X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FLegalizeDAG.cpp;h=7d98872f8af1f6fdd031e47cfb04c414e5a6042c;hb=d72b35710738a3655a3cea808e71659d8ec9d82b;hp=5d17a5f643e307789a66348a50e8dc08a85eb478;hpb=5401ba7099b9f3cd32b3399276b549bea15e5d1e;p=oota-llvm.git diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp index 5d17a5f643e..7d98872f8af 100644 --- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp +++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp @@ -40,7 +40,7 @@ using namespace llvm; #define DEBUG_TYPE "legalizedag" //===----------------------------------------------------------------------===// -/// SelectionDAGLegalize - This takes an arbitrary SelectionDAG as input and +/// This takes an arbitrary SelectionDAG as input and /// hacks on it until the target machine can handle it. This involves /// eliminating value sizes the machine cannot handle (promoting small sizes to /// large sizes or splitting up large values into small values) as well as @@ -86,7 +86,7 @@ private: void LegalizeLoadOps(SDNode *Node); void LegalizeStoreOps(SDNode *Node); - /// PerformInsertVectorEltInMemory - Some target cannot handle a variable + /// Some targets cannot handle a variable /// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it /// is necessary to spill the vector being inserted into to memory, perform /// the insert there, and then read the result back. @@ -95,7 +95,7 @@ private: SDValue ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val, SDValue Idx, SDLoc dl); - /// ShuffleWithNarrowerEltType - Return a vector shuffle operation which + /// Return a vector shuffle operation which /// performs the same shuffe in terms of order or result bytes, but on a type /// whose vector element type is narrower than the original shuffle type. /// e.g. <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3> @@ -200,7 +200,7 @@ public: }; } -/// ShuffleWithNarrowerEltType - Return a vector shuffle operation which +/// Return a vector shuffle operation which /// performs the same shuffe in terms of order or result bytes, but on a type /// whose vector element type is narrower than the original shuffle type. /// e.g. <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3> @@ -232,7 +232,7 @@ SelectionDAGLegalize::ShuffleWithNarrowerEltType(EVT NVT, EVT VT, SDLoc dl, return DAG.getVectorShuffle(NVT, dl, N1, N2, &NewMask[0]); } -/// ExpandConstantFP - Expands the ConstantFP node to an integer constant or +/// Expands the ConstantFP node to an integer constant or /// a load from the constant pool. SDValue SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) { @@ -249,7 +249,7 @@ SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) { ConstantFP *LLVMC = const_cast(CFP->getConstantFPValue()); if (!UseCP) { assert((VT == MVT::f64 || VT == MVT::f32) && "Invalid type expansion"); - return DAG.getConstant(LLVMC->getValueAPF().bitcastToAPInt(), + return DAG.getConstant(LLVMC->getValueAPF().bitcastToAPInt(), dl, (VT == MVT::f64) ? MVT::i64 : MVT::i32); } @@ -260,7 +260,7 @@ SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) { if (ConstantFPSDNode::isValueValidForType(SVT, CFP->getValueAPF()) && // Only do this if the target has a native EXTLOAD instruction from // smaller type. - TLI.isLoadExtLegal(ISD::EXTLOAD, SVT) && + TLI.isLoadExtLegal(ISD::EXTLOAD, OrigVT, SVT) && TLI.ShouldShrinkFPConstant(OrigVT)) { Type *SType = SVT.getTypeForEVT(*DAG.getContext()); LLVMC = cast(ConstantExpr::getFPTrunc(LLVMC, SType)); @@ -286,7 +286,7 @@ SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) { return Result; } -/// ExpandUnalignedStore - Expands an unaligned store to 2 half-size stores. +/// Expands an unaligned store to 2 half-size stores. static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG, const TargetLowering &TLI, SelectionDAGLegalize *DAGLegalize) { @@ -331,7 +331,7 @@ static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG, SDValue Store = DAG.getTruncStore(Chain, dl, Val, StackPtr, MachinePointerInfo(), StoredVT, false, false, 0); - SDValue Increment = DAG.getConstant(RegBytes, TLI.getPointerTy(AS)); + SDValue Increment = DAG.getConstant(RegBytes, dl, TLI.getPointerTy(AS)); SmallVector Stores; unsigned Offset = 0; @@ -385,7 +385,7 @@ static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG, int IncrementSize = NumBits / 8; // Divide the stored value in two parts. - SDValue ShiftAmount = DAG.getConstant(NumBits, + SDValue ShiftAmount = DAG.getConstant(NumBits, dl, TLI.getShiftAmountTy(Val.getValueType())); SDValue Lo = Val; SDValue Hi = DAG.getNode(ISD::SRL, dl, VT, Val, ShiftAmount); @@ -397,7 +397,7 @@ static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG, ST->isVolatile(), ST->isNonTemporal(), Alignment); Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, TLI.getPointerTy(AS))); + DAG.getConstant(IncrementSize, dl, TLI.getPointerTy(AS))); Alignment = MinAlign(Alignment, IncrementSize); Store2 = DAG.getTruncStore(Chain, dl, TLI.isLittleEndian()?Hi:Lo, Ptr, ST->getPointerInfo().getWithOffset(IncrementSize), @@ -409,7 +409,7 @@ static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG, DAGLegalize->ReplaceNode(SDValue(ST, 0), Result); } -/// ExpandUnalignedLoad - Expands an unaligned load to 2 half-size loads. +/// Expands an unaligned load to 2 half-size loads. static void ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, const TargetLowering &TLI, @@ -448,7 +448,7 @@ ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, // Make sure the stack slot is also aligned for the register type. SDValue StackBase = DAG.CreateStackTemporary(LoadedVT, RegVT); - SDValue Increment = DAG.getConstant(RegBytes, TLI.getPointerTy()); + SDValue Increment = DAG.getConstant(RegBytes, dl, TLI.getPointerTy()); SmallVector Stores; SDValue StackPtr = StackBase; unsigned Offset = 0; @@ -528,7 +528,7 @@ ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, LD->isNonTemporal(), LD->isInvariant(), Alignment, LD->getAAInfo()); Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, Ptr.getValueType())); Hi = DAG.getExtLoad(HiExtType, dl, VT, Chain, Ptr, LD->getPointerInfo().getWithOffset(IncrementSize), NewLoadedVT, LD->isVolatile(), @@ -540,7 +540,7 @@ ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, LD->isNonTemporal(), LD->isInvariant(), Alignment, LD->getAAInfo()); Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, Ptr.getValueType())); Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, VT, Chain, Ptr, LD->getPointerInfo().getWithOffset(IncrementSize), NewLoadedVT, LD->isVolatile(), @@ -549,7 +549,7 @@ ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, } // aggregate the two parts - SDValue ShiftAmount = DAG.getConstant(NumBits, + SDValue ShiftAmount = DAG.getConstant(NumBits, dl, TLI.getShiftAmountTy(Hi.getValueType())); SDValue Result = DAG.getNode(ISD::SHL, dl, VT, Hi, ShiftAmount); Result = DAG.getNode(ISD::OR, dl, VT, Result, Lo); @@ -561,8 +561,8 @@ ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG, ChainResult = TF; } -/// PerformInsertVectorEltInMemory - Some target cannot handle a variable -/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it +/// Some target cannot handle a variable insertion index for the +/// INSERT_VECTOR_ELT instruction. In this case, it /// is necessary to spill the vector being inserted into to memory, perform /// the insert there, and then read the result back. SDValue SelectionDAGLegalize:: @@ -596,7 +596,8 @@ PerformInsertVectorEltInMemory(SDValue Vec, SDValue Val, SDValue Idx, Tmp3 = DAG.getNode(CastOpc, dl, PtrVT, Tmp3); // Add the offset to the index. unsigned EltSize = EltVT.getSizeInBits()/8; - Tmp3 = DAG.getNode(ISD::MUL, dl, IdxVT, Tmp3,DAG.getConstant(EltSize, IdxVT)); + Tmp3 = DAG.getNode(ISD::MUL, dl, IdxVT, Tmp3, + DAG.getConstant(EltSize, dl, IdxVT)); SDValue StackPtr2 = DAG.getNode(ISD::ADD, dl, IdxVT, Tmp3, StackPtr); // Store the scalar value. Ch = DAG.getTruncStore(Ch, dl, Tmp2, StackPtr2, MachinePointerInfo(), EltVT, @@ -655,7 +656,7 @@ SDValue SelectionDAGLegalize::OptimizeFloatStore(StoreSDNode* ST) { TLI.isTypeLegal(MVT::i32)) { SDValue Con = DAG.getConstant(CFP->getValueAPF(). bitcastToAPInt().zextOrTrunc(32), - MVT::i32); + SDLoc(CFP), MVT::i32); return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(), isVolatile, isNonTemporal, Alignment, AAInfo); } @@ -664,7 +665,7 @@ SDValue SelectionDAGLegalize::OptimizeFloatStore(StoreSDNode* ST) { // If this target supports 64-bit registers, do a single 64-bit store. if (TLI.isTypeLegal(MVT::i64)) { SDValue Con = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt(). - zextOrTrunc(64), MVT::i64); + zextOrTrunc(64), SDLoc(CFP), MVT::i64); return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(), isVolatile, isNonTemporal, Alignment, AAInfo); } @@ -673,15 +674,15 @@ SDValue SelectionDAGLegalize::OptimizeFloatStore(StoreSDNode* ST) { // Otherwise, if the target supports 32-bit registers, use 2 32-bit // stores. If the target supports neither 32- nor 64-bits, this // xform is certainly not worth it. - const APInt &IntVal =CFP->getValueAPF().bitcastToAPInt(); - SDValue Lo = DAG.getConstant(IntVal.trunc(32), MVT::i32); - SDValue Hi = DAG.getConstant(IntVal.lshr(32).trunc(32), MVT::i32); + const APInt &IntVal = CFP->getValueAPF().bitcastToAPInt(); + SDValue Lo = DAG.getConstant(IntVal.trunc(32), dl, MVT::i32); + SDValue Hi = DAG.getConstant(IntVal.lshr(32).trunc(32), dl, MVT::i32); if (TLI.isBigEndian()) std::swap(Lo, Hi); Lo = DAG.getStore(Chain, dl, Lo, Ptr, ST->getPointerInfo(), isVolatile, isNonTemporal, Alignment, AAInfo); Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(4, Ptr.getValueType())); + DAG.getConstant(4, dl, Ptr.getValueType())); Hi = DAG.getStore(Chain, dl, Hi, Ptr, ST->getPointerInfo().getWithOffset(4), isVolatile, isNonTemporal, MinAlign(Alignment, 4U), @@ -725,14 +726,13 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext()); unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty); if (Align < ABIAlignment) - ExpandUnalignedStore(cast(Node), - DAG, TLI, this); + ExpandUnalignedStore(cast(Node), DAG, TLI, this); } break; } case TargetLowering::Custom: { SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG); - if (Res.getNode()) + if (Res && Res != SDValue(Node, 0)) ReplaceNode(SDValue(Node, 0), Res); return; } @@ -766,8 +766,7 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { Value = DAG.getZeroExtendInReg(Value, dl, StVT); SDValue Result = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(), - NVT, isVolatile, isNonTemporal, Alignment, - AAInfo); + NVT, isVolatile, isNonTemporal, Alignment, AAInfo); ReplaceNode(SDValue(Node, 0), Result); } else if (StWidth & (StWidth - 1)) { // If not storing a power-of-2 number of bits, expand as two stores. @@ -794,9 +793,10 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { // Store the remaining ExtraWidth bits. IncrementSize = RoundWidth / 8; Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, + Ptr.getValueType())); Hi = DAG.getNode(ISD::SRL, dl, Value.getValueType(), Value, - DAG.getConstant(RoundWidth, + DAG.getConstant(RoundWidth, dl, TLI.getShiftAmountTy(Value.getValueType()))); Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr, ST->getPointerInfo().getWithOffset(IncrementSize), @@ -807,7 +807,7 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { // TRUNCSTORE:i24 X -> TRUNCSTORE:i16 (srl X, 8), TRUNCSTORE@+2:i8 X // Store the top RoundWidth bits. Hi = DAG.getNode(ISD::SRL, dl, Value.getValueType(), Value, - DAG.getConstant(ExtraWidth, + DAG.getConstant(ExtraWidth, dl, TLI.getShiftAmountTy(Value.getValueType()))); Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr, ST->getPointerInfo(), RoundVT, isVolatile, isNonTemporal, Alignment, @@ -816,7 +816,8 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { // Store the remaining ExtraWidth bits. IncrementSize = RoundWidth / 8; Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, + Ptr.getValueType())); Lo = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo().getWithOffset(IncrementSize), ExtraVT, isVolatile, isNonTemporal, @@ -845,7 +846,7 @@ void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) { } case TargetLowering::Custom: { SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG); - if (Res.getNode()) + if (Res && Res != SDValue(Node, 0)) ReplaceNode(SDValue(Node, 0), Res); return; } @@ -946,7 +947,8 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { // nice to have an effective generic way of getting these benefits... // Until such a way is found, don't insist on promoting i1 here. (SrcVT != MVT::i1 || - TLI.getLoadExtAction(ExtType, MVT::i1) == TargetLowering::Promote)) { + TLI.getLoadExtAction(ExtType, Node->getValueType(0), MVT::i1) == + TargetLowering::Promote)) { // Promote to a byte-sized load if not loading an integral number of // bytes. For example, promote EXTLOAD:i20 -> EXTLOAD:i24. unsigned NewWidth = SrcVT.getStoreSizeInBits(); @@ -1005,7 +1007,8 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { // Load the remaining ExtraWidth bits. IncrementSize = RoundWidth / 8; Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, + Ptr.getValueType())); Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr, LD->getPointerInfo().getWithOffset(IncrementSize), ExtraVT, isVolatile, isNonTemporal, isInvariant, @@ -1018,7 +1021,7 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { // Move the top bits to the right place. Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi, - DAG.getConstant(RoundWidth, + DAG.getConstant(RoundWidth, dl, TLI.getShiftAmountTy(Hi.getValueType()))); // Join the hi and lo parts. @@ -1034,7 +1037,8 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { // Load the remaining ExtraWidth bits. IncrementSize = RoundWidth / 8; Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, - DAG.getConstant(IncrementSize, Ptr.getValueType())); + DAG.getConstant(IncrementSize, dl, + Ptr.getValueType())); Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0), Chain, Ptr, LD->getPointerInfo().getWithOffset(IncrementSize), @@ -1048,7 +1052,7 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { // Move the top bits to the right place. Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi, - DAG.getConstant(ExtraWidth, + DAG.getConstant(ExtraWidth, dl, TLI.getShiftAmountTy(Hi.getValueType()))); // Join the hi and lo parts. @@ -1058,7 +1062,8 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { Chain = Ch; } else { bool isCustom = false; - switch (TLI.getLoadExtAction(ExtType, SrcVT.getSimpleVT())) { + switch (TLI.getLoadExtAction(ExtType, Node->getValueType(0), + SrcVT.getSimpleVT())) { default: llvm_unreachable("This action is not supported yet!"); case TargetLowering::Custom: isCustom = true; @@ -1080,36 +1085,35 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { unsigned AS = LD->getAddressSpace(); unsigned Align = LD->getAlignment(); if (!TLI.allowsMisalignedMemoryAccesses(MemVT, AS, Align)) { - Type *Ty = - LD->getMemoryVT().getTypeForEVT(*DAG.getContext()); - unsigned ABIAlignment = - TLI.getDataLayout()->getABITypeAlignment(Ty); + Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext()); + unsigned ABIAlignment = TLI.getDataLayout()->getABITypeAlignment(Ty); if (Align < ABIAlignment){ - ExpandUnalignedLoad(cast(Node), - DAG, TLI, Value, Chain); + ExpandUnalignedLoad(cast(Node), DAG, TLI, Value, Chain); } } } break; } case TargetLowering::Expand: - if (!TLI.isLoadExtLegal(ISD::EXTLOAD, SrcVT) && - TLI.isTypeLegal(SrcVT)) { - SDValue Load = DAG.getLoad(SrcVT, dl, Chain, Ptr, - LD->getMemOperand()); - unsigned ExtendOp; - switch (ExtType) { - case ISD::EXTLOAD: - ExtendOp = (SrcVT.isFloatingPoint() ? - ISD::FP_EXTEND : ISD::ANY_EXTEND); + if (!TLI.isLoadExtLegal(ISD::EXTLOAD, Node->getValueType(0), SrcVT)) { + // If the source type is not legal, see if there is a legal extload to + // an intermediate type that we can then extend further. + EVT LoadVT = TLI.getRegisterType(SrcVT.getSimpleVT()); + if (TLI.isTypeLegal(SrcVT) || // Same as SrcVT == LoadVT? + TLI.isLoadExtLegal(ExtType, LoadVT, SrcVT)) { + // If we are loading a legal type, this is a non-extload followed by a + // full extend. + ISD::LoadExtType MidExtType = + (LoadVT == SrcVT) ? ISD::NON_EXTLOAD : ExtType; + + SDValue Load = DAG.getExtLoad(MidExtType, dl, LoadVT, Chain, Ptr, + SrcVT, LD->getMemOperand()); + unsigned ExtendOp = + ISD::getExtForLoadExtType(SrcVT.isFloatingPoint(), ExtType); + Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load); + Chain = Load.getValue(1); break; - case ISD::SEXTLOAD: ExtendOp = ISD::SIGN_EXTEND; break; - case ISD::ZEXTLOAD: ExtendOp = ISD::ZERO_EXTEND; break; - default: llvm_unreachable("Unexpected extend load type!"); } - Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load); - Chain = Load.getValue(1); - break; } assert(!SrcVT.isVector() && @@ -1133,8 +1137,7 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { Result.getValueType(), Result, DAG.getValueType(SrcVT)); else - ValRes = DAG.getZeroExtendInReg(Result, dl, - SrcVT.getScalarType()); + ValRes = DAG.getZeroExtendInReg(Result, dl, SrcVT.getScalarType()); Value = ValRes; Chain = Result.getValue(1); break; @@ -1155,8 +1158,7 @@ void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) { } } -/// LegalizeOp - Return a legal replacement for the given operation, with -/// all legal operands. +/// Return a legal replacement for the given operation, with all legal operands. void SelectionDAGLegalize::LegalizeOp(SDNode *Node) { DEBUG(dbgs() << "\nLegalizing: "; Node->dump(&DAG)); @@ -1243,12 +1245,6 @@ void SelectionDAGLegalize::LegalizeOp(SDNode *Node) { break; case ISD::EXTRACT_ELEMENT: case ISD::FLT_ROUNDS_: - case ISD::SADDO: - case ISD::SSUBO: - case ISD::UADDO: - case ISD::USUBO: - case ISD::SMULO: - case ISD::UMULO: case ISD::FPOWI: case ISD::MERGE_VALUES: case ISD::EH_RETURN: @@ -1440,18 +1436,32 @@ SDValue SelectionDAGLegalize::ExpandExtractFromVectorThroughStack(SDValue Op) { unsigned EltSize = Vec.getValueType().getVectorElementType().getSizeInBits()/8; Idx = DAG.getNode(ISD::MUL, dl, Idx.getValueType(), Idx, - DAG.getConstant(EltSize, Idx.getValueType())); + DAG.getConstant(EltSize, SDLoc(Vec), Idx.getValueType())); Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy()); StackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, StackPtr); + SDValue NewLoad; + if (Op.getValueType().isVector()) - return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr,MachinePointerInfo(), - false, false, false, 0); - return DAG.getExtLoad(ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr, - MachinePointerInfo(), - Vec.getValueType().getVectorElementType(), - false, false, false, 0); + NewLoad = DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, + MachinePointerInfo(), false, false, false, 0); + else + NewLoad = DAG.getExtLoad( + ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr, MachinePointerInfo(), + Vec.getValueType().getVectorElementType(), false, false, false, 0); + + // Replace the chain going out of the store, by the one out of the load. + DAG.ReplaceAllUsesOfValueWith(Ch, SDValue(NewLoad.getNode(), 1)); + + // We introduced a cycle though, so update the loads operands, making sure + // to use the original store's chain as an incoming chain. + SmallVector NewLoadOperands(NewLoad->op_begin(), + NewLoad->op_end()); + NewLoadOperands[0] = Ch; + NewLoad = + SDValue(DAG.UpdateNodeOperands(NewLoad.getNode(), NewLoadOperands), 0); + return NewLoad; } SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) { @@ -1479,7 +1489,7 @@ SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) { Vec.getValueType().getVectorElementType().getSizeInBits()/8; Idx = DAG.getNode(ISD::MUL, dl, Idx.getValueType(), Idx, - DAG.getConstant(EltSize, Idx.getValueType())); + DAG.getConstant(EltSize, SDLoc(Vec), Idx.getValueType())); Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy()); SDValue SubStackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, @@ -1516,7 +1526,7 @@ SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) { unsigned Offset = TypeByteSize*i; - SDValue Idx = DAG.getConstant(Offset, FIPtr.getValueType()); + SDValue Idx = DAG.getConstant(Offset, dl, FIPtr.getValueType()); Idx = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr, Idx); // If the destination vector element type is narrower than the source @@ -1578,7 +1588,8 @@ SDValue SelectionDAGLegalize::ExpandFCOPYSIGN(SDNode* Node) { unsigned Strides = (FloatVT.getSizeInBits()-1)/LoadTy.getSizeInBits(); unsigned ByteOffset = (Strides * LoadTy.getSizeInBits()) / 8; LoadPtr = DAG.getNode(ISD::ADD, dl, LoadPtr.getValueType(), LoadPtr, - DAG.getConstant(ByteOffset, LoadPtr.getValueType())); + DAG.getConstant(ByteOffset, dl, + LoadPtr.getValueType())); // Load a legal integer containing the sign bit. SignBit = DAG.getLoad(LoadTy, dl, Ch, LoadPtr, MachinePointerInfo(), false, false, false, 0); @@ -1588,13 +1599,14 @@ SDValue SelectionDAGLegalize::ExpandFCOPYSIGN(SDNode* Node) { assert(BitShift < LoadTy.getSizeInBits() && "Pointer advanced wrong?"); if (BitShift) SignBit = DAG.getNode(ISD::SHL, dl, LoadTy, SignBit, - DAG.getConstant(BitShift, + DAG.getConstant(BitShift, dl, TLI.getShiftAmountTy(SignBit.getValueType()))); } } // Now get the sign bit proper, by seeing whether the value is negative. SignBit = DAG.getSetCC(dl, getSetCCResultType(SignBit.getValueType()), - SignBit, DAG.getConstant(0, SignBit.getValueType()), + SignBit, + DAG.getConstant(0, dl, SignBit.getValueType()), ISD::SETLT); // Get the absolute value of the result. SDValue AbsVal = DAG.getNode(ISD::FABS, dl, Tmp1.getValueType(), Tmp1); @@ -1619,8 +1631,7 @@ void SelectionDAGLegalize::ExpandDYNAMIC_STACKALLOC(SDNode* Node, // Chain the dynamic stack allocation so that it doesn't modify the stack // pointer when other instructions are using the stack. - Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(0, true), - SDLoc(Node)); + Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(0, dl, true), dl); SDValue Size = Tmp2.getOperand(1); SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT); @@ -1631,19 +1642,18 @@ void SelectionDAGLegalize::ExpandDYNAMIC_STACKALLOC(SDNode* Node, Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value if (Align > StackAlign) Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1, - DAG.getConstant(-(uint64_t)Align, VT)); + DAG.getConstant(-(uint64_t)Align, dl, VT)); Chain = DAG.getCopyToReg(Chain, dl, SPReg, Tmp1); // Output chain - Tmp2 = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(0, true), - DAG.getIntPtrConstant(0, true), SDValue(), - SDLoc(Node)); + Tmp2 = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(0, dl, true), + DAG.getIntPtrConstant(0, dl, true), SDValue(), dl); Results.push_back(Tmp1); Results.push_back(Tmp2); } -/// LegalizeSetCCCondCode - Legalize a SETCC with given LHS and RHS and -/// condition code CC on the current target. +/// Legalize a SETCC with given LHS and RHS and condition code CC on the current +/// target. /// /// If the SETCC has been legalized using AND / OR, then the legalized node /// will be stored in LHS. RHS and CC will be set to SDValue(). NeedInvert @@ -1757,7 +1767,7 @@ bool SelectionDAGLegalize::LegalizeSetCCCondCode(EVT VT, return false; } -/// EmitStackConvert - Emit a store/load combination to the stack. This stores +/// Emit a store/load combination to the stack. This stores /// SrcOp to a stack slot of type SlotVT, truncating it if needed. It then does /// a load from the stack slot to DestVT, extending it if needed. /// The resultant code need not be legal. @@ -1917,7 +1927,7 @@ ExpandBVWithShuffles(SDNode *Node, SelectionDAG &DAG, return true; } -/// ExpandBUILD_VECTOR - Expand a BUILD_VECTOR node on targets that don't +/// Expand a BUILD_VECTOR node on targets that don't /// support the operation, but do support the resultant vector type. SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) { unsigned NumElems = Node->getNumOperands(); @@ -2029,7 +2039,7 @@ SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) { return ExpandVectorBuildThroughStack(Node); } -// ExpandLibCall - Expand a node into a call to a libcall. If the result value +// Expand a node into a call to a libcall. If the result value // does not fit into a register, return the lo part and set the hi part to the // by-reg argument. If it does fit into a single register, return the result // and leave the Hi part unset. @@ -2077,7 +2087,7 @@ SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, SDNode *Node, return CallInfo.first; } -/// ExpandLibCall - Generate a libcall taking the given operands as arguments +/// Generate a libcall taking the given operands as arguments /// and returning a result of type RetVT. SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, EVT RetVT, const SDValue *Ops, unsigned NumOps, @@ -2108,7 +2118,7 @@ SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, EVT RetVT, return CallInfo.first; } -// ExpandChainLibCall - Expand a node into a call to a libcall. Similar to +// Expand a node into a call to a libcall. Similar to // ExpandLibCall except that the first operand is the in-chain. std::pair SelectionDAGLegalize::ExpandChainLibCall(RTLIB::Libcall LC, @@ -2178,7 +2188,7 @@ SDValue SelectionDAGLegalize::ExpandIntLibCall(SDNode* Node, bool isSigned, return ExpandLibCall(LC, Node, isSigned); } -/// isDivRemLibcallAvailable - Return true if divmod libcall is available. +/// Return true if divmod libcall is available. static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned, const TargetLowering &TLI) { RTLIB::Libcall LC; @@ -2194,8 +2204,7 @@ static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned, return TLI.getLibcallName(LC) != nullptr; } -/// useDivRem - Only issue divrem libcall if both quotient and remainder are -/// needed. +/// Only issue divrem libcall if both quotient and remainder are needed. static bool useDivRem(SDNode *Node, bool isSigned, bool isDIV) { // The other use might have been replaced with a divrem already. unsigned DivRemOpc = isSigned ? ISD::SDIVREM : ISD::UDIVREM; @@ -2220,8 +2229,7 @@ static bool useDivRem(SDNode *Node, bool isSigned, bool isDIV) { return false; } -/// ExpandDivRemLibCall - Issue libcalls to __{u}divmod to compute div / rem -/// pairs. +/// Issue libcalls to __{u}divmod to compute div / rem pairs. void SelectionDAGLegalize::ExpandDivRemLibCall(SDNode *Node, SmallVectorImpl &Results) { @@ -2283,7 +2291,7 @@ SelectionDAGLegalize::ExpandDivRemLibCall(SDNode *Node, Results.push_back(Rem); } -/// isSinCosLibcallAvailable - Return true if sincos libcall is available. +/// Return true if sincos libcall is available. static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) { RTLIB::Libcall LC; switch (Node->getSimpleValueType(0).SimpleTy) { @@ -2297,8 +2305,8 @@ static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) { return TLI.getLibcallName(LC) != nullptr; } -/// canCombineSinCosLibcall - Return true if sincos libcall is available and -/// can be used to combine sin and cos. +/// Return true if sincos libcall is available and can be used to combine sin +/// and cos. static bool canCombineSinCosLibcall(SDNode *Node, const TargetLowering &TLI, const TargetMachine &TM) { if (!isSinCosLibcallAvailable(Node, TLI)) @@ -2311,8 +2319,7 @@ static bool canCombineSinCosLibcall(SDNode *Node, const TargetLowering &TLI, return true; } -/// useSinCos - Only issue sincos libcall if both sin and cos are -/// needed. +/// Only issue sincos libcall if both sin and cos are needed. static bool useSinCos(SDNode *Node) { unsigned OtherOpcode = Node->getOpcode() == ISD::FSIN ? ISD::FCOS : ISD::FSIN; @@ -2330,8 +2337,7 @@ static bool useSinCos(SDNode *Node) { return false; } -/// ExpandSinCosLibCall - Issue libcalls to sincos to compute sin / cos -/// pairs. +/// Issue libcalls to sincos to compute sin / cos pairs. void SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node, SmallVectorImpl &Results) { @@ -2396,7 +2402,7 @@ SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node, MachinePointerInfo(), false, false, false, 0)); } -/// ExpandLegalINT_TO_FP - This function is responsible for legalizing a +/// This function is responsible for legalizing a /// INT_TO_FP operation of the specified operand when the target requests that /// we expand it. At this point, we know that the result and operand types are /// legal for the target. @@ -2411,7 +2417,8 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, SDValue StackSlot = DAG.CreateStackTemporary(MVT::f64); // word offset constant for Hi/Lo address computation - SDValue WordOff = DAG.getConstant(sizeof(int), StackSlot.getValueType()); + SDValue WordOff = DAG.getConstant(sizeof(int), dl, + StackSlot.getValueType()); // set up Hi and Lo (into buffer) address based on endian SDValue Hi = StackSlot; SDValue Lo = DAG.getNode(ISD::ADD, dl, StackSlot.getValueType(), @@ -2423,7 +2430,7 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, SDValue Op0Mapped; if (isSigned) { // constant used to invert sign bit (signed to unsigned mapping) - SDValue SignBit = DAG.getConstant(0x80000000u, MVT::i32); + SDValue SignBit = DAG.getConstant(0x80000000u, dl, MVT::i32); Op0Mapped = DAG.getNode(ISD::XOR, dl, MVT::i32, Op0, SignBit); } else { Op0Mapped = Op0; @@ -2433,7 +2440,7 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, Op0Mapped, Lo, MachinePointerInfo(), false, false, 0); // initial hi portion of constructed double - SDValue InitialHi = DAG.getConstant(0x43300000u, MVT::i32); + SDValue InitialHi = DAG.getConstant(0x43300000u, dl, MVT::i32); // store the hi of the constructed double - biased exponent SDValue Store2 = DAG.getStore(Store1, dl, InitialHi, Hi, MachinePointerInfo(), @@ -2445,7 +2452,7 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, SDValue Bias = DAG.getConstantFP(isSigned ? BitsToDouble(0x4330000080000000ULL) : BitsToDouble(0x4330000000000000ULL), - MVT::f64); + dl, MVT::f64); // subtract the bias SDValue Sub = DAG.getNode(ISD::FSUB, dl, MVT::f64, Load, Bias); // final result @@ -2456,7 +2463,7 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, Result = Sub; } else if (DestVT.bitsLT(MVT::f64)) { Result = DAG.getNode(ISD::FP_ROUND, dl, DestVT, Sub, - DAG.getIntPtrConstant(0)); + DAG.getIntPtrConstant(0, dl)); } else if (DestVT.bitsGT(MVT::f64)) { Result = DAG.getNode(ISD::FP_EXTEND, dl, DestVT, Sub); } @@ -2472,15 +2479,16 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, // TODO: Generalize this for use with other types. if (Op0.getValueType() == MVT::i64 && DestVT == MVT::f64) { SDValue TwoP52 = - DAG.getConstant(UINT64_C(0x4330000000000000), MVT::i64); + DAG.getConstant(UINT64_C(0x4330000000000000), dl, MVT::i64); SDValue TwoP84PlusTwoP52 = - DAG.getConstantFP(BitsToDouble(UINT64_C(0x4530000000100000)), MVT::f64); + DAG.getConstantFP(BitsToDouble(UINT64_C(0x4530000000100000)), dl, + MVT::f64); SDValue TwoP84 = - DAG.getConstant(UINT64_C(0x4530000000000000), MVT::i64); + DAG.getConstant(UINT64_C(0x4530000000000000), dl, MVT::i64); SDValue Lo = DAG.getZeroExtendInReg(Op0, dl, MVT::i32); SDValue Hi = DAG.getNode(ISD::SRL, dl, MVT::i64, Op0, - DAG.getConstant(32, MVT::i64)); + DAG.getConstant(32, dl, MVT::i64)); SDValue LoOr = DAG.getNode(ISD::OR, dl, MVT::i64, Lo, TwoP52); SDValue HiOr = DAG.getNode(ISD::OR, dl, MVT::i64, Hi, TwoP84); SDValue LoFlt = DAG.getNode(ISD::BITCAST, dl, MVT::f64, LoOr); @@ -2499,9 +2507,9 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, SDValue Fast = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, Op0); SDValue ShiftConst = - DAG.getConstant(1, TLI.getShiftAmountTy(Op0.getValueType())); + DAG.getConstant(1, dl, TLI.getShiftAmountTy(Op0.getValueType())); SDValue Shr = DAG.getNode(ISD::SRL, dl, MVT::i64, Op0, ShiftConst); - SDValue AndConst = DAG.getConstant(1, MVT::i64); + SDValue AndConst = DAG.getConstant(1, dl, MVT::i64); SDValue And = DAG.getNode(ISD::AND, dl, MVT::i64, Op0, AndConst); SDValue Or = DAG.getNode(ISD::OR, dl, MVT::i64, And, Shr); @@ -2513,47 +2521,52 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, // thing most of the time. This would be a good candidate for a //pseudo-op, or, even better, for whole-function isel. SDValue SignBitTest = DAG.getSetCC(dl, getSetCCResultType(MVT::i64), - Op0, DAG.getConstant(0, MVT::i64), ISD::SETLT); + Op0, DAG.getConstant(0, dl, MVT::i64), ISD::SETLT); return DAG.getSelect(dl, MVT::f32, SignBitTest, Slow, Fast); } // Otherwise, implement the fully general conversion. SDValue And = DAG.getNode(ISD::AND, dl, MVT::i64, Op0, - DAG.getConstant(UINT64_C(0xfffffffffffff800), MVT::i64)); + DAG.getConstant(UINT64_C(0xfffffffffffff800), dl, MVT::i64)); SDValue Or = DAG.getNode(ISD::OR, dl, MVT::i64, And, - DAG.getConstant(UINT64_C(0x800), MVT::i64)); + DAG.getConstant(UINT64_C(0x800), dl, MVT::i64)); SDValue And2 = DAG.getNode(ISD::AND, dl, MVT::i64, Op0, - DAG.getConstant(UINT64_C(0x7ff), MVT::i64)); - SDValue Ne = DAG.getSetCC(dl, getSetCCResultType(MVT::i64), - And2, DAG.getConstant(UINT64_C(0), MVT::i64), ISD::SETNE); + DAG.getConstant(UINT64_C(0x7ff), dl, MVT::i64)); + SDValue Ne = DAG.getSetCC(dl, getSetCCResultType(MVT::i64), And2, + DAG.getConstant(UINT64_C(0), dl, MVT::i64), + ISD::SETNE); SDValue Sel = DAG.getSelect(dl, MVT::i64, Ne, Or, Op0); - SDValue Ge = DAG.getSetCC(dl, getSetCCResultType(MVT::i64), - Op0, DAG.getConstant(UINT64_C(0x0020000000000000), MVT::i64), - ISD::SETUGE); + SDValue Ge = DAG.getSetCC(dl, getSetCCResultType(MVT::i64), Op0, + DAG.getConstant(UINT64_C(0x0020000000000000), dl, + MVT::i64), + ISD::SETUGE); SDValue Sel2 = DAG.getSelect(dl, MVT::i64, Ge, Sel, Op0); EVT SHVT = TLI.getShiftAmountTy(Sel2.getValueType()); SDValue Sh = DAG.getNode(ISD::SRL, dl, MVT::i64, Sel2, - DAG.getConstant(32, SHVT)); + DAG.getConstant(32, dl, SHVT)); SDValue Trunc = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Sh); SDValue Fcvt = DAG.getNode(ISD::UINT_TO_FP, dl, MVT::f64, Trunc); SDValue TwoP32 = - DAG.getConstantFP(BitsToDouble(UINT64_C(0x41f0000000000000)), MVT::f64); + DAG.getConstantFP(BitsToDouble(UINT64_C(0x41f0000000000000)), dl, + MVT::f64); SDValue Fmul = DAG.getNode(ISD::FMUL, dl, MVT::f64, TwoP32, Fcvt); SDValue Lo = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Sel2); SDValue Fcvt2 = DAG.getNode(ISD::UINT_TO_FP, dl, MVT::f64, Lo); SDValue Fadd = DAG.getNode(ISD::FADD, dl, MVT::f64, Fmul, Fcvt2); return DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Fadd, - DAG.getIntPtrConstant(0)); + DAG.getIntPtrConstant(0, dl)); } SDValue Tmp1 = DAG.getNode(ISD::SINT_TO_FP, dl, DestVT, Op0); SDValue SignSet = DAG.getSetCC(dl, getSetCCResultType(Op0.getValueType()), - Op0, DAG.getConstant(0, Op0.getValueType()), + Op0, + DAG.getConstant(0, dl, Op0.getValueType()), ISD::SETLT); - SDValue Zero = DAG.getIntPtrConstant(0), Four = DAG.getIntPtrConstant(4); + SDValue Zero = DAG.getIntPtrConstant(0, dl), + Four = DAG.getIntPtrConstant(4, dl); SDValue CstOffset = DAG.getSelect(dl, Zero.getValueType(), SignSet, Four, Zero); @@ -2594,7 +2607,7 @@ SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned, return DAG.getNode(ISD::FADD, dl, DestVT, Tmp1, FudgeInReg); } -/// PromoteLegalINT_TO_FP - This function is responsible for legalizing a +/// This function is responsible for legalizing a /// *INT_TO_FP operation of the specified operand when the target requests that /// we promote it. At this point, we know that the result and operand types are /// legal for the target, and that there is a legal UINT_TO_FP or SINT_TO_FP @@ -2636,7 +2649,7 @@ SDValue SelectionDAGLegalize::PromoteLegalINT_TO_FP(SDValue LegalOp, dl, NewInTy, LegalOp)); } -/// PromoteLegalFP_TO_INT - This function is responsible for legalizing a +/// This function is responsible for legalizing a /// FP_TO_*INT operation of the specified operand when the target requests that /// we promote it. At this point, we know that the result and operand types are /// legal for the target, and that there is a legal FP_TO_UINT or FP_TO_SINT @@ -2680,8 +2693,7 @@ SDValue SelectionDAGLegalize::PromoteLegalFP_TO_INT(SDValue LegalOp, return DAG.getNode(ISD::TRUNCATE, dl, DestVT, Operation); } -/// ExpandBSWAP - Open code the operations for BSWAP of the specified operation. -/// +/// Open code the operations for BSWAP of the specified operation. SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) { EVT VT = Op.getValueType(); EVT SHVT = TLI.getShiftAmountTy(VT); @@ -2689,34 +2701,41 @@ SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) { switch (VT.getSimpleVT().SimpleTy) { default: llvm_unreachable("Unhandled Expand type in BSWAP!"); case MVT::i16: - Tmp2 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, SHVT)); - Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, SHVT)); + Tmp2 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); + Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); return DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2); case MVT::i32: - Tmp4 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(24, SHVT)); - Tmp3 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, SHVT)); - Tmp2 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, SHVT)); - Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(24, SHVT)); - Tmp3 = DAG.getNode(ISD::AND, dl, VT, Tmp3, DAG.getConstant(0xFF0000, VT)); - Tmp2 = DAG.getNode(ISD::AND, dl, VT, Tmp2, DAG.getConstant(0xFF00, VT)); + Tmp4 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(24, dl, SHVT)); + Tmp3 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); + Tmp2 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); + Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(24, dl, SHVT)); + Tmp3 = DAG.getNode(ISD::AND, dl, VT, Tmp3, + DAG.getConstant(0xFF0000, dl, VT)); + Tmp2 = DAG.getNode(ISD::AND, dl, VT, Tmp2, DAG.getConstant(0xFF00, dl, VT)); Tmp4 = DAG.getNode(ISD::OR, dl, VT, Tmp4, Tmp3); Tmp2 = DAG.getNode(ISD::OR, dl, VT, Tmp2, Tmp1); return DAG.getNode(ISD::OR, dl, VT, Tmp4, Tmp2); case MVT::i64: - Tmp8 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(56, SHVT)); - Tmp7 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(40, SHVT)); - Tmp6 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(24, SHVT)); - Tmp5 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, SHVT)); - Tmp4 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, SHVT)); - Tmp3 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(24, SHVT)); - Tmp2 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(40, SHVT)); - Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(56, SHVT)); - Tmp7 = DAG.getNode(ISD::AND, dl, VT, Tmp7, DAG.getConstant(255ULL<<48, VT)); - Tmp6 = DAG.getNode(ISD::AND, dl, VT, Tmp6, DAG.getConstant(255ULL<<40, VT)); - Tmp5 = DAG.getNode(ISD::AND, dl, VT, Tmp5, DAG.getConstant(255ULL<<32, VT)); - Tmp4 = DAG.getNode(ISD::AND, dl, VT, Tmp4, DAG.getConstant(255ULL<<24, VT)); - Tmp3 = DAG.getNode(ISD::AND, dl, VT, Tmp3, DAG.getConstant(255ULL<<16, VT)); - Tmp2 = DAG.getNode(ISD::AND, dl, VT, Tmp2, DAG.getConstant(255ULL<<8 , VT)); + Tmp8 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(56, dl, SHVT)); + Tmp7 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(40, dl, SHVT)); + Tmp6 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(24, dl, SHVT)); + Tmp5 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); + Tmp4 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, dl, SHVT)); + Tmp3 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(24, dl, SHVT)); + Tmp2 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(40, dl, SHVT)); + Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(56, dl, SHVT)); + Tmp7 = DAG.getNode(ISD::AND, dl, VT, Tmp7, + DAG.getConstant(255ULL<<48, dl, VT)); + Tmp6 = DAG.getNode(ISD::AND, dl, VT, Tmp6, + DAG.getConstant(255ULL<<40, dl, VT)); + Tmp5 = DAG.getNode(ISD::AND, dl, VT, Tmp5, + DAG.getConstant(255ULL<<32, dl, VT)); + Tmp4 = DAG.getNode(ISD::AND, dl, VT, Tmp4, + DAG.getConstant(255ULL<<24, dl, VT)); + Tmp3 = DAG.getNode(ISD::AND, dl, VT, Tmp3, + DAG.getConstant(255ULL<<16, dl, VT)); + Tmp2 = DAG.getNode(ISD::AND, dl, VT, Tmp2, + DAG.getConstant(255ULL<<8 , dl, VT)); Tmp8 = DAG.getNode(ISD::OR, dl, VT, Tmp8, Tmp7); Tmp6 = DAG.getNode(ISD::OR, dl, VT, Tmp6, Tmp5); Tmp4 = DAG.getNode(ISD::OR, dl, VT, Tmp4, Tmp3); @@ -2727,8 +2746,7 @@ SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) { } } -/// ExpandBitCount - Expand the specified bitcount instruction into operations. -/// +/// Expand the specified bitcount instruction into operations. SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, SDLoc dl) { switch (Opc) { @@ -2744,34 +2762,38 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, // This is the "best" algorithm from // http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel - SDValue Mask55 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x55)), VT); - SDValue Mask33 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x33)), VT); - SDValue Mask0F = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x0F)), VT); - SDValue Mask01 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x01)), VT); + SDValue Mask55 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x55)), + dl, VT); + SDValue Mask33 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x33)), + dl, VT); + SDValue Mask0F = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x0F)), + dl, VT); + SDValue Mask01 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x01)), + dl, VT); // v = v - ((v >> 1) & 0x55555555...) Op = DAG.getNode(ISD::SUB, dl, VT, Op, DAG.getNode(ISD::AND, dl, VT, DAG.getNode(ISD::SRL, dl, VT, Op, - DAG.getConstant(1, ShVT)), + DAG.getConstant(1, dl, ShVT)), Mask55)); // v = (v & 0x33333333...) + ((v >> 2) & 0x33333333...) Op = DAG.getNode(ISD::ADD, dl, VT, DAG.getNode(ISD::AND, dl, VT, Op, Mask33), DAG.getNode(ISD::AND, dl, VT, DAG.getNode(ISD::SRL, dl, VT, Op, - DAG.getConstant(2, ShVT)), + DAG.getConstant(2, dl, ShVT)), Mask33)); // v = (v + (v >> 4)) & 0x0F0F0F0F... Op = DAG.getNode(ISD::AND, dl, VT, DAG.getNode(ISD::ADD, dl, VT, Op, DAG.getNode(ISD::SRL, dl, VT, Op, - DAG.getConstant(4, ShVT))), + DAG.getConstant(4, dl, ShVT))), Mask0F); // v = (v * 0x01010101...) >> (Len - 8) Op = DAG.getNode(ISD::SRL, dl, VT, DAG.getNode(ISD::MUL, dl, VT, Op, Mask01), - DAG.getConstant(Len - 8, ShVT)); + DAG.getConstant(Len - 8, dl, ShVT)); return Op; } @@ -2792,7 +2814,7 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, EVT ShVT = TLI.getShiftAmountTy(VT); unsigned len = VT.getSizeInBits(); for (unsigned i = 0; (1U << i) <= (len / 2); ++i) { - SDValue Tmp3 = DAG.getConstant(1ULL << i, ShVT); + SDValue Tmp3 = DAG.getConstant(1ULL << i, dl, ShVT); Op = DAG.getNode(ISD::OR, dl, VT, Op, DAG.getNode(ISD::SRL, dl, VT, Op, Tmp3)); } @@ -2811,12 +2833,12 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, SDValue Tmp3 = DAG.getNode(ISD::AND, dl, VT, DAG.getNOT(dl, Op, VT), DAG.getNode(ISD::SUB, dl, VT, Op, - DAG.getConstant(1, VT))); + DAG.getConstant(1, dl, VT))); // If ISD::CTLZ is legal and CTPOP isn't, then do that instead. if (!TLI.isOperationLegalOrCustom(ISD::CTPOP, VT) && TLI.isOperationLegalOrCustom(ISD::CTLZ, VT)) return DAG.getNode(ISD::SUB, dl, VT, - DAG.getConstant(VT.getSizeInBits(), VT), + DAG.getConstant(VT.getSizeInBits(), dl, VT), DAG.getNode(ISD::CTLZ, dl, VT, Tmp3)); return DAG.getNode(ISD::CTPOP, dl, VT, Tmp3); } @@ -2826,132 +2848,8 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, std::pair SelectionDAGLegalize::ExpandAtomic(SDNode *Node) { unsigned Opc = Node->getOpcode(); MVT VT = cast(Node)->getMemoryVT().getSimpleVT(); - RTLIB::Libcall LC; - - switch (Opc) { - default: - llvm_unreachable("Unhandled atomic intrinsic Expand!"); - case ISD::ATOMIC_SWAP: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_1; break; - case MVT::i16: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_2; break; - case MVT::i32: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_4; break; - case MVT::i64: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_8; break; - case MVT::i128:LC = RTLIB::SYNC_LOCK_TEST_AND_SET_16;break; - } - break; - case ISD::ATOMIC_CMP_SWAP: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_1; break; - case MVT::i16: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_2; break; - case MVT::i32: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_4; break; - case MVT::i64: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_8; break; - case MVT::i128:LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_16;break; - } - break; - case ISD::ATOMIC_LOAD_ADD: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_ADD_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_ADD_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_ADD_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_ADD_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_ADD_16;break; - } - break; - case ISD::ATOMIC_LOAD_SUB: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_SUB_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_SUB_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_SUB_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_SUB_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_SUB_16;break; - } - break; - case ISD::ATOMIC_LOAD_AND: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_AND_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_AND_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_AND_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_AND_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_AND_16;break; - } - break; - case ISD::ATOMIC_LOAD_OR: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_OR_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_OR_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_OR_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_OR_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_OR_16;break; - } - break; - case ISD::ATOMIC_LOAD_XOR: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_XOR_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_XOR_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_XOR_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_XOR_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_XOR_16;break; - } - break; - case ISD::ATOMIC_LOAD_NAND: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_NAND_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_NAND_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_NAND_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_NAND_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_NAND_16;break; - } - break; - case ISD::ATOMIC_LOAD_MAX: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MAX_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MAX_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MAX_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MAX_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MAX_16;break; - } - break; - case ISD::ATOMIC_LOAD_UMAX: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMAX_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMAX_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMAX_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMAX_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMAX_16;break; - } - break; - case ISD::ATOMIC_LOAD_MIN: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MIN_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MIN_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MIN_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MIN_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MIN_16;break; - } - break; - case ISD::ATOMIC_LOAD_UMIN: - switch (VT.SimpleTy) { - default: llvm_unreachable("Unexpected value type for atomic!"); - case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMIN_1; break; - case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMIN_2; break; - case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMIN_4; break; - case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMIN_8; break; - case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMIN_16;break; - } - break; - } + RTLIB::Libcall LC = RTLIB::getATOMIC(Opc, VT); + assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpected atomic op or value type!"); return ExpandChainLibCall(LC, Node, false); } @@ -2976,10 +2874,10 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { case ISD::FRAMEADDR: case ISD::RETURNADDR: case ISD::FRAME_TO_ARGS_OFFSET: - Results.push_back(DAG.getConstant(0, Node->getValueType(0))); + Results.push_back(DAG.getConstant(0, dl, Node->getValueType(0))); break; case ISD::FLT_ROUNDS_: - Results.push_back(DAG.getConstant(1, Node->getValueType(0))); + Results.push_back(DAG.getConstant(1, dl, Node->getValueType(0))); break; case ISD::EH_RETURN: case ISD::EH_LABEL: @@ -2993,7 +2891,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { case ISD::EH_SJLJ_SETJMP: // If the target didn't expand this, just return 'zero' and preserve the // chain. - Results.push_back(DAG.getConstant(0, MVT::i32)); + Results.push_back(DAG.getConstant(0, dl, MVT::i32)); Results.push_back(Node->getOperand(0)); break; case ISD::ATOMIC_FENCE: { @@ -3014,7 +2912,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { } case ISD::ATOMIC_LOAD: { // There is no libcall for atomic load; fake it with ATOMIC_CMP_SWAP. - SDValue Zero = DAG.getConstant(0, Node->getValueType(0)); + SDValue Zero = DAG.getConstant(0, dl, Node->getValueType(0)); SDVTList VTs = DAG.getVTList(Node->getValueType(0), MVT::Other); SDValue Swap = DAG.getAtomicCmpSwap( ISD::ATOMIC_CMP_SWAP, dl, cast(Node)->getMemoryVT(), VTs, @@ -3090,10 +2988,10 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { case ISD::UNDEF: { EVT VT = Node->getValueType(0); if (VT.isInteger()) - Results.push_back(DAG.getConstant(0, VT)); + Results.push_back(DAG.getConstant(0, dl, VT)); else { assert(VT.isFloatingPoint() && "Unknown value type!"); - Results.push_back(DAG.getConstantFP(0, VT)); + Results.push_back(DAG.getConstantFP(0, dl, VT)); } break; } @@ -3132,7 +3030,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { ShiftAmountTy = VT; unsigned BitsDiff = VT.getScalarType().getSizeInBits() - ExtraVT.getScalarType().getSizeInBits(); - SDValue ShiftCst = DAG.getConstant(BitsDiff, ShiftAmountTy); + SDValue ShiftCst = DAG.getConstant(BitsDiff, dl, ShiftAmountTy); Tmp1 = DAG.getNode(ISD::SHL, dl, Node->getValueType(0), Node->getOperand(0), ShiftCst); Tmp1 = DAG.getNode(ISD::SRA, dl, Node->getValueType(0), Tmp1, ShiftCst); @@ -3170,7 +3068,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { APInt::getNullValue(VT.getSizeInBits())); APInt x = APInt::getSignBit(NVT.getSizeInBits()); (void)apf.convertFromAPInt(x, false, APFloat::rmNearestTiesToEven); - Tmp1 = DAG.getConstantFP(apf, VT); + Tmp1 = DAG.getConstantFP(apf, dl, VT); Tmp2 = DAG.getSetCC(dl, getSetCCResultType(VT), Node->getOperand(0), Tmp1, ISD::SETLT); @@ -3179,7 +3077,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { DAG.getNode(ISD::FSUB, dl, VT, Node->getOperand(0), Tmp1)); False = DAG.getNode(ISD::XOR, dl, NVT, False, - DAG.getConstant(x, NVT)); + DAG.getConstant(x, dl, NVT)); Tmp1 = DAG.getSelect(dl, NVT, Tmp2, True, False); Results.push_back(Tmp1); break; @@ -3200,11 +3098,11 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { assert(((Align & (Align-1)) == 0) && "Expected Align to be a power of 2"); VAList = DAG.getNode(ISD::ADD, dl, VAList.getValueType(), VAList, - DAG.getConstant(Align - 1, + DAG.getConstant(Align - 1, dl, VAList.getValueType())); VAList = DAG.getNode(ISD::AND, dl, VAList.getValueType(), VAList, - DAG.getConstant(-(int64_t)Align, + DAG.getConstant(-(int64_t)Align, dl, VAList.getValueType())); } @@ -3212,6 +3110,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { Tmp3 = DAG.getNode(ISD::ADD, dl, VAList.getValueType(), VAList, DAG.getConstant(TLI.getDataLayout()-> getTypeAllocSize(VT.getTypeForEVT(*DAG.getContext())), + dl, VAList.getValueType())); // Store the incremented VAList to the legalized pointer Tmp3 = DAG.getStore(VAListLoad.getValue(1), dl, Tmp3, Tmp2, @@ -3326,11 +3225,11 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { if (Idx < NumElems) Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, Op0, - DAG.getConstant(Idx, TLI.getVectorIdxTy()))); + DAG.getConstant(Idx, dl, TLI.getVectorIdxTy()))); else Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, Op1, - DAG.getConstant(Idx - NumElems, + DAG.getConstant(Idx - NumElems, dl, TLI.getVectorIdxTy()))); } @@ -3345,7 +3244,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { if (cast(Node->getOperand(1))->getZExtValue()) { // 1 -> Hi Tmp1 = DAG.getNode(ISD::SRL, dl, OpTy, Node->getOperand(0), - DAG.getConstant(OpTy.getSizeInBits()/2, + DAG.getConstant(OpTy.getSizeInBits()/2, dl, TLI.getShiftAmountTy(Node->getOperand(0).getValueType()))); Tmp1 = DAG.getNode(ISD::TRUNCATE, dl, Node->getValueType(0), Tmp1); } else { @@ -3383,7 +3282,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { break; case ISD::FNEG: // Expand Y = FNEG(X) -> Y = SUB -0.0, X - Tmp1 = DAG.getConstantFP(-0.0, Node->getValueType(0)); + Tmp1 = DAG.getConstantFP(-0.0, dl, Node->getValueType(0)); Tmp1 = DAG.getNode(ISD::FSUB, dl, Node->getValueType(0), Tmp1, Node->getOperand(0)); Results.push_back(Tmp1); @@ -3392,7 +3291,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { // Expand Y = FABS(X) -> Y = (X >u 0.0) ? X : fneg(X). EVT VT = Node->getValueType(0); Tmp1 = Node->getOperand(0); - Tmp2 = DAG.getConstantFP(0.0, VT); + Tmp2 = DAG.getConstantFP(0.0, dl, VT); Tmp2 = DAG.getSetCC(dl, getSetCCResultType(Tmp1.getValueType()), Tmp1, Tmp2, ISD::SETUGT); Tmp3 = DAG.getNode(ISD::FNEG, dl, VT, Tmp1); @@ -3400,6 +3299,26 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { Results.push_back(Tmp1); break; } + case ISD::SMIN: + case ISD::SMAX: + case ISD::UMIN: + case ISD::UMAX: { + // Expand Y = MAX(A, B) -> Y = (A > B) ? A : B + ISD::CondCode Pred; + switch (Node->getOpcode()) { + default: llvm_unreachable("How did we get here?"); + case ISD::SMAX: Pred = ISD::SETGT; break; + case ISD::SMIN: Pred = ISD::SETLT; break; + case ISD::UMAX: Pred = ISD::SETUGT; break; + case ISD::UMIN: Pred = ISD::SETULT; break; + } + Tmp1 = Node->getOperand(0); + Tmp2 = Node->getOperand(1); + Tmp1 = DAG.getSelectCC(dl, Tmp1, Tmp2, Tmp1, Tmp2, Pred); + Results.push_back(Tmp1); + break; + } + case ISD::FMINNUM: Results.push_back(ExpandFPLibCall(Node, RTLIB::FMIN_F32, RTLIB::FMIN_F64, RTLIB::FMIN_F80, RTLIB::FMIN_F128, @@ -3528,6 +3447,9 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { RTLIB::FMA_F80, RTLIB::FMA_F128, RTLIB::FMA_PPCF128)); break; + case ISD::FMAD: + llvm_unreachable("Illegal fmad should never be formed"); + case ISD::FADD: Results.push_back(ExpandFPLibCall(Node, RTLIB::ADD_F32, RTLIB::ADD_F64, RTLIB::ADD_F80, RTLIB::ADD_F128, @@ -3554,6 +3476,21 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { break; } case ISD::FP_TO_FP16: { + if (!TLI.useSoftFloat() && TM.Options.UnsafeFPMath) { + SDValue Op = Node->getOperand(0); + MVT SVT = Op.getSimpleValueType(); + if ((SVT == MVT::f64 || SVT == MVT::f80) && + TLI.isOperationLegalOrCustom(ISD::FP_TO_FP16, MVT::f32)) { + // Under fastmath, we can expand this node into a fround followed by + // a float-half conversion. + SDValue FloatVal = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Op, + DAG.getIntPtrConstant(0, dl)); + Results.push_back( + DAG.getNode(ISD::FP_TO_FP16, dl, MVT::i16, FloatVal)); + break; + } + } + RTLIB::Libcall LC = RTLIB::getFPROUND(Node->getOperand(0).getValueType(), MVT::f16); assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unable to expand fp_to_fp16"); @@ -3588,8 +3525,9 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { TLI.isOperationLegalOrCustom(ISD::XOR, VT) && "Don't know how to expand this subtraction!"); Tmp1 = DAG.getNode(ISD::XOR, dl, VT, Node->getOperand(1), - DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), VT)); - Tmp1 = DAG.getNode(ISD::ADD, dl, VT, Tmp1, DAG.getConstant(1, VT)); + DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), dl, + VT)); + Tmp1 = DAG.getNode(ISD::ADD, dl, VT, Tmp1, DAG.getConstant(1, dl, VT)); Results.push_back(DAG.getNode(ISD::ADD, dl, VT, Node->getOperand(0), Tmp1)); break; } @@ -3705,7 +3643,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { TLI.expandMUL(Node, Lo, Hi, HalfType, DAG)) { Lo = DAG.getNode(ISD::ZERO_EXTEND, dl, VT, Lo); Hi = DAG.getNode(ISD::ANY_EXTEND, dl, VT, Hi); - SDValue Shift = DAG.getConstant(HalfType.getSizeInBits(), + SDValue Shift = DAG.getConstant(HalfType.getSizeInBits(), dl, TLI.getShiftAmountTy(HalfType)); Hi = DAG.getNode(ISD::SHL, dl, VT, Hi, Shift); Results.push_back(DAG.getNode(ISD::OR, dl, VT, Lo, Hi)); @@ -3730,7 +3668,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { EVT ResultType = Node->getValueType(1); EVT OType = getSetCCResultType(Node->getValueType(0)); - SDValue Zero = DAG.getConstant(0, LHS.getValueType()); + SDValue Zero = DAG.getConstant(0, dl, LHS.getValueType()); // LHSSign -> LHS >= 0 // RHSSign -> RHS >= 0 @@ -3796,9 +3734,9 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { RHS = DAG.getNode(Ops[isSigned][2], dl, WideVT, RHS); Tmp1 = DAG.getNode(ISD::MUL, dl, WideVT, LHS, RHS); BottomHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Tmp1, - DAG.getIntPtrConstant(0)); + DAG.getIntPtrConstant(0, dl)); TopHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Tmp1, - DAG.getIntPtrConstant(1)); + DAG.getIntPtrConstant(1, dl)); } else { // We can fall back to a libcall with an illegal type for the MUL if we // have a libcall big enough. @@ -3819,9 +3757,11 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { // part. unsigned LoSize = VT.getSizeInBits(); SDValue HiLHS = DAG.getNode(ISD::SRA, dl, VT, RHS, - DAG.getConstant(LoSize-1, TLI.getPointerTy())); + DAG.getConstant(LoSize - 1, dl, + TLI.getPointerTy())); SDValue HiRHS = DAG.getNode(ISD::SRA, dl, VT, LHS, - DAG.getConstant(LoSize-1, TLI.getPointerTy())); + DAG.getConstant(LoSize - 1, dl, + TLI.getPointerTy())); // Here we're passing the 2 arguments explicitly as 4 arguments that are // pre-lowered to the correct types. This all depends upon WideVT not @@ -3830,9 +3770,9 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { SDValue Args[] = { LHS, HiLHS, RHS, HiRHS }; SDValue Ret = ExpandLibCall(LC, WideVT, Args, 4, isSigned, dl); BottomHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Ret, - DAG.getIntPtrConstant(0)); + DAG.getIntPtrConstant(0, dl)); TopHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Ret, - DAG.getIntPtrConstant(1)); + DAG.getIntPtrConstant(1, dl)); // Ret is a node with an illegal type. Because such things are not // generally permitted during this phase of legalization, make sure the // node has no more uses. The above EXTRACT_ELEMENT nodes should have been @@ -3842,14 +3782,14 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { } if (isSigned) { - Tmp1 = DAG.getConstant(VT.getSizeInBits() - 1, + Tmp1 = DAG.getConstant(VT.getSizeInBits() - 1, dl, TLI.getShiftAmountTy(BottomHalf.getValueType())); Tmp1 = DAG.getNode(ISD::SRA, dl, VT, BottomHalf, Tmp1); TopHalf = DAG.getSetCC(dl, getSetCCResultType(VT), TopHalf, Tmp1, ISD::SETNE); } else { TopHalf = DAG.getSetCC(dl, getSetCCResultType(VT), TopHalf, - DAG.getConstant(0, VT), ISD::SETNE); + DAG.getConstant(0, dl, VT), ISD::SETNE); } Results.push_back(BottomHalf); Results.push_back(TopHalf); @@ -3860,7 +3800,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { Tmp1 = DAG.getNode(ISD::ZERO_EXTEND, dl, PairTy, Node->getOperand(0)); Tmp2 = DAG.getNode(ISD::ANY_EXTEND, dl, PairTy, Node->getOperand(1)); Tmp2 = DAG.getNode(ISD::SHL, dl, PairTy, Tmp2, - DAG.getConstant(PairTy.getSizeInBits()/2, + DAG.getConstant(PairTy.getSizeInBits()/2, dl, TLI.getShiftAmountTy(PairTy))); Results.push_back(DAG.getNode(ISD::OR, dl, PairTy, Tmp1, Tmp2)); break; @@ -3875,7 +3815,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { cast(Tmp1.getOperand(2))->get()); } else { Tmp1 = DAG.getSelectCC(dl, Tmp1, - DAG.getConstant(0, Tmp1.getValueType()), + DAG.getConstant(0, dl, Tmp1.getValueType()), Tmp2, Tmp3, ISD::SETNE); } Results.push_back(Tmp1); @@ -3891,8 +3831,8 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { unsigned EntrySize = DAG.getMachineFunction().getJumpTableInfo()->getEntrySize(TD); - Index = DAG.getNode(ISD::MUL, dl, Index.getValueType(), - Index, DAG.getConstant(EntrySize, Index.getValueType())); + Index = DAG.getNode(ISD::MUL, dl, Index.getValueType(), Index, + DAG.getConstant(EntrySize, dl, Index.getValueType())); SDValue Addr = DAG.getNode(ISD::ADD, dl, Index.getValueType(), Index, Table); @@ -3926,10 +3866,10 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { // We test only the i1 bit. Skip the AND if UNDEF. Tmp3 = (Tmp2.getOpcode() == ISD::UNDEF) ? Tmp2 : DAG.getNode(ISD::AND, dl, Tmp2.getValueType(), Tmp2, - DAG.getConstant(1, Tmp2.getValueType())); + DAG.getConstant(1, dl, Tmp2.getValueType())); Tmp1 = DAG.getNode(ISD::BR_CC, dl, MVT::Other, Tmp1, DAG.getCondCode(ISD::SETNE), Tmp3, - DAG.getConstant(0, Tmp3.getValueType()), + DAG.getConstant(0, dl, Tmp3.getValueType()), Node->getOperand(2)); } Results.push_back(Tmp1); @@ -3971,7 +3911,8 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { break; } Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, VT, Tmp1, Tmp2, - DAG.getConstant(TrueValue, VT), DAG.getConstant(0, VT), + DAG.getConstant(TrueValue, dl, VT), + DAG.getConstant(0, dl, VT), Tmp3); Results.push_back(Tmp1); break; @@ -4039,7 +3980,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, Node->getValueType(0), Tmp1, Tmp2, Tmp3, Tmp4, CC); } else { - Tmp2 = DAG.getConstant(0, Tmp1.getValueType()); + Tmp2 = DAG.getConstant(0, dl, Tmp1.getValueType()); CC = DAG.getCondCode(ISD::SETNE); Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, Node->getValueType(0), Tmp1, Tmp2, Tmp3, Tmp4, CC); @@ -4070,7 +4011,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { Tmp1 = DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), Tmp1, Tmp4, Tmp2, Tmp3, Node->getOperand(4)); } else { - Tmp3 = DAG.getConstant(0, Tmp2.getValueType()); + Tmp3 = DAG.getConstant(0, dl, Tmp2.getValueType()); Tmp4 = DAG.getCondCode(ISD::SETNE); Tmp1 = DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), Tmp1, Tmp4, Tmp2, Tmp3, Node->getOperand(4)); @@ -4094,12 +4035,12 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) { for (unsigned Idx = 0; Idx < NumElem; Idx++) { SDValue Ex = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT.getScalarType(), - Node->getOperand(0), DAG.getConstant(Idx, - TLI.getVectorIdxTy())); + Node->getOperand(0), + DAG.getConstant(Idx, dl, TLI.getVectorIdxTy())); SDValue Sh = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT.getScalarType(), - Node->getOperand(1), DAG.getConstant(Idx, - TLI.getVectorIdxTy())); + Node->getOperand(1), + DAG.getConstant(Idx, dl, TLI.getVectorIdxTy())); Scalars.push_back(DAG.getNode(Node->getOpcode(), dl, VT.getScalarType(), Ex, Sh)); } @@ -4134,6 +4075,8 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { Node->getOpcode() == ISD::SETCC) { OVT = Node->getOperand(0).getSimpleValueType(); } + if (Node->getOpcode() == ISD::BR_CC) + OVT = Node->getOperand(2).getSimpleValueType(); MVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), OVT); SDLoc dl(Node); SDValue Tmp1, Tmp2, Tmp3; @@ -4151,16 +4094,16 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { if (Node->getOpcode() == ISD::CTTZ) { // FIXME: This should set a bit in the zero extended value instead. Tmp2 = DAG.getSetCC(dl, getSetCCResultType(NVT), - Tmp1, DAG.getConstant(NVT.getSizeInBits(), NVT), + Tmp1, DAG.getConstant(NVT.getSizeInBits(), dl, NVT), ISD::SETEQ); Tmp1 = DAG.getSelect(dl, NVT, Tmp2, - DAG.getConstant(OVT.getSizeInBits(), NVT), Tmp1); + DAG.getConstant(OVT.getSizeInBits(), dl, NVT), Tmp1); } else if (Node->getOpcode() == ISD::CTLZ || Node->getOpcode() == ISD::CTLZ_ZERO_UNDEF) { // Tmp1 = Tmp1 - (sizeinbits(NVT) - sizeinbits(Old VT)) Tmp1 = DAG.getNode(ISD::SUB, dl, NVT, Tmp1, DAG.getConstant(NVT.getSizeInBits() - - OVT.getSizeInBits(), NVT)); + OVT.getSizeInBits(), dl, NVT)); } Results.push_back(DAG.getNode(ISD::TRUNCATE, dl, OVT, Tmp1)); break; @@ -4169,7 +4112,8 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { Tmp1 = DAG.getNode(ISD::ZERO_EXTEND, dl, NVT, Node->getOperand(0)); Tmp1 = DAG.getNode(ISD::BSWAP, dl, NVT, Tmp1); Tmp1 = DAG.getNode(ISD::SRL, dl, NVT, Tmp1, - DAG.getConstant(DiffBits, TLI.getShiftAmountTy(NVT))); + DAG.getConstant(DiffBits, dl, + TLI.getShiftAmountTy(NVT))); Results.push_back(Tmp1); break; } @@ -4259,7 +4203,7 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { Tmp1 = DAG.getNode(TruncOp, dl, Node->getValueType(0), Tmp1); else Tmp1 = DAG.getNode(TruncOp, dl, Node->getValueType(0), Tmp1, - DAG.getIntPtrConstant(0)); + DAG.getIntPtrConstant(0, dl)); Results.push_back(Tmp1); break; } @@ -4289,27 +4233,74 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { Tmp1, Tmp2, Node->getOperand(2))); break; } + case ISD::BR_CC: { + unsigned ExtOp = ISD::FP_EXTEND; + if (NVT.isInteger()) { + ISD::CondCode CCCode = + cast(Node->getOperand(1))->get(); + ExtOp = isSignedIntSetCC(CCCode) ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND; + } + Tmp1 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(2)); + Tmp2 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(3)); + Results.push_back(DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), + Node->getOperand(0), Node->getOperand(1), + Tmp1, Tmp2, Node->getOperand(4))); + break; + } case ISD::FADD: case ISD::FSUB: case ISD::FMUL: case ISD::FDIV: case ISD::FREM: + case ISD::FMINNUM: + case ISD::FMAXNUM: + case ISD::FCOPYSIGN: case ISD::FPOW: { Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0)); Tmp2 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(1)); Tmp3 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1, Tmp2); Results.push_back(DAG.getNode(ISD::FP_ROUND, dl, OVT, - Tmp3, DAG.getIntPtrConstant(0))); + Tmp3, DAG.getIntPtrConstant(0, dl))); break; } - case ISD::FLOG2: - case ISD::FEXP2: + case ISD::FMA: { + Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0)); + Tmp2 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(1)); + Tmp3 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(2)); + Results.push_back( + DAG.getNode(ISD::FP_ROUND, dl, OVT, + DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1, Tmp2, Tmp3), + DAG.getIntPtrConstant(0, dl))); + break; + } + case ISD::FPOWI: { + Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0)); + Tmp2 = Node->getOperand(1); + Tmp3 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1, Tmp2); + Results.push_back(DAG.getNode(ISD::FP_ROUND, dl, OVT, + Tmp3, DAG.getIntPtrConstant(0, dl))); + break; + } + case ISD::FFLOOR: + case ISD::FCEIL: + case ISD::FRINT: + case ISD::FNEARBYINT: + case ISD::FROUND: + case ISD::FTRUNC: + case ISD::FNEG: + case ISD::FSQRT: + case ISD::FSIN: + case ISD::FCOS: case ISD::FLOG: - case ISD::FEXP: { + case ISD::FLOG2: + case ISD::FLOG10: + case ISD::FABS: + case ISD::FEXP: + case ISD::FEXP2: { Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0)); Tmp2 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1); Results.push_back(DAG.getNode(ISD::FP_ROUND, dl, OVT, - Tmp2, DAG.getIntPtrConstant(0))); + Tmp2, DAG.getIntPtrConstant(0, dl))); break; } } @@ -4319,8 +4310,7 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) { ReplaceNode(Node, Results.data()); } -// SelectionDAG::Legalize - This is the entry point for the file. -// +/// This is the entry point for the file. void SelectionDAG::Legalize() { AssignTopologicalOrder();