/// LowerCallResult - Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
///
SDValue
AlphaTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(), RVLocs,
*DAG.getContext());
CCInfo.AnalyzeCallResult(Ins, RetCC_Alpha);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
VA.getLocVT(), InFlag).getValue(1);
SDValue RetValue = Chain.getValue(0);
InFlag = Chain.getValue(2);
// If this is an 8/16/32-bit value, it is really passed promoted to 64
// bits. Insert an assert[sz]ext to capture this, then truncate to the
// right size.
if (VA.getLocInfo() == CCValAssign::SExt)
RetValue = DAG.getNode(ISD::AssertSext, dl, VA.getLocVT(), RetValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::ZExt)
RetValue = DAG.getNode(ISD::AssertZext, dl, VA.getLocVT(), RetValue,
DAG.getValueType(VA.getValVT()));
if (VA.getLocInfo() != CCValAssign::Full)
RetValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), RetValue);
InVals.push_back(RetValue);
}
return Chain;
}
// There is no native floating point division, but we can convert this to a
// reciprocal/multiply operation. If the first parameter is constant 1.0, then
// just a reciprocal will suffice.
SDValue
VectorProcTargetLowering::LowerFDIV(SDValue Op, SelectionDAG &DAG) const
{
DebugLoc dl = Op.getDebugLoc();
EVT type = Op.getOperand(1).getValueType();
SDValue two = DAG.getConstantFP(2.0, type);
SDValue denom = Op.getOperand(1);
SDValue estimate = DAG.getNode(VectorProcISD::RECIPROCAL_EST, dl, type, denom);
// Perform a series of newton/raphson refinements. Each iteration doubles
// the precision. The initial estimate has 6 bits of precision, so two iteration
// results in 24 bits, which is larger than the significand.
for (int i = 0; i < 2; i++)
{
// trial = x * estimate (ideally, x * 1/x should be 1.0)
// error = 2.0 - trial
// estimate = estimate * error
SDValue trial = DAG.getNode(ISD::FMUL, dl, type, estimate, denom);
SDValue error = DAG.getNode(ISD::FSUB, dl, type, two, trial);
estimate = DAG.getNode(ISD::FMUL, dl, type, estimate, error);
}
// Check if the first parameter is constant 1.0. If so, we don't need
// to multiply.
bool isOne = false;
if (type.isVector())
{
if (isSplatVector(Op.getOperand(0).getNode()))
{
ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Op.getOperand(0).getOperand(0));
isOne = C && C->isExactlyValue(1.0);
}
}
else
{
ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Op.getOperand(0));
isOne = C && C->isExactlyValue(1.0);
}
if (!isOne)
estimate = DAG.getNode(ISD::FMUL, dl, type, Op.getOperand(0), estimate);
return estimate;
}
// After instruction selection, insert COPY_TO_REGCLASS nodes to help in
// choosing the proper register classes.
void BlackfinDAGToDAGISel::FixRegisterClasses(SelectionDAG &DAG) {
const BlackfinInstrInfo &TII = getInstrInfo();
const BlackfinRegisterInfo *TRI = getRegisterInfo();
DAG.AssignTopologicalOrder();
HandleSDNode Dummy(DAG.getRoot());
for (SelectionDAG::allnodes_iterator NI = DAG.allnodes_begin();
NI != DAG.allnodes_end(); ++NI) {
if (NI->use_empty() || !NI->isMachineOpcode())
continue;
const TargetInstrDesc &DefTID = TII.get(NI->getMachineOpcode());
for (SDNode::use_iterator UI = NI->use_begin(); !UI.atEnd(); ++UI) {
if (!UI->isMachineOpcode())
continue;
if (UI.getUse().getResNo() >= DefTID.getNumDefs())
continue;
const TargetRegisterClass *DefRC =
DefTID.OpInfo[UI.getUse().getResNo()].getRegClass(TRI);
const TargetInstrDesc &UseTID = TII.get(UI->getMachineOpcode());
if (UseTID.getNumDefs()+UI.getOperandNo() >= UseTID.getNumOperands())
continue;
const TargetRegisterClass *UseRC =
UseTID.OpInfo[UseTID.getNumDefs()+UI.getOperandNo()].getRegClass(TRI);
if (!DefRC || !UseRC)
continue;
// We cannot copy CC <-> !(CC/D)
if ((isCC(DefRC) && !isDCC(UseRC)) || (isCC(UseRC) && !isDCC(DefRC))) {
SDNode *Copy =
DAG.getMachineNode(TargetOpcode::COPY_TO_REGCLASS,
NI->getDebugLoc(),
MVT::i32,
UI.getUse().get(),
DAG.getTargetConstant(BF::DRegClassID, MVT::i32));
UpdateNodeOperand(DAG, *UI, UI.getOperandNo(), SDValue(Copy, 0));
}
}
}
DAG.setRoot(Dummy.getValue());
}
static SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) {
SDValue Chain = Op.getOperand(0); // Legalize the chain.
SDValue Size = Op.getOperand(1); // Legalize the size.
DebugLoc dl = Op.getDebugLoc();
unsigned SPReg = SP::O6;
SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, MVT::i32);
SDValue NewSP = DAG.getNode(ISD::SUB, dl, MVT::i32, SP, Size); // Value
Chain = DAG.getCopyToReg(SP.getValue(1), dl, SPReg, NewSP); // Output chain
// The resultant pointer is actually 16 words from the bottom of the stack,
// to provide a register spill area.
SDValue NewVal = DAG.getNode(ISD::ADD, dl, MVT::i32, NewSP,
DAG.getConstant(96, MVT::i32));
SDValue Ops[2] = { NewVal, Chain };
return DAG.getMergeValues(Ops, 2, dl);
}
unsigned
SparcTargetLowering::getSRetArgSize(SelectionDAG &DAG, SDValue Callee) const
{
const Function *CalleeFn = 0;
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
CalleeFn = dyn_cast<Function>(G->getGlobal());
} else if (ExternalSymbolSDNode *E =
dyn_cast<ExternalSymbolSDNode>(Callee)) {
const Function *Fn = DAG.getMachineFunction().getFunction();
const Module *M = Fn->getParent();
CalleeFn = M->getFunction(E->getSymbol());
}
if (!CalleeFn)
return 0;
assert(CalleeFn->hasStructRetAttr() &&
"Callee does not have the StructRet attribute.");
const PointerType *Ty = cast<PointerType>(CalleeFn->arg_begin()->getType());
const Type *ElementTy = Ty->getElementType();
return getTargetData()->getTypeAllocSize(ElementTy);
}
// Use CLC to compare [Src1, Src1 + Size) with [Src2, Src2 + Size),
// deciding whether to use a loop or straight-line code.
static SDValue emitCLC(SelectionDAG &DAG, SDLoc DL, SDValue Chain,
SDValue Src1, SDValue Src2, uint64_t Size) {
SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
EVT PtrVT = Src1.getValueType();
// A two-CLC sequence is a clear win over a loop, not least because it
// needs only one branch. A three-CLC sequence needs the same number
// of branches as a loop (i.e. 2), but is shorter. That brings us to
// lengths greater than 768 bytes. It seems relatively likely that
// a difference will be found within the first 768 bytes, so we just
// optimize for the smallest number of branch instructions, in order
// to avoid polluting the prediction buffer too much. A loop only ever
// needs 2 branches, whereas a straight-line sequence would need 3 or more.
if (Size > 3 * 256)
return DAG.getNode(SystemZISD::CLC_LOOP, DL, VTs, Chain, Src1, Src2,
DAG.getConstant(Size, PtrVT),
DAG.getConstant(Size / 256, PtrVT));
return DAG.getNode(SystemZISD::CLC, DL, VTs, Chain, Src1, Src2,
DAG.getConstant(Size, PtrVT));
}
SDValue MSP430TargetLowering::LowerRETURNADDR(SDValue Op,
SelectionDAG &DAG) const {
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
MFI->setReturnAddressIsTaken(true);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
DebugLoc dl = Op.getDebugLoc();
if (Depth > 0) {
SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
SDValue Offset =
DAG.getConstant(TD->getPointerSize(), MVT::i16);
return DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, dl, getPointerTy(),
FrameAddr, Offset),
MachinePointerInfo(), false, false, 0);
}
// Just load the return address.
SDValue RetAddrFI = getReturnAddressFrameIndex(DAG);
return DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
RetAddrFI, MachinePointerInfo(), false, false, 0);
}
SDValue LanaiTargetLowering::LowerRETURNADDR(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setReturnAddressIsTaken(true);
EVT VT = Op.getValueType();
SDLoc DL(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
if (Depth) {
SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
const unsigned Offset = -4;
SDValue Ptr = DAG.getNode(ISD::ADD, DL, VT, FrameAddr,
DAG.getIntPtrConstant(Offset, DL));
return DAG.getLoad(VT, DL, DAG.getEntryNode(), Ptr, MachinePointerInfo(),
false, false, false, 0);
}
// Return the link register, which contains the return address.
// Mark it an implicit live-in.
unsigned Reg = MF.addLiveIn(TRI->getRARegister(), getRegClassFor(MVT::i32));
return DAG.getCopyFromReg(DAG.getEntryNode(), DL, Reg, VT);
}
SDValue SystemZTargetLowering::LowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
int64_t Offset = cast<GlobalAddressSDNode>(Op)->getOffset();
bool IsPic = getTargetMachine().getRelocationModel() == Reloc::PIC_;
bool ExtraLoadRequired =
Subtarget.GVRequiresExtraLoad(GV, getTargetMachine(), false);
SDValue Result;
if (!IsPic && !ExtraLoadRequired) {
Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), Offset);
Offset = 0;
} else {
unsigned char OpFlags = 0;
if (ExtraLoadRequired)
OpFlags = SystemZII::MO_GOTENT;
Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), 0, OpFlags);
}
Result = DAG.getNode(SystemZISD::PCRelativeWrapper, dl,
getPointerTy(), Result);
if (ExtraLoadRequired)
Result = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Result,
PseudoSourceValue::getGOT(), 0, false, false, 0);
// If there was a non-zero offset that we didn't fold, create an explicit
// addition for it.
if (Offset != 0)
Result = DAG.getNode(ISD::ADD, dl, getPointerTy(), Result,
DAG.getConstant(Offset, getPointerTy()));
return Result;
}
SDValue XCoreSelectionDAGInfo::EmitTargetCodeForMemcpy(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, bool isVolatile, bool AlwaysInline,
MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
unsigned SizeBitWidth = Size.getValueType().getSizeInBits();
// Call __memcpy_4 if the src, dst and size are all 4 byte aligned.
if (!AlwaysInline && (Align & 3) == 0 &&
DAG.MaskedValueIsZero(Size, APInt(SizeBitWidth, 3))) {
const TargetLowering &TLI = *DAG.getSubtarget().getTargetLowering();
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
Entry.Node = Dst;
Args.push_back(Entry);
Entry.Node = Src;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setCallee(TLI.getLibcallCallingConv(RTLIB::MEMCPY),
Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol("__memcpy_4",
TLI.getPointerTy(DAG.getDataLayout())),
std::move(Args))
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memcpy.
return SDValue();
}
SDValue HexagonSelectionDAGInfo::EmitTargetCodeForMemcpy(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, bool isVolatile, bool AlwaysInline,
MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
if (AlwaysInline || (Align & 0x3) != 0 || !ConstantSize)
return SDValue();
uint64_t SizeVal = ConstantSize->getZExtValue();
if (SizeVal < 32 || (SizeVal % 8) != 0)
return SDValue();
// Special case aligned memcpys with size >= 32 bytes and a multiple of 8.
//
const TargetLowering &TLI = *DAG.getSubtarget().getTargetLowering();
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
Entry.Node = Dst;
Args.push_back(Entry);
Entry.Node = Src;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
const char *SpecialMemcpyName =
"__hexagon_memcpy_likely_aligned_min32bytes_mult8bytes";
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setCallee(TLI.getLibcallCallingConv(RTLIB::MEMCPY),
Type::getVoidTy(*DAG.getContext()),
DAG.getTargetExternalSymbol(
SpecialMemcpyName, TLI.getPointerTy(DAG.getDataLayout())),
std::move(Args))
.setDiscardResult();
std::pair<SDValue, SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
SDValue MipsTargetLowering::
LowerJumpTable(SDValue Op, SelectionDAG &DAG)
{
SDValue ResNode;
SDValue HiPart;
// FIXME there isn't actually debug info here
DebugLoc dl = Op.getDebugLoc();
EVT PtrVT = Op.getValueType();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
if (getTargetMachine().getRelocationModel() != Reloc::PIC_) {
SDVTList VTs = DAG.getVTList(MVT::i32);
SDValue Ops[] = { JTI };
HiPart = DAG.getNode(MipsISD::Hi, dl, VTs, Ops, 1);
} else // Emit Load from Global Pointer
HiPart = DAG.getLoad(MVT::i32, dl, DAG.getEntryNode(), JTI, NULL, 0);
SDValue Lo = DAG.getNode(MipsISD::Lo, dl, MVT::i32, JTI);
ResNode = DAG.getNode(ISD::ADD, dl, MVT::i32, HiPart, Lo);
return ResNode;
}
SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc,
bool HasLo, bool HasHi,
SelectionDAG &DAG) const {
EVT Ty = Op.getOperand(0).getValueType();
DebugLoc DL = Op.getDebugLoc();
SDValue Mult = DAG.getNode(NewOpc, DL, MVT::Untyped,
Op.getOperand(0), Op.getOperand(1));
SDValue Lo, Hi;
if (HasLo)
Lo = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult,
DAG.getConstant(Mips::sub_lo, MVT::i32));
if (HasHi)
Hi = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult,
DAG.getConstant(Mips::sub_hi, MVT::i32));
if (!HasLo || !HasHi)
return HasLo ? Lo : Hi;
SDValue Vals[] = { Lo, Hi };
return DAG.getMergeValues(Vals, 2, DL);
}
static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG& DAG,
TargetLowering::DAGCombinerInfo &DCI,
const Cpu0Subtarget* Subtarget) {
if (DCI.isBeforeLegalizeOps())
return SDValue();
EVT Ty = N->getValueType(0);
unsigned LO = Cpu0::LO;
unsigned HI = Cpu0::HI;
unsigned opc = N->getOpcode() == ISD::SDIVREM ? Cpu0ISD::DivRem :
Cpu0ISD::DivRemU;
SDLoc DL(N);
SDValue DivRem = DAG.getNode(opc, DL, MVT::Glue,
N->getOperand(0), N->getOperand(1));
SDValue InChain = DAG.getEntryNode();
SDValue InGlue = DivRem;
// insert MFLO
if (N->hasAnyUseOfValue(0)) {
SDValue CopyFromLo = DAG.getCopyFromReg(InChain, DL, LO, Ty,
InGlue);
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), CopyFromLo);
InChain = CopyFromLo.getValue(1);
InGlue = CopyFromLo.getValue(2);
}
// insert MFHI
if (N->hasAnyUseOfValue(1)) {
SDValue CopyFromHi = DAG.getCopyFromReg(InChain, DL,
HI, Ty, InGlue);
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), CopyFromHi);
}
return SDValue();
}
SDValue SystemZTargetLowering::EmitCmp(SDValue LHS, SDValue RHS,
ISD::CondCode CC, SDValue &SystemZCC,
SelectionDAG &DAG) {
// FIXME: Emit a test if RHS is zero
bool isUnsigned = false;
SystemZCC::CondCodes TCC;
switch (CC) {
default:
llvm_unreachable("Invalid integer condition!");
case ISD::SETEQ:
case ISD::SETOEQ:
TCC = SystemZCC::E;
break;
case ISD::SETUEQ:
TCC = SystemZCC::NLH;
break;
case ISD::SETNE:
case ISD::SETONE:
TCC = SystemZCC::NE;
break;
case ISD::SETUNE:
TCC = SystemZCC::LH;
break;
case ISD::SETO:
TCC = SystemZCC::O;
break;
case ISD::SETUO:
TCC = SystemZCC::NO;
break;
case ISD::SETULE:
if (LHS.getValueType().isFloatingPoint()) {
TCC = SystemZCC::NH;
break;
}
isUnsigned = true; // FALLTHROUGH
case ISD::SETLE:
case ISD::SETOLE:
TCC = SystemZCC::LE;
break;
case ISD::SETUGE:
if (LHS.getValueType().isFloatingPoint()) {
TCC = SystemZCC::NL;
break;
}
isUnsigned = true; // FALLTHROUGH
case ISD::SETGE:
case ISD::SETOGE:
TCC = SystemZCC::HE;
break;
case ISD::SETUGT:
if (LHS.getValueType().isFloatingPoint()) {
TCC = SystemZCC::NLE;
break;
}
isUnsigned = true; // FALLTHROUGH
case ISD::SETGT:
case ISD::SETOGT:
TCC = SystemZCC::H;
break;
case ISD::SETULT:
if (LHS.getValueType().isFloatingPoint()) {
TCC = SystemZCC::NHE;
break;
}
isUnsigned = true; // FALLTHROUGH
case ISD::SETLT:
case ISD::SETOLT:
TCC = SystemZCC::L;
break;
}
SystemZCC = DAG.getConstant(TCC, MVT::i32);
DebugLoc dl = LHS.getDebugLoc();
return DAG.getNode((isUnsigned ? SystemZISD::UCMP : SystemZISD::CMP),
dl, MVT::i64, LHS, RHS);
}
/// LowerCCCArguments - transform physical registers into virtual registers and
/// generate load operations for arguments places on the stack.
// FIXME: struct return stuff
// FIXME: varargs
SDValue
SystemZTargetLowering::LowerCCCArguments(SDValue Chain,
CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl,
SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_SystemZ);
if (isVarArg)
report_fatal_error("Varargs not supported yet");
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
SDValue ArgValue;
CCValAssign &VA = ArgLocs[i];
EVT LocVT = VA.getLocVT();
if (VA.isRegLoc()) {
// Arguments passed in registers
TargetRegisterClass *RC;
switch (LocVT.getSimpleVT().SimpleTy) {
default:
#ifndef NDEBUG
errs() << "LowerFormalArguments Unhandled argument type: "
<< LocVT.getSimpleVT().SimpleTy
<< "\n";
#endif
llvm_unreachable(0);
case MVT::i64:
RC = SystemZ::GR64RegisterClass;
break;
case MVT::f32:
RC = SystemZ::FP32RegisterClass;
break;
case MVT::f64:
RC = SystemZ::FP64RegisterClass;
break;
}
unsigned VReg = RegInfo.createVirtualRegister(RC);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
} else {
// Sanity check
assert(VA.isMemLoc());
// Create the nodes corresponding to a load from this parameter slot.
// Create the frame index object for this incoming parameter...
int FI = MFI->CreateFixedObject(LocVT.getSizeInBits()/8,
VA.getLocMemOffset(), true, false);
// Create the SelectionDAG nodes corresponding to a load
// from this parameter
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue = DAG.getLoad(LocVT, dl, Chain, FIN,
PseudoSourceValue::getFixedStack(FI), 0,
false, false, 0);
}
// If this is an 8/16/32-bit value, it is really passed promoted to 64
// bits. Insert an assert[sz]ext to capture this, then truncate to the
// right size.
if (VA.getLocInfo() == CCValAssign::SExt)
ArgValue = DAG.getNode(ISD::AssertSext, dl, LocVT, ArgValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, dl, LocVT, ArgValue,
DAG.getValueType(VA.getValVT()));
if (VA.getLocInfo() != CCValAssign::Full)
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
InVals.push_back(ArgValue);
}
return Chain;
}
SDValue BPFTargetLowering::LowerFormalArguments(
SDValue Chain, CallingConv::ID CallConv, bool IsVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, SDLoc DL, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
switch (CallConv) {
default:
llvm_unreachable("Unsupported calling convention");
case CallingConv::C:
case CallingConv::Fast:
break;
}
MachineFunction &MF = DAG.getMachineFunction();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_BPF64);
for (auto &VA : ArgLocs) {
if (VA.isRegLoc()) {
// Arguments passed in registers
EVT RegVT = VA.getLocVT();
switch (RegVT.getSimpleVT().SimpleTy) {
default: {
errs() << "LowerFormalArguments Unhandled argument type: "
<< RegVT.getSimpleVT().SimpleTy << '\n';
llvm_unreachable(0);
}
case MVT::i64:
unsigned VReg = RegInfo.createVirtualRegister(&BPF::GPRRegClass);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, VReg, RegVT);
// If this is an 8/16/32-bit value, it is really passed promoted to 64
// bits. Insert an assert[sz]ext to capture this, then truncate to the
// right size.
if (VA.getLocInfo() == CCValAssign::SExt)
ArgValue = DAG.getNode(ISD::AssertSext, DL, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, DL, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
if (VA.getLocInfo() != CCValAssign::Full)
ArgValue = DAG.getNode(ISD::TRUNCATE, DL, VA.getValVT(), ArgValue);
InVals.push_back(ArgValue);
}
} else {
DiagnosticInfoUnsupported Err(DL, *MF.getFunction(),
"defined with too many args", SDValue());
DAG.getContext()->diagnose(Err);
}
}
if (IsVarArg || MF.getFunction()->hasStructRetAttr()) {
DiagnosticInfoUnsupported Err(
DL, *MF.getFunction(),
"functions with VarArgs or StructRet are not supported", SDValue());
DAG.getContext()->diagnose(Err);
}
return Chain;
}
// Emit, if possible, a specialized version of the given Libcall. Typically this
// means selecting the appropriately aligned version, but we also convert memset
// of 0 into memclr.
SDValue ARMSelectionDAGInfo::EmitSpecializedLibcall(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, RTLIB::Libcall LC) const {
const ARMSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<ARMSubtarget>();
const ARMTargetLowering *TLI = Subtarget.getTargetLowering();
// Only use a specialized AEABI function if the default version of this
// Libcall is an AEABI function.
if (std::strncmp(TLI->getLibcallName(LC), "__aeabi", 7) != 0)
return SDValue();
// Translate RTLIB::Libcall to AEABILibcall. We only do this in order to be
// able to translate memset to memclr and use the value to index the function
// name array.
enum {
AEABI_MEMCPY = 0,
AEABI_MEMMOVE,
AEABI_MEMSET,
AEABI_MEMCLR
} AEABILibcall;
switch (LC) {
case RTLIB::MEMCPY:
AEABILibcall = AEABI_MEMCPY;
break;
case RTLIB::MEMMOVE:
AEABILibcall = AEABI_MEMMOVE;
break;
case RTLIB::MEMSET:
AEABILibcall = AEABI_MEMSET;
if (ConstantSDNode *ConstantSrc = dyn_cast<ConstantSDNode>(Src))
if (ConstantSrc->getZExtValue() == 0)
AEABILibcall = AEABI_MEMCLR;
break;
default:
return SDValue();
}
// Choose the most-aligned libcall variant that we can
enum {
ALIGN1 = 0,
ALIGN4,
ALIGN8
} AlignVariant;
if ((Align & 7) == 0)
AlignVariant = ALIGN8;
else if ((Align & 3) == 0)
AlignVariant = ALIGN4;
else
AlignVariant = ALIGN1;
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
Entry.Node = Dst;
Args.push_back(Entry);
if (AEABILibcall == AEABI_MEMCLR) {
Entry.Node = Size;
Args.push_back(Entry);
} else if (AEABILibcall == AEABI_MEMSET) {
// Adjust parameters for memset, EABI uses format (ptr, size, value),
// GNU library uses (ptr, value, size)
// See RTABI section 4.3.4
Entry.Node = Size;
Args.push_back(Entry);
// Extend or truncate the argument to be an i32 value for the call.
if (Src.getValueType().bitsGT(MVT::i32))
Src = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
else if (Src.getValueType().bitsLT(MVT::i32))
Src = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
Entry.Node = Src;
Entry.Ty = Type::getInt32Ty(*DAG.getContext());
Entry.IsSExt = false;
Args.push_back(Entry);
} else {
Entry.Node = Src;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
}
char const *FunctionNames[4][3] = {
{ "__aeabi_memcpy", "__aeabi_memcpy4", "__aeabi_memcpy8" },
{ "__aeabi_memmove", "__aeabi_memmove4", "__aeabi_memmove8" },
{ "__aeabi_memset", "__aeabi_memset4", "__aeabi_memset8" },
{ "__aeabi_memclr", "__aeabi_memclr4", "__aeabi_memclr8" }
};
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setLibCallee(
TLI->getLibcallCallingConv(LC), Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(FunctionNames[AEABILibcall][AlignVariant],
TLI->getPointerTy(DAG.getDataLayout())),
std::move(Args))
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI->LowerCallTo(CLI);
return CallResult.second;
}
SDValue MSP430TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
DebugLoc dl = Op.getDebugLoc();
// If we are doing an AND and testing against zero, then the CMP
// will not be generated. The AND (or BIT) will generate the condition codes,
// but they are different from CMP.
// FIXME: since we're doing a post-processing, use a pseudoinstr here, so
// lowering & isel wouldn't diverge.
bool andCC = false;
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
if (RHSC->isNullValue() && LHS.hasOneUse() &&
(LHS.getOpcode() == ISD::AND ||
(LHS.getOpcode() == ISD::TRUNCATE &&
LHS.getOperand(0).getOpcode() == ISD::AND))) {
andCC = true;
}
}
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
SDValue TargetCC;
SDValue Flag = EmitCMP(LHS, RHS, TargetCC, CC, dl, DAG);
// Get the condition codes directly from the status register, if its easy.
// Otherwise a branch will be generated. Note that the AND and BIT
// instructions generate different flags than CMP, the carry bit can be used
// for NE/EQ.
bool Invert = false;
bool Shift = false;
bool Convert = true;
switch (cast<ConstantSDNode>(TargetCC)->getZExtValue()) {
default:
Convert = false;
break;
case MSP430CC::COND_HS:
// Res = SRW & 1, no processing is required
break;
case MSP430CC::COND_LO:
// Res = ~(SRW & 1)
Invert = true;
break;
case MSP430CC::COND_NE:
if (andCC) {
// C = ~Z, thus Res = SRW & 1, no processing is required
} else {
// Res = ~((SRW >> 1) & 1)
Shift = true;
Invert = true;
}
break;
case MSP430CC::COND_E:
Shift = true;
// C = ~Z for AND instruction, thus we can put Res = ~(SRW & 1), however,
// Res = (SRW >> 1) & 1 is 1 word shorter.
break;
}
EVT VT = Op.getValueType();
SDValue One = DAG.getConstant(1, VT);
if (Convert) {
SDValue SR = DAG.getCopyFromReg(DAG.getEntryNode(), dl, MSP430::SRW,
MVT::i16, Flag);
if (Shift)
// FIXME: somewhere this is turned into a SRL, lower it MSP specific?
SR = DAG.getNode(ISD::SRA, dl, MVT::i16, SR, One);
SR = DAG.getNode(ISD::AND, dl, MVT::i16, SR, One);
if (Invert)
SR = DAG.getNode(ISD::XOR, dl, MVT::i16, SR, One);
return SR;
} else {
SDValue Zero = DAG.getConstant(0, VT);
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Glue);
SmallVector<SDValue, 4> Ops;
Ops.push_back(One);
Ops.push_back(Zero);
Ops.push_back(TargetCC);
Ops.push_back(Flag);
return DAG.getNode(MSP430ISD::SELECT_CC, dl, VTs, &Ops[0], Ops.size());
}
}
/// 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
BlackfinTargetLowering::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.getMachineFunction(),
DAG.getTarget(), RVLocs, *DAG.getContext());
// Analize return values.
CCInfo.AnalyzeReturn(Outs, RetCC_Blackfin);
// 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)
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!");
SDValue Opi = OutVals[i];
// Expand to i32 if necessary
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
Opi = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Opi);
break;
case CCValAssign::ZExt:
Opi = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Opi);
break;
case CCValAssign::AExt:
Opi = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Opi);
break;
}
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Opi, SDValue());
// Guarantee that all emitted copies are stuck together with flags.
Flag = Chain.getValue(1);
}
if (Flag.getNode()) {
return DAG.getNode(BFISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
} else {
return DAG.getNode(BFISD::RET_FLAG, dl, MVT::Other, Chain);
}
}
SDValue
BlackfinTargetLowering::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();
MachineFrameInfo *MFI = MF.getFrameInfo();
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AllocateStack(12, 4); // ABI requires 12 bytes stack space
CCInfo.AnalyzeFormalArguments(Ins, CC_Blackfin);
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (VA.isRegLoc()) {
EVT RegVT = VA.getLocVT();
TargetRegisterClass *RC = VA.getLocReg() == BF::P0 ?
BF::PRegisterClass : BF::DRegisterClass;
assert(RC->contains(VA.getLocReg()) && "Unexpected regclass in CCState");
assert(RC->hasType(RegVT) && "Unexpected regclass in CCState");
unsigned Reg = MF.getRegInfo().createVirtualRegister(RC);
MF.getRegInfo().addLiveIn(VA.getLocReg(), Reg);
SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
// If this is an 8 or 16-bit value, it is really passed promoted to 32
// bits. Insert an assert[sz]ext to capture this, then truncate to the
// right size.
if (VA.getLocInfo() == CCValAssign::SExt)
ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
if (VA.getLocInfo() != CCValAssign::Full)
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
InVals.push_back(ArgValue);
} else {
assert(VA.isMemLoc() && "CCValAssign must be RegLoc or MemLoc");
unsigned ObjSize = VA.getLocVT().getStoreSize();
int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset(), true);
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
MachinePointerInfo(),
false, false, 0));
}
}
return Chain;
}
static void fail(SDLoc DL, SelectionDAG &DAG, const char *msg) {
MachineFunction &MF = DAG.getMachineFunction();
DAG.getContext()->diagnose(
DiagnosticInfoUnsupported(DL, *MF.getFunction(), msg, SDValue()));
}
/// LowerCCCArguments - transform physical registers into virtual registers and
/// generate load operations for arguments places on the stack.
// FIXME: struct return stuff
// FIXME: varargs
SDValue
MSP430TargetLowering::LowerCCCArguments(SDValue Chain,
CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl,
SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals)
const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_MSP430);
assert(!isVarArg && "Varargs not supported yet");
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (VA.isRegLoc()) {
// Arguments passed in registers
EVT RegVT = VA.getLocVT();
switch (RegVT.getSimpleVT().SimpleTy) {
default:
{
#ifndef NDEBUG
errs() << "LowerFormalArguments Unhandled argument type: "
<< RegVT.getSimpleVT().SimpleTy << "\n";
#endif
llvm_unreachable(0);
}
case MVT::i16:
unsigned VReg =
RegInfo.createVirtualRegister(MSP430::GR16RegisterClass);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, RegVT);
// If this is an 8-bit value, it is really passed promoted to 16
// bits. Insert an assert[sz]ext to capture this, then truncate to the
// right size.
if (VA.getLocInfo() == CCValAssign::SExt)
ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
if (VA.getLocInfo() != CCValAssign::Full)
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
InVals.push_back(ArgValue);
}
} else {
// Sanity check
assert(VA.isMemLoc());
// Load the argument to a virtual register
unsigned ObjSize = VA.getLocVT().getSizeInBits()/8;
if (ObjSize > 2) {
errs() << "LowerFormalArguments Unhandled argument type: "
<< EVT(VA.getLocVT()).getEVTString()
<< "\n";
}
// Create the frame index object for this incoming parameter...
int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset(), true);
// Create the SelectionDAG nodes corresponding to a load
//from this parameter
SDValue FIN = DAG.getFrameIndex(FI, MVT::i16);
InVals.push_back(DAG.getLoad(VA.getLocVT(), dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, 0));
}
}
return Chain;
}
SDValue
X86SelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl,
SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
bool isVolatile, bool AlwaysInline,
MachinePointerInfo DstPtrInfo,
MachinePointerInfo SrcPtrInfo) const {
// This requires the copy size to be a constant, preferably
// within a subtarget-specific limit.
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
if (!ConstantSize)
return SDValue();
uint64_t SizeVal = ConstantSize->getZExtValue();
if (!AlwaysInline && SizeVal > Subtarget->getMaxInlineSizeThreshold())
return SDValue();
/// If not DWORD aligned, it is more efficient to call the library. However
/// if calling the library is not allowed (AlwaysInline), then soldier on as
/// the code generated here is better than the long load-store sequence we
/// would otherwise get.
if (!AlwaysInline && (Align & 3) != 0)
return SDValue();
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256 ||
SrcPtrInfo.getAddrSpace() >= 256)
return SDValue();
// ESI might be used as a base pointer, in that case we can't simply overwrite
// the register. Fall back to generic code.
const X86RegisterInfo *TRI =
static_cast<const X86RegisterInfo *>(DAG.getTarget().getRegisterInfo());
if (TRI->hasBasePointer(DAG.getMachineFunction()) &&
TRI->getBaseRegister() == X86::ESI)
return SDValue();
MVT AVT;
if (Align & 1)
AVT = MVT::i8;
else if (Align & 2)
AVT = MVT::i16;
else if (Align & 4)
// DWORD aligned
AVT = MVT::i32;
else
// QWORD aligned
AVT = Subtarget->is64Bit() ? MVT::i64 : MVT::i32;
unsigned UBytes = AVT.getSizeInBits() / 8;
unsigned CountVal = SizeVal / UBytes;
SDValue Count = DAG.getIntPtrConstant(CountVal);
unsigned BytesLeft = SizeVal % UBytes;
SDValue InFlag;
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RCX :
X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RDI :
X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RSI :
X86::ESI,
Src, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
SDValue RepMovs = DAG.getNode(X86ISD::REP_MOVS, dl, Tys, Ops);
SmallVector<SDValue, 4> Results;
Results.push_back(RepMovs);
if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT DstVT = Dst.getValueType();
EVT SrcVT = Src.getValueType();
EVT SizeVT = Size.getValueType();
Results.push_back(DAG.getMemcpy(Chain, dl,
DAG.getNode(ISD::ADD, dl, DstVT, Dst,
DAG.getConstant(Offset, DstVT)),
DAG.getNode(ISD::ADD, dl, SrcVT, Src,
DAG.getConstant(Offset, SrcVT)),
DAG.getConstant(BytesLeft, SizeVT),
Align, isVolatile, AlwaysInline,
DstPtrInfo.getWithOffset(Offset),
SrcPtrInfo.getWithOffset(Offset)));
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Results);
}
static SDValue EmitCMP(SDValue &LHS, SDValue &RHS, SDValue &TargetCC,
ISD::CondCode CC,
DebugLoc dl, SelectionDAG &DAG) {
// FIXME: Handle bittests someday
assert(!LHS.getValueType().isFloatingPoint() && "We don't handle FP yet");
// FIXME: Handle jump negative someday
MSP430CC::CondCodes TCC = MSP430CC::COND_INVALID;
switch (CC) {
default: llvm_unreachable("Invalid integer condition!");
case ISD::SETEQ:
TCC = MSP430CC::COND_E; // aka COND_Z
// Minor optimization: if LHS is a constant, swap operands, then the
// constant can be folded into comparison.
if (LHS.getOpcode() == ISD::Constant)
std::swap(LHS, RHS);
break;
case ISD::SETNE:
TCC = MSP430CC::COND_NE; // aka COND_NZ
// Minor optimization: if LHS is a constant, swap operands, then the
// constant can be folded into comparison.
if (LHS.getOpcode() == ISD::Constant)
std::swap(LHS, RHS);
break;
case ISD::SETULE:
std::swap(LHS, RHS); // FALLTHROUGH
case ISD::SETUGE:
// Turn lhs u>= rhs with lhs constant into rhs u< lhs+1, this allows us to
// fold constant into instruction.
if (const ConstantSDNode * C = dyn_cast<ConstantSDNode>(LHS)) {
LHS = RHS;
RHS = DAG.getConstant(C->getSExtValue() + 1, C->getValueType(0));
TCC = MSP430CC::COND_LO;
break;
}
TCC = MSP430CC::COND_HS; // aka COND_C
break;
case ISD::SETUGT:
std::swap(LHS, RHS); // FALLTHROUGH
case ISD::SETULT:
// Turn lhs u< rhs with lhs constant into rhs u>= lhs+1, this allows us to
// fold constant into instruction.
if (const ConstantSDNode * C = dyn_cast<ConstantSDNode>(LHS)) {
LHS = RHS;
RHS = DAG.getConstant(C->getSExtValue() + 1, C->getValueType(0));
TCC = MSP430CC::COND_HS;
break;
}
TCC = MSP430CC::COND_LO; // aka COND_NC
break;
case ISD::SETLE:
std::swap(LHS, RHS); // FALLTHROUGH
case ISD::SETGE:
// Turn lhs >= rhs with lhs constant into rhs < lhs+1, this allows us to
// fold constant into instruction.
if (const ConstantSDNode * C = dyn_cast<ConstantSDNode>(LHS)) {
LHS = RHS;
RHS = DAG.getConstant(C->getSExtValue() + 1, C->getValueType(0));
TCC = MSP430CC::COND_L;
break;
}
TCC = MSP430CC::COND_GE;
break;
case ISD::SETGT:
std::swap(LHS, RHS); // FALLTHROUGH
case ISD::SETLT:
// Turn lhs < rhs with lhs constant into rhs >= lhs+1, this allows us to
// fold constant into instruction.
if (const ConstantSDNode * C = dyn_cast<ConstantSDNode>(LHS)) {
LHS = RHS;
RHS = DAG.getConstant(C->getSExtValue() + 1, C->getValueType(0));
TCC = MSP430CC::COND_GE;
break;
}
TCC = MSP430CC::COND_L;
break;
}
TargetCC = DAG.getConstant(TCC, MVT::i8);
return DAG.getNode(MSP430ISD::CMP, dl, MVT::Glue, LHS, RHS);
}
SDValue
X86SelectionDAGInfo::EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
SDValue Chain,
SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256)
return SDValue();
// If not DWORD aligned or size is more than the threshold, call the library.
// The libc version is likely to be faster for these cases. It can use the
// address value and run time information about the CPU.
if ((Align & 3) != 0 ||
!ConstantSize ||
ConstantSize->getZExtValue() >
Subtarget->getMaxInlineSizeThreshold()) {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *V = dyn_cast<ConstantSDNode>(Src);
if (const char *bzeroEntry = V &&
V->isNullValue() ? Subtarget->getBZeroEntry() : nullptr) {
EVT IntPtr = TLI.getPointerTy();
Type *IntPtrTy = getDataLayout()->getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(bzeroEntry, IntPtr), &Args, 0)
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memset.
return SDValue();
}
uint64_t SizeVal = ConstantSize->getZExtValue();
SDValue InFlag;
EVT AVT;
SDValue Count;
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Src);
unsigned BytesLeft = 0;
bool TwoRepStos = false;
if (ValC) {
unsigned ValReg;
uint64_t Val = ValC->getZExtValue() & 255;
// If the value is a constant, then we can potentially use larger sets.
switch (Align & 3) {
case 2: // WORD aligned
AVT = MVT::i16;
ValReg = X86::AX;
Val = (Val << 8) | Val;
break;
case 0: // DWORD aligned
AVT = MVT::i32;
ValReg = X86::EAX;
Val = (Val << 8) | Val;
Val = (Val << 16) | Val;
if (Subtarget->is64Bit() && ((Align & 0x7) == 0)) { // QWORD aligned
AVT = MVT::i64;
ValReg = X86::RAX;
Val = (Val << 32) | Val;
}
break;
default: // Byte aligned
AVT = MVT::i8;
ValReg = X86::AL;
Count = DAG.getIntPtrConstant(SizeVal);
break;
}
if (AVT.bitsGT(MVT::i8)) {
unsigned UBytes = AVT.getSizeInBits() / 8;
Count = DAG.getIntPtrConstant(SizeVal / UBytes);
BytesLeft = SizeVal % UBytes;
}
Chain = DAG.getCopyToReg(Chain, dl, ValReg, DAG.getConstant(Val, AVT),
InFlag);
InFlag = Chain.getValue(1);
} else {
AVT = MVT::i8;
Count = DAG.getIntPtrConstant(SizeVal);
Chain = DAG.getCopyToReg(Chain, dl, X86::AL, Src, InFlag);
InFlag = Chain.getValue(1);
}
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RCX :
X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RDI :
X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
if (TwoRepStos) {
InFlag = Chain.getValue(1);
Count = Size;
EVT CVT = Count.getValueType();
SDValue Left = DAG.getNode(ISD::AND, dl, CVT, Count,
DAG.getConstant((AVT == MVT::i64) ? 7 : 3, CVT));
Chain = DAG.getCopyToReg(Chain, dl, (CVT == MVT::i64) ? X86::RCX :
X86::ECX,
Left, InFlag);
InFlag = Chain.getValue(1);
Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(MVT::i8), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
} else if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT AddrVT = Dst.getValueType();
EVT SizeVT = Size.getValueType();
Chain = DAG.getMemset(Chain, dl,
DAG.getNode(ISD::ADD, dl, AddrVT, Dst,
DAG.getConstant(Offset, AddrVT)),
Src,
DAG.getConstant(BytesLeft, SizeVT),
Align, isVolatile, DstPtrInfo.getWithOffset(Offset));
}
// TODO: Use a Tokenfactor, as in memcpy, instead of a single chain.
return Chain;
}
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;
}
SDValue ARMSelectionDAGInfo::EmitTargetCodeForMemcpy(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, bool isVolatile, bool AlwaysInline,
MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
const ARMSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<ARMSubtarget>();
// Do repeated 4-byte loads and stores. To be improved.
// This requires 4-byte alignment.
if ((Align & 3) != 0)
return SDValue();
// This requires the copy size to be a constant, preferably
// within a subtarget-specific limit.
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
if (!ConstantSize)
return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
RTLIB::MEMCPY);
uint64_t SizeVal = ConstantSize->getZExtValue();
if (!AlwaysInline && SizeVal > Subtarget.getMaxInlineSizeThreshold())
return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
RTLIB::MEMCPY);
unsigned BytesLeft = SizeVal & 3;
unsigned NumMemOps = SizeVal >> 2;
unsigned EmittedNumMemOps = 0;
EVT VT = MVT::i32;
unsigned VTSize = 4;
unsigned i = 0;
// Emit a maximum of 4 loads in Thumb1 since we have fewer registers
const unsigned MaxLoadsInLDM = Subtarget.isThumb1Only() ? 4 : 6;
SDValue TFOps[6];
SDValue Loads[6];
uint64_t SrcOff = 0, DstOff = 0;
// FIXME: We should invent a VMEMCPY pseudo-instruction that lowers to
// VLDM/VSTM and make this code emit it when appropriate. This would reduce
// pressure on the general purpose registers. However this seems harder to map
// onto the register allocator's view of the world.
// The number of MEMCPY pseudo-instructions to emit. We use up to
// MaxLoadsInLDM registers per mcopy, which will get lowered into ldm/stm
// later on. This is a lower bound on the number of MEMCPY operations we must
// emit.
unsigned NumMEMCPYs = (NumMemOps + MaxLoadsInLDM - 1) / MaxLoadsInLDM;
// Code size optimisation: do not inline memcpy if expansion results in
// more instructions than the libary call.
if (NumMEMCPYs > 1 && DAG.getMachineFunction().getFunction().optForMinSize()) {
return SDValue();
}
SDVTList VTs = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other, MVT::Glue);
for (unsigned I = 0; I != NumMEMCPYs; ++I) {
// Evenly distribute registers among MEMCPY operations to reduce register
// pressure.
unsigned NextEmittedNumMemOps = NumMemOps * (I + 1) / NumMEMCPYs;
unsigned NumRegs = NextEmittedNumMemOps - EmittedNumMemOps;
Dst = DAG.getNode(ARMISD::MEMCPY, dl, VTs, Chain, Dst, Src,
DAG.getConstant(NumRegs, dl, MVT::i32));
Src = Dst.getValue(1);
Chain = Dst.getValue(2);
DstPtrInfo = DstPtrInfo.getWithOffset(NumRegs * VTSize);
SrcPtrInfo = SrcPtrInfo.getWithOffset(NumRegs * VTSize);
EmittedNumMemOps = NextEmittedNumMemOps;
}
if (BytesLeft == 0)
return Chain;
// Issue loads / stores for the trailing (1 - 3) bytes.
auto getRemainingValueType = [](unsigned BytesLeft) {
return (BytesLeft >= 2) ? MVT::i16 : MVT::i8;
};
auto getRemainingSize = [](unsigned BytesLeft) {
return (BytesLeft >= 2) ? 2 : 1;
};
unsigned BytesLeftSave = BytesLeft;
i = 0;
while (BytesLeft) {
VT = getRemainingValueType(BytesLeft);
VTSize = getRemainingSize(BytesLeft);
Loads[i] = DAG.getLoad(VT, dl, Chain,
DAG.getNode(ISD::ADD, dl, MVT::i32, Src,
DAG.getConstant(SrcOff, dl, MVT::i32)),
SrcPtrInfo.getWithOffset(SrcOff));
TFOps[i] = Loads[i].getValue(1);
++i;
SrcOff += VTSize;
BytesLeft -= VTSize;
}
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
makeArrayRef(TFOps, i));
i = 0;
BytesLeft = BytesLeftSave;
while (BytesLeft) {
VT = getRemainingValueType(BytesLeft);
VTSize = getRemainingSize(BytesLeft);
TFOps[i] = DAG.getStore(Chain, dl, Loads[i],
DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
DAG.getConstant(DstOff, dl, MVT::i32)),
DstPtrInfo.getWithOffset(DstOff));
++i;
DstOff += VTSize;
BytesLeft -= VTSize;
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
makeArrayRef(TFOps, i));
}
