diff options
Diffstat (limited to 'llvm/lib')
-rw-r--r-- | llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp | 233 | ||||
-rw-r--r-- | llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h | 14 | ||||
-rw-r--r-- | llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp | 2 | ||||
-rw-r--r-- | llvm/lib/CodeGen/TargetLoweringBase.cpp | 6 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsCCState.cpp | 61 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsCCState.h | 33 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsCallingConv.td | 10 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsISelLowering.cpp | 74 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsISelLowering.h | 27 | ||||
-rw-r--r-- | llvm/lib/Target/Mips/MipsRegisterInfo.cpp | 4 |
10 files changed, 383 insertions, 81 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp index 9b2710c82c3..d34ac40b949 100644 --- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp @@ -101,7 +101,8 @@ static const unsigned MaxParallelChains = 64; static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, const SDValue *Parts, unsigned NumParts, - MVT PartVT, EVT ValueVT, const Value *V); + MVT PartVT, EVT ValueVT, const Value *V, + bool IsABIRegCopy); /// getCopyFromParts - Create a value that contains the specified legal parts /// combined into the value they represent. If the parts combine to a type @@ -111,10 +112,11 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, static SDValue getCopyFromParts(SelectionDAG &DAG, const SDLoc &DL, const SDValue *Parts, unsigned NumParts, MVT PartVT, EVT ValueVT, const Value *V, - Optional<ISD::NodeType> AssertOp = None) { + Optional<ISD::NodeType> AssertOp = None, + bool IsABIRegCopy = false) { if (ValueVT.isVector()) return getCopyFromPartsVector(DAG, DL, Parts, NumParts, - PartVT, ValueVT, V); + PartVT, ValueVT, V, IsABIRegCopy); assert(NumParts > 0 && "No parts to assemble!"); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); @@ -258,7 +260,8 @@ static void diagnosePossiblyInvalidConstraint(LLVMContext &Ctx, const Value *V, /// ValueVT (ISD::AssertSext). static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, const SDValue *Parts, unsigned NumParts, - MVT PartVT, EVT ValueVT, const Value *V) { + MVT PartVT, EVT ValueVT, const Value *V, + bool IsABIRegCopy) { assert(ValueVT.isVector() && "Not a vector value"); assert(NumParts > 0 && "No parts to assemble!"); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); @@ -269,9 +272,18 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, EVT IntermediateVT; MVT RegisterVT; unsigned NumIntermediates; - unsigned NumRegs = - TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT, - NumIntermediates, RegisterVT); + unsigned NumRegs; + + if (IsABIRegCopy) { + NumRegs = TLI.getVectorTypeBreakdownForCallingConv( + *DAG.getContext(), ValueVT, IntermediateVT, NumIntermediates, + RegisterVT); + } else { + NumRegs = + TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT, + NumIntermediates, RegisterVT); + } + assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!"); NumParts = NumRegs; // Silence a compiler warning. assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!"); @@ -300,9 +312,14 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the // intermediate operands. + EVT BuiltVectorTy = + EVT::getVectorVT(*DAG.getContext(), IntermediateVT.getScalarType(), + (IntermediateVT.isVector() + ? IntermediateVT.getVectorNumElements() * NumParts + : NumIntermediates)); Val = DAG.getNode(IntermediateVT.isVector() ? ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, - DL, ValueVT, Ops); + DL, BuiltVectorTy, Ops); } // There is now one part, held in Val. Correct it to match ValueVT. @@ -341,13 +358,29 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, TLI.isTypeLegal(ValueVT)) return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val); - // Handle cases such as i8 -> <1 x i1> if (ValueVT.getVectorNumElements() != 1) { - diagnosePossiblyInvalidConstraint(*DAG.getContext(), V, - "non-trivial scalar-to-vector conversion"); - return DAG.getUNDEF(ValueVT); + // Certain ABIs require that vectors are passed as integers. For vectors + // are the same size, this is an obvious bitcast. + if (ValueVT.getSizeInBits() == PartEVT.getSizeInBits()) { + return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val); + } else if (ValueVT.getSizeInBits() < PartEVT.getSizeInBits()) { + // Bitcast Val back the original type and extract the corresponding + // vector we want. + unsigned Elts = PartEVT.getSizeInBits() / ValueVT.getScalarSizeInBits(); + EVT WiderVecType = EVT::getVectorVT(*DAG.getContext(), + ValueVT.getVectorElementType(), Elts); + Val = DAG.getBitcast(WiderVecType, Val); + return DAG.getNode( + ISD::EXTRACT_SUBVECTOR, DL, ValueVT, Val, + DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout()))); + } + + diagnosePossiblyInvalidConstraint( + *DAG.getContext(), V, "non-trivial scalar-to-vector conversion"); + return DAG.getUNDEF(ValueVT); } + // Handle cases such as i8 -> <1 x i1> EVT ValueSVT = ValueVT.getVectorElementType(); if (ValueVT.getVectorNumElements() == 1 && ValueSVT != PartEVT) Val = ValueVT.isFloatingPoint() ? DAG.getFPExtendOrRound(Val, DL, ValueSVT) @@ -358,7 +391,7 @@ static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL, static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &dl, SDValue Val, SDValue *Parts, unsigned NumParts, - MVT PartVT, const Value *V); + MVT PartVT, const Value *V, bool IsABIRegCopy); /// getCopyToParts - Create a series of nodes that contain the specified value /// split into legal parts. If the parts contain more bits than Val, then, for @@ -366,12 +399,14 @@ static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &dl, static void getCopyToParts(SelectionDAG &DAG, const SDLoc &DL, SDValue Val, SDValue *Parts, unsigned NumParts, MVT PartVT, const Value *V, - ISD::NodeType ExtendKind = ISD::ANY_EXTEND) { + ISD::NodeType ExtendKind = ISD::ANY_EXTEND, + bool IsABIRegCopy = false) { EVT ValueVT = Val.getValueType(); // Handle the vector case separately. if (ValueVT.isVector()) - return getCopyToPartsVector(DAG, DL, Val, Parts, NumParts, PartVT, V); + return getCopyToPartsVector(DAG, DL, Val, Parts, NumParts, PartVT, V, + IsABIRegCopy); unsigned PartBits = PartVT.getSizeInBits(); unsigned OrigNumParts = NumParts; @@ -496,7 +531,9 @@ static void getCopyToParts(SelectionDAG &DAG, const SDLoc &DL, SDValue Val, /// value split into legal parts. static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &DL, SDValue Val, SDValue *Parts, unsigned NumParts, - MVT PartVT, const Value *V) { + MVT PartVT, const Value *V, + bool IsABIRegCopy) { + EVT ValueVT = Val.getValueType(); assert(ValueVT.isVector() && "Not a vector"); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); @@ -537,13 +574,20 @@ static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &DL, // Promoted vector extract Val = DAG.getAnyExtOrTrunc(Val, DL, PartVT); - } else{ - // Vector -> scalar conversion. - assert(ValueVT.getVectorNumElements() == 1 && - "Only trivial vector-to-scalar conversions should get here!"); - Val = DAG.getNode( - ISD::EXTRACT_VECTOR_ELT, DL, PartVT, Val, - DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout()))); + } else { + if (ValueVT.getVectorNumElements() == 1) { + Val = DAG.getNode( + ISD::EXTRACT_VECTOR_ELT, DL, PartVT, Val, + DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout()))); + + } else { + assert(PartVT.getSizeInBits() > ValueVT.getSizeInBits() && + "lossy conversion of vector to scalar type"); + EVT IntermediateType = + EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits()); + Val = DAG.getBitcast(IntermediateType, Val); + Val = DAG.getAnyExtOrTrunc(Val, DL, PartVT); + } } assert(Val.getValueType() == PartVT && "Unexpected vector part value type"); @@ -555,15 +599,31 @@ static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &DL, EVT IntermediateVT; MVT RegisterVT; unsigned NumIntermediates; - unsigned NumRegs = TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, - IntermediateVT, - NumIntermediates, RegisterVT); + unsigned NumRegs; + if (IsABIRegCopy) { + NumRegs = TLI.getVectorTypeBreakdownForCallingConv( + *DAG.getContext(), ValueVT, IntermediateVT, NumIntermediates, + RegisterVT); + } else { + NumRegs = + TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT, + NumIntermediates, RegisterVT); + } unsigned NumElements = ValueVT.getVectorNumElements(); assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!"); NumParts = NumRegs; // Silence a compiler warning. assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!"); + // Convert the vector to the appropiate type if necessary. + unsigned DestVectorNoElts = + NumIntermediates * + (IntermediateVT.isVector() ? IntermediateVT.getVectorNumElements() : 1); + EVT BuiltVectorTy = EVT::getVectorVT( + *DAG.getContext(), IntermediateVT.getScalarType(), DestVectorNoElts); + if (Val.getValueType() != BuiltVectorTy) + Val = DAG.getNode(ISD::BITCAST, DL, BuiltVectorTy, Val); + // Split the vector into intermediate operands. SmallVector<SDValue, 8> Ops(NumIntermediates); for (unsigned i = 0; i != NumIntermediates; ++i) { @@ -596,22 +656,31 @@ static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &DL, } } -RegsForValue::RegsForValue() {} +RegsForValue::RegsForValue() { IsABIMangled = false; } RegsForValue::RegsForValue(const SmallVector<unsigned, 4> ®s, MVT regvt, - EVT valuevt) - : ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {} + EVT valuevt, bool IsABIMangledValue) + : ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs), + RegCount(1, regs.size()), IsABIMangled(IsABIMangledValue) {} RegsForValue::RegsForValue(LLVMContext &Context, const TargetLowering &TLI, - const DataLayout &DL, unsigned Reg, Type *Ty) { + const DataLayout &DL, unsigned Reg, Type *Ty, + bool IsABIMangledValue) { ComputeValueVTs(TLI, DL, Ty, ValueVTs); + IsABIMangled = IsABIMangledValue; + for (EVT ValueVT : ValueVTs) { - unsigned NumRegs = TLI.getNumRegisters(Context, ValueVT); - MVT RegisterVT = TLI.getRegisterType(Context, ValueVT); + unsigned NumRegs = IsABIMangledValue + ? TLI.getNumRegistersForCallingConv(Context, ValueVT) + : TLI.getNumRegisters(Context, ValueVT); + MVT RegisterVT = IsABIMangledValue + ? TLI.getRegisterTypeForCallingConv(Context, ValueVT) + : TLI.getRegisterType(Context, ValueVT); for (unsigned i = 0; i != NumRegs; ++i) Regs.push_back(Reg + i); RegVTs.push_back(RegisterVT); + RegCount.push_back(NumRegs); Reg += NumRegs; } } @@ -632,8 +701,10 @@ SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) { // Copy the legal parts from the registers. EVT ValueVT = ValueVTs[Value]; - unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVT); - MVT RegisterVT = RegVTs[Value]; + unsigned NumRegs = RegCount[Value]; + MVT RegisterVT = IsABIMangled + ? TLI.getRegisterTypeForCallingConv(RegVTs[Value]) + : RegVTs[Value]; Parts.resize(NumRegs); for (unsigned i = 0; i != NumRegs; ++i) { @@ -728,9 +799,11 @@ void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, unsigned NumRegs = Regs.size(); SmallVector<SDValue, 8> Parts(NumRegs); for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) { - EVT ValueVT = ValueVTs[Value]; - unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), ValueVT); - MVT RegisterVT = RegVTs[Value]; + unsigned NumParts = RegCount[Value]; + + MVT RegisterVT = IsABIMangled + ? TLI.getRegisterTypeForCallingConv(RegVTs[Value]) + : RegVTs[Value]; if (ExtendKind == ISD::ANY_EXTEND && TLI.isZExtFree(Val, RegisterVT)) ExtendKind = ISD::ZERO_EXTEND; @@ -953,10 +1026,16 @@ SDValue SelectionDAGBuilder::getCopyFromRegs(const Value *V, Type *Ty) { if (It != FuncInfo.ValueMap.end()) { unsigned InReg = It->second; + bool IsABIRegCopy = + V && ((isa<CallInst>(V) && + !(static_cast<const CallInst *>(V))->isInlineAsm()) || + isa<ReturnInst>(V)); + RegsForValue RFV(*DAG.getContext(), DAG.getTargetLoweringInfo(), - DAG.getDataLayout(), InReg, Ty); + DAG.getDataLayout(), InReg, Ty, IsABIRegCopy); SDValue Chain = DAG.getEntryNode(); - Result = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V); + Result = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, + V); resolveDanglingDebugInfo(V, Result); } @@ -1142,8 +1221,13 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) { // If this is an instruction which fast-isel has deferred, select it now. if (const Instruction *Inst = dyn_cast<Instruction>(V)) { unsigned InReg = FuncInfo.InitializeRegForValue(Inst); + bool IsABIRegCopy = + V && ((isa<CallInst>(V) && + !(static_cast<const CallInst *>(V))->isInlineAsm()) || + isa<ReturnInst>(V)); + RegsForValue RFV(*DAG.getContext(), TLI, DAG.getDataLayout(), InReg, - Inst->getType()); + Inst->getType(), IsABIRegCopy); SDValue Chain = DAG.getEntryNode(); return RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V); } @@ -1371,12 +1455,12 @@ void SelectionDAGBuilder::visitRet(const ReturnInst &I) { if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) VT = TLI.getTypeForExtReturn(Context, VT, ExtendKind); - unsigned NumParts = TLI.getNumRegisters(Context, VT); - MVT PartVT = TLI.getRegisterType(Context, VT); + unsigned NumParts = TLI.getNumRegistersForCallingConv(Context, VT); + MVT PartVT = TLI.getRegisterTypeForCallingConv(Context, VT); SmallVector<SDValue, 4> Parts(NumParts); getCopyToParts(DAG, getCurSDLoc(), SDValue(RetOp.getNode(), RetOp.getResNo() + j), - &Parts[0], NumParts, PartVT, &I, ExtendKind); + &Parts[0], NumParts, PartVT, &I, ExtendKind, true); // 'inreg' on function refers to return value ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy(); @@ -7112,8 +7196,8 @@ void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) { SDLoc dl = getCurSDLoc(); // Use the produced MatchedRegs object to - MatchedRegs.getCopyToRegs(InOperandVal, DAG, dl, - Chain, &Flag, CS.getInstruction()); + MatchedRegs.getCopyToRegs(InOperandVal, DAG, dl, Chain, &Flag, + CS.getInstruction()); MatchedRegs.AddInlineAsmOperands(InlineAsm::Kind_RegUse, true, OpInfo.getMatchedOperand(), dl, DAG, AsmNodeOperands); @@ -7799,8 +7883,10 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { } else { for (unsigned I = 0, E = RetTys.size(); I != E; ++I) { EVT VT = RetTys[I]; - MVT RegisterVT = getRegisterType(CLI.RetTy->getContext(), VT); - unsigned NumRegs = getNumRegisters(CLI.RetTy->getContext(), VT); + MVT RegisterVT = + getRegisterTypeForCallingConv(CLI.RetTy->getContext(), VT); + unsigned NumRegs = + getNumRegistersForCallingConv(CLI.RetTy->getContext(), VT); for (unsigned i = 0; i != NumRegs; ++i) { ISD::InputArg MyFlags; MyFlags.VT = RegisterVT; @@ -7849,7 +7935,11 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { SDValue Op = SDValue(Args[i].Node.getNode(), Args[i].Node.getResNo() + Value); ISD::ArgFlagsTy Flags; - unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy); + + // Certain targets (such as MIPS), may have a different ABI alignment + // for a type depending on the context. Give the target a chance to + // specify the alignment it wants. + unsigned OriginalAlignment = getABIAlignmentForCallingConv(ArgTy, DL); if (Args[i].IsZExt) Flags.setZExt(); @@ -7904,8 +7994,9 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { Flags.setInConsecutiveRegs(); Flags.setOrigAlign(OriginalAlignment); - MVT PartVT = getRegisterType(CLI.RetTy->getContext(), VT); - unsigned NumParts = getNumRegisters(CLI.RetTy->getContext(), VT); + MVT PartVT = getRegisterTypeForCallingConv(CLI.RetTy->getContext(), VT); + unsigned NumParts = + getNumRegistersForCallingConv(CLI.RetTy->getContext(), VT); SmallVector<SDValue, 4> Parts(NumParts); ISD::NodeType ExtendKind = ISD::ANY_EXTEND; @@ -7935,7 +8026,8 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { } getCopyToParts(CLI.DAG, CLI.DL, Op, &Parts[0], NumParts, PartVT, - CLI.CS ? CLI.CS->getInstruction() : nullptr, ExtendKind); + CLI.CS ? CLI.CS->getInstruction() : nullptr, ExtendKind, + true); for (unsigned j = 0; j != NumParts; ++j) { // if it isn't first piece, alignment must be 1 @@ -8035,12 +8127,14 @@ TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const { unsigned CurReg = 0; for (unsigned I = 0, E = RetTys.size(); I != E; ++I) { EVT VT = RetTys[I]; - MVT RegisterVT = getRegisterType(CLI.RetTy->getContext(), VT); - unsigned NumRegs = getNumRegisters(CLI.RetTy->getContext(), VT); + MVT RegisterVT = + getRegisterTypeForCallingConv(CLI.RetTy->getContext(), VT); + unsigned NumRegs = + getNumRegistersForCallingConv(CLI.RetTy->getContext(), VT); ReturnValues.push_back(getCopyFromParts(CLI.DAG, CLI.DL, &InVals[CurReg], NumRegs, RegisterVT, VT, nullptr, - AssertOp)); + AssertOp, true)); CurReg += NumRegs; } @@ -8076,8 +8170,15 @@ SelectionDAGBuilder::CopyValueToVirtualRegister(const Value *V, unsigned Reg) { assert(!TargetRegisterInfo::isPhysicalRegister(Reg) && "Is a physreg"); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + // If this is an InlineAsm we have to match the registers required, not the + // notional registers required by the type. + bool IsABIRegCopy = + V && ((isa<CallInst>(V) && + !(static_cast<const CallInst *>(V))->isInlineAsm()) || + isa<ReturnInst>(V)); + RegsForValue RFV(V->getContext(), TLI, DAG.getDataLayout(), Reg, - V->getType()); + V->getType(), IsABIRegCopy); SDValue Chain = DAG.getEntryNode(); ISD::NodeType ExtendType = (FuncInfo.PreferredExtendType.find(V) == @@ -8319,7 +8420,12 @@ void SelectionDAGISel::LowerArguments(const Function &F) { EVT VT = ValueVTs[Value]; Type *ArgTy = VT.getTypeForEVT(*DAG.getContext()); ISD::ArgFlagsTy Flags; - unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy); + + // Certain targets (such as MIPS), may have a different ABI alignment + // for a type depending on the context. Give the target a chance to + // specify the alignment it wants. + unsigned OriginalAlignment = + TLI->getABIAlignmentForCallingConv(ArgTy, DL); if (Arg.hasAttribute(Attribute::ZExt)) Flags.setZExt(); @@ -8381,8 +8487,10 @@ void SelectionDAGISel::LowerArguments(const Function &F) { if (ArgCopyElisionCandidates.count(&Arg)) Flags.setCopyElisionCandidate(); - MVT RegisterVT = TLI->getRegisterType(*CurDAG->getContext(), VT); - unsigned NumRegs = TLI->getNumRegisters(*CurDAG->getContext(), VT); + MVT RegisterVT = + TLI->getRegisterTypeForCallingConv(*CurDAG->getContext(), VT); + unsigned NumRegs = + TLI->getNumRegistersForCallingConv(*CurDAG->getContext(), VT); for (unsigned i = 0; i != NumRegs; ++i) { ISD::InputArg MyFlags(Flags, RegisterVT, VT, isArgValueUsed, ArgNo, PartBase+i*RegisterVT.getStoreSize()); @@ -8486,8 +8594,10 @@ void SelectionDAGISel::LowerArguments(const Function &F) { for (unsigned Val = 0; Val != NumValues; ++Val) { EVT VT = ValueVTs[Val]; - MVT PartVT = TLI->getRegisterType(*CurDAG->getContext(), VT); - unsigned NumParts = TLI->getNumRegisters(*CurDAG->getContext(), VT); + MVT PartVT = + TLI->getRegisterTypeForCallingConv(*CurDAG->getContext(), VT); + unsigned NumParts = + TLI->getNumRegistersForCallingConv(*CurDAG->getContext(), VT); // Even an apparant 'unused' swifterror argument needs to be returned. So // we do generate a copy for it that can be used on return from the @@ -8500,7 +8610,8 @@ void SelectionDAGISel::LowerArguments(const Function &F) { AssertOp = ISD::AssertZext; ArgValues.push_back(getCopyFromParts(DAG, dl, &InVals[i], NumParts, - PartVT, VT, nullptr, AssertOp)); + PartVT, VT, nullptr, AssertOp, + true)); } i += NumParts; diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h index 77e131fa551..431d52b4b9b 100644 --- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h +++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h @@ -975,18 +975,28 @@ struct RegsForValue { /// expanded value requires multiple registers. SmallVector<unsigned, 4> Regs; + /// This list holds the number of registers for each value. + SmallVector<unsigned, 4> RegCount; + + /// Records if this value needs to be treated in an ABI dependant manner, + /// different to normal type legalization. + bool IsABIMangled; + RegsForValue(); - RegsForValue(const SmallVector<unsigned, 4> ®s, MVT regvt, EVT valuevt); + RegsForValue(const SmallVector<unsigned, 4> ®s, MVT regvt, EVT valuevt, + bool IsABIMangledValue = false); RegsForValue(LLVMContext &Context, const TargetLowering &TLI, - const DataLayout &DL, unsigned Reg, Type *Ty); + const DataLayout &DL, unsigned Reg, Type *Ty, + bool IsABIMangledValue = false); /// Add the specified values to this one. void append(const RegsForValue &RHS) { ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end()); RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end()); Regs.append(RHS.Regs.begin(), RHS.Regs.end()); + RegCount.push_back(RHS.Regs.size()); } /// Emit a series of CopyFromReg nodes that copies from this value and returns diff --git a/llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp index ac727ce34d2..5d78bba86d7 100644 --- a/llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp @@ -840,7 +840,7 @@ SelectionDAGBuilder::LowerStatepoint(ImmutableStatepoint ISP, // completely and make statepoint call to return a tuple. unsigned Reg = FuncInfo.CreateRegs(RetTy); RegsForValue RFV(*DAG.getContext(), DAG.getTargetLoweringInfo(), - DAG.getDataLayout(), Reg, RetTy); + DAG.getDataLayout(), Reg, RetTy, true); SDValue Chain = DAG.getEntryNode(); RFV.getCopyToRegs(ReturnValue, DAG, getCurSDLoc(), Chain, nullptr); diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp index bbbf14f2458..581cfaf6075 100644 --- a/llvm/lib/CodeGen/TargetLoweringBase.cpp +++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp @@ -1637,8 +1637,10 @@ void llvm::GetReturnInfo(Type *ReturnType, AttributeList attr, VT = MinVT; } - unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT); - MVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT); + unsigned NumParts = + TLI.getNumRegistersForCallingConv(ReturnType->getContext(), VT); + MVT PartVT = + TLI.getRegisterTypeForCallingConv(ReturnType->getContext(), VT); // 'inreg' on function refers to return value ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy(); diff --git a/llvm/lib/Target/Mips/MipsCCState.cpp b/llvm/lib/Target/Mips/MipsCCState.cpp index cb9f676c237..6a03ee9927d 100644 --- a/llvm/lib/Target/Mips/MipsCCState.cpp +++ b/llvm/lib/Target/Mips/MipsCCState.cpp @@ -51,6 +51,22 @@ static bool originalTypeIsF128(const Type *Ty, const char *Func) { return (Func && Ty->isIntegerTy(128) && isF128SoftLibCall(Func)); } +/// Return true if the original type was vXfXX. +static bool originalEVTTypeIsVectorFloat(EVT Ty) { + if (Ty.isVector() && Ty.getVectorElementType().isFloatingPoint()) + return true; + + return false; +} + +/// Return true if the original type was vXfXX / vXfXX. +static bool originalTypeIsVectorFloat(const Type * Ty) { + if (Ty->isVectorTy() && Ty->isFPOrFPVectorTy()) + return true; + + return false; +} + MipsCCState::SpecialCallingConvType MipsCCState::getSpecialCallingConvForCallee(const SDNode *Callee, const MipsSubtarget &Subtarget) { @@ -78,8 +94,8 @@ void MipsCCState::PreAnalyzeCallResultForF128( } } -/// Identify lowered values that originated from f128 arguments and record -/// this for use by RetCC_MipsN. +/// Identify lowered values that originated from f128 or float arguments and +/// record this for use by RetCC_MipsN. void MipsCCState::PreAnalyzeReturnForF128( const SmallVectorImpl<ISD::OutputArg> &Outs) { const MachineFunction &MF = getMachineFunction(); @@ -91,23 +107,44 @@ void MipsCCState::PreAnalyzeReturnForF128( } } -/// Identify lowered values that originated from f128 arguments and record +/// Identify lower values that originated from vXfXX and record /// this. +void MipsCCState::PreAnalyzeCallResultForVectorFloat( + const SmallVectorImpl<ISD::InputArg> &Ins, const Type *RetTy) { + for (unsigned i = 0; i < Ins.size(); ++i) { + OriginalRetWasFloatVector.push_back(originalTypeIsVectorFloat(RetTy)); + } +} + +/// Identify lowered values that originated from vXfXX arguments and record +/// this. +void MipsCCState::PreAnalyzeReturnForVectorFloat( + const SmallVectorImpl<ISD::OutputArg> &Outs) { + for (unsigned i = 0; i < Outs.size(); ++i) { + ISD::OutputArg Out = Outs[i]; + OriginalRetWasFloatVector.push_back( + originalEVTTypeIsVectorFloat(Out.ArgVT)); + } +} + +/// Identify lowered values that originated from f128, float and sret to vXfXX +/// arguments and record this. void MipsCCState::PreAnalyzeCallOperands( const SmallVectorImpl<ISD::OutputArg> &Outs, std::vector<TargetLowering::ArgListEntry> &FuncArgs, const char *Func) { for (unsigned i = 0; i < Outs.size(); ++i) { - OriginalArgWasF128.push_back( - originalTypeIsF128(FuncArgs[Outs[i].OrigArgIndex].Ty, Func)); - OriginalArgWasFloat.push_back( - FuncArgs[Outs[i].OrigArgIndex].Ty->isFloatingPointTy()); + TargetLowering::ArgListEntry FuncArg = FuncArgs[Outs[i].OrigArgIndex]; + + OriginalArgWasF128.push_back(originalTypeIsF128(FuncArg.Ty, Func)); + OriginalArgWasFloat.push_back(FuncArg.Ty->isFloatingPointTy()); + OriginalArgWasFloatVector.push_back(FuncArg.Ty->isVectorTy()); CallOperandIsFixed.push_back(Outs[i].IsFixed); } } -/// Identify lowered values that originated from f128 arguments and record -/// this. +/// Identify lowered values that originated from f128, float and vXfXX arguments +/// and record this. void MipsCCState::PreAnalyzeFormalArgumentsForF128( const SmallVectorImpl<ISD::InputArg> &Ins) { const MachineFunction &MF = getMachineFunction(); @@ -120,6 +157,7 @@ void MipsCCState::PreAnalyzeFormalArgumentsForF128( if (Ins[i].Flags.isSRet()) { OriginalArgWasF128.push_back(false); OriginalArgWasFloat.push_back(false); + OriginalArgWasFloatVector.push_back(false); continue; } @@ -129,5 +167,10 @@ void MipsCCState::PreAnalyzeFormalArgumentsForF128( OriginalArgWasF128.push_back( originalTypeIsF128(FuncArg->getType(), nullptr)); OriginalArgWasFloat.push_back(FuncArg->getType()->isFloatingPointTy()); + + // The MIPS vector ABI exhibits a corner case of sorts or quirk; if the + // first argument is actually an SRet pointer to a vector, then the next + // argument slot is $a2. + OriginalArgWasFloatVector.push_back(FuncArg->getType()->isVectorTy()); } } diff --git a/llvm/lib/Target/Mips/MipsCCState.h b/llvm/lib/Target/Mips/MipsCCState.h index 77ecc65b2ee..27901699480 100644 --- a/llvm/lib/Target/Mips/MipsCCState.h +++ b/llvm/lib/Target/Mips/MipsCCState.h @@ -45,16 +45,33 @@ private: const char *Func); /// Identify lowered values that originated from f128 arguments and record - /// this. + /// this for use by RetCC_MipsN. void PreAnalyzeFormalArgumentsForF128(const SmallVectorImpl<ISD::InputArg> &Ins); + void + PreAnalyzeCallResultForVectorFloat(const SmallVectorImpl<ISD::InputArg> &Ins, + const Type *RetTy); + + void PreAnalyzeFormalArgumentsForVectorFloat( + const SmallVectorImpl<ISD::InputArg> &Ins); + + void + PreAnalyzeReturnForVectorFloat(const SmallVectorImpl<ISD::OutputArg> &Outs); + /// Records whether the value has been lowered from an f128. SmallVector<bool, 4> OriginalArgWasF128; /// Records whether the value has been lowered from float. SmallVector<bool, 4> OriginalArgWasFloat; + /// Records whether the value has been lowered from a floating point vector. + SmallVector<bool, 4> OriginalArgWasFloatVector; + + /// Records whether the return value has been lowered from a floating point + /// vector. + SmallVector<bool, 4> OriginalRetWasFloatVector; + /// Records whether the value was a fixed argument. /// See ISD::OutputArg::IsFixed, SmallVector<bool, 4> CallOperandIsFixed; @@ -78,6 +95,7 @@ public: CCState::AnalyzeCallOperands(Outs, Fn); OriginalArgWasF128.clear(); OriginalArgWasFloat.clear(); + OriginalArgWasFloatVector.clear(); CallOperandIsFixed.clear(); } @@ -96,31 +114,38 @@ public: CCState::AnalyzeFormalArguments(Ins, Fn); OriginalArgWasFloat.clear(); OriginalArgWasF128.clear(); + OriginalArgWasFloatVector.clear(); } void AnalyzeCallResult(const SmallVectorImpl<ISD::InputArg> &Ins, CCAssignFn Fn, const Type *RetTy, const char *Func) { PreAnalyzeCallResultForF128(Ins, RetTy, Func); + PreAnalyzeCallResultForVectorFloat(Ins, RetTy); CCState::AnalyzeCallResult(Ins, Fn); OriginalArgWasFloat.clear(); OriginalArgWasF128.clear(); + OriginalArgWasFloatVector.clear(); } void AnalyzeReturn(const SmallVectorImpl<ISD::OutputArg> &Outs, CCAssignFn Fn) { PreAnalyzeReturnForF128(Outs); + PreAnalyzeReturnForVectorFloat(Outs); CCState::AnalyzeReturn(Outs, Fn); OriginalArgWasFloat.clear(); OriginalArgWasF128.clear(); + OriginalArgWasFloatVector.clear(); } bool CheckReturn(const SmallVectorImpl<ISD::OutputArg> &ArgsFlags, CCAssignFn Fn) { PreAnalyzeReturnForF128(ArgsFlags); + PreAnalyzeReturnForVectorFloat(ArgsFlags); bool Return = CCState::CheckReturn(ArgsFlags, Fn); OriginalArgWasFloat.clear(); OriginalArgWasF128.clear(); + OriginalArgWasFloatVector.clear(); return Return; } @@ -128,6 +153,12 @@ public: bool WasOriginalArgFloat(unsigned ValNo) { return OriginalArgWasFloat[ValNo]; } + bool WasOriginalArgVectorFloat(unsigned ValNo) const { + return OriginalArgWasFloatVector[ValNo]; + } + bool WasOriginalRetVectorFloat(unsigned ValNo) const { + return OriginalRetWasFloatVector[ValNo]; + } bool IsCallOperandFixed(unsigned ValNo) { return CallOperandIsFixed[ValNo]; } SpecialCallingConvType getSpecialCallingConv() { return SpecialCallingConv; } }; diff --git a/llvm/lib/Target/Mips/MipsCallingConv.td b/llvm/lib/Target/Mips/MipsCallingConv.td index a57cb7badc1..b5df78f89a6 100644 --- a/llvm/lib/Target/Mips/MipsCallingConv.td +++ b/llvm/lib/Target/Mips/MipsCallingConv.td @@ -37,6 +37,10 @@ class CCIfOrigArgWasF128<CCAction A> class CCIfArgIsVarArg<CCAction A> : CCIf<"!static_cast<MipsCCState *>(&State)->IsCallOperandFixed(ValNo)", A>; +/// Match if the return was a floating point vector. +class CCIfOrigArgWasNotVectorFloat<CCAction A> + : CCIf<"!static_cast<MipsCCState *>(&State)" + "->WasOriginalRetVectorFloat(ValNo)", A>; /// Match if the special calling conv is the specified value. class CCIfSpecialCallingConv<string CC, CCAction A> @@ -93,8 +97,10 @@ def RetCC_MipsO32 : CallingConv<[ // Promote i1/i8/i16 return values to i32. CCIfType<[i1, i8, i16], CCPromoteToType<i32>>, - // i32 are returned in registers V0, V1, A0, A1 - CCIfType<[i32], CCAssignToReg<[V0, V1, A0, A1]>>, + // i32 are returned in registers V0, V1, A0, A1, unless the original return + // type was a vector of floats. + CCIfOrigArgWasNotVectorFloat<CCIfType<[i32], + CCAssignToReg<[V0, V1, A0, A1]>>>, // f32 are returned in registers F0, F2 CCIfType<[f32], CCAssignToReg<[F0, F2]>>, diff --git a/llvm/lib/Target/Mips/MipsISelLowering.cpp b/llvm/lib/Target/Mips/MipsISelLowering.cpp index ce7dd09f429..f2193013b7a 100644 --- a/llvm/lib/Target/Mips/MipsISelLowering.cpp +++ b/llvm/lib/Target/Mips/MipsISelLowering.cpp @@ -71,6 +71,48 @@ static bool isShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) { return true; } +// The MIPS MSA ABI passes vector arguments in the integer register set. +// The number of integer registers used is dependant on the ABI used. +MVT MipsTargetLowering::getRegisterTypeForCallingConv(MVT VT) const { + if (VT.isVector() && Subtarget.hasMSA()) + return Subtarget.isABI_O32() ? MVT::i32 : MVT::i64; + return MipsTargetLowering::getRegisterType(VT); +} + +MVT MipsTargetLowering::getRegisterTypeForCallingConv(LLVMContext &Context, + EVT VT) const { + if (VT.isVector()) { + if (Subtarget.isABI_O32()) { + return MVT::i32; + } else { + return (VT.getSizeInBits() == 32) ? MVT::i32 : MVT::i64; + } + } + return MipsTargetLowering::getRegisterType(Context, VT); +} + +unsigned MipsTargetLowering::getNumRegistersForCallingConv(LLVMContext &Context, + EVT VT) const { + if (VT.isVector()) + return std::max((VT.getSizeInBits() / (Subtarget.isABI_O32() ? 32 : 64)), + 1U); + return MipsTargetLowering::getNumRegisters(Context, VT); +} + +unsigned MipsTargetLowering::getVectorTypeBreakdownForCallingConv( + LLVMContext &Context, EVT VT, EVT &IntermediateVT, + unsigned &NumIntermediates, MVT &RegisterVT) const { + + // Break down vector types to either 2 i64s or 4 i32s. + RegisterVT = getRegisterTypeForCallingConv(Context, VT) ; + IntermediateVT = RegisterVT; + NumIntermediates = VT.getSizeInBits() < RegisterVT.getSizeInBits() + ? VT.getVectorNumElements() + : VT.getSizeInBits() / RegisterVT.getSizeInBits(); + + return NumIntermediates; +} + SDValue MipsTargetLowering::getGlobalReg(SelectionDAG &DAG, EVT Ty) const { MipsFunctionInfo *FI = DAG.getMachineFunction().getInfo<MipsFunctionInfo>(); return DAG.getRegister(FI->getGlobalBaseReg(), Ty); @@ -2552,6 +2594,11 @@ SDValue MipsTargetLowering::lowerFP_TO_SINT(SDValue Op, // yet to hold an argument. Otherwise, use A2, A3 and stack. If A1 is // not used, it must be shadowed. If only A3 is available, shadow it and // go to stack. +// vXiX - Received as scalarized i32s, passed in A0 - A3 and the stack. +// vXf32 - Passed in either a pair of registers {A0, A1}, {A2, A3} or {A0 - A3} +// with the remainder spilled to the stack. +// vXf64 - Passed in either {A0, A1, A2, A3} or {A2, A3} and in both cases +// spilling the remainder to the stack. // // For vararg functions, all arguments are passed in A0, A1, A2, A3 and stack. //===----------------------------------------------------------------------===// @@ -2563,8 +2610,13 @@ static bool CC_MipsO32(unsigned ValNo, MVT ValVT, MVT LocVT, State.getMachineFunction().getSubtarget()); static const MCPhysReg IntRegs[] = { Mips::A0, Mips::A1, Mips::A2, Mips::A3 }; + + const MipsCCState * MipsState = static_cast<MipsCCState *>(&State); + static const MCPhysReg F32Regs[] = { Mips::F12, Mips::F14 }; + static const MCPhysReg FloatVectorIntRegs[] = { Mips::A0, Mips::A2 }; + // Do not process byval args here. if (ArgFlags.isByVal()) return true; @@ -2602,8 +2654,26 @@ static bool CC_MipsO32(unsigned ValNo, MVT ValVT, MVT LocVT, State.getFirstUnallocated(F32Regs) != ValNo; unsigned OrigAlign = ArgFlags.getOrigAlign(); bool isI64 = (ValVT == MVT::i32 && OrigAlign == 8); - - if (ValVT == MVT::i32 || (ValVT == MVT::f32 && AllocateFloatsInIntReg)) { + bool isVectorFloat = MipsState->WasOriginalArgVectorFloat(ValNo); + + // The MIPS vector ABI for floats passes them in a pair of registers + if (ValVT == MVT::i32 && isVectorFloat) { + // This is the start of an vector that was scalarized into an unknown number + // of components. It doesn't matter how many there are. Allocate one of the + // notional 8 byte aligned registers which map onto the argument stack, and + // shadow the register lost to alignment requirements. + if (ArgFlags.isSplit()) { + Reg = State.AllocateReg(FloatVectorIntRegs); + if (Reg == Mips::A2) + State.AllocateReg(Mips::A1); + else if (Reg == 0) + State.AllocateReg(Mips::A3); + } else { + // If we're an intermediate component of the split, we can just attempt to + // allocate a register directly. + Reg = State.AllocateReg(IntRegs); + } + } else if (ValVT == MVT::i32 || (ValVT == MVT::f32 && AllocateFloatsInIntReg)) { Reg = State.AllocateReg(IntRegs); // If this is the first part of an i64 arg, // the allocated register must be either A0 or A2. diff --git a/llvm/lib/Target/Mips/MipsISelLowering.h b/llvm/lib/Target/Mips/MipsISelLowering.h index 2dcafd51061..0e47ed38f42 100644 --- a/llvm/lib/Target/Mips/MipsISelLowering.h +++ b/llvm/lib/Target/Mips/MipsISelLowering.h @@ -248,6 +248,33 @@ namespace llvm { bool isCheapToSpeculateCttz() const override; bool isCheapToSpeculateCtlz() const override; + /// Return the register type for a given MVT, ensuring vectors are treated + /// as a series of gpr sized integers. + virtual MVT getRegisterTypeForCallingConv(MVT VT) const override; + + /// Return the register type for a given MVT, ensuring vectors are treated + /// as a series of gpr sized integers. + virtual MVT getRegisterTypeForCallingConv(LLVMContext &Context, + EVT VT) const override; + + /// Return the number of registers for a given MVT, ensuring vectors are + /// treated as a series of gpr sized integers. + virtual unsigned getNumRegistersForCallingConv(LLVMContext &Context, + EVT VT) const override; + + /// Break down vectors to the correct number of gpr sized integers. + virtual unsigned getVectorTypeBreakdownForCallingConv( + LLVMContext &Context, EVT VT, EVT &IntermediateVT, + unsigned &NumIntermediates, MVT &RegisterVT) const override; + + /// Return the correct alignment for the current calling convention. + virtual unsigned + getABIAlignmentForCallingConv(Type *ArgTy, DataLayout DL) const override { + if (ArgTy->isVectorTy()) + return std::min(DL.getABITypeAlignment(ArgTy), 8U); + return DL.getABITypeAlignment(ArgTy); + } + ISD::NodeType getExtendForAtomicOps() const override { return ISD::SIGN_EXTEND; } diff --git a/llvm/lib/Target/Mips/MipsRegisterInfo.cpp b/llvm/lib/Target/Mips/MipsRegisterInfo.cpp index 65be350f259..de3389b5a6b 100644 --- a/llvm/lib/Target/Mips/MipsRegisterInfo.cpp +++ b/llvm/lib/Target/Mips/MipsRegisterInfo.cpp @@ -286,7 +286,9 @@ eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj, DEBUG(errs() << "FrameIndex : " << FrameIndex << "\n" << "spOffset : " << spOffset << "\n" - << "stackSize : " << stackSize << "\n"); + << "stackSize : " << stackSize << "\n" + << "alignment : " + << MF.getFrameInfo().getObjectAlignment(FrameIndex) << "\n"); eliminateFI(MI, FIOperandNum, FrameIndex, stackSize, spOffset); } |