/// Try to find existing copies of the incoming values in stack slots used for
/// statepoint spilling. If we can find a spill slot for the incoming value,
/// mark that slot as allocated, and reuse the same slot for this safepoint.
/// This helps to avoid series of loads and stores that only serve to reshuffle
/// values on the stack between calls.
static void reservePreviousStackSlotForValue(const Value *IncomingValue,
SelectionDAGBuilder &Builder) {
SDValue Incoming = Builder.getValue(IncomingValue);
if (isa<ConstantSDNode>(Incoming) || isa<FrameIndexSDNode>(Incoming)) {
// We won't need to spill this, so no need to check for previously
// allocated stack slots
return;
}
SDValue OldLocation = Builder.StatepointLowering.getLocation(Incoming);
if (OldLocation.getNode())
// Duplicates in input
return;
const int LookUpDepth = 6;
Optional<int> Index =
findPreviousSpillSlot(IncomingValue, Builder, LookUpDepth);
if (!Index.hasValue())
return;
const auto &StatepointSlots = Builder.FuncInfo.StatepointStackSlots;
auto SlotIt = find(StatepointSlots, *Index);
assert(SlotIt != StatepointSlots.end() &&
"Value spilled to the unknown stack slot");
// This is one of our dedicated lowering slots
const int Offset = std::distance(StatepointSlots.begin(), SlotIt);
if (Builder.StatepointLowering.isStackSlotAllocated(Offset)) {
// stack slot already assigned to someone else, can't use it!
// TODO: currently we reserve space for gc arguments after doing
// normal allocation for deopt arguments. We should reserve for
// _all_ deopt and gc arguments, then start allocating. This
// will prevent some moves being inserted when vm state changes,
// but gc state doesn't between two calls.
return;
}
// Reserve this stack slot
Builder.StatepointLowering.reserveStackSlot(Offset);
// Cache this slot so we find it when going through the normal
// assignment loop.
SDValue Loc = Builder.DAG.getTargetFrameIndex(*Index, Incoming.getValueType());
Builder.StatepointLowering.setLocation(Incoming, Loc);
}
// This function expands mips intrinsic nodes which have 64-bit input operands
// or output values.
//
// out64 = intrinsic-node in64
// =>
// lo = copy (extract-element (in64, 0))
// hi = copy (extract-element (in64, 1))
// mips-specific-node
// v0 = copy lo
// v1 = copy hi
// out64 = merge-values (v0, v1)
//
static SDValue lowerDSPIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
DebugLoc DL = Op.getDebugLoc();
bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other;
SmallVector<SDValue, 3> Ops;
unsigned OpNo = 0;
// See if Op has a chain input.
if (HasChainIn)
Ops.push_back(Op->getOperand(OpNo++));
// The next operand is the intrinsic opcode.
assert(Op->getOperand(OpNo).getOpcode() == ISD::TargetConstant);
// See if the next operand has type i64.
SDValue Opnd = Op->getOperand(++OpNo), In64;
if (Opnd.getValueType() == MVT::i64)
In64 = initAccumulator(Opnd, DL, DAG);
else
Ops.push_back(Opnd);
// Push the remaining operands.
for (++OpNo ; OpNo < Op->getNumOperands(); ++OpNo)
Ops.push_back(Op->getOperand(OpNo));
// Add In64 to the end of the list.
if (In64.getNode())
Ops.push_back(In64);
// Scan output.
SmallVector<EVT, 2> ResTys;
for (SDNode::value_iterator I = Op->value_begin(), E = Op->value_end();
I != E; ++I)
ResTys.push_back((*I == MVT::i64) ? MVT::Untyped : *I);
// Create node.
SDValue Val = DAG.getNode(Opc, DL, ResTys, &Ops[0], Ops.size());
SDValue Out = (ResTys[0] == MVT::Untyped) ? extractLOHI(Val, DL, DAG) : Val;
if (!HasChainIn)
return Out;
assert(Val->getValueType(1) == MVT::Other);
SDValue Vals[] = { Out, SDValue(Val.getNode(), 1) };
return DAG.getMergeValues(Vals, 2, DL);
}
SDValue
BPFTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
bool IsVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SDLoc &DL, SelectionDAG &DAG) const {
unsigned Opc = BPFISD::RET_FLAG;
// CCValAssign - represent the assignment of the return value to a location
SmallVector<CCValAssign, 16> RVLocs;
MachineFunction &MF = DAG.getMachineFunction();
// CCState - Info about the registers and stack slot.
CCState CCInfo(CallConv, IsVarArg, MF, RVLocs, *DAG.getContext());
if (MF.getFunction()->getReturnType()->isAggregateType()) {
fail(DL, DAG, "only integer returns supported");
return DAG.getNode(Opc, DL, MVT::Other, Chain);
}
// Analize return values.
CCInfo.AnalyzeReturn(Outs, RetCC_BPF64);
SDValue Flag;
SmallVector<SDValue, 4> RetOps(1, Chain);
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), OutVals[i], Flag);
// Guarantee that all emitted copies are stuck together,
// avoiding something bad.
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
RetOps[0] = Chain; // Update chain.
// Add the flag if we have it.
if (Flag.getNode())
RetOps.push_back(Flag);
return DAG.getNode(Opc, DL, MVT::Other, RetOps);
}
SDValue
AMDGPUTargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const
{
EVT OVT = Op.getValueType();
SDValue DST;
if (OVT.getScalarType() == MVT::i64) {
DST = LowerSDIV64(Op, DAG);
} else if (OVT.getScalarType() == MVT::i32) {
DST = LowerSDIV32(Op, DAG);
} else if (OVT.getScalarType() == MVT::i16
|| OVT.getScalarType() == MVT::i8) {
DST = LowerSDIV24(Op, DAG);
} else {
DST = SDValue(Op.getNode(), 0);
}
return DST;
}
SDValue BlackfinTargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
switch (Op.getOpcode()) {
default:
Op.getNode()->dump();
llvm_unreachable("Should not custom lower this!");
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
case ISD::GlobalTLSAddress:
llvm_unreachable("TLS not implemented for Blackfin.");
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
// Frame & Return address. Currently unimplemented
case ISD::FRAMEADDR: return SDValue();
case ISD::RETURNADDR: return SDValue();
case ISD::ADDE:
case ISD::SUBE: return LowerADDE(Op, DAG);
}
}
bool AMDGPUDAGToDAGISel::SelectU24(SDValue Op, SDValue &U24) {
APInt KnownZero;
APInt KnownOne;
CurDAG->ComputeMaskedBits(Op, KnownZero, KnownOne);
assert (Op.getValueType() == MVT::i32);
// ANY_EXTEND and EXTLOAD operations can only be done on types smaller than
// i32. These smaller types are legal to use with the i24 instructions.
if ((KnownZero & APInt(KnownZero.getBitWidth(), 0xFF000000)) == 0xFF000000 ||
Op.getOpcode() == ISD::ANY_EXTEND ||
ISD::isEXTLoad(Op.getNode())) {
U24 = SimplifyI24(Op);
return true;
}
return false;
}
// Select constant vector splats whose value is a power of 2.
//
// In addition to the requirements of selectVSplat(), this function returns
// true and sets Imm if:
// * The splat value is the same width as the elements of the vector
// * The splat value is a power of two.
//
// This function looks through ISD::BITCAST nodes.
// TODO: This might not be appropriate for big-endian MSA since BITCAST is
// sometimes a shuffle in big-endian mode.
bool MipsSEDAGToDAGISel::selectVSplatUimmPow2(SDValue N, SDValue &Imm) const {
APInt ImmValue;
EVT EltTy = N->getValueType(0).getVectorElementType();
if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0);
if (selectVSplat (N.getNode(), ImmValue) &&
ImmValue.getBitWidth() == EltTy.getSizeInBits()) {
int32_t Log2 = ImmValue.exactLogBase2();
if (Log2 != -1) {
Imm = CurDAG->getTargetConstant(Log2, EltTy);
return true;
}
}
return false;
}
// Select constant vector splats whose value only has a consecutive sequence
// of right-most bits set (e.g. 0b00...0011...11).
//
// In addition to the requirements of selectVSplat(), this function returns
// true and sets Imm if:
// * The splat value is the same width as the elements of the vector
// * The splat value is a consecutive sequence of right-most bits.
//
// This function looks through ISD::BITCAST nodes.
// TODO: This might not be appropriate for big-endian MSA since BITCAST is
// sometimes a shuffle in big-endian mode.
bool MipsSEDAGToDAGISel::selectVSplatMaskR(SDValue N, SDValue &Imm) const {
APInt ImmValue;
EVT EltTy = N->getValueType(0).getVectorElementType();
if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0);
if (selectVSplat(N.getNode(), ImmValue) &&
ImmValue.getBitWidth() == EltTy.getSizeInBits()) {
// Extract the run of set bits starting with bit zero, and test that the
// result is the same as the original value
if (ImmValue == (ImmValue & ~(ImmValue + 1))) {
Imm = CurDAG->getTargetConstant(ImmValue.countPopulation(), EltTy);
return true;
}
}
return false;
}
SDValue MSP430TargetLowering::LowerShifts(SDValue Op,
SelectionDAG &DAG) const {
unsigned Opc = Op.getOpcode();
SDNode* N = Op.getNode();
EVT VT = Op.getValueType();
DebugLoc dl = N->getDebugLoc();
// Expand non-constant shifts to loops:
if (!isa<ConstantSDNode>(N->getOperand(1)))
switch (Opc) {
default:
assert(0 && "Invalid shift opcode!");
case ISD::SHL:
return DAG.getNode(MSP430ISD::SHL, dl,
VT, N->getOperand(0), N->getOperand(1));
case ISD::SRA:
return DAG.getNode(MSP430ISD::SRA, dl,
VT, N->getOperand(0), N->getOperand(1));
case ISD::SRL:
return DAG.getNode(MSP430ISD::SRL, dl,
VT, N->getOperand(0), N->getOperand(1));
}
uint64_t ShiftAmount = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
// Expand the stuff into sequence of shifts.
// FIXME: for some shift amounts this might be done better!
// E.g.: foo >> (8 + N) => sxt(swpb(foo)) >> N
SDValue Victim = N->getOperand(0);
if (Opc == ISD::SRL && ShiftAmount) {
// Emit a special goodness here:
// srl A, 1 => clrc; rrc A
Victim = DAG.getNode(MSP430ISD::RRC, dl, VT, Victim);
ShiftAmount -= 1;
}
while (ShiftAmount--)
Victim = DAG.getNode((Opc == ISD::SHL ? MSP430ISD::RLA : MSP430ISD::RRA),
dl, VT, Victim);
return Victim;
}
SDValue
SparcTargetLowering::LowerReturn(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
DebugLoc dl, SelectionDAG &DAG) const {
// CCValAssign - represent the assignment of the return value to locations.
SmallVector<CCValAssign, 16> RVLocs;
// CCState - Info about the registers and stack slot.
CCState CCInfo(CallConv, isVarArg, DAG.getTarget(),
RVLocs, *DAG.getContext());
// Analize return values.
CCInfo.AnalyzeReturn(Outs, RetCC_Sparc32);
// If this is the first return lowered for this function, add the regs to the
// liveout set for the function.
if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
for (unsigned i = 0; i != RVLocs.size(); ++i)
if (RVLocs[i].isRegLoc())
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
SDValue Flag;
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
OutVals[i], Flag);
// Guarantee that all emitted copies are stuck together with flags.
Flag = Chain.getValue(1);
}
if (Flag.getNode())
return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain);
}
SDValue AVRTargetLowering::LowerShifts(SDValue Op,
SelectionDAG &DAG) const {
unsigned Opc = Op.getOpcode();
SDNode* N = Op.getNode();
EVT VT = Op.getValueType();
DebugLoc dl = N->getDebugLoc();
// Expand non-constant shifts to loops:
if (!isa<ConstantSDNode>(N->getOperand(1)))
switch (Opc) {
default:
assert(0 && "Invalid shift opcode!");
case ISD::SHL:
return DAG.getNode(AVRISD::SHL, dl,
VT, N->getOperand(0), N->getOperand(1));
case ISD::SRA:
return DAG.getNode(AVRISD::SRA, dl,
VT, N->getOperand(0), N->getOperand(1));
case ISD::SRL:
return DAG.getNode(AVRISD::SRL, dl,
VT, N->getOperand(0), N->getOperand(1));
}
uint64_t ShiftAmount = N->getConstantOperandVal(1);
// Expand the stuff into sequence of shifts.
// FIXME: for some shift amounts this might be done better!
// E.g.: foo >> (8 + N) => sxt(swpb(foo)) >> N
SDValue Victim = N->getOperand(0);
unsigned int TargetOpcode = 0;
switch(Opc) {
case ISD::SHL: TargetOpcode = AVRISD::SHLC; break;
case ISD::SRA: TargetOpcode = AVRISD::SRAC; break;
case ISD::SRL: TargetOpcode = AVRISD::SRLC; break;
}
while (ShiftAmount--)
Victim = DAG.getNode(TargetOpcode, dl, VT, Victim);
return Victim;
}
/// getVR - Return the virtual register corresponding to the specified result
/// of the specified node.
unsigned ScheduleDAGSDNodes::getVR(SDValue Op,
DenseMap<SDValue, unsigned> &VRBaseMap) {
if (Op.isMachineOpcode() &&
Op.getMachineOpcode() == TargetInstrInfo::IMPLICIT_DEF) {
// Add an IMPLICIT_DEF instruction before every use.
unsigned VReg = getDstOfOnlyCopyToRegUse(Op.getNode(), Op.getResNo());
// IMPLICIT_DEF can produce any type of result so its TargetInstrDesc
// does not include operand register class info.
if (!VReg) {
const TargetRegisterClass *RC = TLI->getRegClassFor(Op.getValueType());
VReg = MRI.createVirtualRegister(RC);
}
BuildMI(BB, Op.getDebugLoc(), TII->get(TargetInstrInfo::IMPLICIT_DEF),VReg);
return VReg;
}
DenseMap<SDValue, unsigned>::iterator I = VRBaseMap.find(Op);
assert(I != VRBaseMap.end() && "Node emitted out of order - late");
return I->second;
}
// Select constant vector splats.
//
// In addition to the requirements of selectVSplat(), this function returns
// true and sets Imm if:
// * The splat value is the same width as the elements of the vector
// * The splat value fits in an integer with the specified signed-ness and
// width.
//
// This function looks through ISD::BITCAST nodes.
// TODO: This might not be appropriate for big-endian MSA since BITCAST is
// sometimes a shuffle in big-endian mode.
//
// It's worth noting that this function is not used as part of the selection
// of ldi.[bhwd] since it does not permit using the wrong-typed ldi.[bhwd]
// instruction to achieve the desired bit pattern. ldi.[bhwd] is selected in
// MipsSEDAGToDAGISel::selectNode.
bool MipsSEDAGToDAGISel::
selectVSplatCommon(SDValue N, SDValue &Imm, bool Signed,
unsigned ImmBitSize) const {
APInt ImmValue;
EVT EltTy = N->getValueType(0).getVectorElementType();
if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0);
if (selectVSplat (N.getNode(), ImmValue) &&
ImmValue.getBitWidth() == EltTy.getSizeInBits()) {
if (( Signed && ImmValue.isSignedIntN(ImmBitSize)) ||
(!Signed && ImmValue.isIntN(ImmBitSize))) {
Imm = CurDAG->getTargetConstant(ImmValue, EltTy);
return true;
}
}
return false;
}
bool AMDGPUDAGToDAGISel::FoldOperand(SDValue &Src, SDValue &Sel, SDValue &Neg,
SDValue &Abs, const R600InstrInfo *TII) {
switch (Src.getOpcode()) {
case ISD::FNEG:
Src = Src.getOperand(0);
Neg = CurDAG->getTargetConstant(1, MVT::i32);
return true;
case ISD::FABS:
if (!Abs.getNode())
return false;
Src = Src.getOperand(0);
Abs = CurDAG->getTargetConstant(1, MVT::i32);
return true;
case ISD::BITCAST:
Src = Src.getOperand(0);
return true;
default:
return false;
}
}
// Helper for AddGlue to clone node operands.
static void CloneNodeWithValues(SDNode *N, SelectionDAG *DAG, ArrayRef<EVT> VTs,
SDValue ExtraOper = SDValue()) {
SmallVector<SDValue, 8> Ops(N->op_begin(), N->op_end());
if (ExtraOper.getNode())
Ops.push_back(ExtraOper);
SDVTList VTList = DAG->getVTList(VTs);
MachineSDNode *MN = dyn_cast<MachineSDNode>(N);
// Store memory references.
SmallVector<MachineMemOperand *, 2> MMOs;
if (MN)
MMOs.assign(MN->memoperands_begin(), MN->memoperands_end());
DAG->MorphNodeTo(N, N->getOpcode(), VTList, Ops);
// Reset the memory references
if (MN)
DAG->setNodeMemRefs(MN, MMOs);
}
static void AddGlue(SDNode *N, SDValue Glue, bool AddGlue, SelectionDAG *DAG) {
SmallVector<EVT, 4> VTs;
SDNode *GlueDestNode = Glue.getNode();
// Don't add glue from a node to itself.
if (GlueDestNode == N) return;
// Don't add glue to something that already has it, either as a use or value.
if (N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Glue ||
N->getValueType(N->getNumValues() - 1) == MVT::Glue) {
return;
}
for (unsigned I = 0, E = N->getNumValues(); I != E; ++I)
VTs.push_back(N->getValueType(I));
if (AddGlue)
VTs.push_back(MVT::Glue);
SmallVector<SDValue, 4> Ops;
for (unsigned I = 0, E = N->getNumOperands(); I != E; ++I)
Ops.push_back(N->getOperand(I));
if (GlueDestNode)
Ops.push_back(Glue);
SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size());
MachineSDNode::mmo_iterator Begin = 0, End = 0;
MachineSDNode *MN = dyn_cast<MachineSDNode>(N);
// Store memory references.
if (MN) {
Begin = MN->memoperands_begin();
End = MN->memoperands_end();
}
DAG->MorphNodeTo(N, N->getOpcode(), VTList, &Ops[0], Ops.size());
// Reset the memory references
if (MN)
MN->setMemRefs(Begin, End);
}
/// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
/// inline asm expressions.
bool NVPTXDAGToDAGISel::SelectInlineAsmMemoryOperand(
const SDValue &Op, char ConstraintCode, std::vector<SDValue> &OutOps) {
SDValue Op0, Op1;
switch (ConstraintCode) {
default:
return true;
case 'm': // memory
if (SelectDirectAddr(Op, Op0)) {
OutOps.push_back(Op0);
OutOps.push_back(CurDAG->getTargetConstant(0, MVT::i32));
return false;
}
if (SelectADDRri(Op.getNode(), Op, Op0, Op1)) {
OutOps.push_back(Op0);
OutOps.push_back(Op1);
return false;
}
break;
}
return true;
}
static void AddFlags(SDNode *N, SDValue Flag, bool AddFlag,
SelectionDAG *DAG) {
SmallVector<EVT, 4> VTs;
SDNode *FlagDestNode = Flag.getNode();
// Don't add a flag from a node to itself.
if (FlagDestNode == N) return;
// Don't add a flag to something which already has a flag.
if (N->getValueType(N->getNumValues() - 1) == MVT::Flag) return;
for (unsigned I = 0, E = N->getNumValues(); I != E; ++I)
VTs.push_back(N->getValueType(I));
if (AddFlag)
VTs.push_back(MVT::Flag);
SmallVector<SDValue, 4> Ops;
for (unsigned I = 0, E = N->getNumOperands(); I != E; ++I)
Ops.push_back(N->getOperand(I));
if (FlagDestNode)
Ops.push_back(Flag);
SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size());
MachineSDNode::mmo_iterator Begin = 0, End = 0;
MachineSDNode *MN = dyn_cast<MachineSDNode>(N);
// Store memory references.
if (MN) {
Begin = MN->memoperands_begin();
End = MN->memoperands_end();
}
DAG->MorphNodeTo(N, N->getOpcode(), VTList, &Ops[0], Ops.size());
// Reset the memory references
if (MN)
MN->setMemRefs(Begin, End);
}
// Helper for AddGlue to clone node operands.
static void CloneNodeWithValues(SDNode *N, SelectionDAG *DAG, ArrayRef<EVT> VTs,
SDValue ExtraOper = SDValue()) {
SmallVector<SDValue, 8> Ops(N->op_begin(), N->op_end());
if (ExtraOper.getNode())
Ops.push_back(ExtraOper);
SDVTList VTList = DAG->getVTList(VTs);
MachineSDNode::mmo_iterator Begin = nullptr, End = nullptr;
MachineSDNode *MN = dyn_cast<MachineSDNode>(N);
// Store memory references.
if (MN) {
Begin = MN->memoperands_begin();
End = MN->memoperands_end();
}
DAG->MorphNodeTo(N, N->getOpcode(), VTList, Ops);
// Reset the memory references
if (MN)
MN->setMemRefs(Begin, End);
}
static bool AddGlue(SDNode *N, SDValue Glue, bool AddGlue, SelectionDAG *DAG) {
SDNode *GlueDestNode = Glue.getNode();
// Don't add glue from a node to itself.
if (GlueDestNode == N) return false;
// Don't add a glue operand to something that already uses glue.
if (GlueDestNode &&
N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Glue) {
return false;
}
// Don't add glue to something that already has a glue value.
if (N->getValueType(N->getNumValues() - 1) == MVT::Glue) return false;
SmallVector<EVT, 4> VTs(N->value_begin(), N->value_end());
if (AddGlue)
VTs.push_back(MVT::Glue);
CloneNodeWithValues(N, DAG, VTs, Glue);
return true;
}
SDValue Y86TargetLowering::LowerReturn(SDValue Chain,
CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
SDLoc dl,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
// Gather info about the return values.
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, MF, RVLocs, *DAG.getContext());
CCInfo.AnalyzeReturn(Outs, RetCC_Y86);
SDValue Flag;
SmallVector<SDValue, 6> RetOps(1, Chain); // Operand 0 is the chain.
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
SDValue ValToCopy = OutVals[i];
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), ValToCopy, Flag);
Flag = Chain.getValue(1); // Copies are glued together with flags.
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
RetOps[0] = Chain; // Update the chain.
// Add the flag if we have it.
if (Flag.getNode())
RetOps.push_back(Flag);
return DAG.getNode(Y86ISD::RET_FLAG, dl, MVT::Other, RetOps);
}
void SelectionDAGBuilder::visitGCResult(const CallInst &CI) {
// The result value of the gc_result is simply the result of the actual
// call. We've already emitted this, so just grab the value.
Instruction *I = cast<Instruction>(CI.getArgOperand(0));
assert(isStatepoint(I) && "first argument must be a statepoint token");
if (isa<InvokeInst>(I)) {
// For invokes we should have stored call result in a virtual register.
// We can not use default getValue() functionality to copy value from this
// register because statepoint and actuall call return types can be
// different, and getValue() will use CopyFromReg of the wrong type,
// which is always i32 in our case.
PointerType *CalleeType = cast<PointerType>(
ImmutableStatepoint(I).getCalledValue()->getType());
Type *RetTy =
cast<FunctionType>(CalleeType->getElementType())->getReturnType();
SDValue CopyFromReg = getCopyFromRegs(I, RetTy);
assert(CopyFromReg.getNode());
setValue(&CI, CopyFromReg);
} else {
setValue(&CI, getValue(I));
}
}
/// Do extensive, expensive, sanity checking.
void DAGTypeLegalizer::PerformExpensiveChecks() {
// If a node is not processed, then none of its values should be mapped by any
// of PromotedIntegers, ExpandedIntegers, ..., ReplacedValues.
// If a node is processed, then each value with an illegal type must be mapped
// by exactly one of PromotedIntegers, ExpandedIntegers, ..., ReplacedValues.
// Values with a legal type may be mapped by ReplacedValues, but not by any of
// the other maps.
// Note that these invariants may not hold momentarily when processing a node:
// the node being processed may be put in a map before being marked Processed.
// Note that it is possible to have nodes marked NewNode in the DAG. This can
// occur in two ways. Firstly, a node may be created during legalization but
// never passed to the legalization core. This is usually due to the implicit
// folding that occurs when using the DAG.getNode operators. Secondly, a new
// node may be passed to the legalization core, but when analyzed may morph
// into a different node, leaving the original node as a NewNode in the DAG.
// A node may morph if one of its operands changes during analysis. Whether
// it actually morphs or not depends on whether, after updating its operands,
// it is equivalent to an existing node: if so, it morphs into that existing
// node (CSE). An operand can change during analysis if the operand is a new
// node that morphs, or it is a processed value that was mapped to some other
// value (as recorded in ReplacedValues) in which case the operand is turned
// into that other value. If a node morphs then the node it morphed into will
// be used instead of it for legalization, however the original node continues
// to live on in the DAG.
// The conclusion is that though there may be nodes marked NewNode in the DAG,
// all uses of such nodes are also marked NewNode: the result is a fungus of
// NewNodes growing on top of the useful nodes, and perhaps using them, but
// not used by them.
// If a value is mapped by ReplacedValues, then it must have no uses, except
// by nodes marked NewNode (see above).
// The final node obtained by mapping by ReplacedValues is not marked NewNode.
// Note that ReplacedValues should be applied iteratively.
// Note that the ReplacedValues map may also map deleted nodes (by iterating
// over the DAG we never dereference deleted nodes). This means that it may
// also map nodes marked NewNode if the deallocated memory was reallocated as
// another node, and that new node was not seen by the LegalizeTypes machinery
// (for example because it was created but not used). In general, we cannot
// distinguish between new nodes and deleted nodes.
SmallVector<SDNode*, 16> NewNodes;
for (SDNode &Node : DAG.allnodes()) {
// Remember nodes marked NewNode - they are subject to extra checking below.
if (Node.getNodeId() == NewNode)
NewNodes.push_back(&Node);
for (unsigned i = 0, e = Node.getNumValues(); i != e; ++i) {
SDValue Res(&Node, i);
EVT VT = Res.getValueType();
bool Failed = false;
unsigned Mapped = 0;
if (ReplacedValues.find(Res) != ReplacedValues.end()) {
Mapped |= 1;
// Check that remapped values are only used by nodes marked NewNode.
for (SDNode::use_iterator UI = Node.use_begin(), UE = Node.use_end();
UI != UE; ++UI)
if (UI.getUse().getResNo() == i)
assert(UI->getNodeId() == NewNode &&
"Remapped value has non-trivial use!");
// Check that the final result of applying ReplacedValues is not
// marked NewNode.
SDValue NewVal = ReplacedValues[Res];
DenseMap<SDValue, SDValue>::iterator I = ReplacedValues.find(NewVal);
while (I != ReplacedValues.end()) {
NewVal = I->second;
I = ReplacedValues.find(NewVal);
}
assert(NewVal.getNode()->getNodeId() != NewNode &&
"ReplacedValues maps to a new node!");
}
if (PromotedIntegers.find(Res) != PromotedIntegers.end())
Mapped |= 2;
if (SoftenedFloats.find(Res) != SoftenedFloats.end())
Mapped |= 4;
if (ScalarizedVectors.find(Res) != ScalarizedVectors.end())
Mapped |= 8;
if (ExpandedIntegers.find(Res) != ExpandedIntegers.end())
Mapped |= 16;
if (ExpandedFloats.find(Res) != ExpandedFloats.end())
Mapped |= 32;
if (SplitVectors.find(Res) != SplitVectors.end())
Mapped |= 64;
if (WidenedVectors.find(Res) != WidenedVectors.end())
Mapped |= 128;
if (PromotedFloats.find(Res) != PromotedFloats.end())
Mapped |= 256;
if (Node.getNodeId() != Processed) {
// Since we allow ReplacedValues to map deleted nodes, it may map nodes
// marked NewNode too, since a deleted node may have been reallocated as
// another node that has not been seen by the LegalizeTypes machinery.
if ((Node.getNodeId() == NewNode && Mapped > 1) ||
(Node.getNodeId() != NewNode && Mapped != 0)) {
dbgs() << "Unprocessed value in a map!";
Failed = true;
}
} else if (isTypeLegal(VT) || IgnoreNodeResults(&Node)) {
if (Mapped > 1) {
dbgs() << "Value with legal type was transformed!";
Failed = true;
}
} else {
// If the value can be kept in HW registers, softening machinery can
// leave it unchanged and don't put it to any map.
if (Mapped == 0 &&
!(getTypeAction(VT) == TargetLowering::TypeSoftenFloat &&
isLegalInHWReg(VT))) {
dbgs() << "Processed value not in any map!";
Failed = true;
} else if (Mapped & (Mapped - 1)) {
dbgs() << "Value in multiple maps!";
Failed = true;
}
}
if (Failed) {
if (Mapped & 1)
dbgs() << " ReplacedValues";
if (Mapped & 2)
dbgs() << " PromotedIntegers";
if (Mapped & 4)
dbgs() << " SoftenedFloats";
if (Mapped & 8)
dbgs() << " ScalarizedVectors";
if (Mapped & 16)
dbgs() << " ExpandedIntegers";
if (Mapped & 32)
dbgs() << " ExpandedFloats";
if (Mapped & 64)
dbgs() << " SplitVectors";
if (Mapped & 128)
dbgs() << " WidenedVectors";
if (Mapped & 256)
dbgs() << " PromotedFloats";
dbgs() << "\n";
llvm_unreachable(nullptr);
}
}
}
// Checked that NewNodes are only used by other NewNodes.
for (unsigned i = 0, e = NewNodes.size(); i != e; ++i) {
SDNode *N = NewNodes[i];
for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
UI != UE; ++UI)
assert(UI->getNodeId() == NewNode && "NewNode used by non-NewNode!");
}
}
/// LowerCCCCallTo - functions arguments are copied from virtual regs to
/// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
/// TODO: sret.
SDValue
MSP430TargetLowering::LowerCCCCallTo(SDValue Chain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
bool isTailCall,
const SmallVectorImpl<ISD::OutputArg>
&Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeCallOperands(Outs, CC_MSP430);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
Chain = DAG.getCALLSEQ_START(Chain ,DAG.getConstant(NumBytes,
getPointerTy(), true));
SmallVector<std::pair<unsigned, SDValue>, 4> RegsToPass;
SmallVector<SDValue, 12> MemOpChains;
SDValue StackPtr;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[i];
// Promote the value if needed.
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
break;
}
// Arguments that can be passed on register must be kept at RegsToPass
// vector
if (VA.isRegLoc()) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
assert(VA.isMemLoc());
if (StackPtr.getNode() == 0)
StackPtr = DAG.getCopyFromReg(Chain, dl, MSP430::SPW, getPointerTy());
SDValue PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(),
StackPtr,
DAG.getIntPtrConstant(VA.getLocMemOffset()));
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(),false, false, 0));
}
}
// Transform all store nodes into one single node because all store nodes are
// independent of each other.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// Build a sequence of copy-to-reg nodes chained together with token chain and
// flag operands which copy the outgoing args into registers. The InFlag in
// necessary since all emitted instructions must be stuck together.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
// If the callee is a GlobalAddress node (quite common, every direct call is)
// turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
// Likewise ExternalSymbol -> TargetExternalSymbol.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i16);
else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i16);
// Returns a chain & a flag for retval copy to use.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
Ops.push_back(Callee);
// Add argument registers to the end of the list so that they are
// known live into the call.
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
if (InFlag.getNode())
Ops.push_back(InFlag);
Chain = DAG.getNode(MSP430ISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getConstant(NumBytes, getPointerTy(), true),
DAG.getConstant(0, getPointerTy(), true),
InFlag);
InFlag = Chain.getValue(1);
// Handle result values, copying them out of physregs into vregs that we
// return.
return LowerCallResult(Chain, InFlag, CallConv, isVarArg, Ins, dl,
DAG, InVals);
}
SDValue
NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
const DataLayout *TD = getDataLayout();
const Function *F = MF.getFunction();
const AttrListPtr &PAL = F->getAttributes();
SDValue Root = DAG.getRoot();
std::vector<SDValue> OutChains;
bool isKernel = llvm::isKernelFunction(*F);
bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
std::vector<Type *> argTypes;
std::vector<const Argument *> theArgs;
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I) {
theArgs.push_back(I);
argTypes.push_back(I->getType());
}
assert(argTypes.size() == Ins.size() &&
"Ins types and function types did not match");
int idx = 0;
for (unsigned i=0, e=Ins.size(); i!=e; ++i, ++idx) {
Type *Ty = argTypes[i];
EVT ObjectVT = getValueType(Ty);
assert(ObjectVT == Ins[i].VT &&
"Ins type did not match function type");
// If the kernel argument is image*_t or sampler_t, convert it to
// a i32 constant holding the parameter position. This can later
// matched in the AsmPrinter to output the correct mangled name.
if (isImageOrSamplerVal(theArgs[i],
(theArgs[i]->getParent() ?
theArgs[i]->getParent()->getParent() : 0))) {
assert(isKernel && "Only kernels can have image/sampler params");
InVals.push_back(DAG.getConstant(i+1, MVT::i32));
continue;
}
if (theArgs[i]->use_empty()) {
// argument is dead
InVals.push_back(DAG.getNode(ISD::UNDEF, dl, ObjectVT));
continue;
}
// In the following cases, assign a node order of "idx+1"
// to newly created nodes. The SDNOdes for params have to
// appear in the same order as their order of appearance
// in the original function. "idx+1" holds that order.
if (PAL.getParamAttributes(i+1).hasAttribute(Attributes::ByVal) == false) {
// A plain scalar.
if (isABI || isKernel) {
// If ABI, load from the param symbol
SDValue Arg = getParamSymbol(DAG, idx);
Value *srcValue = new Argument(PointerType::get(ObjectVT.getTypeForEVT(
F->getContext()),
llvm::ADDRESS_SPACE_PARAM));
SDValue p = DAG.getLoad(ObjectVT, dl, Root, Arg,
MachinePointerInfo(srcValue), false, false,
false,
TD->getABITypeAlignment(ObjectVT.getTypeForEVT(
F->getContext())));
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
InVals.push_back(p);
}
else {
// If no ABI, just move the param symbol
SDValue Arg = getParamSymbol(DAG, idx, ObjectVT);
SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
InVals.push_back(p);
}
continue;
}
// Param has ByVal attribute
if (isABI || isKernel) {
// Return MoveParam(param symbol).
// Ideally, the param symbol can be returned directly,
// but when SDNode builder decides to use it in a CopyToReg(),
// machine instruction fails because TargetExternalSymbol
// (not lowered) is target dependent, and CopyToReg assumes
// the source is lowered.
SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
if (isKernel)
InVals.push_back(p);
else {
SDValue p2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, ObjectVT,
DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, MVT::i32),
p);
InVals.push_back(p2);
}
} else {
// Have to move a set of param symbols to registers and
// store them locally and return the local pointer in InVals
const PointerType *elemPtrType = dyn_cast<PointerType>(argTypes[i]);
assert(elemPtrType &&
"Byval parameter should be a pointer type");
Type *elemType = elemPtrType->getElementType();
// Compute the constituent parts
SmallVector<EVT, 16> vtparts;
SmallVector<uint64_t, 16> offsets;
ComputeValueVTs(*this, elemType, vtparts, &offsets, 0);
unsigned totalsize = 0;
for (unsigned j=0, je=vtparts.size(); j!=je; ++j)
totalsize += vtparts[j].getStoreSizeInBits();
SDValue localcopy = DAG.getFrameIndex(MF.getFrameInfo()->
CreateStackObject(totalsize/8, 16, false),
getPointerTy());
unsigned sizesofar = 0;
std::vector<SDValue> theChains;
for (unsigned j=0, je=vtparts.size(); j!=je; ++j) {
unsigned numElems = 1;
if (vtparts[j].isVector()) numElems = vtparts[j].getVectorNumElements();
for (unsigned k=0, ke=numElems; k!=ke; ++k) {
EVT tmpvt = vtparts[j];
if (tmpvt.isVector()) tmpvt = tmpvt.getVectorElementType();
SDValue arg = DAG.getNode(NVPTXISD::MoveParam, dl, tmpvt,
getParamSymbol(DAG, idx, tmpvt));
SDValue addr = DAG.getNode(ISD::ADD, dl, getPointerTy(), localcopy,
DAG.getConstant(sizesofar, getPointerTy()));
theChains.push_back(DAG.getStore(Chain, dl, arg, addr,
MachinePointerInfo(), false, false, 0));
sizesofar += tmpvt.getStoreSizeInBits()/8;
++idx;
}
}
--idx;
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &theChains[0],
theChains.size());
InVals.push_back(localcopy);
}
}
// Clang will check explicit VarArg and issue error if any. However, Clang
// will let code with
// implicit var arg like f() pass.
// We treat this case as if the arg list is empty.
//if (F.isVarArg()) {
// assert(0 && "VarArg not supported yet!");
//}
if (!OutChains.empty())
DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size()));
return Chain;
}
SDValue
NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
DebugLoc &dl = CLI.DL;
SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &isTailCall = CLI.IsTailCall;
ArgListTy &Args = CLI.Args;
Type *retTy = CLI.RetTy;
ImmutableCallSite *CS = CLI.CS;
bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
SDValue tempChain = Chain;
Chain = DAG.getCALLSEQ_START(Chain,
DAG.getIntPtrConstant(uniqueCallSite, true));
SDValue InFlag = Chain.getValue(1);
assert((Outs.size() == Args.size()) &&
"Unexpected number of arguments to function call");
unsigned paramCount = 0;
// Declare the .params or .reg need to pass values
// to the function
for (unsigned i=0, e=Outs.size(); i!=e; ++i) {
EVT VT = Outs[i].VT;
if (Outs[i].Flags.isByVal() == false) {
// Plain scalar
// for ABI, declare .param .b<size> .param<n>;
// for nonABI, declare .reg .b<size> .param<n>;
unsigned isReg = 1;
if (isABI)
isReg = 0;
unsigned sz = VT.getSizeInBits();
if (VT.isInteger() && (sz < 32)) sz = 32;
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareParamOps[] = { Chain,
DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(isReg, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(0, MVT::i32), OutVals[i], InFlag };
unsigned opcode = NVPTXISD::StoreParam;
if (isReg)
opcode = NVPTXISD::MoveToParam;
else {
if (Outs[i].Flags.isZExt())
opcode = NVPTXISD::StoreParamU32;
else if (Outs[i].Flags.isSExt())
opcode = NVPTXISD::StoreParamS32;
}
Chain = DAG.getNode(opcode, dl, CopyParamVTs, CopyParamOps, 5);
InFlag = Chain.getValue(1);
++paramCount;
continue;
}
// struct or vector
SmallVector<EVT, 16> vtparts;
const PointerType *PTy = dyn_cast<PointerType>(Args[i].Ty);
assert(PTy &&
"Type of a byval parameter should be pointer");
ComputeValueVTs(*this, PTy->getElementType(), vtparts);
if (isABI) {
// declare .param .align 16 .b8 .param<n>[<size>];
unsigned sz = Outs[i].Flags.getByValSize();
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
// The ByValAlign in the Outs[i].Flags is alway set at this point, so we
// don't need to
// worry about natural alignment or not. See TargetLowering::LowerCallTo()
SDValue DeclareParamOps[] = { Chain,
DAG.getConstant(Outs[i].Flags.getByValAlign(), MVT::i32),
DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(sz, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
unsigned curOffset = 0;
for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
unsigned elems = 1;
EVT elemtype = vtparts[j];
if (vtparts[j].isVector()) {
elems = vtparts[j].getVectorNumElements();
elemtype = vtparts[j].getVectorElementType();
}
for (unsigned k=0,ke=elems; k!=ke; ++k) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 8)) sz = 8;
SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(),
OutVals[i],
DAG.getConstant(curOffset,
getPointerTy()));
SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
MachinePointerInfo(), false, false, false, 0);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount,
MVT::i32),
DAG.getConstant(curOffset, MVT::i32),
theVal, InFlag };
Chain = DAG.getNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
CopyParamOps, 5);
InFlag = Chain.getValue(1);
curOffset += sz/8;
}
}
++paramCount;
continue;
}
// Non-abi, struct or vector
// Declare a bunch or .reg .b<size> .param<n>
unsigned curOffset = 0;
for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
unsigned elems = 1;
EVT elemtype = vtparts[j];
if (vtparts[j].isVector()) {
elems = vtparts[j].getVectorNumElements();
elemtype = vtparts[j].getVectorElementType();
}
for (unsigned k=0,ke=elems; k!=ke; ++k) {
unsigned sz = elemtype.getSizeInBits();
if (elemtype.isInteger() && (sz < 32)) sz = 32;
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareParamOps[] = { Chain, DAG.getConstant(paramCount,
MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(1, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
DeclareParamOps, 5);
InFlag = Chain.getValue(1);
SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[i],
DAG.getConstant(curOffset,
getPointerTy()));
SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
MachinePointerInfo(), false, false, false, 0);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
DAG.getConstant(0, MVT::i32), theVal,
InFlag };
Chain = DAG.getNode(NVPTXISD::MoveToParam, dl, CopyParamVTs,
CopyParamOps, 5);
InFlag = Chain.getValue(1);
++paramCount;
}
}
}
GlobalAddressSDNode *Func = dyn_cast<GlobalAddressSDNode>(Callee.getNode());
unsigned retAlignment = 0;
// Handle Result
unsigned retCount = 0;
if (Ins.size() > 0) {
SmallVector<EVT, 16> resvtparts;
ComputeValueVTs(*this, retTy, resvtparts);
// Declare one .param .align 16 .b8 func_retval0[<size>] for ABI or
// individual .reg .b<size> func_retval<0..> for non ABI
unsigned resultsz = 0;
for (unsigned i=0,e=resvtparts.size(); i!=e; ++i) {
unsigned elems = 1;
EVT elemtype = resvtparts[i];
if (resvtparts[i].isVector()) {
elems = resvtparts[i].getVectorNumElements();
elemtype = resvtparts[i].getVectorElementType();
}
for (unsigned j=0,je=elems; j!=je; ++j) {
unsigned sz = elemtype.getSizeInBits();
if (isABI == false) {
if (elemtype.isInteger() && (sz < 32)) sz = 32;
}
else {
if (elemtype.isInteger() && (sz < 8)) sz = 8;
}
if (isABI == false) {
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(2, MVT::i32),
DAG.getConstant(sz, MVT::i32),
DAG.getConstant(retCount, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
++retCount;
}
resultsz += sz;
}
}
if (isABI) {
if (retTy->isPrimitiveType() || retTy->isIntegerTy() ||
retTy->isPointerTy() ) {
// Scalar needs to be at least 32bit wide
if (resultsz < 32)
resultsz = 32;
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(1, MVT::i32),
DAG.getConstant(resultsz, MVT::i32),
DAG.getConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
}
else {
if (Func) { // direct call
if (!llvm::getAlign(*(CS->getCalledFunction()), 0, retAlignment))
retAlignment = getDataLayout()->getABITypeAlignment(retTy);
} else { // indirect call
const CallInst *CallI = dyn_cast<CallInst>(CS->getInstruction());
if (!llvm::getAlign(*CallI, 0, retAlignment))
retAlignment = getDataLayout()->getABITypeAlignment(retTy);
}
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain, DAG.getConstant(retAlignment,
MVT::i32),
DAG.getConstant(resultsz/8, MVT::i32),
DAG.getConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRetParam, dl, DeclareRetVTs,
DeclareRetOps, 5);
InFlag = Chain.getValue(1);
}
}
}
if (!Func) {
// This is indirect function call case : PTX requires a prototype of the
// form
// proto_0 : .callprototype(.param .b32 _) _ (.param .b32 _);
// to be emitted, and the label has to used as the last arg of call
// instruction.
// The prototype is embedded in a string and put as the operand for an
// INLINEASM SDNode.
SDVTList InlineAsmVTs = DAG.getVTList(MVT::Other, MVT::Glue);
std::string proto_string = getPrototype(retTy, Args, Outs, retAlignment);
const char *asmstr = nvTM->getManagedStrPool()->
getManagedString(proto_string.c_str())->c_str();
SDValue InlineAsmOps[] = { Chain,
DAG.getTargetExternalSymbol(asmstr,
getPointerTy()),
DAG.getMDNode(0),
DAG.getTargetConstant(0, MVT::i32), InFlag };
Chain = DAG.getNode(ISD::INLINEASM, dl, InlineAsmVTs, InlineAsmOps, 5);
InFlag = Chain.getValue(1);
}
// Op to just print "call"
SDVTList PrintCallVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue PrintCallOps[] = { Chain,
DAG.getConstant(isABI ? ((Ins.size()==0) ? 0 : 1)
: retCount, MVT::i32),
InFlag };
Chain = DAG.getNode(Func?(NVPTXISD::PrintCallUni):(NVPTXISD::PrintCall), dl,
PrintCallVTs, PrintCallOps, 3);
InFlag = Chain.getValue(1);
// Ops to print out the function name
SDVTList CallVoidVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallVoidOps[] = { Chain, Callee, InFlag };
Chain = DAG.getNode(NVPTXISD::CallVoid, dl, CallVoidVTs, CallVoidOps, 3);
InFlag = Chain.getValue(1);
// Ops to print out the param list
SDVTList CallArgBeginVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgBeginOps[] = { Chain, InFlag };
Chain = DAG.getNode(NVPTXISD::CallArgBegin, dl, CallArgBeginVTs,
CallArgBeginOps, 2);
InFlag = Chain.getValue(1);
for (unsigned i=0, e=paramCount; i!=e; ++i) {
unsigned opcode;
if (i==(e-1))
opcode = NVPTXISD::LastCallArg;
else
opcode = NVPTXISD::CallArg;
SDVTList CallArgVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgOps[] = { Chain, DAG.getConstant(1, MVT::i32),
DAG.getConstant(i, MVT::i32),
InFlag };
Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps, 4);
InFlag = Chain.getValue(1);
}
SDVTList CallArgEndVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CallArgEndOps[] = { Chain,
DAG.getConstant(Func ? 1 : 0, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps,
3);
InFlag = Chain.getValue(1);
if (!Func) {
SDVTList PrototypeVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue PrototypeOps[] = { Chain,
DAG.getConstant(uniqueCallSite, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps, 3);
InFlag = Chain.getValue(1);
}
// Generate loads from param memory/moves from registers for result
if (Ins.size() > 0) {
if (isABI) {
unsigned resoffset = 0;
for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
unsigned sz = Ins[i].VT.getSizeInBits();
if (Ins[i].VT.isInteger() && (sz < 8)) sz = 8;
std::vector<EVT> LoadRetVTs;
LoadRetVTs.push_back(Ins[i].VT);
LoadRetVTs.push_back(MVT::Other); LoadRetVTs.push_back(MVT::Glue);
std::vector<SDValue> LoadRetOps;
LoadRetOps.push_back(Chain);
LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
LoadRetOps.push_back(DAG.getConstant(resoffset, MVT::i32));
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, LoadRetVTs,
&LoadRetOps[0], LoadRetOps.size());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
InVals.push_back(retval);
resoffset += sz/8;
}
}
else {
SmallVector<EVT, 16> resvtparts;
ComputeValueVTs(*this, retTy, resvtparts);
assert(Ins.size() == resvtparts.size() &&
"Unexpected number of return values in non-ABI case");
unsigned paramNum = 0;
for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
assert(EVT(Ins[i].VT) == resvtparts[i] &&
"Unexpected EVT type in non-ABI case");
unsigned numelems = 1;
EVT elemtype = Ins[i].VT;
if (Ins[i].VT.isVector()) {
numelems = Ins[i].VT.getVectorNumElements();
elemtype = Ins[i].VT.getVectorElementType();
}
std::vector<SDValue> tempRetVals;
for (unsigned j=0; j<numelems; ++j) {
std::vector<EVT> MoveRetVTs;
MoveRetVTs.push_back(elemtype);
MoveRetVTs.push_back(MVT::Other); MoveRetVTs.push_back(MVT::Glue);
std::vector<SDValue> MoveRetOps;
MoveRetOps.push_back(Chain);
MoveRetOps.push_back(DAG.getConstant(0, MVT::i32));
MoveRetOps.push_back(DAG.getConstant(paramNum, MVT::i32));
MoveRetOps.push_back(InFlag);
SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, MoveRetVTs,
&MoveRetOps[0], MoveRetOps.size());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
tempRetVals.push_back(retval);
++paramNum;
}
if (Ins[i].VT.isVector())
InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, dl, Ins[i].VT,
&tempRetVals[0], tempRetVals.size()));
else
InVals.push_back(tempRetVals[0]);
}
}
}
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getIntPtrConstant(uniqueCallSite, true),
DAG.getIntPtrConstant(uniqueCallSite+1, true),
InFlag);
uniqueCallSite++;
// set isTailCall to false for now, until we figure out how to express
// tail call optimization in PTX
isTailCall = false;
return Chain;
}
std::pair<bool, SDNode*> MipsSEDAGToDAGISel::selectNode(SDNode *Node) {
unsigned Opcode = Node->getOpcode();
SDLoc DL(Node);
///
// Instruction Selection not handled by the auto-generated
// tablegen selection should be handled here.
///
SDNode *Result;
switch(Opcode) {
default: break;
case ISD::SUBE: {
SDValue InFlag = Node->getOperand(2);
Result = selectAddESubE(Mips::SUBu, InFlag, InFlag.getOperand(0), DL, Node);
return std::make_pair(true, Result);
}
case ISD::ADDE: {
if (Subtarget.hasDSP()) // Select DSP instructions, ADDSC and ADDWC.
break;
SDValue InFlag = Node->getOperand(2);
Result = selectAddESubE(Mips::ADDu, InFlag, InFlag.getValue(0), DL, Node);
return std::make_pair(true, Result);
}
case ISD::ConstantFP: {
ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(Node);
if (Node->getValueType(0) == MVT::f64 && CN->isExactlyValue(+0.0)) {
if (Subtarget.hasMips64()) {
SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL,
Mips::ZERO_64, MVT::i64);
Result = CurDAG->getMachineNode(Mips::DMTC1, DL, MVT::f64, Zero);
} else {
SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL,
Mips::ZERO, MVT::i32);
Result = CurDAG->getMachineNode(Mips::BuildPairF64, DL, MVT::f64, Zero,
Zero);
}
return std::make_pair(true, Result);
}
break;
}
case ISD::Constant: {
const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Node);
unsigned Size = CN->getValueSizeInBits(0);
if (Size == 32)
break;
MipsAnalyzeImmediate AnalyzeImm;
int64_t Imm = CN->getSExtValue();
const MipsAnalyzeImmediate::InstSeq &Seq =
AnalyzeImm.Analyze(Imm, Size, false);
MipsAnalyzeImmediate::InstSeq::const_iterator Inst = Seq.begin();
SDLoc DL(CN);
SDNode *RegOpnd;
SDValue ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
MVT::i64);
// The first instruction can be a LUi which is different from other
// instructions (ADDiu, ORI and SLL) in that it does not have a register
// operand.
if (Inst->Opc == Mips::LUi64)
RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, ImmOpnd);
else
RegOpnd =
CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
CurDAG->getRegister(Mips::ZERO_64, MVT::i64),
ImmOpnd);
// The remaining instructions in the sequence are handled here.
for (++Inst; Inst != Seq.end(); ++Inst) {
ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
MVT::i64);
RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
SDValue(RegOpnd, 0), ImmOpnd);
}
return std::make_pair(true, RegOpnd);
}
case ISD::INTRINSIC_W_CHAIN: {
switch (cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue()) {
default:
break;
case Intrinsic::mips_cfcmsa: {
SDValue ChainIn = Node->getOperand(0);
SDValue RegIdx = Node->getOperand(2);
SDValue Reg = CurDAG->getCopyFromReg(ChainIn, DL,
getMSACtrlReg(RegIdx), MVT::i32);
return std::make_pair(true, Reg.getNode());
}
}
break;
}
case ISD::INTRINSIC_WO_CHAIN: {
switch (cast<ConstantSDNode>(Node->getOperand(0))->getZExtValue()) {
default:
break;
case Intrinsic::mips_move_v:
// Like an assignment but will always produce a move.v even if
// unnecessary.
return std::make_pair(true,
CurDAG->getMachineNode(Mips::MOVE_V, DL,
Node->getValueType(0),
Node->getOperand(1)));
}
break;
}
case ISD::INTRINSIC_VOID: {
switch (cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue()) {
default:
break;
case Intrinsic::mips_ctcmsa: {
SDValue ChainIn = Node->getOperand(0);
SDValue RegIdx = Node->getOperand(2);
SDValue Value = Node->getOperand(3);
SDValue ChainOut = CurDAG->getCopyToReg(ChainIn, DL,
getMSACtrlReg(RegIdx), Value);
return std::make_pair(true, ChainOut.getNode());
}
}
break;
}
case MipsISD::ThreadPointer: {
EVT PtrVT = getTargetLowering()->getPointerTy();
unsigned RdhwrOpc, DestReg;
if (PtrVT == MVT::i32) {
RdhwrOpc = Mips::RDHWR;
DestReg = Mips::V1;
} else {
RdhwrOpc = Mips::RDHWR64;
DestReg = Mips::V1_64;
}
SDNode *Rdhwr =
CurDAG->getMachineNode(RdhwrOpc, SDLoc(Node),
Node->getValueType(0),
CurDAG->getRegister(Mips::HWR29, MVT::i32));
SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), DL, DestReg,
SDValue(Rdhwr, 0));
SDValue ResNode = CurDAG->getCopyFromReg(Chain, DL, DestReg, PtrVT);
ReplaceUses(SDValue(Node, 0), ResNode);
return std::make_pair(true, ResNode.getNode());
}
case ISD::BUILD_VECTOR: {
// Select appropriate ldi.[bhwd] instructions for constant splats of
// 128-bit when MSA is enabled. Fixup any register class mismatches that
// occur as a result.
//
// This allows the compiler to use a wider range of immediates than would
// otherwise be allowed. If, for example, v4i32 could only use ldi.h then
// it would not be possible to load { 0x01010101, 0x01010101, 0x01010101,
// 0x01010101 } without using a constant pool. This would be sub-optimal
// when // 'ldi.b wd, 1' is capable of producing that bit-pattern in the
// same set/ of registers. Similarly, ldi.h isn't capable of producing {
// 0x00000000, 0x00000001, 0x00000000, 0x00000001 } but 'ldi.d wd, 1' can.
BuildVectorSDNode *BVN = cast<BuildVectorSDNode>(Node);
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
unsigned LdiOp;
EVT ResVecTy = BVN->getValueType(0);
EVT ViaVecTy;
if (!Subtarget.hasMSA() || !BVN->getValueType(0).is128BitVector())
return std::make_pair(false, (SDNode*)NULL);
if (!BVN->isConstantSplat(SplatValue, SplatUndef, SplatBitSize,
HasAnyUndefs, 8,
!Subtarget.isLittle()))
return std::make_pair(false, (SDNode*)NULL);
switch (SplatBitSize) {
default:
return std::make_pair(false, (SDNode*)NULL);
case 8:
LdiOp = Mips::LDI_B;
ViaVecTy = MVT::v16i8;
break;
case 16:
LdiOp = Mips::LDI_H;
ViaVecTy = MVT::v8i16;
break;
case 32:
LdiOp = Mips::LDI_W;
ViaVecTy = MVT::v4i32;
break;
case 64:
LdiOp = Mips::LDI_D;
ViaVecTy = MVT::v2i64;
break;
}
if (!SplatValue.isSignedIntN(10))
return std::make_pair(false, (SDNode*)NULL);
SDValue Imm = CurDAG->getTargetConstant(SplatValue,
ViaVecTy.getVectorElementType());
SDNode *Res = CurDAG->getMachineNode(LdiOp, SDLoc(Node), ViaVecTy, Imm);
if (ResVecTy != ViaVecTy) {
// If LdiOp is writing to a different register class to ResVecTy, then
// fix it up here. This COPY_TO_REGCLASS should never cause a move.v
// since the source and destination register sets contain the same
// registers.
const TargetLowering *TLI = getTargetLowering();
MVT ResVecTySimple = ResVecTy.getSimpleVT();
const TargetRegisterClass *RC = TLI->getRegClassFor(ResVecTySimple);
Res = CurDAG->getMachineNode(Mips::COPY_TO_REGCLASS, SDLoc(Node),
ResVecTy, SDValue(Res, 0),
CurDAG->getTargetConstant(RC->getID(),
MVT::i32));
}
return std::make_pair(true, Res);
}
}
return std::make_pair(false, (SDNode*)NULL);
}
SDValue BPFTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
auto &Outs = CLI.Outs;
auto &OutVals = CLI.OutVals;
auto &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &IsTailCall = CLI.IsTailCall;
CallingConv::ID CallConv = CLI.CallConv;
bool IsVarArg = CLI.IsVarArg;
MachineFunction &MF = DAG.getMachineFunction();
// BPF target does not support tail call optimization.
IsTailCall = false;
switch (CallConv) {
default:
report_fatal_error("Unsupported calling convention");
case CallingConv::Fast:
case CallingConv::C:
break;
}
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
CCInfo.AnalyzeCallOperands(Outs, CC_BPF64);
unsigned NumBytes = CCInfo.getNextStackOffset();
if (Outs.size() >= 6) {
DiagnosticInfoUnsupported Err(CLI.DL, *MF.getFunction(),
"too many args to ", Callee);
DAG.getContext()->diagnose(Err);
}
for (auto &Arg : Outs) {
ISD::ArgFlagsTy Flags = Arg.Flags;
if (!Flags.isByVal())
continue;
DiagnosticInfoUnsupported Err(CLI.DL, *MF.getFunction(),
"pass by value not supported ", Callee);
DAG.getContext()->diagnose(Err);
}
Chain = DAG.getCALLSEQ_START(
Chain, DAG.getConstant(NumBytes, getPointerTy(), true), CLI.DL);
SmallVector<std::pair<unsigned, SDValue>, 5> RegsToPass;
// Walk arg assignments
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[i];
// Promote the value if needed.
switch (VA.getLocInfo()) {
default:
llvm_unreachable("Unknown loc info");
case CCValAssign::Full:
break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, CLI.DL, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, CLI.DL, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, CLI.DL, VA.getLocVT(), Arg);
break;
}
// Push arguments into RegsToPass vector
if (VA.isRegLoc())
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
else
llvm_unreachable("call arg pass bug");
}
SDValue InFlag;
// Build a sequence of copy-to-reg nodes chained together with token chain and
// flag operands which copy the outgoing args into registers. The InFlag in
// necessary since all emitted instructions must be stuck together.
for (auto &Reg : RegsToPass) {
Chain = DAG.getCopyToReg(Chain, CLI.DL, Reg.first, Reg.second, InFlag);
InFlag = Chain.getValue(1);
}
// If the callee is a GlobalAddress node (quite common, every direct call is)
// turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
// Likewise ExternalSymbol -> TargetExternalSymbol.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), CLI.DL, getPointerTy(),
G->getOffset(), 0);
else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(E->getSymbol(), getPointerTy(), 0);
// Returns a chain & a flag for retval copy to use.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
Ops.push_back(Callee);
// Add argument registers to the end of the list so that they are
// known live into the call.
for (auto &Reg : RegsToPass)
Ops.push_back(DAG.getRegister(Reg.first, Reg.second.getValueType()));
if (InFlag.getNode())
Ops.push_back(InFlag);
Chain = DAG.getNode(BPFISD::CALL, CLI.DL, NodeTys, Ops);
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(
Chain, DAG.getConstant(NumBytes, getPointerTy(), true),
DAG.getConstant(0, getPointerTy(), true), InFlag, CLI.DL);
InFlag = Chain.getValue(1);
// Handle result values, copying them out of physregs into vregs that we
// return.
return LowerCallResult(Chain, InFlag, CallConv, IsVarArg, Ins, CLI.DL, DAG,
InVals);
}
static bool isIntS32Immediate(SDValue Op, int32_t &Imm) {
return isIntS32Immediate(Op.getNode(), Imm);
}
SDValue
BlackfinTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// Blackfin target does not yet support tail call optimization.
isTailCall = false;
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
DAG.getTarget(), ArgLocs, *DAG.getContext());
CCInfo.AllocateStack(12, 4); // ABI requires 12 bytes stack space
CCInfo.AnalyzeCallOperands(Outs, CC_Blackfin);
// Get the size of the outgoing arguments stack space requirement.
unsigned ArgsSize = CCInfo.getNextStackOffset();
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(ArgsSize, true));
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[i];
// Promote the value if needed.
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
break;
}
// Arguments that can be passed on register must be kept at
// RegsToPass vector
if (VA.isRegLoc()) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
assert(VA.isMemLoc() && "CCValAssign must be RegLoc or MemLoc");
int Offset = VA.getLocMemOffset();
assert(Offset%4 == 0 && "Unaligned LocMemOffset");
assert(VA.getLocVT()==MVT::i32 && "Illegal CCValAssign type");
SDValue SPN = DAG.getCopyFromReg(Chain, dl, BF::SP, MVT::i32);
SDValue OffsetN = DAG.getIntPtrConstant(Offset);
OffsetN = DAG.getNode(ISD::ADD, dl, MVT::i32, SPN, OffsetN);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, OffsetN,
MachinePointerInfo(),false, false, 0));
}
}
// Transform all store nodes into one single node because
// all store nodes are independent of each other.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// Build a sequence of copy-to-reg nodes chained together with token
// chain and flag operands which copy the outgoing args into registers.
// The InFlag in necessary since all emitted instructions must be
// stuck together.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
// If the callee is a GlobalAddress node (quite common, every direct call is)
// turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
// Likewise ExternalSymbol -> TargetExternalSymbol.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i32);
else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i32);
std::vector<EVT> NodeTys;
NodeTys.push_back(MVT::Other); // Returns a chain
NodeTys.push_back(MVT::Glue); // Returns a flag for retval copy to use.
SDValue Ops[] = { Chain, Callee, InFlag };
Chain = DAG.getNode(BFISD::CALL, dl, NodeTys, Ops,
InFlag.getNode() ? 3 : 2);
InFlag = Chain.getValue(1);
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(ArgsSize, true),
DAG.getIntPtrConstant(0, true), InFlag);
InFlag = Chain.getValue(1);
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
CCState RVInfo(CallConv, isVarArg, DAG.getMachineFunction(),
DAG.getTarget(), RVLocs, *DAG.getContext());
RVInfo.AnalyzeCallResult(Ins, RetCC_Blackfin);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &RV = RVLocs[i];
unsigned Reg = RV.getLocReg();
Chain = DAG.getCopyFromReg(Chain, dl, Reg,
RVLocs[i].getLocVT(), InFlag);
SDValue Val = Chain.getValue(0);
InFlag = Chain.getValue(2);
Chain = Chain.getValue(1);
// Callee is responsible for extending any i16 return values.
switch (RV.getLocInfo()) {
case CCValAssign::SExt:
Val = DAG.getNode(ISD::AssertSext, dl, RV.getLocVT(), Val,
DAG.getValueType(RV.getValVT()));
break;
case CCValAssign::ZExt:
Val = DAG.getNode(ISD::AssertZext, dl, RV.getLocVT(), Val,
DAG.getValueType(RV.getValVT()));
break;
default:
break;
}
// Truncate to valtype
if (RV.getLocInfo() != CCValAssign::Full)
Val = DAG.getNode(ISD::TRUNCATE, dl, RV.getValVT(), Val);
InVals.push_back(Val);
}
return Chain;
}
