SelectionDAGBuilder.cpp [plain text]
#define DEBUG_TYPE "isel"
#include "SDNodeDbgValue.h"
#include "SelectionDAGBuilder.h"
#include "FunctionLoweringInfo.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Constants.h"
#include "llvm/CallingConv.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/GlobalVariable.h"
#include "llvm/InlineAsm.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Module.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/DwarfWriter.h"
#include "llvm/Analysis/DebugInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
using namespace llvm;
static unsigned LimitFloatPrecision;
static cl::opt<unsigned, true>
LimitFPPrecision("limit-float-precision",
cl::desc("Generate low-precision inline sequences "
"for some float libcalls"),
cl::location(LimitFloatPrecision),
cl::init(0));
namespace {
struct RegsForValue {
const TargetLowering *TLI;
SmallVector<EVT, 4> ValueVTs;
SmallVector<EVT, 4> RegVTs;
SmallVector<unsigned, 4> Regs;
RegsForValue() : TLI(0) {}
RegsForValue(const TargetLowering &tli,
const SmallVector<unsigned, 4> ®s,
EVT regvt, EVT valuevt)
: TLI(&tli), ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {}
RegsForValue(const TargetLowering &tli,
const SmallVector<unsigned, 4> ®s,
const SmallVector<EVT, 4> ®vts,
const SmallVector<EVT, 4> &valuevts)
: TLI(&tli), ValueVTs(valuevts), RegVTs(regvts), Regs(regs) {}
RegsForValue(LLVMContext &Context, const TargetLowering &tli,
unsigned Reg, const Type *Ty) : TLI(&tli) {
ComputeValueVTs(tli, Ty, ValueVTs);
for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
EVT ValueVT = ValueVTs[Value];
unsigned NumRegs = TLI->getNumRegisters(Context, ValueVT);
EVT RegisterVT = TLI->getRegisterType(Context, ValueVT);
for (unsigned i = 0; i != NumRegs; ++i)
Regs.push_back(Reg + i);
RegVTs.push_back(RegisterVT);
Reg += NumRegs;
}
}
bool areValueTypesLegal() {
for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
EVT RegisterVT = RegVTs[Value];
if (!TLI->isTypeLegal(RegisterVT))
return false;
}
return true;
}
void append(const RegsForValue &RHS) {
TLI = RHS.TLI;
ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end());
RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
Regs.append(RHS.Regs.begin(), RHS.Regs.end());
}
SDValue getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const;
void getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const;
void AddInlineAsmOperands(unsigned Code,
bool HasMatching, unsigned MatchingIdx,
SelectionDAG &DAG,
std::vector<SDValue> &Ops) const;
};
}
static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl,
const SDValue *Parts,
unsigned NumParts, EVT PartVT, EVT ValueVT,
ISD::NodeType AssertOp = ISD::DELETED_NODE) {
assert(NumParts > 0 && "No parts to assemble!");
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDValue Val = Parts[0];
if (NumParts > 1) {
if (!ValueVT.isVector() && ValueVT.isInteger()) {
unsigned PartBits = PartVT.getSizeInBits();
unsigned ValueBits = ValueVT.getSizeInBits();
unsigned RoundParts = NumParts & (NumParts - 1) ?
1 << Log2_32(NumParts) : NumParts;
unsigned RoundBits = PartBits * RoundParts;
EVT RoundVT = RoundBits == ValueBits ?
ValueVT : EVT::getIntegerVT(*DAG.getContext(), RoundBits);
SDValue Lo, Hi;
EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(), RoundBits/2);
if (RoundParts > 2) {
Lo = getCopyFromParts(DAG, dl, Parts, RoundParts / 2,
PartVT, HalfVT);
Hi = getCopyFromParts(DAG, dl, Parts + RoundParts / 2,
RoundParts / 2, PartVT, HalfVT);
} else {
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[0]);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[1]);
}
if (TLI.isBigEndian())
std::swap(Lo, Hi);
Val = DAG.getNode(ISD::BUILD_PAIR, dl, RoundVT, Lo, Hi);
if (RoundParts < NumParts) {
unsigned OddParts = NumParts - RoundParts;
EVT OddVT = EVT::getIntegerVT(*DAG.getContext(), OddParts * PartBits);
Hi = getCopyFromParts(DAG, dl,
Parts + RoundParts, OddParts, PartVT, OddVT);
Lo = Val;
if (TLI.isBigEndian())
std::swap(Lo, Hi);
EVT TotalVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Hi = DAG.getNode(ISD::ANY_EXTEND, dl, TotalVT, Hi);
Hi = DAG.getNode(ISD::SHL, dl, TotalVT, Hi,
DAG.getConstant(Lo.getValueType().getSizeInBits(),
TLI.getPointerTy()));
Lo = DAG.getNode(ISD::ZERO_EXTEND, dl, TotalVT, Lo);
Val = DAG.getNode(ISD::OR, dl, TotalVT, Lo, Hi);
}
} else if (ValueVT.isVector()) {
EVT IntermediateVT, RegisterVT;
unsigned NumIntermediates;
unsigned NumRegs =
TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
NumIntermediates, RegisterVT);
assert(NumRegs == NumParts
&& "Part count doesn't match vector breakdown!");
NumParts = NumRegs; assert(RegisterVT == PartVT
&& "Part type doesn't match vector breakdown!");
assert(RegisterVT == Parts[0].getValueType() &&
"Part type doesn't match part!");
SmallVector<SDValue, 8> Ops(NumIntermediates);
if (NumIntermediates == NumParts) {
for (unsigned i = 0; i != NumParts; ++i)
Ops[i] = getCopyFromParts(DAG, dl, &Parts[i], 1,
PartVT, IntermediateVT);
} else if (NumParts > 0) {
assert(NumParts % NumIntermediates == 0 &&
"Must expand into a divisible number of parts!");
unsigned Factor = NumParts / NumIntermediates;
for (unsigned i = 0; i != NumIntermediates; ++i)
Ops[i] = getCopyFromParts(DAG, dl, &Parts[i * Factor], Factor,
PartVT, IntermediateVT);
}
Val = DAG.getNode(IntermediateVT.isVector() ?
ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, dl,
ValueVT, &Ops[0], NumIntermediates);
} else if (PartVT.isFloatingPoint()) {
assert(ValueVT == EVT(MVT::ppcf128) && PartVT == EVT(MVT::f64) &&
"Unexpected split");
SDValue Lo, Hi;
Lo = DAG.getNode(ISD::BIT_CONVERT, dl, EVT(MVT::f64), Parts[0]);
Hi = DAG.getNode(ISD::BIT_CONVERT, dl, EVT(MVT::f64), Parts[1]);
if (TLI.isBigEndian())
std::swap(Lo, Hi);
Val = DAG.getNode(ISD::BUILD_PAIR, dl, ValueVT, Lo, Hi);
} else {
assert(ValueVT.isFloatingPoint() && PartVT.isInteger() &&
!PartVT.isVector() && "Unexpected split");
EVT IntVT = EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits());
Val = getCopyFromParts(DAG, dl, Parts, NumParts, PartVT, IntVT);
}
}
PartVT = Val.getValueType();
if (PartVT == ValueVT)
return Val;
if (PartVT.isVector()) {
assert(ValueVT.isVector() && "Unknown vector conversion!");
return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
}
if (ValueVT.isVector()) {
assert(ValueVT.getVectorElementType() == PartVT &&
ValueVT.getVectorNumElements() == 1 &&
"Only trivial scalar-to-vector conversions should get here!");
return DAG.getNode(ISD::BUILD_VECTOR, dl, ValueVT, Val);
}
if (PartVT.isInteger() &&
ValueVT.isInteger()) {
if (ValueVT.bitsLT(PartVT)) {
if (AssertOp != ISD::DELETED_NODE)
Val = DAG.getNode(AssertOp, dl, PartVT, Val,
DAG.getValueType(ValueVT));
return DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
} else {
return DAG.getNode(ISD::ANY_EXTEND, dl, ValueVT, Val);
}
}
if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
if (ValueVT.bitsLT(Val.getValueType())) {
return DAG.getNode(ISD::FP_ROUND, dl, ValueVT, Val,
DAG.getIntPtrConstant(1));
}
return DAG.getNode(ISD::FP_EXTEND, dl, ValueVT, Val);
}
if (PartVT.getSizeInBits() == ValueVT.getSizeInBits())
return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
llvm_unreachable("Unknown mismatch!");
return SDValue();
}
static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl,
SDValue Val, SDValue *Parts, unsigned NumParts,
EVT PartVT,
ISD::NodeType ExtendKind = ISD::ANY_EXTEND) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
EVT PtrVT = TLI.getPointerTy();
EVT ValueVT = Val.getValueType();
unsigned PartBits = PartVT.getSizeInBits();
unsigned OrigNumParts = NumParts;
assert(TLI.isTypeLegal(PartVT) && "Copying to an illegal type!");
if (!NumParts)
return;
if (!ValueVT.isVector()) {
if (PartVT == ValueVT) {
assert(NumParts == 1 && "No-op copy with multiple parts!");
Parts[0] = Val;
return;
}
if (NumParts * PartBits > ValueVT.getSizeInBits()) {
if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
assert(NumParts == 1 && "Do not know what to promote to!");
Val = DAG.getNode(ISD::FP_EXTEND, dl, PartVT, Val);
} else if (PartVT.isInteger() && ValueVT.isInteger()) {
ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Val = DAG.getNode(ExtendKind, dl, ValueVT, Val);
} else {
llvm_unreachable("Unknown mismatch!");
}
} else if (PartBits == ValueVT.getSizeInBits()) {
assert(NumParts == 1 && PartVT != ValueVT);
Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
} else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
if (PartVT.isInteger() && ValueVT.isInteger()) {
ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
} else {
llvm_unreachable("Unknown mismatch!");
}
}
ValueVT = Val.getValueType();
assert(NumParts * PartBits == ValueVT.getSizeInBits() &&
"Failed to tile the value with PartVT!");
if (NumParts == 1) {
assert(PartVT == ValueVT && "Type conversion failed!");
Parts[0] = Val;
return;
}
if (NumParts & (NumParts - 1)) {
assert(PartVT.isInteger() && ValueVT.isInteger() &&
"Do not know what to expand to!");
unsigned RoundParts = 1 << Log2_32(NumParts);
unsigned RoundBits = RoundParts * PartBits;
unsigned OddParts = NumParts - RoundParts;
SDValue OddVal = DAG.getNode(ISD::SRL, dl, ValueVT, Val,
DAG.getConstant(RoundBits,
TLI.getPointerTy()));
getCopyToParts(DAG, dl, OddVal, Parts + RoundParts,
OddParts, PartVT);
if (TLI.isBigEndian())
std::reverse(Parts + RoundParts, Parts + NumParts);
NumParts = RoundParts;
ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
}
Parts[0] = DAG.getNode(ISD::BIT_CONVERT, dl,
EVT::getIntegerVT(*DAG.getContext(),
ValueVT.getSizeInBits()),
Val);
for (unsigned StepSize = NumParts; StepSize > 1; StepSize /= 2) {
for (unsigned i = 0; i < NumParts; i += StepSize) {
unsigned ThisBits = StepSize * PartBits / 2;
EVT ThisVT = EVT::getIntegerVT(*DAG.getContext(), ThisBits);
SDValue &Part0 = Parts[i];
SDValue &Part1 = Parts[i+StepSize/2];
Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
ThisVT, Part0,
DAG.getConstant(1, PtrVT));
Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
ThisVT, Part0,
DAG.getConstant(0, PtrVT));
if (ThisBits == PartBits && ThisVT != PartVT) {
Part0 = DAG.getNode(ISD::BIT_CONVERT, dl,
PartVT, Part0);
Part1 = DAG.getNode(ISD::BIT_CONVERT, dl,
PartVT, Part1);
}
}
}
if (TLI.isBigEndian())
std::reverse(Parts, Parts + OrigNumParts);
return;
}
if (NumParts == 1) {
if (PartVT != ValueVT) {
if (PartVT.getSizeInBits() == ValueVT.getSizeInBits()) {
Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
} else {
assert(ValueVT.getVectorElementType() == PartVT &&
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, PtrVT));
}
}
Parts[0] = Val;
return;
}
EVT IntermediateVT, RegisterVT;
unsigned NumIntermediates;
unsigned 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; assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
SmallVector<SDValue, 8> Ops(NumIntermediates);
for (unsigned i = 0; i != NumIntermediates; ++i) {
if (IntermediateVT.isVector())
Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl,
IntermediateVT, Val,
DAG.getConstant(i * (NumElements / NumIntermediates),
PtrVT));
else
Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
IntermediateVT, Val,
DAG.getConstant(i, PtrVT));
}
if (NumParts == NumIntermediates) {
for (unsigned i = 0; i != NumParts; ++i)
getCopyToParts(DAG, dl, Ops[i], &Parts[i], 1, PartVT);
} else if (NumParts > 0) {
assert(NumParts % NumIntermediates == 0 &&
"Must expand into a divisible number of parts!");
unsigned Factor = NumParts / NumIntermediates;
for (unsigned i = 0; i != NumIntermediates; ++i)
getCopyToParts(DAG, dl, Ops[i], &Parts[i*Factor], Factor, PartVT);
}
}
void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa) {
AA = &aa;
GFI = gfi;
TD = DAG.getTarget().getTargetData();
}
void SelectionDAGBuilder::clear() {
NodeMap.clear();
UnusedArgNodeMap.clear();
PendingLoads.clear();
PendingExports.clear();
EdgeMapping.clear();
DAG.clear();
CurDebugLoc = DebugLoc();
HasTailCall = false;
}
SDValue SelectionDAGBuilder::getRoot() {
if (PendingLoads.empty())
return DAG.getRoot();
if (PendingLoads.size() == 1) {
SDValue Root = PendingLoads[0];
DAG.setRoot(Root);
PendingLoads.clear();
return Root;
}
SDValue Root = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&PendingLoads[0], PendingLoads.size());
PendingLoads.clear();
DAG.setRoot(Root);
return Root;
}
SDValue SelectionDAGBuilder::getControlRoot() {
SDValue Root = DAG.getRoot();
if (PendingExports.empty())
return Root;
if (Root.getOpcode() != ISD::EntryToken) {
unsigned i = 0, e = PendingExports.size();
for (; i != e; ++i) {
assert(PendingExports[i].getNode()->getNumOperands() > 1);
if (PendingExports[i].getNode()->getOperand(0) == Root)
break; }
if (i == e)
PendingExports.push_back(Root);
}
Root = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&PendingExports[0],
PendingExports.size());
PendingExports.clear();
DAG.setRoot(Root);
return Root;
}
void SelectionDAGBuilder::AssignOrderingToNode(const SDNode *Node) {
if (DAG.GetOrdering(Node) != 0) return; DAG.AssignOrdering(Node, SDNodeOrder);
for (unsigned I = 0, E = Node->getNumOperands(); I != E; ++I)
AssignOrderingToNode(Node->getOperand(I).getNode());
}
void SelectionDAGBuilder::visit(Instruction &I) {
visit(I.getOpcode(), I);
}
void SelectionDAGBuilder::visit(unsigned Opcode, User &I) {
switch (Opcode) {
default: llvm_unreachable("Unknown instruction type encountered!");
#define HANDLE_INST(NUM, OPCODE, CLASS) \
case Instruction::OPCODE: visit##OPCODE((CLASS&)I); break;
#include "llvm/Instruction.def"
}
if (NodeMap.count(&I)) {
++SDNodeOrder;
AssignOrderingToNode(getValue(&I).getNode());
}
}
SDValue SelectionDAGBuilder::getValue(const Value *V) {
SDValue &N = NodeMap[V];
if (N.getNode()) return N;
if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(V))) {
EVT VT = TLI.getValueType(V->getType(), true);
if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
return N = DAG.getConstant(*CI, VT);
if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
return N = DAG.getGlobalAddress(GV, VT);
if (isa<ConstantPointerNull>(C))
return N = DAG.getConstant(0, TLI.getPointerTy());
if (ConstantFP *CFP = dyn_cast<ConstantFP>(C))
return N = DAG.getConstantFP(*CFP, VT);
if (isa<UndefValue>(C) && !V->getType()->isAggregateType())
return N = DAG.getUNDEF(VT);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
visit(CE->getOpcode(), *CE);
SDValue N1 = NodeMap[V];
assert(N1.getNode() && "visit didn't populate the ValueMap!");
return N1;
}
if (isa<ConstantStruct>(C) || isa<ConstantArray>(C)) {
SmallVector<SDValue, 4> Constants;
for (User::const_op_iterator OI = C->op_begin(), OE = C->op_end();
OI != OE; ++OI) {
SDNode *Val = getValue(*OI).getNode();
if (!Val) continue;
for (unsigned i = 0, e = Val->getNumValues(); i != e; ++i)
Constants.push_back(SDValue(Val, i));
}
return DAG.getMergeValues(&Constants[0], Constants.size(),
getCurDebugLoc());
}
if (C->getType()->isStructTy() || C->getType()->isArrayTy()) {
assert((isa<ConstantAggregateZero>(C) || isa<UndefValue>(C)) &&
"Unknown struct or array constant!");
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, C->getType(), ValueVTs);
unsigned NumElts = ValueVTs.size();
if (NumElts == 0)
return SDValue(); SmallVector<SDValue, 4> Constants(NumElts);
for (unsigned i = 0; i != NumElts; ++i) {
EVT EltVT = ValueVTs[i];
if (isa<UndefValue>(C))
Constants[i] = DAG.getUNDEF(EltVT);
else if (EltVT.isFloatingPoint())
Constants[i] = DAG.getConstantFP(0, EltVT);
else
Constants[i] = DAG.getConstant(0, EltVT);
}
return DAG.getMergeValues(&Constants[0], NumElts,
getCurDebugLoc());
}
if (BlockAddress *BA = dyn_cast<BlockAddress>(C))
return DAG.getBlockAddress(BA, VT);
const VectorType *VecTy = cast<VectorType>(V->getType());
unsigned NumElements = VecTy->getNumElements();
SmallVector<SDValue, 16> Ops;
if (ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
for (unsigned i = 0; i != NumElements; ++i)
Ops.push_back(getValue(CP->getOperand(i)));
} else {
assert(isa<ConstantAggregateZero>(C) && "Unknown vector constant!");
EVT EltVT = TLI.getValueType(VecTy->getElementType());
SDValue Op;
if (EltVT.isFloatingPoint())
Op = DAG.getConstantFP(0, EltVT);
else
Op = DAG.getConstant(0, EltVT);
Ops.assign(NumElements, Op);
}
return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
VT, &Ops[0], Ops.size());
}
if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
DenseMap<const AllocaInst*, int>::iterator SI =
FuncInfo.StaticAllocaMap.find(AI);
if (SI != FuncInfo.StaticAllocaMap.end())
return DAG.getFrameIndex(SI->second, TLI.getPointerTy());
}
unsigned InReg = FuncInfo.ValueMap[V];
assert(InReg && "Value not in map!");
RegsForValue RFV(*DAG.getContext(), TLI, InReg, V->getType());
SDValue Chain = DAG.getEntryNode();
return RFV.getCopyFromRegs(DAG, getCurDebugLoc(), Chain, NULL);
}
static void getReturnInfo(const Type* ReturnType,
Attributes attr, SmallVectorImpl<EVT> &OutVTs,
SmallVectorImpl<ISD::ArgFlagsTy> &OutFlags,
TargetLowering &TLI,
SmallVectorImpl<uint64_t> *Offsets = 0) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, ReturnType, ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) return;
unsigned Offset = 0;
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
if (attr & Attribute::SExt)
ExtendKind = ISD::SIGN_EXTEND;
else if (attr & Attribute::ZExt)
ExtendKind = ISD::ZERO_EXTEND;
if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
EVT MinVT = TLI.getRegisterType(ReturnType->getContext(), MVT::i32);
if (VT.bitsLT(MinVT))
VT = MinVT;
}
unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
PartVT.getTypeForEVT(ReturnType->getContext()));
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
if (attr & Attribute::InReg)
Flags.setInReg();
if (attr & Attribute::SExt)
Flags.setSExt();
else if (attr & Attribute::ZExt)
Flags.setZExt();
for (unsigned i = 0; i < NumParts; ++i) {
OutVTs.push_back(PartVT);
OutFlags.push_back(Flags);
if (Offsets)
{
Offsets->push_back(Offset);
Offset += PartSize;
}
}
}
}
void SelectionDAGBuilder::visitRet(ReturnInst &I) {
SDValue Chain = getControlRoot();
SmallVector<ISD::OutputArg, 8> Outs;
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
if (!FLI.CanLowerReturn) {
unsigned DemoteReg = FLI.DemoteRegister;
const Function *F = I.getParent()->getParent();
SmallVector<EVT, 1> PtrValueVTs;
ComputeValueVTs(TLI, PointerType::getUnqual(F->getReturnType()),
PtrValueVTs);
SDValue RetPtr = DAG.getRegister(DemoteReg, PtrValueVTs[0]);
SDValue RetOp = getValue(I.getOperand(0));
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs, &Offsets);
unsigned NumValues = ValueVTs.size();
SmallVector<SDValue, 4> Chains(NumValues);
EVT PtrVT = PtrValueVTs[0];
for (unsigned i = 0; i != NumValues; ++i) {
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT, RetPtr,
DAG.getConstant(Offsets[i], PtrVT));
Chains[i] =
DAG.getStore(Chain, getCurDebugLoc(),
SDValue(RetOp.getNode(), RetOp.getResNo() + i),
Add, NULL, Offsets[i], false, false, 0);
}
Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
MVT::Other, &Chains[0], NumValues);
} else if (I.getNumOperands() != 0) {
SmallVector<EVT, 4> 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();
if (F->paramHasAttr(0, Attribute::SExt))
ExtendKind = ISD::SIGN_EXTEND;
else if (F->paramHasAttr(0, Attribute::ZExt))
ExtendKind = ISD::ZERO_EXTEND;
if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
EVT MinVT = TLI.getRegisterType(*DAG.getContext(), MVT::i32);
if (VT.bitsLT(MinVT))
VT = MinVT;
}
unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), VT);
EVT PartVT = TLI.getRegisterType(*DAG.getContext(), VT);
SmallVector<SDValue, 4> Parts(NumParts);
getCopyToParts(DAG, getCurDebugLoc(),
SDValue(RetOp.getNode(), RetOp.getResNo() + j),
&Parts[0], NumParts, PartVT, ExtendKind);
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
if (F->paramHasAttr(0, Attribute::InReg))
Flags.setInReg();
if (F->paramHasAttr(0, Attribute::SExt))
Flags.setSExt();
else if (F->paramHasAttr(0, Attribute::ZExt))
Flags.setZExt();
for (unsigned i = 0; i < NumParts; ++i)
Outs.push_back(ISD::OutputArg(Flags, Parts[i], true));
}
}
}
bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
CallingConv::ID CallConv =
DAG.getMachineFunction().getFunction()->getCallingConv();
Chain = TLI.LowerReturn(Chain, CallConv, isVarArg,
Outs, getCurDebugLoc(), DAG);
assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
"LowerReturn didn't return a valid chain!");
DAG.setRoot(Chain);
}
void SelectionDAGBuilder::CopyToExportRegsIfNeeded(Value *V) {
if (!V->use_empty()) {
DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
if (VMI != FuncInfo.ValueMap.end())
CopyValueToVirtualRegister(V, VMI->second);
}
}
void SelectionDAGBuilder::ExportFromCurrentBlock(Value *V) {
if (!isa<Instruction>(V) && !isa<Argument>(V)) return;
if (FuncInfo.isExportedInst(V)) return;
unsigned Reg = FuncInfo.InitializeRegForValue(V);
CopyValueToVirtualRegister(V, Reg);
}
bool SelectionDAGBuilder::isExportableFromCurrentBlock(Value *V,
const BasicBlock *FromBB) {
if (Instruction *VI = dyn_cast<Instruction>(V)) {
if (VI->getParent() == FromBB)
return true;
return FuncInfo.isExportedInst(V);
}
if (isa<Argument>(V)) {
if (FromBB == &FromBB->getParent()->getEntryBlock())
return true;
return FuncInfo.isExportedInst(V);
}
return true;
}
static bool InBlock(const Value *V, const BasicBlock *BB) {
if (const Instruction *I = dyn_cast<Instruction>(V))
return I->getParent() == BB;
return true;
}
static ISD::CondCode getFCmpCondCode(FCmpInst::Predicate Pred) {
ISD::CondCode FPC, FOC;
switch (Pred) {
case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
default:
llvm_unreachable("Invalid FCmp predicate opcode!");
FOC = FPC = ISD::SETFALSE;
break;
}
if (FiniteOnlyFPMath())
return FOC;
else
return FPC;
}
static ISD::CondCode getICmpCondCode(ICmpInst::Predicate Pred) {
switch (Pred) {
case ICmpInst::ICMP_EQ: return ISD::SETEQ;
case ICmpInst::ICMP_NE: return ISD::SETNE;
case ICmpInst::ICMP_SLE: return ISD::SETLE;
case ICmpInst::ICMP_ULE: return ISD::SETULE;
case ICmpInst::ICMP_SGE: return ISD::SETGE;
case ICmpInst::ICMP_UGE: return ISD::SETUGE;
case ICmpInst::ICMP_SLT: return ISD::SETLT;
case ICmpInst::ICMP_ULT: return ISD::SETULT;
case ICmpInst::ICMP_SGT: return ISD::SETGT;
case ICmpInst::ICMP_UGT: return ISD::SETUGT;
default:
llvm_unreachable("Invalid ICmp predicate opcode!");
return ISD::SETNE;
}
}
void
SelectionDAGBuilder::EmitBranchForMergedCondition(Value *Cond,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
MachineBasicBlock *CurBB) {
const BasicBlock *BB = CurBB->getBasicBlock();
if (CmpInst *BOp = dyn_cast<CmpInst>(Cond)) {
if (CurBB == CurMBB ||
(isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
isExportableFromCurrentBlock(BOp->getOperand(1), BB))) {
ISD::CondCode Condition;
if (ICmpInst *IC = dyn_cast<ICmpInst>(Cond)) {
Condition = getICmpCondCode(IC->getPredicate());
} else if (FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
Condition = getFCmpCondCode(FC->getPredicate());
} else {
Condition = ISD::SETEQ; llvm_unreachable("Unknown compare instruction");
}
CaseBlock CB(Condition, BOp->getOperand(0),
BOp->getOperand(1), NULL, TBB, FBB, CurBB);
SwitchCases.push_back(CB);
return;
}
}
CaseBlock CB(ISD::SETEQ, Cond, ConstantInt::getTrue(*DAG.getContext()),
NULL, TBB, FBB, CurBB);
SwitchCases.push_back(CB);
}
void SelectionDAGBuilder::FindMergedConditions(Value *Cond,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
MachineBasicBlock *CurBB,
unsigned Opc) {
Instruction *BOp = dyn_cast<Instruction>(Cond);
if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
(unsigned)BOp->getOpcode() != Opc || !BOp->hasOneUse() ||
BOp->getParent() != CurBB->getBasicBlock() ||
!InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
!InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB);
return;
}
MachineFunction::iterator BBI = CurBB;
MachineFunction &MF = DAG.getMachineFunction();
MachineBasicBlock *TmpBB = MF.CreateMachineBasicBlock(CurBB->getBasicBlock());
CurBB->getParent()->insert(++BBI, TmpBB);
if (Opc == Instruction::Or) {
FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, Opc);
FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
} else {
assert(Opc == Instruction::And && "Unknown merge op!");
FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, Opc);
FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
}
}
bool
SelectionDAGBuilder::ShouldEmitAsBranches(const std::vector<CaseBlock> &Cases){
if (Cases.size() != 2) return true;
if ((Cases[0].CmpLHS == Cases[1].CmpLHS &&
Cases[0].CmpRHS == Cases[1].CmpRHS) ||
(Cases[0].CmpRHS == Cases[1].CmpLHS &&
Cases[0].CmpLHS == Cases[1].CmpRHS)) {
return false;
}
if (Cases[0].CmpRHS == Cases[1].CmpRHS &&
Cases[0].CC == Cases[1].CC &&
isa<Constant>(Cases[0].CmpRHS) &&
cast<Constant>(Cases[0].CmpRHS)->isNullValue()) {
if (Cases[0].CC == ISD::SETEQ && Cases[0].TrueBB == Cases[1].ThisBB)
return false;
if (Cases[0].CC == ISD::SETNE && Cases[0].FalseBB == Cases[1].ThisBB)
return false;
}
return true;
}
void SelectionDAGBuilder::visitBr(BranchInst &I) {
MachineBasicBlock *Succ0MBB = FuncInfo.MBBMap[I.getSuccessor(0)];
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
if (I.isUnconditional()) {
CurMBB->addSuccessor(Succ0MBB);
if (Succ0MBB != NextBlock)
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
MVT::Other, getControlRoot(),
DAG.getBasicBlock(Succ0MBB)));
return;
}
Value *CondVal = I.getCondition();
MachineBasicBlock *Succ1MBB = FuncInfo.MBBMap[I.getSuccessor(1)];
if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
if (BOp->hasOneUse() &&
(BOp->getOpcode() == Instruction::And ||
BOp->getOpcode() == Instruction::Or)) {
FindMergedConditions(BOp, Succ0MBB, Succ1MBB, CurMBB, BOp->getOpcode());
assert(SwitchCases[0].ThisBB == CurMBB && "Unexpected lowering!");
if (ShouldEmitAsBranches(SwitchCases)) {
for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i) {
ExportFromCurrentBlock(SwitchCases[i].CmpLHS);
ExportFromCurrentBlock(SwitchCases[i].CmpRHS);
}
visitSwitchCase(SwitchCases[0]);
SwitchCases.erase(SwitchCases.begin());
return;
}
for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i)
FuncInfo.MF->erase(SwitchCases[i].ThisBB);
SwitchCases.clear();
}
}
CaseBlock CB(ISD::SETEQ, CondVal, ConstantInt::getTrue(*DAG.getContext()),
NULL, Succ0MBB, Succ1MBB, CurMBB);
visitSwitchCase(CB);
}
void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB) {
SDValue Cond;
SDValue CondLHS = getValue(CB.CmpLHS);
DebugLoc dl = getCurDebugLoc();
if (CB.CmpMHS == NULL) {
if (CB.CmpRHS == ConstantInt::getTrue(*DAG.getContext()) &&
CB.CC == ISD::SETEQ)
Cond = CondLHS;
else if (CB.CmpRHS == ConstantInt::getFalse(*DAG.getContext()) &&
CB.CC == ISD::SETEQ) {
SDValue True = DAG.getConstant(1, CondLHS.getValueType());
Cond = DAG.getNode(ISD::XOR, dl, CondLHS.getValueType(), CondLHS, True);
} else
Cond = DAG.getSetCC(dl, MVT::i1, CondLHS, getValue(CB.CmpRHS), CB.CC);
} else {
assert(CB.CC == ISD::SETLE && "Can handle only LE ranges now");
const APInt& Low = cast<ConstantInt>(CB.CmpLHS)->getValue();
const APInt& High = cast<ConstantInt>(CB.CmpRHS)->getValue();
SDValue CmpOp = getValue(CB.CmpMHS);
EVT VT = CmpOp.getValueType();
if (cast<ConstantInt>(CB.CmpLHS)->isMinValue(true)) {
Cond = DAG.getSetCC(dl, MVT::i1, CmpOp, DAG.getConstant(High, VT),
ISD::SETLE);
} else {
SDValue SUB = DAG.getNode(ISD::SUB, dl,
VT, CmpOp, DAG.getConstant(Low, VT));
Cond = DAG.getSetCC(dl, MVT::i1, SUB,
DAG.getConstant(High-Low, VT), ISD::SETULE);
}
}
CurMBB->addSuccessor(CB.TrueBB);
CurMBB->addSuccessor(CB.FalseBB);
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
if (CB.TrueBB == NextBlock) {
std::swap(CB.TrueBB, CB.FalseBB);
SDValue True = DAG.getConstant(1, Cond.getValueType());
Cond = DAG.getNode(ISD::XOR, dl, Cond.getValueType(), Cond, True);
}
SDValue BrCond = DAG.getNode(ISD::BRCOND, dl,
MVT::Other, getControlRoot(), Cond,
DAG.getBasicBlock(CB.TrueBB));
if (BrCond.getOpcode() == ISD::BR) {
CurMBB->removeSuccessor(CB.FalseBB);
} else {
if (BrCond == getControlRoot())
CurMBB->removeSuccessor(CB.TrueBB);
if (CB.FalseBB != NextBlock)
BrCond = DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
DAG.getBasicBlock(CB.FalseBB));
}
DAG.setRoot(BrCond);
}
void SelectionDAGBuilder::visitJumpTable(JumpTable &JT) {
assert(JT.Reg != -1U && "Should lower JT Header first!");
EVT PTy = TLI.getPointerTy();
SDValue Index = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(),
JT.Reg, PTy);
SDValue Table = DAG.getJumpTable(JT.JTI, PTy);
SDValue BrJumpTable = DAG.getNode(ISD::BR_JT, getCurDebugLoc(),
MVT::Other, Index.getValue(1),
Table, Index);
DAG.setRoot(BrJumpTable);
}
void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT,
JumpTableHeader &JTH) {
SDValue SwitchOp = getValue(JTH.SValue);
EVT VT = SwitchOp.getValueType();
SDValue Sub = DAG.getNode(ISD::SUB, getCurDebugLoc(), VT, SwitchOp,
DAG.getConstant(JTH.First, VT));
SwitchOp = DAG.getZExtOrTrunc(Sub, getCurDebugLoc(), TLI.getPointerTy());
unsigned JumpTableReg = FuncInfo.MakeReg(TLI.getPointerTy());
SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
JumpTableReg, SwitchOp);
JT.Reg = JumpTableReg;
SDValue CMP = DAG.getSetCC(getCurDebugLoc(),
TLI.getSetCCResultType(Sub.getValueType()), Sub,
DAG.getConstant(JTH.Last-JTH.First,VT),
ISD::SETUGT);
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
SDValue BrCond = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
MVT::Other, CopyTo, CMP,
DAG.getBasicBlock(JT.Default));
if (JT.MBB != NextBlock)
BrCond = DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, BrCond,
DAG.getBasicBlock(JT.MBB));
DAG.setRoot(BrCond);
}
void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B) {
SDValue SwitchOp = getValue(B.SValue);
EVT VT = SwitchOp.getValueType();
SDValue Sub = DAG.getNode(ISD::SUB, getCurDebugLoc(), VT, SwitchOp,
DAG.getConstant(B.First, VT));
SDValue RangeCmp = DAG.getSetCC(getCurDebugLoc(),
TLI.getSetCCResultType(Sub.getValueType()),
Sub, DAG.getConstant(B.Range, VT),
ISD::SETUGT);
SDValue ShiftOp = DAG.getZExtOrTrunc(Sub, getCurDebugLoc(),
TLI.getPointerTy());
B.Reg = FuncInfo.MakeReg(TLI.getPointerTy());
SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
B.Reg, ShiftOp);
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
MachineBasicBlock* MBB = B.Cases[0].ThisBB;
CurMBB->addSuccessor(B.Default);
CurMBB->addSuccessor(MBB);
SDValue BrRange = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
MVT::Other, CopyTo, RangeCmp,
DAG.getBasicBlock(B.Default));
if (MBB != NextBlock)
BrRange = DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, CopyTo,
DAG.getBasicBlock(MBB));
DAG.setRoot(BrRange);
}
void SelectionDAGBuilder::visitBitTestCase(MachineBasicBlock* NextMBB,
unsigned Reg,
BitTestCase &B) {
SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(), Reg,
TLI.getPointerTy());
SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurDebugLoc(),
TLI.getPointerTy(),
DAG.getConstant(1, TLI.getPointerTy()),
ShiftOp);
SDValue AndOp = DAG.getNode(ISD::AND, getCurDebugLoc(),
TLI.getPointerTy(), SwitchVal,
DAG.getConstant(B.Mask, TLI.getPointerTy()));
SDValue AndCmp = DAG.getSetCC(getCurDebugLoc(),
TLI.getSetCCResultType(AndOp.getValueType()),
AndOp, DAG.getConstant(0, TLI.getPointerTy()),
ISD::SETNE);
CurMBB->addSuccessor(B.TargetBB);
CurMBB->addSuccessor(NextMBB);
SDValue BrAnd = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
MVT::Other, getControlRoot(),
AndCmp, DAG.getBasicBlock(B.TargetBB));
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
if (NextMBB != NextBlock)
BrAnd = DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, BrAnd,
DAG.getBasicBlock(NextMBB));
DAG.setRoot(BrAnd);
}
void SelectionDAGBuilder::visitInvoke(InvokeInst &I) {
MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];
const Value *Callee(I.getCalledValue());
if (isa<InlineAsm>(Callee))
visitInlineAsm(&I);
else
LowerCallTo(&I, getValue(Callee), false, LandingPad);
CopyToExportRegsIfNeeded(&I);
CurMBB->addSuccessor(Return);
CurMBB->addSuccessor(LandingPad);
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
MVT::Other, getControlRoot(),
DAG.getBasicBlock(Return)));
}
void SelectionDAGBuilder::visitUnwind(UnwindInst &I) {
}
bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR,
CaseRecVector& WorkList,
Value* SV,
MachineBasicBlock* Default) {
Case& BackCase = *(CR.Range.second-1);
size_t Size = CR.Range.second - CR.Range.first;
if (Size > 3)
return false;
MachineFunction *CurMF = FuncInfo.MF;
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CR.CaseBB;
if (++BBI != FuncInfo.MF->end())
NextBlock = BBI;
if (NextBlock && Default != NextBlock && BackCase.BB != NextBlock) {
for (CaseItr I = CR.Range.first, E = CR.Range.second-1; I != E; ++I) {
if (I->BB == NextBlock) {
std::swap(*I, BackCase);
break;
}
}
}
MachineBasicBlock *CurBlock = CR.CaseBB;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) {
MachineBasicBlock *FallThrough;
if (I != E-1) {
FallThrough = CurMF->CreateMachineBasicBlock(CurBlock->getBasicBlock());
CurMF->insert(BBI, FallThrough);
ExportFromCurrentBlock(SV);
} else {
FallThrough = Default;
}
Value *RHS, *LHS, *MHS;
ISD::CondCode CC;
if (I->High == I->Low) {
CC = ISD::SETEQ;
LHS = SV; RHS = I->High; MHS = NULL;
} else {
CC = ISD::SETLE;
LHS = I->Low; MHS = SV; RHS = I->High;
}
CaseBlock CB(CC, LHS, RHS, MHS, I->BB, FallThrough, CurBlock);
if (CurBlock == CurMBB)
visitSwitchCase(CB);
else
SwitchCases.push_back(CB);
CurBlock = FallThrough;
}
return true;
}
static inline bool areJTsAllowed(const TargetLowering &TLI) {
return !DisableJumpTables &&
(TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other));
}
static APInt ComputeRange(const APInt &First, const APInt &Last) {
APInt LastExt(Last), FirstExt(First);
uint32_t BitWidth = std::max(Last.getBitWidth(), First.getBitWidth()) + 1;
LastExt.sext(BitWidth); FirstExt.sext(BitWidth);
return (LastExt - FirstExt + 1ULL);
}
bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
Value* SV,
MachineBasicBlock* Default) {
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
const APInt &First = cast<ConstantInt>(FrontCase.Low)->getValue();
const APInt &Last = cast<ConstantInt>(BackCase.High)->getValue();
APInt TSize(First.getBitWidth(), 0);
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I)
TSize += I->size();
if (!areJTsAllowed(TLI) || TSize.ult(APInt(First.getBitWidth(), 4)))
return false;
APInt Range = ComputeRange(First, Last);
double Density = TSize.roundToDouble() / Range.roundToDouble();
if (Density < 0.4)
return false;
DEBUG(dbgs() << "Lowering jump table\n"
<< "First entry: " << First << ". Last entry: " << Last << '\n'
<< "Range: " << Range
<< "Size: " << TSize << ". Density: " << Density << "\n\n");
MachineFunction *CurMF = FuncInfo.MF;
MachineFunction::iterator BBI = CR.CaseBB;
++BBI;
const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();
MachineBasicBlock *JumpTableBB = CurMF->CreateMachineBasicBlock(LLVMBB);
CurMF->insert(BBI, JumpTableBB);
CR.CaseBB->addSuccessor(Default);
CR.CaseBB->addSuccessor(JumpTableBB);
std::vector<MachineBasicBlock*> DestBBs;
APInt TEI = First;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++TEI) {
const APInt &Low = cast<ConstantInt>(I->Low)->getValue();
const APInt &High = cast<ConstantInt>(I->High)->getValue();
if (Low.sle(TEI) && TEI.sle(High)) {
DestBBs.push_back(I->BB);
if (TEI==High)
++I;
} else {
DestBBs.push_back(Default);
}
}
BitVector SuccsHandled(CR.CaseBB->getParent()->getNumBlockIDs());
for (std::vector<MachineBasicBlock*>::iterator I = DestBBs.begin(),
E = DestBBs.end(); I != E; ++I) {
if (!SuccsHandled[(*I)->getNumber()]) {
SuccsHandled[(*I)->getNumber()] = true;
JumpTableBB->addSuccessor(*I);
}
}
unsigned JTEncoding = TLI.getJumpTableEncoding();
unsigned JTI = CurMF->getOrCreateJumpTableInfo(JTEncoding)
->createJumpTableIndex(DestBBs);
JumpTable JT(-1U, JTI, JumpTableBB, Default);
JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == CurMBB));
if (CR.CaseBB == CurMBB)
visitJumpTableHeader(JT, JTH);
JTCases.push_back(JumpTableBlock(JTH, JT));
return true;
}
bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
Value* SV,
MachineBasicBlock* Default) {
MachineFunction *CurMF = FuncInfo.MF;
MachineFunction::iterator BBI = CR.CaseBB;
++BBI;
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();
unsigned Size = CR.Range.second - CR.Range.first;
const APInt &First = cast<ConstantInt>(FrontCase.Low)->getValue();
const APInt &Last = cast<ConstantInt>(BackCase.High)->getValue();
double FMetric = 0;
CaseItr Pivot = CR.Range.first + Size/2;
APInt TSize(First.getBitWidth(), 0);
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I)
TSize += I->size();
APInt LSize = FrontCase.size();
APInt RSize = TSize-LSize;
DEBUG(dbgs() << "Selecting best pivot: \n"
<< "First: " << First << ", Last: " << Last <<'\n'
<< "LSize: " << LSize << ", RSize: " << RSize << '\n');
for (CaseItr I = CR.Range.first, J=I+1, E = CR.Range.second;
J!=E; ++I, ++J) {
const APInt &LEnd = cast<ConstantInt>(I->High)->getValue();
const APInt &RBegin = cast<ConstantInt>(J->Low)->getValue();
APInt Range = ComputeRange(LEnd, RBegin);
assert((Range - 2ULL).isNonNegative() &&
"Invalid case distance");
double LDensity = (double)LSize.roundToDouble() /
(LEnd - First + 1ULL).roundToDouble();
double RDensity = (double)RSize.roundToDouble() /
(Last - RBegin + 1ULL).roundToDouble();
double Metric = Range.logBase2()*(LDensity+RDensity);
DEBUG(dbgs() <<"=>Step\n"
<< "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n'
<< "LDensity: " << LDensity
<< ", RDensity: " << RDensity << '\n'
<< "Metric: " << Metric << '\n');
if (FMetric < Metric) {
Pivot = J;
FMetric = Metric;
DEBUG(dbgs() << "Current metric set to: " << FMetric << '\n');
}
LSize += J->size();
RSize -= J->size();
}
if (areJTsAllowed(TLI)) {
assert((FMetric > 0) && "Should handle dense range earlier!");
} else {
Pivot = CR.Range.first + Size/2;
}
CaseRange LHSR(CR.Range.first, Pivot);
CaseRange RHSR(Pivot, CR.Range.second);
Constant *C = Pivot->Low;
MachineBasicBlock *FalseBB = 0, *TrueBB = 0;
if ((LHSR.second - LHSR.first) == 1 &&
LHSR.first->High == CR.GE &&
cast<ConstantInt>(C)->getValue() ==
(cast<ConstantInt>(CR.GE)->getValue() + 1LL)) {
TrueBB = LHSR.first->BB;
} else {
TrueBB = CurMF->CreateMachineBasicBlock(LLVMBB);
CurMF->insert(BBI, TrueBB);
WorkList.push_back(CaseRec(TrueBB, C, CR.GE, LHSR));
ExportFromCurrentBlock(SV);
}
if ((RHSR.second - RHSR.first) == 1 && CR.LT &&
cast<ConstantInt>(RHSR.first->Low)->getValue() ==
(cast<ConstantInt>(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));
ExportFromCurrentBlock(SV);
}
CaseBlock CB(ISD::SETLT, SV, C, NULL, TrueBB, FalseBB, CR.CaseBB);
if (CR.CaseBB == CurMBB)
visitSwitchCase(CB);
else
SwitchCases.push_back(CB);
return true;
}
bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
Value* SV,
MachineBasicBlock* Default){
EVT PTy = TLI.getPointerTy();
unsigned IntPtrBits = PTy.getSizeInBits();
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
MachineFunction *CurMF = FuncInfo.MF;
if (!TLI.isOperationLegal(ISD::SHL, TLI.getPointerTy()))
return false;
size_t numCmps = 0;
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I) {
numCmps += (I->Low == I->High ? 1 : 2);
}
SmallSet<MachineBasicBlock*, 4> Dests;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I!=E; ++I) {
Dests.insert(I->BB);
if (Dests.size() > 3)
return false;
}
DEBUG(dbgs() << "Total number of unique destinations: "
<< Dests.size() << '\n'
<< "Total number of comparisons: " << numCmps << '\n');
const APInt& minValue = cast<ConstantInt>(FrontCase.Low)->getValue();
const APInt& maxValue = cast<ConstantInt>(BackCase.High)->getValue();
APInt cmpRange = maxValue - minValue;
DEBUG(dbgs() << "Compare range: " << cmpRange << '\n'
<< "Low bound: " << minValue << '\n'
<< "High bound: " << maxValue << '\n');
if (cmpRange.uge(APInt(cmpRange.getBitWidth(), IntPtrBits)) ||
(!(Dests.size() == 1 && numCmps >= 3) &&
!(Dests.size() == 2 && numCmps >= 5) &&
!(Dests.size() >= 3 && numCmps >= 6)))
return false;
DEBUG(dbgs() << "Emitting bit tests\n");
APInt lowBound = APInt::getNullValue(cmpRange.getBitWidth());
if (minValue.isNonNegative() &&
maxValue.slt(APInt(maxValue.getBitWidth(), IntPtrBits))) {
cmpRange = maxValue;
} else {
lowBound = minValue;
}
CaseBitsVector CasesBits;
unsigned i, count = 0;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I!=E; ++I) {
MachineBasicBlock* Dest = I->BB;
for (i = 0; i < count; ++i)
if (Dest == CasesBits[i].BB)
break;
if (i == count) {
assert((count < 3) && "Too much destinations to test!");
CasesBits.push_back(CaseBits(0, Dest, 0));
count++;
}
const APInt& lowValue = cast<ConstantInt>(I->Low)->getValue();
const APInt& highValue = cast<ConstantInt>(I->High)->getValue();
uint64_t lo = (lowValue - lowBound).getZExtValue();
uint64_t hi = (highValue - lowBound).getZExtValue();
for (uint64_t j = lo; j <= hi; j++) {
CasesBits[i].Mask |= 1ULL << j;
CasesBits[i].Bits++;
}
}
std::sort(CasesBits.begin(), CasesBits.end(), CaseBitsCmp());
BitTestInfo BTC;
MachineFunction::iterator BBI = CR.CaseBB;
++BBI;
const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();
DEBUG(dbgs() << "Cases:\n");
for (unsigned i = 0, e = CasesBits.size(); i!=e; ++i) {
DEBUG(dbgs() << "Mask: " << CasesBits[i].Mask
<< ", Bits: " << CasesBits[i].Bits
<< ", BB: " << CasesBits[i].BB << '\n');
MachineBasicBlock *CaseBB = CurMF->CreateMachineBasicBlock(LLVMBB);
CurMF->insert(BBI, CaseBB);
BTC.push_back(BitTestCase(CasesBits[i].Mask,
CaseBB,
CasesBits[i].BB));
ExportFromCurrentBlock(SV);
}
BitTestBlock BTB(lowBound, cmpRange, SV,
-1U, (CR.CaseBB == CurMBB),
CR.CaseBB, Default, BTC);
if (CR.CaseBB == CurMBB)
visitBitTestHeader(BTB);
BitTestCases.push_back(BTB);
return true;
}
size_t SelectionDAGBuilder::Clusterify(CaseVector& Cases,
const SwitchInst& SI) {
size_t numCmps = 0;
for (size_t i = 1; i < SI.getNumSuccessors(); ++i) {
MachineBasicBlock *SMBB = FuncInfo.MBBMap[SI.getSuccessor(i)];
Cases.push_back(Case(SI.getSuccessorValue(i),
SI.getSuccessorValue(i),
SMBB));
}
std::sort(Cases.begin(), Cases.end(), CaseCmp());
if (Cases.size() >= 2)
for (CaseItr I = Cases.begin(), J = ++(Cases.begin()); J != Cases.end(); ) {
const APInt& nextValue = cast<ConstantInt>(J->Low)->getValue();
const APInt& currentValue = cast<ConstantInt>(I->High)->getValue();
MachineBasicBlock* nextBB = J->BB;
MachineBasicBlock* currentBB = I->BB;
if ((nextValue - currentValue == 1) && (currentBB == nextBB)) {
I->High = J->High;
J = Cases.erase(J);
} else {
I = J++;
}
}
for (CaseItr I=Cases.begin(), E=Cases.end(); I!=E; ++I, ++numCmps) {
if (I->Low != I->High)
++numCmps;
}
return numCmps;
}
void SelectionDAGBuilder::visitSwitch(SwitchInst &SI) {
MachineBasicBlock *NextBlock = 0;
MachineBasicBlock *Default = FuncInfo.MBBMap[SI.getDefaultDest()];
if (SI.getNumOperands() == 2) {
CurMBB->addSuccessor(Default);
if (Default != NextBlock)
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
MVT::Other, getControlRoot(),
DAG.getBasicBlock(Default)));
return;
}
CaseVector Cases;
size_t numCmps = Clusterify(Cases, SI);
DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
<< ". Total compares: " << numCmps << '\n');
numCmps = 0;
Value *SV = SI.getOperand(0);
CaseRecVector WorkList;
WorkList.push_back(CaseRec(CurMBB,0,0,CaseRange(Cases.begin(),Cases.end())));
while (!WorkList.empty()) {
CaseRec CR = WorkList.back();
WorkList.pop_back();
if (handleBitTestsSwitchCase(CR, WorkList, SV, Default))
continue;
if (handleSmallSwitchRange(CR, WorkList, SV, Default))
continue;
if (handleJTSwitchCase(CR, WorkList, SV, Default))
continue;
handleBTSplitSwitchCase(CR, WorkList, SV, Default);
}
}
void SelectionDAGBuilder::visitIndirectBr(IndirectBrInst &I) {
SmallVector<BasicBlock*, 32> succs;
succs.reserve(I.getNumSuccessors());
for (unsigned i = 0, e = I.getNumSuccessors(); i != e; ++i)
succs.push_back(I.getSuccessor(i));
array_pod_sort(succs.begin(), succs.end());
succs.erase(std::unique(succs.begin(), succs.end()), succs.end());
for (unsigned i = 0, e = succs.size(); i != e; ++i)
CurMBB->addSuccessor(FuncInfo.MBBMap[succs[i]]);
DAG.setRoot(DAG.getNode(ISD::BRIND, getCurDebugLoc(),
MVT::Other, getControlRoot(),
getValue(I.getAddress())));
}
void SelectionDAGBuilder::visitFSub(User &I) {
const Type *Ty = I.getType();
if (Ty->isVectorTy()) {
if (ConstantVector *CV = dyn_cast<ConstantVector>(I.getOperand(0))) {
const VectorType *DestTy = cast<VectorType>(I.getType());
const Type *ElTy = DestTy->getElementType();
unsigned VL = DestTy->getNumElements();
std::vector<Constant*> NZ(VL, ConstantFP::getNegativeZero(ElTy));
Constant *CNZ = ConstantVector::get(&NZ[0], NZ.size());
if (CV == CNZ) {
SDValue Op2 = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
Op2.getValueType(), Op2));
return;
}
}
}
if (ConstantFP *CFP = dyn_cast<ConstantFP>(I.getOperand(0)))
if (CFP->isExactlyValue(ConstantFP::getNegativeZero(Ty)->getValueAPF())) {
SDValue Op2 = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
Op2.getValueType(), Op2));
return;
}
visitBinary(I, ISD::FSUB);
}
void SelectionDAGBuilder::visitBinary(User &I, unsigned OpCode) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(OpCode, getCurDebugLoc(),
Op1.getValueType(), Op1, Op2));
}
void SelectionDAGBuilder::visitShift(User &I, unsigned Opcode) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
if (!I.getType()->isVectorTy() &&
Op2.getValueType() != TLI.getShiftAmountTy()) {
EVT PTy = TLI.getPointerTy();
EVT STy = TLI.getShiftAmountTy();
if (STy.bitsGT(Op2.getValueType()))
Op2 = DAG.getNode(ISD::ANY_EXTEND, getCurDebugLoc(),
TLI.getShiftAmountTy(), Op2);
else if (STy.getSizeInBits() >=
Log2_32_Ceil(Op2.getValueType().getSizeInBits()))
Op2 = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
TLI.getShiftAmountTy(), Op2);
else if (PTy.bitsLT(Op2.getValueType()))
Op2 = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
TLI.getPointerTy(), Op2);
else if (PTy.bitsGT(Op2.getValueType()))
Op2 = DAG.getNode(ISD::ANY_EXTEND, getCurDebugLoc(),
TLI.getPointerTy(), Op2);
}
setValue(&I, DAG.getNode(Opcode, getCurDebugLoc(),
Op1.getValueType(), Op1, Op2));
}
void SelectionDAGBuilder::visitICmp(User &I) {
ICmpInst::Predicate predicate = ICmpInst::BAD_ICMP_PREDICATE;
if (ICmpInst *IC = dyn_cast<ICmpInst>(&I))
predicate = IC->getPredicate();
else if (ConstantExpr *IC = dyn_cast<ConstantExpr>(&I))
predicate = ICmpInst::Predicate(IC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Opcode = getICmpCondCode(predicate);
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Opcode));
}
void SelectionDAGBuilder::visitFCmp(User &I) {
FCmpInst::Predicate predicate = FCmpInst::BAD_FCMP_PREDICATE;
if (FCmpInst *FC = dyn_cast<FCmpInst>(&I))
predicate = FC->getPredicate();
else if (ConstantExpr *FC = dyn_cast<ConstantExpr>(&I))
predicate = FCmpInst::Predicate(FC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Condition = getFCmpCondCode(predicate);
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Condition));
}
void SelectionDAGBuilder::visitSelect(User &I) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I.getType(), ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) return;
SmallVector<SDValue, 4> Values(NumValues);
SDValue Cond = getValue(I.getOperand(0));
SDValue TrueVal = getValue(I.getOperand(1));
SDValue FalseVal = getValue(I.getOperand(2));
for (unsigned i = 0; i != NumValues; ++i)
Values[i] = DAG.getNode(ISD::SELECT, getCurDebugLoc(),
TrueVal.getNode()->getValueType(TrueVal.getResNo()+i),
Cond,
SDValue(TrueVal.getNode(),
TrueVal.getResNo() + i),
SDValue(FalseVal.getNode(),
FalseVal.getResNo() + i));
setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&ValueVTs[0], NumValues),
&Values[0], NumValues));
}
void SelectionDAGBuilder::visitTrunc(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitZExt(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitSExt(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitFPTrunc(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurDebugLoc(),
DestVT, N, DAG.getIntPtrConstant(0)));
}
void SelectionDAGBuilder::visitFPExt(User &I){
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitFPToUI(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_TO_UINT, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitFPToSI(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_TO_SINT, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitUIToFP(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::UINT_TO_FP, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitSIToFP(User &I){
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::SINT_TO_FP, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGBuilder::visitPtrToInt(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT SrcVT = N.getValueType();
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
}
void SelectionDAGBuilder::visitIntToPtr(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT SrcVT = N.getValueType();
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
}
void SelectionDAGBuilder::visitBitCast(User &I) {
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
if (DestVT != N.getValueType())
setValue(&I, DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
DestVT, N)); else
setValue(&I, N); }
void SelectionDAGBuilder::visitInsertElement(User &I) {
SDValue InVec = getValue(I.getOperand(0));
SDValue InVal = getValue(I.getOperand(1));
SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
TLI.getPointerTy(),
getValue(I.getOperand(2)));
setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurDebugLoc(),
TLI.getValueType(I.getType()),
InVec, InVal, InIdx));
}
void SelectionDAGBuilder::visitExtractElement(User &I) {
SDValue InVec = getValue(I.getOperand(0));
SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
TLI.getPointerTy(),
getValue(I.getOperand(1)));
setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
TLI.getValueType(I.getType()), InVec, InIdx));
}
static bool SequentialMask(SmallVectorImpl<int> &Mask, unsigned SIndx) {
unsigned MaskNumElts = Mask.size();
for (unsigned i = 0; i != MaskNumElts; ++i)
if ((Mask[i] >= 0) && (Mask[i] != (int)(i + SIndx)))
return false;
return true;
}
void SelectionDAGBuilder::visitShuffleVector(User &I) {
SmallVector<int, 8> Mask;
SDValue Src1 = getValue(I.getOperand(0));
SDValue Src2 = getValue(I.getOperand(1));
SmallVector<Constant*, 8> MaskElts;
cast<Constant>(I.getOperand(2))->getVectorElements(MaskElts);
unsigned MaskNumElts = MaskElts.size();
for (unsigned i = 0; i != MaskNumElts; ++i) {
if (isa<UndefValue>(MaskElts[i]))
Mask.push_back(-1);
else
Mask.push_back(cast<ConstantInt>(MaskElts[i])->getSExtValue());
}
EVT VT = TLI.getValueType(I.getType());
EVT SrcVT = Src1.getValueType();
unsigned SrcNumElts = SrcVT.getVectorNumElements();
if (SrcNumElts == MaskNumElts) {
setValue(&I, DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
&Mask[0]));
return;
}
if (SrcNumElts < MaskNumElts && MaskNumElts % SrcNumElts == 0) {
if (SrcNumElts*2 == MaskNumElts && SequentialMask(Mask, 0)) {
setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
VT, Src1, Src2));
return;
}
unsigned NumConcat = MaskNumElts / SrcNumElts;
bool Src1U = Src1.getOpcode() == ISD::UNDEF;
bool Src2U = Src2.getOpcode() == ISD::UNDEF;
SDValue UndefVal = DAG.getUNDEF(SrcVT);
SmallVector<SDValue, 8> MOps1(NumConcat, UndefVal);
SmallVector<SDValue, 8> MOps2(NumConcat, UndefVal);
MOps1[0] = Src1;
MOps2[0] = Src2;
Src1 = Src1U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
getCurDebugLoc(), VT,
&MOps1[0], NumConcat);
Src2 = Src2U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
getCurDebugLoc(), VT,
&MOps2[0], NumConcat);
SmallVector<int, 8> MappedOps;
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
if (Idx < (int)SrcNumElts)
MappedOps.push_back(Idx);
else
MappedOps.push_back(Idx + MaskNumElts - SrcNumElts);
}
setValue(&I, DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
&MappedOps[0]));
return;
}
if (SrcNumElts > MaskNumElts) {
int MinRange[2] = { SrcNumElts+1, SrcNumElts+1};
int MaxRange[2] = {-1, -1};
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
int Input = 0;
if (Idx < 0)
continue;
if (Idx >= (int)SrcNumElts) {
Input = 1;
Idx -= SrcNumElts;
}
if (Idx > MaxRange[Input])
MaxRange[Input] = Idx;
if (Idx < MinRange[Input])
MinRange[Input] = Idx;
}
int RangeUse[2] = { 2, 2 }; int StartIdx[2]; for (int Input=0; Input < 2; ++Input) {
if (MinRange[Input] == (int)(SrcNumElts+1) && MaxRange[Input] == -1) {
RangeUse[Input] = 0; StartIdx[Input] = 0;
} else if (MaxRange[Input] - MinRange[Input] < (int)MaskNumElts) {
if (MaxRange[Input] < (int)MaskNumElts) {
RangeUse[Input] = 1; StartIdx[Input] = 0;
} else {
StartIdx[Input] = (MinRange[Input]/MaskNumElts)*MaskNumElts;
if (MaxRange[Input] - StartIdx[Input] < (int)MaskNumElts &&
StartIdx[Input] + MaskNumElts < SrcNumElts)
RangeUse[Input] = 1; }
}
}
if (RangeUse[0] == 0 && RangeUse[1] == 0) {
setValue(&I, DAG.getUNDEF(VT)); return;
}
else if (RangeUse[0] < 2 && RangeUse[1] < 2) {
for (int Input=0; Input < 2; ++Input) {
SDValue &Src = Input == 0 ? Src1 : Src2;
if (RangeUse[Input] == 0)
Src = DAG.getUNDEF(VT);
else
Src = DAG.getNode(ISD::EXTRACT_SUBVECTOR, getCurDebugLoc(), VT,
Src, DAG.getIntPtrConstant(StartIdx[Input]));
}
SmallVector<int, 8> MappedOps;
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
if (Idx < 0)
MappedOps.push_back(Idx);
else if (Idx < (int)SrcNumElts)
MappedOps.push_back(Idx - StartIdx[0]);
else
MappedOps.push_back(Idx - SrcNumElts - StartIdx[1] + MaskNumElts);
}
setValue(&I, DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
&MappedOps[0]));
return;
}
}
EVT EltVT = VT.getVectorElementType();
EVT PtrVT = TLI.getPointerTy();
SmallVector<SDValue,8> Ops;
for (unsigned i = 0; i != MaskNumElts; ++i) {
if (Mask[i] < 0) {
Ops.push_back(DAG.getUNDEF(EltVT));
} else {
int Idx = Mask[i];
SDValue Res;
if (Idx < (int)SrcNumElts)
Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
EltVT, Src1, DAG.getConstant(Idx, PtrVT));
else
Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
EltVT, Src2,
DAG.getConstant(Idx - SrcNumElts, PtrVT));
Ops.push_back(Res);
}
}
setValue(&I, DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
VT, &Ops[0], Ops.size()));
}
void SelectionDAGBuilder::visitInsertValue(InsertValueInst &I) {
const Value *Op0 = I.getOperand(0);
const Value *Op1 = I.getOperand(1);
const Type *AggTy = I.getType();
const Type *ValTy = Op1->getType();
bool IntoUndef = isa<UndefValue>(Op0);
bool FromUndef = isa<UndefValue>(Op1);
unsigned LinearIndex = ComputeLinearIndex(TLI, AggTy,
I.idx_begin(), I.idx_end());
SmallVector<EVT, 4> AggValueVTs;
ComputeValueVTs(TLI, AggTy, AggValueVTs);
SmallVector<EVT, 4> ValValueVTs;
ComputeValueVTs(TLI, ValTy, ValValueVTs);
unsigned NumAggValues = AggValueVTs.size();
unsigned NumValValues = ValValueVTs.size();
SmallVector<SDValue, 4> Values(NumAggValues);
SDValue Agg = getValue(Op0);
SDValue Val = getValue(Op1);
unsigned i = 0;
for (; i != LinearIndex; ++i)
Values[i] = IntoUndef ? DAG.getUNDEF(AggValueVTs[i]) :
SDValue(Agg.getNode(), Agg.getResNo() + i);
for (; i != LinearIndex + NumValValues; ++i)
Values[i] = FromUndef ? DAG.getUNDEF(AggValueVTs[i]) :
SDValue(Val.getNode(), Val.getResNo() + i - LinearIndex);
for (; i != NumAggValues; ++i)
Values[i] = IntoUndef ? DAG.getUNDEF(AggValueVTs[i]) :
SDValue(Agg.getNode(), Agg.getResNo() + i);
setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&AggValueVTs[0], NumAggValues),
&Values[0], NumAggValues));
}
void SelectionDAGBuilder::visitExtractValue(ExtractValueInst &I) {
const Value *Op0 = I.getOperand(0);
const Type *AggTy = Op0->getType();
const Type *ValTy = I.getType();
bool OutOfUndef = isa<UndefValue>(Op0);
unsigned LinearIndex = ComputeLinearIndex(TLI, AggTy,
I.idx_begin(), I.idx_end());
SmallVector<EVT, 4> ValValueVTs;
ComputeValueVTs(TLI, ValTy, ValValueVTs);
unsigned NumValValues = ValValueVTs.size();
SmallVector<SDValue, 4> Values(NumValValues);
SDValue Agg = getValue(Op0);
for (unsigned i = LinearIndex; i != LinearIndex + NumValValues; ++i)
Values[i - LinearIndex] =
OutOfUndef ?
DAG.getUNDEF(Agg.getNode()->getValueType(Agg.getResNo() + i)) :
SDValue(Agg.getNode(), Agg.getResNo() + i);
setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&ValValueVTs[0], NumValValues),
&Values[0], NumValValues));
}
void SelectionDAGBuilder::visitGetElementPtr(User &I) {
SDValue N = getValue(I.getOperand(0));
const Type *Ty = I.getOperand(0)->getType();
for (GetElementPtrInst::op_iterator OI = I.op_begin()+1, E = I.op_end();
OI != E; ++OI) {
Value *Idx = *OI;
if (const StructType *StTy = dyn_cast<StructType>(Ty)) {
unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
if (Field) {
uint64_t Offset = TD->getStructLayout(StTy)->getElementOffset(Field);
N = DAG.getNode(ISD::ADD, getCurDebugLoc(), N.getValueType(), N,
DAG.getIntPtrConstant(Offset));
}
Ty = StTy->getElementType(Field);
} else {
Ty = cast<SequentialType>(Ty)->getElementType();
if (ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
if (CI->getZExtValue() == 0) continue;
uint64_t Offs =
TD->getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
SDValue OffsVal;
EVT PTy = TLI.getPointerTy();
unsigned PtrBits = PTy.getSizeInBits();
if (PtrBits < 64)
OffsVal = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
TLI.getPointerTy(),
DAG.getConstant(Offs, MVT::i64));
else
OffsVal = DAG.getIntPtrConstant(Offs);
N = DAG.getNode(ISD::ADD, getCurDebugLoc(), N.getValueType(), N,
OffsVal);
continue;
}
APInt ElementSize = APInt(TLI.getPointerTy().getSizeInBits(),
TD->getTypeAllocSize(Ty));
SDValue IdxN = getValue(Idx);
IdxN = DAG.getSExtOrTrunc(IdxN, getCurDebugLoc(), N.getValueType());
if (ElementSize != 1) {
if (ElementSize.isPowerOf2()) {
unsigned Amt = ElementSize.logBase2();
IdxN = DAG.getNode(ISD::SHL, getCurDebugLoc(),
N.getValueType(), IdxN,
DAG.getConstant(Amt, TLI.getPointerTy()));
} else {
SDValue Scale = DAG.getConstant(ElementSize, TLI.getPointerTy());
IdxN = DAG.getNode(ISD::MUL, getCurDebugLoc(),
N.getValueType(), IdxN, Scale);
}
}
N = DAG.getNode(ISD::ADD, getCurDebugLoc(),
N.getValueType(), N, IdxN);
}
}
setValue(&I, N);
}
void SelectionDAGBuilder::visitAlloca(AllocaInst &I) {
if (FuncInfo.StaticAllocaMap.count(&I))
return;
const Type *Ty = I.getAllocatedType();
uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(Ty);
unsigned Align =
std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty),
I.getAlignment());
SDValue AllocSize = getValue(I.getArraySize());
AllocSize = DAG.getNode(ISD::MUL, getCurDebugLoc(), AllocSize.getValueType(),
AllocSize,
DAG.getConstant(TySize, AllocSize.getValueType()));
EVT IntPtr = TLI.getPointerTy();
AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurDebugLoc(), IntPtr);
unsigned StackAlign =
TLI.getTargetMachine().getFrameInfo()->getStackAlignment();
if (Align <= StackAlign)
Align = 0;
AllocSize = DAG.getNode(ISD::ADD, getCurDebugLoc(),
AllocSize.getValueType(), AllocSize,
DAG.getIntPtrConstant(StackAlign-1));
AllocSize = DAG.getNode(ISD::AND, getCurDebugLoc(),
AllocSize.getValueType(), AllocSize,
DAG.getIntPtrConstant(~(uint64_t)(StackAlign-1)));
SDValue Ops[] = { getRoot(), AllocSize, DAG.getIntPtrConstant(Align) };
SDVTList VTs = DAG.getVTList(AllocSize.getValueType(), MVT::Other);
SDValue DSA = DAG.getNode(ISD::DYNAMIC_STACKALLOC, getCurDebugLoc(),
VTs, Ops, 3);
setValue(&I, DSA);
DAG.setRoot(DSA.getValue(1));
FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject();
}
void SelectionDAGBuilder::visitLoad(LoadInst &I) {
const Value *SV = I.getOperand(0);
SDValue Ptr = getValue(SV);
const Type *Ty = I.getType();
bool isVolatile = I.isVolatile();
bool isNonTemporal = I.getMetadata("nontemporal") != 0;
unsigned Alignment = I.getAlignment();
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
ComputeValueVTs(TLI, Ty, ValueVTs, &Offsets);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0)
return;
SDValue Root;
bool ConstantMemory = false;
if (I.isVolatile())
Root = getRoot();
else if (AA->pointsToConstantMemory(SV)) {
Root = DAG.getEntryNode();
ConstantMemory = true;
} else {
Root = DAG.getRoot();
}
SmallVector<SDValue, 4> Values(NumValues);
SmallVector<SDValue, 4> Chains(NumValues);
EVT PtrVT = Ptr.getValueType();
for (unsigned i = 0; i != NumValues; ++i) {
SDValue A = DAG.getNode(ISD::ADD, getCurDebugLoc(),
PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT));
SDValue L = DAG.getLoad(ValueVTs[i], getCurDebugLoc(), Root,
A, SV, Offsets[i], isVolatile,
isNonTemporal, Alignment);
Values[i] = L;
Chains[i] = L.getValue(1);
}
if (!ConstantMemory) {
SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
MVT::Other, &Chains[0], NumValues);
if (isVolatile)
DAG.setRoot(Chain);
else
PendingLoads.push_back(Chain);
}
setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&ValueVTs[0], NumValues),
&Values[0], NumValues));
}
void SelectionDAGBuilder::visitStore(StoreInst &I) {
Value *SrcV = I.getOperand(0);
Value *PtrV = I.getOperand(1);
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
ComputeValueVTs(TLI, SrcV->getType(), ValueVTs, &Offsets);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0)
return;
SDValue Src = getValue(SrcV);
SDValue Ptr = getValue(PtrV);
SDValue Root = getRoot();
SmallVector<SDValue, 4> Chains(NumValues);
EVT PtrVT = Ptr.getValueType();
bool isVolatile = I.isVolatile();
bool isNonTemporal = I.getMetadata("nontemporal") != 0;
unsigned Alignment = I.getAlignment();
for (unsigned i = 0; i != NumValues; ++i) {
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT));
Chains[i] = DAG.getStore(Root, getCurDebugLoc(),
SDValue(Src.getNode(), Src.getResNo() + i),
Add, PtrV, Offsets[i], isVolatile,
isNonTemporal, Alignment);
}
DAG.setRoot(DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
MVT::Other, &Chains[0], NumValues));
}
void SelectionDAGBuilder::visitTargetIntrinsic(CallInst &I,
unsigned Intrinsic) {
bool HasChain = !I.doesNotAccessMemory();
bool OnlyLoad = HasChain && I.onlyReadsMemory();
SmallVector<SDValue, 8> Ops;
if (HasChain) { if (OnlyLoad) {
Ops.push_back(DAG.getRoot());
} else {
Ops.push_back(getRoot());
}
}
TargetLowering::IntrinsicInfo Info;
bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I, Intrinsic);
if (!IsTgtIntrinsic)
Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
for (unsigned i = 1, e = I.getNumOperands(); i != e; ++i) {
SDValue Op = getValue(I.getOperand(i));
assert(TLI.isTypeLegal(Op.getValueType()) &&
"Intrinsic uses a non-legal type?");
Ops.push_back(Op);
}
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I.getType(), ValueVTs);
#ifndef NDEBUG
for (unsigned Val = 0, E = ValueVTs.size(); Val != E; ++Val) {
assert(TLI.isTypeLegal(ValueVTs[Val]) &&
"Intrinsic uses a non-legal type?");
}
#endif // NDEBUG
if (HasChain)
ValueVTs.push_back(MVT::Other);
SDVTList VTs = DAG.getVTList(ValueVTs.data(), ValueVTs.size());
SDValue Result;
if (IsTgtIntrinsic) {
Result = DAG.getMemIntrinsicNode(Info.opc, getCurDebugLoc(),
VTs, &Ops[0], Ops.size(),
Info.memVT, Info.ptrVal, Info.offset,
Info.align, Info.vol,
Info.readMem, Info.writeMem);
} else if (!HasChain) {
Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, getCurDebugLoc(),
VTs, &Ops[0], Ops.size());
} else if (!I.getType()->isVoidTy()) {
Result = DAG.getNode(ISD::INTRINSIC_W_CHAIN, getCurDebugLoc(),
VTs, &Ops[0], Ops.size());
} else {
Result = DAG.getNode(ISD::INTRINSIC_VOID, getCurDebugLoc(),
VTs, &Ops[0], Ops.size());
}
if (HasChain) {
SDValue Chain = Result.getValue(Result.getNode()->getNumValues()-1);
if (OnlyLoad)
PendingLoads.push_back(Chain);
else
DAG.setRoot(Chain);
}
if (!I.getType()->isVoidTy()) {
if (const VectorType *PTy = dyn_cast<VectorType>(I.getType())) {
EVT VT = TLI.getValueType(PTy);
Result = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(), VT, Result);
}
setValue(&I, Result);
}
}
static SDValue
GetSignificand(SelectionDAG &DAG, SDValue Op, DebugLoc dl) {
SDValue t1 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
DAG.getConstant(0x007fffff, MVT::i32));
SDValue t2 = DAG.getNode(ISD::OR, dl, MVT::i32, t1,
DAG.getConstant(0x3f800000, MVT::i32));
return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t2);
}
static SDValue
GetExponent(SelectionDAG &DAG, SDValue Op, const TargetLowering &TLI,
DebugLoc dl) {
SDValue t0 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
DAG.getConstant(0x7f800000, MVT::i32));
SDValue t1 = DAG.getNode(ISD::SRL, dl, MVT::i32, t0,
DAG.getConstant(23, TLI.getPointerTy()));
SDValue t2 = DAG.getNode(ISD::SUB, dl, MVT::i32, t1,
DAG.getConstant(127, MVT::i32));
return DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, t2);
}
static SDValue
getF32Constant(SelectionDAG &DAG, unsigned Flt) {
return DAG.getConstantFP(APFloat(APInt(32, Flt)), MVT::f32);
}
const char *
SelectionDAGBuilder::implVisitBinaryAtomic(CallInst& I, ISD::NodeType Op) {
SDValue Root = getRoot();
SDValue L =
DAG.getAtomic(Op, getCurDebugLoc(),
getValue(I.getOperand(2)).getValueType().getSimpleVT(),
Root,
getValue(I.getOperand(1)),
getValue(I.getOperand(2)),
I.getOperand(1));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
}
const char *
SelectionDAGBuilder::implVisitAluOverflow(CallInst &I, ISD::NodeType Op) {
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDVTList VTs = DAG.getVTList(Op1.getValueType(), MVT::i1);
setValue(&I, DAG.getNode(Op, getCurDebugLoc(), VTs, Op1, Op2));
return 0;
}
void
SelectionDAGBuilder::visitExp(CallInst &I) {
SDValue result;
DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
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);
SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1);
IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
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);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t5);
SDValue t6 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t6);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
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);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t7);
SDValue t8 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t8);
} else { 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);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::i32, t13);
SDValue t14 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t14);
}
} else {
result = DAG.getNode(ISD::FEXP, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
setValue(&I, result);
}
void
SelectionDAGBuilder::visitLog(CallInst &I) {
SDValue result;
DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
getF32Constant(DAG, 0x3f317218));
SDValue X = GetSignificand(DAG, Op1, dl);
if (LimitFloatPrecision <= 6) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbe74c456));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3fb3a2b1));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f949a29));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, LogOfMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbd67b6d6));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3ee4f4b8));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fbc278b));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40348e95));
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3fdef31a));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, LogOfMantissa);
} else { SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbc91e5ac));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e4350aa));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f60d3e3));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x4011cdf0));
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x406cfd1c));
SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x408797cb));
SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
getF32Constant(DAG, 0x4006dcab));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, LogOfMantissa);
}
} else {
result = DAG.getNode(ISD::FLOG, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
setValue(&I, result);
}
void
SelectionDAGBuilder::visitLog2(CallInst &I) {
SDValue result;
DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl);
SDValue X = GetSignificand(DAG, Op1, dl);
if (LimitFloatPrecision <= 6) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbeb08fe0));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x40019463));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fd6633d));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log2ofMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbda7262e));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3f25280b));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x4007b923));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40823e2f));
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x4020d29c));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log2ofMantissa);
} else { SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbcd2769e));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e8ce0b9));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fa22ae7));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40525723));
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x40aaf200));
SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x40c39dad));
SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
getF32Constant(DAG, 0x4042902c));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log2ofMantissa);
}
} else {
result = DAG.getNode(ISD::FLOG2, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
setValue(&I, result);
}
void
SelectionDAGBuilder::visitLog10(CallInst &I) {
SDValue result;
DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
getF32Constant(DAG, 0x3e9a209a));
SDValue X = GetSignificand(DAG, Op1, dl);
if (LimitFloatPrecision <= 6) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbdd49a13));
SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3f1c0789));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f011300));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log10ofMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3d431f31));
SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3ea21fb2));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f6ae232));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f25f7c3));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log10ofMantissa);
} else { SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3c5d51ce));
SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e00685a));
SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3efb6798));
SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
SDValue t5 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f88d192));
SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3fc4316c));
SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t8,
getF32Constant(DAG, 0x3f57ce70));
result = DAG.getNode(ISD::FADD, dl,
MVT::f32, LogOfExponent, Log10ofMantissa);
}
} else {
result = DAG.getNode(ISD::FLOG10, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
setValue(&I, result);
}
void
SelectionDAGBuilder::visitExp2(CallInst &I) {
SDValue result;
DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Op);
SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, Op, t1);
IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
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);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
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);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
} else { 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);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
}
} else {
result = DAG.getNode(ISD::FEXP2, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
setValue(&I, result);
}
void
SelectionDAGBuilder::visitPow(CallInst &I) {
SDValue result;
Value *Val = I.getOperand(1);
DebugLoc dl = getCurDebugLoc();
bool IsExp10 = false;
if (getValue(Val).getValueType() == MVT::f32 &&
getValue(I.getOperand(2)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(Val))) {
if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
APFloat Ten(10.0f);
IsExp10 = CFP->getValueAPF().bitwiseIsEqual(Ten);
}
}
}
if (IsExp10 && LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(2));
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op,
getF32Constant(DAG, 0x40549a78));
SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0);
SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1);
IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
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);
SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
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);
SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
} else { 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);
SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
result = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f32, TwoToFractionalPartOfX);
}
} else {
result = DAG.getNode(ISD::FPOW, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)),
getValue(I.getOperand(2)));
}
setValue(&I, result);
}
static SDValue ExpandPowI(DebugLoc DL, SDValue LHS, SDValue RHS,
SelectionDAG &DAG) {
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
unsigned Val = RHSC->getSExtValue();
if ((int)Val < 0) Val = -Val;
if (Val == 0)
return DAG.getConstantFP(1.0, LHS.getValueType());
Function *F = DAG.getMachineFunction().getFunction();
if (!F->hasFnAttr(Attribute::OptimizeForSize) ||
CountPopulation_32(Val)+Log2_32(Val) < 7) {
SDValue Res; SDValue CurSquare = LHS;
while (Val) {
if (Val & 1) {
if (Res.getNode())
Res = DAG.getNode(ISD::FMUL, DL,Res.getValueType(), Res, CurSquare);
else
Res = CurSquare; }
CurSquare = DAG.getNode(ISD::FMUL, DL, CurSquare.getValueType(),
CurSquare, CurSquare);
Val >>= 1;
}
if (RHSC->getSExtValue() < 0)
Res = DAG.getNode(ISD::FDIV, DL, LHS.getValueType(),
DAG.getConstantFP(1.0, LHS.getValueType()), Res);
return Res;
}
}
return DAG.getNode(ISD::FPOWI, DL, LHS.getValueType(), LHS, RHS);
}
bool
SelectionDAGBuilder::EmitFuncArgumentDbgValue(const DbgValueInst &DI,
const Value *V, MDNode *Variable,
uint64_t Offset, SDValue &N) {
if (!isa<Argument>(V))
return false;
MachineFunction &MF = DAG.getMachineFunction();
DIVariable DV(Variable);
if (DV.isInlinedFnArgument(MF.getFunction()))
return false;
MachineBasicBlock *MBB = FuncInfo.MBBMap[DI.getParent()];
if (MBB != &MF.front())
return false;
unsigned Reg = 0;
if (N.getOpcode() == ISD::CopyFromReg) {
Reg = cast<RegisterSDNode>(N.getOperand(1))->getReg();
if (Reg && TargetRegisterInfo::isVirtualRegister(Reg)) {
MachineRegisterInfo &RegInfo = MF.getRegInfo();
unsigned PR = RegInfo.getLiveInPhysReg(Reg);
if (PR)
Reg = PR;
}
}
if (!Reg && N.getNode())
if (LoadSDNode *LNode = dyn_cast<LoadSDNode>(N.getNode()))
if (FrameIndexSDNode *FINode =
dyn_cast<FrameIndexSDNode>(LNode->getBasePtr().getNode())) {
const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
const TargetInstrInfo *TII = DAG.getTarget().getInstrInfo();
MachineInstrBuilder MIB = BuildMI(MF, getCurDebugLoc(),
TII->get(TargetOpcode::DBG_VALUE))
.addReg(TRI->getFrameRegister(MF), RegState::Debug)
.addImm(FINode->getIndex())
.addMetadata(Variable);
FuncInfo.ArgDbgValues.push_back(&*MIB);
return true;
}
if (!Reg) {
DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
if (VMI == FuncInfo.ValueMap.end())
return false;
Reg = VMI->second;
}
const TargetInstrInfo *TII = DAG.getTarget().getInstrInfo();
MachineInstrBuilder MIB = BuildMI(MF, getCurDebugLoc(),
TII->get(TargetOpcode::DBG_VALUE))
.addReg(Reg, RegState::Debug).addImm(Offset).addMetadata(Variable);
FuncInfo.ArgDbgValues.push_back(&*MIB);
return true;
}
const char *
SelectionDAGBuilder::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
DebugLoc dl = getCurDebugLoc();
SDValue Res;
switch (Intrinsic) {
default:
visitTargetIntrinsic(I, Intrinsic);
return 0;
case Intrinsic::vastart: visitVAStart(I); return 0;
case Intrinsic::vaend: visitVAEnd(I); return 0;
case Intrinsic::vacopy: visitVACopy(I); return 0;
case Intrinsic::returnaddress:
setValue(&I, DAG.getNode(ISD::RETURNADDR, dl, TLI.getPointerTy(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::frameaddress:
setValue(&I, DAG.getNode(ISD::FRAMEADDR, dl, TLI.getPointerTy(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::setjmp:
return "_setjmp"+!TLI.usesUnderscoreSetJmp();
case Intrinsic::longjmp:
return "_longjmp"+!TLI.usesUnderscoreLongJmp();
case Intrinsic::memcpy: {
assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
< 256 &&
cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol, false,
I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
case Intrinsic::memset: {
assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
DAG.setRoot(DAG.getMemset(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
I.getOperand(1), 0));
return 0;
}
case Intrinsic::memmove: {
assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
< 256 &&
cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
< 256 &&
"Unknown address space");
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
uint64_t Size = -1ULL;
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op3))
Size = C->getZExtValue();
if (AA->alias(I.getOperand(1), Size, I.getOperand(2), Size) ==
AliasAnalysis::NoAlias) {
DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
false, I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
DAG.setRoot(DAG.getMemmove(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
case Intrinsic::dbg_declare: {
DwarfWriter *DW = DAG.getDwarfWriter();
if (!DW)
return 0;
DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
return 0;
MDNode *Variable = DI.getVariable();
bool isParameter =
DIVariable(Variable).getTag() == dwarf::DW_TAG_arg_variable;
const Value *Address = DI.getAddress();
if (!Address)
return 0;
if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
Address = BCI->getOperand(0);
const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
if (AI) {
DenseMap<const AllocaInst*, int>::iterator SI =
FuncInfo.StaticAllocaMap.find(AI);
if (SI == FuncInfo.StaticAllocaMap.end())
return 0; int FI = SI->second;
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
if (!DI.getDebugLoc().isUnknown() && MMI->hasDebugInfo())
MMI->setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
}
++SDNodeOrder;
SDValue &N = NodeMap[Address];
SDDbgValue *SDV;
if (N.getNode()) {
if (isParameter && !AI) {
FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(N.getNode());
if (FINode)
SDV = DAG.getDbgValue(Variable, FINode->getIndex(),
0, dl, SDNodeOrder);
else
return 0;
} else if (AI)
SDV = DAG.getDbgValue(Variable, N.getNode(), N.getResNo(),
0, dl, SDNodeOrder);
else
return 0;
DAG.AddDbgValue(SDV, N.getNode(), isParameter);
} else {
SDV = DAG.getDbgValue(Variable, UndefValue::get(Address->getType()),
0, dl, SDNodeOrder);
DAG.AddDbgValue(SDV, 0, isParameter);
}
return 0;
}
case Intrinsic::dbg_value: {
DwarfWriter *DW = DAG.getDwarfWriter();
if (!DW)
return 0;
DbgValueInst &DI = cast<DbgValueInst>(I);
if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
return 0;
MDNode *Variable = DI.getVariable();
uint64_t Offset = DI.getOffset();
Value *V = DI.getValue();
if (!V)
return 0;
++SDNodeOrder;
SDDbgValue *SDV;
if (isa<ConstantInt>(V) || isa<ConstantFP>(V)) {
SDV = DAG.getDbgValue(Variable, V, Offset, dl, SDNodeOrder);
DAG.AddDbgValue(SDV, 0, false);
} else {
bool createUndef = false;
SDValue N = NodeMap[V];
if (!N.getNode() && isa<Argument>(V))
N = UnusedArgNodeMap[V];
if (N.getNode()) {
if (!EmitFuncArgumentDbgValue(DI, V, Variable, Offset, N)) {
SDV = DAG.getDbgValue(Variable, N.getNode(),
N.getResNo(), Offset, dl, SDNodeOrder);
DAG.AddDbgValue(SDV, N.getNode(), false);
}
} else if (isa<PHINode>(V) && !V->use_empty()) {
SDValue N = getValue(V);
if (N.getNode()) {
if (!EmitFuncArgumentDbgValue(DI, V, Variable, Offset, N)) {
SDV = DAG.getDbgValue(Variable, N.getNode(),
N.getResNo(), Offset, dl, SDNodeOrder);
DAG.AddDbgValue(SDV, N.getNode(), false);
}
} else
createUndef = true;
} else
createUndef = true;
if (createUndef) {
SDV = DAG.getDbgValue(Variable, UndefValue::get(V->getType()),
Offset, dl, SDNodeOrder);
DAG.AddDbgValue(SDV, 0, false);
}
}
if (BitCastInst *BCI = dyn_cast<BitCastInst>(V))
V = BCI->getOperand(0);
AllocaInst *AI = dyn_cast<AllocaInst>(V);
if (!AI)
return 0;
DenseMap<const AllocaInst*, int>::iterator SI =
FuncInfo.StaticAllocaMap.find(AI);
if (SI == FuncInfo.StaticAllocaMap.end())
return 0; int FI = SI->second;
if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo())
if (!DI.getDebugLoc().isUnknown())
MMI->setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
return 0;
}
case Intrinsic::eh_exception: {
assert(CurMBB->isLandingPad() &&"Call to eh.exception not in landing pad!");
SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
SDValue Ops[1];
Ops[0] = DAG.getRoot();
SDValue Op = DAG.getNode(ISD::EXCEPTIONADDR, dl, VTs, Ops, 1);
setValue(&I, Op);
DAG.setRoot(Op.getValue(1));
return 0;
}
case Intrinsic::eh_selector: {
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
if (CurMBB->isLandingPad())
AddCatchInfo(I, MMI, CurMBB);
else {
#ifndef NDEBUG
FuncInfo.CatchInfoLost.insert(&I);
#endif
unsigned Reg = TLI.getExceptionSelectorRegister();
if (Reg) CurMBB->addLiveIn(Reg);
}
SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
SDValue Ops[2];
Ops[0] = getValue(I.getOperand(1));
Ops[1] = getRoot();
SDValue Op = DAG.getNode(ISD::EHSELECTION, dl, VTs, Ops, 2);
DAG.setRoot(Op.getValue(1));
setValue(&I, DAG.getSExtOrTrunc(Op, dl, MVT::i32));
return 0;
}
case Intrinsic::eh_typeid_for: {
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
if (MMI) {
GlobalVariable *GV = ExtractTypeInfo(I.getOperand(1));
unsigned TypeID = MMI->getTypeIDFor(GV);
Res = DAG.getConstant(TypeID, MVT::i32);
} else {
Res = DAG.getConstant(1, MVT::i32);
}
setValue(&I, Res);
return 0;
}
case Intrinsic::eh_return_i32:
case Intrinsic::eh_return_i64:
if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo()) {
MMI->setCallsEHReturn(true);
DAG.setRoot(DAG.getNode(ISD::EH_RETURN, dl,
MVT::Other,
getControlRoot(),
getValue(I.getOperand(1)),
getValue(I.getOperand(2))));
} else {
setValue(&I, DAG.getConstant(0, TLI.getPointerTy()));
}
return 0;
case Intrinsic::eh_unwind_init:
if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo()) {
MMI->setCallsUnwindInit(true);
}
return 0;
case Intrinsic::eh_dwarf_cfa: {
EVT VT = getValue(I.getOperand(1)).getValueType();
SDValue CfaArg = DAG.getSExtOrTrunc(getValue(I.getOperand(1)), dl,
TLI.getPointerTy());
SDValue Offset = DAG.getNode(ISD::ADD, dl,
TLI.getPointerTy(),
DAG.getNode(ISD::FRAME_TO_ARGS_OFFSET, dl,
TLI.getPointerTy()),
CfaArg);
SDValue FA = DAG.getNode(ISD::FRAMEADDR, dl,
TLI.getPointerTy(),
DAG.getConstant(0, TLI.getPointerTy()));
setValue(&I, DAG.getNode(ISD::ADD, dl, TLI.getPointerTy(),
FA, Offset));
return 0;
}
case Intrinsic::eh_sjlj_callsite: {
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(1));
assert(CI && "Non-constant call site value in eh.sjlj.callsite!");
assert(MMI->getCurrentCallSite() == 0 && "Overlapping call sites!");
MMI->setCurrentCallSite(CI->getZExtValue());
return 0;
}
case Intrinsic::eh_sjlj_longjmp: {
DAG.setRoot(DAG.getNode(ISD::EH_SJLJ_LONGJMP, dl, MVT::Other, getRoot(),
getValue(I.getOperand(1))));
return 0;
}
case Intrinsic::convertff:
case Intrinsic::convertfsi:
case Intrinsic::convertfui:
case Intrinsic::convertsif:
case Intrinsic::convertuif:
case Intrinsic::convertss:
case Intrinsic::convertsu:
case Intrinsic::convertus:
case Intrinsic::convertuu: {
ISD::CvtCode Code = ISD::CVT_INVALID;
switch (Intrinsic) {
case Intrinsic::convertff: Code = ISD::CVT_FF; break;
case Intrinsic::convertfsi: Code = ISD::CVT_FS; break;
case Intrinsic::convertfui: Code = ISD::CVT_FU; break;
case Intrinsic::convertsif: Code = ISD::CVT_SF; break;
case Intrinsic::convertuif: Code = ISD::CVT_UF; break;
case Intrinsic::convertss: Code = ISD::CVT_SS; break;
case Intrinsic::convertsu: Code = ISD::CVT_SU; break;
case Intrinsic::convertus: Code = ISD::CVT_US; break;
case Intrinsic::convertuu: Code = ISD::CVT_UU; break;
}
EVT DestVT = TLI.getValueType(I.getType());
Value *Op1 = I.getOperand(1);
Res = DAG.getConvertRndSat(DestVT, getCurDebugLoc(), getValue(Op1),
DAG.getValueType(DestVT),
DAG.getValueType(getValue(Op1).getValueType()),
getValue(I.getOperand(2)),
getValue(I.getOperand(3)),
Code);
setValue(&I, Res);
return 0;
}
case Intrinsic::sqrt:
setValue(&I, DAG.getNode(ISD::FSQRT, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::powi:
setValue(&I, ExpandPowI(dl, getValue(I.getOperand(1)),
getValue(I.getOperand(2)), DAG));
return 0;
case Intrinsic::sin:
setValue(&I, DAG.getNode(ISD::FSIN, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::cos:
setValue(&I, DAG.getNode(ISD::FCOS, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::log:
visitLog(I);
return 0;
case Intrinsic::log2:
visitLog2(I);
return 0;
case Intrinsic::log10:
visitLog10(I);
return 0;
case Intrinsic::exp:
visitExp(I);
return 0;
case Intrinsic::exp2:
visitExp2(I);
return 0;
case Intrinsic::pow:
visitPow(I);
return 0;
case Intrinsic::convert_to_fp16:
setValue(&I, DAG.getNode(ISD::FP32_TO_FP16, dl,
MVT::i16, getValue(I.getOperand(1))));
return 0;
case Intrinsic::convert_from_fp16:
setValue(&I, DAG.getNode(ISD::FP16_TO_FP32, dl,
MVT::f32, getValue(I.getOperand(1))));
return 0;
case Intrinsic::pcmarker: {
SDValue Tmp = getValue(I.getOperand(1));
DAG.setRoot(DAG.getNode(ISD::PCMARKER, dl, MVT::Other, getRoot(), Tmp));
return 0;
}
case Intrinsic::readcyclecounter: {
SDValue Op = getRoot();
Res = DAG.getNode(ISD::READCYCLECOUNTER, dl,
DAG.getVTList(MVT::i64, MVT::Other),
&Op, 1);
setValue(&I, Res);
DAG.setRoot(Res.getValue(1));
return 0;
}
case Intrinsic::bswap:
setValue(&I, DAG.getNode(ISD::BSWAP, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::cttz: {
SDValue Arg = getValue(I.getOperand(1));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTTZ, dl, Ty, Arg));
return 0;
}
case Intrinsic::ctlz: {
SDValue Arg = getValue(I.getOperand(1));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTLZ, dl, Ty, Arg));
return 0;
}
case Intrinsic::ctpop: {
SDValue Arg = getValue(I.getOperand(1));
EVT Ty = Arg.getValueType();
setValue(&I, DAG.getNode(ISD::CTPOP, dl, Ty, Arg));
return 0;
}
case Intrinsic::stacksave: {
SDValue Op = getRoot();
Res = DAG.getNode(ISD::STACKSAVE, dl,
DAG.getVTList(TLI.getPointerTy(), MVT::Other), &Op, 1);
setValue(&I, Res);
DAG.setRoot(Res.getValue(1));
return 0;
}
case Intrinsic::stackrestore: {
Res = getValue(I.getOperand(1));
DAG.setRoot(DAG.getNode(ISD::STACKRESTORE, dl, MVT::Other, getRoot(), Res));
return 0;
}
case Intrinsic::stackprotector: {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
EVT PtrTy = TLI.getPointerTy();
SDValue Src = getValue(I.getOperand(1)); AllocaInst *Slot = cast<AllocaInst>(I.getOperand(2));
int FI = FuncInfo.StaticAllocaMap[Slot];
MFI->setStackProtectorIndex(FI);
SDValue FIN = DAG.getFrameIndex(FI, PtrTy);
Res = DAG.getStore(getRoot(), getCurDebugLoc(), Src, FIN,
PseudoSourceValue::getFixedStack(FI),
0, true, false, 0);
setValue(&I, Res);
DAG.setRoot(Res);
return 0;
}
case Intrinsic::objectsize: {
ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(2));
assert(CI && "Non-constant type in __builtin_object_size?");
SDValue Arg = getValue(I.getOperand(0));
EVT Ty = Arg.getValueType();
if (CI->getZExtValue() == 0)
Res = DAG.getConstant(-1ULL, Ty);
else
Res = DAG.getConstant(0, Ty);
setValue(&I, Res);
return 0;
}
case Intrinsic::var_annotation:
return 0;
case Intrinsic::init_trampoline: {
const Function *F = cast<Function>(I.getOperand(2)->stripPointerCasts());
SDValue Ops[6];
Ops[0] = getRoot();
Ops[1] = getValue(I.getOperand(1));
Ops[2] = getValue(I.getOperand(2));
Ops[3] = getValue(I.getOperand(3));
Ops[4] = DAG.getSrcValue(I.getOperand(1));
Ops[5] = DAG.getSrcValue(F);
Res = DAG.getNode(ISD::TRAMPOLINE, dl,
DAG.getVTList(TLI.getPointerTy(), MVT::Other),
Ops, 6);
setValue(&I, Res);
DAG.setRoot(Res.getValue(1));
return 0;
}
case Intrinsic::gcroot:
if (GFI) {
Value *Alloca = I.getOperand(1);
Constant *TypeMap = cast<Constant>(I.getOperand(2));
FrameIndexSDNode *FI = cast<FrameIndexSDNode>(getValue(Alloca).getNode());
GFI->addStackRoot(FI->getIndex(), TypeMap);
}
return 0;
case Intrinsic::gcread:
case Intrinsic::gcwrite:
llvm_unreachable("GC failed to lower gcread/gcwrite intrinsics!");
return 0;
case Intrinsic::flt_rounds:
setValue(&I, DAG.getNode(ISD::FLT_ROUNDS_, dl, MVT::i32));
return 0;
case Intrinsic::trap:
DAG.setRoot(DAG.getNode(ISD::TRAP, dl,MVT::Other, getRoot()));
return 0;
case Intrinsic::uadd_with_overflow:
return implVisitAluOverflow(I, ISD::UADDO);
case Intrinsic::sadd_with_overflow:
return implVisitAluOverflow(I, ISD::SADDO);
case Intrinsic::usub_with_overflow:
return implVisitAluOverflow(I, ISD::USUBO);
case Intrinsic::ssub_with_overflow:
return implVisitAluOverflow(I, ISD::SSUBO);
case Intrinsic::umul_with_overflow:
return implVisitAluOverflow(I, ISD::UMULO);
case Intrinsic::smul_with_overflow:
return implVisitAluOverflow(I, ISD::SMULO);
case Intrinsic::prefetch: {
SDValue Ops[4];
Ops[0] = getRoot();
Ops[1] = getValue(I.getOperand(1));
Ops[2] = getValue(I.getOperand(2));
Ops[3] = getValue(I.getOperand(3));
DAG.setRoot(DAG.getNode(ISD::PREFETCH, dl, MVT::Other, &Ops[0], 4));
return 0;
}
case Intrinsic::memory_barrier: {
SDValue Ops[6];
Ops[0] = getRoot();
for (int x = 1; x < 6; ++x)
Ops[x] = getValue(I.getOperand(x));
DAG.setRoot(DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, &Ops[0], 6));
return 0;
}
case Intrinsic::atomic_cmp_swap: {
SDValue Root = getRoot();
SDValue L =
DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, getCurDebugLoc(),
getValue(I.getOperand(2)).getValueType().getSimpleVT(),
Root,
getValue(I.getOperand(1)),
getValue(I.getOperand(2)),
getValue(I.getOperand(3)),
I.getOperand(1));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
}
case Intrinsic::atomic_load_add:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD);
case Intrinsic::atomic_load_sub:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB);
case Intrinsic::atomic_load_or:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR);
case Intrinsic::atomic_load_xor:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR);
case Intrinsic::atomic_load_and:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND);
case Intrinsic::atomic_load_nand:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND);
case Intrinsic::atomic_load_max:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX);
case Intrinsic::atomic_load_min:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN);
case Intrinsic::atomic_load_umin:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN);
case Intrinsic::atomic_load_umax:
return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX);
case Intrinsic::atomic_swap:
return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP);
case Intrinsic::invariant_start:
case Intrinsic::lifetime_start:
setValue(&I, DAG.getUNDEF(TLI.getPointerTy()));
return 0;
case Intrinsic::invariant_end:
case Intrinsic::lifetime_end:
return 0;
}
}
static bool
isInTailCallPosition(CallSite CS, Attributes CalleeRetAttr,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
const Function *F = ExitBB->getParent();
if (!Ret &&
(!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!I->isSafeToSpeculativelyExecute())
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!BBI->isSafeToSpeculativelyExecute())
return false;
}
if (!Ret || Ret->getNumOperands() == 0) return true;
if (isa<UndefValue>(Ret->getOperand(0))) return true;
unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
return false;
if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
return false;
for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
U = dyn_cast<Instruction>(U->getOperand(0))) {
if (!U)
return false;
if (!U->hasOneUse())
return false;
if (U == I)
break;
if (isa<TruncInst>(U) &&
TLI.isTruncateFree(U->getOperand(0)->getType(), U->getType()))
continue;
if (isa<BitCastInst>(U) &&
(U->getOperand(0)->getType() == U->getType() ||
(U->getOperand(0)->getType()->isPointerTy() &&
U->getType()->isPointerTy())))
continue;
return false;
}
return true;
}
void SelectionDAGBuilder::LowerCallTo(CallSite CS, SDValue Callee,
bool isTailCall,
MachineBasicBlock *LandingPad) {
const PointerType *PT = cast<PointerType>(CS.getCalledValue()->getType());
const FunctionType *FTy = cast<FunctionType>(PT->getElementType());
const Type *RetTy = FTy->getReturnType();
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
MCSymbol *BeginLabel = 0;
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Args.reserve(CS.arg_size());
SmallVector<EVT, 4> OutVTs;
SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
SmallVector<uint64_t, 4> Offsets;
getReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
OutVTs, OutsFlags, TLI, &Offsets);
bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
FTy->isVarArg(), OutVTs, OutsFlags, DAG);
SDValue DemoteStackSlot;
if (!CanLowerReturn) {
uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(
FTy->getReturnType());
unsigned Align = TLI.getTargetData()->getPrefTypeAlignment(
FTy->getReturnType());
MachineFunction &MF = DAG.getMachineFunction();
int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align, false);
const Type *StackSlotPtrType = PointerType::getUnqual(FTy->getReturnType());
DemoteStackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy());
Entry.Node = DemoteStackSlot;
Entry.Ty = StackSlotPtrType;
Entry.isSExt = false;
Entry.isZExt = false;
Entry.isInReg = false;
Entry.isSRet = true;
Entry.isNest = false;
Entry.isByVal = false;
Entry.Alignment = Align;
Args.push_back(Entry);
RetTy = Type::getVoidTy(FTy->getContext());
}
for (CallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end();
i != e; ++i) {
SDValue ArgNode = getValue(*i);
Entry.Node = ArgNode; Entry.Ty = (*i)->getType();
unsigned attrInd = i - CS.arg_begin() + 1;
Entry.isSExt = CS.paramHasAttr(attrInd, Attribute::SExt);
Entry.isZExt = CS.paramHasAttr(attrInd, Attribute::ZExt);
Entry.isInReg = CS.paramHasAttr(attrInd, Attribute::InReg);
Entry.isSRet = CS.paramHasAttr(attrInd, Attribute::StructRet);
Entry.isNest = CS.paramHasAttr(attrInd, Attribute::Nest);
Entry.isByVal = CS.paramHasAttr(attrInd, Attribute::ByVal);
Entry.Alignment = CS.getParamAlignment(attrInd);
Args.push_back(Entry);
}
if (LandingPad && MMI) {
BeginLabel = MMI->getContext().CreateTempSymbol();
unsigned CallSiteIndex = MMI->getCurrentCallSite();
if (CallSiteIndex) {
MMI->setCallSiteBeginLabel(BeginLabel, CallSiteIndex);
MMI->setCurrentCallSite(0);
}
(void)getRoot();
DAG.setRoot(DAG.getEHLabel(getCurDebugLoc(), getControlRoot(), BeginLabel));
}
if (isTailCall &&
!isInTailCallPosition(CS, CS.getAttributes().getRetAttributes(), TLI))
isTailCall = false;
std::pair<SDValue,SDValue> Result =
TLI.LowerCallTo(getRoot(), RetTy,
CS.paramHasAttr(0, Attribute::SExt),
CS.paramHasAttr(0, Attribute::ZExt), FTy->isVarArg(),
CS.paramHasAttr(0, Attribute::InReg), FTy->getNumParams(),
CS.getCallingConv(),
isTailCall,
!CS.getInstruction()->use_empty(),
Callee, Args, DAG, getCurDebugLoc());
assert((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);
} else if (!CanLowerReturn && Result.second.getNode()) {
SmallVector<EVT, 1> PVTs;
const Type *PtrRetTy = PointerType::getUnqual(FTy->getReturnType());
ComputeValueVTs(TLI, PtrRetTy, PVTs);
assert(PVTs.size() == 1 && "Pointers should fit in one register");
EVT PtrVT = PVTs[0];
unsigned NumValues = OutVTs.size();
SmallVector<SDValue, 4> Values(NumValues);
SmallVector<SDValue, 4> Chains(NumValues);
for (unsigned i = 0; i < NumValues; ++i) {
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT,
DemoteStackSlot,
DAG.getConstant(Offsets[i], PtrVT));
SDValue L = DAG.getLoad(OutVTs[i], getCurDebugLoc(), Result.second,
Add, NULL, Offsets[i], false, false, 1);
Values[i] = L;
Chains[i] = L.getValue(1);
}
SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
MVT::Other, &Chains[0], NumValues);
PendingLoads.push_back(Chain);
SmallVector<EVT, 4> RetTys;
RetTy = FTy->getReturnType();
ComputeValueVTs(TLI, RetTy, RetTys);
ISD::NodeType AssertOp = ISD::DELETED_NODE;
SmallVector<SDValue, 4> ReturnValues;
unsigned CurReg = 0;
for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
EVT VT = RetTys[I];
EVT RegisterVT = TLI.getRegisterType(RetTy->getContext(), VT);
unsigned NumRegs = TLI.getNumRegisters(RetTy->getContext(), VT);
SDValue ReturnValue =
getCopyFromParts(DAG, getCurDebugLoc(), &Values[CurReg], NumRegs,
RegisterVT, VT, AssertOp);
ReturnValues.push_back(ReturnValue);
CurReg += NumRegs;
}
setValue(CS.getInstruction(),
DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&RetTys[0], RetTys.size()),
&ReturnValues[0], ReturnValues.size()));
}
if (Result.second.getNode())
DAG.setRoot(Result.second);
else
HasTailCall = true;
if (LandingPad && MMI) {
MCSymbol *EndLabel = MMI->getContext().CreateTempSymbol();
DAG.setRoot(DAG.getEHLabel(getCurDebugLoc(), getRoot(), EndLabel));
MMI->addInvoke(LandingPad, BeginLabel, EndLabel);
}
}
static bool IsOnlyUsedInZeroEqualityComparison(Value *V) {
for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
UI != E; ++UI) {
if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
if (IC->isEquality())
if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
if (C->isNullValue())
continue;
return false;
}
return true;
}
static SDValue getMemCmpLoad(Value *PtrVal, MVT LoadVT, const Type *LoadTy,
SelectionDAGBuilder &Builder) {
if (Constant *LoadInput = dyn_cast<Constant>(PtrVal)) {
LoadInput = ConstantExpr::getBitCast(LoadInput,
PointerType::getUnqual(LoadTy));
if (Constant *LoadCst = ConstantFoldLoadFromConstPtr(LoadInput, Builder.TD))
return Builder.getValue(LoadCst);
}
SDValue Root;
bool ConstantMemory = false;
if (Builder.AA->pointsToConstantMemory(PtrVal)) {
Root = Builder.DAG.getEntryNode();
ConstantMemory = true;
} else {
Root = Builder.DAG.getRoot();
}
SDValue Ptr = Builder.getValue(PtrVal);
SDValue LoadVal = Builder.DAG.getLoad(LoadVT, Builder.getCurDebugLoc(), Root,
Ptr, PtrVal , 0,
false ,
false , 1 );
if (!ConstantMemory)
Builder.PendingLoads.push_back(LoadVal.getValue(1));
return LoadVal;
}
bool SelectionDAGBuilder::visitMemCmpCall(CallInst &I) {
if (I.getNumOperands() != 4)
return false;
Value *LHS = I.getOperand(1), *RHS = I.getOperand(2);
if (!LHS->getType()->isPointerTy() || !RHS->getType()->isPointerTy() ||
!I.getOperand(3)->getType()->isIntegerTy() ||
!I.getType()->isIntegerTy())
return false;
ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3));
if (Size && IsOnlyUsedInZeroEqualityComparison(&I)) {
bool ActuallyDoIt = true;
MVT LoadVT;
const Type *LoadTy;
switch (Size->getZExtValue()) {
default:
LoadVT = MVT::Other;
LoadTy = 0;
ActuallyDoIt = false;
break;
case 2:
LoadVT = MVT::i16;
LoadTy = Type::getInt16Ty(Size->getContext());
break;
case 4:
LoadVT = MVT::i32;
LoadTy = Type::getInt32Ty(Size->getContext());
break;
case 8:
LoadVT = MVT::i64;
LoadTy = Type::getInt64Ty(Size->getContext());
break;
}
if (ActuallyDoIt && Size->getZExtValue() > 4) {
if (!TLI.isTypeLegal(LoadVT) ||!TLI.allowsUnalignedMemoryAccesses(LoadVT))
ActuallyDoIt = false;
}
if (ActuallyDoIt) {
SDValue LHSVal = getMemCmpLoad(LHS, LoadVT, LoadTy, *this);
SDValue RHSVal = getMemCmpLoad(RHS, LoadVT, LoadTy, *this);
SDValue Res = DAG.getSetCC(getCurDebugLoc(), MVT::i1, LHSVal, RHSVal,
ISD::SETNE);
EVT CallVT = TLI.getValueType(I.getType(), true);
setValue(&I, DAG.getZExtOrTrunc(Res, getCurDebugLoc(), CallVT));
return true;
}
}
return false;
}
void SelectionDAGBuilder::visitCall(CallInst &I) {
const char *RenameFn = 0;
if (Function *F = I.getCalledFunction()) {
if (F->isDeclaration()) {
const TargetIntrinsicInfo *II = TLI.getTargetMachine().getIntrinsicInfo();
if (II) {
if (unsigned IID = II->getIntrinsicID(F)) {
RenameFn = visitIntrinsicCall(I, IID);
if (!RenameFn)
return;
}
}
if (unsigned IID = F->getIntrinsicID()) {
RenameFn = visitIntrinsicCall(I, IID);
if (!RenameFn)
return;
}
}
if (!F->hasLocalLinkage() && F->hasName()) {
StringRef Name = F->getName();
if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
if (I.getNumOperands() == 3 && I.getOperand(1)->getType()->isFloatingPointTy() &&
I.getType() == I.getOperand(1)->getType() &&
I.getType() == I.getOperand(2)->getType()) {
SDValue LHS = getValue(I.getOperand(1));
SDValue RHS = getValue(I.getOperand(2));
setValue(&I, DAG.getNode(ISD::FCOPYSIGN, getCurDebugLoc(),
LHS.getValueType(), LHS, RHS));
return;
}
} else if (Name == "fabs" || Name == "fabsf" || Name == "fabsl") {
if (I.getNumOperands() == 2 && I.getOperand(1)->getType()->isFloatingPointTy() &&
I.getType() == I.getOperand(1)->getType()) {
SDValue Tmp = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FABS, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
return;
}
} else if (Name == "sin" || Name == "sinf" || Name == "sinl") {
if (I.getNumOperands() == 2 && I.getOperand(1)->getType()->isFloatingPointTy() &&
I.getType() == I.getOperand(1)->getType() &&
I.onlyReadsMemory()) {
SDValue Tmp = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FSIN, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
return;
}
} else if (Name == "cos" || Name == "cosf" || Name == "cosl") {
if (I.getNumOperands() == 2 && I.getOperand(1)->getType()->isFloatingPointTy() &&
I.getType() == I.getOperand(1)->getType() &&
I.onlyReadsMemory()) {
SDValue Tmp = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FCOS, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
return;
}
} else if (Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl") {
if (I.getNumOperands() == 2 && I.getOperand(1)->getType()->isFloatingPointTy() &&
I.getType() == I.getOperand(1)->getType() &&
I.onlyReadsMemory()) {
SDValue Tmp = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(ISD::FSQRT, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
return;
}
} else if (Name == "memcmp") {
if (visitMemCmpCall(I))
return;
}
}
} else if (isa<InlineAsm>(I.getOperand(0))) {
visitInlineAsm(&I);
return;
}
SDValue Callee;
if (!RenameFn)
Callee = getValue(I.getOperand(0));
else
Callee = DAG.getExternalSymbol(RenameFn, TLI.getPointerTy());
LowerCallTo(&I, Callee, I.isTailCall());
}
SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const {
SmallVector<SDValue, 4> Values(ValueVTs.size());
SmallVector<SDValue, 8> Parts;
for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
EVT ValueVT = ValueVTs[Value];
unsigned NumRegs = TLI->getNumRegisters(*DAG.getContext(), ValueVT);
EVT RegisterVT = RegVTs[Value];
Parts.resize(NumRegs);
for (unsigned i = 0; i != NumRegs; ++i) {
SDValue P;
if (Flag == 0) {
P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT);
} else {
P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT, *Flag);
*Flag = P.getValue(2);
}
Chain = P.getValue(1);
if (TargetRegisterInfo::isVirtualRegister(Regs[Part+i]) &&
RegisterVT.isInteger() && !RegisterVT.isVector()) {
unsigned SlotNo = Regs[Part+i]-TargetRegisterInfo::FirstVirtualRegister;
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
if (FLI.LiveOutRegInfo.size() > SlotNo) {
FunctionLoweringInfo::LiveOutInfo &LOI = FLI.LiveOutRegInfo[SlotNo];
unsigned RegSize = RegisterVT.getSizeInBits();
unsigned NumSignBits = LOI.NumSignBits;
unsigned NumZeroBits = LOI.KnownZero.countLeadingOnes();
bool isSExt = true;
EVT FromVT(MVT::Other);
if (NumSignBits == RegSize)
isSExt = true, FromVT = MVT::i1; else if (NumZeroBits >= RegSize-1)
isSExt = false, FromVT = MVT::i1; else if (NumSignBits > RegSize-8)
isSExt = true, FromVT = MVT::i8; else if (NumZeroBits >= RegSize-8)
isSExt = false, FromVT = MVT::i8; else if (NumSignBits > RegSize-16)
isSExt = true, FromVT = MVT::i16; else if (NumZeroBits >= RegSize-16)
isSExt = false, FromVT = MVT::i16; else if (NumSignBits > RegSize-32)
isSExt = true, FromVT = MVT::i32; else if (NumZeroBits >= RegSize-32)
isSExt = false, FromVT = MVT::i32;
if (FromVT != MVT::Other)
P = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext, dl,
RegisterVT, P, DAG.getValueType(FromVT));
}
}
Parts[i] = P;
}
Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(),
NumRegs, RegisterVT, ValueVT);
Part += NumRegs;
Parts.clear();
}
return DAG.getNode(ISD::MERGE_VALUES, dl,
DAG.getVTList(&ValueVTs[0], ValueVTs.size()),
&Values[0], ValueVTs.size());
}
void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const {
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);
EVT RegisterVT = RegVTs[Value];
getCopyToParts(DAG, dl,
Val.getValue(Val.getResNo() + Value),
&Parts[Part], NumParts, RegisterVT);
Part += NumParts;
}
SmallVector<SDValue, 8> Chains(NumRegs);
for (unsigned i = 0; i != NumRegs; ++i) {
SDValue Part;
if (Flag == 0) {
Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i]);
} else {
Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i], *Flag);
*Flag = Part.getValue(1);
}
Chains[i] = Part.getValue(0);
}
if (NumRegs == 1 || Flag)
Chain = Chains[NumRegs-1];
else
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0], NumRegs);
}
void RegsForValue::AddInlineAsmOperands(unsigned Code,
bool HasMatching,unsigned MatchingIdx,
SelectionDAG &DAG,
std::vector<SDValue> &Ops) const {
assert(Regs.size() < (1 << 13) && "Too many inline asm outputs!");
unsigned Flag = Code | (Regs.size() << 3);
if (HasMatching)
Flag |= 0x80000000 | (MatchingIdx << 16);
SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
Ops.push_back(Res);
for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
unsigned NumRegs = TLI->getNumRegisters(*DAG.getContext(), ValueVTs[Value]);
EVT RegisterVT = RegVTs[Value];
for (unsigned i = 0; i != NumRegs; ++i) {
assert(Reg < Regs.size() && "Mismatch in # registers expected");
Ops.push_back(DAG.getRegister(Regs[Reg++], RegisterVT));
}
}
}
static const TargetRegisterClass *
isAllocatableRegister(unsigned Reg, MachineFunction &MF,
const TargetLowering &TLI,
const TargetRegisterInfo *TRI) {
EVT FoundVT = MVT::Other;
const TargetRegisterClass *FoundRC = 0;
for (TargetRegisterInfo::regclass_iterator RCI = TRI->regclass_begin(),
E = TRI->regclass_end(); RCI != E; ++RCI) {
EVT ThisVT = MVT::Other;
const TargetRegisterClass *RC = *RCI;
for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
I != E; ++I) {
if (TLI.isTypeLegal(*I)) {
if (FoundVT == MVT::Other || FoundVT.bitsLT(*I)) {
ThisVT = *I;
break;
}
}
}
if (ThisVT == MVT::Other) continue;
for (TargetRegisterClass::iterator I = RC->allocation_order_begin(MF),
E = RC->allocation_order_end(MF); I != E; ++I)
if (*I == Reg) {
FoundRC = RC;
FoundVT = ThisVT;
break;
}
}
return FoundRC;
}
namespace llvm {
class VISIBILITY_HIDDEN SDISelAsmOperandInfo :
public TargetLowering::AsmOperandInfo {
public:
SDValue CallOperand;
RegsForValue AssignedRegs;
explicit SDISelAsmOperandInfo(const InlineAsm::ConstraintInfo &info)
: TargetLowering::AsmOperandInfo(info), CallOperand(0,0) {
}
void MarkAllocatedRegs(bool isOutReg, bool isInReg,
std::set<unsigned> &OutputRegs,
std::set<unsigned> &InputRegs,
const TargetRegisterInfo &TRI) const {
if (isOutReg) {
for (unsigned i = 0, e = AssignedRegs.Regs.size(); i != e; ++i)
MarkRegAndAliases(AssignedRegs.Regs[i], OutputRegs, TRI);
}
if (isInReg) {
for (unsigned i = 0, e = AssignedRegs.Regs.size(); i != e; ++i)
MarkRegAndAliases(AssignedRegs.Regs[i], InputRegs, TRI);
}
}
EVT getCallOperandValEVT(LLVMContext &Context,
const TargetLowering &TLI,
const TargetData *TD) const {
if (CallOperandVal == 0) return MVT::Other;
if (isa<BasicBlock>(CallOperandVal))
return TLI.getPointerTy();
const llvm::Type *OpTy = CallOperandVal->getType();
if (isIndirect) {
const llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
if (!PtrTy)
llvm_report_error("Indirect operand for inline asm not a pointer!");
OpTy = PtrTy->getElementType();
}
if (!OpTy->isSingleValueType() && OpTy->isSized()) {
unsigned BitSize = TD->getTypeSizeInBits(OpTy);
switch (BitSize) {
default: break;
case 1:
case 8:
case 16:
case 32:
case 64:
case 128:
OpTy = IntegerType::get(Context, BitSize);
break;
}
}
return TLI.getValueType(OpTy, true);
}
private:
static void MarkRegAndAliases(unsigned Reg, std::set<unsigned> &Regs,
const TargetRegisterInfo &TRI) {
assert(TargetRegisterInfo::isPhysicalRegister(Reg) && "Isn't a physreg");
Regs.insert(Reg);
if (const unsigned *Aliases = TRI.getAliasSet(Reg))
for (; *Aliases; ++Aliases)
Regs.insert(*Aliases);
}
};
}
void SelectionDAGBuilder::
GetRegistersForValue(SDISelAsmOperandInfo &OpInfo,
std::set<unsigned> &OutputRegs,
std::set<unsigned> &InputRegs) {
LLVMContext &Context = FuncInfo.Fn->getContext();
bool isOutReg = false;
bool isInReg = false;
switch (OpInfo.Type) {
case InlineAsm::isOutput:
isOutReg = true;
isInReg = OpInfo.hasMatchingInput();
break;
case InlineAsm::isInput:
isInReg = true;
isOutReg = false;
break;
case InlineAsm::isClobber:
isOutReg = true;
isInReg = true;
break;
}
MachineFunction &MF = DAG.getMachineFunction();
SmallVector<unsigned, 4> Regs;
std::pair<unsigned, const TargetRegisterClass*> PhysReg =
TLI.getRegForInlineAsmConstraint(OpInfo.ConstraintCode,
OpInfo.ConstraintVT);
unsigned NumRegs = 1;
if (OpInfo.ConstraintVT != MVT::Other) {
if (OpInfo.Type == InlineAsm::isInput &&
PhysReg.second && !PhysReg.second->hasType(OpInfo.ConstraintVT)) {
EVT RegVT = *PhysReg.second->vt_begin();
if (RegVT.getSizeInBits() == OpInfo.ConstraintVT.getSizeInBits()) {
OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
RegVT, OpInfo.CallOperand);
OpInfo.ConstraintVT = RegVT;
} else if (RegVT.isInteger() && OpInfo.ConstraintVT.isFloatingPoint()) {
RegVT = EVT::getIntegerVT(Context,
OpInfo.ConstraintVT.getSizeInBits());
OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
RegVT, OpInfo.CallOperand);
OpInfo.ConstraintVT = RegVT;
}
}
NumRegs = TLI.getNumRegisters(Context, OpInfo.ConstraintVT);
}
EVT RegVT;
EVT ValueVT = OpInfo.ConstraintVT;
if (unsigned AssignedReg = PhysReg.first) {
const TargetRegisterClass *RC = PhysReg.second;
if (OpInfo.ConstraintVT == MVT::Other)
ValueVT = *RC->vt_begin();
RegVT = *RC->vt_begin();
Regs.push_back(AssignedReg);
if (NumRegs != 1) {
TargetRegisterClass::iterator I = RC->begin();
for (; *I != AssignedReg; ++I)
assert(I != RC->end() && "Didn't find reg!");
--NumRegs; ++I;
for (; NumRegs; --NumRegs, ++I) {
assert(I != RC->end() && "Ran out of registers to allocate!");
Regs.push_back(*I);
}
}
OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
}
if (const TargetRegisterClass *RC = PhysReg.second) {
RegVT = *RC->vt_begin();
if (OpInfo.ConstraintVT == MVT::Other)
ValueVT = RegVT;
MachineRegisterInfo &RegInfo = MF.getRegInfo();
for (; NumRegs; --NumRegs)
Regs.push_back(RegInfo.createVirtualRegister(RC));
OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
return;
}
std::vector<unsigned> RegClassRegs
= TLI.getRegClassForInlineAsmConstraint(OpInfo.ConstraintCode,
OpInfo.ConstraintVT);
const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
unsigned NumAllocated = 0;
for (unsigned i = 0, e = RegClassRegs.size(); i != e; ++i) {
unsigned Reg = RegClassRegs[i];
if ((isOutReg && OutputRegs.count(Reg)) || (isInReg && InputRegs.count(Reg))) { NumAllocated = 0;
continue;
}
const TargetRegisterClass *RC = isAllocatableRegister(Reg, MF, TLI, TRI);
if (!RC) { NumAllocated = 0;
continue;
}
++NumAllocated;
if (NumAllocated == NumRegs) {
unsigned RegStart = (i-NumAllocated)+1;
unsigned RegEnd = i+1;
for (unsigned i = RegStart; i != RegEnd; ++i)
Regs.push_back(RegClassRegs[i]);
OpInfo.AssignedRegs = RegsForValue(TLI, Regs, *RC->vt_begin(),
OpInfo.ConstraintVT);
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
}
}
}
static bool
hasInlineAsmMemConstraint(std::vector<InlineAsm::ConstraintInfo> &CInfos,
const TargetLowering &TLI) {
for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
InlineAsm::ConstraintInfo &CI = CInfos[i];
for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
TargetLowering::ConstraintType CType = TLI.getConstraintType(CI.Codes[j]);
if (CType == TargetLowering::C_Memory)
return true;
}
if (CI.isIndirect)
return true;
}
return false;
}
void SelectionDAGBuilder::visitInlineAsm(CallSite CS) {
InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
std::vector<SDISelAsmOperandInfo> ConstraintOperands;
std::set<unsigned> OutputRegs, InputRegs;
std::vector<InlineAsm::ConstraintInfo>
ConstraintInfos = IA->ParseConstraints();
bool hasMemory = hasInlineAsmMemConstraint(ConstraintInfos, TLI);
SDValue Chain, Flag;
if (hasMemory || IA->hasSideEffects())
Chain = getRoot();
else
Chain = DAG.getRoot();
unsigned ArgNo = 0; unsigned ResNo = 0; for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
ConstraintOperands.push_back(SDISelAsmOperandInfo(ConstraintInfos[i]));
SDISelAsmOperandInfo &OpInfo = ConstraintOperands.back();
EVT OpVT = MVT::Other;
switch (OpInfo.Type) {
case InlineAsm::isOutput:
if (OpInfo.isIndirect) {
OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
break;
}
assert(!CS.getType()->isVoidTy() &&
"Bad inline asm!");
if (const StructType *STy = dyn_cast<StructType>(CS.getType())) {
OpVT = TLI.getValueType(STy->getElementType(ResNo));
} else {
assert(ResNo == 0 && "Asm only has one result!");
OpVT = TLI.getValueType(CS.getType());
}
++ResNo;
break;
case InlineAsm::isInput:
OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
break;
case InlineAsm::isClobber:
break;
}
if (OpInfo.CallOperandVal) {
OpInfo.CallOperandVal = OpInfo.CallOperandVal->stripPointerCasts();
if (BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal)) {
OpInfo.CallOperand = DAG.getBasicBlock(FuncInfo.MBBMap[BB]);
} else {
OpInfo.CallOperand = getValue(OpInfo.CallOperandVal);
}
OpVT = OpInfo.getCallOperandValEVT(*DAG.getContext(), TLI, TD);
}
OpInfo.ConstraintVT = OpVT;
}
for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
if (OpInfo.hasMatchingInput()) {
SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
if ((OpInfo.ConstraintVT.isInteger() !=
Input.ConstraintVT.isInteger()) ||
(OpInfo.ConstraintVT.getSizeInBits() !=
Input.ConstraintVT.getSizeInBits())) {
llvm_report_error("Unsupported asm: input constraint"
" with a matching output constraint of incompatible"
" type!");
}
Input.ConstraintVT = OpInfo.ConstraintVT;
}
}
TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, hasMemory, &DAG);
if (OpInfo.ConstraintType == TargetLowering::C_Memory &&
!OpInfo.isIndirect) {
assert(OpInfo.Type == InlineAsm::isInput &&
"Can only indirectify direct input operands!");
Value *OpVal = OpInfo.CallOperandVal;
if (isa<ConstantFP>(OpVal) || isa<ConstantInt>(OpVal) ||
isa<ConstantVector>(OpVal)) {
OpInfo.CallOperand = DAG.getConstantPool(cast<Constant>(OpVal),
TLI.getPointerTy());
} else {
const Type *Ty = OpVal->getType();
uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(Ty);
unsigned Align = TLI.getTargetData()->getPrefTypeAlignment(Ty);
MachineFunction &MF = DAG.getMachineFunction();
int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align, false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy());
Chain = DAG.getStore(Chain, getCurDebugLoc(),
OpInfo.CallOperand, StackSlot, NULL, 0,
false, false, 0);
OpInfo.CallOperand = StackSlot;
}
OpInfo.CallOperandVal = 0;
OpInfo.isIndirect = true;
}
if (OpInfo.ConstraintType == TargetLowering::C_Register)
GetRegistersForValue(OpInfo, OutputRegs, InputRegs);
}
ConstraintInfos.clear();
for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
if (OpInfo.ConstraintType == TargetLowering::C_RegisterClass)
GetRegistersForValue(OpInfo, OutputRegs, InputRegs);
}
std::vector<SDValue> AsmNodeOperands;
AsmNodeOperands.push_back(SDValue()); AsmNodeOperands.push_back(
DAG.getTargetExternalSymbol(IA->getAsmString().c_str(),
TLI.getPointerTy()));
AsmNodeOperands.push_back(DAG.getTargetConstant(IA->isAlignStack() ? 1 : 0,
MVT::i1));
RegsForValue RetValRegs;
std::vector<std::pair<RegsForValue, Value*> > IndirectStoresToEmit;
for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
switch (OpInfo.Type) {
case InlineAsm::isOutput: {
if (OpInfo.ConstraintType != TargetLowering::C_RegisterClass &&
OpInfo.ConstraintType != TargetLowering::C_Register) {
assert(OpInfo.isIndirect && "Memory output must be indirect operand");
unsigned ResOpType = 4 | (1<<3);
AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
TLI.getPointerTy()));
AsmNodeOperands.push_back(OpInfo.CallOperand);
break;
}
if (OpInfo.AssignedRegs.Regs.empty()) {
llvm_report_error("Couldn't allocate output reg for"
" constraint '" + OpInfo.ConstraintCode + "'!");
}
if (OpInfo.isIndirect) {
IndirectStoresToEmit.push_back(std::make_pair(OpInfo.AssignedRegs,
OpInfo.CallOperandVal));
} else {
assert(!CS.getType()->isVoidTy() && "Bad inline asm!");
RetValRegs.append(OpInfo.AssignedRegs);
}
OpInfo.AssignedRegs.AddInlineAsmOperands(OpInfo.isEarlyClobber ?
6 :
2 ,
false,
0,
DAG,
AsmNodeOperands);
break;
}
case InlineAsm::isInput: {
SDValue InOperandVal = OpInfo.CallOperand;
if (OpInfo.isMatchingInputConstraint()) { unsigned OperandNo = OpInfo.getMatchedOperand();
unsigned CurOp = 3; for (; OperandNo; --OperandNo) {
unsigned OpFlag =
cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
assert(((OpFlag & 7) == 2 ||
(OpFlag & 7) == 6 ||
(OpFlag & 7) == 4 ) &&
"Skipped past definitions?");
CurOp += InlineAsm::getNumOperandRegisters(OpFlag)+1;
}
unsigned OpFlag =
cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
if ((OpFlag & 7) == 2
|| (OpFlag & 7) == 6 ) {
if (OpInfo.isIndirect) {
llvm_report_error("Don't know how to handle tied indirect "
"register inputs yet!");
}
RegsForValue MatchedRegs;
MatchedRegs.TLI = &TLI;
MatchedRegs.ValueVTs.push_back(InOperandVal.getValueType());
EVT RegVT = AsmNodeOperands[CurOp+1].getValueType();
MatchedRegs.RegVTs.push_back(RegVT);
MachineRegisterInfo &RegInfo = DAG.getMachineFunction().getRegInfo();
for (unsigned i = 0, e = InlineAsm::getNumOperandRegisters(OpFlag);
i != e; ++i)
MatchedRegs.Regs.push_back
(RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT)));
MatchedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
Chain, &Flag);
MatchedRegs.AddInlineAsmOperands(1 ,
true, OpInfo.getMatchedOperand(),
DAG, AsmNodeOperands);
break;
} else {
assert(((OpFlag & 7) == 4) && "Unknown matching constraint!");
assert((InlineAsm::getNumOperandRegisters(OpFlag)) == 1 &&
"Unexpected number of operands");
OpFlag |= 0x80000000 | (OpInfo.getMatchedOperand() << 16);
AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlag,
TLI.getPointerTy()));
AsmNodeOperands.push_back(AsmNodeOperands[CurOp+1]);
break;
}
}
if (OpInfo.ConstraintType == TargetLowering::C_Other) {
assert(!OpInfo.isIndirect &&
"Don't know how to handle indirect other inputs yet!");
std::vector<SDValue> Ops;
TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode[0],
hasMemory, Ops, DAG);
if (Ops.empty()) {
llvm_report_error("Invalid operand for inline asm"
" constraint '" + OpInfo.ConstraintCode + "'!");
}
unsigned ResOpType = 3 | (Ops.size() << 3);
AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
TLI.getPointerTy()));
AsmNodeOperands.insert(AsmNodeOperands.end(), Ops.begin(), Ops.end());
break;
} else if (OpInfo.ConstraintType == TargetLowering::C_Memory) {
assert(OpInfo.isIndirect && "Operand must be indirect to be a mem!");
assert(InOperandVal.getValueType() == TLI.getPointerTy() &&
"Memory operands expect pointer values");
unsigned ResOpType = 4 | (1<<3);
AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
TLI.getPointerTy()));
AsmNodeOperands.push_back(InOperandVal);
break;
}
assert((OpInfo.ConstraintType == TargetLowering::C_RegisterClass ||
OpInfo.ConstraintType == TargetLowering::C_Register) &&
"Unknown constraint type!");
assert(!OpInfo.isIndirect &&
"Don't know how to handle indirect register inputs yet!");
if (OpInfo.AssignedRegs.Regs.empty() ||
!OpInfo.AssignedRegs.areValueTypesLegal()) {
llvm_report_error("Couldn't allocate input reg for"
" constraint '"+ OpInfo.ConstraintCode +"'!");
}
OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
Chain, &Flag);
OpInfo.AssignedRegs.AddInlineAsmOperands(1, false, 0,
DAG, AsmNodeOperands);
break;
}
case InlineAsm::isClobber: {
if (!OpInfo.AssignedRegs.Regs.empty())
OpInfo.AssignedRegs.AddInlineAsmOperands(6 ,
false, 0, DAG,
AsmNodeOperands);
break;
}
}
}
AsmNodeOperands[0] = Chain;
if (Flag.getNode()) AsmNodeOperands.push_back(Flag);
Chain = DAG.getNode(ISD::INLINEASM, getCurDebugLoc(),
DAG.getVTList(MVT::Other, MVT::Flag),
&AsmNodeOperands[0], AsmNodeOperands.size());
Flag = Chain.getValue(1);
if (!RetValRegs.Regs.empty()) {
SDValue Val = RetValRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
Chain, &Flag);
if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) {
EVT ResultType = TLI.getValueType(CS.getType());
if (ResultType != Val.getValueType() && Val.getValueType().isVector()) {
Val = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
ResultType, Val);
} else if (ResultType != Val.getValueType() &&
ResultType.isInteger() && Val.getValueType().isInteger()) {
Val = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), ResultType, Val);
}
assert(ResultType == Val.getValueType() && "Asm result value mismatch!");
}
setValue(CS.getInstruction(), Val);
if (!IA->hasSideEffects() && !hasMemory && IndirectStoresToEmit.empty())
return;
}
std::vector<std::pair<SDValue, Value*> > StoresToEmit;
for (unsigned i = 0, e = IndirectStoresToEmit.size(); i != e; ++i) {
RegsForValue &OutRegs = IndirectStoresToEmit[i].first;
Value *Ptr = IndirectStoresToEmit[i].second;
SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
Chain, &Flag);
StoresToEmit.push_back(std::make_pair(OutVal, Ptr));
}
SmallVector<SDValue, 8> OutChains;
for (unsigned i = 0, e = StoresToEmit.size(); i != e; ++i) {
SDValue Val = DAG.getStore(Chain, getCurDebugLoc(),
StoresToEmit[i].first,
getValue(StoresToEmit[i].second),
StoresToEmit[i].second, 0,
false, false, 0);
OutChains.push_back(Val);
}
if (!OutChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&OutChains[0], OutChains.size());
DAG.setRoot(Chain);
}
void SelectionDAGBuilder::visitVAStart(CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VASTART, getCurDebugLoc(),
MVT::Other, getRoot(),
getValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(1))));
}
void SelectionDAGBuilder::visitVAArg(VAArgInst &I) {
SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getCurDebugLoc(),
getRoot(), getValue(I.getOperand(0)),
DAG.getSrcValue(I.getOperand(0)));
setValue(&I, V);
DAG.setRoot(V.getValue(1));
}
void SelectionDAGBuilder::visitVAEnd(CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VAEND, getCurDebugLoc(),
MVT::Other, getRoot(),
getValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(1))));
}
void SelectionDAGBuilder::visitVACopy(CallInst &I) {
DAG.setRoot(DAG.getNode(ISD::VACOPY, getCurDebugLoc(),
MVT::Other, getRoot(),
getValue(I.getOperand(1)),
getValue(I.getOperand(2)),
DAG.getSrcValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(2))));
}
std::pair<SDValue, SDValue>
TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy,
bool RetSExt, bool RetZExt, bool isVarArg,
bool isInreg, unsigned NumFixedArgs,
CallingConv::ID CallConv, bool isTailCall,
bool isReturnValueUsed,
SDValue Callee,
ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl) {
SmallVector<ISD::OutputArg, 32> Outs;
for (unsigned i = 0, e = Args.size(); i != e; ++i) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(*this, Args[i].Ty, ValueVTs);
for (unsigned Value = 0, NumValues = ValueVTs.size();
Value != NumValues; ++Value) {
EVT VT = ValueVTs[Value];
const Type *ArgTy = VT.getTypeForEVT(RetTy->getContext());
SDValue Op = SDValue(Args[i].Node.getNode(),
Args[i].Node.getResNo() + Value);
ISD::ArgFlagsTy Flags;
unsigned OriginalAlignment =
getTargetData()->getABITypeAlignment(ArgTy);
if (Args[i].isZExt)
Flags.setZExt();
if (Args[i].isSExt)
Flags.setSExt();
if (Args[i].isInReg)
Flags.setInReg();
if (Args[i].isSRet)
Flags.setSRet();
if (Args[i].isByVal) {
Flags.setByVal();
const PointerType *Ty = cast<PointerType>(Args[i].Ty);
const Type *ElementTy = Ty->getElementType();
unsigned FrameAlign = getByValTypeAlignment(ElementTy);
unsigned FrameSize = getTargetData()->getTypeAllocSize(ElementTy);
if (Args[i].Alignment)
FrameAlign = Args[i].Alignment;
Flags.setByValAlign(FrameAlign);
Flags.setByValSize(FrameSize);
}
if (Args[i].isNest)
Flags.setNest();
Flags.setOrigAlign(OriginalAlignment);
EVT PartVT = getRegisterType(RetTy->getContext(), VT);
unsigned NumParts = getNumRegisters(RetTy->getContext(), VT);
SmallVector<SDValue, 4> Parts(NumParts);
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
if (Args[i].isSExt)
ExtendKind = ISD::SIGN_EXTEND;
else if (Args[i].isZExt)
ExtendKind = ISD::ZERO_EXTEND;
getCopyToParts(DAG, dl, Op, &Parts[0], NumParts,
PartVT, ExtendKind);
for (unsigned j = 0; j != NumParts; ++j) {
ISD::OutputArg MyFlags(Flags, Parts[j], i < NumFixedArgs);
if (NumParts > 1 && j == 0)
MyFlags.Flags.setSplit();
else if (j != 0)
MyFlags.Flags.setOrigAlign(1);
Outs.push_back(MyFlags);
}
}
}
SmallVector<ISD::InputArg, 32> Ins;
SmallVector<EVT, 4> RetTys;
ComputeValueVTs(*this, RetTy, RetTys);
for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
EVT VT = RetTys[I];
EVT RegisterVT = getRegisterType(RetTy->getContext(), VT);
unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT);
for (unsigned i = 0; i != NumRegs; ++i) {
ISD::InputArg MyFlags;
MyFlags.VT = RegisterVT;
MyFlags.Used = isReturnValueUsed;
if (RetSExt)
MyFlags.Flags.setSExt();
if (RetZExt)
MyFlags.Flags.setZExt();
if (isInreg)
MyFlags.Flags.setInReg();
Ins.push_back(MyFlags);
}
}
SmallVector<SDValue, 4> InVals;
Chain = LowerCall(Chain, Callee, CallConv, isVarArg, isTailCall,
Outs, Ins, dl, DAG, InVals);
assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
"LowerCall didn't return a valid chain!");
assert((!isTailCall || InVals.empty()) &&
"LowerCall emitted a return value for a tail call!");
assert((isTailCall || InVals.size() == Ins.size()) &&
"LowerCall didn't emit the correct number of values!");
if (isTailCall) {
DAG.setRoot(Chain);
return std::make_pair(SDValue(), SDValue());
}
DEBUG(for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
assert(InVals[i].getNode() &&
"LowerCall emitted a null value!");
assert(Ins[i].VT == InVals[i].getValueType() &&
"LowerCall emitted a value with the wrong type!");
});
ISD::NodeType AssertOp = ISD::DELETED_NODE;
if (RetSExt)
AssertOp = ISD::AssertSext;
else if (RetZExt)
AssertOp = ISD::AssertZext;
SmallVector<SDValue, 4> ReturnValues;
unsigned CurReg = 0;
for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
EVT VT = RetTys[I];
EVT RegisterVT = getRegisterType(RetTy->getContext(), VT);
unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT);
ReturnValues.push_back(getCopyFromParts(DAG, dl, &InVals[CurReg],
NumRegs, RegisterVT, VT,
AssertOp));
CurReg += NumRegs;
}
if (ReturnValues.empty())
return std::make_pair(SDValue(), Chain);
SDValue Res = DAG.getNode(ISD::MERGE_VALUES, dl,
DAG.getVTList(&RetTys[0], RetTys.size()),
&ReturnValues[0], ReturnValues.size());
return std::make_pair(Res, Chain);
}
void TargetLowering::LowerOperationWrapper(SDNode *N,
SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) {
SDValue Res = LowerOperation(SDValue(N, 0), DAG);
if (Res.getNode())
Results.push_back(Res);
}
SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
llvm_unreachable("LowerOperation not implemented for this target!");
return SDValue();
}
void SelectionDAGBuilder::CopyValueToVirtualRegister(Value *V, unsigned Reg) {
SDValue Op = getValue(V);
assert((Op.getOpcode() != ISD::CopyFromReg ||
cast<RegisterSDNode>(Op.getOperand(1))->getReg() != Reg) &&
"Copy from a reg to the same reg!");
assert(!TargetRegisterInfo::isPhysicalRegister(Reg) && "Is a physreg");
RegsForValue RFV(V->getContext(), TLI, Reg, V->getType());
SDValue Chain = DAG.getEntryNode();
RFV.getCopyToRegs(Op, DAG, getCurDebugLoc(), Chain, 0);
PendingExports.push_back(Chain);
}
#include "llvm/CodeGen/SelectionDAGISel.h"
void SelectionDAGISel::LowerArguments(BasicBlock *LLVMBB) {
Function &F = *LLVMBB->getParent();
SelectionDAG &DAG = SDB->DAG;
SDValue OldRoot = DAG.getRoot();
DebugLoc dl = SDB->getCurDebugLoc();
const TargetData *TD = TLI.getTargetData();
SmallVector<ISD::InputArg, 16> Ins;
SmallVector<EVT, 4> OutVTs;
SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
getReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
OutVTs, OutsFlags, TLI);
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
FLI.CanLowerReturn = TLI.CanLowerReturn(F.getCallingConv(), F.isVarArg(),
OutVTs, OutsFlags, DAG);
if (!FLI.CanLowerReturn) {
SmallVector<EVT, 1> ValueVTs;
ComputeValueVTs(TLI, PointerType::getUnqual(F.getReturnType()), ValueVTs);
ISD::ArgFlagsTy Flags;
Flags.setSRet();
EVT RegisterVT = TLI.getRegisterType(*CurDAG->getContext(), ValueVTs[0]);
ISD::InputArg RetArg(Flags, RegisterVT, true);
Ins.push_back(RetArg);
}
unsigned Idx = 1;
for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end();
I != E; ++I, ++Idx) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I->getType(), ValueVTs);
bool isArgValueUsed = !I->use_empty();
for (unsigned Value = 0, NumValues = ValueVTs.size();
Value != NumValues; ++Value) {
EVT VT = ValueVTs[Value];
const Type *ArgTy = VT.getTypeForEVT(*DAG.getContext());
ISD::ArgFlagsTy Flags;
unsigned OriginalAlignment =
TD->getABITypeAlignment(ArgTy);
if (F.paramHasAttr(Idx, Attribute::ZExt))
Flags.setZExt();
if (F.paramHasAttr(Idx, Attribute::SExt))
Flags.setSExt();
if (F.paramHasAttr(Idx, Attribute::InReg))
Flags.setInReg();
if (F.paramHasAttr(Idx, Attribute::StructRet))
Flags.setSRet();
if (F.paramHasAttr(Idx, Attribute::ByVal)) {
Flags.setByVal();
const PointerType *Ty = cast<PointerType>(I->getType());
const Type *ElementTy = Ty->getElementType();
unsigned FrameAlign = TLI.getByValTypeAlignment(ElementTy);
unsigned FrameSize = TD->getTypeAllocSize(ElementTy);
if (F.getParamAlignment(Idx))
FrameAlign = F.getParamAlignment(Idx);
Flags.setByValAlign(FrameAlign);
Flags.setByValSize(FrameSize);
}
if (F.paramHasAttr(Idx, Attribute::Nest))
Flags.setNest();
Flags.setOrigAlign(OriginalAlignment);
EVT RegisterVT = TLI.getRegisterType(*CurDAG->getContext(), VT);
unsigned NumRegs = TLI.getNumRegisters(*CurDAG->getContext(), VT);
for (unsigned i = 0; i != NumRegs; ++i) {
ISD::InputArg MyFlags(Flags, RegisterVT, isArgValueUsed);
if (NumRegs > 1 && i == 0)
MyFlags.Flags.setSplit();
else if (i > 0)
MyFlags.Flags.setOrigAlign(1);
Ins.push_back(MyFlags);
}
}
}
SmallVector<SDValue, 8> InVals;
SDValue NewRoot = TLI.LowerFormalArguments(DAG.getRoot(), F.getCallingConv(),
F.isVarArg(), Ins,
dl, DAG, InVals);
assert(NewRoot.getNode() && NewRoot.getValueType() == MVT::Other &&
"LowerFormalArguments didn't return a valid chain!");
assert(InVals.size() == Ins.size() &&
"LowerFormalArguments didn't emit the correct number of values!");
DEBUG({
for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
assert(InVals[i].getNode() &&
"LowerFormalArguments emitted a null value!");
assert(Ins[i].VT == InVals[i].getValueType() &&
"LowerFormalArguments emitted a value with the wrong type!");
}
});
DAG.setRoot(NewRoot);
unsigned i = 0;
Idx = 1;
if (!FLI.CanLowerReturn) {
SmallVector<EVT, 1> ValueVTs;
ComputeValueVTs(TLI, PointerType::getUnqual(F.getReturnType()), ValueVTs);
EVT VT = ValueVTs[0];
EVT RegVT = TLI.getRegisterType(*CurDAG->getContext(), VT);
ISD::NodeType AssertOp = ISD::DELETED_NODE;
SDValue ArgValue = getCopyFromParts(DAG, dl, &InVals[0], 1,
RegVT, VT, AssertOp);
MachineFunction& MF = SDB->DAG.getMachineFunction();
MachineRegisterInfo& RegInfo = MF.getRegInfo();
unsigned SRetReg = RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT));
FLI.DemoteRegister = SRetReg;
NewRoot = SDB->DAG.getCopyToReg(NewRoot, SDB->getCurDebugLoc(),
SRetReg, ArgValue);
DAG.setRoot(NewRoot);
++i;
}
for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E;
++I, ++Idx) {
SmallVector<SDValue, 4> ArgValues;
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I->getType(), ValueVTs);
unsigned NumValues = ValueVTs.size();
if (I->use_empty() && NumValues)
SDB->setUnusedArgValue(I, InVals[i]);
for (unsigned Value = 0; Value != NumValues; ++Value) {
EVT VT = ValueVTs[Value];
EVT PartVT = TLI.getRegisterType(*CurDAG->getContext(), VT);
unsigned NumParts = TLI.getNumRegisters(*CurDAG->getContext(), VT);
if (!I->use_empty()) {
ISD::NodeType AssertOp = ISD::DELETED_NODE;
if (F.paramHasAttr(Idx, Attribute::SExt))
AssertOp = ISD::AssertSext;
else if (F.paramHasAttr(Idx, Attribute::ZExt))
AssertOp = ISD::AssertZext;
ArgValues.push_back(getCopyFromParts(DAG, dl, &InVals[i],
NumParts, PartVT, VT,
AssertOp));
}
i += NumParts;
}
if (!I->use_empty()) {
SDValue Res = DAG.getMergeValues(&ArgValues[0], NumValues,
SDB->getCurDebugLoc());
SDB->setValue(I, Res);
SDB->CopyToExportRegsIfNeeded(I);
}
}
assert(i == InVals.size() && "Argument register count mismatch!");
EmitFunctionEntryCode(F, SDB->DAG.getMachineFunction());
}
void
SelectionDAGISel::HandlePHINodesInSuccessorBlocks(BasicBlock *LLVMBB) {
TerminatorInst *TI = LLVMBB->getTerminator();
SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
BasicBlock *SuccBB = TI->getSuccessor(succ);
if (!isa<PHINode>(SuccBB->begin())) continue;
MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
if (!SuccsHandled.insert(SuccMBB)) continue;
MachineBasicBlock::iterator MBBI = SuccMBB->begin();
PHINode *PN;
for (BasicBlock::iterator I = SuccBB->begin();
(PN = dyn_cast<PHINode>(I)); ++I) {
if (PN->use_empty()) continue;
unsigned Reg;
Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
if (Constant *C = dyn_cast<Constant>(PHIOp)) {
unsigned &RegOut = SDB->ConstantsOut[C];
if (RegOut == 0) {
RegOut = FuncInfo->CreateRegForValue(C);
SDB->CopyValueToVirtualRegister(C, RegOut);
}
Reg = RegOut;
} else {
Reg = FuncInfo->ValueMap[PHIOp];
if (Reg == 0) {
assert(isa<AllocaInst>(PHIOp) &&
FuncInfo->StaticAllocaMap.count(cast<AllocaInst>(PHIOp)) &&
"Didn't codegen value into a register!??");
Reg = FuncInfo->CreateRegForValue(PHIOp);
SDB->CopyValueToVirtualRegister(PHIOp, Reg);
}
}
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, PN->getType(), ValueVTs);
for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) {
EVT VT = ValueVTs[vti];
unsigned NumRegisters = TLI.getNumRegisters(*CurDAG->getContext(), VT);
for (unsigned i = 0, e = NumRegisters; i != e; ++i)
SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i));
Reg += NumRegisters;
}
}
}
SDB->ConstantsOut.clear();
}
bool
SelectionDAGISel::HandlePHINodesInSuccessorBlocksFast(BasicBlock *LLVMBB,
FastISel *F) {
TerminatorInst *TI = LLVMBB->getTerminator();
SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
unsigned OrigNumPHINodesToUpdate = SDB->PHINodesToUpdate.size();
for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
BasicBlock *SuccBB = TI->getSuccessor(succ);
if (!isa<PHINode>(SuccBB->begin())) continue;
MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
if (!SuccsHandled.insert(SuccMBB)) continue;
MachineBasicBlock::iterator MBBI = SuccMBB->begin();
PHINode *PN;
for (BasicBlock::iterator I = SuccBB->begin();
(PN = dyn_cast<PHINode>(I)); ++I) {
if (PN->use_empty()) continue;
EVT VT = TLI.getValueType(PN->getType(), true);
if (VT == MVT::Other || !TLI.isTypeLegal(VT)) {
if (VT == MVT::i1)
VT = TLI.getTypeToTransformTo(*CurDAG->getContext(), VT);
else {
SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
return false;
}
}
Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
unsigned Reg = F->getRegForValue(PHIOp);
if (Reg == 0) {
SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
return false;
}
SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg));
}
}
return true;
}