// 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));
}
//===----------------------------------------------------------------------===//
// Misc Lower Operation implementation
//===----------------------------------------------------------------------===//
SDValue Cpu0TargetLowering::
LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const
{
MachineFunction &MF = DAG.getMachineFunction();
Cpu0FunctionInfo *Cpu0FI = MF.getInfo<Cpu0FunctionInfo>();
unsigned SP = Cpu0::SP;
assert(getTargetMachine().getFrameLowering()->getStackAlignment() >=
cast<ConstantSDNode>(Op.getOperand(2).getNode())->getZExtValue() &&
"Cannot lower if the alignment of the allocated space is larger than \
that of the stack.");
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
DebugLoc dl = Op.getDebugLoc();
// Get a reference from Cpu0 stack pointer
SDValue StackPointer = DAG.getCopyFromReg(Chain, dl, SP, getPointerTy());
// Subtract the dynamic size from the actual stack size to
// obtain the new stack size.
SDValue Sub = DAG.getNode(ISD::SUB, dl, getPointerTy(), StackPointer, Size);
// The Sub result contains the new stack start address, so it
// must be placed in the stack pointer register.
Chain = DAG.getCopyToReg(StackPointer.getValue(1), dl, SP, Sub, SDValue());
// This node always has two return values: a new stack pointer
// value and a chain
SDVTList VTLs = DAG.getVTList(getPointerTy(), MVT::Other);
SDValue Ptr = DAG.getFrameIndex(Cpu0FI->getDynAllocFI(), getPointerTy());
SDValue Ops[] = { Chain, Ptr, Chain.getValue(1) };
return DAG.getNode(Cpu0ISD::DynAlloc, dl, VTLs, Ops, 3);
}
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;
}
SDValue ARCompactTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
ARCompactFunctionInfo *FuncInfo = MF.getInfo<ARCompactFunctionInfo>();
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
DebugLoc dl = Op.getDebugLoc();
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1),
MachinePointerInfo(SV), false, false, 0);
}
static SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) {
MachineFunction &MF = DAG.getMachineFunction();
auto *FuncInfo = MF.getInfo<ARCFunctionInfo>();
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
SDLoc dl(Op);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1),
MachinePointerInfo(SV));
}
SDValue LanaiTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
LanaiMachineFunctionInfo *FuncInfo = MF.getInfo<LanaiMachineFunctionInfo>();
SDLoc DL(Op);
SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
getPointerTy(DAG.getDataLayout()));
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1),
MachinePointerInfo(SV), false, false, 0);
}
SDValue MSP430TargetLowering::LowerVASTART(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
MSP430MachineFunctionInfo *FuncInfo = MF.getInfo<MSP430MachineFunctionInfo>();
// Frame index of first vararg argument
SDValue FrameIndex = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
getPointerTy());
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// Create a store of the frame index to the location operand
return DAG.getStore(Op.getOperand(0), Op.getDebugLoc(), FrameIndex,
Op.getOperand(1), MachinePointerInfo(SV),
false, false, 0);
}
SDValue
MSP430TargetLowering::getReturnAddressFrameIndex(SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
MSP430MachineFunctionInfo *FuncInfo = MF.getInfo<MSP430MachineFunctionInfo>();
int ReturnAddrIndex = FuncInfo->getRAIndex();
if (ReturnAddrIndex == 0) {
// Set up a frame object for the return address.
uint64_t SlotSize = TD->getPointerSize();
ReturnAddrIndex = MF.getFrameInfo()->CreateFixedObject(SlotSize, -SlotSize,
true);
FuncInfo->setRAIndex(ReturnAddrIndex);
}
return DAG.getFrameIndex(ReturnAddrIndex, getPointerTy());
}
/// LowerCall - functions arguments are copied from virtual regs to
/// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
/// TODO: isVarArg, isTailCall.
SDValue MBlazeTargetLowering::
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 {
// MBlaze does not yet support tail call optimization
isTailCall = false;
// The MBlaze requires stack slots for arguments passed to var arg
// functions even if they are passed in registers.
bool needsRegArgSlots = isVarArg;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
const TargetFrameInfo &TFI = *MF.getTarget().getFrameInfo();
// 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_MBlaze);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
// Variable argument function calls require a minimum of 24-bytes of stack
if (isVarArg && NumBytes < 24) NumBytes = 24;
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, 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];
MVT RegVT = VA.getLocVT();
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, RegVT, Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, RegVT, 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 {
// Register can't get to this point...
assert(VA.isMemLoc());
// Since we are alread passing values on the stack we don't
// need to worry about creating additional slots for the
// values passed via registers.
needsRegArgSlots = false;
// Create the frame index object for this incoming parameter
unsigned ArgSize = VA.getValVT().getSizeInBits()/8;
unsigned StackLoc = VA.getLocMemOffset() + 4;
int FI = MFI->CreateFixedObject(ArgSize, StackLoc, true);
SDValue PtrOff = DAG.getFrameIndex(FI,getPointerTy());
// emit ISD::STORE whichs stores the
// parameter value to a stack Location
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(),
false, false, 0));
}
}
// If we need to reserve stack space for the arguments passed via registers
// then create a fixed stack object at the beginning of the stack.
if (needsRegArgSlots && TFI.hasReservedCallFrame(MF))
MFI->CreateFixedObject(28,0,true);
// 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 emited 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/ExternalSymbol node (quite common, every
// direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
// node so that legalize doesn't hack it.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl,
getPointerTy(), 0, 0);
else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(S->getSymbol(),
getPointerTy(), 0);
// MBlazeJmpLink = #chain, #target_address, #opt_in_flags...
// = Chain, Callee, Reg#1, Reg#2, ...
//
// Returns a chain & a flag for retval copy to use.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
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(MBlazeISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(0, true), InFlag);
if (!Ins.empty())
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 AVM2TargetLowering::
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();
AVM2MachineFunctionInfo *FuncInfo = MF.getInfo<AVM2MachineFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_AVM2_32);
SDValue Root = DAG.getRoot();
unsigned ArgOffset = 0;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
SDValue ArgValue;
CCValAssign &VA = ArgLocs[i];
// FIXME: We ignore the register assignments of AnalyzeFormalArguments
// because it doesn't know how to split a double into two i32 registers.
EVT ObjectVT = VA.getValVT();
switch (ObjectVT.getSimpleVT().SimpleTy) {
default:
assert(0 && "Unhandled argument type!");
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
if (!Ins[i].Used) { // Argument is dead.
InVals.push_back(DAG.getNode(ISD::UNDEF, DL, ObjectVT));
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset, true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Load;
if (ObjectVT == MVT::i32) {
Load = DAG.getLoad(MVT::i32, DL, Root, FIPtr, MachinePointerInfo(), false, false, 0);
} else {
ISD::LoadExtType LoadOp = ISD::SEXTLOAD;
unsigned Offset = 0; // LE
FIPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, FIPtr,
DAG.getConstant(Offset, MVT::i32));
Load = DAG.getExtLoad(LoadOp, DL, MVT::i32, Root, FIPtr,
MachinePointerInfo(), ObjectVT, false, false, 0);
Load = DAG.getNode(ISD::TRUNCATE, DL, ObjectVT, Load);
}
InVals.push_back(Load);
}
ArgOffset += 4;
break;
case MVT::f32:
if (!Ins[i].Used) { // Argument is dead.
InVals.push_back(DAG.getNode(ISD::UNDEF, DL, ObjectVT));
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset, true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
InVals.push_back(DAG.getLoad(ObjectVT, DL, Root, FIPtr, MachinePointerInfo(), false, false, 0));
}
ArgOffset += 4;
break;
case MVT::i64:
if (!Ins[i].Used) { // Argument is dead.
InVals.push_back(DAG.getNode(ISD::UNDEF, DL, ObjectVT));
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(8, ArgOffset, true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue FIPtr4 = DAG.getNode(ISD::ADD, DL, MVT::i32, FIPtr, DAG.getConstant(4, MVT::i32));
SDValue Lo = DAG.getLoad(MVT::i32, DL, Root, FIPtr, MachinePointerInfo(), false, false, 0);
SDValue Hi = DAG.getLoad(MVT::i32, DL, Root, FIPtr4, MachinePointerInfo(), false, false, 0);
InVals.push_back(DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi));
}
ArgOffset += 8;
break;
case MVT::f64:
if (!Ins[i].Used) { // Argument is dead.
InVals.push_back(DAG.getNode(ISD::UNDEF, DL, ObjectVT));
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(8, ArgOffset, true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
InVals.push_back(DAG.getLoad(ObjectVT, DL, Root, FIPtr, MachinePointerInfo(), false, false, 0));
}
ArgOffset += 8;
break;
}
}
if (isVarArg) {
// Remember the vararg offset for the va_start implementation.
FuncInfo->setVarArgsFrameOffset(ArgOffset);
}
return Chain;
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDOperand AlphaTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::FORMAL_ARGUMENTS: return LowerFORMAL_ARGUMENTS(Op, DAG,
VarArgsBase,
VarArgsOffset);
case ISD::RET: return LowerRET(Op,DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SINT_TO_FP: {
assert(MVT::i64 == Op.getOperand(0).getValueType() &&
"Unhandled SINT_TO_FP type in custom expander!");
SDOperand LD;
bool isDouble = MVT::f64 == Op.getValueType();
LD = DAG.getNode(ISD::BIT_CONVERT, MVT::f64, Op.getOperand(0));
SDOperand FP = DAG.getNode(isDouble?AlphaISD::CVTQT_:AlphaISD::CVTQS_,
isDouble?MVT::f64:MVT::f32, LD);
return FP;
}
case ISD::FP_TO_SINT: {
bool isDouble = MVT::f64 == Op.getOperand(0).getValueType();
SDOperand src = Op.getOperand(0);
if (!isDouble) //Promote
src = DAG.getNode(ISD::FP_EXTEND, MVT::f64, src);
src = DAG.getNode(AlphaISD::CVTTQ_, MVT::f64, src);
return DAG.getNode(ISD::BIT_CONVERT, MVT::i64, src);
}
case ISD::ConstantPool: {
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDOperand CPI = DAG.getTargetConstantPool(C, MVT::i64, CP->getAlignment());
SDOperand Hi = DAG.getNode(AlphaISD::GPRelHi, MVT::i64, CPI,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
SDOperand Lo = DAG.getNode(AlphaISD::GPRelLo, MVT::i64, CPI, Hi);
return Lo;
}
case ISD::GlobalAddress: {
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDOperand GA = DAG.getTargetGlobalAddress(GV, MVT::i64, GSDN->getOffset());
// if (!GV->hasWeakLinkage() && !GV->isDeclaration() && !GV->hasLinkOnceLinkage()) {
if (GV->hasInternalLinkage()) {
SDOperand Hi = DAG.getNode(AlphaISD::GPRelHi, MVT::i64, GA,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
SDOperand Lo = DAG.getNode(AlphaISD::GPRelLo, MVT::i64, GA, Hi);
return Lo;
} else
return DAG.getNode(AlphaISD::RelLit, MVT::i64, GA,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
}
case ISD::ExternalSymbol: {
return DAG.getNode(AlphaISD::RelLit, MVT::i64,
DAG.getTargetExternalSymbol(cast<ExternalSymbolSDNode>(Op)
->getSymbol(), MVT::i64),
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
}
case ISD::UREM:
case ISD::SREM:
//Expand only on constant case
if (Op.getOperand(1).getOpcode() == ISD::Constant) {
MVT::ValueType VT = Op.Val->getValueType(0);
SDOperand Tmp1 = Op.Val->getOpcode() == ISD::UREM ?
BuildUDIV(Op.Val, DAG, NULL) :
BuildSDIV(Op.Val, DAG, NULL);
Tmp1 = DAG.getNode(ISD::MUL, VT, Tmp1, Op.getOperand(1));
Tmp1 = DAG.getNode(ISD::SUB, VT, Op.getOperand(0), Tmp1);
return Tmp1;
}
//fall through
case ISD::SDIV:
case ISD::UDIV:
if (MVT::isInteger(Op.getValueType())) {
if (Op.getOperand(1).getOpcode() == ISD::Constant)
return Op.getOpcode() == ISD::SDIV ? BuildSDIV(Op.Val, DAG, NULL)
: BuildUDIV(Op.Val, DAG, NULL);
const char* opstr = 0;
switch (Op.getOpcode()) {
case ISD::UREM: opstr = "__remqu"; break;
case ISD::SREM: opstr = "__remq"; break;
case ISD::UDIV: opstr = "__divqu"; break;
case ISD::SDIV: opstr = "__divq"; break;
}
SDOperand Tmp1 = Op.getOperand(0),
Tmp2 = Op.getOperand(1),
Addr = DAG.getExternalSymbol(opstr, MVT::i64);
return DAG.getNode(AlphaISD::DivCall, MVT::i64, Addr, Tmp1, Tmp2);
}
break;
case ISD::VAARG: {
SDOperand Chain = Op.getOperand(0);
SDOperand VAListP = Op.getOperand(1);
SrcValueSDNode *VAListS = cast<SrcValueSDNode>(Op.getOperand(2));
SDOperand Base = DAG.getLoad(MVT::i64, Chain, VAListP, VAListS->getValue(),
VAListS->getOffset());
SDOperand Tmp = DAG.getNode(ISD::ADD, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
SDOperand Offset = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Base.getValue(1),
Tmp, NULL, 0, MVT::i32);
SDOperand DataPtr = DAG.getNode(ISD::ADD, MVT::i64, Base, Offset);
if (MVT::isFloatingPoint(Op.getValueType()))
{
//if fp && Offset < 6*8, then subtract 6*8 from DataPtr
SDOperand FPDataPtr = DAG.getNode(ISD::SUB, MVT::i64, DataPtr,
DAG.getConstant(8*6, MVT::i64));
SDOperand CC = DAG.getSetCC(MVT::i64, Offset,
DAG.getConstant(8*6, MVT::i64), ISD::SETLT);
DataPtr = DAG.getNode(ISD::SELECT, MVT::i64, CC, FPDataPtr, DataPtr);
}
SDOperand NewOffset = DAG.getNode(ISD::ADD, MVT::i64, Offset,
DAG.getConstant(8, MVT::i64));
SDOperand Update = DAG.getTruncStore(Offset.getValue(1), NewOffset,
Tmp, NULL, 0, MVT::i32);
SDOperand Result;
if (Op.getValueType() == MVT::i32)
Result = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Update, DataPtr,
NULL, 0, MVT::i32);
else
Result = DAG.getLoad(Op.getValueType(), Update, DataPtr, NULL, 0);
return Result;
}
case ISD::VACOPY: {
SDOperand Chain = Op.getOperand(0);
SDOperand DestP = Op.getOperand(1);
SDOperand SrcP = Op.getOperand(2);
SrcValueSDNode *DestS = cast<SrcValueSDNode>(Op.getOperand(3));
SrcValueSDNode *SrcS = cast<SrcValueSDNode>(Op.getOperand(4));
SDOperand Val = DAG.getLoad(getPointerTy(), Chain, SrcP,
SrcS->getValue(), SrcS->getOffset());
SDOperand Result = DAG.getStore(Val.getValue(1), Val, DestP, DestS->getValue(),
DestS->getOffset());
SDOperand NP = DAG.getNode(ISD::ADD, MVT::i64, SrcP,
DAG.getConstant(8, MVT::i64));
Val = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Result, NP, NULL,0, MVT::i32);
SDOperand NPD = DAG.getNode(ISD::ADD, MVT::i64, DestP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(Val.getValue(1), Val, NPD, NULL, 0, MVT::i32);
}
case ISD::VASTART: {
SDOperand Chain = Op.getOperand(0);
SDOperand VAListP = Op.getOperand(1);
SrcValueSDNode *VAListS = cast<SrcValueSDNode>(Op.getOperand(2));
// vastart stores the address of the VarArgsBase and VarArgsOffset
SDOperand FR = DAG.getFrameIndex(VarArgsBase, MVT::i64);
SDOperand S1 = DAG.getStore(Chain, FR, VAListP, VAListS->getValue(),
VAListS->getOffset());
SDOperand SA2 = DAG.getNode(ISD::ADD, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(S1, DAG.getConstant(VarArgsOffset, MVT::i64),
SA2, NULL, 0, MVT::i32);
}
case ISD::RETURNADDR:
return DAG.getNode(AlphaISD::GlobalRetAddr, MVT::i64);
//FIXME: implement
case ISD::FRAMEADDR: break;
}
return SDOperand();
}
/// LowerFormalArguments - V8 uses a very simple ABI, where all values are
/// passed in either one or two GPRs, including FP values. TODO: we should
/// pass FP values in FP registers for fastcc functions.
SDValue
SparcTargetLowering::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();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
SparcMachineFunctionInfo *FuncInfo = MF.getInfo<SparcMachineFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_Sparc32);
static const unsigned ArgRegs[] = {
SP::I0, SP::I1, SP::I2, SP::I3, SP::I4, SP::I5
};
const unsigned *CurArgReg = ArgRegs, *ArgRegEnd = ArgRegs+6;
unsigned ArgOffset = 68;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
SDValue ArgValue;
CCValAssign &VA = ArgLocs[i];
// FIXME: We ignore the register assignments of AnalyzeFormalArguments
// because it doesn't know how to split a double into two i32 registers.
EVT ObjectVT = VA.getValVT();
switch (ObjectVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unhandled argument type!");
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
if (!Ins[i].Used) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
InVals.push_back(DAG.getUNDEF(ObjectVT));
} else if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VReg);
SDValue Arg = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
if (ObjectVT != MVT::i32) {
unsigned AssertOp = ISD::AssertSext;
Arg = DAG.getNode(AssertOp, dl, MVT::i32, Arg,
DAG.getValueType(ObjectVT));
Arg = DAG.getNode(ISD::TRUNCATE, dl, ObjectVT, Arg);
}
InVals.push_back(Arg);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset,
true, false);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Load;
if (ObjectVT == MVT::i32) {
Load = DAG.getLoad(MVT::i32, dl, Chain, FIPtr, NULL, 0,
false, false, 0);
} else {
ISD::LoadExtType LoadOp = ISD::SEXTLOAD;
// Sparc is big endian, so add an offset based on the ObjectVT.
unsigned Offset = 4-std::max(1U, ObjectVT.getSizeInBits()/8);
FIPtr = DAG.getNode(ISD::ADD, dl, MVT::i32, FIPtr,
DAG.getConstant(Offset, MVT::i32));
Load = DAG.getExtLoad(LoadOp, dl, MVT::i32, Chain, FIPtr,
NULL, 0, ObjectVT, false, false, 0);
Load = DAG.getNode(ISD::TRUNCATE, dl, ObjectVT, Load);
}
InVals.push_back(Load);
}
ArgOffset += 4;
break;
case MVT::f32:
if (!Ins[i].Used) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
InVals.push_back(DAG.getUNDEF(ObjectVT));
} else if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
// FP value is passed in an integer register.
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VReg);
SDValue Arg = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, Arg);
InVals.push_back(Arg);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset,
true, false);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Load = DAG.getLoad(MVT::f32, dl, Chain, FIPtr, NULL, 0,
false, false, 0);
InVals.push_back(Load);
}
ArgOffset += 4;
break;
case MVT::i64:
case MVT::f64:
if (!Ins[i].Used) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
if (CurArgReg < ArgRegEnd) ++CurArgReg;
InVals.push_back(DAG.getUNDEF(ObjectVT));
} else {
SDValue HiVal;
if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VRegHi = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VRegHi);
HiVal = DAG.getCopyFromReg(Chain, dl, VRegHi, MVT::i32);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset,
true, false);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
HiVal = DAG.getLoad(MVT::i32, dl, Chain, FIPtr, NULL, 0,
false, false, 0);
}
SDValue LoVal;
if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VRegLo = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VRegLo);
LoVal = DAG.getCopyFromReg(Chain, dl, VRegLo, MVT::i32);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset+4,
true, false);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
LoVal = DAG.getLoad(MVT::i32, dl, Chain, FIPtr, NULL, 0,
false, false, 0);
}
// Compose the two halves together into an i64 unit.
SDValue WholeValue =
DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, LoVal, HiVal);
// If we want a double, do a bit convert.
if (ObjectVT == MVT::f64)
WholeValue = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f64, WholeValue);
InVals.push_back(WholeValue);
}
ArgOffset += 8;
break;
}
}
// Store remaining ArgRegs to the stack if this is a varargs function.
if (isVarArg) {
// Remember the vararg offset for the va_start implementation.
FuncInfo->setVarArgsFrameOffset(ArgOffset);
std::vector<SDValue> OutChains;
for (; CurArgReg != ArgRegEnd; ++CurArgReg) {
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg, VReg);
SDValue Arg = DAG.getCopyFromReg(DAG.getRoot(), dl, VReg, MVT::i32);
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset,
true, false);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
OutChains.push_back(DAG.getStore(DAG.getRoot(), dl, Arg, FIPtr, NULL, 0,
false, false, 0));
ArgOffset += 4;
}
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size());
}
}
return Chain;
}
/// 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;
}
/// 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;
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDValue AlphaTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
switch (Op.getOpcode()) {
default: llvm_unreachable("Wasn't expecting to be able to lower this!");
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
switch (IntNo) {
default: break; // Don't custom lower most intrinsics.
case Intrinsic::alpha_umulh:
return DAG.getNode(ISD::MULHU, dl, MVT::i64,
Op.getOperand(1), Op.getOperand(2));
}
}
case ISD::SRL_PARTS: {
SDValue ShOpLo = Op.getOperand(0);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
SDValue bm = DAG.getNode(ISD::SUB, dl, MVT::i64,
DAG.getConstant(64, MVT::i64), ShAmt);
SDValue BMCC = DAG.getSetCC(dl, MVT::i64, bm,
DAG.getConstant(0, MVT::i64), ISD::SETLE);
// if 64 - shAmt <= 0
SDValue Hi_Neg = DAG.getConstant(0, MVT::i64);
SDValue ShAmt_Neg = DAG.getNode(ISD::SUB, dl, MVT::i64,
DAG.getConstant(0, MVT::i64), bm);
SDValue Lo_Neg = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpHi, ShAmt_Neg);
// else
SDValue carries = DAG.getNode(ISD::SHL, dl, MVT::i64, ShOpHi, bm);
SDValue Hi_Pos = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpHi, ShAmt);
SDValue Lo_Pos = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpLo, ShAmt);
Lo_Pos = DAG.getNode(ISD::OR, dl, MVT::i64, Lo_Pos, carries);
// Merge
SDValue Hi = DAG.getNode(ISD::SELECT, dl, MVT::i64, BMCC, Hi_Neg, Hi_Pos);
SDValue Lo = DAG.getNode(ISD::SELECT, dl, MVT::i64, BMCC, Lo_Neg, Lo_Pos);
SDValue Ops[2] = { Lo, Hi };
return DAG.getMergeValues(Ops, 2, dl);
}
// case ISD::SRA_PARTS:
// case ISD::SHL_PARTS:
case ISD::SINT_TO_FP: {
assert(Op.getOperand(0).getValueType() == MVT::i64 &&
"Unhandled SINT_TO_FP type in custom expander!");
SDValue LD;
bool isDouble = Op.getValueType() == MVT::f64;
LD = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f64, Op.getOperand(0));
SDValue FP = DAG.getNode(isDouble?AlphaISD::CVTQT_:AlphaISD::CVTQS_, dl,
isDouble?MVT::f64:MVT::f32, LD);
return FP;
}
case ISD::FP_TO_SINT: {
bool isDouble = Op.getOperand(0).getValueType() == MVT::f64;
SDValue src = Op.getOperand(0);
if (!isDouble) //Promote
src = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, src);
src = DAG.getNode(AlphaISD::CVTTQ_, dl, MVT::f64, src);
return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, src);
}
case ISD::ConstantPool: {
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDValue CPI = DAG.getTargetConstantPool(C, MVT::i64, CP->getAlignment());
// FIXME there isn't really any debug info here
SDValue Hi = DAG.getNode(AlphaISD::GPRelHi, dl, MVT::i64, CPI,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
SDValue Lo = DAG.getNode(AlphaISD::GPRelLo, dl, MVT::i64, CPI, Hi);
return Lo;
}
case ISD::GlobalTLSAddress:
llvm_unreachable("TLS not implemented for Alpha.");
case ISD::GlobalAddress: {
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDValue GA = DAG.getTargetGlobalAddress(GV, MVT::i64, GSDN->getOffset());
// FIXME there isn't really any debug info here
// if (!GV->hasWeakLinkage() && !GV->isDeclaration() && !GV->hasLinkOnceLinkage()) {
if (GV->hasLocalLinkage()) {
SDValue Hi = DAG.getNode(AlphaISD::GPRelHi, dl, MVT::i64, GA,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
SDValue Lo = DAG.getNode(AlphaISD::GPRelLo, dl, MVT::i64, GA, Hi);
return Lo;
} else
return DAG.getNode(AlphaISD::RelLit, dl, MVT::i64, GA,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
}
case ISD::ExternalSymbol: {
return DAG.getNode(AlphaISD::RelLit, dl, MVT::i64,
DAG.getTargetExternalSymbol(cast<ExternalSymbolSDNode>(Op)
->getSymbol(), MVT::i64),
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
}
case ISD::UREM:
case ISD::SREM:
//Expand only on constant case
if (Op.getOperand(1).getOpcode() == ISD::Constant) {
EVT VT = Op.getNode()->getValueType(0);
SDValue Tmp1 = Op.getNode()->getOpcode() == ISD::UREM ?
BuildUDIV(Op.getNode(), DAG, NULL) :
BuildSDIV(Op.getNode(), DAG, NULL);
Tmp1 = DAG.getNode(ISD::MUL, dl, VT, Tmp1, Op.getOperand(1));
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, Op.getOperand(0), Tmp1);
return Tmp1;
}
//fall through
case ISD::SDIV:
case ISD::UDIV:
if (Op.getValueType().isInteger()) {
if (Op.getOperand(1).getOpcode() == ISD::Constant)
return Op.getOpcode() == ISD::SDIV ? BuildSDIV(Op.getNode(), DAG, NULL)
: BuildUDIV(Op.getNode(), DAG, NULL);
const char* opstr = 0;
switch (Op.getOpcode()) {
case ISD::UREM: opstr = "__remqu"; break;
case ISD::SREM: opstr = "__remq"; break;
case ISD::UDIV: opstr = "__divqu"; break;
case ISD::SDIV: opstr = "__divq"; break;
}
SDValue Tmp1 = Op.getOperand(0),
Tmp2 = Op.getOperand(1),
Addr = DAG.getExternalSymbol(opstr, MVT::i64);
return DAG.getNode(AlphaISD::DivCall, dl, MVT::i64, Addr, Tmp1, Tmp2);
}
break;
case ISD::VAARG: {
SDValue Chain, DataPtr;
LowerVAARG(Op.getNode(), Chain, DataPtr, DAG);
SDValue Result;
if (Op.getValueType() == MVT::i32)
Result = DAG.getExtLoad(ISD::SEXTLOAD, dl, MVT::i64, Chain, DataPtr,
NULL, 0, MVT::i32);
else
Result = DAG.getLoad(Op.getValueType(), dl, Chain, DataPtr, NULL, 0);
return Result;
}
case ISD::VACOPY: {
SDValue Chain = Op.getOperand(0);
SDValue DestP = Op.getOperand(1);
SDValue SrcP = Op.getOperand(2);
const Value *DestS = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
const Value *SrcS = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
SDValue Val = DAG.getLoad(getPointerTy(), dl, Chain, SrcP, SrcS, 0);
SDValue Result = DAG.getStore(Val.getValue(1), dl, Val, DestP, DestS, 0);
SDValue NP = DAG.getNode(ISD::ADD, dl, MVT::i64, SrcP,
DAG.getConstant(8, MVT::i64));
Val = DAG.getExtLoad(ISD::SEXTLOAD, dl, MVT::i64, Result,
NP, NULL,0, MVT::i32);
SDValue NPD = DAG.getNode(ISD::ADD, dl, MVT::i64, DestP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(Val.getValue(1), dl, Val, NPD, NULL, 0, MVT::i32);
}
case ISD::VASTART: {
SDValue Chain = Op.getOperand(0);
SDValue VAListP = Op.getOperand(1);
const Value *VAListS = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// vastart stores the address of the VarArgsBase and VarArgsOffset
SDValue FR = DAG.getFrameIndex(VarArgsBase, MVT::i64);
SDValue S1 = DAG.getStore(Chain, dl, FR, VAListP, VAListS, 0);
SDValue SA2 = DAG.getNode(ISD::ADD, dl, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(S1, dl, DAG.getConstant(VarArgsOffset, MVT::i64),
SA2, NULL, 0, MVT::i32);
}
case ISD::RETURNADDR:
return DAG.getNode(AlphaISD::GlobalRetAddr, DebugLoc::getUnknownLoc(),
MVT::i64);
//FIXME: implement
case ISD::FRAMEADDR: break;
}
return 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 Op,
SelectionDAG &DAG) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
SDValue Root = Op.getOperand(0);
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() != 0;
unsigned CC = MF.getFunction()->getCallingConv();
DebugLoc dl = Op.getDebugLoc();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
CCInfo.AnalyzeFormalArguments(Op.getNode(), CC_MSP430);
assert(!isVarArg && "Varargs not supported yet");
SmallVector<SDValue, 16> ArgValues;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (VA.isRegLoc()) {
// Arguments passed in registers
MVT RegVT = VA.getLocVT();
switch (RegVT.getSimpleVT()) {
default:
cerr << "LowerFORMAL_ARGUMENTS Unhandled argument type: "
<< RegVT.getSimpleVT()
<< "\n";
abort();
case MVT::i16:
unsigned VReg =
RegInfo.createVirtualRegister(MSP430::GR16RegisterClass);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
SDValue ArgValue = DAG.getCopyFromReg(Root, 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);
ArgValues.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) {
cerr << "LowerFORMAL_ARGUMENTS Unhandled argument type: "
<< VA.getLocVT().getSimpleVT()
<< "\n";
}
// Create the frame index object for this incoming parameter...
int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset());
// Create the SelectionDAG nodes corresponding to a load
//from this parameter
SDValue FIN = DAG.getFrameIndex(FI, MVT::i16);
ArgValues.push_back(DAG.getLoad(VA.getLocVT(), dl, Root, FIN,
PseudoSourceValue::getFixedStack(FI), 0));
}
}
ArgValues.push_back(Root);
// Return the new list of results.
return DAG.getNode(ISD::MERGE_VALUES, dl, Op.getNode()->getVTList(),
&ArgValues[0], ArgValues.size()).getValue(Op.getResNo());
}
// 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);
}
// LowerCCCArguments - transform physical registers into virtual registers and
// generate load operations for arguments places on the stack.
SDValue LanaiTargetLowering::LowerCCCArguments(
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();
LanaiMachineFunctionInfo *LanaiMFI = MF.getInfo<LanaiMachineFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext());
if (CallConv == CallingConv::Fast) {
CCInfo.AnalyzeFormalArguments(Ins, CC_Lanai32_Fast);
} else {
CCInfo.AnalyzeFormalArguments(Ins, CC_Lanai32);
}
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) {
case MVT::i32: {
unsigned VReg = RegInfo.createVirtualRegister(&Lanai::GPRRegClass);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, VReg, RegVT);
// If this is an 8/16-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);
break;
}
default:
DEBUG(dbgs() << "LowerFormalArguments Unhandled argument type: "
<< RegVT.getSimpleVT().SimpleTy << "\n");
llvm_unreachable("unhandled argument type");
}
} else {
// Sanity check
assert(VA.isMemLoc());
// Load the argument to a virtual register
unsigned ObjSize = VA.getLocVT().getSizeInBits() / 8;
// Check that the argument fits in stack slot
if (ObjSize > 4) {
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::i32);
InVals.push_back(DAG.getLoad(
VA.getLocVT(), DL, Chain, FIN,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI),
false, false, false, 0));
}
}
// The Lanai ABI for returning structs by value requires that we copy
// the sret argument into rv for the return. Save the argument into
// a virtual register so that we can access it from the return points.
if (MF.getFunction()->hasStructRetAttr()) {
unsigned Reg = LanaiMFI->getSRetReturnReg();
if (!Reg) {
Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i32));
LanaiMFI->setSRetReturnReg(Reg);
}
SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), DL, Reg, InVals[0]);
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Copy, Chain);
}
if (IsVarArg) {
// Record the frame index of the first variable argument
// which is a value necessary to VASTART.
int FI = MFI->CreateFixedObject(4, CCInfo.getNextStackOffset(), true);
LanaiMFI->setVarArgsFrameIndex(FI);
}
return Chain;
}
SDValue Y86TargetLowering::
LowerFormalArguments(SDValue Chain,
CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SDLoc dl,
SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
// Gather info about the formals.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, MF, ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_Y86);
// Push corresponding SDValues for the arguments.
SDValue ArgValue;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (VA.isRegLoc()) {
EVT RegVT = VA.getLocVT();
const TargetRegisterClass *RC;
if (RegVT == MVT::i32)
RC = &Y86::GPRRegClass;
else
llvm_unreachable("Unknown EVT");
unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
} else {
assert(VA.isMemLoc());
EVT ValVT;
if (VA.getLocInfo() == CCValAssign::Indirect)
ValVT = VA.getLocVT();
else
ValVT = VA.getValVT();
// Create space in the stack (and an associated frame index)
// accordingly: Either the complete size of the object is it's
// a call by value or a pointer size otherwise.
MachineFrameInfo *MFI = MF.getFrameInfo();
ISD::ArgFlagsTy Flags = Ins[i].Flags;
if (Flags.isByVal()) {
unsigned Bytes = Flags.getByValSize();
if (Bytes == 0)
Bytes = 1; // Don't create zero-sized stack objects.
int FI = MFI->CreateFixedObject(Bytes, VA.getLocMemOffset(), false);
ArgValue = DAG.getFrameIndex(FI, getPointerTy());
} else {
int FI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), true);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue = DAG.getLoad(ValVT, dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, false, 0);
}
}
InVals.push_back(ArgValue);
}
return Chain;
}
SDValue
ARCTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
bool IsVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SDLoc &dl, SelectionDAG &DAG) const {
auto *AFI = DAG.getMachineFunction().getInfo<ARCFunctionInfo>();
MachineFrameInfo &MFI = DAG.getMachineFunction().getFrameInfo();
// CCValAssign - represent the assignment of
// the return value to a location
SmallVector<CCValAssign, 16> RVLocs;
// CCState - Info about the registers and stack slot.
CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext());
// Analyze return values.
if (!IsVarArg)
CCInfo.AllocateStack(AFI->getReturnStackOffset(), 4);
CCInfo.AnalyzeReturn(Outs, RetCC_ARC);
SDValue Flag;
SmallVector<SDValue, 4> RetOps(1, Chain);
SmallVector<SDValue, 4> MemOpChains;
// Handle return values that must be copied to memory.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
if (VA.isRegLoc())
continue;
assert(VA.isMemLoc());
if (IsVarArg) {
report_fatal_error("Can't return value from vararg function in memory");
}
int Offset = VA.getLocMemOffset();
unsigned ObjSize = VA.getLocVT().getStoreSize();
// Create the frame index object for the memory location.
int FI = MFI.CreateFixedObject(ObjSize, Offset, false);
// Create a SelectionDAG node corresponding to a store
// to this memory location.
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
MemOpChains.push_back(DAG.getStore(
Chain, dl, OutVals[i], FIN,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI)));
}
// Transform all store nodes into one single node because
// all stores are independent of each other.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
// Now handle return values copied to registers.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
if (!VA.isRegLoc())
continue;
// Copy the result values into the output registers.
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), OutVals[i], Flag);
// guarantee that all emitted copies are
// stuck together, avoiding something bad
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
RetOps[0] = Chain; // Update chain.
// Add the flag if we have it.
if (Flag.getNode())
RetOps.push_back(Flag);
// What to do with the RetOps?
return DAG.getNode(ARCISD::RET, dl, MVT::Other, RetOps);
}
/// 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;
}
/// LowerFormalArguments - transform physical registers into
/// virtual registers and generate load operations for
/// arguments places on the stack.
SDValue MBlazeTargetLowering::
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();
MBlazeFunctionInfo *MBlazeFI = MF.getInfo<MBlazeFunctionInfo>();
unsigned StackReg = MF.getTarget().getRegisterInfo()->getFrameRegister(MF);
MBlazeFI->setVarArgsFrameIndex(0);
// Used with vargs to acumulate store chains.
std::vector<SDValue> OutChains;
// Keep track of the last register used for arguments
unsigned ArgRegEnd = 0;
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_MBlaze);
SDValue StackPtr;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
// Arguments stored on registers
if (VA.isRegLoc()) {
MVT RegVT = VA.getLocVT();
ArgRegEnd = VA.getLocReg();
TargetRegisterClass *RC = 0;
if (RegVT == MVT::i32)
RC = MBlaze::GPRRegisterClass;
else if (RegVT == MVT::f32)
RC = MBlaze::GPRRegisterClass;
else
llvm_unreachable("RegVT not supported by LowerFormalArguments");
// Transform the arguments stored on
// physical registers into virtual ones
unsigned Reg = MF.addLiveIn(ArgRegEnd, RC);
SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
// If this is an 8 or 16-bit value, it has been passed promoted
// to 32 bits. Insert an assert[sz]ext to capture this, then
// truncate to the right size. If if is a floating point value
// then convert to the correct type.
if (VA.getLocInfo() != CCValAssign::Full) {
unsigned Opcode = 0;
if (VA.getLocInfo() == CCValAssign::SExt)
Opcode = ISD::AssertSext;
else if (VA.getLocInfo() == CCValAssign::ZExt)
Opcode = ISD::AssertZext;
if (Opcode)
ArgValue = DAG.getNode(Opcode, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
}
InVals.push_back(ArgValue);
} else { // VA.isRegLoc()
// sanity check
assert(VA.isMemLoc());
// The last argument is not a register
ArgRegEnd = 0;
// The stack pointer offset is relative to the caller stack frame.
// Since the real stack size is unknown here, a negative SPOffset
// is used so there's a way to adjust these offsets when the stack
// size get known (on EliminateFrameIndex). A dummy SPOffset is
// used instead of a direct negative address (which is recorded to
// be used on emitPrologue) to avoid mis-calc of the first stack
// offset on PEI::calculateFrameObjectOffsets.
// Arguments are always 32-bit.
unsigned ArgSize = VA.getLocVT().getSizeInBits()/8;
unsigned StackLoc = VA.getLocMemOffset() + 4;
int FI = MFI->CreateFixedObject(ArgSize, 0, true);
MBlazeFI->recordLoadArgsFI(FI, -StackLoc);
// Create load nodes to retrieve arguments from the stack
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, 0));
}
}
// To meet ABI, when VARARGS are passed on registers, the registers
// must have their values written to the caller stack frame. If the last
// argument was placed in the stack, there's no need to save any register.
if ((isVarArg) && ArgRegEnd) {
if (StackPtr.getNode() == 0)
StackPtr = DAG.getRegister(StackReg, getPointerTy());
// The last register argument that must be saved is MBlaze::R10
TargetRegisterClass *RC = MBlaze::GPRRegisterClass;
unsigned Begin = MBlazeRegisterInfo::getRegisterNumbering(MBlaze::R5);
unsigned Start = MBlazeRegisterInfo::getRegisterNumbering(ArgRegEnd+1);
unsigned End = MBlazeRegisterInfo::getRegisterNumbering(MBlaze::R10);
unsigned StackLoc = Start - Begin + 1;
for (; Start <= End; ++Start, ++StackLoc) {
unsigned Reg = MBlazeRegisterInfo::getRegisterFromNumbering(Start);
unsigned LiveReg = MF.addLiveIn(Reg, RC);
SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, LiveReg, MVT::i32);
int FI = MFI->CreateFixedObject(4, 0, true);
MBlazeFI->recordStoreVarArgsFI(FI, -(StackLoc*4));
SDValue PtrOff = DAG.getFrameIndex(FI, getPointerTy());
OutChains.push_back(DAG.getStore(Chain, dl, ArgValue, PtrOff,
MachinePointerInfo(),
false, false, 0));
// Record the frame index of the first variable argument
// which is a value necessary to VASTART.
if (!MBlazeFI->getVarArgsFrameIndex())
MBlazeFI->setVarArgsFrameIndex(FI);
}
}
// All stores are grouped in one node to allow the matching between
// the size of Ins and InVals. This only happens when on varg functions
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size());
}
return Chain;
}
SDValue
AlphaTargetLowering::LowerFormalArguments(SDValue Chain,
unsigned CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
unsigned args_int[] = {
Alpha::R16, Alpha::R17, Alpha::R18, Alpha::R19, Alpha::R20, Alpha::R21};
unsigned args_float[] = {
Alpha::F16, Alpha::F17, Alpha::F18, Alpha::F19, Alpha::F20, Alpha::F21};
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
SDValue argt;
EVT ObjectVT = Ins[ArgNo].VT;
SDValue ArgVal;
if (ArgNo < 6) {
switch (ObjectVT.getSimpleVT().SimpleTy) {
default:
assert(false && "Invalid value type!");
case MVT::f64:
args_float[ArgNo] = AddLiveIn(MF, args_float[ArgNo],
&Alpha::F8RCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, args_float[ArgNo], ObjectVT);
break;
case MVT::f32:
args_float[ArgNo] = AddLiveIn(MF, args_float[ArgNo],
&Alpha::F4RCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, args_float[ArgNo], ObjectVT);
break;
case MVT::i64:
args_int[ArgNo] = AddLiveIn(MF, args_int[ArgNo],
&Alpha::GPRCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, args_int[ArgNo], MVT::i64);
break;
}
} else { //more args
// Create the frame index object for this incoming parameter...
int FI = MFI->CreateFixedObject(8, 8 * (ArgNo - 6));
// Create the SelectionDAG nodes corresponding to a load
//from this parameter
SDValue FIN = DAG.getFrameIndex(FI, MVT::i64);
ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, NULL, 0);
}
InVals.push_back(ArgVal);
}
// If the functions takes variable number of arguments, copy all regs to stack
if (isVarArg) {
VarArgsOffset = Ins.size() * 8;
std::vector<SDValue> LS;
for (int i = 0; i < 6; ++i) {
if (TargetRegisterInfo::isPhysicalRegister(args_int[i]))
args_int[i] = AddLiveIn(MF, args_int[i], &Alpha::GPRCRegClass);
SDValue argt = DAG.getCopyFromReg(Chain, dl, args_int[i], MVT::i64);
int FI = MFI->CreateFixedObject(8, -8 * (6 - i));
if (i == 0) VarArgsBase = FI;
SDValue SDFI = DAG.getFrameIndex(FI, MVT::i64);
LS.push_back(DAG.getStore(Chain, dl, argt, SDFI, NULL, 0));
if (TargetRegisterInfo::isPhysicalRegister(args_float[i]))
args_float[i] = AddLiveIn(MF, args_float[i], &Alpha::F8RCRegClass);
argt = DAG.getCopyFromReg(Chain, dl, args_float[i], MVT::f64);
FI = MFI->CreateFixedObject(8, - 8 * (12 - i));
SDFI = DAG.getFrameIndex(FI, MVT::i64);
LS.push_back(DAG.getStore(Chain, dl, argt, SDFI, NULL, 0));
}
//Set up a token factor with all the stack traffic
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &LS[0], LS.size());
}
return Chain;
}
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;
}
/// LowerFormalArguments - V8 uses a very simple ABI, where all values are
/// passed in either one or two GPRs, including FP values. TODO: we should
/// pass FP values in FP registers for fastcc functions.
SDValue
SparcTargetLowering::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();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
SparcMachineFunctionInfo *FuncInfo = MF.getInfo<SparcMachineFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_Sparc32);
const unsigned StackOffset = 92;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (i == 0 && Ins[i].Flags.isSRet()) {
//Get SRet from [%fp+64]
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, 64, true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Arg = DAG.getLoad(MVT::i32, dl, Chain, FIPtr,
MachinePointerInfo(),
false, false, 0);
InVals.push_back(Arg);
continue;
}
if (VA.isRegLoc()) {
if (VA.needsCustom()) {
assert(VA.getLocVT() == MVT::f64);
unsigned VRegHi = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(VA.getLocReg(), VRegHi);
SDValue HiVal = DAG.getCopyFromReg(Chain, dl, VRegHi, MVT::i32);
assert(i+1 < e);
CCValAssign &NextVA = ArgLocs[++i];
SDValue LoVal;
if (NextVA.isMemLoc()) {
int FrameIdx = MF.getFrameInfo()->
CreateFixedObject(4, StackOffset+NextVA.getLocMemOffset(),true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
LoVal = DAG.getLoad(MVT::i32, dl, Chain, FIPtr,
MachinePointerInfo(),
false, false, 0);
} else {
unsigned loReg = MF.addLiveIn(NextVA.getLocReg(),
&SP::IntRegsRegClass);
LoVal = DAG.getCopyFromReg(Chain, dl, loReg, MVT::i32);
}
SDValue WholeValue =
DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, LoVal, HiVal);
WholeValue = DAG.getNode(ISD::BITCAST, dl, MVT::f64, WholeValue);
InVals.push_back(WholeValue);
continue;
}
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(VA.getLocReg(), VReg);
SDValue Arg = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
if (VA.getLocVT() == MVT::f32)
Arg = DAG.getNode(ISD::BITCAST, dl, MVT::f32, Arg);
else if (VA.getLocVT() != MVT::i32) {
Arg = DAG.getNode(ISD::AssertSext, dl, MVT::i32, Arg,
DAG.getValueType(VA.getLocVT()));
Arg = DAG.getNode(ISD::TRUNCATE, dl, VA.getLocVT(), Arg);
}
InVals.push_back(Arg);
continue;
}
assert(VA.isMemLoc());
unsigned Offset = VA.getLocMemOffset()+StackOffset;
if (VA.needsCustom()) {
assert(VA.getValVT() == MVT::f64);
//If it is double-word aligned, just load.
if (Offset % 8 == 0) {
int FI = MF.getFrameInfo()->CreateFixedObject(8,
Offset,
true);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
SDValue Load = DAG.getLoad(VA.getValVT(), dl, Chain, FIPtr,
MachinePointerInfo(),
false,false, 0);
InVals.push_back(Load);
continue;
}
int FI = MF.getFrameInfo()->CreateFixedObject(4,
Offset,
true);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
SDValue HiVal = DAG.getLoad(MVT::i32, dl, Chain, FIPtr,
MachinePointerInfo(),
false, false, 0);
int FI2 = MF.getFrameInfo()->CreateFixedObject(4,
Offset+4,
true);
SDValue FIPtr2 = DAG.getFrameIndex(FI2, getPointerTy());
SDValue LoVal = DAG.getLoad(MVT::i32, dl, Chain, FIPtr2,
MachinePointerInfo(),
false, false, 0);
SDValue WholeValue =
DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, LoVal, HiVal);
WholeValue = DAG.getNode(ISD::BITCAST, dl, MVT::f64, WholeValue);
InVals.push_back(WholeValue);
continue;
}
int FI = MF.getFrameInfo()->CreateFixedObject(4,
Offset,
true);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
SDValue Load ;
if (VA.getValVT() == MVT::i32 || VA.getValVT() == MVT::f32) {
Load = DAG.getLoad(VA.getValVT(), dl, Chain, FIPtr,
MachinePointerInfo(),
false, false, 0);
} else {
ISD::LoadExtType LoadOp = ISD::SEXTLOAD;
// Sparc is big endian, so add an offset based on the ObjectVT.
unsigned Offset = 4-std::max(1U, VA.getValVT().getSizeInBits()/8);
FIPtr = DAG.getNode(ISD::ADD, dl, MVT::i32, FIPtr,
DAG.getConstant(Offset, MVT::i32));
Load = DAG.getExtLoad(LoadOp, dl, MVT::i32, Chain, FIPtr,
MachinePointerInfo(),
VA.getValVT(), false, false,0);
Load = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Load);
}
InVals.push_back(Load);
}
if (MF.getFunction()->hasStructRetAttr()) {
//Copy the SRet Argument to SRetReturnReg
SparcMachineFunctionInfo *SFI = MF.getInfo<SparcMachineFunctionInfo>();
unsigned Reg = SFI->getSRetReturnReg();
if (!Reg) {
Reg = MF.getRegInfo().createVirtualRegister(&SP::IntRegsRegClass);
SFI->setSRetReturnReg(Reg);
}
SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Chain);
}
// Store remaining ArgRegs to the stack if this is a varargs function.
if (isVarArg) {
static const unsigned ArgRegs[] = {
SP::I0, SP::I1, SP::I2, SP::I3, SP::I4, SP::I5
};
unsigned NumAllocated = CCInfo.getFirstUnallocated(ArgRegs, 6);
const unsigned *CurArgReg = ArgRegs+NumAllocated, *ArgRegEnd = ArgRegs+6;
unsigned ArgOffset = CCInfo.getNextStackOffset();
if (NumAllocated == 6)
ArgOffset += StackOffset;
else {
assert(!ArgOffset);
ArgOffset = 68+4*NumAllocated;
}
// Remember the vararg offset for the va_start implementation.
FuncInfo->setVarArgsFrameOffset(ArgOffset);
std::vector<SDValue> OutChains;
for (; CurArgReg != ArgRegEnd; ++CurArgReg) {
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg, VReg);
SDValue Arg = DAG.getCopyFromReg(DAG.getRoot(), dl, VReg, MVT::i32);
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset,
true);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
OutChains.push_back(DAG.getStore(DAG.getRoot(), dl, Arg, FIPtr,
MachinePointerInfo(),
false, false, 0));
ArgOffset += 4;
}
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size());
}
}
return Chain;
}
SDValue
Cpu0TargetLowering::LowerCall(SDValue InChain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
bool doesNotRet, bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
#if 1
// Cpu0 target does not yet support tail call optimization.
isTailCall = false;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
const TargetFrameLowering *TFL = MF.getTarget().getFrameLowering();
bool IsPIC = getTargetMachine().getRelocationModel() == Reloc::PIC_;
Cpu0FunctionInfo *Cpu0FI = MF.getInfo<Cpu0FunctionInfo>();
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeCallOperands(Outs, CC_Cpu0);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NextStackOffset = CCInfo.getNextStackOffset();
// Chain is the output chain of the last Load/Store or CopyToReg node.
// ByValChain is the output chain of the last Memcpy node created for copying
// byval arguments to the stack.
SDValue Chain, CallSeqStart, ByValChain;
SDValue NextStackOffsetVal = DAG.getIntPtrConstant(NextStackOffset, true);
Chain = CallSeqStart = DAG.getCALLSEQ_START(InChain, NextStackOffsetVal);
ByValChain = InChain;
#if 0
// If this is the first call, create a stack frame object that points to
// a location to which .cprestore saves $gp.
if (IsO32 && IsPIC && Cpu0FI->globalBaseRegFixed() && !Cpu0FI->getGPFI())
Cpu0FI->setGPFI(MFI->CreateFixedObject(4, 0, true));
#endif
// Get the frame index of the stack frame object that points to the location
// of dynamically allocated area on the stack.
int DynAllocFI = Cpu0FI->getDynAllocFI();
#if 0
// Update size of the maximum argument space.
// For O32, a minimum of four words (16 bytes) of argument space is
// allocated.
if (IsO32)
NextStackOffset = std::max(NextStackOffset, (unsigned)16);
#endif
unsigned MaxCallFrameSize = Cpu0FI->getMaxCallFrameSize();
if (MaxCallFrameSize < NextStackOffset) {
Cpu0FI->setMaxCallFrameSize(NextStackOffset);
// Set the offsets relative to $sp of the $gp restore slot and dynamically
// allocated stack space. These offsets must be aligned to a boundary
// determined by the stack alignment of the ABI.
unsigned StackAlignment = TFL->getStackAlignment();
NextStackOffset = (NextStackOffset + StackAlignment - 1) /
StackAlignment * StackAlignment;
if (Cpu0FI->needGPSaveRestore())
MFI->setObjectOffset(Cpu0FI->getGPFI(), NextStackOffset);
MFI->setObjectOffset(DynAllocFI, NextStackOffset);
}
// With EABI is it possible to have 16 args on registers.
SmallVector<std::pair<unsigned, SDValue>, 16> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
int FirstFI = -MFI->getNumFixedObjects() - 1, LastFI = 0;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
SDValue Arg = OutVals[i];
CCValAssign &VA = ArgLocs[i];
MVT ValVT = VA.getValVT(), LocVT = VA.getLocVT();
ISD::ArgFlagsTy Flags = Outs[i].Flags;
// ByVal Arg.
if (Flags.isByVal()) {
assert(Flags.getByValSize() &&
"ByVal args of size 0 should have been ignored by front-end.");
#if 0
if (IsO32)
WriteByValArg(ByValChain, Chain, dl, RegsToPass, MemOpChains, LastFI,
MFI, DAG, Arg, VA, Flags, getPointerTy(),
Subtarget->isLittle());
#endif
#if 0
else
PassByValArg64(ByValChain, Chain, dl, RegsToPass, MemOpChains, LastFI,
MFI, DAG, Arg, VA, Flags, getPointerTy(),
Subtarget->isLittle());
#endif
continue;
}
// Promote the value if needed.
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full:
#if 0
if (VA.isRegLoc()) {
if ((ValVT == MVT::f32 && LocVT == MVT::i32) ||
(ValVT == MVT::f64 && LocVT == MVT::i64))
Arg = DAG.getNode(ISD::BITCAST, dl, LocVT, Arg);
else if (ValVT == MVT::f64 && LocVT == MVT::i32) {
SDValue Lo = DAG.getNode(Cpu0ISD::ExtractElementF64, dl, MVT::i32,
Arg, DAG.getConstant(0, MVT::i32));
SDValue Hi = DAG.getNode(Cpu0ISD::ExtractElementF64, dl, MVT::i32,
Arg, DAG.getConstant(1, MVT::i32));
if (!Subtarget->isLittle())
std::swap(Lo, Hi);
unsigned LocRegLo = VA.getLocReg();
unsigned LocRegHigh = getNextIntArgReg(LocRegLo);
RegsToPass.push_back(std::make_pair(LocRegLo, Lo));
RegsToPass.push_back(std::make_pair(LocRegHigh, Hi));
continue;
}
}
#else
assert("CCValAssign::Full:"); // Gamma debug
#endif
break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, LocVT, Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, LocVT, Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, LocVT, 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));
continue;
}
// Register can't get to this point...
assert(VA.isMemLoc());
// Create the frame index object for this incoming parameter
LastFI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), true);
SDValue PtrOff = DAG.getFrameIndex(LastFI, getPointerTy());
// emit ISD::STORE whichs stores the
// parameter value to a stack Location
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(), false, false, 0));
}
// Extend range of indices of frame objects for outgoing arguments that were
// created during this function call. Skip this step if no such objects were
// created.
if (LastFI)
Cpu0FI->extendOutArgFIRange(FirstFI, LastFI);
// If a memcpy has been created to copy a byval arg to a stack, replace the
// chain input of CallSeqStart with ByValChain.
if (InChain != ByValChain)
DAG.UpdateNodeOperands(CallSeqStart.getNode(), ByValChain,
NextStackOffsetVal);
// 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());
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
// direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
// node so that legalize doesn't hack it.
unsigned char OpFlag;
#if 0 // cpu0 int 32 only
bool IsPICCall = (IsN64 || IsPIC); // true if calls are translated to jalr $25
#else
bool IsPICCall = IsPIC; // true if calls are translated to jalr $25
#endif
bool GlobalOrExternal = false;
SDValue CalleeLo;
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
if (IsPICCall && G->getGlobal()->hasInternalLinkage()) {
OpFlag = Cpu0II::MO_GOT;
#if 0
unsigned char LoFlag = IsO32 ? Cpu0II::MO_ABS_LO : Cpu0II::MO_GOT_OFST;
#else
unsigned char LoFlag = Cpu0II::MO_ABS_LO;
#endif
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, getPointerTy(), 0,
OpFlag);
CalleeLo = DAG.getTargetGlobalAddress(G->getGlobal(), dl, getPointerTy(),
0, LoFlag);
} else {
OpFlag = IsPICCall ? Cpu0II::MO_GOT_CALL : Cpu0II::MO_NO_FLAG;
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl,
getPointerTy(), 0, OpFlag);
}
GlobalOrExternal = true;
}
else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
if (!IsPIC) // static
OpFlag = Cpu0II::MO_NO_FLAG;
else // O32 & PIC
OpFlag = Cpu0II::MO_GOT_CALL;
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(),
OpFlag);
GlobalOrExternal = true;
}
SDValue InFlag;
// Create nodes that load address of callee and copy it to T9
if (IsPICCall) {
if (GlobalOrExternal) {
// Load callee address
Callee = DAG.getNode(Cpu0ISD::Wrapper, dl, getPointerTy(),
GetGlobalReg(DAG, getPointerTy()), Callee);
SDValue LoadValue = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
Callee, MachinePointerInfo::getGOT(),
false, false, false, 0);
// Use GOT+LO if callee has internal linkage.
if (CalleeLo.getNode()) {
SDValue Lo = DAG.getNode(Cpu0ISD::Lo, dl, getPointerTy(), CalleeLo);
Callee = DAG.getNode(ISD::ADD, dl, getPointerTy(), LoadValue, Lo);
} else
Callee = LoadValue;
}
}
// T9 should contain the address of the callee function if
// -reloction-model=pic or it is an indirect call.
if (IsPICCall || !GlobalOrExternal) {
// copy to T9
unsigned T9Reg = Cpu0::T9;
Chain = DAG.getCopyToReg(Chain, dl, T9Reg, Callee, SDValue(0, 0));
InFlag = Chain.getValue(1);
Callee = DAG.getRegister(T9Reg, getPointerTy());
}
// 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);
}
// Cpu0JmpLink = #chain, #target_address, #opt_in_flags...
// = Chain, Callee, Reg#1, Reg#2, ...
//
// 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()));
// Add a register mask operand representing the call-preserved registers.
const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
if (InFlag.getNode())
Ops.push_back(InFlag);
Chain = DAG.getNode(Cpu0ISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getIntPtrConstant(NextStackOffset, true),
DAG.getIntPtrConstant(0, 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);
#else
return InChain;
#endif
}
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;
}
/// LowerFormalArguments - transform physical registers into virtual registers
/// and generate load operations for arguments places on the stack.
SDValue
Cpu0TargetLowering::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();
Cpu0FunctionInfo *Cpu0FI = MF.getInfo<Cpu0FunctionInfo>();
Cpu0FI->setVarArgsFrameIndex(0);
// Used with vargs to acumulate store chains.
std::vector<SDValue> OutChains;
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_Cpu0);
Function::const_arg_iterator FuncArg =
DAG.getMachineFunction().getFunction()->arg_begin();
int LastFI = 0;// Cpu0FI->LastInArgFI is 0 at the entry of this function.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i, ++FuncArg) {
CCValAssign &VA = ArgLocs[i];
EVT ValVT = VA.getValVT();
ISD::ArgFlagsTy Flags = Ins[i].Flags;
bool IsRegLoc = VA.isRegLoc();
if (Flags.isByVal()) {
assert(Flags.getByValSize() &&
"ByVal args of size 0 should have been ignored by front-end.");
continue;
}
{ // VA.isRegLoc()
// sanity check
assert(VA.isMemLoc());
// The stack pointer offset is relative to the caller stack frame.
LastFI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), true);
// Create load nodes to retrieve arguments from the stack
SDValue FIN = DAG.getFrameIndex(LastFI, getPointerTy());
InVals.push_back(DAG.getLoad(ValVT, dl, Chain, FIN,
MachinePointerInfo::getFixedStack(LastFI),
false, false, false, 0));
}
}
Cpu0FI->setLastInArgFI(LastFI);
#if 1
// All stores are grouped in one node to allow the matching between
// the size of Ins and InVals. This only happens when on varg functions
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size());
}
#endif
return Chain;
}
static SDOperand LowerFORMAL_ARGUMENTS(SDOperand Op, SelectionDAG &DAG,
int &VarArgsBase,
int &VarArgsOffset) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
std::vector<SDOperand> ArgValues;
SDOperand Root = Op.getOperand(0);
AddLiveIn(MF, Alpha::R29, &Alpha::GPRCRegClass); //GP
AddLiveIn(MF, Alpha::R26, &Alpha::GPRCRegClass); //RA
unsigned args_int[] = {
Alpha::R16, Alpha::R17, Alpha::R18, Alpha::R19, Alpha::R20, Alpha::R21};
unsigned args_float[] = {
Alpha::F16, Alpha::F17, Alpha::F18, Alpha::F19, Alpha::F20, Alpha::F21};
for (unsigned ArgNo = 0, e = Op.Val->getNumValues()-1; ArgNo != e; ++ArgNo) {
SDOperand argt;
MVT::ValueType ObjectVT = Op.getValue(ArgNo).getValueType();
SDOperand ArgVal;
if (ArgNo < 6) {
switch (ObjectVT) {
default:
cerr << "Unknown Type " << ObjectVT << "\n";
abort();
case MVT::f64:
args_float[ArgNo] = AddLiveIn(MF, args_float[ArgNo],
&Alpha::F8RCRegClass);
ArgVal = DAG.getCopyFromReg(Root, args_float[ArgNo], ObjectVT);
break;
case MVT::f32:
args_float[ArgNo] = AddLiveIn(MF, args_float[ArgNo],
&Alpha::F4RCRegClass);
ArgVal = DAG.getCopyFromReg(Root, args_float[ArgNo], ObjectVT);
break;
case MVT::i64:
args_int[ArgNo] = AddLiveIn(MF, args_int[ArgNo],
&Alpha::GPRCRegClass);
ArgVal = DAG.getCopyFromReg(Root, args_int[ArgNo], MVT::i64);
break;
}
} else { //more args
// Create the frame index object for this incoming parameter...
int FI = MFI->CreateFixedObject(8, 8 * (ArgNo - 6));
// Create the SelectionDAG nodes corresponding to a load
//from this parameter
SDOperand FIN = DAG.getFrameIndex(FI, MVT::i64);
ArgVal = DAG.getLoad(ObjectVT, Root, FIN, NULL, 0);
}
ArgValues.push_back(ArgVal);
}
// If the functions takes variable number of arguments, copy all regs to stack
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
if (isVarArg) {
VarArgsOffset = (Op.Val->getNumValues()-1) * 8;
std::vector<SDOperand> LS;
for (int i = 0; i < 6; ++i) {
if (MRegisterInfo::isPhysicalRegister(args_int[i]))
args_int[i] = AddLiveIn(MF, args_int[i], &Alpha::GPRCRegClass);
SDOperand argt = DAG.getCopyFromReg(Root, args_int[i], MVT::i64);
int FI = MFI->CreateFixedObject(8, -8 * (6 - i));
if (i == 0) VarArgsBase = FI;
SDOperand SDFI = DAG.getFrameIndex(FI, MVT::i64);
LS.push_back(DAG.getStore(Root, argt, SDFI, NULL, 0));
if (MRegisterInfo::isPhysicalRegister(args_float[i]))
args_float[i] = AddLiveIn(MF, args_float[i], &Alpha::F8RCRegClass);
argt = DAG.getCopyFromReg(Root, args_float[i], MVT::f64);
FI = MFI->CreateFixedObject(8, - 8 * (12 - i));
SDFI = DAG.getFrameIndex(FI, MVT::i64);
LS.push_back(DAG.getStore(Root, argt, SDFI, NULL, 0));
}
//Set up a token factor with all the stack traffic
Root = DAG.getNode(ISD::TokenFactor, MVT::Other, &LS[0], LS.size());
}
ArgValues.push_back(Root);
// Return the new list of results.
std::vector<MVT::ValueType> RetVT(Op.Val->value_begin(),
Op.Val->value_end());
return DAG.getNode(ISD::MERGE_VALUES, RetVT, &ArgValues[0], ArgValues.size());
}
/// LowerArguments - V8 uses a very simple ABI, where all values are passed in
/// either one or two GPRs, including FP values. TODO: we should pass FP values
/// in FP registers for fastcc functions.
void
SparcTargetLowering::LowerArguments(Function &F, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &ArgValues,
DebugLoc dl) {
MachineFunction &MF = DAG.getMachineFunction();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
static const unsigned ArgRegs[] = {
SP::I0, SP::I1, SP::I2, SP::I3, SP::I4, SP::I5
};
const unsigned *CurArgReg = ArgRegs, *ArgRegEnd = ArgRegs+6;
unsigned ArgOffset = 68;
SDValue Root = DAG.getRoot();
std::vector<SDValue> OutChains;
for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E; ++I) {
MVT ObjectVT = getValueType(I->getType());
switch (ObjectVT.getSimpleVT()) {
default: assert(0 && "Unhandled argument type!");
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
if (I->use_empty()) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
ArgValues.push_back(DAG.getUNDEF(ObjectVT));
} else if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VReg);
SDValue Arg = DAG.getCopyFromReg(Root, dl, VReg, MVT::i32);
if (ObjectVT != MVT::i32) {
unsigned AssertOp = ISD::AssertSext;
Arg = DAG.getNode(AssertOp, dl, MVT::i32, Arg,
DAG.getValueType(ObjectVT));
Arg = DAG.getNode(ISD::TRUNCATE, dl, ObjectVT, Arg);
}
ArgValues.push_back(Arg);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Load;
if (ObjectVT == MVT::i32) {
Load = DAG.getLoad(MVT::i32, dl, Root, FIPtr, NULL, 0);
} else {
ISD::LoadExtType LoadOp = ISD::SEXTLOAD;
// Sparc is big endian, so add an offset based on the ObjectVT.
unsigned Offset = 4-std::max(1U, ObjectVT.getSizeInBits()/8);
FIPtr = DAG.getNode(ISD::ADD, dl, MVT::i32, FIPtr,
DAG.getConstant(Offset, MVT::i32));
Load = DAG.getExtLoad(LoadOp, dl, MVT::i32, Root, FIPtr,
NULL, 0, ObjectVT);
Load = DAG.getNode(ISD::TRUNCATE, dl, ObjectVT, Load);
}
ArgValues.push_back(Load);
}
ArgOffset += 4;
break;
case MVT::f32:
if (I->use_empty()) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
ArgValues.push_back(DAG.getUNDEF(ObjectVT));
} else if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
// FP value is passed in an integer register.
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VReg);
SDValue Arg = DAG.getCopyFromReg(Root, dl, VReg, MVT::i32);
Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, Arg);
ArgValues.push_back(Arg);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
SDValue Load = DAG.getLoad(MVT::f32, dl, Root, FIPtr, NULL, 0);
ArgValues.push_back(Load);
}
ArgOffset += 4;
break;
case MVT::i64:
case MVT::f64:
if (I->use_empty()) { // Argument is dead.
if (CurArgReg < ArgRegEnd) ++CurArgReg;
if (CurArgReg < ArgRegEnd) ++CurArgReg;
ArgValues.push_back(DAG.getUNDEF(ObjectVT));
} else {
SDValue HiVal;
if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VRegHi = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VRegHi);
HiVal = DAG.getCopyFromReg(Root, dl, VRegHi, MVT::i32);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
HiVal = DAG.getLoad(MVT::i32, dl, Root, FIPtr, NULL, 0);
}
SDValue LoVal;
if (CurArgReg < ArgRegEnd) { // Lives in an incoming GPR
unsigned VRegLo = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg++, VRegLo);
LoVal = DAG.getCopyFromReg(Root, dl, VRegLo, MVT::i32);
} else {
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset+4);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
LoVal = DAG.getLoad(MVT::i32, dl, Root, FIPtr, NULL, 0);
}
// Compose the two halves together into an i64 unit.
SDValue WholeValue =
DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, LoVal, HiVal);
// If we want a double, do a bit convert.
if (ObjectVT == MVT::f64)
WholeValue = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f64, WholeValue);
ArgValues.push_back(WholeValue);
}
ArgOffset += 8;
break;
}
}
// Store remaining ArgRegs to the stack if this is a varargs function.
if (F.isVarArg()) {
// Remember the vararg offset for the va_start implementation.
VarArgsFrameOffset = ArgOffset;
for (; CurArgReg != ArgRegEnd; ++CurArgReg) {
unsigned VReg = RegInfo.createVirtualRegister(&SP::IntRegsRegClass);
MF.getRegInfo().addLiveIn(*CurArgReg, VReg);
SDValue Arg = DAG.getCopyFromReg(DAG.getRoot(), dl, VReg, MVT::i32);
int FrameIdx = MF.getFrameInfo()->CreateFixedObject(4, ArgOffset);
SDValue FIPtr = DAG.getFrameIndex(FrameIdx, MVT::i32);
OutChains.push_back(DAG.getStore(DAG.getRoot(), dl, Arg, FIPtr, NULL, 0));
ArgOffset += 4;
}
}
if (!OutChains.empty())
DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size()));
}