X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FSelectionDAG.cpp;h=1d7378b5dcd0504d97564e94dfc8ef3b20ab6c69;hb=3574eca1b02600bac4e625297f4ecf745f4c4f32;hp=dd33205c5de576ec20af8dfd11d3979af9e9d9de;hpb=9728ea2447e689ae6e884cf044f226b99c4a47da;p=oota-llvm.git diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp index dd33205c5de..1d7378b5dcd 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp @@ -14,23 +14,22 @@ #include "llvm/CodeGen/SelectionDAG.h" #include "SDNodeOrdering.h" #include "SDNodeDbgValue.h" +#include "llvm/CallingConv.h" #include "llvm/Constants.h" -#include "llvm/Analysis/DebugInfo.h" -#include "llvm/Analysis/ValueTracking.h" +#include "llvm/DebugInfo.h" +#include "llvm/DerivedTypes.h" #include "llvm/Function.h" #include "llvm/GlobalAlias.h" #include "llvm/GlobalVariable.h" #include "llvm/Intrinsics.h" -#include "llvm/DerivedTypes.h" +#include "llvm/Analysis/ValueTracking.h" #include "llvm/Assembly/Writer.h" -#include "llvm/CallingConv.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineModuleInfo.h" -#include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/Target/TargetRegisterInfo.h" -#include "llvm/Target/TargetData.h" +#include "llvm/DataLayout.h" #include "llvm/Target/TargetLowering.h" #include "llvm/Target/TargetSelectionDAGInfo.h" #include "llvm/Target/TargetOptions.h" @@ -63,6 +62,7 @@ static SDVTList makeVTList(const EVT *VTs, unsigned NumVTs) { static const fltSemantics *EVTToAPFloatSemantics(EVT VT) { switch (VT.getSimpleVT().SimpleTy) { default: llvm_unreachable("Unknown FP format"); + case MVT::f16: return &APFloat::IEEEhalf; case MVT::f32: return &APFloat::IEEEsingle; case MVT::f64: return &APFloat::IEEEdouble; case MVT::f80: return &APFloat::x87DoubleExtended; @@ -71,7 +71,9 @@ static const fltSemantics *EVTToAPFloatSemantics(EVT VT) { } } -SelectionDAG::DAGUpdateListener::~DAGUpdateListener() {} +// Default null implementations of the callbacks. +void SelectionDAG::DAGUpdateListener::NodeDeleted(SDNode*, SDNode*) {} +void SelectionDAG::DAGUpdateListener::NodeUpdated(SDNode*) {} //===----------------------------------------------------------------------===// // ConstantFPSDNode Class @@ -125,20 +127,27 @@ bool ISD::isBuildVectorAllOnes(const SDNode *N) { if (i == e) return false; // Do not accept build_vectors that aren't all constants or which have non-~0 - // elements. + // elements. We have to be a bit careful here, as the type of the constant + // may not be the same as the type of the vector elements due to type + // legalization (the elements are promoted to a legal type for the target and + // a vector of a type may be legal when the base element type is not). + // We only want to check enough bits to cover the vector elements, because + // we care if the resultant vector is all ones, not whether the individual + // constants are. SDValue NotZero = N->getOperand(i); - if (isa(NotZero)) { - if (!cast(NotZero)->isAllOnesValue()) + unsigned EltSize = N->getValueType(0).getVectorElementType().getSizeInBits(); + if (ConstantSDNode *CN = dyn_cast(NotZero)) { + if (CN->getAPIntValue().countTrailingOnes() < EltSize) return false; - } else if (isa(NotZero)) { - if (!cast(NotZero)->getValueAPF(). - bitcastToAPInt().isAllOnesValue()) + } else if (ConstantFPSDNode *CFPN = dyn_cast(NotZero)) { + if (CFPN->getValueAPF().bitcastToAPInt().countTrailingOnes() < EltSize) return false; } else return false; // Okay, we have at least one ~0 value, check to see if the rest match or are - // undefs. + // undefs. Even with the above element type twiddling, this should be OK, as + // the same type legalization should have applied to all the elements. for (++i; i != e; ++i) if (N->getOperand(i) != NotZero && N->getOperand(i).getOpcode() != ISD::UNDEF) @@ -168,11 +177,11 @@ bool ISD::isBuildVectorAllZeros(const SDNode *N) { // Do not accept build_vectors that aren't all constants or which have non-0 // elements. SDValue Zero = N->getOperand(i); - if (isa(Zero)) { - if (!cast(Zero)->isNullValue()) + if (ConstantSDNode *CN = dyn_cast(Zero)) { + if (!CN->isNullValue()) return false; - } else if (isa(Zero)) { - if (!cast(Zero)->getValueAPF().isPosZero()) + } else if (ConstantFPSDNode *CFPN = dyn_cast(Zero)) { + if (!CFPN->getValueAPF().isPosZero()) return false; } else return false; @@ -208,6 +217,22 @@ bool ISD::isScalarToVector(const SDNode *N) { return true; } +/// allOperandsUndef - Return true if the node has at least one operand +/// and all operands of the specified node are ISD::UNDEF. +bool ISD::allOperandsUndef(const SDNode *N) { + // Return false if the node has no operands. + // This is "logically inconsistent" with the definition of "all" but + // is probably the desired behavior. + if (N->getNumOperands() == 0) + return false; + + for (unsigned i = 0, e = N->getNumOperands(); i != e ; ++i) + if (N->getOperand(i).getOpcode() != ISD::UNDEF) + return false; + + return true; +} + /// getSetCCSwappedOperands - Return the operation corresponding to (Y op X) /// when given the operation for (X op Y). ISD::CondCode ISD::getSetCCSwappedOperands(ISD::CondCode Operation) { @@ -376,6 +401,7 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) { ID.AddPointer(GA->getGlobal()); ID.AddInteger(GA->getOffset()); ID.AddInteger(GA->getTargetFlags()); + ID.AddInteger(GA->getAddressSpace()); break; } case ISD::BasicBlock: @@ -384,7 +410,9 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) { case ISD::Register: ID.AddInteger(cast(N)->getReg()); break; - + case ISD::RegisterMask: + ID.AddPointer(cast(N)->getRegMask()); + break; case ISD::SRCVALUE: ID.AddPointer(cast(N)->getValue()); break; @@ -403,22 +431,31 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) { ID.AddInteger(CP->getAlignment()); ID.AddInteger(CP->getOffset()); if (CP->isMachineConstantPoolEntry()) - CP->getMachineCPVal()->AddSelectionDAGCSEId(ID); + CP->getMachineCPVal()->addSelectionDAGCSEId(ID); else ID.AddPointer(CP->getConstVal()); ID.AddInteger(CP->getTargetFlags()); break; } + case ISD::TargetIndex: { + const TargetIndexSDNode *TI = cast(N); + ID.AddInteger(TI->getIndex()); + ID.AddInteger(TI->getOffset()); + ID.AddInteger(TI->getTargetFlags()); + break; + } case ISD::LOAD: { const LoadSDNode *LD = cast(N); ID.AddInteger(LD->getMemoryVT().getRawBits()); ID.AddInteger(LD->getRawSubclassData()); + ID.AddInteger(LD->getPointerInfo().getAddrSpace()); break; } case ISD::STORE: { const StoreSDNode *ST = cast(N); ID.AddInteger(ST->getMemoryVT().getRawBits()); ID.AddInteger(ST->getRawSubclassData()); + ID.AddInteger(ST->getPointerInfo().getAddrSpace()); break; } case ISD::ATOMIC_CMP_SWAP: @@ -432,10 +469,18 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) { case ISD::ATOMIC_LOAD_MIN: case ISD::ATOMIC_LOAD_MAX: case ISD::ATOMIC_LOAD_UMIN: - case ISD::ATOMIC_LOAD_UMAX: { + case ISD::ATOMIC_LOAD_UMAX: + case ISD::ATOMIC_LOAD: + case ISD::ATOMIC_STORE: { const AtomicSDNode *AT = cast(N); ID.AddInteger(AT->getMemoryVT().getRawBits()); ID.AddInteger(AT->getRawSubclassData()); + ID.AddInteger(AT->getPointerInfo().getAddrSpace()); + break; + } + case ISD::PREFETCH: { + const MemSDNode *PF = cast(N); + ID.AddInteger(PF->getPointerInfo().getAddrSpace()); break; } case ISD::VECTOR_SHUFFLE: { @@ -447,11 +492,17 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) { } case ISD::TargetBlockAddress: case ISD::BlockAddress: { - ID.AddPointer(cast(N)->getBlockAddress()); - ID.AddInteger(cast(N)->getTargetFlags()); + const BlockAddressSDNode *BA = cast(N); + ID.AddPointer(BA->getBlockAddress()); + ID.AddInteger(BA->getOffset()); + ID.AddInteger(BA->getTargetFlags()); break; } } // end switch (N->getOpcode()) + + // Target specific memory nodes could also have address spaces to check. + if (N->isTargetMemoryOpcode()) + ID.AddInteger(cast(N)->getPointerInfo().getAddrSpace()); } /// AddNodeIDNode - Generic routine for adding a nodes info to the NodeID @@ -473,7 +524,7 @@ static void AddNodeIDNode(FoldingSetNodeID &ID, const SDNode *N) { /// static inline unsigned encodeMemSDNodeFlags(int ConvType, ISD::MemIndexedMode AM, bool isVolatile, - bool isNonTemporal) { + bool isNonTemporal, bool isInvariant) { assert((ConvType & 3) == ConvType && "ConvType may not require more than 2 bits!"); assert((AM & 7) == AM && @@ -481,7 +532,8 @@ encodeMemSDNodeFlags(int ConvType, ISD::MemIndexedMode AM, bool isVolatile, return ConvType | (AM << 2) | (isVolatile << 5) | - (isNonTemporal << 6); + (isNonTemporal << 6) | + (isInvariant << 7); } //===----------------------------------------------------------------------===// @@ -530,16 +582,15 @@ void SelectionDAG::RemoveDeadNodes() { /// RemoveDeadNodes - This method deletes the unreachable nodes in the /// given list, and any nodes that become unreachable as a result. -void SelectionDAG::RemoveDeadNodes(SmallVectorImpl &DeadNodes, - DAGUpdateListener *UpdateListener) { +void SelectionDAG::RemoveDeadNodes(SmallVectorImpl &DeadNodes) { // Process the worklist, deleting the nodes and adding their uses to the // worklist. while (!DeadNodes.empty()) { SDNode *N = DeadNodes.pop_back_val(); - if (UpdateListener) - UpdateListener->NodeDeleted(N, 0); + for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next) + DUL->NodeDeleted(N, 0); // Take the node out of the appropriate CSE map. RemoveNodeFromCSEMaps(N); @@ -560,9 +611,15 @@ void SelectionDAG::RemoveDeadNodes(SmallVectorImpl &DeadNodes, } } -void SelectionDAG::RemoveDeadNode(SDNode *N, DAGUpdateListener *UpdateListener){ +void SelectionDAG::RemoveDeadNode(SDNode *N){ SmallVector DeadNodes(1, N); - RemoveDeadNodes(DeadNodes, UpdateListener); + + // Create a dummy node that adds a reference to the root node, preventing + // it from being deleted. (This matters if the root is an operand of the + // dead node.) + HandleSDNode Dummy(getRoot()); + + RemoveDeadNodes(DeadNodes); } void SelectionDAG::DeleteNode(SDNode *N) { @@ -598,7 +655,7 @@ void SelectionDAG::DeallocateNode(SDNode *N) { Ordering->remove(N); // If any of the SDDbgValue nodes refer to this SDNode, invalidate them. - SmallVector &DbgVals = DbgInfo->getSDDbgValues(N); + ArrayRef DbgVals = DbgInfo->getSDDbgValues(N); for (unsigned i = 0, e = DbgVals.size(); i != e; ++i) DbgVals[i]->setIsInvalidated(); } @@ -664,8 +721,7 @@ bool SelectionDAG::RemoveNodeFromCSEMaps(SDNode *N) { /// node. This transfer can potentially trigger recursive merging. /// void -SelectionDAG::AddModifiedNodeToCSEMaps(SDNode *N, - DAGUpdateListener *UpdateListener) { +SelectionDAG::AddModifiedNodeToCSEMaps(SDNode *N) { // For node types that aren't CSE'd, just act as if no identical node // already exists. if (!doNotCSE(N)) { @@ -674,20 +730,19 @@ SelectionDAG::AddModifiedNodeToCSEMaps(SDNode *N, // If there was already an existing matching node, use ReplaceAllUsesWith // to replace the dead one with the existing one. This can cause // recursive merging of other unrelated nodes down the line. - ReplaceAllUsesWith(N, Existing, UpdateListener); + ReplaceAllUsesWith(N, Existing); - // N is now dead. Inform the listener if it exists and delete it. - if (UpdateListener) - UpdateListener->NodeDeleted(N, Existing); + // N is now dead. Inform the listeners and delete it. + for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next) + DUL->NodeDeleted(N, Existing); DeleteNodeNotInCSEMaps(N); return; } } - // If the node doesn't already exist, we updated it. Inform a listener if - // it exists. - if (UpdateListener) - UpdateListener->NodeUpdated(N); + // If the node doesn't already exist, we updated it. Inform listeners. + for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next) + DUL->NodeUpdated(N); } /// FindModifiedNodeSlot - Find a slot for the specified node if its operands @@ -769,11 +824,14 @@ static void VerifyNodeCommon(SDNode *N) { assert(N->getNumOperands() == N->getValueType(0).getVectorNumElements() && "Wrong number of operands!"); EVT EltVT = N->getValueType(0).getVectorElementType(); - for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) + for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) { assert((I->getValueType() == EltVT || (EltVT.isInteger() && I->getValueType().isInteger() && EltVT.bitsLE(I->getValueType()))) && "Wrong operand type!"); + assert(I->getValueType() == N->getOperand(0).getValueType() && + "Operands must all have the same type"); + } break; } } @@ -821,18 +879,18 @@ static void VerifyMachineNode(SDNode *N) { /// given type. /// unsigned SelectionDAG::getEVTAlignment(EVT VT) const { - const Type *Ty = VT == MVT::iPTR ? + Type *Ty = VT == MVT::iPTR ? PointerType::get(Type::getInt8Ty(*getContext()), 0) : VT.getTypeForEVT(*getContext()); - return TLI.getTargetData()->getABITypeAlignment(Ty); + return TLI.getDataLayout()->getABITypeAlignment(Ty); } // EntryNode could meaningfully have debug info if we can find it... -SelectionDAG::SelectionDAG(const TargetMachine &tm) +SelectionDAG::SelectionDAG(const TargetMachine &tm, CodeGenOpt::Level OL) : TM(tm), TLI(*tm.getTargetLowering()), TSI(*tm.getSelectionDAGInfo()), - EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)), - Root(getEntryNode()), Ordering(0) { + OptLevel(OL), EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)), + Root(getEntryNode()), Ordering(0), UpdateListeners(0) { AllNodes.push_back(&EntryNode); Ordering = new SDNodeOrdering(); DbgInfo = new SDDbgInfo(); @@ -844,6 +902,7 @@ void SelectionDAG::init(MachineFunction &mf) { } SelectionDAG::~SelectionDAG() { + assert(!UpdateListeners && "Dangling registered DAGUpdateListeners"); allnodes_clear(); delete Ordering; delete DbgInfo; @@ -876,6 +935,12 @@ void SelectionDAG::clear() { DbgInfo->clear(); } +SDValue SelectionDAG::getAnyExtOrTrunc(SDValue Op, DebugLoc DL, EVT VT) { + return VT.bitsGT(Op.getValueType()) ? + getNode(ISD::ANY_EXTEND, DL, VT, Op) : + getNode(ISD::TRUNCATE, DL, VT, Op); +} + SDValue SelectionDAG::getSExtOrTrunc(SDValue Op, DebugLoc DL, EVT VT) { return VT.bitsGT(Op.getValueType()) ? getNode(ISD::SIGN_EXTEND, DL, VT, Op) : @@ -925,13 +990,25 @@ SDValue SelectionDAG::getConstant(const ConstantInt &Val, EVT VT, bool isT) { assert(VT.isInteger() && "Cannot create FP integer constant!"); EVT EltVT = VT.getScalarType(); - assert(Val.getBitWidth() == EltVT.getSizeInBits() && - "APInt size does not match type size!"); + const ConstantInt *Elt = &Val; + + // In some cases the vector type is legal but the element type is illegal and + // needs to be promoted, for example v8i8 on ARM. In this case, promote the + // inserted value (the type does not need to match the vector element type). + // Any extra bits introduced will be truncated away. + if (VT.isVector() && TLI.getTypeAction(*getContext(), EltVT) == + TargetLowering::TypePromoteInteger) { + EltVT = TLI.getTypeToTransformTo(*getContext(), EltVT); + APInt NewVal = Elt->getValue().zext(EltVT.getSizeInBits()); + Elt = ConstantInt::get(*getContext(), NewVal); + } + assert(Elt->getBitWidth() == EltVT.getSizeInBits() && + "APInt size does not match type size!"); unsigned Opc = isT ? ISD::TargetConstant : ISD::Constant; FoldingSetNodeID ID; AddNodeIDNode(ID, Opc, getVTList(EltVT), 0, 0); - ID.AddPointer(&Val); + ID.AddPointer(Elt); void *IP = 0; SDNode *N = NULL; if ((N = CSEMap.FindNodeOrInsertPos(ID, IP))) @@ -939,7 +1016,7 @@ SDValue SelectionDAG::getConstant(const ConstantInt &Val, EVT VT, bool isT) { return SDValue(N, 0); if (!N) { - N = new (NodeAllocator) ConstantSDNode(isT, &Val, EltVT); + N = new (NodeAllocator) ConstantSDNode(isT, Elt, EltVT); CSEMap.InsertNode(N, IP); AllNodes.push_back(N); } @@ -1002,16 +1079,14 @@ SDValue SelectionDAG::getConstantFP(double Val, EVT VT, bool isTarget) { return getConstantFP(APFloat((float)Val), VT, isTarget); else if (EltVT==MVT::f64) return getConstantFP(APFloat(Val), VT, isTarget); - else if (EltVT==MVT::f80 || EltVT==MVT::f128) { + else if (EltVT==MVT::f80 || EltVT==MVT::f128 || EltVT==MVT::f16) { bool ignored; APFloat apf = APFloat(Val); apf.convert(*EVTToAPFloatSemantics(EltVT), APFloat::rmNearestTiesToEven, &ignored); return getConstantFP(apf, VT, isTarget); - } else { - assert(0 && "Unsupported type in getConstantFP"); - return SDValue(); - } + } else + llvm_unreachable("Unsupported type in getConstantFP"); } SDValue SelectionDAG::getGlobalAddress(const GlobalValue *GV, DebugLoc DL, @@ -1022,10 +1097,9 @@ SDValue SelectionDAG::getGlobalAddress(const GlobalValue *GV, DebugLoc DL, "Cannot set target flags on target-independent globals"); // Truncate (with sign-extension) the offset value to the pointer size. - EVT PTy = TLI.getPointerTy(); - unsigned BitWidth = PTy.getSizeInBits(); + unsigned BitWidth = TLI.getPointerTy().getSizeInBits(); if (BitWidth < 64) - Offset = (Offset << (64 - BitWidth) >> (64 - BitWidth)); + Offset = SignExtend64(Offset, BitWidth); const GlobalVariable *GVar = dyn_cast(GV); if (!GVar) { @@ -1045,6 +1119,7 @@ SDValue SelectionDAG::getGlobalAddress(const GlobalValue *GV, DebugLoc DL, ID.AddPointer(GV); ID.AddInteger(Offset); ID.AddInteger(TargetFlags); + ID.AddInteger(GV->getType()->getAddressSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) return SDValue(E, 0); @@ -1098,7 +1173,7 @@ SDValue SelectionDAG::getConstantPool(const Constant *C, EVT VT, assert((TargetFlags == 0 || isTarget) && "Cannot set target flags on target-independent globals"); if (Alignment == 0) - Alignment = TLI.getTargetData()->getPrefTypeAlignment(C->getType()); + Alignment = TLI.getDataLayout()->getPrefTypeAlignment(C->getType()); unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool; FoldingSetNodeID ID; AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0); @@ -1125,13 +1200,13 @@ SDValue SelectionDAG::getConstantPool(MachineConstantPoolValue *C, EVT VT, assert((TargetFlags == 0 || isTarget) && "Cannot set target flags on target-independent globals"); if (Alignment == 0) - Alignment = TLI.getTargetData()->getPrefTypeAlignment(C->getType()); + Alignment = TLI.getDataLayout()->getPrefTypeAlignment(C->getType()); unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool; FoldingSetNodeID ID; AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0); ID.AddInteger(Alignment); ID.AddInteger(Offset); - C->AddSelectionDAGCSEId(ID); + C->addSelectionDAGCSEId(ID); ID.AddInteger(TargetFlags); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) @@ -1144,6 +1219,24 @@ SDValue SelectionDAG::getConstantPool(MachineConstantPoolValue *C, EVT VT, return SDValue(N, 0); } +SDValue SelectionDAG::getTargetIndex(int Index, EVT VT, int64_t Offset, + unsigned char TargetFlags) { + FoldingSetNodeID ID; + AddNodeIDNode(ID, ISD::TargetIndex, getVTList(VT), 0, 0); + ID.AddInteger(Index); + ID.AddInteger(Offset); + ID.AddInteger(TargetFlags); + void *IP = 0; + if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) + return SDValue(E, 0); + + SDNode *N = new (NodeAllocator) TargetIndexSDNode(Index, VT, Offset, + TargetFlags); + CSEMap.InsertNode(N, IP); + AllNodes.push_back(N); + return SDValue(N, 0); +} + SDValue SelectionDAG::getBasicBlock(MachineBasicBlock *MBB) { FoldingSetNodeID ID; AddNodeIDNode(ID, ISD::BasicBlock, getVTList(MVT::Other), 0, 0); @@ -1346,6 +1439,20 @@ SDValue SelectionDAG::getRegister(unsigned RegNo, EVT VT) { return SDValue(N, 0); } +SDValue SelectionDAG::getRegisterMask(const uint32_t *RegMask) { + FoldingSetNodeID ID; + AddNodeIDNode(ID, ISD::RegisterMask, getVTList(MVT::Untyped), 0, 0); + ID.AddPointer(RegMask); + void *IP = 0; + if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) + return SDValue(E, 0); + + SDNode *N = new (NodeAllocator) RegisterMaskSDNode(RegMask); + CSEMap.InsertNode(N, IP); + AllNodes.push_back(N); + return SDValue(N, 0); +} + SDValue SelectionDAG::getEHLabel(DebugLoc dl, SDValue Root, MCSymbol *Label) { FoldingSetNodeID ID; SDValue Ops[] = { Root }; @@ -1363,6 +1470,7 @@ SDValue SelectionDAG::getEHLabel(DebugLoc dl, SDValue Root, MCSymbol *Label) { SDValue SelectionDAG::getBlockAddress(const BlockAddress *BA, EVT VT, + int64_t Offset, bool isTarget, unsigned char TargetFlags) { unsigned Opc = isTarget ? ISD::TargetBlockAddress : ISD::BlockAddress; @@ -1370,12 +1478,14 @@ SDValue SelectionDAG::getBlockAddress(const BlockAddress *BA, EVT VT, FoldingSetNodeID ID; AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0); ID.AddPointer(BA); + ID.AddInteger(Offset); ID.AddInteger(TargetFlags); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) return SDValue(E, 0); - SDNode *N = new (NodeAllocator) BlockAddressSDNode(Opc, VT, BA, TargetFlags); + SDNode *N = new (NodeAllocator) BlockAddressSDNode(Opc, VT, BA, Offset, + TargetFlags); CSEMap.InsertNode(N, IP); AllNodes.push_back(N); return SDValue(N, 0); @@ -1432,9 +1542,9 @@ SDValue SelectionDAG::getShiftAmountOperand(EVT LHSTy, SDValue Op) { SDValue SelectionDAG::CreateStackTemporary(EVT VT, unsigned minAlign) { MachineFrameInfo *FrameInfo = getMachineFunction().getFrameInfo(); unsigned ByteSize = VT.getStoreSize(); - const Type *Ty = VT.getTypeForEVT(*getContext()); + Type *Ty = VT.getTypeForEVT(*getContext()); unsigned StackAlign = - std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty), minAlign); + std::max((unsigned)TLI.getDataLayout()->getPrefTypeAlignment(Ty), minAlign); int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false); return getFrameIndex(FrameIdx, TLI.getPointerTy()); @@ -1445,9 +1555,9 @@ SDValue SelectionDAG::CreateStackTemporary(EVT VT, unsigned minAlign) { SDValue SelectionDAG::CreateStackTemporary(EVT VT1, EVT VT2) { unsigned Bytes = std::max(VT1.getStoreSizeInBits(), VT2.getStoreSizeInBits())/8; - const Type *Ty1 = VT1.getTypeForEVT(*getContext()); - const Type *Ty2 = VT2.getTypeForEVT(*getContext()); - const TargetData *TD = TLI.getTargetData(); + Type *Ty1 = VT1.getTypeForEVT(*getContext()); + Type *Ty2 = VT2.getTypeForEVT(*getContext()); + const DataLayout *TD = TLI.getDataLayout(); unsigned Align = std::max(TD->getPrefTypeAlignment(Ty1), TD->getPrefTypeAlignment(Ty2)); @@ -1575,7 +1685,7 @@ bool SelectionDAG::SignBitIsZero(SDValue Op, unsigned Depth) const { bool SelectionDAG::MaskedValueIsZero(SDValue Op, const APInt &Mask, unsigned Depth) const { APInt KnownZero, KnownOne; - ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth); + ComputeMaskedBits(Op, KnownZero, KnownOne, Depth); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); return (KnownZero & Mask) == Mask; } @@ -1584,15 +1694,12 @@ bool SelectionDAG::MaskedValueIsZero(SDValue Op, const APInt &Mask, /// known to be either zero or one and return them in the KnownZero/KnownOne /// bitsets. This code only analyzes bits in Mask, in order to short-circuit /// processing. -void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, - APInt &KnownZero, APInt &KnownOne, - unsigned Depth) const { - unsigned BitWidth = Mask.getBitWidth(); - assert(BitWidth == Op.getValueType().getScalarType().getSizeInBits() && - "Mask size mismatches value type size!"); +void SelectionDAG::ComputeMaskedBits(SDValue Op, APInt &KnownZero, + APInt &KnownOne, unsigned Depth) const { + unsigned BitWidth = Op.getValueType().getScalarType().getSizeInBits(); KnownZero = KnownOne = APInt(BitWidth, 0); // Don't know anything. - if (Depth == 6 || Mask == 0) + if (Depth == 6) return; // Limit search depth. APInt KnownZero2, KnownOne2; @@ -1600,14 +1707,13 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, switch (Op.getOpcode()) { case ISD::Constant: // We know all of the bits for a constant! - KnownOne = cast(Op)->getAPIntValue() & Mask; - KnownZero = ~KnownOne & Mask; + KnownOne = cast(Op)->getAPIntValue(); + KnownZero = ~KnownOne; return; case ISD::AND: // If either the LHS or the RHS are Zero, the result is zero. - ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownZero, - KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1617,9 +1723,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, KnownZero |= KnownZero2; return; case ISD::OR: - ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownOne, - KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1629,8 +1734,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, KnownOne |= KnownOne2; return; case ISD::XOR: { - ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1642,9 +1747,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, return; } case ISD::MUL: { - APInt Mask2 = APInt::getAllOnesValue(BitWidth); - ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1663,33 +1767,29 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, LeadZ = std::min(LeadZ, BitWidth); KnownZero = APInt::getLowBitsSet(BitWidth, TrailZ) | APInt::getHighBitsSet(BitWidth, LeadZ); - KnownZero &= Mask; return; } case ISD::UDIV: { // For the purposes of computing leading zeros we can conservatively // treat a udiv as a logical right shift by the power of 2 known to // be less than the denominator. - APInt AllOnes = APInt::getAllOnesValue(BitWidth); - ComputeMaskedBits(Op.getOperand(0), - AllOnes, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); unsigned LeadZ = KnownZero2.countLeadingOnes(); KnownOne2.clearAllBits(); KnownZero2.clearAllBits(); - ComputeMaskedBits(Op.getOperand(1), - AllOnes, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1); unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros(); if (RHSUnknownLeadingOnes != BitWidth) LeadZ = std::min(BitWidth, LeadZ + BitWidth - RHSUnknownLeadingOnes - 1); - KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ) & Mask; + KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ); return; } case ISD::SELECT: - ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(2), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1698,8 +1798,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, KnownZero &= KnownZero2; return; case ISD::SELECT_CC: - ComputeMaskedBits(Op.getOperand(3), Mask, KnownZero, KnownOne, Depth+1); - ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(3), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(2), KnownZero2, KnownOne2, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); @@ -1718,8 +1818,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, // The boolean result conforms to getBooleanContents. Fall through. case ISD::SETCC: // If we know the result of a setcc has the top bits zero, use this info. - if (TLI.getBooleanContents() == TargetLowering::ZeroOrOneBooleanContent && - BitWidth > 1) + if (TLI.getBooleanContents(Op.getValueType().isVector()) == + TargetLowering::ZeroOrOneBooleanContent && BitWidth > 1) KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1); return; case ISD::SHL: @@ -1731,8 +1831,7 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if (ShAmt >= BitWidth) return; - ComputeMaskedBits(Op.getOperand(0), Mask.lshr(ShAmt), - KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); KnownZero <<= ShAmt; KnownOne <<= ShAmt; @@ -1749,13 +1848,12 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if (ShAmt >= BitWidth) return; - ComputeMaskedBits(Op.getOperand(0), (Mask << ShAmt), - KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); KnownZero = KnownZero.lshr(ShAmt); KnownOne = KnownOne.lshr(ShAmt); - APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask; + APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt); KnownZero |= HighBits; // High bits known zero. } return; @@ -1767,15 +1865,11 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if (ShAmt >= BitWidth) return; - APInt InDemandedMask = (Mask << ShAmt); // If any of the demanded bits are produced by the sign extension, we also // demand the input sign bit. - APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask; - if (HighBits.getBoolValue()) - InDemandedMask |= APInt::getSignBit(BitWidth); + APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt); - ComputeMaskedBits(Op.getOperand(0), InDemandedMask, KnownZero, KnownOne, - Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); KnownZero = KnownZero.lshr(ShAmt); KnownOne = KnownOne.lshr(ShAmt); @@ -1797,10 +1891,10 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, // Sign extension. Compute the demanded bits in the result that are not // present in the input. - APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - EBits) & Mask; + APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - EBits); APInt InSignBit = APInt::getSignBit(EBits); - APInt InputDemandedBits = Mask & APInt::getLowBitsSet(BitWidth, EBits); + APInt InputDemandedBits = APInt::getLowBitsSet(BitWidth, EBits); // If the sign extended bits are demanded, we know that the sign // bit is demanded. @@ -1808,8 +1902,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if (NewBits.getBoolValue()) InputDemandedBits |= InSignBit; - ComputeMaskedBits(Op.getOperand(0), InputDemandedBits, - KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); + KnownOne &= InputDemandedBits; + KnownZero &= InputDemandedBits; assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); // If the sign bit of the input is known set or clear, then we know the @@ -1827,7 +1922,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, return; } case ISD::CTTZ: + case ISD::CTTZ_ZERO_UNDEF: case ISD::CTLZ: + case ISD::CTLZ_ZERO_UNDEF: case ISD::CTPOP: { unsigned LowBits = Log2_32(BitWidth)+1; KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits); @@ -1835,22 +1932,23 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, return; } case ISD::LOAD: { + LoadSDNode *LD = cast(Op); if (ISD::isZEXTLoad(Op.getNode())) { - LoadSDNode *LD = cast(Op); EVT VT = LD->getMemoryVT(); unsigned MemBits = VT.getScalarType().getSizeInBits(); - KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits) & Mask; + KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits); + } else if (const MDNode *Ranges = LD->getRanges()) { + computeMaskedBitsLoad(*Ranges, KnownZero); } return; } case ISD::ZERO_EXTEND: { EVT InVT = Op.getOperand(0).getValueType(); unsigned InBits = InVT.getScalarType().getSizeInBits(); - APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask; - APInt InMask = Mask.trunc(InBits); + APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits); KnownZero = KnownZero.trunc(InBits); KnownOne = KnownOne.trunc(InBits); - ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); KnownZero = KnownZero.zext(BitWidth); KnownOne = KnownOne.zext(BitWidth); KnownZero |= NewBits; @@ -1860,17 +1958,11 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, EVT InVT = Op.getOperand(0).getValueType(); unsigned InBits = InVT.getScalarType().getSizeInBits(); APInt InSignBit = APInt::getSignBit(InBits); - APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask; - APInt InMask = Mask.trunc(InBits); - - // If any of the sign extended bits are demanded, we know that the sign - // bit is demanded. Temporarily set this bit in the mask for our callee. - if (NewBits.getBoolValue()) - InMask |= InSignBit; + APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits); KnownZero = KnownZero.trunc(InBits); KnownOne = KnownOne.trunc(InBits); - ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); // Note if the sign bit is known to be zero or one. bool SignBitKnownZero = KnownZero.isNegative(); @@ -1878,13 +1970,6 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, assert(!(SignBitKnownZero && SignBitKnownOne) && "Sign bit can't be known to be both zero and one!"); - // If the sign bit wasn't actually demanded by our caller, we don't - // want it set in the KnownZero and KnownOne result values. Reset the - // mask and reapply it to the result values. - InMask = Mask.trunc(InBits); - KnownZero &= InMask; - KnownOne &= InMask; - KnownZero = KnownZero.zext(BitWidth); KnownOne = KnownOne.zext(BitWidth); @@ -1898,10 +1983,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, case ISD::ANY_EXTEND: { EVT InVT = Op.getOperand(0).getValueType(); unsigned InBits = InVT.getScalarType().getSizeInBits(); - APInt InMask = Mask.trunc(InBits); KnownZero = KnownZero.trunc(InBits); KnownOne = KnownOne.trunc(InBits); - ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); KnownZero = KnownZero.zext(BitWidth); KnownOne = KnownOne.zext(BitWidth); return; @@ -1909,10 +1993,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, case ISD::TRUNCATE: { EVT InVT = Op.getOperand(0).getValueType(); unsigned InBits = InVT.getScalarType().getSizeInBits(); - APInt InMask = Mask.zext(InBits); KnownZero = KnownZero.zext(InBits); KnownOne = KnownOne.zext(InBits); - ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); KnownZero = KnownZero.trunc(BitWidth); KnownOne = KnownOne.trunc(BitWidth); @@ -1921,9 +2004,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, case ISD::AssertZext: { EVT VT = cast(Op.getOperand(1))->getVT(); APInt InMask = APInt::getLowBitsSet(BitWidth, VT.getSizeInBits()); - ComputeMaskedBits(Op.getOperand(0), Mask & InMask, KnownZero, - KnownOne, Depth+1); - KnownZero |= (~InMask) & Mask; + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); + KnownZero |= (~InMask); + KnownOne &= (~KnownZero); return; } case ISD::FGETSIGN: @@ -1940,8 +2023,7 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, unsigned NLZ = (CLHS->getAPIntValue()+1).countLeadingZeros(); // NLZ can't be BitWidth with no sign bit APInt MaskV = APInt::getHighBitsSet(BitWidth, NLZ+1); - ComputeMaskedBits(Op.getOperand(1), MaskV, KnownZero2, KnownOne2, - Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1); // If all of the MaskV bits are known to be zero, then we know the // output top bits are zero, because we now know that the output is @@ -1949,7 +2031,7 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if ((KnownZero2 & MaskV) == MaskV) { unsigned NLZ2 = CLHS->getAPIntValue().countLeadingZeros(); // Top bits known zero. - KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2) & Mask; + KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2); } } } @@ -1960,13 +2042,11 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, // Output known-0 bits are known if clear or set in both the low clear bits // common to both LHS & RHS. For example, 8+(X<<3) is known to have the // low 3 bits clear. - APInt Mask2 = APInt::getLowBitsSet(BitWidth, - BitWidth - Mask.countLeadingZeros()); - ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); unsigned KnownZeroOut = KnownZero2.countTrailingOnes(); - ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero2, KnownOne2, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1); assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); KnownZeroOut = std::min(KnownZeroOut, KnownZero2.countTrailingOnes()); @@ -1990,7 +2070,7 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, if (RA.isPowerOf2()) { APInt LowBits = RA - 1; APInt Mask2 = LowBits | APInt::getSignBit(BitWidth); - ComputeMaskedBits(Op.getOperand(0), Mask2,KnownZero2,KnownOne2,Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero2,KnownOne2,Depth+1); // The low bits of the first operand are unchanged by the srem. KnownZero = KnownZero2 & LowBits; @@ -2005,10 +2085,6 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, // the upper bits are all one. if (KnownOne2[BitWidth-1] && ((KnownOne2 & LowBits) != 0)) KnownOne |= ~LowBits; - - KnownZero &= Mask; - KnownOne &= Mask; - assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?"); } } @@ -2018,9 +2094,8 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, const APInt &RA = Rem->getAPIntValue(); if (RA.isPowerOf2()) { APInt LowBits = (RA - 1); - APInt Mask2 = LowBits & Mask; - KnownZero |= ~LowBits & Mask; - ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero, KnownOne,Depth+1); + KnownZero |= ~LowBits; + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne,Depth+1); assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?"); break; } @@ -2028,16 +2103,13 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, // Since the result is less than or equal to either operand, any leading // zero bits in either operand must also exist in the result. - APInt AllOnes = APInt::getAllOnesValue(BitWidth); - ComputeMaskedBits(Op.getOperand(0), AllOnes, KnownZero, KnownOne, - Depth+1); - ComputeMaskedBits(Op.getOperand(1), AllOnes, KnownZero2, KnownOne2, - Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1); uint32_t Leaders = std::max(KnownZero.countLeadingOnes(), KnownZero2.countLeadingOnes()); KnownOne.clearAllBits(); - KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & Mask; + KnownZero = APInt::getHighBitsSet(BitWidth, Leaders); return; } case ISD::FrameIndex: @@ -2050,14 +2122,14 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask, break; default: - // Allow the target to implement this method for its nodes. - if (Op.getOpcode() >= ISD::BUILTIN_OP_END) { + if (Op.getOpcode() < ISD::BUILTIN_OP_END) + break; + // Fallthrough case ISD::INTRINSIC_WO_CHAIN: case ISD::INTRINSIC_W_CHAIN: case ISD::INTRINSIC_VOID: - TLI.computeMaskedBitsForTargetNode(Op, Mask, KnownZero, KnownOne, *this, - Depth); - } + // Allow the target to implement this method for its nodes. + TLI.computeMaskedBitsForTargetNode(Op, KnownZero, KnownOne, *this, Depth); return; } } @@ -2152,7 +2224,7 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ // The boolean result conforms to getBooleanContents. Fall through. case ISD::SETCC: // If setcc returns 0/-1, all bits are sign bits. - if (TLI.getBooleanContents() == + if (TLI.getBooleanContents(Op.getValueType().isVector()) == TargetLowering::ZeroOrNegativeOneBooleanContent) return VTBits; break; @@ -2181,12 +2253,11 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ if (ConstantSDNode *CRHS = dyn_cast(Op.getOperand(1))) if (CRHS->isAllOnesValue()) { APInt KnownZero, KnownOne; - APInt Mask = APInt::getAllOnesValue(VTBits); - ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1); // If the input is known to be 0 or 1, the output is 0/-1, which is all // sign bits set. - if ((KnownZero | APInt(VTBits, 1)) == Mask) + if ((KnownZero | APInt(VTBits, 1)).isAllOnesValue()) return VTBits; // If we are subtracting one from a positive number, there is no carry @@ -2197,8 +2268,7 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1); if (Tmp2 == 1) return 1; - return std::min(Tmp, Tmp2)-1; - break; + return std::min(Tmp, Tmp2)-1; case ISD::SUB: Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1); @@ -2208,11 +2278,10 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ if (ConstantSDNode *CLHS = dyn_cast(Op.getOperand(0))) if (CLHS->isNullValue()) { APInt KnownZero, KnownOne; - APInt Mask = APInt::getAllOnesValue(VTBits); - ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1); + ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1); // If the input is known to be 0 or 1, the output is 0/-1, which is all // sign bits set. - if ((KnownZero | APInt(VTBits, 1)) == Mask) + if ((KnownZero | APInt(VTBits, 1)).isAllOnesValue()) return VTBits; // If the input is known to be positive (the sign bit is known clear), @@ -2227,8 +2296,7 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ // is, at worst, one more bit than the inputs. Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1); if (Tmp == 1) return 1; // Early out. - return std::min(Tmp, Tmp2)-1; - break; + return std::min(Tmp, Tmp2)-1; case ISD::TRUNCATE: // FIXME: it's tricky to do anything useful for this, but it is an important // case for targets like X86. @@ -2236,8 +2304,7 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ } // Handle LOADX separately here. EXTLOAD case will fallthrough. - if (Op.getOpcode() == ISD::LOAD) { - LoadSDNode *LD = cast(Op); + if (LoadSDNode *LD = dyn_cast(Op)) { unsigned ExtType = LD->getExtensionType(); switch (ExtType) { default: break; @@ -2262,9 +2329,9 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{ // Finally, if we can prove that the top bits of the result are 0's or 1's, // use this information. APInt KnownZero, KnownOne; - APInt Mask = APInt::getAllOnesValue(VTBits); - ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth); + ComputeMaskedBits(Op, KnownZero, KnownOne, Depth); + APInt Mask; if (KnownZero.isNegative()) { // sign bit is 0 Mask = KnownZero; } else if (KnownOne.isNegative()) { // sign bit is 1; @@ -2304,7 +2371,7 @@ bool SelectionDAG::isBaseWithConstantOffset(SDValue Op) const { bool SelectionDAG::isKnownNeverNaN(SDValue Op) const { // If we're told that NaNs won't happen, assume they won't. - if (NoNaNsFPMath) + if (getTarget().Options.NoNaNsFPMath) return true; // If the value is a constant, we can obviously see if it is a NaN or not. @@ -2322,6 +2389,13 @@ bool SelectionDAG::isKnownNeverZero(SDValue Op) const { return !C->isZero(); // TODO: Recognize more cases here. + switch (Op.getOpcode()) { + default: break; + case ISD::OR: + if (const ConstantSDNode *C = dyn_cast(Op.getOperand(1))) + return !C->isNullValue(); + break; + } return false; } @@ -2339,16 +2413,6 @@ bool SelectionDAG::isEqualTo(SDValue A, SDValue B) const { return false; } -bool SelectionDAG::isVerifiedDebugInfoDesc(SDValue Op) const { - GlobalAddressSDNode *GA = dyn_cast(Op); - if (!GA) return false; - if (GA->getOffset() != 0) return false; - const GlobalVariable *GV = dyn_cast(GA->getGlobal()); - if (!GV) return false; - return MF->getMMI().hasDebugInfo(); -} - - /// getNode - Gets or creates the specified node. /// SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT) { @@ -2393,17 +2457,19 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, } case ISD::BITCAST: if (VT == MVT::f32 && C->getValueType(0) == MVT::i32) - return getConstantFP(Val.bitsToFloat(), VT); + return getConstantFP(APFloat(Val), VT); else if (VT == MVT::f64 && C->getValueType(0) == MVT::i64) - return getConstantFP(Val.bitsToDouble(), VT); + return getConstantFP(APFloat(Val), VT); break; case ISD::BSWAP: return getConstant(Val.byteSwap(), VT); case ISD::CTPOP: return getConstant(Val.countPopulation(), VT); case ISD::CTLZ: + case ISD::CTLZ_ZERO_UNDEF: return getConstant(Val.countLeadingZeros(), VT); case ISD::CTTZ: + case ISD::CTTZ_ZERO_UNDEF: return getConstant(Val.countTrailingZeros(), VT); } } @@ -2419,7 +2485,24 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, case ISD::FABS: V.clearSign(); return getConstantFP(V, VT); - case ISD::FP_ROUND: + case ISD::FCEIL: { + APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardPositive); + if (fs == APFloat::opOK || fs == APFloat::opInexact) + return getConstantFP(V, VT); + break; + } + case ISD::FTRUNC: { + APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardZero); + if (fs == APFloat::opOK || fs == APFloat::opInexact) + return getConstantFP(V, VT); + break; + } + case ISD::FFLOOR: { + APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardNegative); + if (fs == APFloat::opOK || fs == APFloat::opInexact) + return getConstantFP(V, VT); + break; + } case ISD::FP_EXTEND: { bool ignored; // This can return overflow, underflow, or inexact; we don't care. @@ -2439,7 +2522,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, APFloat::rmTowardZero, &ignored); if (s==APFloat::opInvalidOp) // inexact is OK, in fact usual break; - APInt api(VT.getSizeInBits(), 2, x); + APInt api(VT.getSizeInBits(), x); return getConstant(api, VT); } case ISD::BITCAST: @@ -2540,17 +2623,18 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, "Vector element count mismatch!"); if (OpOpcode == ISD::TRUNCATE) return getNode(ISD::TRUNCATE, DL, VT, Operand.getNode()->getOperand(0)); - else if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND || - OpOpcode == ISD::ANY_EXTEND) { + if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND || + OpOpcode == ISD::ANY_EXTEND) { // If the source is smaller than the dest, we still need an extend. if (Operand.getNode()->getOperand(0).getValueType().getScalarType() .bitsLT(VT.getScalarType())) return getNode(OpOpcode, DL, VT, Operand.getNode()->getOperand(0)); - else if (Operand.getNode()->getOperand(0).getValueType().bitsGT(VT)) + if (Operand.getNode()->getOperand(0).getValueType().bitsGT(VT)) return getNode(ISD::TRUNCATE, DL, VT, Operand.getNode()->getOperand(0)); - else - return Operand.getNode()->getOperand(0); + return Operand.getNode()->getOperand(0); } + if (OpOpcode == ISD::UNDEF) + return getUNDEF(VT); break; case ISD::BITCAST: // Basic sanity checking. @@ -2580,7 +2664,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, break; case ISD::FNEG: // -(X-Y) -> (Y-X) is unsafe because when X==Y, -0.0 != +0.0 - if (UnsafeFPMath && OpOpcode == ISD::FSUB) + if (getTarget().Options.UnsafeFPMath && OpOpcode == ISD::FSUB) return getNode(ISD::FSUB, DL, VT, Operand.getNode()->getOperand(1), Operand.getNode()->getOperand(0)); if (OpOpcode == ISD::FNEG) // --X -> X @@ -2666,6 +2750,11 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, if (N1 == N2) return N1; break; case ISD::CONCAT_VECTORS: + // Concat of UNDEFs is UNDEF. + if (N1.getOpcode() == ISD::UNDEF && + N2.getOpcode() == ISD::UNDEF) + return getUNDEF(VT); + // A CONCAT_VECTOR with all operands BUILD_VECTOR can be simplified to // one big BUILD_VECTOR. if (N1.getOpcode() == ISD::BUILD_VECTOR && @@ -2715,7 +2804,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, case ISD::FMUL: case ISD::FDIV: case ISD::FREM: - if (UnsafeFPMath) { + if (getTarget().Options.UnsafeFPMath) { if (Opcode == ISD::FADD) { // 0+x --> x if (ConstantFPSDNode *CFP = dyn_cast(N1)) @@ -2730,6 +2819,24 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, if (ConstantFPSDNode *CFP = dyn_cast(N2)) if (CFP->getValueAPF().isZero()) return N1; + } else if (Opcode == ISD::FMUL) { + ConstantFPSDNode *CFP = dyn_cast(N1); + SDValue V = N2; + + // If the first operand isn't the constant, try the second + if (!CFP) { + CFP = dyn_cast(N2); + V = N1; + } + + if (CFP) { + // 0*x --> 0 + if (CFP->isZero()) + return SDValue(CFP,0); + // 1*x --> x + if (CFP->isExactlyValue(1.0)) + return V; + } } } assert(VT.isFloatingPoint() && "This operator only applies to FP types!"); @@ -2779,6 +2886,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, EVT.getVectorNumElements() == VT.getVectorNumElements()) && "Vector element counts must match in FP_ROUND_INREG"); assert(EVT.bitsLE(VT) && "Not rounding down!"); + (void)EVT; if (cast(N2)->getVT() == VT) return N1; // Not actually rounding. break; } @@ -2847,17 +2955,13 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, // expanding large vector constants. if (N2C && N1.getOpcode() == ISD::BUILD_VECTOR) { SDValue Elt = N1.getOperand(N2C->getZExtValue()); - EVT VEltTy = N1.getValueType().getVectorElementType(); - if (Elt.getValueType() != VEltTy) { + + if (VT != Elt.getValueType()) // If the vector element type is not legal, the BUILD_VECTOR operands - // are promoted and implicitly truncated. Make that explicit here. - Elt = getNode(ISD::TRUNCATE, DL, VEltTy, Elt); - } - if (VT != VEltTy) { - // If the vector element type is not legal, the EXTRACT_VECTOR_ELT - // result is implicitly extended. - Elt = getNode(ISD::ANY_EXTEND, DL, VT, Elt); - } + // are promoted and implicitly truncated, and the result implicitly + // extended. Make that explicit here. + Elt = getAnyExtOrTrunc(Elt, DL, VT); + return Elt; } @@ -2886,6 +2990,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, assert(N2C && (unsigned)N2C->getZExtValue() < 2 && "Bad EXTRACT_ELEMENT!"); assert(!N1.getValueType().isVector() && !VT.isVector() && (N1.getValueType().isInteger() == VT.isInteger()) && + N1.getValueType() != VT && "Wrong types for EXTRACT_ELEMENT!"); // EXTRACT_ELEMENT of BUILD_PAIR is often formed while legalize is expanding @@ -2982,6 +3087,16 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, default: break; } } + + if (Opcode == ISD::FP_ROUND) { + APFloat V = N1CFP->getValueAPF(); // make copy + bool ignored; + // This can return overflow, underflow, or inexact; we don't care. + // FIXME need to be more flexible about rounding mode. + (void)V.convert(*EVTToAPFloatSemantics(VT), + APFloat::rmNearestTiesToEven, &ignored); + return getConstantFP(V, VT); + } } // Canonicalize an UNDEF to the RHS, even over a constant. @@ -3036,7 +3151,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, case ISD::FMUL: case ISD::FDIV: case ISD::FREM: - if (UnsafeFPMath) + if (getTarget().Options.UnsafeFPMath) return N2; break; case ISD::MUL: @@ -3110,16 +3225,14 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT, case ISD::SELECT: if (N1C) { if (N1C->getZExtValue()) - return N2; // select true, X, Y -> X - else - return N3; // select false, X, Y -> Y + return N2; // select true, X, Y -> X + return N3; // select false, X, Y -> Y } if (N2 == N3) return N2; // select C, X, X -> X break; case ISD::VECTOR_SHUFFLE: llvm_unreachable("should use getVectorShuffle constructor!"); - break; case ISD::INSERT_SUBVECTOR: { SDValue Index = N3; if (VT.isSimple() && N1.getValueType().isSimple() @@ -3252,8 +3365,7 @@ static SDValue getMemsetValue(SDValue Value, EVT VT, SelectionDAG &DAG, /// used when a memcpy is turned into a memset when the source is a constant /// string ptr. static SDValue getMemsetStringVal(EVT VT, DebugLoc dl, SelectionDAG &DAG, - const TargetLowering &TLI, - std::string &Str, unsigned Offset) { + const TargetLowering &TLI, StringRef Str) { // Handle vector with all elements zero. if (Str.empty()) { if (VT.isInteger()) @@ -3271,15 +3383,18 @@ static SDValue getMemsetStringVal(EVT VT, DebugLoc dl, SelectionDAG &DAG, } assert(!VT.isVector() && "Can't handle vector type here!"); - unsigned NumBits = VT.getSizeInBits(); - unsigned MSB = NumBits / 8; + unsigned NumVTBytes = VT.getSizeInBits() / 8; + unsigned NumBytes = std::min(NumVTBytes, unsigned(Str.size())); + uint64_t Val = 0; - if (TLI.isLittleEndian()) - Offset = Offset + MSB - 1; - for (unsigned i = 0; i != MSB; ++i) { - Val = (Val << 8) | (unsigned char)Str[Offset]; - Offset += TLI.isLittleEndian() ? -1 : 1; + if (TLI.isLittleEndian()) { + for (unsigned i = 0; i != NumBytes; ++i) + Val |= (uint64_t)(unsigned char)Str[i] << i*8; + } else { + for (unsigned i = 0; i != NumBytes; ++i) + Val |= (uint64_t)(unsigned char)Str[i] << (NumVTBytes-i-1)*8; } + return DAG.getConstant(Val, VT); } @@ -3294,7 +3409,7 @@ static SDValue getMemBasePlusOffset(SDValue Base, unsigned Offset, /// isMemSrcFromString - Returns true if memcpy source is a string constant. /// -static bool isMemSrcFromString(SDValue Src, std::string &Str) { +static bool isMemSrcFromString(SDValue Src, StringRef &Str) { unsigned SrcDelta = 0; GlobalAddressSDNode *G = NULL; if (Src.getOpcode() == ISD::GlobalAddress) @@ -3308,11 +3423,7 @@ static bool isMemSrcFromString(SDValue Src, std::string &Str) { if (!G) return false; - const GlobalVariable *GV = dyn_cast(G->getGlobal()); - if (GV && GetConstantStringInfo(GV, Str, SrcDelta, false)) - return true; - - return false; + return getConstantStringInfo(G->getGlobal(), Str, SrcDelta, false); } /// FindOptimalMemOpLowering - Determines the optimial series memory ops @@ -3322,25 +3433,25 @@ static bool isMemSrcFromString(SDValue Src, std::string &Str) { static bool FindOptimalMemOpLowering(std::vector &MemOps, unsigned Limit, uint64_t Size, unsigned DstAlign, unsigned SrcAlign, - bool NonScalarIntSafe, + bool IsZeroVal, bool MemcpyStrSrc, SelectionDAG &DAG, const TargetLowering &TLI) { assert((SrcAlign == 0 || SrcAlign >= DstAlign) && "Expecting memcpy / memset source to meet alignment requirement!"); - // If 'SrcAlign' is zero, that means the memory operation does not need load - // the value, i.e. memset or memcpy from constant string. Otherwise, it's - // the inferred alignment of the source. 'DstAlign', on the other hand, is the - // specified alignment of the memory operation. If it is zero, that means - // it's possible to change the alignment of the destination. 'MemcpyStrSrc' - // indicates whether the memcpy source is constant so it does not need to be - // loaded. + // If 'SrcAlign' is zero, that means the memory operation does not need to + // load the value, i.e. memset or memcpy from constant string. Otherwise, + // it's the inferred alignment of the source. 'DstAlign', on the other hand, + // is the specified alignment of the memory operation. If it is zero, that + // means it's possible to change the alignment of the destination. + // 'MemcpyStrSrc' indicates whether the memcpy source is constant so it does + // not need to be loaded. EVT VT = TLI.getOptimalMemOpType(Size, DstAlign, SrcAlign, - NonScalarIntSafe, MemcpyStrSrc, + IsZeroVal, MemcpyStrSrc, DAG.getMachineFunction()); if (VT == MVT::Other) { - if (DstAlign >= TLI.getTargetData()->getPointerPrefAlignment() || + if (DstAlign >= TLI.getDataLayout()->getPointerPrefAlignment() || TLI.allowsUnalignedMemoryAccesses(VT)) { VT = TLI.getPointerTy(); } else { @@ -3408,14 +3519,14 @@ static SDValue getMemcpyLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, bool DstAlignCanChange = false; MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo *MFI = MF.getFrameInfo(); - bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize); + bool OptSize = MF.getFunction()->getFnAttributes().hasOptimizeForSizeAttr(); FrameIndexSDNode *FI = dyn_cast(Dst); if (FI && !MFI->isFixedObjectIndex(FI->getIndex())) DstAlignCanChange = true; unsigned SrcAlign = DAG.InferPtrAlignment(Src); if (Align > SrcAlign) SrcAlign = Align; - std::string Str; + StringRef Str; bool CopyFromStr = isMemSrcFromString(Src, Str); bool isZeroStr = CopyFromStr && Str.empty(); unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemcpy(OptSize); @@ -3427,8 +3538,8 @@ static SDValue getMemcpyLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, return SDValue(); if (DstAlignCanChange) { - const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); - unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty); + Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); + unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty); if (NewAlign > Align) { // Give the stack frame object a larger alignment if needed. if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign) @@ -3452,7 +3563,7 @@ static SDValue getMemcpyLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, // We only handle zero vectors here. // FIXME: Handle other cases where store of vector immediate is done in // a single instruction. - Value = getMemsetStringVal(VT, dl, DAG, TLI, Str, SrcOff); + Value = getMemsetStringVal(VT, dl, DAG, TLI, Str.substr(SrcOff)); Store = DAG.getStore(Chain, dl, Value, getMemBasePlusOffset(Dst, DstOff, DAG), DstPtrInfo.getWithOffset(DstOff), isVol, @@ -3501,7 +3612,7 @@ static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, bool DstAlignCanChange = false; MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo *MFI = MF.getFrameInfo(); - bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize); + bool OptSize = MF.getFunction()->getFnAttributes().hasOptimizeForSizeAttr(); FrameIndexSDNode *FI = dyn_cast(Dst); if (FI && !MFI->isFixedObjectIndex(FI->getIndex())) DstAlignCanChange = true; @@ -3516,8 +3627,8 @@ static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, return SDValue(); if (DstAlignCanChange) { - const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); - unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty); + Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); + unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty); if (NewAlign > Align) { // Give the stack frame object a larger alignment if needed. if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign) @@ -3539,7 +3650,7 @@ static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, DebugLoc dl, Value = DAG.getLoad(VT, dl, Chain, getMemBasePlusOffset(Src, SrcOff, DAG), SrcPtrInfo.getWithOffset(SrcOff), isVol, - false, SrcAlign); + false, false, SrcAlign); LoadValues.push_back(Value); LoadChains.push_back(Value.getValue(1)); SrcOff += VTSize; @@ -3579,20 +3690,20 @@ static SDValue getMemsetStores(SelectionDAG &DAG, DebugLoc dl, bool DstAlignCanChange = false; MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo *MFI = MF.getFrameInfo(); - bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize); + bool OptSize = MF.getFunction()->getFnAttributes().hasOptimizeForSizeAttr(); FrameIndexSDNode *FI = dyn_cast(Dst); if (FI && !MFI->isFixedObjectIndex(FI->getIndex())) DstAlignCanChange = true; - bool NonScalarIntSafe = + bool IsZeroVal = isa(Src) && cast(Src)->isNullValue(); if (!FindOptimalMemOpLowering(MemOps, TLI.getMaxStoresPerMemset(OptSize), Size, (DstAlignCanChange ? 0 : Align), 0, - NonScalarIntSafe, false, DAG, TLI)) + IsZeroVal, false, DAG, TLI)) return SDValue(); if (DstAlignCanChange) { - const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); - unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty); + Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext()); + unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty); if (NewAlign > Align) { // Give the stack frame object a larger alignment if needed. if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign) @@ -3686,19 +3797,22 @@ SDValue SelectionDAG::getMemcpy(SDValue Chain, DebugLoc dl, SDValue Dst, // Emit a library call. TargetLowering::ArgListTy Args; TargetLowering::ArgListEntry Entry; - Entry.Ty = TLI.getTargetData()->getIntPtrType(*getContext()); + Entry.Ty = TLI.getDataLayout()->getIntPtrType(*getContext()); Entry.Node = Dst; Args.push_back(Entry); Entry.Node = Src; Args.push_back(Entry); Entry.Node = Size; Args.push_back(Entry); // FIXME: pass in DebugLoc - std::pair CallResult = - TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()), + TargetLowering:: + CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()), false, false, false, false, 0, - TLI.getLibcallCallingConv(RTLIB::MEMCPY), false, - /*isReturnValueUsed=*/false, + TLI.getLibcallCallingConv(RTLIB::MEMCPY), + /*isTailCall=*/false, + /*doesNotReturn=*/false, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMCPY), TLI.getPointerTy()), Args, *this, dl); + std::pair CallResult = TLI.LowerCallTo(CLI); + return CallResult.second; } @@ -3738,19 +3852,22 @@ SDValue SelectionDAG::getMemmove(SDValue Chain, DebugLoc dl, SDValue Dst, // Emit a library call. TargetLowering::ArgListTy Args; TargetLowering::ArgListEntry Entry; - Entry.Ty = TLI.getTargetData()->getIntPtrType(*getContext()); + Entry.Ty = TLI.getDataLayout()->getIntPtrType(*getContext()); Entry.Node = Dst; Args.push_back(Entry); Entry.Node = Src; Args.push_back(Entry); Entry.Node = Size; Args.push_back(Entry); // FIXME: pass in DebugLoc - std::pair CallResult = - TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()), + TargetLowering:: + CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()), false, false, false, false, 0, - TLI.getLibcallCallingConv(RTLIB::MEMMOVE), false, - /*isReturnValueUsed=*/false, + TLI.getLibcallCallingConv(RTLIB::MEMMOVE), + /*isTailCall=*/false, + /*doesNotReturn=*/false, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMMOVE), TLI.getPointerTy()), Args, *this, dl); + std::pair CallResult = TLI.LowerCallTo(CLI); + return CallResult.second; } @@ -3784,7 +3901,7 @@ SDValue SelectionDAG::getMemset(SDValue Chain, DebugLoc dl, SDValue Dst, return Result; // Emit a library call. - const Type *IntPtrTy = TLI.getTargetData()->getIntPtrType(*getContext()); + Type *IntPtrTy = TLI.getDataLayout()->getIntPtrType(*getContext()); TargetLowering::ArgListTy Args; TargetLowering::ArgListEntry Entry; Entry.Node = Dst; Entry.Ty = IntPtrTy; @@ -3803,40 +3920,54 @@ SDValue SelectionDAG::getMemset(SDValue Chain, DebugLoc dl, SDValue Dst, Entry.isSExt = false; Args.push_back(Entry); // FIXME: pass in DebugLoc - std::pair CallResult = - TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()), + TargetLowering:: + CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()), false, false, false, false, 0, - TLI.getLibcallCallingConv(RTLIB::MEMSET), false, - /*isReturnValueUsed=*/false, + TLI.getLibcallCallingConv(RTLIB::MEMSET), + /*isTailCall=*/false, + /*doesNotReturn*/false, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMSET), TLI.getPointerTy()), Args, *this, dl); + std::pair CallResult = TLI.LowerCallTo(CLI); + return CallResult.second; } SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, SDValue Chain, SDValue Ptr, SDValue Cmp, SDValue Swp, MachinePointerInfo PtrInfo, - unsigned Alignment) { + unsigned Alignment, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { if (Alignment == 0) // Ensure that codegen never sees alignment 0 Alignment = getEVTAlignment(MemVT); MachineFunction &MF = getMachineFunction(); - unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore; + // All atomics are load and store, except for ATMOIC_LOAD and ATOMIC_STORE. // For now, atomics are considered to be volatile always. - Flags |= MachineMemOperand::MOVolatile; + // FIXME: Volatile isn't really correct; we should keep track of atomic + // orderings in the memoperand. + unsigned Flags = MachineMemOperand::MOVolatile; + if (Opcode != ISD::ATOMIC_STORE) + Flags |= MachineMemOperand::MOLoad; + if (Opcode != ISD::ATOMIC_LOAD) + Flags |= MachineMemOperand::MOStore; MachineMemOperand *MMO = MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment); - return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Cmp, Swp, MMO); + return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Cmp, Swp, MMO, + Ordering, SynchScope); } SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, SDValue Chain, SDValue Ptr, SDValue Cmp, - SDValue Swp, MachineMemOperand *MMO) { + SDValue Swp, MachineMemOperand *MMO, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { assert(Opcode == ISD::ATOMIC_CMP_SWAP && "Invalid Atomic Op"); assert(Cmp.getValueType() == Swp.getValueType() && "Invalid Atomic Op Types"); @@ -3847,13 +3978,15 @@ SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, ID.AddInteger(MemVT.getRawBits()); SDValue Ops[] = {Chain, Ptr, Cmp, Swp}; AddNodeIDNode(ID, Opcode, VTs, Ops, 4); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void* IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); return SDValue(E, 0); } SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain, - Ptr, Cmp, Swp, MMO); + Ptr, Cmp, Swp, MMO, Ordering, + SynchScope); CSEMap.InsertNode(N, IP); AllNodes.push_back(N); return SDValue(N, 0); @@ -3863,27 +3996,39 @@ SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, SDValue Chain, SDValue Ptr, SDValue Val, const Value* PtrVal, - unsigned Alignment) { + unsigned Alignment, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { if (Alignment == 0) // Ensure that codegen never sees alignment 0 Alignment = getEVTAlignment(MemVT); MachineFunction &MF = getMachineFunction(); - unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore; - - // For now, atomics are considered to be volatile always. - Flags |= MachineMemOperand::MOVolatile; + // An atomic store does not load. An atomic load does not store. + // (An atomicrmw obviously both loads and stores.) + // For now, atomics are considered to be volatile always, and they are + // chained as such. + // FIXME: Volatile isn't really correct; we should keep track of atomic + // orderings in the memoperand. + unsigned Flags = MachineMemOperand::MOVolatile; + if (Opcode != ISD::ATOMIC_STORE) + Flags |= MachineMemOperand::MOLoad; + if (Opcode != ISD::ATOMIC_LOAD) + Flags |= MachineMemOperand::MOStore; MachineMemOperand *MMO = MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags, MemVT.getStoreSize(), Alignment); - return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Val, MMO); + return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Val, MMO, + Ordering, SynchScope); } SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, SDValue Chain, SDValue Ptr, SDValue Val, - MachineMemOperand *MMO) { + MachineMemOperand *MMO, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { assert((Opcode == ISD::ATOMIC_LOAD_ADD || Opcode == ISD::ATOMIC_LOAD_SUB || Opcode == ISD::ATOMIC_LOAD_AND || @@ -3894,23 +4039,84 @@ SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, Opcode == ISD::ATOMIC_LOAD_MAX || Opcode == ISD::ATOMIC_LOAD_UMIN || Opcode == ISD::ATOMIC_LOAD_UMAX || - Opcode == ISD::ATOMIC_SWAP) && + Opcode == ISD::ATOMIC_SWAP || + Opcode == ISD::ATOMIC_STORE) && "Invalid Atomic Op"); EVT VT = Val.getValueType(); - SDVTList VTs = getVTList(VT, MVT::Other); + SDVTList VTs = Opcode == ISD::ATOMIC_STORE ? getVTList(MVT::Other) : + getVTList(VT, MVT::Other); FoldingSetNodeID ID; ID.AddInteger(MemVT.getRawBits()); SDValue Ops[] = {Chain, Ptr, Val}; AddNodeIDNode(ID, Opcode, VTs, Ops, 3); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void* IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); return SDValue(E, 0); } SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain, - Ptr, Val, MMO); + Ptr, Val, MMO, + Ordering, SynchScope); + CSEMap.InsertNode(N, IP); + AllNodes.push_back(N); + return SDValue(N, 0); +} + +SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, + EVT VT, SDValue Chain, + SDValue Ptr, + const Value* PtrVal, + unsigned Alignment, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { + if (Alignment == 0) // Ensure that codegen never sees alignment 0 + Alignment = getEVTAlignment(MemVT); + + MachineFunction &MF = getMachineFunction(); + // An atomic store does not load. An atomic load does not store. + // (An atomicrmw obviously both loads and stores.) + // For now, atomics are considered to be volatile always, and they are + // chained as such. + // FIXME: Volatile isn't really correct; we should keep track of atomic + // orderings in the memoperand. + unsigned Flags = MachineMemOperand::MOVolatile; + if (Opcode != ISD::ATOMIC_STORE) + Flags |= MachineMemOperand::MOLoad; + if (Opcode != ISD::ATOMIC_LOAD) + Flags |= MachineMemOperand::MOStore; + + MachineMemOperand *MMO = + MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags, + MemVT.getStoreSize(), Alignment); + + return getAtomic(Opcode, dl, MemVT, VT, Chain, Ptr, MMO, + Ordering, SynchScope); +} + +SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT, + EVT VT, SDValue Chain, + SDValue Ptr, + MachineMemOperand *MMO, + AtomicOrdering Ordering, + SynchronizationScope SynchScope) { + assert(Opcode == ISD::ATOMIC_LOAD && "Invalid Atomic Op"); + + SDVTList VTs = getVTList(VT, MVT::Other); + FoldingSetNodeID ID; + ID.AddInteger(MemVT.getRawBits()); + SDValue Ops[] = {Chain, Ptr}; + AddNodeIDNode(ID, Opcode, VTs, Ops, 2); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); + void* IP = 0; + if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { + cast(E)->refineAlignment(MMO); + return SDValue(E, 0); + } + SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain, + Ptr, MMO, Ordering, SynchScope); CSEMap.InsertNode(N, IP); AllNodes.push_back(N); return SDValue(N, 0); @@ -3972,6 +4178,8 @@ SelectionDAG::getMemIntrinsicNode(unsigned Opcode, DebugLoc dl, SDVTList VTList, assert((Opcode == ISD::INTRINSIC_VOID || Opcode == ISD::INTRINSIC_W_CHAIN || Opcode == ISD::PREFETCH || + Opcode == ISD::LIFETIME_START || + Opcode == ISD::LIFETIME_END || (Opcode <= INT_MAX && (int)Opcode >= ISD::FIRST_TARGET_MEMORY_OPCODE)) && "Opcode is not a memory-accessing opcode!"); @@ -3981,6 +4189,7 @@ SelectionDAG::getMemIntrinsicNode(unsigned Opcode, DebugLoc dl, SDVTList VTList, if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) { FoldingSetNodeID ID; AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); @@ -4037,8 +4246,11 @@ SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType, EVT VT, DebugLoc dl, SDValue Chain, SDValue Ptr, SDValue Offset, MachinePointerInfo PtrInfo, EVT MemVT, - bool isVolatile, bool isNonTemporal, - unsigned Alignment, const MDNode *TBAAInfo) { + bool isVolatile, bool isNonTemporal, bool isInvariant, + unsigned Alignment, const MDNode *TBAAInfo, + const MDNode *Ranges) { + assert(Chain.getValueType() == MVT::Other && + "Invalid chain type"); if (Alignment == 0) // Ensure that codegen never sees alignment 0 Alignment = getEVTAlignment(VT); @@ -4047,6 +4259,8 @@ SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType, Flags |= MachineMemOperand::MOVolatile; if (isNonTemporal) Flags |= MachineMemOperand::MONonTemporal; + if (isInvariant) + Flags |= MachineMemOperand::MOInvariant; // If we don't have a PtrInfo, infer the trivial frame index case to simplify // clients. @@ -4056,7 +4270,7 @@ SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType, MachineFunction &MF = getMachineFunction(); MachineMemOperand *MMO = MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment, - TBAAInfo); + TBAAInfo, Ranges); return getLoad(AM, ExtType, VT, dl, Chain, Ptr, Offset, MemVT, MMO); } @@ -4093,7 +4307,9 @@ SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType, AddNodeIDNode(ID, ISD::LOAD, VTs, Ops, 3); ID.AddInteger(MemVT.getRawBits()); ID.AddInteger(encodeMemSDNodeFlags(ExtType, AM, MMO->isVolatile(), - MMO->isNonTemporal())); + MMO->isNonTemporal(), + MMO->isInvariant())); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); @@ -4110,10 +4326,13 @@ SDValue SelectionDAG::getLoad(EVT VT, DebugLoc dl, SDValue Chain, SDValue Ptr, MachinePointerInfo PtrInfo, bool isVolatile, bool isNonTemporal, - unsigned Alignment, const MDNode *TBAAInfo) { + bool isInvariant, unsigned Alignment, + const MDNode *TBAAInfo, + const MDNode *Ranges) { SDValue Undef = getUNDEF(Ptr.getValueType()); return getLoad(ISD::UNINDEXED, ISD::NON_EXTLOAD, VT, dl, Chain, Ptr, Undef, - PtrInfo, VT, isVolatile, isNonTemporal, Alignment, TBAAInfo); + PtrInfo, VT, isVolatile, isNonTemporal, isInvariant, Alignment, + TBAAInfo, Ranges); } SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, DebugLoc dl, EVT VT, @@ -4123,7 +4342,7 @@ SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, DebugLoc dl, EVT VT, unsigned Alignment, const MDNode *TBAAInfo) { SDValue Undef = getUNDEF(Ptr.getValueType()); return getLoad(ISD::UNINDEXED, ExtType, VT, dl, Chain, Ptr, Undef, - PtrInfo, MemVT, isVolatile, isNonTemporal, Alignment, + PtrInfo, MemVT, isVolatile, isNonTemporal, false, Alignment, TBAAInfo); } @@ -4136,14 +4355,16 @@ SelectionDAG::getIndexedLoad(SDValue OrigLoad, DebugLoc dl, SDValue Base, "Load is already a indexed load!"); return getLoad(AM, LD->getExtensionType(), OrigLoad.getValueType(), dl, LD->getChain(), Base, Offset, LD->getPointerInfo(), - LD->getMemoryVT(), - LD->isVolatile(), LD->isNonTemporal(), LD->getAlignment()); + LD->getMemoryVT(), LD->isVolatile(), LD->isNonTemporal(), + false, LD->getAlignment()); } SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val, SDValue Ptr, MachinePointerInfo PtrInfo, bool isVolatile, bool isNonTemporal, unsigned Alignment, const MDNode *TBAAInfo) { + assert(Chain.getValueType() == MVT::Other && + "Invalid chain type"); if (Alignment == 0) // Ensure that codegen never sees alignment 0 Alignment = getEVTAlignment(Val.getValueType()); @@ -4167,6 +4388,8 @@ SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val, SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val, SDValue Ptr, MachineMemOperand *MMO) { + assert(Chain.getValueType() == MVT::Other && + "Invalid chain type"); EVT VT = Val.getValueType(); SDVTList VTs = getVTList(MVT::Other); SDValue Undef = getUNDEF(Ptr.getValueType()); @@ -4175,7 +4398,8 @@ SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val, AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4); ID.AddInteger(VT.getRawBits()); ID.AddInteger(encodeMemSDNodeFlags(false, ISD::UNINDEXED, MMO->isVolatile(), - MMO->isNonTemporal())); + MMO->isNonTemporal(), MMO->isInvariant())); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); @@ -4193,6 +4417,8 @@ SDValue SelectionDAG::getTruncStore(SDValue Chain, DebugLoc dl, SDValue Val, EVT SVT,bool isVolatile, bool isNonTemporal, unsigned Alignment, const MDNode *TBAAInfo) { + assert(Chain.getValueType() == MVT::Other && + "Invalid chain type"); if (Alignment == 0) // Ensure that codegen never sees alignment 0 Alignment = getEVTAlignment(SVT); @@ -4218,6 +4444,8 @@ SDValue SelectionDAG::getTruncStore(SDValue Chain, DebugLoc dl, SDValue Val, MachineMemOperand *MMO) { EVT VT = Val.getValueType(); + assert(Chain.getValueType() == MVT::Other && + "Invalid chain type"); if (VT == SVT) return getStore(Chain, dl, Val, Ptr, MMO); @@ -4238,7 +4466,8 @@ SDValue SelectionDAG::getTruncStore(SDValue Chain, DebugLoc dl, SDValue Val, AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4); ID.AddInteger(SVT.getRawBits()); ID.AddInteger(encodeMemSDNodeFlags(true, ISD::UNINDEXED, MMO->isVolatile(), - MMO->isNonTemporal())); + MMO->isNonTemporal(), MMO->isInvariant())); + ID.AddInteger(MMO->getPointerInfo().getAddrSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast(E)->refineAlignment(MMO); @@ -4263,6 +4492,7 @@ SelectionDAG::getIndexedStore(SDValue OrigStore, DebugLoc dl, SDValue Base, AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4); ID.AddInteger(ST->getMemoryVT().getRawBits()); ID.AddInteger(ST->getRawSubclassData()); + ID.AddInteger(ST->getPointerInfo().getAddrSpace()); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) return SDValue(E, 0); @@ -4539,13 +4769,7 @@ SDVTList SelectionDAG::getVTList(const EVT *VTs, unsigned NumVTs) { if (I->NumVTs != NumVTs || VTs[0] != I->VTs[0] || VTs[1] != I->VTs[1]) continue; - bool NoMatch = false; - for (unsigned i = 2; i != NumVTs; ++i) - if (VTs[i] != I->VTs[i]) { - NoMatch = true; - break; - } - if (!NoMatch) + if (std::equal(&VTs[2], &VTs[NumVTs], &I->VTs[2])) return *I; } @@ -4792,6 +5016,20 @@ SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc, return N; } +/// UpdadeDebugLocOnMergedSDNode - If the opt level is -O0 then it throws away +/// the line number information on the merged node since it is not possible to +/// preserve the information that operation is associated with multiple lines. +/// This will make the debugger working better at -O0, were there is a higher +/// probability having other instructions associated with that line. +/// +SDNode *SelectionDAG::UpdadeDebugLocOnMergedSDNode(SDNode *N, DebugLoc OLoc) { + DebugLoc NLoc = N->getDebugLoc(); + if (!(NLoc.isUnknown()) && (OptLevel == CodeGenOpt::None) && (OLoc != NLoc)) { + N->setDebugLoc(DebugLoc()); + } + return N; +} + /// MorphNodeTo - This *mutates* the specified node to have the specified /// return type, opcode, and operands. /// @@ -4813,7 +5051,7 @@ SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, FoldingSetNodeID ID; AddNodeIDNode(ID, Opc, VTs, Ops, NumOps); if (SDNode *ON = CSEMap.FindNodeOrInsertPos(ID, IP)) - return ON; + return UpdadeDebugLocOnMergedSDNode(ON, N->getDebugLoc()); } if (!RemoveNodeFromCSEMaps(N)) @@ -5017,8 +5255,9 @@ SelectionDAG::getMachineNode(unsigned Opcode, DebugLoc DL, SDVTList VTs, FoldingSetNodeID ID; AddNodeIDNode(ID, ~Opcode, VTs, Ops, NumOps); IP = 0; - if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) - return cast(E); + if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { + return cast(UpdadeDebugLocOnMergedSDNode(E, DL)); + } } // Allocate a new MachineSDNode. @@ -5107,11 +5346,7 @@ namespace { /// pointed to by a use iterator is deleted, increment the use iterator /// so that it doesn't dangle. /// -/// This class also manages a "downlink" DAGUpdateListener, to forward -/// messages to ReplaceAllUsesWith's callers. -/// class RAUWUpdateListener : public SelectionDAG::DAGUpdateListener { - SelectionDAG::DAGUpdateListener *DownLink; SDNode::use_iterator &UI; SDNode::use_iterator &UE; @@ -5119,21 +5354,13 @@ class RAUWUpdateListener : public SelectionDAG::DAGUpdateListener { // Increment the iterator as needed. while (UI != UE && N == *UI) ++UI; - - // Then forward the message. - if (DownLink) DownLink->NodeDeleted(N, E); - } - - virtual void NodeUpdated(SDNode *N) { - // Just forward the message. - if (DownLink) DownLink->NodeUpdated(N); } public: - RAUWUpdateListener(SelectionDAG::DAGUpdateListener *dl, + RAUWUpdateListener(SelectionDAG &d, SDNode::use_iterator &ui, SDNode::use_iterator &ue) - : DownLink(dl), UI(ui), UE(ue) {} + : SelectionDAG::DAGUpdateListener(d), UI(ui), UE(ue) {} }; } @@ -5143,8 +5370,7 @@ public: /// /// This version assumes From has a single result value. /// -void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To, - DAGUpdateListener *UpdateListener) { +void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To) { SDNode *From = FromN.getNode(); assert(From->getNumValues() == 1 && FromN.getResNo() == 0 && "Cannot replace with this method!"); @@ -5158,7 +5384,7 @@ void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To, // is replaced by To, we don't want to replace of all its users with To // too. See PR3018 for more info. SDNode::use_iterator UI = From->use_begin(), UE = From->use_end(); - RAUWUpdateListener Listener(UpdateListener, UI, UE); + RAUWUpdateListener Listener(*this, UI, UE); while (UI != UE) { SDNode *User = *UI; @@ -5177,8 +5403,12 @@ void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To, // Now that we have modified User, add it back to the CSE maps. If it // already exists there, recursively merge the results together. - AddModifiedNodeToCSEMaps(User, &Listener); + AddModifiedNodeToCSEMaps(User); } + + // If we just RAUW'd the root, take note. + if (FromN == getRoot()) + setRoot(To); } /// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead. @@ -5187,8 +5417,7 @@ void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To, /// This version assumes that for each value of From, there is a /// corresponding value in To in the same position with the same type. /// -void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To, - DAGUpdateListener *UpdateListener) { +void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To) { #ifndef NDEBUG for (unsigned i = 0, e = From->getNumValues(); i != e; ++i) assert((!From->hasAnyUseOfValue(i) || @@ -5203,7 +5432,7 @@ void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To, // Iterate over just the existing users of From. See the comments in // the ReplaceAllUsesWith above. SDNode::use_iterator UI = From->use_begin(), UE = From->use_end(); - RAUWUpdateListener Listener(UpdateListener, UI, UE); + RAUWUpdateListener Listener(*this, UI, UE); while (UI != UE) { SDNode *User = *UI; @@ -5222,8 +5451,12 @@ void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To, // Now that we have modified User, add it back to the CSE maps. If it // already exists there, recursively merge the results together. - AddModifiedNodeToCSEMaps(User, &Listener); + AddModifiedNodeToCSEMaps(User); } + + // If we just RAUW'd the root, take note. + if (From == getRoot().getNode()) + setRoot(SDValue(To, getRoot().getResNo())); } /// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead. @@ -5231,16 +5464,14 @@ void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To, /// /// This version can replace From with any result values. To must match the /// number and types of values returned by From. -void SelectionDAG::ReplaceAllUsesWith(SDNode *From, - const SDValue *To, - DAGUpdateListener *UpdateListener) { +void SelectionDAG::ReplaceAllUsesWith(SDNode *From, const SDValue *To) { if (From->getNumValues() == 1) // Handle the simple case efficiently. - return ReplaceAllUsesWith(SDValue(From, 0), To[0], UpdateListener); + return ReplaceAllUsesWith(SDValue(From, 0), To[0]); // Iterate over just the existing users of From. See the comments in // the ReplaceAllUsesWith above. SDNode::use_iterator UI = From->use_begin(), UE = From->use_end(); - RAUWUpdateListener Listener(UpdateListener, UI, UE); + RAUWUpdateListener Listener(*this, UI, UE); while (UI != UE) { SDNode *User = *UI; @@ -5260,21 +5491,24 @@ void SelectionDAG::ReplaceAllUsesWith(SDNode *From, // Now that we have modified User, add it back to the CSE maps. If it // already exists there, recursively merge the results together. - AddModifiedNodeToCSEMaps(User, &Listener); + AddModifiedNodeToCSEMaps(User); } + + // If we just RAUW'd the root, take note. + if (From == getRoot().getNode()) + setRoot(SDValue(To[getRoot().getResNo()])); } /// ReplaceAllUsesOfValueWith - Replace any uses of From with To, leaving /// uses of other values produced by From.getNode() alone. The Deleted /// vector is handled the same way as for ReplaceAllUsesWith. -void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To, - DAGUpdateListener *UpdateListener){ +void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To){ // Handle the really simple, really trivial case efficiently. if (From == To) return; // Handle the simple, trivial, case efficiently. if (From.getNode()->getNumValues() == 1) { - ReplaceAllUsesWith(From, To, UpdateListener); + ReplaceAllUsesWith(From, To); return; } @@ -5282,7 +5516,7 @@ void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To, // the ReplaceAllUsesWith above. SDNode::use_iterator UI = From.getNode()->use_begin(), UE = From.getNode()->use_end(); - RAUWUpdateListener Listener(UpdateListener, UI, UE); + RAUWUpdateListener Listener(*this, UI, UE); while (UI != UE) { SDNode *User = *UI; bool UserRemovedFromCSEMaps = false; @@ -5318,8 +5552,12 @@ void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To, // Now that we have modified User, add it back to the CSE maps. If it // already exists there, recursively merge the results together. - AddModifiedNodeToCSEMaps(User, &Listener); + AddModifiedNodeToCSEMaps(User); } + + // If we just RAUW'd the root, take note. + if (From == getRoot()) + setRoot(To); } namespace { @@ -5343,11 +5581,10 @@ namespace { /// handled the same way as for ReplaceAllUsesWith. void SelectionDAG::ReplaceAllUsesOfValuesWith(const SDValue *From, const SDValue *To, - unsigned Num, - DAGUpdateListener *UpdateListener){ + unsigned Num){ // Handle the simple, trivial case efficiently. if (Num == 1) - return ReplaceAllUsesOfValueWith(*From, *To, UpdateListener); + return ReplaceAllUsesOfValueWith(*From, *To); // Read up all the uses and make records of them. This helps // processing new uses that are introduced during the @@ -5392,7 +5629,7 @@ void SelectionDAG::ReplaceAllUsesOfValuesWith(const SDValue *From, // Now that we have modified User, add it back to the CSE maps. If it // already exists there, recursively merge the results together. - AddModifiedNodeToCSEMaps(User, UpdateListener); + AddModifiedNodeToCSEMaps(User); } } @@ -5433,7 +5670,7 @@ unsigned SelectionDAG::AssignTopologicalOrder() { } } - // Visit all the nodes. As we iterate, moves nodes into sorted order, + // Visit all the nodes. As we iterate, move nodes into sorted order, // such that by the time the end is reached all nodes will be sorted. for (allnodes_iterator I = allnodes_begin(),E = allnodes_end(); I != E; ++I) { SDNode *N = I; @@ -5510,9 +5747,9 @@ void SelectionDAG::TransferDbgValues(SDValue From, SDValue To) { return; SDNode *FromNode = From.getNode(); SDNode *ToNode = To.getNode(); - SmallVector &DVs = GetDbgValues(FromNode); + ArrayRef DVs = GetDbgValues(FromNode); SmallVector ClonedDVs; - for (SmallVector::iterator I = DVs.begin(), E = DVs.end(); + for (ArrayRef::iterator I = DVs.begin(), E = DVs.end(); I != E; ++I) { SDDbgValue *Dbg = *I; if (Dbg->getKind() == SDDbgValue::SDNODE) { @@ -5546,7 +5783,7 @@ MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs, EVT memvt, MachineMemOperand *mmo) : SDNode(Opc, dl, VTs), MemoryVT(memvt), MMO(mmo) { SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(), - MMO->isNonTemporal()); + MMO->isNonTemporal(), MMO->isInvariant()); assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!"); assert(isNonTemporal() == MMO->isNonTemporal() && "Non-temporal encoding error!"); @@ -5559,7 +5796,7 @@ MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs, : SDNode(Opc, dl, VTs, Ops, NumOps), MemoryVT(memvt), MMO(mmo) { SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(), - MMO->isNonTemporal()); + MMO->isNonTemporal(), MMO->isInvariant()); assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!"); assert(memvt.getStoreSize() == MMO->getSize() && "Size mismatch!"); } @@ -5693,24 +5930,39 @@ bool SDValue::reachesChainWithoutSideEffects(SDValue Dest, return false; } -/// isPredecessorOf - Return true if this node is a predecessor of N. This node -/// is either an operand of N or it can be reached by traversing up the operands. -/// NOTE: this is an expensive method. Use it carefully. -bool SDNode::isPredecessorOf(SDNode *N) const { - SmallPtrSet Visited; - SmallVector Worklist; - Worklist.push_back(N); +/// hasPredecessor - Return true if N is a predecessor of this node. +/// N is either an operand of this node, or can be reached by recursively +/// traversing up the operands. +/// NOTE: This is an expensive method. Use it carefully. +bool SDNode::hasPredecessor(const SDNode *N) const { + SmallPtrSet Visited; + SmallVector Worklist; + return hasPredecessorHelper(N, Visited, Worklist); +} - do { - N = Worklist.pop_back_val(); - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) { - SDNode *Op = N->getOperand(i).getNode(); - if (Op == this) - return true; +bool SDNode::hasPredecessorHelper(const SDNode *N, + SmallPtrSet &Visited, + SmallVector &Worklist) const { + if (Visited.empty()) { + Worklist.push_back(this); + } else { + // Take a look in the visited set. If we've already encountered this node + // we needn't search further. + if (Visited.count(N)) + return true; + } + + // Haven't visited N yet. Continue the search. + while (!Worklist.empty()) { + const SDNode *M = Worklist.pop_back_val(); + for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) { + SDNode *Op = M->getOperand(i).getNode(); if (Visited.insert(Op)) Worklist.push_back(Op); + if (Op == N) + return true; } - } while (!Worklist.empty()); + } return false; } @@ -5720,561 +5972,6 @@ uint64_t SDNode::getConstantOperandVal(unsigned Num) const { return cast(OperandList[Num])->getZExtValue(); } -std::string SDNode::getOperationName(const SelectionDAG *G) const { - switch (getOpcode()) { - default: - if (getOpcode() < ISD::BUILTIN_OP_END) - return "<>"; - if (isMachineOpcode()) { - if (G) - if (const TargetInstrInfo *TII = G->getTarget().getInstrInfo()) - if (getMachineOpcode() < TII->getNumOpcodes()) - return TII->get(getMachineOpcode()).getName(); - return "<>"; - } - if (G) { - const TargetLowering &TLI = G->getTargetLoweringInfo(); - const char *Name = TLI.getTargetNodeName(getOpcode()); - if (Name) return Name; - return "<>"; - } - return "<>"; - -#ifndef NDEBUG - case ISD::DELETED_NODE: - return "<>"; -#endif - case ISD::PREFETCH: return "Prefetch"; - case ISD::MEMBARRIER: return "MemBarrier"; - case ISD::ATOMIC_CMP_SWAP: return "AtomicCmpSwap"; - case ISD::ATOMIC_SWAP: return "AtomicSwap"; - case ISD::ATOMIC_LOAD_ADD: return "AtomicLoadAdd"; - case ISD::ATOMIC_LOAD_SUB: return "AtomicLoadSub"; - case ISD::ATOMIC_LOAD_AND: return "AtomicLoadAnd"; - case ISD::ATOMIC_LOAD_OR: return "AtomicLoadOr"; - case ISD::ATOMIC_LOAD_XOR: return "AtomicLoadXor"; - case ISD::ATOMIC_LOAD_NAND: return "AtomicLoadNand"; - case ISD::ATOMIC_LOAD_MIN: return "AtomicLoadMin"; - case ISD::ATOMIC_LOAD_MAX: return "AtomicLoadMax"; - case ISD::ATOMIC_LOAD_UMIN: return "AtomicLoadUMin"; - case ISD::ATOMIC_LOAD_UMAX: return "AtomicLoadUMax"; - case ISD::PCMARKER: return "PCMarker"; - case ISD::READCYCLECOUNTER: return "ReadCycleCounter"; - case ISD::SRCVALUE: return "SrcValue"; - case ISD::MDNODE_SDNODE: return "MDNode"; - case ISD::EntryToken: return "EntryToken"; - case ISD::TokenFactor: return "TokenFactor"; - case ISD::AssertSext: return "AssertSext"; - case ISD::AssertZext: return "AssertZext"; - - case ISD::BasicBlock: return "BasicBlock"; - case ISD::VALUETYPE: return "ValueType"; - case ISD::Register: return "Register"; - - case ISD::Constant: return "Constant"; - case ISD::ConstantFP: return "ConstantFP"; - case ISD::GlobalAddress: return "GlobalAddress"; - case ISD::GlobalTLSAddress: return "GlobalTLSAddress"; - case ISD::FrameIndex: return "FrameIndex"; - case ISD::JumpTable: return "JumpTable"; - case ISD::GLOBAL_OFFSET_TABLE: return "GLOBAL_OFFSET_TABLE"; - case ISD::RETURNADDR: return "RETURNADDR"; - case ISD::FRAMEADDR: return "FRAMEADDR"; - case ISD::FRAME_TO_ARGS_OFFSET: return "FRAME_TO_ARGS_OFFSET"; - case ISD::EXCEPTIONADDR: return "EXCEPTIONADDR"; - case ISD::LSDAADDR: return "LSDAADDR"; - case ISD::EHSELECTION: return "EHSELECTION"; - case ISD::EH_RETURN: return "EH_RETURN"; - case ISD::EH_SJLJ_SETJMP: return "EH_SJLJ_SETJMP"; - case ISD::EH_SJLJ_LONGJMP: return "EH_SJLJ_LONGJMP"; - case ISD::EH_SJLJ_DISPATCHSETUP: return "EH_SJLJ_DISPATCHSETUP"; - case ISD::ConstantPool: return "ConstantPool"; - case ISD::ExternalSymbol: return "ExternalSymbol"; - case ISD::BlockAddress: return "BlockAddress"; - case ISD::INTRINSIC_WO_CHAIN: - case ISD::INTRINSIC_VOID: - case ISD::INTRINSIC_W_CHAIN: { - unsigned OpNo = getOpcode() == ISD::INTRINSIC_WO_CHAIN ? 0 : 1; - unsigned IID = cast(getOperand(OpNo))->getZExtValue(); - if (IID < Intrinsic::num_intrinsics) - return Intrinsic::getName((Intrinsic::ID)IID); - else if (const TargetIntrinsicInfo *TII = G->getTarget().getIntrinsicInfo()) - return TII->getName(IID); - llvm_unreachable("Invalid intrinsic ID"); - } - - case ISD::BUILD_VECTOR: return "BUILD_VECTOR"; - case ISD::TargetConstant: return "TargetConstant"; - case ISD::TargetConstantFP:return "TargetConstantFP"; - case ISD::TargetGlobalAddress: return "TargetGlobalAddress"; - case ISD::TargetGlobalTLSAddress: return "TargetGlobalTLSAddress"; - case ISD::TargetFrameIndex: return "TargetFrameIndex"; - case ISD::TargetJumpTable: return "TargetJumpTable"; - case ISD::TargetConstantPool: return "TargetConstantPool"; - case ISD::TargetExternalSymbol: return "TargetExternalSymbol"; - case ISD::TargetBlockAddress: return "TargetBlockAddress"; - - case ISD::CopyToReg: return "CopyToReg"; - case ISD::CopyFromReg: return "CopyFromReg"; - case ISD::UNDEF: return "undef"; - case ISD::MERGE_VALUES: return "merge_values"; - case ISD::INLINEASM: return "inlineasm"; - case ISD::EH_LABEL: return "eh_label"; - case ISD::HANDLENODE: return "handlenode"; - - // Unary operators - case ISD::FABS: return "fabs"; - case ISD::FNEG: return "fneg"; - case ISD::FSQRT: return "fsqrt"; - case ISD::FSIN: return "fsin"; - case ISD::FCOS: return "fcos"; - case ISD::FTRUNC: return "ftrunc"; - case ISD::FFLOOR: return "ffloor"; - case ISD::FCEIL: return "fceil"; - case ISD::FRINT: return "frint"; - case ISD::FNEARBYINT: return "fnearbyint"; - case ISD::FEXP: return "fexp"; - case ISD::FEXP2: return "fexp2"; - case ISD::FLOG: return "flog"; - case ISD::FLOG2: return "flog2"; - case ISD::FLOG10: return "flog10"; - - // Binary operators - case ISD::ADD: return "add"; - case ISD::SUB: return "sub"; - case ISD::MUL: return "mul"; - case ISD::MULHU: return "mulhu"; - case ISD::MULHS: return "mulhs"; - case ISD::SDIV: return "sdiv"; - case ISD::UDIV: return "udiv"; - case ISD::SREM: return "srem"; - case ISD::UREM: return "urem"; - case ISD::SMUL_LOHI: return "smul_lohi"; - case ISD::UMUL_LOHI: return "umul_lohi"; - case ISD::SDIVREM: return "sdivrem"; - case ISD::UDIVREM: return "udivrem"; - case ISD::AND: return "and"; - case ISD::OR: return "or"; - case ISD::XOR: return "xor"; - case ISD::SHL: return "shl"; - case ISD::SRA: return "sra"; - case ISD::SRL: return "srl"; - case ISD::ROTL: return "rotl"; - case ISD::ROTR: return "rotr"; - case ISD::FADD: return "fadd"; - case ISD::FSUB: return "fsub"; - case ISD::FMUL: return "fmul"; - case ISD::FDIV: return "fdiv"; - case ISD::FREM: return "frem"; - case ISD::FCOPYSIGN: return "fcopysign"; - case ISD::FGETSIGN: return "fgetsign"; - case ISD::FPOW: return "fpow"; - - case ISD::FPOWI: return "fpowi"; - case ISD::SETCC: return "setcc"; - case ISD::VSETCC: return "vsetcc"; - case ISD::SELECT: return "select"; - case ISD::SELECT_CC: return "select_cc"; - case ISD::INSERT_VECTOR_ELT: return "insert_vector_elt"; - case ISD::EXTRACT_VECTOR_ELT: return "extract_vector_elt"; - case ISD::CONCAT_VECTORS: return "concat_vectors"; - case ISD::INSERT_SUBVECTOR: return "insert_subvector"; - case ISD::EXTRACT_SUBVECTOR: return "extract_subvector"; - case ISD::SCALAR_TO_VECTOR: return "scalar_to_vector"; - case ISD::VECTOR_SHUFFLE: return "vector_shuffle"; - case ISD::CARRY_FALSE: return "carry_false"; - case ISD::ADDC: return "addc"; - case ISD::ADDE: return "adde"; - case ISD::SADDO: return "saddo"; - case ISD::UADDO: return "uaddo"; - case ISD::SSUBO: return "ssubo"; - case ISD::USUBO: return "usubo"; - case ISD::SMULO: return "smulo"; - case ISD::UMULO: return "umulo"; - case ISD::SUBC: return "subc"; - case ISD::SUBE: return "sube"; - case ISD::SHL_PARTS: return "shl_parts"; - case ISD::SRA_PARTS: return "sra_parts"; - case ISD::SRL_PARTS: return "srl_parts"; - - // Conversion operators. - case ISD::SIGN_EXTEND: return "sign_extend"; - case ISD::ZERO_EXTEND: return "zero_extend"; - case ISD::ANY_EXTEND: return "any_extend"; - case ISD::SIGN_EXTEND_INREG: return "sign_extend_inreg"; - case ISD::TRUNCATE: return "truncate"; - case ISD::FP_ROUND: return "fp_round"; - case ISD::FLT_ROUNDS_: return "flt_rounds"; - case ISD::FP_ROUND_INREG: return "fp_round_inreg"; - case ISD::FP_EXTEND: return "fp_extend"; - - case ISD::SINT_TO_FP: return "sint_to_fp"; - case ISD::UINT_TO_FP: return "uint_to_fp"; - case ISD::FP_TO_SINT: return "fp_to_sint"; - case ISD::FP_TO_UINT: return "fp_to_uint"; - case ISD::BITCAST: return "bitcast"; - case ISD::FP16_TO_FP32: return "fp16_to_fp32"; - case ISD::FP32_TO_FP16: return "fp32_to_fp16"; - - case ISD::CONVERT_RNDSAT: { - switch (cast(this)->getCvtCode()) { - default: llvm_unreachable("Unknown cvt code!"); - case ISD::CVT_FF: return "cvt_ff"; - case ISD::CVT_FS: return "cvt_fs"; - case ISD::CVT_FU: return "cvt_fu"; - case ISD::CVT_SF: return "cvt_sf"; - case ISD::CVT_UF: return "cvt_uf"; - case ISD::CVT_SS: return "cvt_ss"; - case ISD::CVT_SU: return "cvt_su"; - case ISD::CVT_US: return "cvt_us"; - case ISD::CVT_UU: return "cvt_uu"; - } - } - - // Control flow instructions - case ISD::BR: return "br"; - case ISD::BRIND: return "brind"; - case ISD::BR_JT: return "br_jt"; - case ISD::BRCOND: return "brcond"; - case ISD::BR_CC: return "br_cc"; - case ISD::CALLSEQ_START: return "callseq_start"; - case ISD::CALLSEQ_END: return "callseq_end"; - - // Other operators - case ISD::LOAD: return "load"; - case ISD::STORE: return "store"; - case ISD::VAARG: return "vaarg"; - case ISD::VACOPY: return "vacopy"; - case ISD::VAEND: return "vaend"; - case ISD::VASTART: return "vastart"; - case ISD::DYNAMIC_STACKALLOC: return "dynamic_stackalloc"; - case ISD::EXTRACT_ELEMENT: return "extract_element"; - case ISD::BUILD_PAIR: return "build_pair"; - case ISD::STACKSAVE: return "stacksave"; - case ISD::STACKRESTORE: return "stackrestore"; - case ISD::TRAP: return "trap"; - - // Bit manipulation - case ISD::BSWAP: return "bswap"; - case ISD::CTPOP: return "ctpop"; - case ISD::CTTZ: return "cttz"; - case ISD::CTLZ: return "ctlz"; - - // Trampolines - case ISD::TRAMPOLINE: return "trampoline"; - - case ISD::CONDCODE: - switch (cast(this)->get()) { - default: llvm_unreachable("Unknown setcc condition!"); - case ISD::SETOEQ: return "setoeq"; - case ISD::SETOGT: return "setogt"; - case ISD::SETOGE: return "setoge"; - case ISD::SETOLT: return "setolt"; - case ISD::SETOLE: return "setole"; - case ISD::SETONE: return "setone"; - - case ISD::SETO: return "seto"; - case ISD::SETUO: return "setuo"; - case ISD::SETUEQ: return "setue"; - case ISD::SETUGT: return "setugt"; - case ISD::SETUGE: return "setuge"; - case ISD::SETULT: return "setult"; - case ISD::SETULE: return "setule"; - case ISD::SETUNE: return "setune"; - - case ISD::SETEQ: return "seteq"; - case ISD::SETGT: return "setgt"; - case ISD::SETGE: return "setge"; - case ISD::SETLT: return "setlt"; - case ISD::SETLE: return "setle"; - case ISD::SETNE: return "setne"; - } - } -} - -const char *SDNode::getIndexedModeName(ISD::MemIndexedMode AM) { - switch (AM) { - default: - return ""; - case ISD::PRE_INC: - return ""; - case ISD::PRE_DEC: - return ""; - case ISD::POST_INC: - return ""; - case ISD::POST_DEC: - return ""; - } -} - -std::string ISD::ArgFlagsTy::getArgFlagsString() { - std::string S = "< "; - - if (isZExt()) - S += "zext "; - if (isSExt()) - S += "sext "; - if (isInReg()) - S += "inreg "; - if (isSRet()) - S += "sret "; - if (isByVal()) - S += "byval "; - if (isNest()) - S += "nest "; - if (getByValAlign()) - S += "byval-align:" + utostr(getByValAlign()) + " "; - if (getOrigAlign()) - S += "orig-align:" + utostr(getOrigAlign()) + " "; - if (getByValSize()) - S += "byval-size:" + utostr(getByValSize()) + " "; - return S + ">"; -} - -void SDNode::dump() const { dump(0); } -void SDNode::dump(const SelectionDAG *G) const { - print(dbgs(), G); - dbgs() << '\n'; -} - -void SDNode::print_types(raw_ostream &OS, const SelectionDAG *G) const { - OS << (void*)this << ": "; - - for (unsigned i = 0, e = getNumValues(); i != e; ++i) { - if (i) OS << ","; - if (getValueType(i) == MVT::Other) - OS << "ch"; - else - OS << getValueType(i).getEVTString(); - } - OS << " = " << getOperationName(G); -} - -void SDNode::print_details(raw_ostream &OS, const SelectionDAG *G) const { - if (const MachineSDNode *MN = dyn_cast(this)) { - if (!MN->memoperands_empty()) { - OS << "<"; - OS << "Mem:"; - for (MachineSDNode::mmo_iterator i = MN->memoperands_begin(), - e = MN->memoperands_end(); i != e; ++i) { - OS << **i; - if (llvm::next(i) != e) - OS << " "; - } - OS << ">"; - } - } else if (const ShuffleVectorSDNode *SVN = - dyn_cast(this)) { - OS << "<"; - for (unsigned i = 0, e = ValueList[0].getVectorNumElements(); i != e; ++i) { - int Idx = SVN->getMaskElt(i); - if (i) OS << ","; - if (Idx < 0) - OS << "u"; - else - OS << Idx; - } - OS << ">"; - } else if (const ConstantSDNode *CSDN = dyn_cast(this)) { - OS << '<' << CSDN->getAPIntValue() << '>'; - } else if (const ConstantFPSDNode *CSDN = dyn_cast(this)) { - if (&CSDN->getValueAPF().getSemantics()==&APFloat::IEEEsingle) - OS << '<' << CSDN->getValueAPF().convertToFloat() << '>'; - else if (&CSDN->getValueAPF().getSemantics()==&APFloat::IEEEdouble) - OS << '<' << CSDN->getValueAPF().convertToDouble() << '>'; - else { - OS << "getValueAPF().bitcastToAPInt().dump(); - OS << ")>"; - } - } else if (const GlobalAddressSDNode *GADN = - dyn_cast(this)) { - int64_t offset = GADN->getOffset(); - OS << '<'; - WriteAsOperand(OS, GADN->getGlobal()); - OS << '>'; - if (offset > 0) - OS << " + " << offset; - else - OS << " " << offset; - if (unsigned int TF = GADN->getTargetFlags()) - OS << " [TF=" << TF << ']'; - } else if (const FrameIndexSDNode *FIDN = dyn_cast(this)) { - OS << "<" << FIDN->getIndex() << ">"; - } else if (const JumpTableSDNode *JTDN = dyn_cast(this)) { - OS << "<" << JTDN->getIndex() << ">"; - if (unsigned int TF = JTDN->getTargetFlags()) - OS << " [TF=" << TF << ']'; - } else if (const ConstantPoolSDNode *CP = dyn_cast(this)){ - int offset = CP->getOffset(); - if (CP->isMachineConstantPoolEntry()) - OS << "<" << *CP->getMachineCPVal() << ">"; - else - OS << "<" << *CP->getConstVal() << ">"; - if (offset > 0) - OS << " + " << offset; - else - OS << " " << offset; - if (unsigned int TF = CP->getTargetFlags()) - OS << " [TF=" << TF << ']'; - } else if (const BasicBlockSDNode *BBDN = dyn_cast(this)) { - OS << "<"; - const Value *LBB = (const Value*)BBDN->getBasicBlock()->getBasicBlock(); - if (LBB) - OS << LBB->getName() << " "; - OS << (const void*)BBDN->getBasicBlock() << ">"; - } else if (const RegisterSDNode *R = dyn_cast(this)) { - OS << ' ' << PrintReg(R->getReg(), G ? G->getTarget().getRegisterInfo() :0); - } else if (const ExternalSymbolSDNode *ES = - dyn_cast(this)) { - OS << "'" << ES->getSymbol() << "'"; - if (unsigned int TF = ES->getTargetFlags()) - OS << " [TF=" << TF << ']'; - } else if (const SrcValueSDNode *M = dyn_cast(this)) { - if (M->getValue()) - OS << "<" << M->getValue() << ">"; - else - OS << ""; - } else if (const MDNodeSDNode *MD = dyn_cast(this)) { - if (MD->getMD()) - OS << "<" << MD->getMD() << ">"; - else - OS << ""; - } else if (const VTSDNode *N = dyn_cast(this)) { - OS << ":" << N->getVT().getEVTString(); - } - else if (const LoadSDNode *LD = dyn_cast(this)) { - OS << "<" << *LD->getMemOperand(); - - bool doExt = true; - switch (LD->getExtensionType()) { - default: doExt = false; break; - case ISD::EXTLOAD: OS << ", anyext"; break; - case ISD::SEXTLOAD: OS << ", sext"; break; - case ISD::ZEXTLOAD: OS << ", zext"; break; - } - if (doExt) - OS << " from " << LD->getMemoryVT().getEVTString(); - - const char *AM = getIndexedModeName(LD->getAddressingMode()); - if (*AM) - OS << ", " << AM; - - OS << ">"; - } else if (const StoreSDNode *ST = dyn_cast(this)) { - OS << "<" << *ST->getMemOperand(); - - if (ST->isTruncatingStore()) - OS << ", trunc to " << ST->getMemoryVT().getEVTString(); - - const char *AM = getIndexedModeName(ST->getAddressingMode()); - if (*AM) - OS << ", " << AM; - - OS << ">"; - } else if (const MemSDNode* M = dyn_cast(this)) { - OS << "<" << *M->getMemOperand() << ">"; - } else if (const BlockAddressSDNode *BA = - dyn_cast(this)) { - OS << "<"; - WriteAsOperand(OS, BA->getBlockAddress()->getFunction(), false); - OS << ", "; - WriteAsOperand(OS, BA->getBlockAddress()->getBasicBlock(), false); - OS << ">"; - if (unsigned int TF = BA->getTargetFlags()) - OS << " [TF=" << TF << ']'; - } - - if (G) - if (unsigned Order = G->GetOrdering(this)) - OS << " [ORD=" << Order << ']'; - - if (getNodeId() != -1) - OS << " [ID=" << getNodeId() << ']'; - - DebugLoc dl = getDebugLoc(); - if (G && !dl.isUnknown()) { - DIScope - Scope(dl.getScope(G->getMachineFunction().getFunction()->getContext())); - OS << " dbg:"; - // Omit the directory, since it's usually long and uninteresting. - if (Scope.Verify()) - OS << Scope.getFilename(); - else - OS << ""; - OS << ':' << dl.getLine(); - if (dl.getCol() != 0) - OS << ':' << dl.getCol(); - } -} - -void SDNode::print(raw_ostream &OS, const SelectionDAG *G) const { - print_types(OS, G); - for (unsigned i = 0, e = getNumOperands(); i != e; ++i) { - if (i) OS << ", "; else OS << " "; - OS << (void*)getOperand(i).getNode(); - if (unsigned RN = getOperand(i).getResNo()) - OS << ":" << RN; - } - print_details(OS, G); -} - -static void printrWithDepthHelper(raw_ostream &OS, const SDNode *N, - const SelectionDAG *G, unsigned depth, - unsigned indent) -{ - if (depth == 0) - return; - - OS.indent(indent); - - N->print(OS, G); - - if (depth < 1) - return; - - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) { - // Don't follow chain operands. - if (N->getOperand(i).getValueType() == MVT::Other) - continue; - OS << '\n'; - printrWithDepthHelper(OS, N->getOperand(i).getNode(), G, depth-1, indent+2); - } -} - -void SDNode::printrWithDepth(raw_ostream &OS, const SelectionDAG *G, - unsigned depth) const { - printrWithDepthHelper(OS, this, G, depth, 0); -} - -void SDNode::printrFull(raw_ostream &OS, const SelectionDAG *G) const { - // Don't print impossibly deep things. - printrWithDepth(OS, G, 10); -} - -void SDNode::dumprWithDepth(const SelectionDAG *G, unsigned depth) const { - printrWithDepth(dbgs(), G, depth); -} - -void SDNode::dumprFull(const SelectionDAG *G) const { - // Don't print impossibly deep things. - dumprWithDepth(G, 10); -} - -static void DumpNodes(const SDNode *N, unsigned indent, const SelectionDAG *G) { - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) - if (N->getOperand(i).getNode()->hasOneUse()) - DumpNodes(N->getOperand(i).getNode(), indent+2, G); - else - dbgs() << "\n" << std::string(indent+2, ' ') - << (void*)N->getOperand(i).getNode() << ": "; - - - dbgs() << "\n"; - dbgs().indent(indent); - N->dump(G); -} - SDValue SelectionDAG::UnrollVectorOp(SDNode *N, unsigned ResNE) { assert(N->getNumValues() == 1 && "Can't unroll a vector with multiple results!"); @@ -6316,6 +6013,10 @@ SDValue SelectionDAG::UnrollVectorOp(SDNode *N, unsigned ResNE) { Scalars.push_back(getNode(N->getOpcode(), dl, EltVT, &Operands[0], Operands.size())); break; + case ISD::VSELECT: + Scalars.push_back(getNode(ISD::SELECT, dl, EltVT, + &Operands[0], Operands.size())); + break; case ISD::SHL: case ISD::SRA: case ISD::SRL: @@ -6393,18 +6094,14 @@ unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const { const GlobalValue *GV; int64_t GVOffset = 0; if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset)) { - // If GV has specified alignment, then use it. Otherwise, use the preferred - // alignment. - unsigned Align = GV->getAlignment(); - if (!Align) { - if (const GlobalVariable *GVar = dyn_cast(GV)) { - if (GVar->hasInitializer()) { - const TargetData *TD = TLI.getTargetData(); - Align = TD->getPreferredAlignment(GVar); - } - } - } - return MinAlign(Align, GVOffset); + unsigned PtrWidth = TLI.getPointerTy().getSizeInBits(); + APInt KnownZero(PtrWidth, 0), KnownOne(PtrWidth, 0); + llvm::ComputeMaskedBits(const_cast(GV), KnownZero, KnownOne, + TLI.getDataLayout()); + unsigned AlignBits = KnownZero.countTrailingOnes(); + unsigned Align = AlignBits ? 1 << std::min(31U, AlignBits) : 0; + if (Align) + return MinAlign(Align, GVOffset); } // If this is a direct reference to a stack slot, use information about the @@ -6430,81 +6127,13 @@ unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const { return 0; } -void SelectionDAG::dump() const { - dbgs() << "SelectionDAG has " << AllNodes.size() << " nodes:"; - - for (allnodes_const_iterator I = allnodes_begin(), E = allnodes_end(); - I != E; ++I) { - const SDNode *N = I; - if (!N->hasOneUse() && N != getRoot().getNode()) - DumpNodes(N, 2, this); - } - - if (getRoot().getNode()) DumpNodes(getRoot().getNode(), 2, this); - - dbgs() << "\n\n"; -} - -void SDNode::printr(raw_ostream &OS, const SelectionDAG *G) const { - print_types(OS, G); - print_details(OS, G); -} - -typedef SmallPtrSet VisitedSDNodeSet; -static void DumpNodesr(raw_ostream &OS, const SDNode *N, unsigned indent, - const SelectionDAG *G, VisitedSDNodeSet &once) { - if (!once.insert(N)) // If we've been here before, return now. - return; - - // Dump the current SDNode, but don't end the line yet. - OS << std::string(indent, ' '); - N->printr(OS, G); - - // Having printed this SDNode, walk the children: - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) { - const SDNode *child = N->getOperand(i).getNode(); - - if (i) OS << ","; - OS << " "; - - if (child->getNumOperands() == 0) { - // This child has no grandchildren; print it inline right here. - child->printr(OS, G); - once.insert(child); - } else { // Just the address. FIXME: also print the child's opcode. - OS << (void*)child; - if (unsigned RN = N->getOperand(i).getResNo()) - OS << ":" << RN; - } - } - - OS << "\n"; - - // Dump children that have grandchildren on their own line(s). - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) { - const SDNode *child = N->getOperand(i).getNode(); - DumpNodesr(OS, child, indent+2, G, once); - } -} - -void SDNode::dumpr() const { - VisitedSDNodeSet once; - DumpNodesr(dbgs(), this, 0, 0, once); -} - -void SDNode::dumpr(const SelectionDAG *G) const { - VisitedSDNodeSet once; - DumpNodesr(dbgs(), this, 0, G, once); -} - - // getAddressSpace - Return the address space this GlobalAddress belongs to. unsigned GlobalAddressSDNode::getAddressSpace() const { return getGlobal()->getType()->getAddressSpace(); } -const Type *ConstantPoolSDNode::getType() const { +Type *ConstantPoolSDNode::getType() const { if (isMachineConstantPoolEntry()) return Val.MachineCPVal->getType(); return Val.ConstVal->getType();