2) }; - return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops); + return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops); } } return SDValue(); } +// Fold (fp_to_{s/u}int ({s/u}int_to_fpx)) -> zext x, sext x, trunc x, or x +static SDValue FoldIntToFPToInt(SDNode *N, SelectionDAG &DAG) { + SDValue N0 = N->getOperand(0); + EVT VT = N->getValueType(0); + + if (N0.getOpcode() != ISD::UINT_TO_FP && N0.getOpcode() != ISD::SINT_TO_FP) + return SDValue(); + + SDValue Src = N0.getOperand(0); + EVT SrcVT = Src.getValueType(); + bool IsInputSigned = N0.getOpcode() == ISD::SINT_TO_FP; + bool IsOutputSigned = N->getOpcode() == ISD::FP_TO_SINT; + + // We can safely assume the conversion won't overflow the output range, + // because (for example) (uint8_t)18293.f is undefined behavior. + + // Since we can assume the conversion won't overflow, our decision as to + // whether the input will fit in the float should depend on the minimum + // of the input range and output range. + + // This means this is also safe for a signed input and unsigned output, since + // a negative input would lead to undefined behavior. + unsigned InputSize = (int)SrcVT.getScalarSizeInBits() - IsInputSigned; + unsigned OutputSize = (int)VT.getScalarSizeInBits() - IsOutputSigned; + unsigned ActualSize = std::min(InputSize, OutputSize); + const fltSemantics &sem = DAG.EVTToAPFloatSemantics(N0.getValueType()); + + // We can only fold away the float conversion if the input range can be + // represented exactly in the float range. + if (APFloat::semanticsPrecision(sem) >= ActualSize) { + if (VT.getScalarSizeInBits() > SrcVT.getScalarSizeInBits()) { + unsigned ExtOp = IsInputSigned && IsOutputSigned ? ISD::SIGN_EXTEND + : ISD::ZERO_EXTEND; + return DAG.getNode(ExtOp, SDLoc(N), VT, Src); + } + if (VT.getScalarSizeInBits() < SrcVT.getScalarSizeInBits()) + return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, Src); + if (SrcVT == VT) + return Src; + return DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Src); + } + return SDValue(); +} + SDValue DAGCombiner::visitFP_TO_SINT(SDNode *N) { SDValue N0 = N->getOperand(0); - ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0); EVT VT = N->getValueType(0); // fold (fp_to_sint c1fp) -> c1 - if (N0CFP) + if (isConstantFPBuildVectorOrConstantFP(N0)) return DAG.getNode(ISD::FP_TO_SINT, SDLoc(N), VT, N0); - return SDValue(); + return FoldIntToFPToInt(N, DAG); } SDValue DAGCombiner::visitFP_TO_UINT(SDNode *N) { SDValue N0 = N->getOperand(0); - ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0); EVT VT = N->getValueType(0); // fold (fp_to_uint c1fp) -> c1 - if (N0CFP) + if (isConstantFPBuildVectorOrConstantFP(N0)) return DAG.getNode(ISD::FP_TO_UINT, SDLoc(N), VT, N0); - return SDValue(); + return FoldIntToFPToInt(N, DAG); } SDValue DAGCombiner::visitFP_ROUND(SDNode *N) { @@ -7742,11 +8616,18 @@ SDValue DAGCombiner::visitFP_ROUND(SDNode *N) { // fold (fp_round (fp_round x)) -> (fp_round x) if (N0.getOpcode() == ISD::FP_ROUND) { - // This is a value preserving truncation if both round's are. - bool IsTrunc = N->getConstantOperandVal(1) == 1 && - N0.getNode()->getConstantOperandVal(1) == 1; - return DAG.getNode(ISD::FP_ROUND, SDLoc(N), VT, N0.getOperand(0), - DAG.getIntPtrConstant(IsTrunc)); + const bool NIsTrunc = N->getConstantOperandVal(1) == 1; + const bool N0IsTrunc = N0.getNode()->getConstantOperandVal(1) == 1; + // If the first fp_round isn't a value preserving truncation, it might + // introduce a tie in the second fp_round, that wouldn't occur in the + // single-step fp_round we want to fold to. + // In other words, double rounding isn't the same as rounding. + // Also, this is a value preserving truncation iff both fp_round's are. + if (DAG.getTarget().Options.UnsafeFPMath || N0IsTrunc) { + SDLoc DL(N); + return DAG.getNode(ISD::FP_ROUND, DL, VT, N0.getOperand(0), + DAG.getIntPtrConstant(NIsTrunc && N0IsTrunc, DL)); + } } // fold (fp_round (copysign X, Y)) -> (copysign (fp_round X), Y) @@ -7769,8 +8650,9 @@ SDValue DAGCombiner::visitFP_ROUND_INREG(SDNode *N) { // fold (fp_round_inreg c1fp) -> c1fp if (N0CFP && isTypeLegal(EVT)) { - SDValue Round = DAG.getConstantFP(*N0CFP->getConstantFPValue(), EVT); - return DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT, Round); + SDLoc DL(N); + SDValue Round = DAG.getConstantFP(*N0CFP->getConstantFPValue(), DL, EVT); + return DAG.getNode(ISD::FP_EXTEND, DL, VT, Round); } return SDValue(); @@ -7778,7 +8660,6 @@ SDValue DAGCombiner::visitFP_ROUND_INREG(SDNode *N) { 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. @@ -7787,9 +8668,14 @@ SDValue DAGCombiner::visitFP_EXTEND(SDNode *N) { 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 @@ -7813,7 +8699,8 @@ SDValue DAGCombiner::visitFP_EXTEND(SDNode *N) { CombineTo(N, ExtLoad); CombineTo(N0.getNode(), DAG.getNode(ISD::FP_ROUND, SDLoc(N0), - N0.getValueType(), ExtLoad, DAG.getIntPtrConstant(1)), + N0.getValueType(), ExtLoad, + DAG.getIntPtrConstant(1, SDLoc(N0))), ExtLoad.getValue(1)); return SDValue(N, 0); // Return N so it doesn't get rechecked! } @@ -7823,11 +8710,10 @@ SDValue DAGCombiner::visitFP_EXTEND(SDNode *N) { 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(); @@ -7835,11 +8721,10 @@ SDValue DAGCombiner::visitFCEIL(SDNode *N) { 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(); @@ -7847,11 +8732,10 @@ SDValue DAGCombiner::visitFTRUNC(SDNode *N) { 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(); @@ -7862,14 +8746,9 @@ SDValue DAGCombiner::visitFNEG(SDNode *N) { 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)) @@ -7893,15 +8772,17 @@ SDValue DAGCombiner::visitFNEG(SDNode *N) { // For a scalar, just generate 0x80... SignMask = APInt::getSignBit(IntVT.getSizeInBits()); } - Int = DAG.getNode(ISD::XOR, SDLoc(N0), IntVT, Int, - DAG.getConstant(SignMask, IntVT)); + SDLoc DL0(N0); + Int = DAG.getNode(ISD::XOR, DL0, IntVT, Int, + DAG.getConstant(SignMask, DL0, IntVT)); AddToWorklist(Int.getNode()); return DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Int); } } // (fneg (fmul c, x)) -> (fmul -c, x) - if (N0.getOpcode() == ISD::FMUL) { + if (N0.getOpcode() == ISD::FMUL && + (N0.getNode()->hasOneUse() || !TLI.isFNegFree(VT))) { ConstantFPSDNode *CFP1 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1)); if (CFP1) { APFloat CVal = CFP1->getValueAPF(); @@ -7927,7 +8808,7 @@ SDValue DAGCombiner::visitFMINNUM(SDNode *N) { if (N0CFP && N1CFP) { const APFloat &C0 = N0CFP->getValueAPF(); const APFloat &C1 = N1CFP->getValueAPF(); - return DAG.getConstantFP(minnum(C0, C1), N->getValueType(0)); + return DAG.getConstantFP(minnum(C0, C1), SDLoc(N), N->getValueType(0)); } if (N0CFP) { @@ -7948,7 +8829,7 @@ SDValue DAGCombiner::visitFMAXNUM(SDNode *N) { if (N0CFP && N1CFP) { const APFloat &C0 = N0CFP->getValueAPF(); const APFloat &C1 = N1CFP->getValueAPF(); - return DAG.getConstantFP(maxnum(C0, C1), N->getValueType(0)); + return DAG.getConstantFP(maxnum(C0, C1), SDLoc(N), N->getValueType(0)); } if (N0CFP) { @@ -7964,13 +8845,8 @@ SDValue DAGCombiner::visitFABS(SDNode *N) { 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) @@ -8000,8 +8876,9 @@ SDValue DAGCombiner::visitFABS(SDNode *N) { // For a scalar, just generate 0x7f... SignMask = ~APInt::getSignBit(IntVT.getSizeInBits()); } - Int = DAG.getNode(ISD::AND, SDLoc(N0), IntVT, Int, - DAG.getConstant(SignMask, IntVT)); + SDLoc DL(N0); + Int = DAG.getNode(ISD::AND, DL, IntVT, Int, + DAG.getConstant(SignMask, DL, IntVT)); AddToWorklist(Int.getNode()); return DAG.getNode(ISD::BITCAST, SDLoc(N), N->getValueType(0), Int); } @@ -8071,13 +8948,14 @@ SDValue DAGCombiner::visitBRCOND(SDNode *N) { if (AndConst.isPowerOf2() && cast<ConstantSDNode>(Op1)->getAPIntValue()==AndConst.logBase2()) { + SDLoc DL(N); SDValue SetCC = - DAG.getSetCC(SDLoc(N), + DAG.getSetCC(DL, getSetCCResultType(Op0.getValueType()), - Op0, DAG.getConstant(0, Op0.getValueType()), + Op0, DAG.getConstant(0, DL, Op0.getValueType()), ISD::SETNE); - SDValue NewBRCond = DAG.getNode(ISD::BRCOND, SDLoc(N), + SDValue NewBRCond = DAG.getNode(ISD::BRCOND, DL, MVT::Other, Chain, SetCC, N2); // Don't add the new BRCond into the worklist or else SimplifySelectCC // will convert it back to (X & C1) >> C2. @@ -8130,12 +9008,11 @@ SDValue DAGCombiner::visitBRCOND(SDNode *N) { if (Op0.getOpcode() != ISD::SETCC && Op1.getOpcode() != ISD::SETCC) { bool Equal = false; - if (ConstantSDNode *RHSCI = dyn_cast<ConstantSDNode>(Op0)) - if (RHSCI->getAPIntValue() == 1 && Op0.hasOneUse() && - Op0.getOpcode() == ISD::XOR) { - TheXor = Op0.getNode(); - Equal = true; - } + if (isOneConstant(Op0) && Op0.hasOneUse() && + Op0.getOpcode() == ISD::XOR) { + TheXor = Op0.getNode(); + Equal = true; + } EVT SetCCVT = N1.getValueType(); if (LegalTypes) @@ -8190,14 +9067,18 @@ static bool canFoldInAddressingMode(SDNode *N, SDNode *Use, SelectionDAG &DAG, const TargetLowering &TLI) { EVT VT; + unsigned AS; + if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Use)) { if (LD->isIndexed() || LD->getBasePtr().getNode() != N) return false; - VT = Use->getValueType(0); + VT = LD->getMemoryVT(); + AS = LD->getAddressSpace(); } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(Use)) { if (ST->isIndexed() || ST->getBasePtr().getNode() != N) return false; - VT = ST->getValue().getValueType(); + VT = ST->getMemoryVT(); + AS = ST->getAddressSpace(); } else return false; @@ -8221,7 +9102,7 @@ static bool canFoldInAddressingMode(SDNode *N, SDNode *Use, } else return false; - return TLI.isLegalAddressingMode(AM, VT.getTypeForEVT(*DAG.getContext())); + return TLI.isLegalAddressingMode(AM, VT.getTypeForEVT(*DAG.getContext()), AS); } /// Try turning a load/store into a pre-indexed load/store when the base @@ -8280,8 +9161,7 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) { } // Don't create a indexed load / store with zero offset. - if (isa<ConstantSDNode>(Offset) && - cast<ConstantSDNode>(Offset)->isNullValue()) + if (isNullConstant(Offset)) return false; // Try turning it into a pre-indexed load / store except when: @@ -8309,24 +9189,25 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) { // a copy of the original base pointer. SmallVector<SDNode *, 16> OtherUses; if (isa<ConstantSDNode>(Offset)) - for (SDNode *Use : BasePtr.getNode()->uses()) { - if (Use == Ptr.getNode()) + for (SDNode::use_iterator UI = BasePtr.getNode()->use_begin(), + UE = BasePtr.getNode()->use_end(); + UI != UE; ++UI) { + SDUse &Use = UI.getUse(); + // Skip the use that is Ptr and uses of other results from BasePtr's + // node (important for nodes that return multiple results). + if (Use.getUser() == Ptr.getNode() || Use != BasePtr) continue; - if (Use->isPredecessorOf(N)) + if (Use.getUser()->isPredecessorOf(N)) continue; - if (Use->getOpcode() != ISD::ADD && Use->getOpcode() != ISD::SUB) { + if (Use.getUser()->getOpcode() != ISD::ADD && + Use.getUser()->getOpcode() != ISD::SUB) { OtherUses.clear(); break; } - SDValue Op0 = Use->getOperand(0), Op1 = Use->getOperand(1); - if (Op1.getNode() == BasePtr.getNode()) - std::swap(Op0, Op1); - assert(Op0.getNode() == BasePtr.getNode() && - "Use of ADD/SUB but not an operand"); - + SDValue Op1 = Use.getUser()->getOperand((UI.getOperandNo() + 1) & 1); if (!isa<ConstantSDNode>(Op1)) { OtherUses.clear(); break; @@ -8338,7 +9219,7 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) { break; } - OtherUses.push_back(Use); + OtherUses.push_back(Use.getUser()); } if (Swapped) @@ -8431,12 +9312,14 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) { if (X1 * Y0 * Y1 < 0) CNV = CNV + Offset1; else CNV = CNV - Offset1; + SDLoc DL(OtherUses[i]); + // We can now generate the new expression. - SDValue NewOp1 = DAG.getConstant(CNV, CN->getValueType(0)); + SDValue NewOp1 = DAG.getConstant(CNV, DL, CN->getValueType(0)); SDValue NewOp2 = Result.getValue(isLoad ? 1 : 0); SDValue NewUse = DAG.getNode(Opcode, - SDLoc(OtherUses[i]), + DL, OtherUses[i]->getValueType(0), NewOp1, NewOp2); DAG.ReplaceAllUsesOfValueWith(SDValue(OtherUses[i], 0), NewUse); deleteAndRecombine(OtherUses[i]); @@ -8494,8 +9377,7 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) { ISD::MemIndexedMode AM = ISD::UNINDEXED; if (TLI.getPostIndexedAddressParts(N, Op, BasePtr, Offset, AM, DAG)) { // Don't create a indexed load / store with zero offset. - if (isa<ConstantSDNode>(Offset) && - cast<ConstantSDNode>(Offset)->isNullValue()) + if (isNullConstant(Offset)) continue; // Try turning it into a post-indexed load / store except when @@ -8585,7 +9467,7 @@ SDValue DAGCombiner::SplitIndexingFromLoad(LoadSDNode *LD) { "Cannot split out indexing using opaque target constants"); if (Inc.getOpcode() == ISD::TargetConstant) { ConstantSDNode *ConstInc = cast<ConstantSDNode>(Inc); - Inc = DAG.getConstant(*ConstInc->getConstantIntValue(), + Inc = DAG.getConstant(*ConstInc->getConstantIntValue(), SDLoc(Inc), ConstInc->getValueType(0)); } @@ -8686,7 +9568,8 @@ SDValue DAGCombiner::visitLOAD(SDNode *N) { 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); } } } @@ -8851,9 +9734,6 @@ struct LoadedSlice { 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. @@ -8980,8 +9860,9 @@ struct LoadedSlice { if (Offset) { // BaseAddr = BaseAddr + Offset. EVT ArithType = BaseAddr.getValueType(); - BaseAddr = DAG->getNode(ISD::ADD, SDLoc(Origin), ArithType, BaseAddr, - DAG->getConstant(Offset, ArithType)); + SDLoc DL(Origin); + BaseAddr = DAG->getNode(ISD::ADD, DL, ArithType, BaseAddr, + DAG->getConstant(Offset, DL, ArithType)); } // Create the type of the loaded slice according to its size. @@ -9336,7 +10217,7 @@ CheckForMaskedLoad(SDValue V, SDValue Ptr, SDValue Chain) { if (NotMaskLZ == 64) return Result; // All zero mask. // See if we have a continuous run of bits. If so, we have 0*1+0* - if (CountTrailingOnes_64(NotMask >> NotMaskTZ)+NotMaskTZ+NotMaskLZ != 64) + if (countTrailingOnes(NotMask >> NotMaskTZ) + NotMaskTZ + NotMaskLZ != 64) return Result; // Adjust NotMaskLZ down to be from the actual size of the int instead of i64. @@ -9387,10 +10268,12 @@ ShrinkLoadReplaceStoreWithStore(const std::pair<unsigned, unsigned> &MaskInfo, // Okay, we can do this! Replace the 'St' store with a store of IVal that is // shifted by ByteShift and truncated down to NumBytes. - if (ByteShift) - IVal = DAG.getNode(ISD::SRL, SDLoc(IVal), IVal.getValueType(), IVal, - DAG.getConstant(ByteShift*8, + if (ByteShift) { + SDLoc DL(IVal); + IVal = DAG.getNode(ISD::SRL, DL, IVal.getValueType(), IVal, + DAG.getConstant(ByteShift*8, DL, DC->getShiftAmountTy(IVal.getValueType()))); + } // Figure out the offset for the store and the alignment of the access. unsigned StOffset; @@ -9403,8 +10286,9 @@ ShrinkLoadReplaceStoreWithStore(const std::pair<unsigned, unsigned> &MaskInfo, SDValue Ptr = St->getBasePtr(); if (StOffset) { - Ptr = DAG.getNode(ISD::ADD, SDLoc(IVal), Ptr.getValueType(), - Ptr, DAG.getConstant(StOffset, Ptr.getValueType())); + SDLoc DL(IVal); + Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), + Ptr, DAG.getConstant(StOffset, DL, Ptr.getValueType())); NewAlign = MinAlign(NewAlign, StOffset); } @@ -9486,8 +10370,9 @@ SDValue DAGCombiner::ReduceLoadOpStoreWidth(SDNode *N) { // The narrowing should be profitable, the load/store operation should be // legal (or custom) and the store size should be equal to the NewVT width. while (NewBW < BitWidth && - !(TLI.isOperationLegalOrCustom(Opc, NewVT) && - TLI.isNarrowingProfitable(VT, NewVT))) { + (NewVT.getStoreSizeInBits() != NewBW || + !TLI.isOperationLegalOrCustom(Opc, NewVT) || + !TLI.isNarrowingProfitable(VT, NewVT))) { NewBW = NextPowerOf2(NewBW); NewVT = EVT::getIntegerVT(*DAG.getContext(), NewBW); } @@ -9517,7 +10402,8 @@ SDValue DAGCombiner::ReduceLoadOpStoreWidth(SDNode *N) { SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(LD), Ptr.getValueType(), Ptr, - DAG.getConstant(PtrOff, Ptr.getValueType())); + DAG.getConstant(PtrOff, SDLoc(LD), + Ptr.getValueType())); SDValue NewLD = DAG.getLoad(NewVT, SDLoc(N0), LD->getChain(), NewPtr, LD->getPointerInfo().getWithOffset(PtrOff), @@ -9525,7 +10411,8 @@ SDValue DAGCombiner::ReduceLoadOpStoreWidth(SDNode *N) { LD->isInvariant(), NewAlign, LD->getAAInfo()); SDValue NewVal = DAG.getNode(Opc, SDLoc(Value), NewVT, NewLD, - DAG.getConstant(NewImm, NewVT)); + DAG.getConstant(NewImm, SDLoc(Value), + NewVT)); SDValue NewST = DAG.getStore(Chain, SDLoc(N), NewVal, NewPtr, ST->getPointerInfo().getWithOffset(PtrOff), @@ -9599,6 +10486,7 @@ SDValue DAGCombiner::TransformFPLoadStorePair(SDNode *N) { 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 @@ -9686,37 +10574,156 @@ struct BaseIndexOffset { return BaseIndexOffset(Base, Index, Off, IsIndexSignExt); } }; +} // namespace -/// Holds a pointer to an LSBaseSDNode as well as information on where it -/// is located in a sequence of memory operations connected by a chain. -struct MemOpLink { - MemOpLink (LSBaseSDNode *N, int64_t Offset, unsigned Seq): - MemNode(N), OffsetFromBase(Offset), SequenceNum(Seq) { } - // Ptr to the mem node. - LSBaseSDNode *MemNode; - // Offset from the base ptr. - int64_t OffsetFromBase; - // What is the sequence number of this mem node. - // Lowest mem operand in the DAG starts at zero. - unsigned SequenceNum; -}; +bool DAGCombiner::MergeStoresOfConstantsOrVecElts( + SmallVectorImpl<MemOpLink> &StoreNodes, EVT MemVT, + unsigned NumElem, bool IsConstantSrc, bool UseVector) { + // Make sure we have something to merge. + if (NumElem < 2) + return false; + + int64_t ElementSizeBytes = MemVT.getSizeInBits() / 8; + LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode; + 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 + // latest store node which is *used* and replaced by the wide store. + if (StoreNodes[i].SequenceNum < StoreNodes[LatestNodeUsed].SequenceNum) + LatestNodeUsed = i; + } + + // 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. + EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem); + assert(TLI.isTypeLegal(Ty) && "Illegal vector store"); + if (IsConstantSrc) { + // A vector store with a constant source implies that the constant is + // zero; we only handle merging stores of constant zeros because the zero + // can be materialized without a load. + // It may be beneficial to loosen this restriction to allow non-zero + // store merging. + StoredVal = DAG.getConstant(0, DL, Ty); + } else { + SmallVector<SDValue, 8> Ops; + for (unsigned i = 0; i < NumElem ; ++i) { + StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode); + SDValue Val = St->getValue(); + // All of the operands of a BUILD_VECTOR must have the same type. + if (Val.getValueType() != MemVT) + return false; + Ops.push_back(Val); + } + + // Build the extracted vector elements back into a vector. + StoredVal = DAG.getNode(ISD::BUILD_VECTOR, DL, Ty, Ops); + } + } else { + // We should always use a vector store when merging extracted vector + // elements, so this path implies a store of constants. + assert(IsConstantSrc && "Merged vector elements should use vector store"); + + 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(); + for (unsigned i = 0; i < NumElem ; ++i) { + unsigned Idx = IsLE ? (NumElem - 1 - i) : i; + StoreSDNode *St = cast<StoreSDNode>(StoreNodes[Idx].MemNode); + SDValue Val = St->getValue(); + StoreInt <<= ElementSizeBytes*8; + if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Val)) { + StoreInt |= C->getAPIntValue().zext(StoreBW); + } else if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Val)) { + StoreInt |= C->getValueAPF().bitcastToAPInt().zext(StoreBW); + } else { + llvm_unreachable("Invalid constant element type"); + } + } + + // Create the new Load and Store operations. + EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW); + StoredVal = DAG.getConstant(StoreInt, DL, StoreTy); + } + + SDValue NewStore = DAG.getStore(LatestOp->getChain(), DL, StoredVal, + FirstInChain->getBasePtr(), + FirstInChain->getPointerInfo(), + false, false, + FirstInChain->getAlignment()); + + // 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 == LatestOp) + continue; + StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode); + // ReplaceAllUsesWith will replace all uses that existed when it was + // called, but graph optimizations may cause new ones to appear. For + // example, the case in pr14333 looks like + // + // St's chain -> St -> another store -> X + // + // And the only difference from St to the other store is the chain. + // When we change it's chain to be St's chain they become identical, + // get CSEed and the net result is that X is now a use of St. + // Since we know that St is redundant, just iterate. + while (!St->use_empty()) + DAG.ReplaceAllUsesWith(SDValue(St, 0), St->getChain()); + deleteAndRecombine(St); + } + + return true; +} + +static bool allowableAlignment(const SelectionDAG &DAG, + const TargetLowering &TLI, EVT EVTTy, + unsigned AS, unsigned Align) { + if (TLI.allowsMisalignedMemoryAccesses(EVTTy, AS, Align)) + return true; + + Type *Ty = EVTTy.getTypeForEVT(*DAG.getContext()); + unsigned ABIAlignment = TLI.getDataLayout()->getPrefTypeAlignment(Ty); + return (Align >= ABIAlignment); +} 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()->getAttributes(). - hasAttribute(AttributeSet::FunctionIndex, Attribute::NoImplicitFloat); + bool NoVectors = DAG.getMachineFunction().getFunction()->hasFnAttribute( + Attribute::NoImplicitFloat); + + // This function cannot currently deal with non-byte-sized memory sizes. + if (ElementSizeBytes * 8 != MemVT.getSizeInBits()) + return false; // Don't merge vectors into wider inputs. if (MemVT.isVector() || !MemVT.isSimple()) return false; // Perform an early exit check. Do not bother looking at stored values that - // are not constants or loads. + // are not constants, loads, or extracted vector elements. SDValue StoredVal = St->getValue(); bool IsLoadSrc = isa<LoadSDNode>(StoredVal); - if (!isa<ConstantSDNode>(StoredVal) && !isa<ConstantFPSDNode>(StoredVal) && - !IsLoadSrc) + bool IsConstantSrc = isa<ConstantSDNode>(StoredVal) || + isa<ConstantFPSDNode>(StoredVal); + bool IsExtractVecEltSrc = (StoredVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT); + + if (!IsConstantSrc && !IsLoadSrc && !IsExtractVecEltSrc) return false; // Only look at ends of store sequences. @@ -9761,10 +10768,6 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { if (!Ptr.equalBaseIndex(BasePtr)) break; - // Check that the alignment is the same. - if (Index->getAlignment() != St->getAlignment()) - break; - // The memory operands must not be volatile. if (Index->isVolatile() || Index->isIndexed()) break; @@ -9778,11 +10781,6 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { if (Index->getMemoryVT() != MemVT) break; - // We do not allow unaligned stores because we want to prevent overriding - // stores. - if (Index->getAlignment()*8 != MemVT.getSizeInBits()) - break; - // We found a potential memory operand to merge. StoreNodes.push_back(MemOpLink(Index, Ptr.Offset, Seq++)); @@ -9856,9 +10854,11 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { // The node with the lowest store address. LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode; + unsigned FirstStoreAS = FirstInChain->getAddressSpace(); + unsigned FirstStoreAlign = FirstInChain->getAlignment(); // Store the constants into memory as one consecutive store. - if (!IsLoadSrc) { + if (IsConstantSrc) { unsigned LastLegalType = 0; unsigned LastLegalVectorType = 0; bool NonZero = false; @@ -9878,27 +10878,41 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { // Find a legal type for the constant store. unsigned StoreBW = (i+1) * ElementSizeBytes * 8; EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW); - if (TLI.isTypeLegal(StoreTy)) + if (TLI.isTypeLegal(StoreTy) && + allowableAlignment(DAG, TLI, StoreTy, FirstStoreAS, + FirstStoreAlign)) { LastLegalType = i+1; // Or check whether a truncstore is legal. - else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) == - TargetLowering::TypePromoteInteger) { + } else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) == + TargetLowering::TypePromoteInteger) { EVT LegalizedStoredValueTy = TLI.getTypeToTransformTo(*DAG.getContext(), StoredVal.getValueType()); - if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy)) - LastLegalType = i+1; + if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy) && + allowableAlignment(DAG, TLI, LegalizedStoredValueTy, FirstStoreAS, + FirstStoreAlign)) { + LastLegalType = i + 1; + } } // Find a legal type for the vector store. EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1); - if (TLI.isTypeLegal(Ty)) + if (TLI.isTypeLegal(Ty) && + allowableAlignment(DAG, TLI, Ty, FirstStoreAS, FirstStoreAlign)) { LastLegalVectorType = i + 1; + } } - // We only use vectors if the constant is known to be zero and the - // function is not marked with the noimplicitfloat attribute. - if (NonZero || NoVectors) + + // We only use vectors if the constant is known to be zero or the target + // allows it and the function is not marked with the noimplicitfloat + // attribute. + if (NoVectors) { LastLegalVectorType = 0; + } else if (NonZero && !TLI.storeOfVectorConstantIsCheap(MemVT, + LastLegalVectorType, + FirstStoreAS)) { + LastLegalVectorType = 0; + } // Check if we found a legal integer type to store. if (LastLegalType == 0 && LastLegalVectorType == 0) @@ -9907,85 +10921,34 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { bool UseVector = (LastLegalVectorType > LastLegalType) && !NoVectors; unsigned NumElem = UseVector ? LastLegalVectorType : LastLegalType; - // Make sure we have something to merge. - if (NumElem < 2) - return false; - - unsigned EarliestNodeUsed = 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; - } + return MergeStoresOfConstantsOrVecElts(StoreNodes, MemVT, NumElem, + true, UseVector); + } - // The earliest Node in the DAG. - LSBaseSDNode *EarliestOp = StoreNodes[EarliestNodeUsed].MemNode; - SDLoc DL(StoreNodes[0].MemNode); + // When extracting multiple vector elements, try to store them + // in one vector store rather than a sequence of scalar stores. + if (IsExtractVecEltSrc) { + unsigned NumElem = 0; + for (unsigned i = 0; i < LastConsecutiveStore + 1; ++i) { + StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode); + SDValue StoredVal = St->getValue(); + // This restriction could be loosened. + // Bail out if any stored values are not elements extracted from a vector. + // It should be possible to handle mixed sources, but load sources need + // more careful handling (see the block of code below that handles + // consecutive loads). + if (StoredVal.getOpcode() != ISD::EXTRACT_VECTOR_ELT) + return false; - SDValue StoredVal; - if (UseVector) { // Find a legal type for the vector store. - EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem); - assert(TLI.isTypeLegal(Ty) && "Illegal vector store"); - StoredVal = DAG.getConstant(0, Ty); - } else { - 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(); - for (unsigned i = 0; i < NumElem ; ++i) { - unsigned Idx = IsLE ?(NumElem - 1 - i) : i; - StoreSDNode *St = cast<StoreSDNode>(StoreNodes[Idx].MemNode); - SDValue Val = St->getValue(); - StoreInt<<=ElementSizeBytes*8; - if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Val)) { - StoreInt|=C->getAPIntValue().zext(StoreBW); - } else if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Val)) { - StoreInt|= C->getValueAPF().bitcastToAPInt().zext(StoreBW); - } else { - 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, - FirstInChain->getBasePtr(), - FirstInChain->getPointerInfo(), - false, false, - FirstInChain->getAlignment()); - - // Replace the first store with the new store - CombineTo(EarliestOp, NewStore); - // Erase all other stores. - for (unsigned i = 0; i < NumElem ; ++i) { - if (StoreNodes[i].MemNode == EarliestOp) - continue; - StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode); - // ReplaceAllUsesWith will replace all uses that existed when it was - // called, but graph optimizations may cause new ones to appear. For - // example, the case in pr14333 looks like - // - // St's chain -> St -> another store -> X - // - // And the only difference from St to the other store is the chain. - // When we change it's chain to be St's chain they become identical, - // get CSEed and the net result is that X is now a use of St. - // Since we know that St is redundant, just iterate. - while (!St->use_empty()) - DAG.ReplaceAllUsesWith(SDValue(St, 0), St->getChain()); - deleteAndRecombine(St); + EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1); + if (TLI.isTypeLegal(Ty) && + allowableAlignment(DAG, TLI, Ty, FirstStoreAS, FirstStoreAlign)) + NumElem = i + 1; } - return true; + return MergeStoresOfConstantsOrVecElts(StoreNodes, MemVT, NumElem, + false, true); } // Below we handle the case of multiple consecutive stores that @@ -10007,10 +10970,6 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { if (!Ld->hasNUsesOfValue(1, 0)) break; - // Check that the alignment is the same as the stores. - if (Ld->getAlignment() != St->getAlignment()) - break; - // The memory operands must not be volatile. if (Ld->isVolatile() || Ld->isIndexed()) break; @@ -10048,6 +11007,10 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { St->getAlignment() >= RequiredAlignment) return false; + LoadSDNode *FirstLoad = cast<LoadSDNode>(LoadNodes[0].MemNode); + unsigned FirstLoadAS = FirstLoad->getAddressSpace(); + unsigned FirstLoadAlign = FirstLoad->getAlignment(); + // Scan the memory operations on the chain and find the first non-consecutive // load memory address. These variables hold the index in the store node // array. @@ -10056,7 +11019,7 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { unsigned LastLegalVectorType = 0; unsigned LastLegalIntegerType = 0; StartAddress = LoadNodes[0].OffsetFromBase; - SDValue FirstChain = LoadNodes[0].MemNode->getChain(); + SDValue FirstChain = FirstLoad->getChain(); for (unsigned i = 1; i < LoadNodes.size(); ++i) { // All loads much share the same chain. if (LoadNodes[i].MemNode->getChain() != FirstChain) @@ -10069,13 +11032,18 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { // Find a legal type for the vector store. EVT StoreTy = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1); - if (TLI.isTypeLegal(StoreTy)) + if (TLI.isTypeLegal(StoreTy) && + allowableAlignment(DAG, TLI, StoreTy, FirstStoreAS, FirstStoreAlign) && + allowableAlignment(DAG, TLI, StoreTy, FirstLoadAS, FirstLoadAlign)) { LastLegalVectorType = i + 1; + } // Find a legal type for the integer store. unsigned StoreBW = (i+1) * ElementSizeBytes * 8; StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW); - if (TLI.isTypeLegal(StoreTy)) + if (TLI.isTypeLegal(StoreTy) && + allowableAlignment(DAG, TLI, StoreTy, FirstStoreAS, FirstStoreAlign) && + allowableAlignment(DAG, TLI, StoreTy, FirstLoadAS, FirstLoadAlign)) LastLegalIntegerType = i + 1; // Or check whether a truncstore and extload is legal. else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) == @@ -10085,7 +11053,11 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy) && TLI.isLoadExtLegal(ISD::ZEXTLOAD, LegalizedStoredValueTy, StoreTy) && TLI.isLoadExtLegal(ISD::SEXTLOAD, LegalizedStoredValueTy, StoreTy) && - TLI.isLoadExtLegal(ISD::EXTLOAD, LegalizedStoredValueTy, StoreTy)) + TLI.isLoadExtLegal(ISD::EXTLOAD, LegalizedStoredValueTy, StoreTy) && + allowableAlignment(DAG, TLI, LegalizedStoredValueTy, FirstStoreAS, + FirstStoreAlign) && + allowableAlignment(DAG, TLI, LegalizedStoredValueTy, FirstLoadAS, + FirstLoadAlign)) LastLegalIntegerType = i+1; } } @@ -10103,18 +11075,19 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { 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; @@ -10128,18 +11101,13 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { SDLoc LoadDL(LoadNodes[0].MemNode); SDLoc StoreDL(StoreNodes[0].MemNode); - LoadSDNode *FirstLoad = cast<LoadSDNode>(LoadNodes[0].MemNode); - SDValue NewLoad = DAG.getLoad(JointMemOpVT, LoadDL, - FirstLoad->getChain(), - FirstLoad->getBasePtr(), - FirstLoad->getPointerInfo(), - false, false, false, - FirstLoad->getAlignment()); - - SDValue NewStore = DAG.getStore(EarliestOp->getChain(), StoreDL, NewLoad, - FirstInChain->getBasePtr(), - FirstInChain->getPointerInfo(), false, false, - FirstInChain->getAlignment()); + SDValue NewLoad = DAG.getLoad( + JointMemOpVT, LoadDL, FirstLoad->getChain(), FirstLoad->getBasePtr(), + FirstLoad->getPointerInfo(), false, false, false, FirstLoadAlign); + + SDValue NewStore = DAG.getStore( + LatestOp->getChain(), StoreDL, NewLoad, FirstInChain->getBasePtr(), + FirstInChain->getPointerInfo(), false, false, FirstStoreAlign); // Replace one of the loads with the new load. LoadSDNode *Ld = cast<LoadSDNode>(LoadNodes[0].MemNode); @@ -10154,12 +11122,12 @@ bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) { 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()); @@ -10214,8 +11182,10 @@ SDValue DAGCombiner::visitSTORE(SDNode *N) { case MVT::f32: if ((isTypeLegal(MVT::i32) && !LegalOperations && !ST->isVolatile()) || TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) { + ; Tmp = DAG.getConstant((uint32_t)CFP->getValueAPF(). - bitcastToAPInt().getZExtValue(), MVT::i32); + bitcastToAPInt().getZExtValue(), SDLoc(CFP), + MVT::i32); return DAG.getStore(Chain, SDLoc(N), Tmp, Ptr, ST->getMemOperand()); } @@ -10224,8 +11194,9 @@ SDValue DAGCombiner::visitSTORE(SDNode *N) { if ((TLI.isTypeLegal(MVT::i64) && !LegalOperations && !ST->isVolatile()) || TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i64)) { + ; Tmp = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt(). - getZExtValue(), MVT::i64); + getZExtValue(), SDLoc(CFP), MVT::i64); return DAG.getStore(Chain, SDLoc(N), Tmp, Ptr, ST->getMemOperand()); } @@ -10236,8 +11207,8 @@ SDValue DAGCombiner::visitSTORE(SDNode *N) { // argument passing. Since this is so common, custom legalize the // 64-bit integer store into two 32-bit stores. uint64_t Val = CFP->getValueAPF().bitcastToAPInt().getZExtValue(); - SDValue Lo = DAG.getConstant(Val & 0xFFFFFFFF, MVT::i32); - SDValue Hi = DAG.getConstant(Val >> 32, MVT::i32); + SDValue Lo = DAG.getConstant(Val & 0xFFFFFFFF, SDLoc(CFP), MVT::i32); + SDValue Hi = DAG.getConstant(Val >> 32, SDLoc(CFP), MVT::i32); if (TLI.isBigEndian()) std::swap(Lo, Hi); unsigned Alignment = ST->getAlignment(); @@ -10245,18 +11216,20 @@ SDValue DAGCombiner::visitSTORE(SDNode *N) { bool isNonTemporal = ST->isNonTemporal(); AAMDNodes AAInfo = ST->getAAInfo(); + SDLoc DL(N); + SDValue St0 = DAG.getStore(Chain, SDLoc(ST), Lo, Ptr, ST->getPointerInfo(), isVolatile, isNonTemporal, ST->getAlignment(), AAInfo); - Ptr = DAG.getNode(ISD::ADD, SDLoc(N), Ptr.getValueType(), Ptr, - DAG.getConstant(4, Ptr.getValueType())); + Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr, + DAG.getConstant(4, DL, Ptr.getValueType())); Alignment = MinAlign(Alignment, 4U); SDValue St1 = DAG.getStore(Chain, SDLoc(ST), Hi, Ptr, ST->getPointerInfo().getWithOffset(4), isVolatile, isNonTemporal, Alignment, AAInfo); - return DAG.getNode(ISD::TokenFactor, SDLoc(N), MVT::Other, + return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, St0, St1); } @@ -10268,11 +11241,15 @@ SDValue DAGCombiner::visitSTORE(SDNode *N) { // 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); + } } } @@ -10493,24 +11470,20 @@ SDValue DAGCombiner::ReplaceExtractVectorEltOfLoadWithNarrowedLoad( SDValue Offset; EVT PtrType = NewPtr.getValueType(); MachinePointerInfo MPI; + SDLoc DL(EVE); if (auto *ConstEltNo = dyn_cast<ConstantSDNode>(EltNo)) { int Elt = ConstEltNo->getZExtValue(); unsigned PtrOff = VecEltVT.getSizeInBits() * Elt / 8; - if (TLI.isBigEndian()) - PtrOff = InVecVT.getSizeInBits() / 8 - PtrOff; - Offset = DAG.getConstant(PtrOff, PtrType); + Offset = DAG.getConstant(PtrOff, DL, PtrType); MPI = OriginalLoad->getPointerInfo().getWithOffset(PtrOff); } else { + Offset = DAG.getZExtOrTrunc(EltNo, DL, PtrType); Offset = DAG.getNode( - ISD::MUL, SDLoc(EVE), EltNo.getValueType(), EltNo, - DAG.getConstant(VecEltVT.getStoreSize(), EltNo.getValueType())); - if (TLI.isBigEndian()) - Offset = DAG.getNode( - ISD::SUB, SDLoc(EVE), EltNo.getValueType(), - DAG.getConstant(InVecVT.getStoreSize(), EltNo.getValueType()), Offset); + ISD::MUL, DL, PtrType, Offset, + DAG.getConstant(VecEltVT.getStoreSize(), DL, PtrType)); MPI = OriginalLoad->getPointerInfo(); } - NewPtr = DAG.getNode(ISD::ADD, SDLoc(EVE), PtrType, NewPtr, Offset); + NewPtr = DAG.getNode(ISD::ADD, DL, PtrType, NewPtr, Offset); // The replacement we need to do here is a little tricky: we need to // replace an extractelement of a load with a load. @@ -10620,8 +11593,8 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) { if (!LegalOperations) { EVT IndexTy = TLI.getVectorIdxTy(); - return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N), NVT, - SVInVec, DAG.getConstant(OrigElt, IndexTy)); + return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N), NVT, SVInVec, + DAG.getConstant(OrigElt, SDLoc(SVOp), IndexTy)); } } @@ -10710,7 +11683,7 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) { if (ISD::isNormalLoad(InVec.getNode())) { LN0 = cast<LoadSDNode>(InVec); Elt = (Idx < (int)NumElems) ? Idx : Idx - (int)NumElems; - EltNo = DAG.getConstant(Elt, EltNo.getValueType()); + EltNo = DAG.getConstant(Elt, SDLoc(EltNo), EltNo.getValueType()); } } @@ -10800,7 +11773,7 @@ SDValue DAGCombiner::reduceBuildVecExtToExtBuildVec(SDNode *N) { unsigned ElemRatio = OutScalarTy.getSizeInBits()/SourceType.getSizeInBits(); assert(ElemRatio > 1 && "Invalid element size ratio"); SDValue Filler = AllAnyExt ? DAG.getUNDEF(SourceType): - DAG.getConstant(0, SourceType); + DAG.getConstant(0, SDLoc(N), SourceType); unsigned NewBVElems = ElemRatio * VT.getVectorNumElements(); SmallVector<SDValue, 8> Ops(NewBVElems, Filler); @@ -10890,6 +11863,11 @@ SDValue DAGCombiner::reduceBuildVecConvertToConvertBuildVec(SDNode *N) { if (!TLI.isOperationLegalOrCustom(Opcode, NVT)) return SDValue(); + // 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(); + SmallVector<SDValue, 8> Opnds; for (unsigned i = 0; i != NumInScalars; ++i) { SDValue In = N->getOperand(i); @@ -10914,12 +11892,10 @@ SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) { 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 @@ -10942,10 +11918,7 @@ SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) { if (Op.getOpcode() == ISD::UNDEF) continue; // See if we can combine this build_vector into a blend with a zero vector. - if (!VecIn2.getNode() && ((Op.getOpcode() == ISD::Constant && - cast<ConstantSDNode>(Op.getNode())->isNullValue()) || - (Op.getOpcode() == ISD::ConstantFP && - cast<ConstantFPSDNode>(Op.getNode())->getValueAPF().isZero()))) { + if (!VecIn2.getNode() && (isNullConstant(Op) || isNullFPConstant(Op))) { UsesZeroVector = true; continue; } @@ -11047,20 +12020,20 @@ SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) { 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, - DAG.getConstant(VT.getVectorNumElements(), TLI.getVectorIdxTy())); + VecIn2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1, + DAG.getConstant(VT.getVectorNumElements(), dl, TLI.getVectorIdxTy())); VecIn1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1, - DAG.getConstant(0, TLI.getVectorIdxTy())); + DAG.getConstant(0, dl, TLI.getVectorIdxTy())); } else return SDValue(); } if (UsesZeroVector) - VecIn2 = VT.isInteger() ? DAG.getConstant(0, VT) : - DAG.getConstantFP(0.0, VT); + VecIn2 = VT.isInteger() ? DAG.getConstant(0, dl, VT) : + DAG.getConstantFP(0.0, dl, VT); else // If VecIn2 is unused then change it to undef. VecIn2 = VecIn2.getNode() ? VecIn2 : DAG.getUNDEF(VT); @@ -11081,6 +12054,68 @@ SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) { 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 @@ -11096,9 +12131,10 @@ SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) { 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"); @@ -11106,6 +12142,15 @@ SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) { 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()) @@ -11122,39 +12167,61 @@ SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) { } } + // 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 @@ -11216,7 +12283,7 @@ SDValue DAGCombiner::visitEXTRACT_SUBVECTOR(SDNode* N) { // 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."); @@ -11347,7 +12414,8 @@ static SDValue simplifyShuffleOperands(ShuffleVectorSDNode *SVN, SDValue N0, return DAG.getVectorShuffle(VT, SDLoc(SVN), S0, S1, SVN->getMask()); } -// Tries to turn a shuffle of two CONCAT_VECTORS into a single concat. +// Tries to turn a shuffle of two CONCAT_VECTORS into a single concat, +// or turn a shuffle of a single concat into simpler shuffle then concat. static SDValue partitionShuffleOfConcats(SDNode *N, SelectionDAG &DAG) { EVT VT = N->getValueType(0); unsigned NumElts = VT.getVectorNumElements(); @@ -11361,6 +12429,18 @@ static SDValue partitionShuffleOfConcats(SDNode *N, SelectionDAG &DAG) { unsigned NumElemsPerConcat = ConcatVT.getVectorNumElements(); unsigned NumConcats = NumElts / NumElemsPerConcat; + // Special case: shuffle(concat(A,B)) can be more efficiently represented + // as concat(shuffle(A,B),UNDEF) if the shuffle doesn't set any of the high + // half vector elements. + if (NumElemsPerConcat * 2 == NumElts && N1.getOpcode() == ISD::UNDEF && + std::all_of(SVN->getMask().begin() + NumElemsPerConcat, + SVN->getMask().end(), [](int i) { return i == -1; })) { + N0 = DAG.getVectorShuffle(ConcatVT, SDLoc(N), N0.getOperand(0), N0.getOperand(1), + ArrayRef<int>(SVN->getMask().begin(), NumElemsPerConcat)); + N1 = DAG.getUNDEF(ConcatVT); + return DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(N), VT, N0, N1); + } + // Look at every vector that's inserted. We're looking for exact // subvector-sized copies from a concatenated vector for (unsigned I = 0; I != NumConcats; ++I) { @@ -11459,7 +12539,7 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { } // If it is a splat, check if the argument vector is another splat or a - // build_vector with all scalar elements the same. + // build_vector. if (SVN->isSplat() && SVN->getSplatIndex() < (int)NumElts) { SDNode *V = N0.getNode(); @@ -11496,6 +12576,18 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { // Splat of <x, x, x, x>, return <x, x, x, x> if (AllSame) return N0; + + // Canonicalize any other splat as a build_vector. + const SDValue &Splatted = V->getOperand(SVN->getSplatIndex()); + SmallVector<SDValue, 8> Ops(NumElts, Splatted); + SDValue NewBV = DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), + V->getValueType(0), Ops); + + // 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); + return NewBV; } } @@ -11516,6 +12608,118 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { 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) @@ -11543,8 +12747,9 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { // shuffle(shuffle(A, B, M0), C, M1) -> shuffle(A, C, M2) // shuffle(shuffle(A, B, M0), C, M1) -> shuffle(B, C, M2) // Don't try to fold shuffles with illegal type. - if (N0.getOpcode() == ISD::VECTOR_SHUFFLE && Level < AfterLegalizeDAG && - TLI.isTypeLegal(VT)) { + // Only fold if this shuffle is the only user of the other shuffle. + if (N0.getOpcode() == ISD::VECTOR_SHUFFLE && N->isOnlyUserOf(N0.getNode()) && + Level < AfterLegalizeDAG && TLI.isTypeLegal(VT)) { ShuffleVectorSDNode *OtherSV = cast<ShuffleVectorSDNode>(N0); // The incoming shuffle must be of the same type as the result of the @@ -11624,20 +12829,11 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { // 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) @@ -11653,6 +12849,34 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) { 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); @@ -11680,48 +12904,64 @@ SDValue DAGCombiner::visitINSERT_SUBVECTOR(SDNode *N) { 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(); - // Let's see if the target supports this vector_shuffle. - EVT RVT = RHS.getValueType(); - if (!TLI.isVectorClearMaskLegal(Indices, RVT)) - return SDValue(); + if (N->getOpcode() != ISD::AND) + 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 (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 (isAllOnesConstant(Elt)) + Indices.push_back(i); + else if (isNullConstant(Elt)) + 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, dl, EltVT)); + SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, dl, 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(); @@ -11734,8 +12974,9 @@ SDValue DAGCombiner::SimplifyVBinOp(SDNode *N) { 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. @@ -11754,10 +12995,8 @@ SDValue DAGCombiner::SimplifyVBinOp(SDNode *N) { // Can't fold divide by zero. if (N->getOpcode() == ISD::SDIV || N->getOpcode() == ISD::UDIV || N->getOpcode() == ISD::FDIV) { - if ((RHSOp.getOpcode() == ISD::Constant && - cast<ConstantSDNode>(RHSOp.getNode())->isNullValue()) || - (RHSOp.getOpcode() == ISD::ConstantFP && - cast<ConstantFPSDNode>(RHSOp.getNode())->getValueAPF().isZero())) + if (isNullConstant(RHSOp) || (RHSOp.getOpcode() == ISD::ConstantFP && + cast<ConstantFPSDNode>(RHSOp.getNode())->isZero())) break; } @@ -11813,38 +13052,6 @@ SDValue DAGCombiner::SimplifyVBinOp(SDNode *N) { 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!"); @@ -11881,6 +13088,38 @@ SDValue DAGCombiner::SimplifySelect(SDLoc DL, SDValue N0, 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; @@ -11902,6 +13141,9 @@ bool DAGCombiner::SimplifySelectOps(SDNode *TheSelect, SDValue LHS, 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. @@ -12003,25 +13245,22 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, EVT VT = N2.getValueType(); ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode()); ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.getNode()); - ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N3.getNode()); // Determine if the condition we're dealing with is constant SDValue SCC = SimplifySetCC(getSetCCResultType(N0.getValueType()), N0, N1, CC, DL, false); if (SCC.getNode()) AddToWorklist(SCC.getNode()); - ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.getNode()); - // fold select_cc true, x, y -> x - if (SCCC && !SCCC->isNullValue()) - return N2; - // fold select_cc false, x, y -> y - if (SCCC && SCCC->isNullValue()) - return N3; + if (ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.getNode())) { + // fold select_cc true, x, y -> x + // fold select_cc false, x, y -> y + return !SCCC->isNullValue() ? N2 : N3; + } // Check to see if we can simplify the select into an fabs node if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N1)) { // Allow either -0.0 or 0.0 - if (CFP->getValueAPF().isZero()) { + if (CFP->isZero()) { // select (setg[te] X, +/-0.0), X, fneg(X) -> fabs if ((CC == ISD::SETGE || CC == ISD::SETGT) && N0 == N2 && N3.getOpcode() == ISD::FNEG && @@ -12069,9 +13308,9 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, // Get the offsets to the 0 and 1 element of the array so that we can // select between them. - SDValue Zero = DAG.getIntPtrConstant(0); + SDValue Zero = DAG.getIntPtrConstant(0, DL); unsigned EltSize = (unsigned)TD.getTypeAllocSize(Elts[0]->getType()); - SDValue One = DAG.getIntPtrConstant(EltSize); + SDValue One = DAG.getIntPtrConstant(EltSize, SDLoc(FV)); SDValue Cond = DAG.getSetCC(DL, getSetCCResultType(N0.getValueType()), @@ -12086,24 +13325,23 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, return DAG.getLoad(TV->getValueType(0), DL, DAG.getEntryNode(), CPIdx, MachinePointerInfo::getConstantPool(), false, false, false, Alignment); - } } // Check to see if we can perform the "gzip trick", transforming // (select_cc setlt X, 0, A, 0) -> (and (sra X, (sub size(X), 1), A) - if (N1C && N3C && N3C->isNullValue() && CC == ISD::SETLT && - (N1C->isNullValue() || // (a < 0) ? b : 0 - (N1C->getAPIntValue() == 1 && N0 == N2))) { // (a < 1) ? a : 0 + if (isNullConstant(N3) && CC == ISD::SETLT && + (isNullConstant(N1) || // (a < 0) ? b : 0 + (isOneConstant(N1) && N0 == N2))) { // (a < 1) ? a : 0 EVT XType = N0.getValueType(); EVT AType = N2.getValueType(); if (XType.bitsGE(AType)) { // and (sra X, size(X)-1, A) -> "and (srl X, C2), A" iff A is a // single-bit constant. - if (N2C && ((N2C->getAPIntValue() & (N2C->getAPIntValue()-1)) == 0)) { + if (N2C && ((N2C->getAPIntValue() & (N2C->getAPIntValue() - 1)) == 0)) { unsigned ShCtV = N2C->getAPIntValue().logBase2(); - ShCtV = XType.getSizeInBits()-ShCtV-1; - SDValue ShCt = DAG.getConstant(ShCtV, + ShCtV = XType.getSizeInBits() - ShCtV - 1; + SDValue ShCt = DAG.getConstant(ShCtV, SDLoc(N0), getShiftAmountTy(N0.getValueType())); SDValue Shift = DAG.getNode(ISD::SRL, SDLoc(N0), XType, N0, ShCt); @@ -12119,7 +13357,8 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, SDValue Shift = DAG.getNode(ISD::SRA, SDLoc(N0), XType, N0, - DAG.getConstant(XType.getSizeInBits()-1, + DAG.getConstant(XType.getSizeInBits() - 1, + SDLoc(N0), getShiftAmountTy(N0.getValueType()))); AddToWorklist(Shift.getNode()); @@ -12139,23 +13378,21 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, // single bit-test can be materialized as an all-ones register with // shift-left and shift-right-arith. if (CC == ISD::SETEQ && N0->getOpcode() == ISD::AND && - N0->getValueType(0) == VT && - N1C && N1C->isNullValue() && - N2C && N2C->isNullValue()) { + N0->getValueType(0) == VT && isNullConstant(N1) && isNullConstant(N2)) { SDValue AndLHS = N0->getOperand(0); ConstantSDNode *ConstAndRHS = dyn_cast<ConstantSDNode>(N0->getOperand(1)); if (ConstAndRHS && ConstAndRHS->getAPIntValue().countPopulation() == 1) { // Shift the tested bit over the sign bit. APInt AndMask = ConstAndRHS->getAPIntValue(); SDValue ShlAmt = - DAG.getConstant(AndMask.countLeadingZeros(), + DAG.getConstant(AndMask.countLeadingZeros(), SDLoc(AndLHS), getShiftAmountTy(AndLHS.getValueType())); SDValue Shl = DAG.getNode(ISD::SHL, SDLoc(N0), VT, AndLHS, ShlAmt); // Now arithmetic right shift it all the way over, so the result is either // all-ones, or zero. SDValue ShrAmt = - DAG.getConstant(AndMask.getBitWidth()-1, + DAG.getConstant(AndMask.getBitWidth() - 1, SDLoc(Shl), getShiftAmountTy(Shl.getValueType())); SDValue Shr = DAG.getNode(ISD::SRA, SDLoc(N0), VT, Shl, ShrAmt); @@ -12164,13 +13401,13 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, } // fold select C, 16, 0 -> shl C, 4 - if (N2C && N3C && N3C->isNullValue() && N2C->getAPIntValue().isPowerOf2() && + if (N2C && isNullConstant(N3) && N2C->getAPIntValue().isPowerOf2() && TLI.getBooleanContents(N0.getValueType()) == TargetLowering::ZeroOrOneBooleanContent) { // If the caller doesn't want us to simplify this into a zext of a compare, // don't do it. - if (NotExtCompare && N2C->getAPIntValue() == 1) + if (NotExtCompare && N2C->isOne()) return SDValue(); // Get a SetCC of the condition @@ -12198,13 +13435,13 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, AddToWorklist(SCC.getNode()); AddToWorklist(Temp.getNode()); - if (N2C->getAPIntValue() == 1) + if (N2C->isOne()) return Temp; // shl setcc result by log2 n2c return DAG.getNode( ISD::SHL, DL, N2.getValueType(), Temp, - DAG.getConstant(N2C->getAPIntValue().logBase2(), + DAG.getConstant(N2C->getAPIntValue().logBase2(), SDLoc(Temp), getShiftAmountTy(Temp.getValueType()))); } } @@ -12212,7 +13449,7 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, // Check to see if this is the equivalent of setcc // FIXME: Turn all of these into setcc if setcc if setcc is legal // otherwise, go ahead with the folds. - if (0 && N3C && N3C->isNullValue() && N2C && (N2C->getAPIntValue() == 1ULL)) { + if (0 && isNullConstant(N3) && isOneConstant(N2)) { EVT XType = N0.getValueType(); if (!LegalOperations || TLI.isOperationLegal(ISD::SETCC, getSetCCResultType(XType))) { @@ -12223,30 +13460,34 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, } // fold (seteq X, 0) -> (srl (ctlz X, log2(size(X)))) - if (N1C && N1C->isNullValue() && CC == ISD::SETEQ && + if (isNullConstant(N1) && CC == ISD::SETEQ && (!LegalOperations || TLI.isOperationLegal(ISD::CTLZ, XType))) { SDValue Ctlz = DAG.getNode(ISD::CTLZ, SDLoc(N0), XType, N0); return DAG.getNode(ISD::SRL, DL, XType, Ctlz, DAG.getConstant(Log2_32(XType.getSizeInBits()), + SDLoc(Ctlz), getShiftAmountTy(Ctlz.getValueType()))); } // fold (setgt X, 0) -> (srl (and (-X, ~X), size(X)-1)) - if (N1C && N1C->isNullValue() && CC == ISD::SETGT) { - SDValue NegN0 = DAG.getNode(ISD::SUB, SDLoc(N0), - XType, DAG.getConstant(0, XType), N0); - SDValue NotN0 = DAG.getNOT(SDLoc(N0), N0, XType); + if (isNullConstant(N1) && CC == ISD::SETGT) { + SDLoc DL(N0); + SDValue NegN0 = DAG.getNode(ISD::SUB, DL, + XType, DAG.getConstant(0, DL, XType), N0); + SDValue NotN0 = DAG.getNOT(DL, N0, XType); return DAG.getNode(ISD::SRL, DL, XType, DAG.getNode(ISD::AND, DL, XType, NegN0, NotN0), - DAG.getConstant(XType.getSizeInBits()-1, + DAG.getConstant(XType.getSizeInBits() - 1, DL, getShiftAmountTy(XType))); } // fold (setgt X, -1) -> (xor (srl (X, size(X)-1), 1)) - if (N1C && N1C->isAllOnesValue() && CC == ISD::SETGT) { - SDValue Sign = DAG.getNode(ISD::SRL, SDLoc(N0), XType, N0, - DAG.getConstant(XType.getSizeInBits()-1, + if (isAllOnesConstant(N1) && CC == ISD::SETGT) { + SDLoc DL(N0); + SDValue Sign = DAG.getNode(ISD::SRL, DL, XType, N0, + DAG.getConstant(XType.getSizeInBits() - 1, DL, getShiftAmountTy(N0.getValueType()))); - return DAG.getNode(ISD::XOR, DL, XType, Sign, DAG.getConstant(1, XType)); + return DAG.getNode(ISD::XOR, DL, XType, Sign, DAG.getConstant(1, DL, + XType)); } } @@ -12269,11 +13510,12 @@ SDValue DAGCombiner::SimplifySelectCC(SDLoc DL, SDValue N0, SDValue N1, EVT XType = N0.getValueType(); if (SubC && SubC->isNullValue() && XType.isInteger()) { - SDValue Shift = DAG.getNode(ISD::SRA, SDLoc(N0), XType, + SDLoc DL(N0); + SDValue Shift = DAG.getNode(ISD::SRA, DL, XType, N0, - DAG.getConstant(XType.getSizeInBits()-1, + DAG.getConstant(XType.getSizeInBits() - 1, DL, getShiftAmountTy(N0.getValueType()))); - SDValue Add = DAG.getNode(ISD::ADD, SDLoc(N0), + SDValue Add = DAG.getNode(ISD::ADD, DL, XType, N0, Shift); AddToWorklist(Shift.getNode()); AddToWorklist(Add.getNode()); @@ -12303,7 +13545,7 @@ SDValue DAGCombiner::BuildSDIV(SDNode *N) { return SDValue(); // Avoid division by zero. - if (!C->getAPIntValue()) + if (C->isNullValue()) return SDValue(); std::vector<SDNode*> Built; @@ -12323,7 +13565,7 @@ SDValue DAGCombiner::BuildSDIVPow2(SDNode *N) { return SDValue(); // Avoid division by zero. - if (!C->getAPIntValue()) + if (C->isNullValue()) return SDValue(); std::vector<SDNode *> Built; @@ -12344,7 +13586,7 @@ SDValue DAGCombiner::BuildUDIV(SDNode *N) { return SDValue(); // Avoid division by zero. - if (!C->getAPIntValue()) + if (C->isNullValue()) return SDValue(); std::vector<SDNode*> Built; @@ -12374,7 +13616,7 @@ SDValue DAGCombiner::BuildReciprocalEstimate(SDValue Op) { // does not require additional intermediate precision] EVT VT = Op.getValueType(); SDLoc DL(Op); - SDValue FPOne = DAG.getConstantFP(1.0, VT); + SDValue FPOne = DAG.getConstantFP(1.0, DL, VT); AddToWorklist(Est.getNode()); @@ -12409,7 +13651,7 @@ SDValue DAGCombiner::BuildRsqrtNROneConst(SDValue Arg, SDValue Est, unsigned Iterations) { EVT VT = Arg.getValueType(); SDLoc DL(Arg); - SDValue ThreeHalves = DAG.getConstantFP(1.5, VT); + SDValue ThreeHalves = DAG.getConstantFP(1.5, DL, VT); // We now need 0.5 * Arg which we can write as (1.5 * Arg - Arg) so that // this entire sequence requires only one FP constant. @@ -12445,8 +13687,8 @@ SDValue DAGCombiner::BuildRsqrtNRTwoConst(SDValue Arg, SDValue Est, unsigned Iterations) { EVT VT = Arg.getValueType(); SDLoc DL(Arg); - SDValue MinusThree = DAG.getConstantFP(-3.0, VT); - SDValue MinusHalf = DAG.getConstantFP(-0.5, VT); + SDValue MinusThree = DAG.getConstantFP(-3.0, DL, VT); + SDValue MinusHalf = DAG.getConstantFP(-0.5, DL, VT); // Newton iterations: Est = -0.5 * Est * (-3.0 + Arg * Est * Est) for (unsigned i = 0; i < Iterations; ++i) { |
