SDValue
AlphaTargetLowering::LowerReturn(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
DebugLoc dl, SelectionDAG &DAG) const {
SDValue Copy = DAG.getCopyToReg(Chain, dl, Alpha::R26,
DAG.getNode(AlphaISD::GlobalRetAddr,
DebugLoc(), MVT::i64),
SDValue());
switch (Outs.size()) {
default:
llvm_unreachable("Do not know how to return this many arguments!");
case 0:
break;
//return SDValue(); // ret void is legal
case 1: {
EVT ArgVT = Outs[0].Val.getValueType();
unsigned ArgReg;
if (ArgVT.isInteger())
ArgReg = Alpha::R0;
else {
assert(ArgVT.isFloatingPoint());
ArgReg = Alpha::F0;
}
Copy = DAG.getCopyToReg(Copy, dl, ArgReg,
Outs[0].Val, Copy.getValue(1));
if (DAG.getMachineFunction().getRegInfo().liveout_empty())
DAG.getMachineFunction().getRegInfo().addLiveOut(ArgReg);
break;
}
case 2: {
EVT ArgVT = Outs[0].Val.getValueType();
unsigned ArgReg1, ArgReg2;
if (ArgVT.isInteger()) {
ArgReg1 = Alpha::R0;
ArgReg2 = Alpha::R1;
} else {
assert(ArgVT.isFloatingPoint());
ArgReg1 = Alpha::F0;
ArgReg2 = Alpha::F1;
}
Copy = DAG.getCopyToReg(Copy, dl, ArgReg1,
Outs[0].Val, Copy.getValue(1));
if (std::find(DAG.getMachineFunction().getRegInfo().liveout_begin(),
DAG.getMachineFunction().getRegInfo().liveout_end(), ArgReg1)
== DAG.getMachineFunction().getRegInfo().liveout_end())
DAG.getMachineFunction().getRegInfo().addLiveOut(ArgReg1);
Copy = DAG.getCopyToReg(Copy, dl, ArgReg2,
Outs[1].Val, Copy.getValue(1));
if (std::find(DAG.getMachineFunction().getRegInfo().liveout_begin(),
DAG.getMachineFunction().getRegInfo().liveout_end(), ArgReg2)
== DAG.getMachineFunction().getRegInfo().liveout_end())
DAG.getMachineFunction().getRegInfo().addLiveOut(ArgReg2);
break;
}
}
return DAG.getNode(AlphaISD::RET_FLAG, dl,
MVT::Other, Copy, Copy.getValue(1));
}
/// Get the EVTs and ArgFlags collections that represent the legalized return
/// type of the given function. This does not require a DAG or a return value,
/// and is suitable for use before any DAGs for the function are constructed.
/// TODO: Move this out of TargetLowering.cpp.
void llvm::GetReturnInfo(Type* ReturnType, AttributeSet attr,
SmallVectorImpl<ISD::OutputArg> &Outs,
const TargetLowering &TLI) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, ReturnType, ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) return;
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::SExt))
ExtendKind = ISD::SIGN_EXTEND;
else if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::ZExt))
ExtendKind = ISD::ZERO_EXTEND;
// FIXME: C calling convention requires the return type to be promoted to
// at least 32-bit. But this is not necessary for non-C calling
// conventions. The frontend should mark functions whose return values
// require promoting with signext or zeroext attributes.
if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
MVT MinVT = TLI.getRegisterType(ReturnType->getContext(), MVT::i32);
if (VT.bitsLT(MinVT))
VT = MinVT;
}
unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
MVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
// 'inreg' on function refers to return value
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::InReg))
Flags.setInReg();
// Propagate extension type if any
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::SExt))
Flags.setSExt();
else if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::ZExt))
Flags.setZExt();
for (unsigned i = 0; i < NumParts; ++i)
Outs.push_back(ISD::OutputArg(Flags, PartVT, /*isFixed=*/true, 0, 0));
}
}
Value *GenericToNVVM::getOrInsertCVTA(Module *M, Function *F,
GlobalVariable *GV,
IRBuilder<> &Builder) {
PointerType *GVType = GV->getType();
Value *CVTA = nullptr;
// See if the address space conversion requires the operand to be bitcast
// to i8 addrspace(n)* first.
EVT ExtendedGVType = EVT::getEVT(GVType->getElementType(), true);
if (!ExtendedGVType.isInteger() && !ExtendedGVType.isFloatingPoint()) {
// A bitcast to i8 addrspace(n)* on the operand is needed.
LLVMContext &Context = M->getContext();
unsigned int AddrSpace = GVType->getAddressSpace();
Type *DestTy = PointerType::get(Type::getInt8Ty(Context), AddrSpace);
CVTA = Builder.CreateBitCast(GV, DestTy, "cvta");
// Insert the address space conversion.
Type *ResultType =
PointerType::get(Type::getInt8Ty(Context), llvm::ADDRESS_SPACE_GENERIC);
SmallVector<Type *, 2> ParamTypes;
ParamTypes.push_back(ResultType);
ParamTypes.push_back(DestTy);
Function *CVTAFunction = Intrinsic::getDeclaration(
M, Intrinsic::nvvm_ptr_global_to_gen, ParamTypes);
CVTA = Builder.CreateCall(CVTAFunction, CVTA, "cvta");
// Another bitcast from i8 * to <the element type of GVType> * is
// required.
DestTy =
PointerType::get(GVType->getElementType(), llvm::ADDRESS_SPACE_GENERIC);
CVTA = Builder.CreateBitCast(CVTA, DestTy, "cvta");
} else {
// A simple CVTA is enough.
SmallVector<Type *, 2> ParamTypes;
ParamTypes.push_back(PointerType::get(GVType->getElementType(),
llvm::ADDRESS_SPACE_GENERIC));
ParamTypes.push_back(GVType);
Function *CVTAFunction = Intrinsic::getDeclaration(
M, Intrinsic::nvvm_ptr_global_to_gen, ParamTypes);
CVTA = Builder.CreateCall(CVTAFunction, GV, "cvta");
}
return CVTA;
}
bool MSP430TargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
if (!VT1.isInteger() || !VT2.isInteger())
return false;
return (VT1.getSizeInBits() > VT2.getSizeInBits());
}
unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
Type *Src) const {
int ISD = TLI->InstructionOpcodeToISD(Opcode);
assert(ISD && "Invalid opcode");
// Single to/from double precision conversions.
static const CostTblEntry<MVT::SimpleValueType> NEONFltDblTbl[] = {
// Vector fptrunc/fpext conversions.
{ ISD::FP_ROUND, MVT::v2f64, 2 },
{ ISD::FP_EXTEND, MVT::v2f32, 2 },
{ ISD::FP_EXTEND, MVT::v4f32, 4 }
};
if (Src->isVectorTy() && ST->hasNEON() && (ISD == ISD::FP_ROUND ||
ISD == ISD::FP_EXTEND)) {
std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
int Idx = CostTableLookup(NEONFltDblTbl, ISD, LT.second);
if (Idx != -1)
return LT.first * NEONFltDblTbl[Idx].Cost;
}
EVT SrcTy = TLI->getValueType(Src);
EVT DstTy = TLI->getValueType(Dst);
if (!SrcTy.isSimple() || !DstTy.isSimple())
return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
// Some arithmetic, load and store operations have specific instructions
// to cast up/down their types automatically at no extra cost.
// TODO: Get these tables to know at least what the related operations are.
static const TypeConversionCostTblEntry<MVT::SimpleValueType>
NEONVectorConversionTbl[] = {
{ ISD::SIGN_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
{ ISD::ZERO_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
{ ISD::SIGN_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
{ ISD::ZERO_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
{ ISD::TRUNCATE, MVT::v4i32, MVT::v4i64, 0 },
{ ISD::TRUNCATE, MVT::v4i16, MVT::v4i32, 1 },
// The number of vmovl instructions for the extension.
{ ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
{ ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
{ ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
{ ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
{ ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
{ ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
{ ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
{ ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
{ ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
{ ISD::ZERO_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
// Operations that we legalize using splitting.
{ ISD::TRUNCATE, MVT::v16i8, MVT::v16i32, 6 },
{ ISD::TRUNCATE, MVT::v8i8, MVT::v8i32, 3 },
// Vector float <-> i32 conversions.
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 },
{ ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i8, 3 },
{ ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i8, 3 },
{ ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i16, 2 },
{ ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i16, 2 },
{ ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i32, 1 },
{ ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i32, 1 },
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i1, 3 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i1, 3 },
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i8, 3 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i8, 3 },
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 },
{ ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i16, 4 },
{ ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i16, 4 },
{ ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i32, 2 },
{ ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i32, 2 },
{ ISD::SINT_TO_FP, MVT::v16f32, MVT::v16i16, 8 },
{ ISD::UINT_TO_FP, MVT::v16f32, MVT::v16i16, 8 },
{ ISD::SINT_TO_FP, MVT::v16f32, MVT::v16i32, 4 },
{ ISD::UINT_TO_FP, MVT::v16f32, MVT::v16i32, 4 },
{ ISD::FP_TO_SINT, MVT::v4i32, MVT::v4f32, 1 },
{ ISD::FP_TO_UINT, MVT::v4i32, MVT::v4f32, 1 },
{ ISD::FP_TO_SINT, MVT::v4i8, MVT::v4f32, 3 },
{ ISD::FP_TO_UINT, MVT::v4i8, MVT::v4f32, 3 },
{ ISD::FP_TO_SINT, MVT::v4i16, MVT::v4f32, 2 },
{ ISD::FP_TO_UINT, MVT::v4i16, MVT::v4f32, 2 },
// Vector double <-> i32 conversions.
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },
{ ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i8, 4 },
{ ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i8, 4 },
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i16, 3 },
{ ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i16, 3 },
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },
{ ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },
{ ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f64, 2 },
{ ISD::FP_TO_UINT, MVT::v2i32, MVT::v2f64, 2 },
{ ISD::FP_TO_SINT, MVT::v8i16, MVT::v8f32, 4 },
{ ISD::FP_TO_UINT, MVT::v8i16, MVT::v8f32, 4 },
{ ISD::FP_TO_SINT, MVT::v16i16, MVT::v16f32, 8 },
{ ISD::FP_TO_UINT, MVT::v16i16, MVT::v16f32, 8 }
};
if (SrcTy.isVector() && ST->hasNEON()) {
int Idx = ConvertCostTableLookup(NEONVectorConversionTbl, ISD,
DstTy.getSimpleVT(), SrcTy.getSimpleVT());
if (Idx != -1)
return NEONVectorConversionTbl[Idx].Cost;
}
// Scalar float to integer conversions.
static const TypeConversionCostTblEntry<MVT::SimpleValueType>
NEONFloatConversionTbl[] = {
{ ISD::FP_TO_SINT, MVT::i1, MVT::f32, 2 },
{ ISD::FP_TO_UINT, MVT::i1, MVT::f32, 2 },
{ ISD::FP_TO_SINT, MVT::i1, MVT::f64, 2 },
{ ISD::FP_TO_UINT, MVT::i1, MVT::f64, 2 },
{ ISD::FP_TO_SINT, MVT::i8, MVT::f32, 2 },
{ ISD::FP_TO_UINT, MVT::i8, MVT::f32, 2 },
{ ISD::FP_TO_SINT, MVT::i8, MVT::f64, 2 },
{ ISD::FP_TO_UINT, MVT::i8, MVT::f64, 2 },
{ ISD::FP_TO_SINT, MVT::i16, MVT::f32, 2 },
{ ISD::FP_TO_UINT, MVT::i16, MVT::f32, 2 },
{ ISD::FP_TO_SINT, MVT::i16, MVT::f64, 2 },
{ ISD::FP_TO_UINT, MVT::i16, MVT::f64, 2 },
{ ISD::FP_TO_SINT, MVT::i32, MVT::f32, 2 },
{ ISD::FP_TO_UINT, MVT::i32, MVT::f32, 2 },
{ ISD::FP_TO_SINT, MVT::i32, MVT::f64, 2 },
{ ISD::FP_TO_UINT, MVT::i32, MVT::f64, 2 },
{ ISD::FP_TO_SINT, MVT::i64, MVT::f32, 10 },
{ ISD::FP_TO_UINT, MVT::i64, MVT::f32, 10 },
{ ISD::FP_TO_SINT, MVT::i64, MVT::f64, 10 },
{ ISD::FP_TO_UINT, MVT::i64, MVT::f64, 10 }
};
if (SrcTy.isFloatingPoint() && ST->hasNEON()) {
int Idx = ConvertCostTableLookup(NEONFloatConversionTbl, ISD,
DstTy.getSimpleVT(), SrcTy.getSimpleVT());
if (Idx != -1)
return NEONFloatConversionTbl[Idx].Cost;
}
// Scalar integer to float conversions.
static const TypeConversionCostTblEntry<MVT::SimpleValueType>
NEONIntegerConversionTbl[] = {
{ ISD::SINT_TO_FP, MVT::f32, MVT::i1, 2 },
{ ISD::UINT_TO_FP, MVT::f32, MVT::i1, 2 },
{ ISD::SINT_TO_FP, MVT::f64, MVT::i1, 2 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i1, 2 },
{ ISD::SINT_TO_FP, MVT::f32, MVT::i8, 2 },
{ ISD::UINT_TO_FP, MVT::f32, MVT::i8, 2 },
{ ISD::SINT_TO_FP, MVT::f64, MVT::i8, 2 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i8, 2 },
{ ISD::SINT_TO_FP, MVT::f32, MVT::i16, 2 },
{ ISD::UINT_TO_FP, MVT::f32, MVT::i16, 2 },
{ ISD::SINT_TO_FP, MVT::f64, MVT::i16, 2 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i16, 2 },
{ ISD::SINT_TO_FP, MVT::f32, MVT::i32, 2 },
{ ISD::UINT_TO_FP, MVT::f32, MVT::i32, 2 },
{ ISD::SINT_TO_FP, MVT::f64, MVT::i32, 2 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i32, 2 },
{ ISD::SINT_TO_FP, MVT::f32, MVT::i64, 10 },
{ ISD::UINT_TO_FP, MVT::f32, MVT::i64, 10 },
{ ISD::SINT_TO_FP, MVT::f64, MVT::i64, 10 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i64, 10 }
};
if (SrcTy.isInteger() && ST->hasNEON()) {
int Idx = ConvertCostTableLookup(NEONIntegerConversionTbl, ISD,
DstTy.getSimpleVT(), SrcTy.getSimpleVT());
if (Idx != -1)
return NEONIntegerConversionTbl[Idx].Cost;
}
// Scalar integer conversion costs.
static const TypeConversionCostTblEntry<MVT::SimpleValueType>
ARMIntegerConversionTbl[] = {
// i16 -> i64 requires two dependent operations.
{ ISD::SIGN_EXTEND, MVT::i64, MVT::i16, 2 },
// Truncates on i64 are assumed to be free.
{ ISD::TRUNCATE, MVT::i32, MVT::i64, 0 },
{ ISD::TRUNCATE, MVT::i16, MVT::i64, 0 },
{ ISD::TRUNCATE, MVT::i8, MVT::i64, 0 },
{ ISD::TRUNCATE, MVT::i1, MVT::i64, 0 }
};
if (SrcTy.isInteger()) {
int Idx = ConvertCostTableLookup(ARMIntegerConversionTbl, ISD,
DstTy.getSimpleVT(), SrcTy.getSimpleVT());
if (Idx != -1)
return ARMIntegerConversionTbl[Idx].Cost;
}
return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
}
SDValue
NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
DebugLoc &dl = CLI.DL;
SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &isTailCall = CLI.IsTailCall;
ArgListTy &Args = CLI.Args;
Type *retTy = CLI.RetTy;
ImmutableCallSite *CS = CLI.CS;
bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
SDValue tempChain = Chain;
Chain = DAG.getCALLSEQ_START(Chain,
DAG.getIntPtrConstant(uniqueCallSite, true));
SDValue InFlag = Chain.getValue(1);
assert((Outs.size() == Args.size()) &&
"Unexpected number of arguments to function call");
unsigned paramCount = 0;
// Declare the .params or .reg need to pass values
// to the function
for (unsigned i=0, e=Outs.size(); i!=e; ++i) {
EVT VT = Outs[i].VT;
if (Outs[i].Flags.isByVal() == false) {
// Plain scalar
// for ABI, declare .param .b<size> .param<n>;
// for nonABI, declare .reg .b<size> .param<n>;
unsigned isReg = 1;
if (isABI)
isReg = 0;
unsigned sz = VT.getSizeInBits();
if (VT.isInteger() && (sz < 32)) sz = 32;
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareParamOps[] = { Chain,
DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(isReg, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(0, MVT::i32), OutVals[i], InFlag };
unsigned opcode = NVPTXISD::StoreParam;
if (isReg)
opcode = NVPTXISD::MoveToParam;
else {
if (Outs[i].Flags.isZExt())
opcode = NVPTXISD::StoreParamU32;
else if (Outs[i].Flags.isSExt())
opcode = NVPTXISD::StoreParamS32;
}
Chain = DAG.getNode(opcode, dl, CopyParamVTs, CopyParamOps, 5);
InFlag = Chain.getValue(1);
++paramCount;
continue;
}
// struct or vector
SmallVector<EVT, 16> vtparts;
const PointerType *PTy = dyn_cast<PointerType>(Args[i].Ty);
assert(PTy &&
"Type of a byval parameter should be pointer");
ComputeValueVTs(*this, PTy->getElementType(), vtparts);
if (isABI) {
// declare .param .align 16 .b8 .param<n>[<size>];
unsigned sz = Outs[i].Flags.getByValSize();
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
// The ByValAlign in the Outs[i].Flags is alway set at this point, so we
// don't need to
// worry about natural alignment or not. See TargetLowering::LowerCallTo()
SDValue DeclareParamOps[] = { Chain,
DAG.getConstant(Outs[i].Flags.getByValAlign(), MVT::i32),
DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(sz, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
unsigned curOffset = 0;
for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
unsigned elems = 1;
EVT elemtype = vtparts[j];
if (vtparts[j].isVector()) {
elems = vtparts[j].getVectorNumElements();
elemtype = vtparts[j].getVectorElementType();
}
for (unsigned k=0,ke=elems; k!=ke; ++k) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 8)) sz = 8;
SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(),
OutVals[i],
DAG.getConstant(curOffset,
getPointerTy()));
SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
MachinePointerInfo(), false, false, false, 0);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount,
MVT::i32),
DAG.getConstant(curOffset, MVT::i32),
theVal, InFlag };
Chain = DAG.getNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
CopyParamOps, 5);
InFlag = Chain.getValue(1);
curOffset += sz/8;
}
}
++paramCount;
continue;
}
// Non-abi, struct or vector
// Declare a bunch or .reg .b<size> .param<n>
unsigned curOffset = 0;
for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
unsigned elems = 1;
EVT elemtype = vtparts[j];
if (vtparts[j].isVector()) {
elems = vtparts[j].getVectorNumElements();
elemtype = vtparts[j].getVectorElementType();
}
for (unsigned k=0,ke=elems; k!=ke; ++k) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 32)) sz = 32;
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareParamOps[] = { Chain, DAG.getConstant(paramCount,
MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(1, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[i],
DAG.getConstant(curOffset,
getPointerTy()));
SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
MachinePointerInfo(), false, false, false, 0);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(0, MVT::i32), theVal,
InFlag };
Chain = DAG.getNode(NVPTXISD::MoveToParam, dl, CopyParamVTs,
CopyParamOps, 5);
InFlag = Chain.getValue(1);
++paramCount;
}
}
}
GlobalAddressSDNode *Func = dyn_cast<GlobalAddressSDNode>(Callee.getNode());
unsigned retAlignment = 0;
// Handle Result
unsigned retCount = 0;
if (Ins.size() > 0) {
SmallVector<EVT, 16> resvtparts;
ComputeValueVTs(*this, retTy, resvtparts);
// Declare one .param .align 16 .b8 func_retval0[<size>] for ABI or
// individual .reg .b<size> func_retval<0..> for non ABI
unsigned resultsz = 0;
for (unsigned i=0,e=resvtparts.size(); i!=e; ++i) {
unsigned elems = 1;
EVT elemtype = resvtparts[i];
if (resvtparts[i].isVector()) {
elems = resvtparts[i].getVectorNumElements();
elemtype = resvtparts[i].getVectorElementType();
}
for (unsigned j=0,je=elems; j!=je; ++j) {
unsigned sz = elemtype.getSizeInBits();
if (isABI == false) {
if (elemtype.isInteger() && (sz < 32)) sz = 32;
}
else {
if (elemtype.isInteger() && (sz < 8)) sz = 8;
}
if (isABI == false) {
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(2, MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(retCount, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
++retCount;
}
resultsz += sz;
}
}
if (isABI) {
if (retTy->isPrimitiveType() || retTy->isIntegerTy() ||
retTy->isPointerTy() ) {
// Scalar needs to be at least 32bit wide
if (resultsz < 32)
resultsz = 32;
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(1, MVT::i32),
DAG.getConstant(resultsz, MVT::i32),
DAG.getConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
}
else {
if (Func) { // direct call
if (!llvm::getAlign(*(CS->getCalledFunction()), 0, retAlignment))
retAlignment = getDataLayout()->getABITypeAlignment(retTy);
} else { // indirect call
const CallInst *CallI = dyn_cast<CallInst>(CS->getInstruction());
if (!llvm::getAlign(*CallI, 0, retAlignment))
retAlignment = getDataLayout()->getABITypeAlignment(retTy);
}
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(retAlignment,
MVT::i32),
DAG.getConstant(resultsz/8, MVT::i32),
DAG.getConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRetParam, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
}
}
}
if (!Func) {
// This is indirect function call case : PTX requires a prototype of the
// form
// proto_0 : .callprototype(.param .b32 _) _ (.param .b32 _);
// to be emitted, and the label has to used as the last arg of call
// instruction.
// The prototype is embedded in a string and put as the operand for an
// INLINEASM SDNode.
SDVTList InlineAsmVTs = DAG.getVTList(MVT::Other, MVT::Glue);
std::string proto_string = getPrototype(retTy, Args, Outs, retAlignment);
const char *asmstr = nvTM->getManagedStrPool()->
getManagedString(proto_string.c_str())->c_str();
SDValue InlineAsmOps[] = { Chain,
DAG.getTargetExternalSymbol(asmstr,
getPointerTy()),
DAG.getMDNode(0),
DAG.getTargetConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(ISD::INLINEASM, dl, InlineAsmVTs, InlineAsmOps, 5);
InFlag = Chain.getValue(1);
}
// Op to just print "call"
SDVTList PrintCallVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue PrintCallOps[] = { Chain,
DAG.getConstant(isABI ? ((Ins.size()==0) ? 0 : 1)
: retCount, MVT::i32),
InFlag };
Chain = DAG.getNode(Func?(NVPTXISD::PrintCallUni):(NVPTXISD::PrintCall), dl,
PrintCallVTs, PrintCallOps, 3);
InFlag = Chain.getValue(1);
// Ops to print out the function name
SDVTList CallVoidVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallVoidOps[] = { Chain, Callee, InFlag };
Chain = DAG.getNode(NVPTXISD::CallVoid, dl, CallVoidVTs, CallVoidOps, 3);
InFlag = Chain.getValue(1);
// Ops to print out the param list
SDVTList CallArgBeginVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgBeginOps[] = { Chain, InFlag };
Chain = DAG.getNode(NVPTXISD::CallArgBegin, dl, CallArgBeginVTs,
CallArgBeginOps, 2);
InFlag = Chain.getValue(1);
for (unsigned i=0, e=paramCount; i!=e; ++i) {
unsigned opcode;
if (i==(e-1))
opcode = NVPTXISD::LastCallArg;
else
opcode = NVPTXISD::CallArg;
SDVTList CallArgVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgOps[] = { Chain, DAG.getConstant(1, MVT::i32),
DAG.getConstant(i, MVT::i32),
InFlag };
Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps, 4);
InFlag = Chain.getValue(1);
}
SDVTList CallArgEndVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgEndOps[] = { Chain,
DAG.getConstant(Func ? 1 : 0, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps,
3);
InFlag = Chain.getValue(1);
if (!Func) {
SDVTList PrototypeVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue PrototypeOps[] = { Chain,
DAG.getConstant(uniqueCallSite, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps, 3);
InFlag = Chain.getValue(1);
}
// Generate loads from param memory/moves from registers for result
if (Ins.size() > 0) {
if (isABI) {
unsigned resoffset = 0;
for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
unsigned sz = Ins[i].VT.getSizeInBits();
if (Ins[i].VT.isInteger() && (sz < 8)) sz = 8;
std::vector<EVT> LoadRetVTs;
LoadRetVTs.push_back(Ins[i].VT);
LoadRetVTs.push_back(MVT::Other); LoadRetVTs.push_back(MVT::Glue);
std::vector<SDValue> LoadRetOps;
LoadRetOps.push_back(Chain);
LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
LoadRetOps.push_back(DAG.getConstant(resoffset, MVT::i32));
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, LoadRetVTs,
&LoadRetOps[0], LoadRetOps.size());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
InVals.push_back(retval);
resoffset += sz/8;
}
}
else {
SmallVector<EVT, 16> resvtparts;
ComputeValueVTs(*this, retTy, resvtparts);
assert(Ins.size() == resvtparts.size() &&
"Unexpected number of return values in non-ABI case");
unsigned paramNum = 0;
for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
assert(EVT(Ins[i].VT) == resvtparts[i] &&
"Unexpected EVT type in non-ABI case");
unsigned numelems = 1;
EVT elemtype = Ins[i].VT;
if (Ins[i].VT.isVector()) {
numelems = Ins[i].VT.getVectorNumElements();
elemtype = Ins[i].VT.getVectorElementType();
}
std::vector<SDValue> tempRetVals;
for (unsigned j=0; j<numelems; ++j) {
std::vector<EVT> MoveRetVTs;
MoveRetVTs.push_back(elemtype);
MoveRetVTs.push_back(MVT::Other); MoveRetVTs.push_back(MVT::Glue);
std::vector<SDValue> MoveRetOps;
MoveRetOps.push_back(Chain);
MoveRetOps.push_back(DAG.getConstant(0, MVT::i32));
MoveRetOps.push_back(DAG.getConstant(paramNum, MVT::i32));
MoveRetOps.push_back(InFlag);
SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, MoveRetVTs,
&MoveRetOps[0], MoveRetOps.size());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
tempRetVals.push_back(retval);
++paramNum;
}
if (Ins[i].VT.isVector())
InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, dl, Ins[i].VT,
&tempRetVals[0], tempRetVals.size()));
else
InVals.push_back(tempRetVals[0]);
}
}
}
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getIntPtrConstant(uniqueCallSite, true),
DAG.getIntPtrConstant(uniqueCallSite+1, true),
InFlag);
uniqueCallSite++;
// set isTailCall to false for now, until we figure out how to express
// tail call optimization in PTX
isTailCall = false;
return Chain;
}
std::string NVPTXTargetLowering::getPrototype(Type *retTy,
const ArgListTy &Args,
const SmallVectorImpl<ISD::OutputArg> &Outs,
unsigned retAlignment) const {
bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
std::stringstream O;
O << "prototype_" << uniqueCallSite << " : .callprototype ";
if (retTy->getTypeID() == Type::VoidTyID)
O << "()";
else {
O << "(";
if (isABI) {
if (retTy->isPrimitiveType() || retTy->isIntegerTy()) {
unsigned size = 0;
if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
size = ITy->getBitWidth();
if (size < 32) size = 32;
}
else {
assert(retTy->isFloatingPointTy() &&
"Floating point type expected here");
size = retTy->getPrimitiveSizeInBits();
}
O << ".param .b" << size << " _";
}
else if (isa<PointerType>(retTy))
O << ".param .b" << getPointerTy().getSizeInBits()
<< " _";
else {
if ((retTy->getTypeID() == Type::StructTyID) ||
isa<VectorType>(retTy)) {
SmallVector<EVT, 16> vtparts;
ComputeValueVTs(*this, retTy, vtparts);
unsigned totalsz = 0;
for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
unsigned elems = 1;
EVT elemtype = vtparts[i];
if (vtparts[i].isVector()) {
elems = vtparts[i].getVectorNumElements();
elemtype = vtparts[i].getVectorElementType();
}
for (unsigned j=0, je=elems; j!=je; ++j) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 8)) sz = 8;
totalsz += sz/8;
}
}
O << ".param .align "
<< retAlignment
<< " .b8 _["
<< totalsz << "]";
}
else {
assert(false &&
"Unknown return type");
}
}
}
else {
SmallVector<EVT, 16> vtparts;
ComputeValueVTs(*this, retTy, vtparts);
unsigned idx = 0;
for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
unsigned elems = 1;
EVT elemtype = vtparts[i];
if (vtparts[i].isVector()) {
elems = vtparts[i].getVectorNumElements();
elemtype = vtparts[i].getVectorElementType();
}
for (unsigned j=0, je=elems; j!=je; ++j) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 32)) sz = 32;
O << ".reg .b" << sz << " _";
if (j<je-1) O << ", ";
++idx;
}
if (i < e-1)
O << ", ";
}
}
O << ") ";
}
O << "_ (";
bool first = true;
MVT thePointerTy = getPointerTy();
for (unsigned i=0,e=Args.size(); i!=e; ++i) {
const Type *Ty = Args[i].Ty;
if (!first) {
O << ", ";
}
first = false;
if (Outs[i].Flags.isByVal() == false) {
unsigned sz = 0;
if (isa<IntegerType>(Ty)) {
sz = cast<IntegerType>(Ty)->getBitWidth();
if (sz < 32) sz = 32;
}
else if (isa<PointerType>(Ty))
sz = thePointerTy.getSizeInBits();
else
sz = Ty->getPrimitiveSizeInBits();
if (isABI)
O << ".param .b" << sz << " ";
else
O << ".reg .b" << sz << " ";
O << "_";
continue;
}
const PointerType *PTy = dyn_cast<PointerType>(Ty);
assert(PTy &&
"Param with byval attribute should be a pointer type");
Type *ETy = PTy->getElementType();
if (isABI) {
unsigned align = Outs[i].Flags.getByValAlign();
unsigned sz = getDataLayout()->getTypeAllocSize(ETy);
O << ".param .align " << align
<< " .b8 ";
O << "_";
O << "[" << sz << "]";
continue;
}
else {
SmallVector<EVT, 16> vtparts;
ComputeValueVTs(*this, ETy, vtparts);
for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
unsigned elems = 1;
EVT elemtype = vtparts[i];
if (vtparts[i].isVector()) {
elems = vtparts[i].getVectorNumElements();
elemtype = vtparts[i].getVectorElementType();
}
for (unsigned j=0,je=elems; j!=je; ++j) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 32)) sz = 32;
O << ".reg .b" << sz << " ";
O << "_";
if (j<je-1) O << ", ";
}
if (i<e-1)
O << ", ";
}
continue;
}
}
O << ");";
return O.str();
}
/// OptimizeNoopCopyExpression - If the specified cast instruction is a noop
/// copy (e.g. it's casting from one pointer type to another, i32->i8 on PPC),
/// sink it into user blocks to reduce the number of virtual
/// registers that must be created and coalesced.
///
/// Return true if any changes are made.
///
static bool OptimizeNoopCopyExpression(CastInst *CI, const TargetLowering &TLI) {
// If this is a noop copy,
EVT SrcVT = TLI.getValueType(CI->getOperand(0)->getType());
EVT DstVT = TLI.getValueType(CI->getType());
// This is an fp<->int conversion?
if (SrcVT.isInteger() != DstVT.isInteger())
return false;
// If this is an extension, it will be a zero or sign extension, which
// isn't a noop.
if (SrcVT.bitsLT(DstVT)) return false;
// If these values will be promoted, find out what they will be promoted
// to. This helps us consider truncates on PPC as noop copies when they
// are.
if (TLI.getTypeAction(CI->getContext(), SrcVT) ==
TargetLowering::TypePromoteInteger)
SrcVT = TLI.getTypeToTransformTo(CI->getContext(), SrcVT);
if (TLI.getTypeAction(CI->getContext(), DstVT) ==
TargetLowering::TypePromoteInteger)
DstVT = TLI.getTypeToTransformTo(CI->getContext(), DstVT);
// If, after promotion, these are the same types, this is a noop copy.
if (SrcVT != DstVT)
return false;
BasicBlock *DefBB = CI->getParent();
/// InsertedCasts - Only insert a cast in each block once.
DenseMap<BasicBlock*, CastInst*> InsertedCasts;
bool MadeChange = false;
for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end();
UI != E; ) {
Use &TheUse = UI.getUse();
Instruction *User = cast<Instruction>(*UI);
// Figure out which BB this cast is used in. For PHI's this is the
// appropriate predecessor block.
BasicBlock *UserBB = User->getParent();
if (PHINode *PN = dyn_cast<PHINode>(User)) {
UserBB = PN->getIncomingBlock(UI);
}
// Preincrement use iterator so we don't invalidate it.
++UI;
// If this user is in the same block as the cast, don't change the cast.
if (UserBB == DefBB) continue;
// If we have already inserted a cast into this block, use it.
CastInst *&InsertedCast = InsertedCasts[UserBB];
if (!InsertedCast) {
BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedCast =
CastInst::Create(CI->getOpcode(), CI->getOperand(0), CI->getType(), "",
InsertPt);
MadeChange = true;
}
// Replace a use of the cast with a use of the new cast.
TheUse = InsertedCast;
++NumCastUses;
}
// If we removed all uses, nuke the cast.
if (CI->use_empty()) {
CI->eraseFromParent();
MadeChange = true;
}
return MadeChange;
}
void DAGTypeLegalizer::ExpandRes_BITCAST(SDNode *N, SDValue &Lo, SDValue &Hi) {
EVT OutVT = N->getValueType(0);
EVT NOutVT = TLI.getTypeToTransformTo(*DAG.getContext(), OutVT);
SDValue InOp = N->getOperand(0);
EVT InVT = InOp.getValueType();
SDLoc dl(N);
// Handle some special cases efficiently.
switch (getTypeAction(InVT)) {
case TargetLowering::TypeLegal:
case TargetLowering::TypePromoteInteger:
break;
case TargetLowering::TypePromoteFloat:
llvm_unreachable("Bitcast of a promotion-needing float should never need"
"expansion");
case TargetLowering::TypeSoftenFloat:
// Convert the integer operand instead.
SplitInteger(GetSoftenedFloat(InOp), Lo, Hi);
Lo = DAG.getNode(ISD::BITCAST, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BITCAST, dl, NOutVT, Hi);
return;
case TargetLowering::TypeExpandInteger:
case TargetLowering::TypeExpandFloat:
// Convert the expanded pieces of the input.
GetExpandedOp(InOp, Lo, Hi);
if (TLI.hasBigEndianPartOrdering(InVT) !=
TLI.hasBigEndianPartOrdering(OutVT))
std::swap(Lo, Hi);
Lo = DAG.getNode(ISD::BITCAST, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BITCAST, dl, NOutVT, Hi);
return;
case TargetLowering::TypeSplitVector:
GetSplitVector(InOp, Lo, Hi);
if (TLI.hasBigEndianPartOrdering(OutVT))
std::swap(Lo, Hi);
Lo = DAG.getNode(ISD::BITCAST, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BITCAST, dl, NOutVT, Hi);
return;
case TargetLowering::TypeScalarizeVector:
// Convert the element instead.
SplitInteger(BitConvertToInteger(GetScalarizedVector(InOp)), Lo, Hi);
Lo = DAG.getNode(ISD::BITCAST, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BITCAST, dl, NOutVT, Hi);
return;
case TargetLowering::TypeWidenVector: {
assert(!(InVT.getVectorNumElements() & 1) && "Unsupported BITCAST");
InOp = GetWidenedVector(InOp);
EVT LoVT, HiVT;
std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(InVT);
std::tie(Lo, Hi) = DAG.SplitVector(InOp, dl, LoVT, HiVT);
if (TLI.hasBigEndianPartOrdering(OutVT))
std::swap(Lo, Hi);
Lo = DAG.getNode(ISD::BITCAST, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BITCAST, dl, NOutVT, Hi);
return;
}
}
if (InVT.isVector() && OutVT.isInteger()) {
// Handle cases like i64 = BITCAST v1i64 on x86, where the operand
// is legal but the result is not.
unsigned NumElems = 2;
EVT ElemVT = NOutVT;
EVT NVT = EVT::getVectorVT(*DAG.getContext(), ElemVT, NumElems);
// If <ElemVT * N> is not a legal type, try <ElemVT/2 * (N*2)>.
while (!isTypeLegal(NVT)) {
unsigned NewSizeInBits = ElemVT.getSizeInBits() / 2;
// If the element size is smaller than byte, bail.
if (NewSizeInBits < 8)
break;
NumElems *= 2;
ElemVT = EVT::getIntegerVT(*DAG.getContext(), NewSizeInBits);
NVT = EVT::getVectorVT(*DAG.getContext(), ElemVT, NumElems);
}
if (isTypeLegal(NVT)) {
SDValue CastInOp = DAG.getNode(ISD::BITCAST, dl, NVT, InOp);
SmallVector<SDValue, 8> Vals;
for (unsigned i = 0; i < NumElems; ++i)
Vals.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ElemVT,
CastInOp, DAG.getConstant(i, dl,
TLI.getVectorIdxTy())));
// Build Lo, Hi pair by pairing extracted elements if needed.
unsigned Slot = 0;
for (unsigned e = Vals.size(); e - Slot > 2; Slot += 2, e += 1) {
// Each iteration will BUILD_PAIR two nodes and append the result until
// there are only two nodes left, i.e. Lo and Hi.
SDValue LHS = Vals[Slot];
SDValue RHS = Vals[Slot + 1];
if (TLI.isBigEndian())
std::swap(LHS, RHS);
Vals.push_back(DAG.getNode(ISD::BUILD_PAIR, dl,
EVT::getIntegerVT(
*DAG.getContext(),
LHS.getValueType().getSizeInBits() << 1),
LHS, RHS));
}
Lo = Vals[Slot++];
Hi = Vals[Slot++];
if (TLI.isBigEndian())
std::swap(Lo, Hi);
return;
}
}
// Lower the bit-convert to a store/load from the stack.
assert(NOutVT.isByteSized() && "Expanded type not byte sized!");
// Create the stack frame object. Make sure it is aligned for both
// the source and expanded destination types.
unsigned Alignment =
TLI.getDataLayout()->getPrefTypeAlignment(NOutVT.
getTypeForEVT(*DAG.getContext()));
SDValue StackPtr = DAG.CreateStackTemporary(InVT, Alignment);
int SPFI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
MachinePointerInfo PtrInfo = MachinePointerInfo::getFixedStack(SPFI);
// Emit a store to the stack slot.
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, InOp, StackPtr, PtrInfo,
false, false, 0);
// Load the first half from the stack slot.
Lo = DAG.getLoad(NOutVT, dl, Store, StackPtr, PtrInfo,
false, false, false, 0);
// Increment the pointer to the other half.
unsigned IncrementSize = NOutVT.getSizeInBits() / 8;
StackPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
DAG.getConstant(IncrementSize, dl,
StackPtr.getValueType()));
// Load the second half from the stack slot.
Hi = DAG.getLoad(NOutVT, dl, Store, StackPtr,
PtrInfo.getWithOffset(IncrementSize), false,
false, false, MinAlign(Alignment, IncrementSize));
// Handle endianness of the load.
if (TLI.hasBigEndianPartOrdering(OutVT))
std::swap(Lo, Hi);
}
void DAGTypeLegalizer::ExpandRes_BIT_CONVERT(SDNode *N, SDValue &Lo,
SDValue &Hi) {
EVT OutVT = N->getValueType(0);
EVT NOutVT = TLI.getTypeToTransformTo(*DAG.getContext(), OutVT);
SDValue InOp = N->getOperand(0);
EVT InVT = InOp.getValueType();
DebugLoc dl = N->getDebugLoc();
// Handle some special cases efficiently.
switch (getTypeAction(InVT)) {
default:
assert(false && "Unknown type action!");
case Legal:
case PromoteInteger:
break;
case SoftenFloat:
// Convert the integer operand instead.
SplitInteger(GetSoftenedFloat(InOp), Lo, Hi);
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Hi);
return;
case ExpandInteger:
case ExpandFloat:
// Convert the expanded pieces of the input.
GetExpandedOp(InOp, Lo, Hi);
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Hi);
return;
case SplitVector:
GetSplitVector(InOp, Lo, Hi);
if (TLI.isBigEndian())
std::swap(Lo, Hi);
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Hi);
return;
case ScalarizeVector:
// Convert the element instead.
SplitInteger(BitConvertToInteger(GetScalarizedVector(InOp)), Lo, Hi);
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Hi);
return;
case WidenVector: {
assert(!(InVT.getVectorNumElements() & 1) && "Unsupported BIT_CONVERT");
InOp = GetWidenedVector(InOp);
EVT InNVT = EVT::getVectorVT(*DAG.getContext(), InVT.getVectorElementType(),
InVT.getVectorNumElements()/2);
Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, InNVT, InOp,
DAG.getIntPtrConstant(0));
Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, InNVT, InOp,
DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
if (TLI.isBigEndian())
std::swap(Lo, Hi);
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Lo);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, Hi);
return;
}
}
if (InVT.isVector() && OutVT.isInteger()) {
// Handle cases like i64 = BIT_CONVERT v1i64 on x86, where the operand
// is legal but the result is not.
EVT NVT = EVT::getVectorVT(*DAG.getContext(), NOutVT, 2);
if (isTypeLegal(NVT)) {
SDValue CastInOp = DAG.getNode(ISD::BIT_CONVERT, dl, NVT, InOp);
Lo = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, NOutVT, CastInOp,
DAG.getIntPtrConstant(0));
Hi = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, NOutVT, CastInOp,
DAG.getIntPtrConstant(1));
if (TLI.isBigEndian())
std::swap(Lo, Hi);
return;
}
}
// Lower the bit-convert to a store/load from the stack.
assert(NOutVT.isByteSized() && "Expanded type not byte sized!");
// Create the stack frame object. Make sure it is aligned for both
// the source and expanded destination types.
unsigned Alignment =
TLI.getTargetData()->getPrefTypeAlignment(NOutVT.getTypeForEVT(*DAG.getContext()));
SDValue StackPtr = DAG.CreateStackTemporary(InVT, Alignment);
int SPFI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
const Value *SV = PseudoSourceValue::getFixedStack(SPFI);
// Emit a store to the stack slot.
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, InOp, StackPtr, SV, 0);
// Load the first half from the stack slot.
Lo = DAG.getLoad(NOutVT, dl, Store, StackPtr, SV, 0);
// Increment the pointer to the other half.
unsigned IncrementSize = NOutVT.getSizeInBits() / 8;
StackPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
DAG.getIntPtrConstant(IncrementSize));
// Load the second half from the stack slot.
Hi = DAG.getLoad(NOutVT, dl, Store, StackPtr, SV, IncrementSize, false,
MinAlign(Alignment, IncrementSize));
// Handle endianness of the load.
if (TLI.isBigEndian())
std::swap(Lo, Hi);
}
EVT TargetLoweringBase::getShiftAmountTy(EVT LHSTy) const {
assert(LHSTy.isInteger() && "Shift amount is not an integer type!");
if (LHSTy.isVector())
return LHSTy;
return getScalarShiftAmountTy(LHSTy);
}