// Write ByVal Arg to arg registers and stack.
static void
WriteByValArg(SDValue& ByValChain, SDValue Chain, SDLoc DL,
SmallVector<std::pair<unsigned, SDValue>, 16>& RegsToPass,
SmallVector<SDValue, 8>& MemOpChains, int& LastFI,
MachineFrameInfo *MFI, SelectionDAG &DAG, SDValue Arg,
const CCValAssign &VA, const ISD::ArgFlagsTy& Flags,
MVT PtrType, bool isLittle) {
unsigned LocMemOffset = VA.getLocMemOffset();
unsigned Offset = 0;
uint32_t RemainingSize = Flags.getByValSize();
unsigned ByValAlign = Flags.getByValAlign();
if (RemainingSize == 0)
return;
// Create a fixed object on stack at offset LocMemOffset and copy
// remaining part of byval arg to it using memcpy.
SDValue Src = DAG.getNode(ISD::ADD, DL, MVT::i32, Arg,
DAG.getConstant(Offset, MVT::i32));
LastFI = MFI->CreateFixedObject(RemainingSize, LocMemOffset, true);
SDValue Dst = DAG.getFrameIndex(LastFI, PtrType);
ByValChain = DAG.getMemcpy(ByValChain, DL, Dst, Src,
DAG.getConstant(RemainingSize, MVT::i32),
std::min(ByValAlign, (unsigned)4),
/*isVolatile=*/false, /*AlwaysInline=*/false,
MachinePointerInfo(0), MachinePointerInfo(0));
}
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);
}
/// 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, const SDLoc &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, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext());
AnalyzeArguments(CCInfo, ArgLocs, Outs);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
auto PtrVT = getPointerTy(DAG.getDataLayout());
Chain = DAG.getCALLSEQ_START(Chain,
DAG.getConstant(NumBytes, dl, PtrVT, true), dl);
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())
StackPtr = DAG.getCopyFromReg(Chain, dl, MSP430::SP, PtrVT);
SDValue PtrOff =
DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
DAG.getIntPtrConstant(VA.getLocMemOffset(), dl));
SDValue MemOp;
ISD::ArgFlagsTy Flags = Outs[i].Flags;
if (Flags.isByVal()) {
SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), dl, MVT::i16);
MemOp = DAG.getMemcpy(Chain, dl, PtrOff, Arg, SizeNode,
Flags.getByValAlign(),
/*isVolatile*/false,
/*AlwaysInline=*/true,
/*isTailCall=*/false,
MachinePointerInfo(),
MachinePointerInfo());
} else {
MemOp = DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo(),
false, false, 0);
}
MemOpChains.push_back(MemOp);
}
}
// 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);
// 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);
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(Chain, DAG.getConstant(NumBytes, dl, PtrVT, true),
DAG.getConstant(0, dl, PtrVT, true), InFlag, 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, dl,
DAG, InVals);
}
SDValue X86SelectionDAGInfo::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 {
// This requires the copy size to be a constant, preferably
// within a subtarget-specific limit.
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
const X86Subtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<X86Subtarget>();
if (!ConstantSize)
return SDValue();
RepMovsRepeats Repeats(ConstantSize->getZExtValue());
if (!AlwaysInline && Repeats.Size > 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();
// If the base register might conflict with our physical registers, bail out.
const MCPhysReg ClobberSet[] = {X86::RCX, X86::RSI, X86::RDI,
X86::ECX, X86::ESI, X86::EDI};
if (isBaseRegConflictPossible(DAG, ClobberSet))
return SDValue();
// If the target has enhanced REPMOVSB, then it's at least as fast to use
// REP MOVSB instead of REP MOVS{W,D,Q}, and it avoids having to handle
// BytesLeft.
if (!Subtarget.hasERMSB() && !(Align & 1)) {
if (Align & 2)
// WORD aligned
Repeats.AVT = MVT::i16;
else if (Align & 4)
// DWORD aligned
Repeats.AVT = MVT::i32;
else
// QWORD aligned
Repeats.AVT = Subtarget.is64Bit() ? MVT::i64 : MVT::i32;
if (Repeats.BytesLeft() > 0 &&
DAG.getMachineFunction().getFunction().optForMinSize()) {
// When agressively optimizing for size, avoid generating the code to
// handle BytesLeft.
Repeats.AVT = MVT::i8;
}
}
SDValue InFlag;
Chain = DAG.getCopyToReg(Chain, dl, Subtarget.is64Bit() ? X86::RCX : X86::ECX,
DAG.getIntPtrConstant(Repeats.Count(), dl), 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(Repeats.AVT), InFlag };
SDValue RepMovs = DAG.getNode(X86ISD::REP_MOVS, dl, Tys, Ops);
SmallVector<SDValue, 4> Results;
Results.push_back(RepMovs);
if (Repeats.BytesLeft()) {
// Handle the last 1 - 7 bytes.
unsigned Offset = Repeats.Size - Repeats.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, dl,
DstVT)),
DAG.getNode(ISD::ADD, dl, SrcVT, Src,
DAG.getConstant(Offset, dl,
SrcVT)),
DAG.getConstant(Repeats.BytesLeft(), dl,
SizeVT),
Align, isVolatile, AlwaysInline, false,
DstPtrInfo.getWithOffset(Offset),
SrcPtrInfo.getWithOffset(Offset)));
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Results);
}
// LowerCCCCallTo - functions arguments are copied from virtual regs to
// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
SDValue LanaiTargetLowering::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, SDLoc 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, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext());
GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
NumFixedArgs = 0;
if (IsVarArg && G) {
const Function *CalleeFn = dyn_cast<Function>(G->getGlobal());
if (CalleeFn)
NumFixedArgs = CalleeFn->getFunctionType()->getNumParams();
}
if (NumFixedArgs)
CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32_VarArg);
else {
if (CallConv == CallingConv::Fast)
CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32_Fast);
else
CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32);
}
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
// Create local copies for byval args.
SmallVector<SDValue, 8> ByValArgs;
for (unsigned I = 0, E = Outs.size(); I != E; ++I) {
ISD::ArgFlagsTy Flags = Outs[I].Flags;
if (!Flags.isByVal())
continue;
SDValue Arg = OutVals[I];
unsigned Size = Flags.getByValSize();
unsigned Align = Flags.getByValAlign();
int FI = MFI->CreateStackObject(Size, Align, false);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy(DAG.getDataLayout()));
SDValue SizeNode = DAG.getConstant(Size, DL, MVT::i32);
Chain = DAG.getMemcpy(Chain, DL, FIPtr, Arg, SizeNode, Align,
/*IsVolatile=*/false,
/*AlwaysInline=*/false,
/*IsTailCall=*/false, MachinePointerInfo(),
MachinePointerInfo());
ByValArgs.push_back(FIPtr);
}
Chain = DAG.getCALLSEQ_START(
Chain,
DAG.getConstant(NumBytes, DL, getPointerTy(DAG.getDataLayout()), true),
DL);
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, J = 0, E = ArgLocs.size(); I != E; ++I) {
CCValAssign &VA = ArgLocs[I];
SDValue Arg = OutVals[I];
ISD::ArgFlagsTy Flags = Outs[I].Flags;
// Promote the value if needed.
switch (VA.getLocInfo()) {
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;
default:
llvm_unreachable("Unknown loc info!");
}
// Use local copy if it is a byval arg.
if (Flags.isByVal())
Arg = ByValArgs[J++];
// 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, Lanai::SP,
getPointerTy(DAG.getDataLayout()));
SDValue PtrOff =
DAG.getNode(ISD::ADD, DL, getPointerTy(DAG.getDataLayout()), StackPtr,
DAG.getIntPtrConstant(VA.getLocMemOffset(), DL));
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,
ArrayRef<SDValue>(&MemOpChains[0], MemOpChains.size()));
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 (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.
uint8_t OpFlag = LanaiII::MO_NO_FLAG;
if (G) {
Callee = DAG.getTargetGlobalAddress(
G->getGlobal(), DL, getPointerTy(DAG.getDataLayout()), 0, OpFlag);
} else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee)) {
Callee = DAG.getTargetExternalSymbol(
E->getSymbol(), getPointerTy(DAG.getDataLayout()), OpFlag);
}
// 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 a register mask operand representing the call-preserved registers.
// TODO: Should return-twice functions be handled?
const uint32_t *Mask =
TRI->getCallPreservedMask(DAG.getMachineFunction(), CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
// 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(LanaiISD::CALL, DL, NodeTys,
ArrayRef<SDValue>(&Ops[0], Ops.size()));
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(
Chain,
DAG.getConstant(NumBytes, DL, getPointerTy(DAG.getDataLayout()), true),
DAG.getConstant(0, DL, getPointerTy(DAG.getDataLayout()), true), InFlag,
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, DL, DAG,
InVals);
}
/// Transform physical registers into virtual registers, and generate load
/// operations for argument places on the stack.
SDValue ARCTargetLowering::LowerCallArguments(
SDValue Chain, CallingConv::ID CallConv, bool IsVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
auto *AFI = MF.getInfo<ARCFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_ARC);
unsigned StackSlotSize = 4;
if (!IsVarArg)
AFI->setReturnStackOffset(CCInfo.getNextStackOffset());
// All getCopyFromReg ops must precede any getMemcpys to prevent the
// scheduler clobbering a register before it has been copied.
// The stages are:
// 1. CopyFromReg (and load) arg & vararg registers.
// 2. Chain CopyFromReg nodes into a TokenFactor.
// 3. Memcpy 'byVal' args & push final InVals.
// 4. Chain mem ops nodes into a TokenFactor.
SmallVector<SDValue, 4> CFRegNode;
SmallVector<ArgDataPair, 4> ArgData;
SmallVector<SDValue, 4> MemOps;
// 1a. CopyFromReg (and load) arg registers.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
SDValue ArgIn;
if (VA.isRegLoc()) {
// Arguments passed in registers
EVT RegVT = VA.getLocVT();
switch (RegVT.getSimpleVT().SimpleTy) {
default: {
DEBUG(errs() << "LowerFormalArguments Unhandled argument type: "
<< RegVT.getSimpleVT().SimpleTy << "\n");
llvm_unreachable("Unhandled LowerFormalArguments type.");
}
case MVT::i32:
unsigned VReg = RegInfo.createVirtualRegister(&ARC::GPR32RegClass);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
ArgIn = DAG.getCopyFromReg(Chain, dl, VReg, RegVT);
CFRegNode.push_back(ArgIn.getValue(ArgIn->getNumValues() - 1));
}
} else {
// sanity check
assert(VA.isMemLoc());
// Load the argument to a virtual register
unsigned ObjSize = VA.getLocVT().getStoreSize();
assert((ObjSize <= StackSlotSize) && "Unhandled argument");
// 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::i32);
ArgIn = DAG.getLoad(VA.getLocVT(), dl, Chain, FIN,
MachinePointerInfo::getFixedStack(MF, FI));
}
const ArgDataPair ADP = {ArgIn, Ins[i].Flags};
ArgData.push_back(ADP);
}
// 1b. CopyFromReg vararg registers.
if (IsVarArg) {
// Argument registers
static const MCPhysReg ArgRegs[] = {ARC::R0, ARC::R1, ARC::R2, ARC::R3,
ARC::R4, ARC::R5, ARC::R6, ARC::R7};
auto *AFI = MF.getInfo<ARCFunctionInfo>();
unsigned FirstVAReg = CCInfo.getFirstUnallocated(ArgRegs);
if (FirstVAReg < array_lengthof(ArgRegs)) {
int Offset = 0;
// Save remaining registers, storing higher register numbers at a higher
// address
// There are (array_lengthof(ArgRegs) - FirstVAReg) registers which
// need to be saved.
int VarFI =
MFI.CreateFixedObject((array_lengthof(ArgRegs) - FirstVAReg) * 4,
CCInfo.getNextStackOffset(), true);
AFI->setVarArgsFrameIndex(VarFI);
SDValue FIN = DAG.getFrameIndex(VarFI, MVT::i32);
for (unsigned i = FirstVAReg; i < array_lengthof(ArgRegs); i++) {
// Move argument from phys reg -> virt reg
unsigned VReg = RegInfo.createVirtualRegister(&ARC::GPR32RegClass);
RegInfo.addLiveIn(ArgRegs[i], VReg);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
CFRegNode.push_back(Val.getValue(Val->getNumValues() - 1));
SDValue VAObj = DAG.getNode(ISD::ADD, dl, MVT::i32, FIN,
DAG.getConstant(Offset, dl, MVT::i32));
// Move argument from virt reg -> stack
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, VAObj, MachinePointerInfo());
MemOps.push_back(Store);
Offset += 4;
}
} else {
llvm_unreachable("Too many var args parameters.");
}
}
// 2. Chain CopyFromReg nodes into a TokenFactor.
if (!CFRegNode.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, CFRegNode);
// 3. Memcpy 'byVal' args & push final InVals.
// Aggregates passed "byVal" need to be copied by the callee.
// The callee will use a pointer to this copy, rather than the original
// pointer.
for (const auto &ArgDI : ArgData) {
if (ArgDI.Flags.isByVal() && ArgDI.Flags.getByValSize()) {
unsigned Size = ArgDI.Flags.getByValSize();
unsigned Align = std::max(StackSlotSize, ArgDI.Flags.getByValAlign());
// Create a new object on the stack and copy the pointee into it.
int FI = MFI.CreateStackObject(Size, Align, false);
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
InVals.push_back(FIN);
MemOps.push_back(DAG.getMemcpy(
Chain, dl, FIN, ArgDI.SDV, DAG.getConstant(Size, dl, MVT::i32), Align,
false, false, false, MachinePointerInfo(), MachinePointerInfo()));
} else {
InVals.push_back(ArgDI.SDV);
}
}
// 4. Chain mem ops nodes into a TokenFactor.
if (!MemOps.empty()) {
MemOps.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
}
return Chain;
}
SDValue
X86SelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
bool isVolatile, bool AlwaysInline,
const Value *DstSV,
uint64_t DstSVOff,
const Value *SrcSV,
uint64_t SrcSVOff) const {
// This requires the copy size to be a constant, preferrably
// 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, call the library.
if ((Align & 3) != 0)
return SDValue();
// DWORD aligned
EVT AVT = MVT::i32;
if (Subtarget->is64Bit() && ((Align & 0x7) == 0)) // QWORD aligned
AVT = MVT::i64;
unsigned UBytes = AVT.getSizeInBits() / 8;
unsigned CountVal = SizeVal / UBytes;
SDValue Count = DAG.getIntPtrConstant(CountVal);
unsigned BytesLeft = SizeVal % UBytes;
SDValue InFlag(0, 0);
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::Flag);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
SDValue RepMovs = DAG.getNode(X86ISD::REP_MOVS, dl, Tys, Ops,
array_lengthof(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,
DstSV, DstSVOff + Offset,
SrcSV, SrcSVOff + Offset));
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&Results[0], Results.size());
}
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();
// If ESI is used as a base pointer, we must preserve it when doing rep movs.
const X86RegisterInfo *TRI =
static_cast<const X86RegisterInfo *>(DAG.getTarget().getRegisterInfo());
bool PreserveESI = TRI->hasBasePointer(DAG.getMachineFunction()) &&
TRI->getBaseRegister() == X86::ESI;
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;
if (PreserveESI) {
// Save ESI to a physical register. (We cannot use a virtual register
// because if it is spilled we wouldn't be able to reload it.)
// We don't glue this because the register dependencies are explicit.
Chain = DAG.getCopyToReg(Chain, dl, X86::EDX,
DAG.getRegister(X86::ESI, MVT::i32));
}
SDValue InGlue(0, 0);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RCX :
X86::ECX,
Count, InGlue);
InGlue = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RDI :
X86::EDI,
Dst, InGlue);
InGlue = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RSI :
X86::ESI,
Src, InGlue);
InGlue = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InGlue };
// FIXME: Make X86rep_movs explicitly use FCX, RDI, RSI instead of glue.
SDValue RepMovs = DAG.getNode(X86ISD::REP_MOVS, dl, Tys, Ops,
array_lengthof(Ops));
if (PreserveESI) {
InGlue = RepMovs.getValue(1);
RepMovs = DAG.getCopyToReg(RepMovs, dl, X86::ESI,
DAG.getRegister(X86::EDX, MVT::i32), InGlue);
}
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[0], Results.size());
}
SDValue
SparcTargetLowering::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 {
// Sparc 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.getTarget(), ArgLocs,
*DAG.getContext());
CCInfo.AnalyzeCallOperands(Outs, CC_Sparc32);
// Get the size of the outgoing arguments stack space requirement.
unsigned ArgsSize = CCInfo.getNextStackOffset();
// Keep stack frames 8-byte aligned.
ArgsSize = (ArgsSize+7) & ~7;
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
//Create local copies for byval args.
SmallVector<SDValue, 8> ByValArgs;
for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
ISD::ArgFlagsTy Flags = Outs[i].Flags;
if (!Flags.isByVal())
continue;
SDValue Arg = OutVals[i];
unsigned Size = Flags.getByValSize();
unsigned Align = Flags.getByValAlign();
int FI = MFI->CreateStackObject(Size, Align, false);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
SDValue SizeNode = DAG.getConstant(Size, MVT::i32);
Chain = DAG.getMemcpy(Chain, dl, FIPtr, Arg, SizeNode, Align,
false, //isVolatile,
(Size <= 32), //AlwaysInline if size <= 32
MachinePointerInfo(), MachinePointerInfo());
ByValArgs.push_back(FIPtr);
}
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(ArgsSize, true));
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
const unsigned StackOffset = 92;
bool hasStructRetAttr = false;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, realArgIdx = 0, byvalArgIdx = 0, e = ArgLocs.size();
i != e;
++i, ++realArgIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[realArgIdx];
ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
//Use local copy if it is a byval arg.
if (Flags.isByVal())
Arg = ByValArgs[byvalArgIdx++];
// 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;
case CCValAssign::BCvt:
Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
break;
}
if (Flags.isSRet()) {
assert(VA.needsCustom());
// store SRet argument in %sp+64
SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
SDValue PtrOff = DAG.getIntPtrConstant(64);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(),
false, false, 0));
hasStructRetAttr = true;
continue;
}
if (VA.needsCustom()) {
assert(VA.getLocVT() == MVT::f64);
if (VA.isMemLoc()) {
unsigned Offset = VA.getLocMemOffset() + StackOffset;
//if it is double-word aligned, just store.
if (Offset % 8 == 0) {
SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
SDValue PtrOff = DAG.getIntPtrConstant(Offset);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(),
false, false, 0));
continue;
}
}
SDValue StackPtr = DAG.CreateStackTemporary(MVT::f64, MVT::i32);
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl,
Arg, StackPtr, MachinePointerInfo(),
false, false, 0);
// Sparc is big-endian, so the high part comes first.
SDValue Hi = DAG.getLoad(MVT::i32, dl, Store, StackPtr,
MachinePointerInfo(), false, false, 0);
// Increment the pointer to the other half.
StackPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
DAG.getIntPtrConstant(4));
// Load the low part.
SDValue Lo = DAG.getLoad(MVT::i32, dl, Store, StackPtr,
MachinePointerInfo(), false, false, 0);
if (VA.isRegLoc()) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Hi));
assert(i+1 != e);
CCValAssign &NextVA = ArgLocs[++i];
if (NextVA.isRegLoc()) {
RegsToPass.push_back(std::make_pair(NextVA.getLocReg(), Lo));
} else {
//Store the low part in stack.
unsigned Offset = NextVA.getLocMemOffset() + StackOffset;
SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
SDValue PtrOff = DAG.getIntPtrConstant(Offset);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Lo, PtrOff,
MachinePointerInfo(),
false, false, 0));
}
} else {
unsigned Offset = VA.getLocMemOffset() + StackOffset;
// Store the high part.
SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
SDValue PtrOff = DAG.getIntPtrConstant(Offset);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Hi, PtrOff,
MachinePointerInfo(),
false, false, 0));
// Store the low part.
PtrOff = DAG.getIntPtrConstant(Offset+4);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Lo, PtrOff,
MachinePointerInfo(),
false, false, 0));
}
continue;
}
// Arguments that can be passed on register must be kept at
// RegsToPass vector
if (VA.isRegLoc()) {
if (VA.getLocVT() != MVT::f32) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
continue;
}
Arg = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Arg);
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
continue;
}
assert(VA.isMemLoc());
// Create a store off the stack pointer for this argument.
SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
SDValue PtrOff = DAG.getIntPtrConstant(VA.getLocMemOffset()+StackOffset);
PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(),
false, false, 0));
}
// Emit all stores, make sure the occur before any copies into physregs.
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) {
unsigned Reg = RegsToPass[i].first;
// Remap I0->I7 -> O0->O7.
if (Reg >= SP::I0 && Reg <= SP::I7)
Reg = Reg-SP::I0+SP::O0;
Chain = DAG.getCopyToReg(Chain, dl, Reg, RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
unsigned SRetArgSize = (hasStructRetAttr)? getSRetArgSize(DAG, Callee):0;
// 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);
// 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);
if (hasStructRetAttr)
Ops.push_back(DAG.getTargetConstant(SRetArgSize, MVT::i32));
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
unsigned Reg = RegsToPass[i].first;
if (Reg >= SP::I0 && Reg <= SP::I7)
Reg = Reg-SP::I0+SP::O0;
Ops.push_back(DAG.getRegister(Reg, RegsToPass[i].second.getValueType()));
}
if (InFlag.getNode())
Ops.push_back(InFlag);
Chain = DAG.getNode(SPISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
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.getTarget(),
RVLocs, *DAG.getContext());
RVInfo.AnalyzeCallResult(Ins, RetCC_Sparc32);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
unsigned Reg = RVLocs[i].getLocReg();
// Remap I0->I7 -> O0->O7.
if (Reg >= SP::I0 && Reg <= SP::I7)
Reg = Reg-SP::I0+SP::O0;
Chain = DAG.getCopyFromReg(Chain, dl, Reg,
RVLocs[i].getValVT(), InFlag).getValue(1);
InFlag = Chain.getValue(2);
InVals.push_back(Chain.getValue(0));
}
return Chain;
}