X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FSelectionDAGBuilder.cpp;h=6f5ea7c746910affa72f6bdc51f24e1b1444d70c;hp=cc6eac70ba7b64e2920cffcc47c962813355f4a7;hb=11053bd878d0a25ddbc791ddc4e898120bd985b8;hpb=aa5b9c0f6f3a99f955fe0ded13d61d7eb4e1a0b5 diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp index cc6eac70ba7..6f5ea7c7469 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp @@ -16,9 +16,11 @@ #include "llvm/ADT/BitVector.h" #include "llvm/ADT/Optional.h" #include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/BranchProbabilityInfo.h" #include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/CodeGen/Analysis.h" #include "llvm/CodeGen/FastISel.h" @@ -46,6 +48,8 @@ #include "llvm/IR/Intrinsics.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/Module.h" +#include "llvm/IR/Statepoint.h" +#include "llvm/MC/MCSymbol.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" @@ -54,7 +58,6 @@ #include "llvm/Target/TargetFrameLowering.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/Target/TargetIntrinsicInfo.h" -#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/Target/TargetLowering.h" #include "llvm/Target/TargetOptions.h" #include "llvm/Target/TargetSelectionDAGInfo.h" @@ -564,6 +567,7 @@ static void getCopyToPartsVector(SelectionDAG &DAG, SDLoc DL, } else if (NumParts > 0) { // If the intermediate type was expanded, split each the value into // legal parts. + assert(NumIntermediates != 0 && "division by zero"); assert(NumParts % NumIntermediates == 0 && "Must expand into a divisible number of parts!"); unsigned Factor = NumParts / NumIntermediates; @@ -646,8 +650,10 @@ namespace { /// specified value into the registers specified by this object. This uses /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. - void getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl, - SDValue &Chain, SDValue *Flag, const Value *V) const; + void + getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl, SDValue &Chain, + SDValue *Flag, const Value *V, + ISD::NodeType PreferredExtendType = ISD::ANY_EXTEND) const; /// AddInlineAsmOperands - Add this value to the specified inlineasm node /// operand list. This adds the code marker, matching input operand index @@ -762,9 +768,10 @@ SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl, - SDValue &Chain, SDValue *Flag, - const Value *V) const { + SDValue &Chain, SDValue *Flag, const Value *V, + ISD::NodeType PreferredExtendType) const { const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + ISD::NodeType ExtendKind = PreferredExtendType; // Get the list of the values's legal parts. unsigned NumRegs = Regs.size(); @@ -773,8 +780,9 @@ void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl, EVT ValueVT = ValueVTs[Value]; unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), ValueVT); MVT RegisterVT = RegVTs[Value]; - ISD::NodeType ExtendKind = - TLI.isZExtFree(Val, RegisterVT)? ISD::ZERO_EXTEND: ISD::ANY_EXTEND; + + if (ExtendKind == ISD::ANY_EXTEND && TLI.isZExtFree(Val, RegisterVT)) + ExtendKind = ISD::ZERO_EXTEND; getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value), &Parts[Part], NumParts, RegisterVT, V, ExtendKind); @@ -861,7 +869,7 @@ void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa, AA = &aa; GFI = gfi; LibInfo = li; - DL = DAG.getSubtarget().getDataLayout(); + DL = DAG.getTarget().getDataLayout(); Context = DAG.getContext(); LPadToCallSiteMap.clear(); } @@ -880,6 +888,7 @@ void SelectionDAGBuilder::clear() { CurInst = nullptr; HasTailCall = false; SDNodeOrder = LowestSDNodeOrder; + StatepointLowering.clear(); } /// clearDanglingDebugInfo - Clear the dangling debug information @@ -989,15 +998,16 @@ void SelectionDAGBuilder::resolveDanglingDebugInfo(const Value *V, DebugLoc dl = DDI.getdl(); unsigned DbgSDNodeOrder = DDI.getSDNodeOrder(); MDNode *Variable = DI->getVariable(); + MDNode *Expr = DI->getExpression(); uint64_t Offset = DI->getOffset(); // A dbg.value for an alloca is always indirect. bool IsIndirect = isa(V) || Offset != 0; SDDbgValue *SDV; if (Val.getNode()) { - if (!EmitFuncArgumentDbgValue(V, Variable, Offset, IsIndirect, Val)) { - SDV = DAG.getDbgValue(Variable, Val.getNode(), - Val.getResNo(), IsIndirect, - Offset, dl, DbgSDNodeOrder); + if (!EmitFuncArgumentDbgValue(V, Variable, Expr, Offset, IsIndirect, + Val)) { + SDV = DAG.getDbgValue(Variable, Expr, Val.getNode(), Val.getResNo(), + IsIndirect, Offset, dl, DbgSDNodeOrder); DAG.AddDbgValue(SDV, Val.getNode(), false); } } else @@ -1006,6 +1016,24 @@ void SelectionDAGBuilder::resolveDanglingDebugInfo(const Value *V, } } +/// getCopyFromRegs - If there was virtual register allocated for the value V +/// emit CopyFromReg of the specified type Ty. Return empty SDValue() otherwise. +SDValue SelectionDAGBuilder::getCopyFromRegs(const Value *V, Type *Ty) { + DenseMap::iterator It = FuncInfo.ValueMap.find(V); + SDValue res; + + if (It != FuncInfo.ValueMap.end()) { + unsigned InReg = It->second; + RegsForValue RFV(*DAG.getContext(), DAG.getTargetLoweringInfo(), InReg, + Ty); + SDValue Chain = DAG.getEntryNode(); + res = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V); + resolveDanglingDebugInfo(V, res); + } + + return res; +} + /// getValue - Return an SDValue for the given Value. SDValue SelectionDAGBuilder::getValue(const Value *V) { // If we already have an SDValue for this value, use it. It's important @@ -1016,16 +1044,9 @@ SDValue SelectionDAGBuilder::getValue(const Value *V) { // If there's a virtual register allocated and initialized for this // value, use it. - DenseMap::iterator It = FuncInfo.ValueMap.find(V); - if (It != FuncInfo.ValueMap.end()) { - unsigned InReg = It->second; - RegsForValue RFV(*DAG.getContext(), - *TM.getSubtargetImpl()->getTargetLowering(), InReg, - V->getType()); - SDValue Chain = DAG.getEntryNode(); - N = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V); - resolveDanglingDebugInfo(V, N); - return N; + SDValue copyFromReg = getCopyFromRegs(V, V->getType()); + if (copyFromReg.getNode()) { + return copyFromReg; } // Otherwise create a new SDValue and remember it. @@ -1052,10 +1073,10 @@ SDValue SelectionDAGBuilder::getNonRegisterValue(const Value *V) { /// getValueImpl - Helper function for getValue and getNonRegisterValue. /// Create an SDValue for the given value. SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); if (const Constant *C = dyn_cast(V)) { - EVT VT = TLI->getValueType(V->getType(), true); + EVT VT = TLI.getValueType(V->getType(), true); if (const ConstantInt *CI = dyn_cast(C)) return DAG.getConstant(*CI, VT); @@ -1065,7 +1086,7 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { if (isa(C)) { unsigned AS = V->getType()->getPointerAddressSpace(); - return DAG.getConstant(0, TLI->getPointerTy(AS)); + return DAG.getConstant(0, TLI.getPointerTy(AS)); } if (const ConstantFP *CFP = dyn_cast(C)) @@ -1119,7 +1140,7 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { "Unknown struct or array constant!"); SmallVector ValueVTs; - ComputeValueVTs(*TLI, C->getType(), ValueVTs); + ComputeValueVTs(TLI, C->getType(), ValueVTs); unsigned NumElts = ValueVTs.size(); if (NumElts == 0) return SDValue(); // empty struct @@ -1151,7 +1172,7 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { Ops.push_back(getValue(CV->getOperand(i))); } else { assert(isa(C) && "Unknown vector constant!"); - EVT EltVT = TLI->getValueType(VecTy->getElementType()); + EVT EltVT = TLI.getValueType(VecTy->getElementType()); SDValue Op; if (EltVT.isFloatingPoint()) @@ -1171,13 +1192,13 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { DenseMap::iterator SI = FuncInfo.StaticAllocaMap.find(AI); if (SI != FuncInfo.StaticAllocaMap.end()) - return DAG.getFrameIndex(SI->second, TLI->getPointerTy()); + return DAG.getFrameIndex(SI->second, TLI.getPointerTy()); } // If this is an instruction which fast-isel has deferred, select it now. if (const Instruction *Inst = dyn_cast(V)) { unsigned InReg = FuncInfo.InitializeRegForValue(Inst); - RegsForValue RFV(*DAG.getContext(), *TLI, InReg, Inst->getType()); + RegsForValue RFV(*DAG.getContext(), TLI, InReg, Inst->getType()); SDValue Chain = DAG.getEntryNode(); return RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V); } @@ -1186,7 +1207,7 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { } void SelectionDAGBuilder::visitRet(const ReturnInst &I) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SDValue Chain = getControlRoot(); SmallVector Outs; SmallVector OutVals; @@ -1199,7 +1220,7 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { // Leave Outs empty so that LowerReturn won't try to load return // registers the usual way. SmallVector PtrValueVTs; - ComputeValueVTs(*TLI, PointerType::getUnqual(F->getReturnType()), + ComputeValueVTs(TLI, PointerType::getUnqual(F->getReturnType()), PtrValueVTs); SDValue RetPtr = DAG.getRegister(DemoteReg, PtrValueVTs[0]); @@ -1207,7 +1228,7 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { SmallVector ValueVTs; SmallVector Offsets; - ComputeValueVTs(*TLI, I.getOperand(0)->getType(), ValueVTs, &Offsets); + ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs, &Offsets); unsigned NumValues = ValueVTs.size(); SmallVector Chains(NumValues); @@ -1226,28 +1247,33 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { MVT::Other, Chains); } else if (I.getNumOperands() != 0) { SmallVector ValueVTs; - ComputeValueVTs(*TLI, I.getOperand(0)->getType(), ValueVTs); + ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs); unsigned NumValues = ValueVTs.size(); if (NumValues) { SDValue RetOp = getValue(I.getOperand(0)); - for (unsigned j = 0, f = NumValues; j != f; ++j) { - EVT VT = ValueVTs[j]; - ISD::NodeType ExtendKind = ISD::ANY_EXTEND; + const Function *F = I.getParent()->getParent(); + + ISD::NodeType ExtendKind = ISD::ANY_EXTEND; + if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, + Attribute::SExt)) + ExtendKind = ISD::SIGN_EXTEND; + else if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, + Attribute::ZExt)) + ExtendKind = ISD::ZERO_EXTEND; - const Function *F = I.getParent()->getParent(); - if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, - Attribute::SExt)) - ExtendKind = ISD::SIGN_EXTEND; - else if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, - Attribute::ZExt)) - ExtendKind = ISD::ZERO_EXTEND; + LLVMContext &Context = F->getContext(); + bool RetInReg = F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, + Attribute::InReg); + + for (unsigned j = 0; j != NumValues; ++j) { + EVT VT = ValueVTs[j]; if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) - VT = TLI->getTypeForExtArgOrReturn(VT.getSimpleVT(), ExtendKind); + VT = TLI.getTypeForExtArgOrReturn(Context, VT, ExtendKind); - unsigned NumParts = TLI->getNumRegisters(*DAG.getContext(), VT); - MVT PartVT = TLI->getRegisterType(*DAG.getContext(), VT); + unsigned NumParts = TLI.getNumRegisters(Context, VT); + MVT PartVT = TLI.getRegisterType(Context, VT); SmallVector Parts(NumParts); getCopyToParts(DAG, getCurSDLoc(), SDValue(RetOp.getNode(), RetOp.getResNo() + j), @@ -1255,8 +1281,7 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { // 'inreg' on function refers to return value ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy(); - if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex, - Attribute::InReg)) + if (RetInReg) Flags.setInReg(); // Propagate extension type if any @@ -1277,7 +1302,7 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg(); CallingConv::ID CallConv = DAG.getMachineFunction().getFunction()->getCallingConv(); - Chain = TM.getSubtargetImpl()->getTargetLowering()->LowerReturn( + Chain = DAG.getTargetLoweringInfo().LowerReturn( Chain, CallConv, isVarArg, Outs, OutVals, getCurSDLoc(), DAG); // Verify that the target's LowerReturn behaved as expected. @@ -1401,7 +1426,7 @@ SelectionDAGBuilder::EmitBranchForMergedCondition(const Value *Cond, if (TM.Options.NoNaNsFPMath) Condition = getFCmpCodeWithoutNaN(Condition); } else { - Condition = ISD::SETEQ; // silence warning. + (void)Condition; // silence warning. llvm_unreachable("Unknown compare instruction"); } @@ -1560,19 +1585,13 @@ void SelectionDAGBuilder::visitBr(const BranchInst &I) { // Update machine-CFG edges. MachineBasicBlock *Succ0MBB = FuncInfo.MBBMap[I.getSuccessor(0)]; - // Figure out which block is immediately after the current one. - MachineBasicBlock *NextBlock = nullptr; - MachineFunction::iterator BBI = BrMBB; - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; - if (I.isUnconditional()) { // Update machine-CFG edges. BrMBB->addSuccessor(Succ0MBB); // If this is not a fall-through branch or optimizations are switched off, // emit the branch. - if (Succ0MBB != NextBlock || TM.getOptLevel() == CodeGenOpt::None) + if (Succ0MBB != NextBlock(BrMBB) || TM.getOptLevel() == CodeGenOpt::None) DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, getControlRoot(), DAG.getBasicBlock(Succ0MBB))); @@ -1602,7 +1621,7 @@ void SelectionDAGBuilder::visitBr(const BranchInst &I) { // jle foo // if (const BinaryOperator *BOp = dyn_cast(CondVal)) { - if (!TM.getSubtargetImpl()->getTargetLowering()->isJumpExpensive() && + if (!DAG.getTargetLoweringInfo().isJumpExpensive() && BOp->hasOneUse() && (BOp->getOpcode() == Instruction::And || BOp->getOpcode() == Instruction::Or)) { FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB, @@ -1669,7 +1688,7 @@ void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB, assert(CB.CC == ISD::SETLE && "Can handle only LE ranges now"); const APInt& Low = cast(CB.CmpLHS)->getValue(); - const APInt& High = cast(CB.CmpRHS)->getValue(); + const APInt& High = cast(CB.CmpRHS)->getValue(); SDValue CmpOp = getValue(CB.CmpMHS); EVT VT = CmpOp.getValueType(); @@ -1692,16 +1711,9 @@ void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB, if (CB.TrueBB != CB.FalseBB) addSuccessorWithWeight(SwitchBB, CB.FalseBB, CB.FalseWeight); - // Set NextBlock to be the MBB immediately after the current one, if any. - // This is used to avoid emitting unnecessary branches to the next block. - MachineBasicBlock *NextBlock = nullptr; - MachineFunction::iterator BBI = SwitchBB; - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; - // If the lhs block is the next block, invert the condition so that we can // fall through to the lhs instead of the rhs block. - if (CB.TrueBB == NextBlock) { + if (CB.TrueBB == NextBlock(SwitchBB)) { std::swap(CB.TrueBB, CB.FalseBB); SDValue True = DAG.getConstant(1, Cond.getValueType()); Cond = DAG.getNode(ISD::XOR, dl, Cond.getValueType(), Cond, True); @@ -1724,7 +1736,7 @@ void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB, void SelectionDAGBuilder::visitJumpTable(JumpTable &JT) { // Emit the code for the jump table assert(JT.Reg != -1U && "Should lower JT Header first!"); - EVT PTy = TM.getSubtargetImpl()->getTargetLowering()->getPointerTy(); + EVT PTy = DAG.getTargetLoweringInfo().getPointerTy(); SDValue Index = DAG.getCopyFromReg(getControlRoot(), getCurSDLoc(), JT.Reg, PTy); SDValue Table = DAG.getJumpTable(JT.JTI, PTy); @@ -1752,10 +1764,10 @@ void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT, // can be used as an index into the jump table in a subsequent basic block. // This value may be smaller or larger than the target's pointer type, and // therefore require extension or truncating. - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - SwitchOp = DAG.getZExtOrTrunc(Sub, getCurSDLoc(), TLI->getPointerTy()); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + SwitchOp = DAG.getZExtOrTrunc(Sub, getCurSDLoc(), TLI.getPointerTy()); - unsigned JumpTableReg = FuncInfo.CreateReg(TLI->getPointerTy()); + unsigned JumpTableReg = FuncInfo.CreateReg(TLI.getPointerTy()); SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurSDLoc(), JumpTableReg, SwitchOp); JT.Reg = JumpTableReg; @@ -1763,26 +1775,17 @@ void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT, // Emit the range check for the jump table, and branch to the default block // for the switch statement if the value being switched on exceeds the largest // case in the switch. - SDValue CMP = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), - Sub.getValueType()), - Sub, - DAG.getConstant(JTH.Last - JTH.First,VT), - ISD::SETUGT); - - // Set NextBlock to be the MBB immediately after the current one, if any. - // This is used to avoid emitting unnecessary branches to the next block. - MachineBasicBlock *NextBlock = nullptr; - MachineFunction::iterator BBI = SwitchBB; - - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; + SDValue CMP = + DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), + Sub.getValueType()), + Sub, DAG.getConstant(JTH.Last - JTH.First, VT), ISD::SETUGT); SDValue BrCond = DAG.getNode(ISD::BRCOND, getCurSDLoc(), MVT::Other, CopyTo, CMP, DAG.getBasicBlock(JT.Default)); - if (JT.MBB != NextBlock) + // Avoid emitting unnecessary branches to the next block. + if (JT.MBB != NextBlock(SwitchBB)) BrCond = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, BrCond, DAG.getBasicBlock(JT.MBB)); @@ -1799,8 +1802,8 @@ void SelectionDAGBuilder::visitSPDescriptorParent(StackProtectorDescriptor &SPD, MachineBasicBlock *ParentBB) { // First create the loads to the guard/stack slot for the comparison. - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT PtrTy = TLI->getPointerTy(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT PtrTy = TLI.getPointerTy(); MachineFrameInfo *MFI = ParentBB->getParent()->getFrameInfo(); int FI = MFI->getStackProtectorIndex(); @@ -1810,16 +1813,18 @@ void SelectionDAGBuilder::visitSPDescriptorParent(StackProtectorDescriptor &SPD, SDValue StackSlotPtr = DAG.getFrameIndex(FI, PtrTy); unsigned Align = - TLI->getDataLayout()->getPrefTypeAlignment(IRGuard->getType()); + TLI.getDataLayout()->getPrefTypeAlignment(IRGuard->getType()); SDValue Guard; - // If useLoadStackGuardNode returns true, retrieve the guard value from - // the virtual register holding the value. Otherwise, emit a volatile load - // to retrieve the stack guard value. - if (TLI->useLoadStackGuardNode()) - Guard = DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), - SPD.getGuardReg(), PtrTy); + // If GuardReg is set and useLoadStackGuardNode returns true, retrieve the + // guard value from the virtual register holding the value. Otherwise, emit a + // volatile load to retrieve the stack guard value. + unsigned GuardReg = SPD.getGuardReg(); + + if (GuardReg && TLI.useLoadStackGuardNode()) + Guard = DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), GuardReg, + PtrTy); else Guard = DAG.getLoad(PtrTy, getCurSDLoc(), DAG.getEntryNode(), GuardPtr, MachinePointerInfo(IRGuard, 0), @@ -1834,11 +1839,10 @@ void SelectionDAGBuilder::visitSPDescriptorParent(StackProtectorDescriptor &SPD, EVT VT = Guard.getValueType(); SDValue Sub = DAG.getNode(ISD::SUB, getCurSDLoc(), VT, Guard, StackSlot); - SDValue Cmp = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), - Sub.getValueType()), - Sub, DAG.getConstant(0, VT), - ISD::SETNE); + SDValue Cmp = + DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), + Sub.getValueType()), + Sub, DAG.getConstant(0, VT), ISD::SETNE); // If the sub is not 0, then we know the guard/stackslot do not equal, so // branch to failure MBB. @@ -1863,10 +1867,10 @@ void SelectionDAGBuilder::visitSPDescriptorParent(StackProtectorDescriptor &SPD, /// StackProtectorDescriptor. void SelectionDAGBuilder::visitSPDescriptorFailure(StackProtectorDescriptor &SPD) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - SDValue Chain = TLI->makeLibCall(DAG, RTLIB::STACKPROTECTOR_CHECK_FAIL, - MVT::isVoid, nullptr, 0, false, - getCurSDLoc(), false, false).second; + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + SDValue Chain = + TLI.makeLibCall(DAG, RTLIB::STACKPROTECTOR_CHECK_FAIL, MVT::isVoid, + nullptr, 0, false, getCurSDLoc(), false, false).second; DAG.setRoot(Chain); } @@ -1881,16 +1885,15 @@ void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B, DAG.getConstant(B.First, VT)); // Check range - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - SDValue RangeCmp = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + SDValue RangeCmp = + DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), Sub.getValueType()), - Sub, DAG.getConstant(B.Range, VT), - ISD::SETUGT); + Sub, DAG.getConstant(B.Range, VT), ISD::SETUGT); // Determine the type of the test operands. bool UsePtrType = false; - if (!TLI->isTypeLegal(VT)) + if (!TLI.isTypeLegal(VT)) UsePtrType = true; else { for (unsigned i = 0, e = B.Cases.size(); i != e; ++i) @@ -1902,7 +1905,7 @@ void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B, } } if (UsePtrType) { - VT = TLI->getPointerTy(); + VT = TLI.getPointerTy(); Sub = DAG.getZExtOrTrunc(Sub, getCurSDLoc(), VT); } @@ -1911,13 +1914,6 @@ void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B, SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurSDLoc(), B.Reg, Sub); - // Set NextBlock to be the MBB immediately after the current one, if any. - // This is used to avoid emitting unnecessary branches to the next block. - MachineBasicBlock *NextBlock = nullptr; - MachineFunction::iterator BBI = SwitchBB; - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; - MachineBasicBlock* MBB = B.Cases[0].ThisBB; addSuccessorWithWeight(SwitchBB, B.Default); @@ -1927,7 +1923,8 @@ void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B, MVT::Other, CopyTo, RangeCmp, DAG.getBasicBlock(B.Default)); - if (MBB != NextBlock) + // Avoid emitting unnecessary branches to the next block. + if (MBB != NextBlock(SwitchBB)) BrRange = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, CopyTo, DAG.getBasicBlock(MBB)); @@ -1945,23 +1942,19 @@ void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB, SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurSDLoc(), Reg, VT); SDValue Cmp; - unsigned PopCount = CountPopulation_64(B.Mask); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + unsigned PopCount = countPopulation(B.Mask); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); if (PopCount == 1) { // Testing for a single bit; just compare the shift count with what it // would need to be to shift a 1 bit in that position. - Cmp = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), VT), - ShiftOp, - DAG.getConstant(countTrailingZeros(B.Mask), VT), - ISD::SETEQ); + Cmp = DAG.getSetCC( + getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), VT), ShiftOp, + DAG.getConstant(countTrailingZeros(B.Mask), VT), ISD::SETEQ); } else if (PopCount == BB.Range) { // There is only one zero bit in the range, test for it directly. - Cmp = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), VT), - ShiftOp, - DAG.getConstant(CountTrailingOnes_64(B.Mask), VT), - ISD::SETNE); + Cmp = DAG.getSetCC( + getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), VT), ShiftOp, + DAG.getConstant(countTrailingOnes(B.Mask), VT), ISD::SETNE); } else { // Make desired shift SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurSDLoc(), VT, @@ -1971,9 +1964,8 @@ void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB, SDValue AndOp = DAG.getNode(ISD::AND, getCurSDLoc(), VT, SwitchVal, DAG.getConstant(B.Mask, VT)); Cmp = DAG.getSetCC(getCurSDLoc(), - TLI->getSetCCResultType(*DAG.getContext(), VT), - AndOp, DAG.getConstant(0, VT), - ISD::SETNE); + TLI.getSetCCResultType(*DAG.getContext(), VT), AndOp, + DAG.getConstant(0, VT), ISD::SETNE); } // The branch weight from SwitchBB to B.TargetBB is B.ExtraWeight. @@ -1985,14 +1977,8 @@ void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB, MVT::Other, getControlRoot(), Cmp, DAG.getBasicBlock(B.TargetBB)); - // Set NextBlock to be the MBB immediately after the current one, if any. - // This is used to avoid emitting unnecessary branches to the next block. - MachineBasicBlock *NextBlock = nullptr; - MachineFunction::iterator BBI = SwitchBB; - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; - - if (NextMBB != NextBlock) + // Avoid emitting unnecessary branches to the next block. + if (NextMBB != NextBlock(SwitchBB)) BrAnd = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, BrAnd, DAG.getBasicBlock(NextMBB)); @@ -2011,14 +1997,30 @@ void SelectionDAGBuilder::visitInvoke(const InvokeInst &I) { if (isa(Callee)) visitInlineAsm(&I); else if (Fn && Fn->isIntrinsic()) { - assert(Fn->getIntrinsicID() == Intrinsic::donothing); - // Ignore invokes to @llvm.donothing: jump directly to the next BB. + switch (Fn->getIntrinsicID()) { + default: + llvm_unreachable("Cannot invoke this intrinsic"); + case Intrinsic::donothing: + // Ignore invokes to @llvm.donothing: jump directly to the next BB. + break; + case Intrinsic::experimental_patchpoint_void: + case Intrinsic::experimental_patchpoint_i64: + visitPatchpoint(&I, LandingPad); + break; + case Intrinsic::experimental_gc_statepoint: + LowerStatepoint(ImmutableStatepoint(&I), LandingPad); + break; + } } else LowerCallTo(&I, getValue(Callee), false, LandingPad); // If the value of the invoke is used outside of its defining block, make it // available as a virtual register. - CopyToExportRegsIfNeeded(&I); + // We already took care of the exported value for the statepoint instruction + // during call to the LowerStatepoint. + if (!isStatepoint(I)) { + CopyToExportRegsIfNeeded(&I); + } // Update successor info addSuccessorWithWeight(InvokeMBB, Return); @@ -2044,26 +2046,30 @@ void SelectionDAGBuilder::visitLandingPad(const LandingPadInst &LP) { // If there aren't registers to copy the values into (e.g., during SjLj // exceptions), then don't bother to create these DAG nodes. - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (TLI->getExceptionPointerRegister() == 0 && - TLI->getExceptionSelectorRegister() == 0) + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + if (TLI.getExceptionPointerRegister() == 0 && + TLI.getExceptionSelectorRegister() == 0) return; SmallVector ValueVTs; - ComputeValueVTs(*TLI, LP.getType(), ValueVTs); + ComputeValueVTs(TLI, LP.getType(), ValueVTs); assert(ValueVTs.size() == 2 && "Only two-valued landingpads are supported"); // Get the two live-in registers as SDValues. The physregs have already been // copied into virtual registers. SDValue Ops[2]; - Ops[0] = DAG.getZExtOrTrunc( - DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), - FuncInfo.ExceptionPointerVirtReg, TLI->getPointerTy()), - getCurSDLoc(), ValueVTs[0]); + if (FuncInfo.ExceptionPointerVirtReg) { + Ops[0] = DAG.getZExtOrTrunc( + DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), + FuncInfo.ExceptionPointerVirtReg, TLI.getPointerTy()), + getCurSDLoc(), ValueVTs[0]); + } else { + Ops[0] = DAG.getConstant(0, TLI.getPointerTy()); + } Ops[1] = DAG.getZExtOrTrunc( - DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), - FuncInfo.ExceptionSelectorVirtReg, TLI->getPointerTy()), - getCurSDLoc(), ValueVTs[1]); + DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), + FuncInfo.ExceptionSelectorVirtReg, TLI.getPointerTy()), + getCurSDLoc(), ValueVTs[1]); // Merge into one. SDValue Res = DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(), @@ -2071,6 +2077,27 @@ void SelectionDAGBuilder::visitLandingPad(const LandingPadInst &LP) { setValue(&LP, Res); } +unsigned +SelectionDAGBuilder::visitLandingPadClauseBB(GlobalValue *ClauseGV, + MachineBasicBlock *LPadBB) { + SDValue Chain = getControlRoot(); + + // Get the typeid that we will dispatch on later. + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + const TargetRegisterClass *RC = TLI.getRegClassFor(TLI.getPointerTy()); + unsigned VReg = FuncInfo.MF->getRegInfo().createVirtualRegister(RC); + unsigned TypeID = DAG.getMachineFunction().getMMI().getTypeIDFor(ClauseGV); + SDValue Sel = DAG.getConstant(TypeID, TLI.getPointerTy()); + Chain = DAG.getCopyToReg(Chain, getCurSDLoc(), VReg, Sel); + + // Branch to the main landing pad block. + MachineBasicBlock *ClauseMBB = FuncInfo.MBB; + ClauseMBB->addSuccessor(LPadBB); + DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, Chain, + DAG.getBasicBlock(LPadBB))); + return VReg; +} + /// handleSmallSwitchCaseRange - Emit a series of specific tests (suitable for /// small case ranges). bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR, @@ -2088,11 +2115,10 @@ bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR, MachineFunction *CurMF = FuncInfo.MF; // Figure out which block is immediately after the current one. - MachineBasicBlock *NextBlock = nullptr; + MachineBasicBlock *NextMBB = nullptr; MachineFunction::iterator BBI = CR.CaseBB; - if (++BBI != FuncInfo.MF->end()) - NextBlock = BBI; + NextMBB = BBI; BranchProbabilityInfo *BPI = FuncInfo.BPI; // If any two of the cases has the same destination, and if one value @@ -2106,8 +2132,8 @@ bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR, Case &Big = *(CR.Range.second-1); if (Small.Low == Small.High && Big.Low == Big.High && Small.BB == Big.BB) { - const APInt& SmallValue = cast(Small.Low)->getValue(); - const APInt& BigValue = cast(Big.Low)->getValue(); + const APInt& SmallValue = Small.Low->getValue(); + const APInt& BigValue = Big.Low->getValue(); // Check that there is only one bit different. if (BigValue.countPopulation() == SmallValue.countPopulation() + 1 && @@ -2165,13 +2191,12 @@ bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR, } // Rearrange the case blocks so that the last one falls through if possible. Case &BackCase = *(CR.Range.second-1); - if (Size > 1 && - NextBlock && Default != NextBlock && BackCase.BB != NextBlock) { - // The last case block won't fall through into 'NextBlock' if we emit the + if (Size > 1 && NextMBB && Default != NextMBB && BackCase.BB != NextMBB) { + // The last case block won't fall through into 'NextMBB' if we emit the // branches in this order. See if rearranging a case value would help. // We start at the bottom as it's the case with the least weight. for (Case *I = &*(CR.Range.second-2), *E = &*CR.Range.first-1; I != E; --I) - if (I->BB == NextBlock) { + if (I->BB == NextMBB) { std::swap(*I, BackCase); break; } @@ -2228,9 +2253,8 @@ bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR, } static inline bool areJTsAllowed(const TargetLowering &TLI) { - return TLI.supportJumpTables() && - (TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) || - TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other)); + return TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) || + TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other); } static APInt ComputeRange(const APInt &First, const APInt &Last) { @@ -2248,15 +2272,15 @@ bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR, Case& FrontCase = *CR.Range.first; Case& BackCase = *(CR.Range.second-1); - const APInt &First = cast(FrontCase.Low)->getValue(); - const APInt &Last = cast(BackCase.High)->getValue(); + const APInt &First = FrontCase.Low->getValue(); + const APInt &Last = BackCase.High->getValue(); APInt TSize(First.getBitWidth(), 0); for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) TSize += I->size(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (!areJTsAllowed(*TLI) || TSize.ult(TLI->getMinimumJumpTableEntries())) + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + if (!areJTsAllowed(TLI) || TSize.ult(TLI.getMinimumJumpTableEntries())) return false; APInt Range = ComputeRange(First, Last); @@ -2299,8 +2323,8 @@ bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR, std::vector DestBBs; APInt TEI = First; for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++TEI) { - const APInt &Low = cast(I->Low)->getValue(); - const APInt &High = cast(I->High)->getValue(); + const APInt &Low = I->Low->getValue(); + const APInt &High = I->High->getValue(); if (Low.sle(TEI) && TEI.sle(High)) { DestBBs.push_back(I->BB); @@ -2313,31 +2337,24 @@ bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR, // Calculate weight for each unique destination in CR. DenseMap DestWeights; - if (FuncInfo.BPI) - for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) { - DenseMap::iterator Itr = - DestWeights.find(I->BB); - if (Itr != DestWeights.end()) - Itr->second += I->ExtraWeight; - else - DestWeights[I->BB] = I->ExtraWeight; - } + if (FuncInfo.BPI) { + for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) + DestWeights[I->BB] += I->ExtraWeight; + } // Update successor info. Add one edge to each unique successor. BitVector SuccsHandled(CR.CaseBB->getParent()->getNumBlockIDs()); - for (std::vector::iterator I = DestBBs.begin(), - E = DestBBs.end(); I != E; ++I) { - if (!SuccsHandled[(*I)->getNumber()]) { - SuccsHandled[(*I)->getNumber()] = true; - DenseMap::iterator Itr = - DestWeights.find(*I); - addSuccessorWithWeight(JumpTableBB, *I, - Itr != DestWeights.end() ? Itr->second : 0); + for (MachineBasicBlock *DestBB : DestBBs) { + if (!SuccsHandled[DestBB->getNumber()]) { + SuccsHandled[DestBB->getNumber()] = true; + auto I = DestWeights.find(DestBB); + addSuccessorWithWeight(JumpTableBB, DestBB, + I != DestWeights.end() ? I->second : 0); } } // Create a jump table index for this jump table. - unsigned JTEncoding = TLI->getJumpTableEncoding(); + unsigned JTEncoding = TLI.getJumpTableEncoding(); unsigned JTI = CurMF->getOrCreateJumpTableInfo(JTEncoding) ->createJumpTableIndex(DestBBs); @@ -2357,25 +2374,15 @@ bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR, bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, CaseRecVector& WorkList, const Value* SV, - MachineBasicBlock* Default, MachineBasicBlock* SwitchBB) { - // Get the MachineFunction which holds the current MBB. This is used when - // inserting any additional MBBs necessary to represent the switch. - MachineFunction *CurMF = FuncInfo.MF; - - // Figure out which block is immediately after the current one. - MachineFunction::iterator BBI = CR.CaseBB; - ++BBI; - Case& FrontCase = *CR.Range.first; Case& BackCase = *(CR.Range.second-1); - const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock(); // Size is the number of Cases represented by this range. unsigned Size = CR.Range.second - CR.Range.first; - const APInt &First = cast(FrontCase.Low)->getValue(); - const APInt &Last = cast(BackCase.High)->getValue(); + const APInt &First = FrontCase.Low->getValue(); + const APInt &Last = BackCase.High->getValue(); double FMetric = 0; CaseItr Pivot = CR.Range.first + Size/2; @@ -2391,22 +2398,27 @@ bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, DEBUG(dbgs() << "Selecting best pivot: \n" << "First: " << First << ", Last: " << Last <<'\n' << "LSize: " << LSize << ", RSize: " << RSize << '\n'); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); for (CaseItr I = CR.Range.first, J=I+1, E = CR.Range.second; J!=E; ++I, ++J) { - const APInt &LEnd = cast(I->High)->getValue(); - const APInt &RBegin = cast(J->Low)->getValue(); + const APInt &LEnd = I->High->getValue(); + const APInt &RBegin = J->Low->getValue(); APInt Range = ComputeRange(LEnd, RBegin); assert((Range - 2ULL).isNonNegative() && "Invalid case distance"); // Use volatile double here to avoid excess precision issues on some hosts, // e.g. that use 80-bit X87 registers. + // Only consider the density of sub-ranges that actually have sufficient + // entries to be lowered as a jump table. volatile double LDensity = - (double)LSize.roundToDouble() / - (LEnd - First + 1ULL).roundToDouble(); + LSize.ult(TLI.getMinimumJumpTableEntries()) + ? 0.0 + : LSize.roundToDouble() / (LEnd - First + 1ULL).roundToDouble(); volatile double RDensity = - (double)RSize.roundToDouble() / - (Last - RBegin + 1ULL).roundToDouble(); - volatile double Metric = Range.logBase2()*(LDensity+RDensity); + RSize.ult(TLI.getMinimumJumpTableEntries()) + ? 0.0 + : RSize.roundToDouble() / (Last - RBegin + 1ULL).roundToDouble(); + volatile double Metric = Range.logBase2() * (LDensity + RDensity); // Should always split in some non-trivial place DEBUG(dbgs() <<"=>Step\n" << "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n' @@ -2423,17 +2435,29 @@ bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, RSize -= J->size(); } - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (areJTsAllowed(*TLI)) { - // If our case is dense we *really* should handle it earlier! - assert((FMetric > 0) && "Should handle dense range earlier!"); - } else { + if (FMetric == 0 || !areJTsAllowed(TLI)) Pivot = CR.Range.first + Size/2; - } + splitSwitchCase(CR, Pivot, WorkList, SV, SwitchBB); + return true; +} + +void SelectionDAGBuilder::splitSwitchCase(CaseRec &CR, CaseItr Pivot, + CaseRecVector &WorkList, + const Value *SV, + MachineBasicBlock *SwitchBB) { + // Get the MachineFunction which holds the current MBB. This is used when + // inserting any additional MBBs necessary to represent the switch. + MachineFunction *CurMF = FuncInfo.MF; + + // Figure out which block is immediately after the current one. + MachineFunction::iterator BBI = CR.CaseBB; + ++BBI; + + const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock(); CaseRange LHSR(CR.Range.first, Pivot); CaseRange RHSR(Pivot, CR.Range.second); - const Constant *C = Pivot->Low; + const ConstantInt *C = Pivot->Low; MachineBasicBlock *FalseBB = nullptr, *TrueBB = nullptr; // We know that we branch to the LHS if the Value being switched on is @@ -2442,10 +2466,8 @@ bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, // LHS's Case Value, and that Case Value is exactly one less than the // Pivot's Value, then we can branch directly to the LHS's Target, // rather than creating a leaf node for it. - if ((LHSR.second - LHSR.first) == 1 && - LHSR.first->High == CR.GE && - cast(C)->getValue() == - (cast(CR.GE)->getValue() + 1LL)) { + if ((LHSR.second - LHSR.first) == 1 && LHSR.first->High == CR.GE && + C->getValue() == (CR.GE->getValue() + 1LL)) { TrueBB = LHSR.first->BB; } else { TrueBB = CurMF->CreateMachineBasicBlock(LLVMBB); @@ -2461,13 +2483,12 @@ bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, // is CR.LT - 1, then we can branch directly to the target block for // the current Case Value, rather than emitting a RHS leaf node for it. if ((RHSR.second - RHSR.first) == 1 && CR.LT && - cast(RHSR.first->Low)->getValue() == - (cast(CR.LT)->getValue() - 1LL)) { + RHSR.first->Low->getValue() == (CR.LT->getValue() - 1LL)) { FalseBB = RHSR.first->BB; } else { FalseBB = CurMF->CreateMachineBasicBlock(LLVMBB); CurMF->insert(BBI, FalseBB); - WorkList.push_back(CaseRec(FalseBB,CR.LT,C,RHSR)); + WorkList.push_back(CaseRec(FalseBB, CR.LT, C, RHSR)); // Put SV in a virtual register to make it available from the new blocks. ExportFromCurrentBlock(SV); @@ -2482,8 +2503,6 @@ bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR, visitSwitchCase(CB, SwitchBB); else SwitchCases.push_back(CB); - - return true; } /// handleBitTestsSwitchCase - if current case range has few destination and @@ -2494,8 +2513,8 @@ bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR, const Value* SV, MachineBasicBlock* Default, MachineBasicBlock* SwitchBB) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT PTy = TLI->getPointerTy(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT PTy = TLI.getPointerTy(); unsigned IntPtrBits = PTy.getSizeInBits(); Case& FrontCase = *CR.Range.first; @@ -2506,19 +2525,18 @@ bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR, MachineFunction *CurMF = FuncInfo.MF; // If target does not have legal shift left, do not emit bit tests at all. - if (!TLI->isOperationLegal(ISD::SHL, PTy)) + if (!TLI.isOperationLegal(ISD::SHL, PTy)) return false; size_t numCmps = 0; - for (CaseItr I = CR.Range.first, E = CR.Range.second; - I!=E; ++I) { + for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) { // Single case counts one, case range - two. numCmps += (I->Low == I->High ? 1 : 2); } // Count unique destinations SmallSet Dests; - for (CaseItr I = CR.Range.first, E = CR.Range.second; I!=E; ++I) { + for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) { Dests.insert(I->BB); if (Dests.size() > 3) // Don't bother the code below, if there are too much unique destinations @@ -2529,8 +2547,8 @@ bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR, << "Total number of comparisons: " << numCmps << '\n'); // Compute span of values. - const APInt& minValue = cast(FrontCase.Low)->getValue(); - const APInt& maxValue = cast(BackCase.High)->getValue(); + const APInt& minValue = FrontCase.Low->getValue(); + const APInt& maxValue = BackCase.High->getValue(); APInt cmpRange = maxValue - minValue; DEBUG(dbgs() << "Compare range: " << cmpRange << '\n' @@ -2570,8 +2588,8 @@ bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR, count++; } - const APInt& lowValue = cast(I->Low)->getValue(); - const APInt& highValue = cast(I->High)->getValue(); + const APInt& lowValue = I->Low->getValue(); + const APInt& highValue = I->High->getValue(); uint64_t lo = (lowValue - lowBound).getZExtValue(); uint64_t hi = (highValue - lowBound).getZExtValue(); @@ -2611,65 +2629,61 @@ bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR, BitTestBlock BTB(lowBound, cmpRange, SV, -1U, MVT::Other, (CR.CaseBB == SwitchBB), - CR.CaseBB, Default, BTC); + CR.CaseBB, Default, std::move(BTC)); if (CR.CaseBB == SwitchBB) visitBitTestHeader(BTB, SwitchBB); - BitTestCases.push_back(BTB); + BitTestCases.push_back(std::move(BTB)); return true; } -/// Clusterify - Transform simple list of Cases into list of CaseRange's -size_t SelectionDAGBuilder::Clusterify(CaseVector& Cases, - const SwitchInst& SI) { - size_t numCmps = 0; - +void SelectionDAGBuilder::Clusterify(CaseVector &Cases, const SwitchInst *SI) { BranchProbabilityInfo *BPI = FuncInfo.BPI; - // Start with "simple" cases - for (SwitchInst::ConstCaseIt i = SI.case_begin(), e = SI.case_end(); - i != e; ++i) { - const BasicBlock *SuccBB = i.getCaseSuccessor(); - MachineBasicBlock *SMBB = FuncInfo.MBBMap[SuccBB]; - - uint32_t ExtraWeight = - BPI ? BPI->getEdgeWeight(SI.getParent(), i.getSuccessorIndex()) : 0; - - Cases.push_back(Case(i.getCaseValue(), i.getCaseValue(), - SMBB, ExtraWeight)); - } - std::sort(Cases.begin(), Cases.end(), CaseCmp()); - - // Merge case into clusters - if (Cases.size() >= 2) - // Must recompute end() each iteration because it may be - // invalidated by erase if we hold on to it - for (CaseItr I = Cases.begin(), J = std::next(Cases.begin()); - J != Cases.end(); ) { - const APInt& nextValue = cast(J->Low)->getValue(); - const APInt& currentValue = cast(I->High)->getValue(); - MachineBasicBlock* nextBB = J->BB; - MachineBasicBlock* currentBB = I->BB; - - // If the two neighboring cases go to the same destination, merge them - // into a single case. - if ((nextValue - currentValue == 1) && (currentBB == nextBB)) { - I->High = J->High; - I->ExtraWeight += J->ExtraWeight; - J = Cases.erase(J); - } else { - I = J++; - } + + // Extract cases from the switch and sort them. + typedef std::pair CasePair; + std::vector Sorted; + Sorted.reserve(SI->getNumCases()); + for (auto I : SI->cases()) + Sorted.push_back(std::make_pair(I.getCaseValue(), I.getSuccessorIndex())); + std::sort(Sorted.begin(), Sorted.end(), [](CasePair a, CasePair b) { + return a.first->getValue().slt(b.first->getValue()); + }); + + // Merge adjacent cases with the same destination, build Cases vector. + assert(Cases.empty() && "Cases should be empty before Clusterify;"); + Cases.reserve(SI->getNumCases()); + MachineBasicBlock *PreviousSucc = nullptr; + for (CasePair &CP : Sorted) { + const ConstantInt *CaseVal = CP.first; + unsigned SuccIndex = CP.second; + MachineBasicBlock *Succ = FuncInfo.MBBMap[SI->getSuccessor(SuccIndex)]; + uint32_t Weight = BPI ? BPI->getEdgeWeight(SI->getParent(), SuccIndex) : 0; + + if (PreviousSucc == Succ && + (CaseVal->getValue() - Cases.back().High->getValue()) == 1) { + // If this case has the same successor and is a neighbour, merge it into + // the previous cluster. + Cases.back().High = CaseVal; + Cases.back().ExtraWeight += Weight; + } else { + Cases.push_back(Case(CaseVal, CaseVal, Succ, Weight)); } - for (CaseItr I=Cases.begin(), E=Cases.end(); I!=E; ++I, ++numCmps) { - if (I->Low != I->High) - // A range counts double, since it requires two compares. - ++numCmps; + PreviousSucc = Succ; } - return numCmps; + DEBUG({ + size_t numCmps = 0; + for (auto &I : Cases) + // A range counts double, since it requires two compares. + numCmps += I.Low != I.High ? 2 : 1; + + dbgs() << "Clusterify finished. Total clusters: " << Cases.size() + << ". Total compares: " << numCmps << '\n'; + }); } void SelectionDAGBuilder::UpdateSplitBlock(MachineBasicBlock *First, @@ -2688,37 +2702,54 @@ void SelectionDAGBuilder::UpdateSplitBlock(MachineBasicBlock *First, void SelectionDAGBuilder::visitSwitch(const SwitchInst &SI) { MachineBasicBlock *SwitchMBB = FuncInfo.MBB; - // Figure out which block is immediately after the current one. - MachineBasicBlock *NextBlock = nullptr; + // Create a vector of Cases, sorted so that we can efficiently create a binary + // search tree from them. + CaseVector Cases; + Clusterify(Cases, &SI); + + // Get the default destination MBB. MachineBasicBlock *Default = FuncInfo.MBBMap[SI.getDefaultDest()]; - // If there is only the default destination, branch to it if it is not the - // next basic block. Otherwise, just fall through. - if (!SI.getNumCases()) { - // Update machine-CFG edges. + if (isa(SI.getDefaultDest()->getFirstNonPHIOrDbg()) && + !Cases.empty()) { + // Replace an unreachable default destination with the most popular case + // destination. + DenseMap Popularity; + unsigned MaxPop = 0; + const BasicBlock *MaxBB = nullptr; + for (auto I : SI.cases()) { + const BasicBlock *BB = I.getCaseSuccessor(); + if (++Popularity[BB] > MaxPop) { + MaxPop = Popularity[BB]; + MaxBB = BB; + } + } - // If this is not a fall-through branch, emit the branch. + // Set new default. + assert(MaxPop > 0); + assert(MaxBB); + Default = FuncInfo.MBBMap[MaxBB]; + + // Remove cases that were pointing to the destination that is now the default. + Cases.erase(std::remove_if(Cases.begin(), Cases.end(), + [&](const Case &C) { return C.BB == Default; }), + Cases.end()); + } + + // If there is only the default destination, go there directly. + if (Cases.empty()) { + // Update machine-CFG edges. SwitchMBB->addSuccessor(Default); - if (Default != NextBlock) - DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), - MVT::Other, getControlRoot(), - DAG.getBasicBlock(Default))); + // If this is not a fall-through branch, emit the branch. + if (Default != NextBlock(SwitchMBB)) { + DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, + getControlRoot(), DAG.getBasicBlock(Default))); + } return; } - // If there are any non-default case statements, create a vector of Cases - // representing each one, and sort the vector so that we can efficiently - // create a binary search tree from them. - CaseVector Cases; - size_t numCmps = Clusterify(Cases, SI); - DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size() - << ". Total compares: " << numCmps << '\n'); - (void)numCmps; - - // Get the Value to be switched on and default basic blocks, which will be - // inserted into CaseBlock records, representing basic blocks in the binary - // search tree. + // Get the Value to be switched on. const Value *SV = SI.getCondition(); // Push the initial CaseRec onto the worklist @@ -2748,7 +2779,7 @@ void SelectionDAGBuilder::visitSwitch(const SwitchInst &SI) { // Emit binary tree. We need to pick a pivot, and push left and right ranges // onto the worklist. Leafs are handled via handleSmallSwitchRange() call. - handleBTSplitSwitchCase(CR, WorkList, SV, Default, SwitchMBB); + handleBTSplitSwitchCase(CR, WorkList, SV, SwitchMBB); } } @@ -2759,7 +2790,7 @@ void SelectionDAGBuilder::visitIndirectBr(const IndirectBrInst &I) { SmallSet Done; for (unsigned i = 0, e = I.getNumSuccessors(); i != e; ++i) { BasicBlock *BB = I.getSuccessor(i); - bool Inserted = Done.insert(BB); + bool Inserted = Done.insert(BB).second; if (!Inserted) continue; @@ -2816,8 +2847,8 @@ void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) { SDValue Op1 = getValue(I.getOperand(0)); SDValue Op2 = getValue(I.getOperand(1)); - EVT ShiftTy = TM.getSubtargetImpl()->getTargetLowering()->getShiftAmountTy( - Op2.getValueType()); + EVT ShiftTy = + DAG.getTargetLoweringInfo().getShiftAmountTy(Op2.getValueType()); // Coerce the shift amount to the right type if we can. if (!I.getType()->isVectorTy() && Op2.getValueType() != ShiftTy) { @@ -2872,8 +2903,8 @@ void SelectionDAGBuilder::visitSDiv(const User &I) { if (isa(&I) && cast(&I)->isExact() && !isa(Op1) && isa(Op2) && !cast(Op2)->isNullValue()) - setValue(&I, TM.getSubtargetImpl()->getTargetLowering()->BuildExactSDIV( - Op1, Op2, getCurSDLoc(), DAG)); + setValue(&I, DAG.getTargetLoweringInfo() + .BuildExactSDIV(Op1, Op2, getCurSDLoc(), DAG)); else setValue(&I, DAG.getNode(ISD::SDIV, getCurSDLoc(), Op1.getValueType(), Op1, Op2)); @@ -2889,8 +2920,7 @@ void SelectionDAGBuilder::visitICmp(const User &I) { SDValue Op2 = getValue(I.getOperand(1)); ISD::CondCode Opcode = getICmpCondCode(predicate); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getSetCC(getCurSDLoc(), DestVT, Op1, Op2, Opcode)); } @@ -2905,15 +2935,13 @@ void SelectionDAGBuilder::visitFCmp(const User &I) { ISD::CondCode Condition = getFCmpCondCode(predicate); if (TM.Options.NoNaNsFPMath) Condition = getFCmpCodeWithoutNaN(Condition); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getSetCC(getCurSDLoc(), DestVT, Op1, Op2, Condition)); } void SelectionDAGBuilder::visitSelect(const User &I) { SmallVector ValueVTs; - ComputeValueVTs(*TM.getSubtargetImpl()->getTargetLowering(), I.getType(), - ValueVTs); + ComputeValueVTs(DAG.getTargetLoweringInfo(), I.getType(), ValueVTs); unsigned NumValues = ValueVTs.size(); if (NumValues == 0) return; @@ -2940,8 +2968,7 @@ void SelectionDAGBuilder::visitSelect(const User &I) { void SelectionDAGBuilder::visitTrunc(const User &I) { // TruncInst cannot be a no-op cast because sizeof(src) > sizeof(dest). SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurSDLoc(), DestVT, N)); } @@ -2949,8 +2976,7 @@ void SelectionDAGBuilder::visitZExt(const User &I) { // ZExt cannot be a no-op cast because sizeof(src) < sizeof(dest). // ZExt also can't be a cast to bool for same reason. So, nothing much to do SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurSDLoc(), DestVT, N)); } @@ -2958,58 +2984,51 @@ void SelectionDAGBuilder::visitSExt(const User &I) { // SExt cannot be a no-op cast because sizeof(src) < sizeof(dest). // SExt also can't be a cast to bool for same reason. So, nothing much to do SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, getCurSDLoc(), DestVT, N)); } void SelectionDAGBuilder::visitFPTrunc(const User &I) { // FPTrunc is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT DestVT = TLI->getValueType(I.getType()); - setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurSDLoc(), - DestVT, N, - DAG.getTargetConstant(0, TLI->getPointerTy()))); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT DestVT = TLI.getValueType(I.getType()); + setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurSDLoc(), DestVT, N, + DAG.getTargetConstant(0, TLI.getPointerTy()))); } void SelectionDAGBuilder::visitFPExt(const User &I) { // FPExt is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurSDLoc(), DestVT, N)); } void SelectionDAGBuilder::visitFPToUI(const User &I) { // FPToUI is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::FP_TO_UINT, getCurSDLoc(), DestVT, N)); } void SelectionDAGBuilder::visitFPToSI(const User &I) { // FPToSI is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::FP_TO_SINT, getCurSDLoc(), DestVT, N)); } void SelectionDAGBuilder::visitUIToFP(const User &I) { // UIToFP is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::UINT_TO_FP, getCurSDLoc(), DestVT, N)); } void SelectionDAGBuilder::visitSIToFP(const User &I) { // SIToFP is never a no-op cast, no need to check SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::SINT_TO_FP, getCurSDLoc(), DestVT, N)); } @@ -3017,8 +3036,7 @@ void SelectionDAGBuilder::visitPtrToInt(const User &I) { // What to do depends on the size of the integer and the size of the pointer. // We can either truncate, zero extend, or no-op, accordingly. SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getZExtOrTrunc(N, getCurSDLoc(), DestVT)); } @@ -3026,15 +3044,13 @@ void SelectionDAGBuilder::visitIntToPtr(const User &I) { // What to do depends on the size of the integer and the size of the pointer. // We can either truncate, zero extend, or no-op, accordingly. SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); setValue(&I, DAG.getZExtOrTrunc(N, getCurSDLoc(), DestVT)); } void SelectionDAGBuilder::visitBitCast(const User &I) { SDValue N = getValue(I.getOperand(0)); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType()); // BitCast assures us that source and destination are the same size so this is // either a BITCAST or a no-op. @@ -3056,8 +3072,7 @@ void SelectionDAGBuilder::visitAddrSpaceCast(const User &I) { const TargetLowering &TLI = DAG.getTargetLoweringInfo(); const Value *SV = I.getOperand(0); SDValue N = getValue(SV); - EVT DestVT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + EVT DestVT = TLI.getValueType(I.getType()); unsigned SrcAS = SV->getType()->getPointerAddressSpace(); unsigned DestAS = I.getType()->getPointerAddressSpace(); @@ -3074,11 +3089,8 @@ void SelectionDAGBuilder::visitInsertElement(const User &I) { SDValue InVal = getValue(I.getOperand(1)); SDValue InIdx = DAG.getSExtOrTrunc(getValue(I.getOperand(2)), getCurSDLoc(), TLI.getVectorIdxTy()); - setValue(&I, - DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurSDLoc(), - TM.getSubtargetImpl()->getTargetLowering()->getValueType( - I.getType()), - InVec, InVal, InIdx)); + setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurSDLoc(), + TLI.getValueType(I.getType()), InVec, InVal, InIdx)); } void SelectionDAGBuilder::visitExtractElement(const User &I) { @@ -3086,11 +3098,8 @@ void SelectionDAGBuilder::visitExtractElement(const User &I) { SDValue InVec = getValue(I.getOperand(0)); SDValue InIdx = DAG.getSExtOrTrunc(getValue(I.getOperand(1)), getCurSDLoc(), TLI.getVectorIdxTy()); - setValue(&I, - DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurSDLoc(), - TM.getSubtargetImpl()->getTargetLowering()->getValueType( - I.getType()), - InVec, InIdx)); + setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurSDLoc(), + TLI.getValueType(I.getType()), InVec, InIdx)); } // Utility for visitShuffleVector - Return true if every element in Mask, @@ -3112,8 +3121,8 @@ void SelectionDAGBuilder::visitShuffleVector(const User &I) { ShuffleVectorInst::getShuffleMask(cast(I.getOperand(2)), Mask); unsigned MaskNumElts = Mask.size(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT VT = TLI->getValueType(I.getType()); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT VT = TLI.getValueType(I.getType()); EVT SrcVT = Src1.getValueType(); unsigned SrcNumElts = SrcVT.getVectorNumElements(); @@ -3232,9 +3241,9 @@ void SelectionDAGBuilder::visitShuffleVector(const User &I) { if (RangeUse[Input] == 0) Src = DAG.getUNDEF(VT); else - Src = DAG.getNode(ISD::EXTRACT_SUBVECTOR, getCurSDLoc(), VT, - Src, DAG.getConstant(StartIdx[Input], - TLI->getVectorIdxTy())); + Src = DAG.getNode( + ISD::EXTRACT_SUBVECTOR, getCurSDLoc(), VT, Src, + DAG.getConstant(StartIdx[Input], TLI.getVectorIdxTy())); } // Calculate new mask. @@ -3260,7 +3269,7 @@ void SelectionDAGBuilder::visitShuffleVector(const User &I) { // replacing the shuffle with extract and build vector. // to insert and build vector. EVT EltVT = VT.getVectorElementType(); - EVT IdxVT = TLI->getVectorIdxTy(); + EVT IdxVT = TLI.getVectorIdxTy(); SmallVector Ops; for (unsigned i = 0; i != MaskNumElts; ++i) { int Idx = Mask[i]; @@ -3292,16 +3301,22 @@ void SelectionDAGBuilder::visitInsertValue(const InsertValueInst &I) { unsigned LinearIndex = ComputeLinearIndex(AggTy, I.getIndices()); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SmallVector AggValueVTs; - ComputeValueVTs(*TLI, AggTy, AggValueVTs); + ComputeValueVTs(TLI, AggTy, AggValueVTs); SmallVector ValValueVTs; - ComputeValueVTs(*TLI, ValTy, ValValueVTs); + ComputeValueVTs(TLI, ValTy, ValValueVTs); unsigned NumAggValues = AggValueVTs.size(); unsigned NumValValues = ValValueVTs.size(); SmallVector Values(NumAggValues); + // Ignore an insertvalue that produces an empty object + if (!NumAggValues) { + setValue(&I, DAG.getUNDEF(MVT(MVT::Other))); + return; + } + SDValue Agg = getValue(Op0); unsigned i = 0; // Copy the beginning value(s) from the original aggregate. @@ -3332,9 +3347,9 @@ void SelectionDAGBuilder::visitExtractValue(const ExtractValueInst &I) { unsigned LinearIndex = ComputeLinearIndex(AggTy, I.getIndices()); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SmallVector ValValueVTs; - ComputeValueVTs(*TLI, ValTy, ValValueVTs); + ComputeValueVTs(TLI, ValTy, ValValueVTs); unsigned NumValValues = ValValueVTs.size(); @@ -3381,30 +3396,21 @@ void SelectionDAGBuilder::visitGetElementPtr(const User &I) { Ty = StTy->getElementType(Field); } else { Ty = cast(Ty)->getElementType(); + MVT PtrTy = DAG.getTargetLoweringInfo().getPointerTy(AS); + unsigned PtrSize = PtrTy.getSizeInBits(); + APInt ElementSize(PtrSize, DL->getTypeAllocSize(Ty)); // If this is a constant subscript, handle it quickly. - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (const ConstantInt *CI = dyn_cast(Idx)) { - if (CI->isZero()) continue; - uint64_t Offs = - DL->getTypeAllocSize(Ty)*cast(CI)->getSExtValue(); - SDValue OffsVal; - EVT PTy = TLI->getPointerTy(AS); - unsigned PtrBits = PTy.getSizeInBits(); - if (PtrBits < 64) - OffsVal = DAG.getNode(ISD::TRUNCATE, getCurSDLoc(), PTy, - DAG.getConstant(Offs, MVT::i64)); - else - OffsVal = DAG.getConstant(Offs, PTy); - - N = DAG.getNode(ISD::ADD, getCurSDLoc(), N.getValueType(), N, - OffsVal); + if (const auto *CI = dyn_cast(Idx)) { + if (CI->isZero()) + continue; + APInt Offs = ElementSize * CI->getValue().sextOrTrunc(PtrSize); + SDValue OffsVal = DAG.getConstant(Offs, PtrTy); + N = DAG.getNode(ISD::ADD, getCurSDLoc(), N.getValueType(), N, OffsVal); continue; } // N = N + Idx * ElementSize; - APInt ElementSize = APInt(TLI->getPointerSizeInBits(AS), - DL->getTypeAllocSize(Ty)); SDValue IdxN = getValue(Idx); // If the index is smaller or larger than intptr_t, truncate or extend @@ -3441,15 +3447,15 @@ void SelectionDAGBuilder::visitAlloca(const AllocaInst &I) { return; // getValue will auto-populate this. Type *Ty = I.getAllocatedType(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(Ty); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + uint64_t TySize = TLI.getDataLayout()->getTypeAllocSize(Ty); unsigned Align = - std::max((unsigned)TLI->getDataLayout()->getPrefTypeAlignment(Ty), - I.getAlignment()); + std::max((unsigned)TLI.getDataLayout()->getPrefTypeAlignment(Ty), + I.getAlignment()); SDValue AllocSize = getValue(I.getArraySize()); - EVT IntPtr = TLI->getPointerTy(); + EVT IntPtr = TLI.getPointerTy(); if (AllocSize.getValueType() != IntPtr) AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurSDLoc(), IntPtr); @@ -3461,7 +3467,7 @@ void SelectionDAGBuilder::visitAlloca(const AllocaInst &I) { // the stack alignment, ignore it. If the size is greater than or equal to // the stack alignment, we note this in the DYNAMIC_STACKALLOC node. unsigned StackAlign = - TM.getSubtargetImpl()->getFrameLowering()->getStackAlignment(); + DAG.getSubtarget().getFrameLowering()->getStackAlignment(); if (Align <= StackAlign) Align = 0; @@ -3495,18 +3501,18 @@ void SelectionDAGBuilder::visitLoad(const LoadInst &I) { Type *Ty = I.getType(); bool isVolatile = I.isVolatile(); - bool isNonTemporal = I.getMetadata("nontemporal") != nullptr; - bool isInvariant = I.getMetadata("invariant.load") != nullptr; + bool isNonTemporal = I.getMetadata(LLVMContext::MD_nontemporal) != nullptr; + bool isInvariant = I.getMetadata(LLVMContext::MD_invariant_load) != nullptr; unsigned Alignment = I.getAlignment(); AAMDNodes AAInfo; I.getAAMetadata(AAInfo); const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SmallVector ValueVTs; SmallVector Offsets; - ComputeValueVTs(*TM.getSubtargetImpl()->getTargetLowering(), Ty, ValueVTs, - &Offsets); + ComputeValueVTs(TLI, Ty, ValueVTs, &Offsets); unsigned NumValues = ValueVTs.size(); if (NumValues == 0) return; @@ -3526,9 +3532,8 @@ void SelectionDAGBuilder::visitLoad(const LoadInst &I) { Root = DAG.getRoot(); } - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); if (isVolatile) - Root = TLI->prepareVolatileOrAtomicLoad(Root, getCurSDLoc(), DAG); + Root = TLI.prepareVolatileOrAtomicLoad(Root, getCurSDLoc(), DAG); SmallVector Values(NumValues); SmallVector Chains(std::min(unsigned(MaxParallelChains), @@ -3583,7 +3588,7 @@ void SelectionDAGBuilder::visitStore(const StoreInst &I) { SmallVector ValueVTs; SmallVector Offsets; - ComputeValueVTs(*TM.getSubtargetImpl()->getTargetLowering(), SrcV->getType(), + ComputeValueVTs(DAG.getTargetLoweringInfo(), SrcV->getType(), ValueVTs, &Offsets); unsigned NumValues = ValueVTs.size(); if (NumValues == 0) @@ -3600,7 +3605,7 @@ void SelectionDAGBuilder::visitStore(const StoreInst &I) { NumValues)); EVT PtrVT = Ptr.getValueType(); bool isVolatile = I.isVolatile(); - bool isNonTemporal = I.getMetadata("nontemporal") != nullptr; + bool isNonTemporal = I.getMetadata(LLVMContext::MD_nontemporal) != nullptr; unsigned Alignment = I.getAlignment(); AAMDNodes AAInfo; @@ -3629,28 +3634,72 @@ void SelectionDAGBuilder::visitStore(const StoreInst &I) { DAG.setRoot(StoreNode); } -static SDValue InsertFenceForAtomic(SDValue Chain, AtomicOrdering Order, - SynchronizationScope Scope, - bool Before, SDLoc dl, - SelectionDAG &DAG, - const TargetLowering &TLI) { - // Fence, if necessary - if (Before) { - if (Order == AcquireRelease || Order == SequentiallyConsistent) - Order = Release; - else if (Order == Acquire || Order == Monotonic || Order == Unordered) - return Chain; - } else { - if (Order == AcquireRelease) - Order = Acquire; - else if (Order == Release || Order == Monotonic || Order == Unordered) - return Chain; +void SelectionDAGBuilder::visitMaskedStore(const CallInst &I) { + SDLoc sdl = getCurSDLoc(); + + // llvm.masked.store.*(Src0, Ptr, alignemt, Mask) + Value *PtrOperand = I.getArgOperand(1); + SDValue Ptr = getValue(PtrOperand); + SDValue Src0 = getValue(I.getArgOperand(0)); + SDValue Mask = getValue(I.getArgOperand(3)); + EVT VT = Src0.getValueType(); + unsigned Alignment = (cast(I.getArgOperand(2)))->getZExtValue(); + if (!Alignment) + Alignment = DAG.getEVTAlignment(VT); + + AAMDNodes AAInfo; + I.getAAMetadata(AAInfo); + + MachineMemOperand *MMO = + DAG.getMachineFunction(). + getMachineMemOperand(MachinePointerInfo(PtrOperand), + MachineMemOperand::MOStore, VT.getStoreSize(), + Alignment, AAInfo); + SDValue StoreNode = DAG.getMaskedStore(getRoot(), sdl, Src0, Ptr, Mask, VT, + MMO, false); + DAG.setRoot(StoreNode); + setValue(&I, StoreNode); +} + +void SelectionDAGBuilder::visitMaskedLoad(const CallInst &I) { + SDLoc sdl = getCurSDLoc(); + + // @llvm.masked.load.*(Ptr, alignment, Mask, Src0) + Value *PtrOperand = I.getArgOperand(0); + SDValue Ptr = getValue(PtrOperand); + SDValue Src0 = getValue(I.getArgOperand(3)); + SDValue Mask = getValue(I.getArgOperand(2)); + + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT VT = TLI.getValueType(I.getType()); + unsigned Alignment = (cast(I.getArgOperand(1)))->getZExtValue(); + if (!Alignment) + Alignment = DAG.getEVTAlignment(VT); + + AAMDNodes AAInfo; + I.getAAMetadata(AAInfo); + const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range); + + SDValue InChain = DAG.getRoot(); + if (AA->pointsToConstantMemory( + AliasAnalysis::Location(PtrOperand, + AA->getTypeStoreSize(I.getType()), + AAInfo))) { + // Do not serialize (non-volatile) loads of constant memory with anything. + InChain = DAG.getEntryNode(); } - SDValue Ops[3]; - Ops[0] = Chain; - Ops[1] = DAG.getConstant(Order, TLI.getPointerTy()); - Ops[2] = DAG.getConstant(Scope, TLI.getPointerTy()); - return DAG.getNode(ISD::ATOMIC_FENCE, dl, MVT::Other, Ops); + + MachineMemOperand *MMO = + DAG.getMachineFunction(). + getMachineMemOperand(MachinePointerInfo(PtrOperand), + MachineMemOperand::MOLoad, VT.getStoreSize(), + Alignment, AAInfo, Ranges); + + SDValue Load = DAG.getMaskedLoad(VT, sdl, InChain, Ptr, Mask, Src0, VT, MMO, + ISD::NON_EXTLOAD); + SDValue OutChain = Load.getValue(1); + DAG.setRoot(OutChain); + setValue(&I, Load); } void SelectionDAGBuilder::visitAtomicCmpXchg(const AtomicCmpXchgInst &I) { @@ -3661,27 +3710,16 @@ void SelectionDAGBuilder::visitAtomicCmpXchg(const AtomicCmpXchgInst &I) { SDValue InChain = getRoot(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (TLI->getInsertFencesForAtomic()) - InChain = InsertFenceForAtomic(InChain, SuccessOrder, Scope, true, dl, - DAG, *TLI); - MVT MemVT = getValue(I.getCompareOperand()).getSimpleValueType(); SDVTList VTs = DAG.getVTList(MemVT, MVT::i1, MVT::Other); SDValue L = DAG.getAtomicCmpSwap( ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS, dl, MemVT, VTs, InChain, getValue(I.getPointerOperand()), getValue(I.getCompareOperand()), getValue(I.getNewValOperand()), MachinePointerInfo(I.getPointerOperand()), - 0 /* Alignment */, - TLI->getInsertFencesForAtomic() ? Monotonic : SuccessOrder, - TLI->getInsertFencesForAtomic() ? Monotonic : FailureOrder, Scope); + /*Alignment=*/ 0, SuccessOrder, FailureOrder, Scope); SDValue OutChain = L.getValue(2); - if (TLI->getInsertFencesForAtomic()) - OutChain = InsertFenceForAtomic(OutChain, SuccessOrder, Scope, false, dl, - DAG, *TLI); - setValue(&I, L); DAG.setRoot(OutChain); } @@ -3708,38 +3746,28 @@ void SelectionDAGBuilder::visitAtomicRMW(const AtomicRMWInst &I) { SDValue InChain = getRoot(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - if (TLI->getInsertFencesForAtomic()) - InChain = InsertFenceForAtomic(InChain, Order, Scope, true, dl, - DAG, *TLI); - SDValue L = DAG.getAtomic(NT, dl, getValue(I.getValOperand()).getSimpleValueType(), InChain, getValue(I.getPointerOperand()), getValue(I.getValOperand()), - I.getPointerOperand(), 0 /* Alignment */, - TLI->getInsertFencesForAtomic() ? Monotonic : Order, - Scope); + I.getPointerOperand(), + /* Alignment=*/ 0, Order, Scope); SDValue OutChain = L.getValue(1); - if (TLI->getInsertFencesForAtomic()) - OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl, - DAG, *TLI); - setValue(&I, L); DAG.setRoot(OutChain); } void SelectionDAGBuilder::visitFence(const FenceInst &I) { SDLoc dl = getCurSDLoc(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SDValue Ops[3]; Ops[0] = getRoot(); - Ops[1] = DAG.getConstant(I.getOrdering(), TLI->getPointerTy()); - Ops[2] = DAG.getConstant(I.getSynchScope(), TLI->getPointerTy()); + Ops[1] = DAG.getConstant(I.getOrdering(), TLI.getPointerTy()); + Ops[2] = DAG.getConstant(I.getSynchScope(), TLI.getPointerTy()); DAG.setRoot(DAG.getNode(ISD::ATOMIC_FENCE, dl, MVT::Other, Ops)); } @@ -3750,8 +3778,8 @@ void SelectionDAGBuilder::visitAtomicLoad(const LoadInst &I) { SDValue InChain = getRoot(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT VT = TLI->getValueType(I.getType()); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT VT = TLI.getValueType(I.getType()); if (I.getAlignment() < VT.getSizeInBits() / 8) report_fatal_error("Cannot generate unaligned atomic load"); @@ -3765,19 +3793,14 @@ void SelectionDAGBuilder::visitAtomicLoad(const LoadInst &I) { I.getAlignment() ? I.getAlignment() : DAG.getEVTAlignment(VT)); - InChain = TLI->prepareVolatileOrAtomicLoad(InChain, dl, DAG); + InChain = TLI.prepareVolatileOrAtomicLoad(InChain, dl, DAG); SDValue L = DAG.getAtomic(ISD::ATOMIC_LOAD, dl, VT, VT, InChain, getValue(I.getPointerOperand()), MMO, - TLI->getInsertFencesForAtomic() ? Monotonic : Order, - Scope); + Order, Scope); SDValue OutChain = L.getValue(1); - if (TLI->getInsertFencesForAtomic()) - OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl, - DAG, *TLI); - setValue(&I, L); DAG.setRoot(OutChain); } @@ -3790,28 +3813,19 @@ void SelectionDAGBuilder::visitAtomicStore(const StoreInst &I) { SDValue InChain = getRoot(); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - EVT VT = TLI->getValueType(I.getValueOperand()->getType()); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + EVT VT = TLI.getValueType(I.getValueOperand()->getType()); if (I.getAlignment() < VT.getSizeInBits() / 8) report_fatal_error("Cannot generate unaligned atomic store"); - if (TLI->getInsertFencesForAtomic()) - InChain = InsertFenceForAtomic(InChain, Order, Scope, true, dl, - DAG, *TLI); - SDValue OutChain = DAG.getAtomic(ISD::ATOMIC_STORE, dl, VT, InChain, getValue(I.getPointerOperand()), getValue(I.getValueOperand()), I.getPointerOperand(), I.getAlignment(), - TLI->getInsertFencesForAtomic() ? Monotonic : Order, - Scope); - - if (TLI->getInsertFencesForAtomic()) - OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl, - DAG, *TLI); + Order, Scope); DAG.setRoot(OutChain); } @@ -3836,13 +3850,13 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I, // Info is set by getTgtMemInstrinsic TargetLowering::IntrinsicInfo Info; - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - bool IsTgtIntrinsic = TLI->getTgtMemIntrinsic(Info, I, Intrinsic); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I, Intrinsic); // Add the intrinsic ID as an integer operand if it's not a target intrinsic. if (!IsTgtIntrinsic || Info.opc == ISD::INTRINSIC_VOID || Info.opc == ISD::INTRINSIC_W_CHAIN) - Ops.push_back(DAG.getTargetConstant(Intrinsic, TLI->getPointerTy())); + Ops.push_back(DAG.getTargetConstant(Intrinsic, TLI.getPointerTy())); // Add all operands of the call to the operand list. for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) { @@ -3851,7 +3865,7 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I, } SmallVector ValueVTs; - ComputeValueVTs(*TLI, I.getType(), ValueVTs); + ComputeValueVTs(TLI, I.getType(), ValueVTs); if (HasChain) ValueVTs.push_back(MVT::Other); @@ -3866,7 +3880,7 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I, VTs, Ops, Info.memVT, MachinePointerInfo(Info.ptrVal, Info.offset), Info.align, Info.vol, - Info.readMem, Info.writeMem); + Info.readMem, Info.writeMem, Info.size); } else if (!HasChain) { Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, getCurSDLoc(), VTs, Ops); } else if (!I.getType()->isVoidTy()) { @@ -3885,7 +3899,7 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I, if (!I.getType()->isVoidTy()) { if (VectorType *PTy = dyn_cast(I.getType())) { - EVT VT = TLI->getValueType(PTy); + EVT VT = TLI.getValueType(PTy); Result = DAG.getNode(ISD::BITCAST, getCurSDLoc(), VT, Result); } @@ -3932,6 +3946,93 @@ getF32Constant(SelectionDAG &DAG, unsigned Flt) { MVT::f32); } +static SDValue getLimitedPrecisionExp2(SDValue t0, SDLoc dl, + SelectionDAG &DAG) { + // IntegerPartOfX = ((int32_t)(t0); + SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0); + + // FractionalPartOfX = t0 - (float)IntegerPartOfX; + SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX); + SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1); + + // IntegerPartOfX <<= 23; + IntegerPartOfX = DAG.getNode( + ISD::SHL, dl, MVT::i32, IntegerPartOfX, + DAG.getConstant(23, DAG.getTargetLoweringInfo().getPointerTy())); + + SDValue TwoToFractionalPartOfX; + if (LimitFloatPrecision <= 6) { + // For floating-point precision of 6: + // + // TwoToFractionalPartOfX = + // 0.997535578f + + // (0.735607626f + 0.252464424f * x) * x; + // + // error 0.0144103317, which is 6 bits + SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, + getF32Constant(DAG, 0x3e814304)); + SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, + getF32Constant(DAG, 0x3f3c50c8)); + SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); + TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, + getF32Constant(DAG, 0x3f7f5e7e)); + } else if (LimitFloatPrecision <= 12) { + // For floating-point precision of 12: + // + // TwoToFractionalPartOfX = + // 0.999892986f + + // (0.696457318f + + // (0.224338339f + 0.792043434e-1f * x) * x) * x; + // + // error 0.000107046256, which is 13 to 14 bits + SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, + getF32Constant(DAG, 0x3da235e3)); + SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, + getF32Constant(DAG, 0x3e65b8f3)); + SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); + SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, + getF32Constant(DAG, 0x3f324b07)); + SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); + TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, + getF32Constant(DAG, 0x3f7ff8fd)); + } else { // LimitFloatPrecision <= 18 + // For floating-point precision of 18: + // + // TwoToFractionalPartOfX = + // 0.999999982f + + // (0.693148872f + + // (0.240227044f + + // (0.554906021e-1f + + // (0.961591928e-2f + + // (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x; + // error 2.47208000*10^(-7), which is better than 18 bits + SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, + getF32Constant(DAG, 0x3924b03e)); + SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, + getF32Constant(DAG, 0x3ab24b87)); + SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); + SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, + getF32Constant(DAG, 0x3c1d8c17)); + SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); + SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, + getF32Constant(DAG, 0x3d634a1d)); + SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X); + SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8, + getF32Constant(DAG, 0x3e75fe14)); + SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X); + SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10, + getF32Constant(DAG, 0x3f317234)); + SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X); + TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t12, + getF32Constant(DAG, 0x3f800000)); + } + + // Add the exponent into the result in integer domain. + SDValue t13 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, TwoToFractionalPartOfX); + return DAG.getNode(ISD::BITCAST, dl, MVT::f32, + DAG.getNode(ISD::ADD, dl, MVT::i32, t13, IntegerPartOfX)); +} + /// expandExp - Lower an exp intrinsic. Handles the special sequences for /// limited-precision mode. static SDValue expandExp(SDLoc dl, SDValue Op, SelectionDAG &DAG, @@ -3943,92 +4044,10 @@ static SDValue expandExp(SDLoc dl, SDValue Op, SelectionDAG &DAG, // final result: // // #define LOG2OFe 1.4426950f - // IntegerPartOfX = ((int32_t)(X * LOG2OFe)); + // t0 = Op * LOG2OFe SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op, getF32Constant(DAG, 0x3fb8aa3b)); - SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0); - - // FractionalPartOfX = (X * LOG2OFe) - (float)IntegerPartOfX; - SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX); - SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1); - - // IntegerPartOfX <<= 23; - IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX, - DAG.getConstant(23, TLI.getPointerTy())); - - SDValue TwoToFracPartOfX; - if (LimitFloatPrecision <= 6) { - // For floating-point precision of 6: - // - // TwoToFractionalPartOfX = - // 0.997535578f + - // (0.735607626f + 0.252464424f * x) * x; - // - // error 0.0144103317, which is 6 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3e814304)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3f3c50c8)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - TwoToFracPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f7f5e7e)); - } else if (LimitFloatPrecision <= 12) { - // For floating-point precision of 12: - // - // TwoToFractionalPartOfX = - // 0.999892986f + - // (0.696457318f + - // (0.224338339f + 0.792043434e-1f * x) * x) * x; - // - // 0.000107046256 error, which is 13 to 14 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3da235e3)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3e65b8f3)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f324b07)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - TwoToFracPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3f7ff8fd)); - } else { // LimitFloatPrecision <= 18 - // For floating-point precision of 18: - // - // TwoToFractionalPartOfX = - // 0.999999982f + - // (0.693148872f + - // (0.240227044f + - // (0.554906021e-1f + - // (0.961591928e-2f + - // (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x; - // - // error 2.47208000*10^(-7), which is better than 18 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3924b03e)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3ab24b87)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3c1d8c17)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3d634a1d)); - SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X); - SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8, - getF32Constant(DAG, 0x3e75fe14)); - SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X); - SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10, - getF32Constant(DAG, 0x3f317234)); - SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X); - TwoToFracPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t12, - getF32Constant(DAG, 0x3f800000)); - } - - // Add the exponent into the result in integer domain. - SDValue t13 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, TwoToFracPartOfX); - return DAG.getNode(ISD::BITCAST, dl, MVT::f32, - DAG.getNode(ISD::ADD, dl, MVT::i32, - t13, IntegerPartOfX)); + return getLimitedPrecisionExp2(t0, dl, DAG); } // No special expansion. @@ -4319,91 +4338,8 @@ static SDValue expandLog10(SDLoc dl, SDValue Op, SelectionDAG &DAG, static SDValue expandExp2(SDLoc dl, SDValue Op, SelectionDAG &DAG, const TargetLowering &TLI) { if (Op.getValueType() == MVT::f32 && - LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) { - SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Op); - - // FractionalPartOfX = x - (float)IntegerPartOfX; - SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX); - SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, Op, t1); - - // IntegerPartOfX <<= 23; - IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX, - DAG.getConstant(23, TLI.getPointerTy())); - - SDValue TwoToFractionalPartOfX; - if (LimitFloatPrecision <= 6) { - // For floating-point precision of 6: - // - // TwoToFractionalPartOfX = - // 0.997535578f + - // (0.735607626f + 0.252464424f * x) * x; - // - // error 0.0144103317, which is 6 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3e814304)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3f3c50c8)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f7f5e7e)); - } else if (LimitFloatPrecision <= 12) { - // For floating-point precision of 12: - // - // TwoToFractionalPartOfX = - // 0.999892986f + - // (0.696457318f + - // (0.224338339f + 0.792043434e-1f * x) * x) * x; - // - // error 0.000107046256, which is 13 to 14 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3da235e3)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3e65b8f3)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f324b07)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3f7ff8fd)); - } else { // LimitFloatPrecision <= 18 - // For floating-point precision of 18: - // - // TwoToFractionalPartOfX = - // 0.999999982f + - // (0.693148872f + - // (0.240227044f + - // (0.554906021e-1f + - // (0.961591928e-2f + - // (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x; - // error 2.47208000*10^(-7), which is better than 18 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3924b03e)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3ab24b87)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3c1d8c17)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3d634a1d)); - SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X); - SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8, - getF32Constant(DAG, 0x3e75fe14)); - SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X); - SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10, - getF32Constant(DAG, 0x3f317234)); - SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t12, - getF32Constant(DAG, 0x3f800000)); - } - - // Add the exponent into the result in integer domain. - SDValue t13 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, - TwoToFractionalPartOfX); - return DAG.getNode(ISD::BITCAST, dl, MVT::f32, - DAG.getNode(ISD::ADD, dl, MVT::i32, - t13, IntegerPartOfX)); - } + LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) + return getLimitedPrecisionExp2(Op, dl, DAG); // No special expansion. return DAG.getNode(ISD::FEXP2, dl, Op.getValueType(), Op); @@ -4427,90 +4363,10 @@ static SDValue expandPow(SDLoc dl, SDValue LHS, SDValue RHS, // final result: // // #define LOG2OF10 3.3219281f - // IntegerPartOfX = (int32_t)(x * LOG2OF10); + // t0 = Op * LOG2OF10; SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, RHS, getF32Constant(DAG, 0x40549a78)); - SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0); - - // FractionalPartOfX = x - (float)IntegerPartOfX; - SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX); - SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1); - - // IntegerPartOfX <<= 23; - IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX, - DAG.getConstant(23, TLI.getPointerTy())); - - SDValue TwoToFractionalPartOfX; - if (LimitFloatPrecision <= 6) { - // For floating-point precision of 6: - // - // twoToFractionalPartOfX = - // 0.997535578f + - // (0.735607626f + 0.252464424f * x) * x; - // - // error 0.0144103317, which is 6 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3e814304)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3f3c50c8)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f7f5e7e)); - } else if (LimitFloatPrecision <= 12) { - // For floating-point precision of 12: - // - // TwoToFractionalPartOfX = - // 0.999892986f + - // (0.696457318f + - // (0.224338339f + 0.792043434e-1f * x) * x) * x; - // - // error 0.000107046256, which is 13 to 14 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3da235e3)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3e65b8f3)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3f324b07)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3f7ff8fd)); - } else { // LimitFloatPrecision <= 18 - // For floating-point precision of 18: - // - // TwoToFractionalPartOfX = - // 0.999999982f + - // (0.693148872f + - // (0.240227044f + - // (0.554906021e-1f + - // (0.961591928e-2f + - // (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x; - // error 2.47208000*10^(-7), which is better than 18 bits - SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X, - getF32Constant(DAG, 0x3924b03e)); - SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2, - getF32Constant(DAG, 0x3ab24b87)); - SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X); - SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4, - getF32Constant(DAG, 0x3c1d8c17)); - SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X); - SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6, - getF32Constant(DAG, 0x3d634a1d)); - SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X); - SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8, - getF32Constant(DAG, 0x3e75fe14)); - SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X); - SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10, - getF32Constant(DAG, 0x3f317234)); - SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X); - TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t12, - getF32Constant(DAG, 0x3f800000)); - } - - SDValue t13 = DAG.getNode(ISD::BITCAST, dl,MVT::i32,TwoToFractionalPartOfX); - return DAG.getNode(ISD::BITCAST, dl, MVT::f32, - DAG.getNode(ISD::ADD, dl, MVT::i32, - t13, IntegerPartOfX)); + return getLimitedPrecisionExp2(t0, dl, DAG); } // No special expansion. @@ -4535,11 +4391,10 @@ static SDValue ExpandPowI(SDLoc DL, SDValue LHS, SDValue RHS, return DAG.getConstantFP(1.0, LHS.getValueType()); const Function *F = DAG.getMachineFunction().getFunction(); - if (!F->getAttributes().hasAttribute(AttributeSet::FunctionIndex, - Attribute::OptimizeForSize) || + if (!F->hasFnAttribute(Attribute::OptimizeForSize) || // If optimizing for size, don't insert too many multiplies. This // inserts up to 5 multiplies. - CountPopulation_32(Val)+Log2_32(Val) < 7) { + countPopulation(Val) + Log2_32(Val) < 7) { // We use the simple binary decomposition method to generate the multiply // sequence. There are more optimal ways to do this (for example, // powi(x,15) generates one more multiply than it should), but this has @@ -4592,10 +4447,11 @@ static unsigned getTruncatedArgReg(const SDValue &N) { /// EmitFuncArgumentDbgValue - If the DbgValueInst is a dbg_value of a function /// argument, create the corresponding DBG_VALUE machine instruction for it now. /// At the end of instruction selection, they will be inserted to the entry BB. -bool -SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable, - int64_t Offset, bool IsIndirect, - const SDValue &N) { +bool SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, + MDNode *Variable, + MDNode *Expr, int64_t Offset, + bool IsIndirect, + const SDValue &N) { const Argument *Arg = dyn_cast(V); if (!Arg) return false; @@ -4647,14 +4503,16 @@ SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable, return false; if (Op->isReg()) - FuncInfo.ArgDbgValues.push_back(BuildMI(MF, getCurDebugLoc(), - TII->get(TargetOpcode::DBG_VALUE), - IsIndirect, - Op->getReg(), Offset, Variable)); + FuncInfo.ArgDbgValues.push_back( + BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE), + IsIndirect, Op->getReg(), Offset, Variable, Expr)); else FuncInfo.ArgDbgValues.push_back( - BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE)) - .addOperand(*Op).addImm(Offset).addMetadata(Variable)); + BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE)) + .addOperand(*Op) + .addImm(Offset) + .addMetadata(Variable) + .addMetadata(Expr)); return true; } @@ -4672,7 +4530,7 @@ SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable, /// otherwise lower it and return null. const char * SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SDLoc sdl = getCurSDLoc(); DebugLoc dl = getCurDebugLoc(); SDValue Res; @@ -4686,18 +4544,18 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { case Intrinsic::vaend: visitVAEnd(I); return nullptr; case Intrinsic::vacopy: visitVACopy(I); return nullptr; case Intrinsic::returnaddress: - setValue(&I, DAG.getNode(ISD::RETURNADDR, sdl, TLI->getPointerTy(), + setValue(&I, DAG.getNode(ISD::RETURNADDR, sdl, TLI.getPointerTy(), getValue(I.getArgOperand(0)))); return nullptr; case Intrinsic::frameaddress: - setValue(&I, DAG.getNode(ISD::FRAMEADDR, sdl, TLI->getPointerTy(), + setValue(&I, DAG.getNode(ISD::FRAMEADDR, sdl, TLI.getPointerTy(), getValue(I.getArgOperand(0)))); return nullptr; case Intrinsic::read_register: { Value *Reg = I.getArgOperand(0); - SDValue RegName = DAG.getMDNode(cast(Reg)); - EVT VT = - TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType()); + SDValue RegName = + DAG.getMDNode(cast(cast(Reg)->getMetadata())); + EVT VT = TLI.getValueType(I.getType()); setValue(&I, DAG.getNode(ISD::READ_REGISTER, sdl, VT, RegName)); return nullptr; } @@ -4705,16 +4563,18 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { Value *Reg = I.getArgOperand(0); Value *RegValue = I.getArgOperand(1); SDValue Chain = getValue(RegValue).getOperand(0); - SDValue RegName = DAG.getMDNode(cast(Reg)); + SDValue RegName = + DAG.getMDNode(cast(cast(Reg)->getMetadata())); DAG.setRoot(DAG.getNode(ISD::WRITE_REGISTER, sdl, MVT::Other, Chain, RegName, getValue(RegValue))); return nullptr; } case Intrinsic::setjmp: - return &"_setjmp"[!TLI->usesUnderscoreSetJmp()]; + return &"_setjmp"[!TLI.usesUnderscoreSetJmp()]; case Intrinsic::longjmp: - return &"_longjmp"[!TLI->usesUnderscoreLongJmp()]; + return &"_longjmp"[!TLI.usesUnderscoreLongJmp()]; case Intrinsic::memcpy: { + // FIXME: this definition of "user defined address space" is x86-specific // Assert for address < 256 since we support only user defined address // spaces. assert(cast(I.getArgOperand(0)->getType())->getAddressSpace() @@ -4735,6 +4595,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; } case Intrinsic::memset: { + // FIXME: this definition of "user defined address space" is x86-specific // Assert for address < 256 since we support only user defined address // spaces. assert(cast(I.getArgOperand(0)->getType())->getAddressSpace() @@ -4752,6 +4613,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; } case Intrinsic::memmove: { + // FIXME: this definition of "user defined address space" is x86-specific // Assert for address < 256 since we support only user defined address // spaces. assert(cast(I.getArgOperand(0)->getType())->getAddressSpace() @@ -4774,6 +4636,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { case Intrinsic::dbg_declare: { const DbgDeclareInst &DI = cast(I); MDNode *Variable = DI.getVariable(); + MDNode *Expression = DI.getExpression(); const Value *Address = DI.getAddress(); DIVariable DIVar(Variable); assert((!DIVar || DIVar.isVariable()) && @@ -4809,16 +4672,16 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { FrameIndexSDNode *FINode = dyn_cast(N.getNode()); if (FINode) // Byval parameter. We have a frame index at this point. - SDV = DAG.getFrameIndexDbgValue(Variable, FINode->getIndex(), - 0, dl, SDNodeOrder); + SDV = DAG.getFrameIndexDbgValue( + Variable, Expression, FINode->getIndex(), 0, dl, SDNodeOrder); else { // Address is an argument, so try to emit its dbg value using // virtual register info from the FuncInfo.ValueMap. - EmitFuncArgumentDbgValue(Address, Variable, 0, false, N); + EmitFuncArgumentDbgValue(Address, Variable, Expression, 0, false, N); return nullptr; } } else if (AI) - SDV = DAG.getDbgValue(Variable, N.getNode(), N.getResNo(), + SDV = DAG.getDbgValue(Variable, Expression, N.getNode(), N.getResNo(), true, 0, dl, SDNodeOrder); else { // Can't do anything with other non-AI cases yet. @@ -4831,7 +4694,8 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { } else { // If Address is an argument then try to emit its dbg value using // virtual register info from the FuncInfo.ValueMap. - if (!EmitFuncArgumentDbgValue(Address, Variable, 0, false, N)) { + if (!EmitFuncArgumentDbgValue(Address, Variable, Expression, 0, false, + N)) { // If variable is pinned by a alloca in dominating bb then // use StaticAllocaMap. if (const AllocaInst *AI = dyn_cast(Address)) { @@ -4839,7 +4703,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { DenseMap::iterator SI = FuncInfo.StaticAllocaMap.find(AI); if (SI != FuncInfo.StaticAllocaMap.end()) { - SDV = DAG.getFrameIndexDbgValue(Variable, SI->second, + SDV = DAG.getFrameIndexDbgValue(Variable, Expression, SI->second, 0, dl, SDNodeOrder); DAG.AddDbgValue(SDV, nullptr, false); return nullptr; @@ -4860,6 +4724,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; MDNode *Variable = DI.getVariable(); + MDNode *Expression = DI.getExpression(); uint64_t Offset = DI.getOffset(); const Value *V = DI.getValue(); if (!V) @@ -4867,7 +4732,8 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { SDDbgValue *SDV; if (isa(V) || isa(V) || isa(V)) { - SDV = DAG.getConstantDbgValue(Variable, V, Offset, dl, SDNodeOrder); + SDV = DAG.getConstantDbgValue(Variable, Expression, V, Offset, dl, + SDNodeOrder); DAG.AddDbgValue(SDV, nullptr, false); } else { // Do not use getValue() in here; we don't want to generate code at @@ -4879,10 +4745,10 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { if (N.getNode()) { // A dbg.value for an alloca is always indirect. bool IsIndirect = isa(V) || Offset != 0; - if (!EmitFuncArgumentDbgValue(V, Variable, Offset, IsIndirect, N)) { - SDV = DAG.getDbgValue(Variable, N.getNode(), - N.getResNo(), IsIndirect, - Offset, dl, SDNodeOrder); + if (!EmitFuncArgumentDbgValue(V, Variable, Expression, Offset, + IsIndirect, N)) { + SDV = DAG.getDbgValue(Variable, Expression, N.getNode(), N.getResNo(), + IsIndirect, Offset, dl, SDNodeOrder); DAG.AddDbgValue(SDV, N.getNode(), false); } } else if (!V->use_empty() ) { @@ -4916,7 +4782,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { case Intrinsic::eh_typeid_for: { // Find the type id for the given typeinfo. - GlobalVariable *GV = ExtractTypeInfo(I.getArgOperand(0)); + GlobalValue *GV = ExtractTypeInfo(I.getArgOperand(0)); unsigned TypeID = DAG.getMachineFunction().getMMI().getTypeIDFor(GV); Res = DAG.getConstant(TypeID, MVT::i32); setValue(&I, Res); @@ -4937,15 +4803,14 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; case Intrinsic::eh_dwarf_cfa: { SDValue CfaArg = DAG.getSExtOrTrunc(getValue(I.getArgOperand(0)), sdl, - TLI->getPointerTy()); + TLI.getPointerTy()); SDValue Offset = DAG.getNode(ISD::ADD, sdl, CfaArg.getValueType(), DAG.getNode(ISD::FRAME_TO_ARGS_OFFSET, sdl, CfaArg.getValueType()), CfaArg); - SDValue FA = DAG.getNode(ISD::FRAMEADDR, sdl, - TLI->getPointerTy(), - DAG.getConstant(0, TLI->getPointerTy())); + SDValue FA = DAG.getNode(ISD::FRAMEADDR, sdl, TLI.getPointerTy(), + DAG.getConstant(0, TLI.getPointerTy())); setValue(&I, DAG.getNode(ISD::ADD, sdl, FA.getValueType(), FA, Offset)); return nullptr; @@ -4984,6 +4849,12 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; } + case Intrinsic::masked_load: + visitMaskedLoad(I); + return nullptr; + case Intrinsic::masked_store: + visitMaskedStore(I); + return nullptr; case Intrinsic::x86_mmx_pslli_w: case Intrinsic::x86_mmx_pslli_d: case Intrinsic::x86_mmx_pslli_q: @@ -5035,7 +4906,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { ShOps[0] = ShAmt; ShOps[1] = DAG.getConstant(0, MVT::i32); ShAmt = DAG.getNode(ISD::BUILD_VECTOR, sdl, ShAmtVT, ShOps); - EVT DestVT = TLI->getValueType(I.getType()); + EVT DestVT = TLI.getValueType(I.getType()); ShAmt = DAG.getNode(ISD::BITCAST, sdl, DestVT, ShAmt); Res = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, sdl, DestVT, DAG.getConstant(NewIntrinsic, MVT::i32), @@ -5043,34 +4914,6 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { setValue(&I, Res); return nullptr; } - case Intrinsic::x86_avx_vinsertf128_pd_256: - case Intrinsic::x86_avx_vinsertf128_ps_256: - case Intrinsic::x86_avx_vinsertf128_si_256: - case Intrinsic::x86_avx2_vinserti128: { - EVT DestVT = TLI->getValueType(I.getType()); - EVT ElVT = TLI->getValueType(I.getArgOperand(1)->getType()); - uint64_t Idx = (cast(I.getArgOperand(2))->getZExtValue() & 1) * - ElVT.getVectorNumElements(); - Res = DAG.getNode(ISD::INSERT_SUBVECTOR, sdl, DestVT, - getValue(I.getArgOperand(0)), - getValue(I.getArgOperand(1)), - DAG.getConstant(Idx, TLI->getVectorIdxTy())); - setValue(&I, Res); - return nullptr; - } - case Intrinsic::x86_avx_vextractf128_pd_256: - case Intrinsic::x86_avx_vextractf128_ps_256: - case Intrinsic::x86_avx_vextractf128_si_256: - case Intrinsic::x86_avx2_vextracti128: { - EVT DestVT = TLI->getValueType(I.getType()); - uint64_t Idx = (cast(I.getArgOperand(1))->getZExtValue() & 1) * - DestVT.getVectorNumElements(); - Res = DAG.getNode(ISD::EXTRACT_SUBVECTOR, sdl, DestVT, - getValue(I.getArgOperand(0)), - DAG.getConstant(Idx, TLI->getVectorIdxTy())); - setValue(&I, Res); - return nullptr; - } case Intrinsic::convertff: case Intrinsic::convertfsi: case Intrinsic::convertfui: @@ -5093,7 +4936,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { case Intrinsic::convertus: Code = ISD::CVT_US; break; case Intrinsic::convertuu: Code = ISD::CVT_UU; break; } - EVT DestVT = TLI->getValueType(I.getType()); + EVT DestVT = TLI.getValueType(I.getType()); const Value *Op1 = I.getArgOperand(0); Res = DAG.getConvertRndSat(DestVT, sdl, getValue(Op1), DAG.getValueType(DestVT), @@ -5109,23 +4952,23 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { getValue(I.getArgOperand(1)), DAG)); return nullptr; case Intrinsic::log: - setValue(&I, expandLog(sdl, getValue(I.getArgOperand(0)), DAG, *TLI)); + setValue(&I, expandLog(sdl, getValue(I.getArgOperand(0)), DAG, TLI)); return nullptr; case Intrinsic::log2: - setValue(&I, expandLog2(sdl, getValue(I.getArgOperand(0)), DAG, *TLI)); + setValue(&I, expandLog2(sdl, getValue(I.getArgOperand(0)), DAG, TLI)); return nullptr; case Intrinsic::log10: - setValue(&I, expandLog10(sdl, getValue(I.getArgOperand(0)), DAG, *TLI)); + setValue(&I, expandLog10(sdl, getValue(I.getArgOperand(0)), DAG, TLI)); return nullptr; case Intrinsic::exp: - setValue(&I, expandExp(sdl, getValue(I.getArgOperand(0)), DAG, *TLI)); + setValue(&I, expandExp(sdl, getValue(I.getArgOperand(0)), DAG, TLI)); return nullptr; case Intrinsic::exp2: - setValue(&I, expandExp2(sdl, getValue(I.getArgOperand(0)), DAG, *TLI)); + setValue(&I, expandExp2(sdl, getValue(I.getArgOperand(0)), DAG, TLI)); return nullptr; case Intrinsic::pow: setValue(&I, expandPow(sdl, getValue(I.getArgOperand(0)), - getValue(I.getArgOperand(1)), DAG, *TLI)); + getValue(I.getArgOperand(1)), DAG, TLI)); return nullptr; case Intrinsic::sqrt: case Intrinsic::fabs: @@ -5157,6 +5000,18 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { getValue(I.getArgOperand(0)))); return nullptr; } + case Intrinsic::minnum: + setValue(&I, DAG.getNode(ISD::FMINNUM, sdl, + getValue(I.getArgOperand(0)).getValueType(), + getValue(I.getArgOperand(0)), + getValue(I.getArgOperand(1)))); + return nullptr; + case Intrinsic::maxnum: + setValue(&I, DAG.getNode(ISD::FMAXNUM, sdl, + getValue(I.getArgOperand(0)).getValueType(), + getValue(I.getArgOperand(0)), + getValue(I.getArgOperand(1)))); + return nullptr; case Intrinsic::copysign: setValue(&I, DAG.getNode(ISD::FCOPYSIGN, sdl, getValue(I.getArgOperand(0)).getValueType(), @@ -5171,9 +5026,9 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { getValue(I.getArgOperand(2)))); return nullptr; case Intrinsic::fmuladd: { - EVT VT = TLI->getValueType(I.getType()); + EVT VT = TLI.getValueType(I.getType()); if (TM.Options.AllowFPOpFusion != FPOpFusion::Strict && - TLI->isFMAFasterThanFMulAndFAdd(VT)) { + TLI.isFMAFasterThanFMulAndFAdd(VT)) { setValue(&I, DAG.getNode(ISD::FMA, sdl, getValue(I.getArgOperand(0)).getValueType(), getValue(I.getArgOperand(0)), @@ -5200,7 +5055,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; case Intrinsic::convert_from_fp16: setValue(&I, - DAG.getNode(ISD::FP_EXTEND, sdl, TLI->getValueType(I.getType()), + DAG.getNode(ISD::FP_EXTEND, sdl, TLI.getValueType(I.getType()), DAG.getNode(ISD::BITCAST, sdl, MVT::f16, getValue(I.getArgOperand(0))))); return nullptr; @@ -5247,7 +5102,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { case Intrinsic::stacksave: { SDValue Op = getRoot(); Res = DAG.getNode(ISD::STACKSAVE, sdl, - DAG.getVTList(TLI->getPointerTy(), MVT::Other), Op); + DAG.getVTList(TLI.getPointerTy(), MVT::Other), Op); setValue(&I, Res); DAG.setRoot(Res.getValue(1)); return nullptr; @@ -5261,15 +5116,23 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { // Emit code into the DAG to store the stack guard onto the stack. MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo *MFI = MF.getFrameInfo(); - EVT PtrTy = TLI->getPointerTy(); + EVT PtrTy = TLI.getPointerTy(); SDValue Src, Chain = getRoot(); + const Value *Ptr = cast(I.getArgOperand(0))->getPointerOperand(); + const GlobalVariable *GV = dyn_cast(Ptr); + + // See if Ptr is a bitcast. If it is, look through it and see if we can get + // global variable __stack_chk_guard. + if (!GV) + if (const Operator *BC = dyn_cast(Ptr)) + if (BC->getOpcode() == Instruction::BitCast) + GV = dyn_cast(BC->getOperand(0)); - if (TLI->useLoadStackGuardNode()) { + if (GV && TLI.useLoadStackGuardNode()) { // Emit a LOAD_STACK_GUARD node. MachineSDNode *Node = DAG.getMachineNode(TargetOpcode::LOAD_STACK_GUARD, sdl, PtrTy, Chain); - LoadInst *LI = cast(I.getArgOperand(0)); - MachinePointerInfo MPInfo(LI->getPointerOperand()); + MachinePointerInfo MPInfo(GV); MachineInstr::mmo_iterator MemRefs = MF.allocateMemRefsArray(1); unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOInvariant; @@ -5282,7 +5145,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { // retrieved in the epilogue. Src = SDValue(Node, 0); const TargetRegisterClass *RC = - TLI->getRegClassFor(Src.getSimpleValueType()); + TLI.getRegClassFor(Src.getSimpleValueType()); unsigned Reg = MF.getRegInfo().createVirtualRegister(RC); SPDescriptor.setGuardReg(Reg); @@ -5351,7 +5214,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { } case Intrinsic::adjust_trampoline: { setValue(&I, DAG.getNode(ISD::ADJUST_TRAMPOLINE, sdl, - TLI->getPointerTy(), + TLI.getPointerTy(), getValue(I.getArgOperand(0)))); return nullptr; } @@ -5391,10 +5254,10 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { TargetLowering::CallLoweringInfo CLI(DAG); CLI.setDebugLoc(sdl).setChain(getRoot()) .setCallee(CallingConv::C, I.getType(), - DAG.getExternalSymbol(TrapFuncName.data(), TLI->getPointerTy()), + DAG.getExternalSymbol(TrapFuncName.data(), TLI.getPointerTy()), std::move(Args), 0); - std::pair Result = TLI->LowerCallTo(CLI); + std::pair Result = TLI.LowerCallTo(CLI); DAG.setRoot(Result.second); return nullptr; } @@ -5448,7 +5311,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; SmallVector Allocas; - GetUnderlyingObjects(I.getArgOperand(1), Allocas, DL); + GetUnderlyingObjects(I.getArgOperand(1), Allocas, *DL); for (SmallVectorImpl::iterator Object = Allocas.begin(), E = Allocas.end(); Object != E; ++Object) { @@ -5458,11 +5321,17 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { if (!LifetimeObject) continue; - int FI = FuncInfo.StaticAllocaMap[LifetimeObject]; + // First check that the Alloca is static, otherwise it won't have a + // valid frame index. + auto SI = FuncInfo.StaticAllocaMap.find(LifetimeObject); + if (SI == FuncInfo.StaticAllocaMap.end()) + return nullptr; + + int FI = SI->second; SDValue Ops[2]; Ops[0] = getRoot(); - Ops[1] = DAG.getFrameIndex(FI, TLI->getPointerTy(), true); + Ops[1] = DAG.getFrameIndex(FI, TLI.getPointerTy(), true); unsigned Opcode = (IsStart ? ISD::LIFETIME_START : ISD::LIFETIME_END); Res = DAG.getNode(Opcode, sdl, MVT::Other, Ops); @@ -5472,7 +5341,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { } case Intrinsic::invariant_start: // Discard region information. - setValue(&I, DAG.getUNDEF(TLI->getPointerTy())); + setValue(&I, DAG.getUNDEF(TLI.getPointerTy())); return nullptr; case Intrinsic::invariant_end: // Discard region information. @@ -5490,7 +5359,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { return nullptr; } case Intrinsic::clear_cache: - return TLI->getClearCacheBuiltinName(); + return TLI.getClearCacheBuiltinName(); case Intrinsic::donothing: // ignore return nullptr; @@ -5500,41 +5369,91 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) { } case Intrinsic::experimental_patchpoint_void: case Intrinsic::experimental_patchpoint_i64: { - visitPatchpoint(I); + visitPatchpoint(&I); return nullptr; } + case Intrinsic::experimental_gc_statepoint: { + visitStatepoint(I); + return nullptr; } -} - -void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee, - bool isTailCall, - MachineBasicBlock *LandingPad) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - PointerType *PT = cast(CS.getCalledValue()->getType()); - FunctionType *FTy = cast(PT->getElementType()); - Type *RetTy = FTy->getReturnType(); - MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI(); - MCSymbol *BeginLabel = nullptr; - - TargetLowering::ArgListTy Args; - TargetLowering::ArgListEntry Entry; - Args.reserve(CS.arg_size()); + case Intrinsic::experimental_gc_result_int: + case Intrinsic::experimental_gc_result_float: + case Intrinsic::experimental_gc_result_ptr: + case Intrinsic::experimental_gc_result: { + visitGCResult(I); + return nullptr; + } + case Intrinsic::experimental_gc_relocate: { + visitGCRelocate(I); + return nullptr; + } + case Intrinsic::instrprof_increment: + llvm_unreachable("instrprof failed to lower an increment"); - for (ImmutableCallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end(); - i != e; ++i) { - const Value *V = *i; + case Intrinsic::frameescape: { + MachineFunction &MF = DAG.getMachineFunction(); + const TargetInstrInfo *TII = DAG.getSubtarget().getInstrInfo(); + + // Directly emit some FRAME_ALLOC machine instrs. Label assignment emission + // is the same on all targets. + for (unsigned Idx = 0, E = I.getNumArgOperands(); Idx < E; ++Idx) { + AllocaInst *Slot = + cast(I.getArgOperand(Idx)->stripPointerCasts()); + assert(FuncInfo.StaticAllocaMap.count(Slot) && + "can only escape static allocas"); + int FI = FuncInfo.StaticAllocaMap[Slot]; + MCSymbol *FrameAllocSym = + MF.getMMI().getContext().getOrCreateFrameAllocSymbol(MF.getName(), + Idx); + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, dl, + TII->get(TargetOpcode::FRAME_ALLOC)) + .addSym(FrameAllocSym) + .addFrameIndex(FI); + } - // Skip empty types - if (V->getType()->isEmptyTy()) - continue; + return nullptr; + } - SDValue ArgNode = getValue(V); - Entry.Node = ArgNode; Entry.Ty = V->getType(); + case Intrinsic::framerecover: { + // i8* @llvm.framerecover(i8* %fn, i8* %fp, i32 %idx) + MachineFunction &MF = DAG.getMachineFunction(); + MVT PtrVT = TLI.getPointerTy(0); + + // Get the symbol that defines the frame offset. + auto *Fn = cast(I.getArgOperand(0)->stripPointerCasts()); + auto *Idx = cast(I.getArgOperand(2)); + unsigned IdxVal = unsigned(Idx->getLimitedValue(INT_MAX)); + MCSymbol *FrameAllocSym = + MF.getMMI().getContext().getOrCreateFrameAllocSymbol(Fn->getName(), + IdxVal); + + // Create a TargetExternalSymbol for the label to avoid any target lowering + // that would make this PC relative. + StringRef Name = FrameAllocSym->getName(); + assert(Name.data()[Name.size()] == '\0' && "not null terminated"); + SDValue OffsetSym = DAG.getTargetExternalSymbol(Name.data(), PtrVT); + SDValue OffsetVal = + DAG.getNode(ISD::FRAME_ALLOC_RECOVER, sdl, PtrVT, OffsetSym); + + // Add the offset to the FP. + Value *FP = I.getArgOperand(1); + SDValue FPVal = getValue(FP); + SDValue Add = DAG.getNode(ISD::ADD, sdl, PtrVT, FPVal, OffsetVal); + setValue(&I, Add); - // Skip the first return-type Attribute to get to params. - Entry.setAttributes(&CS, i - CS.arg_begin() + 1); - Args.push_back(Entry); + return nullptr; + } + case Intrinsic::eh_begincatch: + case Intrinsic::eh_endcatch: + llvm_unreachable("begin/end catch intrinsics not lowered in codegen"); } +} + +std::pair +SelectionDAGBuilder::lowerInvokable(TargetLowering::CallLoweringInfo &CLI, + MachineBasicBlock *LandingPad) { + MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI(); + MCSymbol *BeginLabel = nullptr; if (LandingPad) { // Insert a label before the invoke call to mark the try range. This can be @@ -5556,24 +5475,16 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee, // this call might not return. (void)getRoot(); DAG.setRoot(DAG.getEHLabel(getCurSDLoc(), getControlRoot(), BeginLabel)); - } - // Check if target-independent constraints permit a tail call here. - // Target-dependent constraints are checked within TLI->LowerCallTo. - if (isTailCall && !isInTailCallPosition(CS, DAG.getTarget())) - isTailCall = false; - - TargetLowering::CallLoweringInfo CLI(DAG); - CLI.setDebugLoc(getCurSDLoc()).setChain(getRoot()) - .setCallee(RetTy, FTy, Callee, std::move(Args), CS).setTailCall(isTailCall); + CLI.setChain(getRoot()); + } + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + std::pair Result = TLI.LowerCallTo(CLI); - std::pair Result = TLI->LowerCallTo(CLI); - assert((isTailCall || Result.second.getNode()) && + assert((CLI.IsTailCall || Result.second.getNode()) && "Non-null chain expected with non-tail call!"); assert((Result.second.getNode() || !Result.first.getNode()) && "Null value expected with tail call!"); - if (Result.first.getNode()) - setValue(CS.getInstruction(), Result.first); if (!Result.second.getNode()) { // As a special case, a null chain means that a tail call has been emitted @@ -5596,6 +5507,50 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee, // Inform MachineModuleInfo of range. MMI.addInvoke(LandingPad, BeginLabel, EndLabel); } + + return Result; +} + +void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee, + bool isTailCall, + MachineBasicBlock *LandingPad) { + PointerType *PT = cast(CS.getCalledValue()->getType()); + FunctionType *FTy = cast(PT->getElementType()); + Type *RetTy = FTy->getReturnType(); + + TargetLowering::ArgListTy Args; + TargetLowering::ArgListEntry Entry; + Args.reserve(CS.arg_size()); + + for (ImmutableCallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end(); + i != e; ++i) { + const Value *V = *i; + + // Skip empty types + if (V->getType()->isEmptyTy()) + continue; + + SDValue ArgNode = getValue(V); + Entry.Node = ArgNode; Entry.Ty = V->getType(); + + // Skip the first return-type Attribute to get to params. + Entry.setAttributes(&CS, i - CS.arg_begin() + 1); + Args.push_back(Entry); + } + + // Check if target-independent constraints permit a tail call here. + // Target-dependent constraints are checked within TLI->LowerCallTo. + if (isTailCall && !isInTailCallPosition(CS, DAG.getTarget())) + isTailCall = false; + + TargetLowering::CallLoweringInfo CLI(DAG); + CLI.setDebugLoc(getCurSDLoc()).setChain(getRoot()) + .setCallee(RetTy, FTy, Callee, std::move(Args), CS) + .setTailCall(isTailCall); + std::pair Result = lowerInvokable(CLI, LandingPad); + + if (Result.first.getNode()) + setValue(CS.getInstruction(), Result.first); } /// IsOnlyUsedInZeroEqualityComparison - Return true if it only matters that the @@ -5624,9 +5579,8 @@ static SDValue getMemCmpLoad(const Value *PtrVal, MVT LoadVT, LoadInput = ConstantExpr::getBitCast(const_cast(LoadInput), PointerType::getUnqual(LoadTy)); - if (const Constant *LoadCst = - ConstantFoldLoadFromConstPtr(const_cast(LoadInput), - Builder.DL)) + if (const Constant *LoadCst = ConstantFoldLoadFromConstPtr( + const_cast(LoadInput), *Builder.DL)) return Builder.getValue(LoadCst); } @@ -5661,8 +5615,7 @@ static SDValue getMemCmpLoad(const Value *PtrVal, MVT LoadVT, void SelectionDAGBuilder::processIntegerCallValue(const Instruction &I, SDValue Value, bool IsSigned) { - EVT VT = TM.getSubtargetImpl()->getTargetLowering()->getValueType(I.getType(), - true); + EVT VT = DAG.getTargetLoweringInfo().getValueType(I.getType(), true); if (IsSigned) Value = DAG.getSExtOrTrunc(Value, getCurSDLoc(), VT); else @@ -5687,8 +5640,7 @@ bool SelectionDAGBuilder::visitMemCmpCall(const CallInst &I) { const Value *Size = I.getArgOperand(2); const ConstantInt *CSize = dyn_cast(Size); if (CSize && CSize->getZExtValue() == 0) { - EVT CallVT = TM.getSubtargetImpl()->getTargetLowering()->getValueType( - I.getType(), true); + EVT CallVT = DAG.getTargetLoweringInfo().getValueType(I.getType(), true); setValue(&I, DAG.getConstant(0, CallVT)); return true; } @@ -5745,16 +5697,16 @@ bool SelectionDAGBuilder::visitMemCmpCall(const CallInst &I) { // Require that we can find a legal MVT, and only do this if the target // supports unaligned loads of that type. Expanding into byte loads would // bloat the code. - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); if (ActuallyDoIt && CSize->getZExtValue() > 4) { unsigned DstAS = LHS->getType()->getPointerAddressSpace(); unsigned SrcAS = RHS->getType()->getPointerAddressSpace(); // TODO: Handle 5 byte compare as 4-byte + 1 byte. // TODO: Handle 8 byte compare on x86-32 as two 32-bit loads. // TODO: Check alignment of src and dest ptrs. - if (!TLI->isTypeLegal(LoadVT) || - !TLI->allowsMisalignedMemoryAccesses(LoadVT, SrcAS) || - !TLI->allowsMisalignedMemoryAccesses(LoadVT, DstAS)) + if (!TLI.isTypeLegal(LoadVT) || + !TLI.allowsMisalignedMemoryAccesses(LoadVT, SrcAS) || + !TLI.allowsMisalignedMemoryAccesses(LoadVT, DstAS)) ActuallyDoIt = false; } @@ -5932,6 +5884,26 @@ bool SelectionDAGBuilder::visitUnaryFloatCall(const CallInst &I, return true; } +/// visitBinaryFloatCall - If a call instruction is a binary floating-point +/// operation (as expected), translate it to an SDNode with the specified opcode +/// and return true. +bool SelectionDAGBuilder::visitBinaryFloatCall(const CallInst &I, + unsigned Opcode) { + // Sanity check that it really is a binary floating-point call. + if (I.getNumArgOperands() != 2 || + !I.getArgOperand(0)->getType()->isFloatingPointTy() || + I.getType() != I.getArgOperand(0)->getType() || + I.getType() != I.getArgOperand(1)->getType() || + !I.onlyReadsMemory()) + return false; + + SDValue Tmp0 = getValue(I.getArgOperand(0)); + SDValue Tmp1 = getValue(I.getArgOperand(1)); + EVT VT = Tmp0.getValueType(); + setValue(&I, DAG.getNode(Opcode, getCurSDLoc(), VT, Tmp0, Tmp1)); + return true; +} + void SelectionDAGBuilder::visitCall(const CallInst &I) { // Handle inline assembly differently. if (isa(I.getCalledValue())) { @@ -5988,6 +5960,18 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) { if (visitUnaryFloatCall(I, ISD::FABS)) return; break; + case LibFunc::fmin: + case LibFunc::fminf: + case LibFunc::fminl: + if (visitBinaryFloatCall(I, ISD::FMINNUM)) + return; + break; + case LibFunc::fmax: + case LibFunc::fmaxf: + case LibFunc::fmaxl: + if (visitBinaryFloatCall(I, ISD::FMAXNUM)) + return; + break; case LibFunc::sin: case LibFunc::sinf: case LibFunc::sinl: @@ -6093,8 +6077,8 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) { if (!RenameFn) Callee = getValue(I.getCalledValue()); else - Callee = DAG.getExternalSymbol( - RenameFn, TM.getSubtargetImpl()->getTargetLowering()->getPointerTy()); + Callee = DAG.getExternalSymbol(RenameFn, + DAG.getTargetLoweringInfo().getPointerTy()); // Check if we can potentially perform a tail call. More detailed checking is // be done within LowerCallTo, after more information about the call is known. @@ -6192,9 +6176,10 @@ static void GetRegistersForValue(SelectionDAG &DAG, // If this is a constraint for a single physreg, or a constraint for a // register class, find it. - std::pair PhysReg = - TLI.getRegForInlineAsmConstraint(OpInfo.ConstraintCode, - OpInfo.ConstraintVT); + std::pair PhysReg = + TLI.getRegForInlineAsmConstraint(MF.getSubtarget().getRegisterInfo(), + OpInfo.ConstraintCode, + OpInfo.ConstraintVT); unsigned NumRegs = 1; if (OpInfo.ConstraintVT != MVT::Other) { @@ -6289,9 +6274,9 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { /// ConstraintOperands - Information about all of the constraints. SDISelAsmOperandInfoVector ConstraintOperands; - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - TargetLowering::AsmOperandInfoVector - TargetConstraints = TLI->ParseConstraints(CS); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + TargetLowering::AsmOperandInfoVector TargetConstraints = + TLI.ParseConstraints(DAG.getSubtarget().getRegisterInfo(), CS); bool hasMemory = false; @@ -6316,10 +6301,10 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { // corresponding argument. assert(!CS.getType()->isVoidTy() && "Bad inline asm!"); if (StructType *STy = dyn_cast(CS.getType())) { - OpVT = TLI->getSimpleValueType(STy->getElementType(ResNo)); + OpVT = TLI.getSimpleValueType(STy->getElementType(ResNo)); } else { assert(ResNo == 0 && "Asm only has one result!"); - OpVT = TLI->getSimpleValueType(CS.getType()); + OpVT = TLI.getSimpleValueType(CS.getType()); } ++ResNo; break; @@ -6340,8 +6325,8 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { OpInfo.CallOperand = getValue(OpInfo.CallOperandVal); } - OpVT = OpInfo.getCallOperandValEVT(*DAG.getContext(), *TLI, DL). - getSimpleVT(); + OpVT = + OpInfo.getCallOperandValEVT(*DAG.getContext(), TLI, DL).getSimpleVT(); } OpInfo.ConstraintVT = OpVT; @@ -6352,7 +6337,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { else { for (unsigned j = 0, ee = OpInfo.Codes.size(); j != ee; ++j) { TargetLowering::ConstraintType - CType = TLI->getConstraintType(OpInfo.Codes[j]); + CType = TLI.getConstraintType(OpInfo.Codes[j]); if (CType == TargetLowering::C_Memory) { hasMemory = true; break; @@ -6383,12 +6368,13 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput]; if (OpInfo.ConstraintVT != Input.ConstraintVT) { - std::pair MatchRC = - TLI->getRegForInlineAsmConstraint(OpInfo.ConstraintCode, - OpInfo.ConstraintVT); - std::pair InputRC = - TLI->getRegForInlineAsmConstraint(Input.ConstraintCode, - Input.ConstraintVT); + const TargetRegisterInfo *TRI = DAG.getSubtarget().getRegisterInfo(); + std::pair MatchRC = + TLI.getRegForInlineAsmConstraint(TRI, OpInfo.ConstraintCode, + OpInfo.ConstraintVT); + std::pair InputRC = + TLI.getRegForInlineAsmConstraint(TRI, Input.ConstraintCode, + Input.ConstraintVT); if ((OpInfo.ConstraintVT.isInteger() != Input.ConstraintVT.isInteger()) || (MatchRC.second != InputRC.second)) { @@ -6401,7 +6387,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { } // Compute the constraint code and ConstraintType to use. - TLI->ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, &DAG); + TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, &DAG); if (OpInfo.ConstraintType == TargetLowering::C_Memory && OpInfo.Type == InlineAsm::isClobber) @@ -6429,16 +6415,16 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { if (isa(OpVal) || isa(OpVal) || isa(OpVal) || isa(OpVal)) { OpInfo.CallOperand = DAG.getConstantPool(cast(OpVal), - TLI->getPointerTy()); + TLI.getPointerTy()); } else { // Otherwise, create a stack slot and emit a store to it before the // asm. Type *Ty = OpVal->getType(); - uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(Ty); - unsigned Align = TLI->getDataLayout()->getPrefTypeAlignment(Ty); + uint64_t TySize = TLI.getDataLayout()->getTypeAllocSize(Ty); + unsigned Align = TLI.getDataLayout()->getPrefTypeAlignment(Ty); MachineFunction &MF = DAG.getMachineFunction(); int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align, false); - SDValue StackSlot = DAG.getFrameIndex(SSFI, TLI->getPointerTy()); + SDValue StackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy()); Chain = DAG.getStore(Chain, getCurSDLoc(), OpInfo.CallOperand, StackSlot, MachinePointerInfo::getFixedStack(SSFI), @@ -6456,7 +6442,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { // If this constraint is for a specific register, allocate it before // anything else. if (OpInfo.ConstraintType == TargetLowering::C_Register) - GetRegistersForValue(DAG, *TLI, getCurSDLoc(), OpInfo); + GetRegistersForValue(DAG, TLI, getCurSDLoc(), OpInfo); } // Second pass - Loop over all of the operands, assigning virtual or physregs @@ -6467,7 +6453,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { // C_Register operands have already been allocated, Other/Memory don't need // to be. if (OpInfo.ConstraintType == TargetLowering::C_RegisterClass) - GetRegistersForValue(DAG, *TLI, getCurSDLoc(), OpInfo); + GetRegistersForValue(DAG, TLI, getCurSDLoc(), OpInfo); } // AsmNodeOperands - The operands for the ISD::INLINEASM node. @@ -6475,7 +6461,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { AsmNodeOperands.push_back(SDValue()); // reserve space for input chain AsmNodeOperands.push_back( DAG.getTargetExternalSymbol(IA->getAsmString().c_str(), - TLI->getPointerTy())); + TLI.getPointerTy())); // If we have a !srcloc metadata node associated with it, we want to attach // this to the ultimately generated inline asm machineinstr. To do this, we @@ -6498,7 +6484,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { TargetLowering::AsmOperandInfo &OpInfo = TargetConstraints[i]; // Compute the constraint code and ConstraintType to use. - TLI->ComputeConstraintToUse(OpInfo, SDValue()); + TLI.ComputeConstraintToUse(OpInfo, SDValue()); // Ideally, we would only check against memory constraints. However, the // meaning of an other constraint can be target-specific and we can't easily @@ -6516,7 +6502,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { } AsmNodeOperands.push_back(DAG.getTargetConstant(ExtraInfo, - TLI->getPointerTy())); + TLI.getPointerTy())); // Loop over all of the inputs, copying the operand values into the // appropriate registers and processing the output regs. @@ -6535,10 +6521,15 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { // Memory output, or 'other' output (e.g. 'X' constraint). assert(OpInfo.isIndirect && "Memory output must be indirect operand"); + unsigned ConstraintID = + TLI.getInlineAsmMemConstraint(OpInfo.ConstraintCode); + assert(ConstraintID != InlineAsm::Constraint_Unknown && + "Failed to convert memory constraint code to constraint id."); + // Add information to the INLINEASM node to know about this output. unsigned OpFlags = InlineAsm::getFlagWord(InlineAsm::Kind_Mem, 1); - AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlags, - TLI->getPointerTy())); + OpFlags = InlineAsm::getFlagWordForMem(OpFlags, ConstraintID); + AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlags, MVT::i32)); AsmNodeOperands.push_back(OpInfo.CallOperand); break; } @@ -6618,7 +6609,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { MachineRegisterInfo &RegInfo = DAG.getMachineFunction().getRegInfo(); for (unsigned i = 0, e = InlineAsm::getNumOperandRegisters(OpFlag); i != e; ++i) { - if (const TargetRegisterClass *RC = TLI->getRegClassFor(RegVT)) + if (const TargetRegisterClass *RC = TLI.getRegClassFor(RegVT)) MatchedRegs.Regs.push_back(RegInfo.createVirtualRegister(RC)); else { LLVMContext &Ctx = *DAG.getContext(); @@ -6642,10 +6633,11 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { "Unexpected number of operands"); // Add information to the INLINEASM node to know about this input. // See InlineAsm.h isUseOperandTiedToDef. + OpFlag = InlineAsm::convertMemFlagWordToMatchingFlagWord(OpFlag); OpFlag = InlineAsm::getFlagWordForMatchingOp(OpFlag, OpInfo.getMatchedOperand()); AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlag, - TLI->getPointerTy())); + TLI.getPointerTy())); AsmNodeOperands.push_back(AsmNodeOperands[CurOp+1]); break; } @@ -6657,7 +6649,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { if (OpInfo.ConstraintType == TargetLowering::C_Other) { std::vector Ops; - TLI->LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode, + TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode, Ops, DAG); if (Ops.empty()) { LLVMContext &Ctx = *DAG.getContext(); @@ -6671,20 +6663,25 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { unsigned ResOpType = InlineAsm::getFlagWord(InlineAsm::Kind_Imm, Ops.size()); AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType, - TLI->getPointerTy())); + TLI.getPointerTy())); AsmNodeOperands.insert(AsmNodeOperands.end(), Ops.begin(), Ops.end()); break; } if (OpInfo.ConstraintType == TargetLowering::C_Memory) { assert(OpInfo.isIndirect && "Operand must be indirect to be a mem!"); - assert(InOperandVal.getValueType() == TLI->getPointerTy() && + assert(InOperandVal.getValueType() == TLI.getPointerTy() && "Memory operands expect pointer values"); + unsigned ConstraintID = + TLI.getInlineAsmMemConstraint(OpInfo.ConstraintCode); + assert(ConstraintID != InlineAsm::Constraint_Unknown && + "Failed to convert memory constraint code to constraint id."); + // Add information to the INLINEASM node to know about this input. unsigned ResOpType = InlineAsm::getFlagWord(InlineAsm::Kind_Mem, 1); - AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType, - TLI->getPointerTy())); + ResOpType = InlineAsm::getFlagWordForMem(ResOpType, ConstraintID); + AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType, MVT::i32)); AsmNodeOperands.push_back(InOperandVal); break; } @@ -6747,7 +6744,7 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { // FIXME: Why don't we do this for inline asms with MRVs? if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) { - EVT ResultType = TLI->getValueType(CS.getType()); + EVT ResultType = TLI.getValueType(CS.getType()); // If any of the results of the inline asm is a vector, it may have the // wrong width/num elts. This can happen for register classes that can @@ -6812,9 +6809,9 @@ void SelectionDAGBuilder::visitVAStart(const CallInst &I) { } void SelectionDAGBuilder::visitVAArg(const VAArgInst &I) { - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - const DataLayout &DL = *TLI->getDataLayout(); - SDValue V = DAG.getVAArg(TLI->getValueType(I.getType()), getCurSDLoc(), + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + const DataLayout &DL = *TLI.getDataLayout(); + SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getCurSDLoc(), getRoot(), getValue(I.getOperand(0)), DAG.getSrcValue(I.getOperand(0)), DL.getABITypeAlignment(I.getType())); @@ -6846,18 +6843,19 @@ void SelectionDAGBuilder::visitVACopy(const CallInst &I) { /// convention or require stack pointer adjustment. Only a subset of the /// intrinsic's operands need to participate in the calling convention. std::pair -SelectionDAGBuilder::LowerCallOperands(const CallInst &CI, unsigned ArgIdx, +SelectionDAGBuilder::lowerCallOperands(ImmutableCallSite CS, unsigned ArgIdx, unsigned NumArgs, SDValue Callee, - bool useVoidTy) { + bool UseVoidTy, + MachineBasicBlock *LandingPad, + bool IsPatchPoint) { TargetLowering::ArgListTy Args; Args.reserve(NumArgs); // Populate the argument list. // Attributes for args start at offset 1, after the return attribute. - ImmutableCallSite CS(&CI); for (unsigned ArgI = ArgIdx, ArgE = ArgIdx + NumArgs, AttrI = ArgIdx + 1; ArgI != ArgE; ++ArgI) { - const Value *V = CI.getOperand(ArgI); + const Value *V = CS->getOperand(ArgI); assert(!V->getType()->isEmptyTy() && "Empty type passed to intrinsic."); @@ -6868,14 +6866,13 @@ SelectionDAGBuilder::LowerCallOperands(const CallInst &CI, unsigned ArgIdx, Args.push_back(Entry); } - Type *retTy = useVoidTy ? Type::getVoidTy(*DAG.getContext()) : CI.getType(); + Type *retTy = UseVoidTy ? Type::getVoidTy(*DAG.getContext()) : CS->getType(); TargetLowering::CallLoweringInfo CLI(DAG); CLI.setDebugLoc(getCurSDLoc()).setChain(getRoot()) - .setCallee(CI.getCallingConv(), retTy, Callee, std::move(Args), NumArgs) - .setDiscardResult(!CI.use_empty()); + .setCallee(CS.getCallingConv(), retTy, Callee, std::move(Args), NumArgs) + .setDiscardResult(CS->use_empty()).setIsPatchPoint(IsPatchPoint); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - return TLI->LowerCallTo(CLI); + return lowerInvokable(CLI, LandingPad); } /// \brief Add a stack map intrinsic call's live variable operands to a stackmap @@ -6895,11 +6892,11 @@ SelectionDAGBuilder::LowerCallOperands(const CallInst &CI, unsigned ArgIdx, /// assumption made by the llvm.gcroot intrinsic). If the alloca's location were /// only available in a register, then the runtime would need to trap when /// execution reaches the StackMap in order to read the alloca's location. -static void addStackMapLiveVars(const CallInst &CI, unsigned StartIdx, +static void addStackMapLiveVars(ImmutableCallSite CS, unsigned StartIdx, SmallVectorImpl &Ops, SelectionDAGBuilder &Builder) { - for (unsigned i = StartIdx, e = CI.getNumArgOperands(); i != e; ++i) { - SDValue OpVal = Builder.getValue(CI.getArgOperand(i)); + for (unsigned i = StartIdx, e = CS.arg_size(); i != e; ++i) { + SDValue OpVal = Builder.getValue(CS.getArgument(i)); if (ConstantSDNode *C = dyn_cast(OpVal)) { Ops.push_back( Builder.DAG.getTargetConstant(StackMaps::ConstantOp, MVT::i64)); @@ -6950,7 +6947,7 @@ void SelectionDAGBuilder::visitStackmap(const CallInst &CI) { cast(NBytesVal)->getZExtValue(), MVT::i32)); // Push live variables for the stack map. - addStackMapLiveVars(CI, 2, Ops, *this); + addStackMapLiveVars(&CI, 2, Ops, *this); // We are not pushing any register mask info here on the operands list, // because the stackmap doesn't clobber anything. @@ -6977,7 +6974,8 @@ void SelectionDAGBuilder::visitStackmap(const CallInst &CI) { } /// \brief Lower llvm.experimental.patchpoint directly to its target opcode. -void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { +void SelectionDAGBuilder::visitPatchpoint(ImmutableCallSite CS, + MachineBasicBlock *LandingPad) { // void|i64 @llvm.experimental.patchpoint.void|i64(i64 , // i32 , // i8* , @@ -6985,32 +6983,29 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { // [Args...], // [live variables...]) - CallingConv::ID CC = CI.getCallingConv(); - bool isAnyRegCC = CC == CallingConv::AnyReg; - bool hasDef = !CI.getType()->isVoidTy(); - SDValue Callee = getValue(CI.getOperand(2)); // + CallingConv::ID CC = CS.getCallingConv(); + bool IsAnyRegCC = CC == CallingConv::AnyReg; + bool HasDef = !CS->getType()->isVoidTy(); + SDValue Callee = getValue(CS->getOperand(2)); // // Get the real number of arguments participating in the call - SDValue NArgVal = getValue(CI.getArgOperand(PatchPointOpers::NArgPos)); + SDValue NArgVal = getValue(CS.getArgument(PatchPointOpers::NArgPos)); unsigned NumArgs = cast(NArgVal)->getZExtValue(); // Skip the four meta args: , , , // Intrinsics include all meta-operands up to but not including CC. unsigned NumMetaOpers = PatchPointOpers::CCPos; - assert(CI.getNumArgOperands() >= NumMetaOpers + NumArgs && + assert(CS.arg_size() >= NumMetaOpers + NumArgs && "Not enough arguments provided to the patchpoint intrinsic"); // For AnyRegCC the arguments are lowered later on manually. - unsigned NumCallArgs = isAnyRegCC ? 0 : NumArgs; + unsigned NumCallArgs = IsAnyRegCC ? 0 : NumArgs; std::pair Result = - LowerCallOperands(CI, NumMetaOpers, NumCallArgs, Callee, isAnyRegCC); - - // Set the root to the target-lowered call chain. - SDValue Chain = Result.second; - DAG.setRoot(Chain); + lowerCallOperands(CS, NumMetaOpers, NumCallArgs, Callee, IsAnyRegCC, + LandingPad, true); - SDNode *CallEnd = Chain.getNode(); - if (hasDef && (CallEnd->getOpcode() == ISD::CopyFromReg)) + SDNode *CallEnd = Result.second.getNode(); + if (HasDef && (CallEnd->getOpcode() == ISD::CopyFromReg)) CallEnd = CallEnd->getOperand(0).getNode(); /// Get a call instruction from the call sequence chain. @@ -7018,16 +7013,16 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { assert(CallEnd->getOpcode() == ISD::CALLSEQ_END && "Expected a callseq node."); SDNode *Call = CallEnd->getOperand(0).getNode(); - bool hasGlue = Call->getGluedNode(); + bool HasGlue = Call->getGluedNode(); // Replace the target specific call node with the patchable intrinsic. SmallVector Ops; // Add the and constants. - SDValue IDVal = getValue(CI.getOperand(PatchPointOpers::IDPos)); + SDValue IDVal = getValue(CS->getOperand(PatchPointOpers::IDPos)); Ops.push_back(DAG.getTargetConstant( cast(IDVal)->getZExtValue(), MVT::i64)); - SDValue NBytesVal = getValue(CI.getOperand(PatchPointOpers::NBytesPos)); + SDValue NBytesVal = getValue(CS->getOperand(PatchPointOpers::NBytesPos)); Ops.push_back(DAG.getTargetConstant( cast(NBytesVal)->getZExtValue(), MVT::i32)); @@ -7040,8 +7035,8 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { // Adjust to account for any arguments that have been passed on the // stack instead. // Call Node: Chain, Target, {Args}, RegMask, [Glue] - unsigned NumCallRegArgs = Call->getNumOperands() - (hasGlue ? 4 : 3); - NumCallRegArgs = isAnyRegCC ? NumArgs : NumCallRegArgs; + unsigned NumCallRegArgs = Call->getNumOperands() - (HasGlue ? 4 : 3); + NumCallRegArgs = IsAnyRegCC ? NumArgs : NumCallRegArgs; Ops.push_back(DAG.getTargetConstant(NumCallRegArgs, MVT::i32)); // Add the calling convention @@ -7049,20 +7044,19 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { // Add the arguments we omitted previously. The register allocator should // place these in any free register. - if (isAnyRegCC) + if (IsAnyRegCC) for (unsigned i = NumMetaOpers, e = NumMetaOpers + NumArgs; i != e; ++i) - Ops.push_back(getValue(CI.getArgOperand(i))); + Ops.push_back(getValue(CS.getArgument(i))); // Push the arguments from the call instruction up to the register mask. - SDNode::op_iterator e = hasGlue ? Call->op_end()-2 : Call->op_end()-1; - for (SDNode::op_iterator i = Call->op_begin()+2; i != e; ++i) - Ops.push_back(*i); + SDNode::op_iterator e = HasGlue ? Call->op_end()-2 : Call->op_end()-1; + Ops.append(Call->op_begin() + 2, e); // Push live variables for the stack map. - addStackMapLiveVars(CI, NumMetaOpers + NumArgs, Ops, *this); + addStackMapLiveVars(CS, NumMetaOpers + NumArgs, Ops, *this); // Push the register mask info. - if (hasGlue) + if (HasGlue) Ops.push_back(*(Call->op_end()-2)); else Ops.push_back(*(Call->op_end()-1)); @@ -7072,15 +7066,15 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { Ops.push_back(*(Call->op_begin())); // Push the glue flag (last operand). - if (hasGlue) + if (HasGlue) Ops.push_back(*(Call->op_end()-1)); SDVTList NodeTys; - if (isAnyRegCC && hasDef) { + if (IsAnyRegCC && HasDef) { // Create the return types based on the intrinsic definition const TargetLowering &TLI = DAG.getTargetLoweringInfo(); SmallVector ValueVTs; - ComputeValueVTs(TLI, CI.getType(), ValueVTs); + ComputeValueVTs(TLI, CS->getType(), ValueVTs); assert(ValueVTs.size() == 1 && "Expected only one return value type."); // There is always a chain and a glue type at the end @@ -7095,18 +7089,18 @@ void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) { getCurSDLoc(), NodeTys, Ops); // Update the NodeMap. - if (hasDef) { - if (isAnyRegCC) - setValue(&CI, SDValue(MN, 0)); + if (HasDef) { + if (IsAnyRegCC) + setValue(CS.getInstruction(), SDValue(MN, 0)); else - setValue(&CI, Result.first); + setValue(CS.getInstruction(), Result.first); } // Fixup the consumers of the intrinsic. The chain and glue may be used in the // call sequence. Furthermore the location of the chain and glue can change // when the AnyReg calling convention is used and the intrinsic returns a // value. - if (isAnyRegCC && hasDef) { + if (IsAnyRegCC && HasDef) { SDValue From[] = {SDValue(Call, 0), SDValue(Call, 1)}; SDValue To[] = {SDValue(MN, 1), SDValue(MN, 2)}; DAG.ReplaceAllUsesOfValuesWith(From, To, 2); @@ -7302,13 +7296,12 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { else if (j != 0) MyFlags.Flags.setOrigAlign(1); - // Only mark the end at the last register of the last value. - if (NeedsRegBlock && Value == NumValues - 1 && j == NumParts - 1) - MyFlags.Flags.setInConsecutiveRegsLast(); - CLI.Outs.push_back(MyFlags); CLI.OutVals.push_back(Parts[j]); } + + if (NeedsRegBlock && Value == NumValues - 1) + CLI.Outs[CLI.Outs.size() - 1].Flags.setInConsecutiveRegsLast(); } } @@ -7418,10 +7411,15 @@ SelectionDAGBuilder::CopyValueToVirtualRegister(const Value *V, unsigned Reg) { "Copy from a reg to the same reg!"); assert(!TargetRegisterInfo::isPhysicalRegister(Reg) && "Is a physreg"); - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - RegsForValue RFV(V->getContext(), *TLI, Reg, V->getType()); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + RegsForValue RFV(V->getContext(), TLI, Reg, V->getType()); SDValue Chain = DAG.getEntryNode(); - RFV.getCopyToRegs(Op, DAG, getCurSDLoc(), Chain, nullptr, V); + + ISD::NodeType ExtendType = (FuncInfo.PreferredExtendType.find(V) == + FuncInfo.PreferredExtendType.end()) + ? ISD::ANY_EXTEND + : FuncInfo.PreferredExtendType[V]; + RFV.getCopyToRegs(Op, DAG, getCurSDLoc(), Chain, nullptr, V, ExtendType); PendingExports.push_back(Chain); } @@ -7447,22 +7445,21 @@ static bool isOnlyUsedInEntryBlock(const Argument *A, bool FastISel) { void SelectionDAGISel::LowerArguments(const Function &F) { SelectionDAG &DAG = SDB->DAG; SDLoc dl = SDB->getCurSDLoc(); - const TargetLowering *TLI = getTargetLowering(); const DataLayout *DL = TLI->getDataLayout(); SmallVector Ins; if (!FuncInfo->CanLowerReturn) { // Put in an sret pointer parameter before all the other parameters. SmallVector ValueVTs; - ComputeValueVTs(*getTargetLowering(), - PointerType::getUnqual(F.getReturnType()), ValueVTs); + ComputeValueVTs(*TLI, PointerType::getUnqual(F.getReturnType()), ValueVTs); // NOTE: Assuming that a pointer will never break down to more than one VT // or one register. ISD::ArgFlagsTy Flags; Flags.setSRet(); MVT RegisterVT = TLI->getRegisterType(*DAG.getContext(), ValueVTs[0]); - ISD::InputArg RetArg(Flags, RegisterVT, ValueVTs[0], true, 0, 0); + ISD::InputArg RetArg(Flags, RegisterVT, ValueVTs[0], true, + ISD::InputArg::NoArgIndex, 0); Ins.push_back(RetArg); } @@ -7534,22 +7531,18 @@ void SelectionDAGISel::LowerArguments(const Function &F) { // if it isn't first piece, alignment must be 1 else if (i > 0) MyFlags.Flags.setOrigAlign(1); - - // Only mark the end at the last register of the last value. - if (NeedsRegBlock && Value == NumValues - 1 && i == NumRegs - 1) - MyFlags.Flags.setInConsecutiveRegsLast(); - Ins.push_back(MyFlags); } + if (NeedsRegBlock && Value == NumValues - 1) + Ins[Ins.size() - 1].Flags.setInConsecutiveRegsLast(); PartBase += VT.getStoreSize(); } } // Call the target to set up the argument values. SmallVector InVals; - SDValue NewRoot = TLI->LowerFormalArguments(DAG.getRoot(), F.getCallingConv(), - F.isVarArg(), Ins, - dl, DAG, InVals); + SDValue NewRoot = TLI->LowerFormalArguments( + DAG.getRoot(), F.getCallingConv(), F.isVarArg(), Ins, dl, DAG, InVals); // Verify that the target's LowerFormalArguments behaved as expected. assert(NewRoot.getNode() && NewRoot.getValueType() == MVT::Other && @@ -7586,8 +7579,8 @@ void SelectionDAGISel::LowerArguments(const Function &F) { MachineRegisterInfo& RegInfo = MF.getRegInfo(); unsigned SRetReg = RegInfo.createVirtualRegister(TLI->getRegClassFor(RegVT)); FuncInfo->DemoteRegister = SRetReg; - NewRoot = SDB->DAG.getCopyToReg(NewRoot, SDB->getCurSDLoc(), - SRetReg, ArgValue); + NewRoot = + SDB->DAG.getCopyToReg(NewRoot, SDB->getCurSDLoc(), SRetReg, ArgValue); DAG.setRoot(NewRoot); // i indexes lowered arguments. Bump it past the hidden sret argument. @@ -7676,7 +7669,6 @@ void SelectionDAGISel::LowerArguments(const Function &F) { assert(i == InVals.size() && "Argument register count mismatch!"); // Finally, if the target has anything special to do, allow it to do so. - // FIXME: this should insert code into the DAG! EmitFunctionEntryCode(); } @@ -7693,8 +7685,8 @@ SelectionDAGBuilder::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) { SmallPtrSet SuccsHandled; - // Check successor nodes' PHI nodes that expect a constant to be available - // from this block. + // Check PHI nodes in successors that expect a value to be available from this + // block. for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) { const BasicBlock *SuccBB = TI->getSuccessor(succ); if (!isa(SuccBB->begin())) continue; @@ -7702,7 +7694,8 @@ SelectionDAGBuilder::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) { // If this terminator has multiple identical successors (common for // switches), only handle each succ once. - if (!SuccsHandled.insert(SuccMBB)) continue; + if (!SuccsHandled.insert(SuccMBB).second) + continue; MachineBasicBlock::iterator MBBI = SuccMBB->begin(); @@ -7745,11 +7738,11 @@ SelectionDAGBuilder::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) { // Remember that this register needs to added to the machine PHI node as // the input for this MBB. SmallVector ValueVTs; - const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering(); - ComputeValueVTs(*TLI, PN->getType(), ValueVTs); + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + ComputeValueVTs(TLI, PN->getType(), ValueVTs); for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) { EVT VT = ValueVTs[vti]; - unsigned NumRegisters = TLI->getNumRegisters(*DAG.getContext(), VT); + unsigned NumRegisters = TLI.getNumRegisters(*DAG.getContext(), VT); for (unsigned i = 0, e = NumRegisters; i != e; ++i) FuncInfo.PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i)); Reg += NumRegisters; @@ -7766,6 +7759,7 @@ MachineBasicBlock * SelectionDAGBuilder::StackProtectorDescriptor:: AddSuccessorMBB(const BasicBlock *BB, MachineBasicBlock *ParentMBB, + bool IsLikely, MachineBasicBlock *SuccMBB) { // If SuccBB has not been created yet, create it. if (!SuccMBB) { @@ -7775,6 +7769,14 @@ AddSuccessorMBB(const BasicBlock *BB, MF->insert(++BBI, SuccMBB); } // Add it as a successor of ParentMBB. - ParentMBB->addSuccessor(SuccMBB); + ParentMBB->addSuccessor( + SuccMBB, BranchProbabilityInfo::getBranchWeightStackProtector(IsLikely)); return SuccMBB; } + +MachineBasicBlock *SelectionDAGBuilder::NextBlock(MachineBasicBlock *MBB) { + MachineFunction::iterator I = MBB; + if (++I == FuncInfo.MF->end()) + return nullptr; + return I; +}