static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
if (TM.getSubtargetImpl()->isDarwin())
return new TargetLoweringObjectFileMachO();
- return new TargetLoweringObjectFileELF(true);
+ return new TargetLoweringObjectFileELF();
}
// We cannot sextinreg(i1). Expand to shifts.
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
- // Support label based line numbers.
- setOperationAction(ISD::DBG_STOPPOINT, MVT::Other, Expand);
- setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
-
setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
setOperationAction(ISD::EHSELECTION, MVT::i64, Expand);
setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
// appropriate instructions to materialize the address.
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
+ setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
setOperationAction(ISD::JumpTable, MVT::i32, Custom);
setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
+ setOperationAction(ISD::BlockAddress, MVT::i64, Custom);
setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
setOperationAction(ISD::JumpTable, MVT::i64, Custom);
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
- // VAARG is custom lowered with the SVR4 ABI
- if (TM.getSubtarget<PPCSubtarget>().isSVR4ABI())
+ // VAARG is custom lowered with the 32-bit SVR4 ABI.
+ if ( TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
+ && !TM.getSubtarget<PPCSubtarget>().isPPC64())
setOperationAction(ISD::VAARG, MVT::Other, Custom);
else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
case PPCISD::VPERM: return "PPCISD::VPERM";
case PPCISD::Hi: return "PPCISD::Hi";
case PPCISD::Lo: return "PPCISD::Lo";
+ case PPCISD::TOC_ENTRY: return "PPCISD::TOC_ENTRY";
+ case PPCISD::TOC_RESTORE: return "PPCISD::TOC_RESTORE";
+ case PPCISD::LOAD: return "PPCISD::LOAD";
+ case PPCISD::LOAD_TOC: return "PPCISD::LOAD_TOC";
case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC";
case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg";
case PPCISD::SRL: return "PPCISD::SRL";
case PPCISD::STD_32: return "PPCISD::STD_32";
case PPCISD::CALL_SVR4: return "PPCISD::CALL_SVR4";
case PPCISD::CALL_Darwin: return "PPCISD::CALL_Darwin";
+ case PPCISD::NOP: return "PPCISD::NOP";
case PPCISD::MTCTR: return "PPCISD::MTCTR";
case PPCISD::BCTRL_Darwin: return "PPCISD::BCTRL_Darwin";
case PPCISD::BCTRL_SVR4: return "PPCISD::BCTRL_SVR4";
unsigned BitSize;
bool HasAnyUndefs;
- if (BV->isConstantSplat(APVal, APUndef, BitSize, HasAnyUndefs, 32))
+ if (BV->isConstantSplat(APVal, APUndef, BitSize, HasAnyUndefs, 32, true))
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N->getOperand(0)))
return CFP->getValueAPF().isNegZero();
Base = DAG.getTargetConstant((Addr - (signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
- Base = SDValue(DAG.getTargetNode(Opc, dl, CN->getValueType(0), Base), 0);
+ Base = SDValue(DAG.getMachineNode(Opc, dl, CN->getValueType(0), Base), 0);
return true;
}
}
Disp = DAG.getTargetConstant((short)Addr >> 2, MVT::i32);
Base = DAG.getTargetConstant((Addr-(signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
- Base = SDValue(DAG.getTargetNode(Opc, dl, CN->getValueType(0), Base),0);
+ Base = SDValue(DAG.getMachineNode(Opc, dl, CN->getValueType(0), Base),0);
return true;
}
}
return SDValue(); // Not reached
}
+SDValue PPCTargetLowering::LowerBlockAddress(SDValue Op, SelectionDAG &DAG) {
+ EVT PtrVT = Op.getValueType();
+ DebugLoc DL = Op.getDebugLoc();
+
+ BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
+ SDValue TgtBA = DAG.getBlockAddress(BA, PtrVT, /*isTarget=*/true);
+ SDValue Zero = DAG.getConstant(0, PtrVT);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, DL, PtrVT, TgtBA, Zero);
+ SDValue Lo = DAG.getNode(PPCISD::Lo, DL, PtrVT, TgtBA, Zero);
+
+ // If this is a non-darwin platform, we don't support non-static relo models
+ // yet.
+ const TargetMachine &TM = DAG.getTarget();
+ if (TM.getRelocationModel() == Reloc::Static ||
+ !TM.getSubtarget<PPCSubtarget>().isDarwin()) {
+ // Generate non-pic code that has direct accesses to globals.
+ // The address of the global is just (hi(&g)+lo(&g)).
+ return DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo);
+ }
+
+ if (TM.getRelocationModel() == Reloc::PIC_) {
+ // With PIC, the first instruction is actually "GR+hi(&G)".
+ Hi = DAG.getNode(ISD::ADD, DL, PtrVT,
+ DAG.getNode(PPCISD::GlobalBaseReg,
+ DebugLoc::getUnknownLoc(), PtrVT), Hi);
+ }
+
+ return DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo);
+}
+
SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) {
EVT PtrVT = Op.getValueType();
const TargetMachine &TM = DAG.getTarget();
+ // 64-bit SVR4 ABI code is always position-independent.
+ // The actual address of the GlobalValue is stored in the TOC.
+ if (PPCSubTarget.isSVR4ABI() && PPCSubTarget.isPPC64()) {
+ return DAG.getNode(PPCISD::TOC_ENTRY, dl, MVT::i64, GA,
+ DAG.getRegister(PPC::X2, MVT::i64));
+ }
+
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, GA, Zero);
SDValue Lo = DAG.getNode(PPCISD::Lo, dl, PtrVT, GA, Zero);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = (PtrVT == MVT::i64);
const Type *IntPtrTy =
- DAG.getTargetLoweringInfo().getTargetData()->getIntPtrType();
+ DAG.getTargetLoweringInfo().getTargetData()->getIntPtrType(
+ *DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
std::pair<SDValue, SDValue> CallResult =
- LowerCallTo(Chain, Op.getValueType().getTypeForEVT(),
+ LowerCallTo(Chain, Op.getValueType().getTypeForEVT(*DAG.getContext()),
false, false, false, false, 0, CallingConv::C, false,
/*isReturnValueUsed=*/true,
DAG.getExternalSymbol("__trampoline_setup", PtrVT),
- Args, DAG, dl);
+ Args, DAG, dl, DAG.GetOrdering(Chain.getNode()));
SDValue Ops[] =
{ CallResult.first, CallResult.second };
const PPCSubtarget &Subtarget) {
DebugLoc dl = Op.getDebugLoc();
- if (Subtarget.isDarwinABI()) {
+ if (Subtarget.isDarwinABI() || Subtarget.isPPC64()) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0);
}
- // For the SVR4 ABI we follow the layout of the va_list struct.
+ // For the 32-bit SVR4 ABI we follow the layout of the va_list struct.
// We suppose the given va_list is already allocated.
//
// typedef struct {
}
/// GetFPR - Get the set of FP registers that should be allocated for arguments,
-/// depending on which subtarget is selected.
-static const unsigned *GetFPR(const PPCSubtarget &Subtarget) {
- if (Subtarget.isDarwinABI()) {
- static const unsigned FPR[] = {
- PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
- PPC::F8, PPC::F9, PPC::F10, PPC::F11, PPC::F12, PPC::F13
- };
- return FPR;
- }
-
-
+/// on Darwin.
+static const unsigned *GetFPR() {
static const unsigned FPR[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
- PPC::F8
+ PPC::F8, PPC::F9, PPC::F10, PPC::F11, PPC::F12, PPC::F13
};
+
return FPR;
}
SDValue
PPCTargetLowering::LowerFormalArguments(SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
- if (PPCSubTarget.isSVR4ABI()) {
+ if (PPCSubTarget.isSVR4ABI() && !PPCSubTarget.isPPC64()) {
return LowerFormalArguments_SVR4(Chain, CallConv, isVarArg, Ins,
dl, DAG, InVals);
} else {
SDValue
PPCTargetLowering::LowerFormalArguments_SVR4(
SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
- // SVR4 ABI Stack Frame Layout:
+ // 32-bit SVR4 ABI Stack Frame Layout:
// +-----------------------------------+
// +--> | Back chain |
// | +-----------------------------------+
unsigned ArgSize = VA.getLocVT().getSizeInBits() / 8;
int FI = MFI->CreateFixedObject(ArgSize, VA.getLocMemOffset(),
- isImmutable);
+ isImmutable, false);
// Create load nodes to retrieve arguments from the stack.
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
NumFPArgRegs * EVT(MVT::f64).getSizeInBits()/8;
VarArgsStackOffset = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
- CCInfo.getNextStackOffset());
+ CCInfo.getNextStackOffset(),
+ true, false);
- VarArgsFrameIndex = MFI->CreateStackObject(Depth, 8);
+ VarArgsFrameIndex = MFI->CreateStackObject(Depth, 8, false);
SDValue FIN = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
// The fixed integer arguments of a variadic function are
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
- // FIXME SVR4: We only need to save FP argument registers if CR bit 6 is
- // set.
+ // FIXME 32-bit SVR4: We only need to save FP argument registers if CR bit 6
+ // is set.
// The double arguments are stored to the VarArgsFrameIndex
// on the stack.
SDValue
PPCTargetLowering::LowerFormalArguments_Darwin(
SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
-
// TODO: add description of PPC stack frame format, or at least some docs.
//
MachineFunction &MF = DAG.getMachineFunction();
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
- static const unsigned *FPR = GetFPR(PPCSubTarget);
+ static const unsigned *FPR = GetFPR();
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
CurArgOffset = CurArgOffset + (4 - ObjSize);
}
// The value of the object is its address.
- int FI = MFI->CreateFixedObject(ObjSize, CurArgOffset);
+ int FI = MFI->CreateFixedObject(ObjSize, CurArgOffset, true, false);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
InVals.push_back(FIN);
if (ObjSize==1 || ObjSize==2) {
// the object.
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
- int FI = MFI->CreateFixedObject(PtrByteSize, ArgOffset);
+ int FI = MFI->CreateFixedObject(PtrByteSize, ArgOffset, true, false);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0);
GPR_idx++;
}
ArgOffset += 16;
- GPR_idx = std::min(GPR_idx+4, Num_GPR_Regs);
+ GPR_idx = std::min(GPR_idx+4, Num_GPR_Regs); // FIXME correct for ppc64?
}
++VR_idx;
} else {
if (needsLoad) {
int FI = MFI->CreateFixedObject(ObjSize,
CurArgOffset + (ArgSize - ObjSize),
- isImmutable);
+ isImmutable, false);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, NULL, 0);
}
int Depth = ArgOffset;
VarArgsFrameIndex = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
- Depth);
+ Depth, true, false);
SDValue FIN = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
// If this function is vararg, store any remaining integer argument regs
/// CalculateTailCallSPDiff - Get the amount the stack pointer has to be
/// adjusted to accomodate the arguments for the tailcall.
-static int CalculateTailCallSPDiff(SelectionDAG& DAG, bool IsTailCall,
+static int CalculateTailCallSPDiff(SelectionDAG& DAG, bool isTailCall,
unsigned ParamSize) {
- if (!IsTailCall) return 0;
+ if (!isTailCall) return 0;
PPCFunctionInfo *FI = DAG.getMachineFunction().getInfo<PPCFunctionInfo>();
unsigned CallerMinReservedArea = FI->getMinReservedArea();
/// optimization should implement this function.
bool
PPCTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
- unsigned CalleeCC,
+ CallingConv::ID CalleeCC,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SelectionDAG& DAG) const {
+ if (!PerformTailCallOpt)
+ return false;
+
// Variable argument functions are not supported.
if (isVarArg)
return false;
MachineFunction &MF = DAG.getMachineFunction();
- unsigned CallerCC = MF.getFunction()->getCallingConv();
+ CallingConv::ID CallerCC = MF.getFunction()->getCallingConv();
if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC) {
// Functions containing by val parameters are not supported.
for (unsigned i = 0; i != Ins.size(); i++) {
int NewRetAddrLoc = SPDiff + PPCFrameInfo::getReturnSaveOffset(isPPC64,
isDarwinABI);
int NewRetAddr = MF.getFrameInfo()->CreateFixedObject(SlotSize,
- NewRetAddrLoc);
+ NewRetAddrLoc,
+ true, false);
EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
SDValue NewRetAddrFrIdx = DAG.getFrameIndex(NewRetAddr, VT);
Chain = DAG.getStore(Chain, dl, OldRetAddr, NewRetAddrFrIdx,
PseudoSourceValue::getFixedStack(NewRetAddr), 0);
- // When using the SVR4 ABI there is no need to move the FP stack slot
- // as the FP is never overwritten.
+ // When using the 32/64-bit SVR4 ABI there is no need to move the FP stack
+ // slot as the FP is never overwritten.
if (isDarwinABI) {
int NewFPLoc =
SPDiff + PPCFrameInfo::getFramePointerSaveOffset(isPPC64, isDarwinABI);
- int NewFPIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize, NewFPLoc);
+ int NewFPIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize, NewFPLoc,
+ true, false);
SDValue NewFramePtrIdx = DAG.getFrameIndex(NewFPIdx, VT);
Chain = DAG.getStore(Chain, dl, OldFP, NewFramePtrIdx,
PseudoSourceValue::getFixedStack(NewFPIdx), 0);
SmallVector<TailCallArgumentInfo, 8>& TailCallArguments) {
int Offset = ArgOffset + SPDiff;
uint32_t OpSize = (Arg.getValueType().getSizeInBits()+7)/8;
- int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset);
+ int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset, true,false);
EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
SDValue FIN = DAG.getFrameIndex(FI, VT);
TailCallArgumentInfo Info;
LROpOut = DAG.getLoad(VT, dl, Chain, LROpOut, NULL, 0);
Chain = SDValue(LROpOut.getNode(), 1);
- // When using the SVR4 ABI there is no need to load the FP stack slot
- // as the FP is never overwritten.
+ // When using the 32/64-bit SVR4 ABI there is no need to load the FP stack
+ // slot as the FP is never overwritten.
if (isDarwinABI) {
FPOpOut = getFramePointerFrameIndex(DAG);
FPOpOut = DAG.getLoad(VT, dl, Chain, FPOpOut, NULL, 0);
SDValue &Chain, DebugLoc dl, int SPDiff, bool isTailCall,
SmallVector<std::pair<unsigned, SDValue>, 8> &RegsToPass,
SmallVector<SDValue, 8> &Ops, std::vector<EVT> &NodeTys,
- bool isSVR4ABI) {
+ bool isPPC64, bool isSVR4ABI) {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
NodeTys.push_back(MVT::Other); // Returns a chain
NodeTys.push_back(MVT::Flag); // Returns a flag for retval copy to use.
// Otherwise, this is an indirect call. We have to use a MTCTR/BCTRL pair
// to do the call, we can't use PPCISD::CALL.
SDValue MTCTROps[] = {Chain, Callee, InFlag};
+
+ if (isSVR4ABI && isPPC64) {
+ // Function pointers in the 64-bit SVR4 ABI do not point to the function
+ // entry point, but to the function descriptor (the function entry point
+ // address is part of the function descriptor though).
+ // The function descriptor is a three doubleword structure with the
+ // following fields: function entry point, TOC base address and
+ // environment pointer.
+ // Thus for a call through a function pointer, the following actions need
+ // to be performed:
+ // 1. Save the TOC of the caller in the TOC save area of its stack
+ // frame (this is done in LowerCall_Darwin()).
+ // 2. Load the address of the function entry point from the function
+ // descriptor.
+ // 3. Load the TOC of the callee from the function descriptor into r2.
+ // 4. Load the environment pointer from the function descriptor into
+ // r11.
+ // 5. Branch to the function entry point address.
+ // 6. On return of the callee, the TOC of the caller needs to be
+ // restored (this is done in FinishCall()).
+ //
+ // All those operations are flagged together to ensure that no other
+ // operations can be scheduled in between. E.g. without flagging the
+ // operations together, a TOC access in the caller could be scheduled
+ // between the load of the callee TOC and the branch to the callee, which
+ // results in the TOC access going through the TOC of the callee instead
+ // of going through the TOC of the caller, which leads to incorrect code.
+
+ // Load the address of the function entry point from the function
+ // descriptor.
+ SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Other, MVT::Flag);
+ SDValue LoadFuncPtr = DAG.getNode(PPCISD::LOAD, dl, VTs, MTCTROps,
+ InFlag.getNode() ? 3 : 2);
+ Chain = LoadFuncPtr.getValue(1);
+ InFlag = LoadFuncPtr.getValue(2);
+
+ // Load environment pointer into r11.
+ // Offset of the environment pointer within the function descriptor.
+ SDValue PtrOff = DAG.getIntPtrConstant(16);
+
+ SDValue AddPtr = DAG.getNode(ISD::ADD, dl, MVT::i64, Callee, PtrOff);
+ SDValue LoadEnvPtr = DAG.getNode(PPCISD::LOAD, dl, VTs, Chain, AddPtr,
+ InFlag);
+ Chain = LoadEnvPtr.getValue(1);
+ InFlag = LoadEnvPtr.getValue(2);
+
+ SDValue EnvVal = DAG.getCopyToReg(Chain, dl, PPC::X11, LoadEnvPtr,
+ InFlag);
+ Chain = EnvVal.getValue(0);
+ InFlag = EnvVal.getValue(1);
+
+ // Load TOC of the callee into r2. We are using a target-specific load
+ // with r2 hard coded, because the result of a target-independent load
+ // would never go directly into r2, since r2 is a reserved register (which
+ // prevents the register allocator from allocating it), resulting in an
+ // additional register being allocated and an unnecessary move instruction
+ // being generated.
+ VTs = DAG.getVTList(MVT::Other, MVT::Flag);
+ SDValue LoadTOCPtr = DAG.getNode(PPCISD::LOAD_TOC, dl, VTs, Chain,
+ Callee, InFlag);
+ Chain = LoadTOCPtr.getValue(0);
+ InFlag = LoadTOCPtr.getValue(1);
+
+ MTCTROps[0] = Chain;
+ MTCTROps[1] = LoadFuncPtr;
+ MTCTROps[2] = InFlag;
+ }
+
Chain = DAG.getNode(PPCISD::MTCTR, dl, NodeTys, MTCTROps,
2 + (InFlag.getNode() != 0));
InFlag = Chain.getValue(1);
SDValue
PPCTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
}
SDValue
-PPCTargetLowering::FinishCall(unsigned CallConv, DebugLoc dl, bool isTailCall,
- bool isVarArg,
+PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
+ bool isTailCall, bool isVarArg,
SelectionDAG &DAG,
SmallVector<std::pair<unsigned, SDValue>, 8>
&RegsToPass,
int SPDiff, unsigned NumBytes,
const SmallVectorImpl<ISD::InputArg> &Ins,
SmallVectorImpl<SDValue> &InVals) {
-
std::vector<EVT> NodeTys;
SmallVector<SDValue, 8> Ops;
unsigned CallOpc = PrepareCall(DAG, Callee, InFlag, Chain, dl, SPDiff,
isTailCall, RegsToPass, Ops, NodeTys,
+ PPCSubTarget.isPPC64(),
PPCSubTarget.isSVR4ABI());
// When performing tail call optimization the callee pops its arguments off
Chain = DAG.getNode(CallOpc, dl, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
+ // Add a NOP immediately after the branch instruction when using the 64-bit
+ // SVR4 ABI. At link time, if caller and callee are in a different module and
+ // thus have a different TOC, the call will be replaced with a call to a stub
+ // function which saves the current TOC, loads the TOC of the callee and
+ // branches to the callee. The NOP will be replaced with a load instruction
+ // which restores the TOC of the caller from the TOC save slot of the current
+ // stack frame. If caller and callee belong to the same module (and have the
+ // same TOC), the NOP will remain unchanged.
+ if (!isTailCall && PPCSubTarget.isSVR4ABI()&& PPCSubTarget.isPPC64()) {
+ SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Flag);
+ if (CallOpc == PPCISD::BCTRL_SVR4) {
+ // This is a call through a function pointer.
+ // Restore the caller TOC from the save area into R2.
+ // See PrepareCall() for more information about calls through function
+ // pointers in the 64-bit SVR4 ABI.
+ // We are using a target-specific load with r2 hard coded, because the
+ // result of a target-independent load would never go directly into r2,
+ // since r2 is a reserved register (which prevents the register allocator
+ // from allocating it), resulting in an additional register being
+ // allocated and an unnecessary move instruction being generated.
+ Chain = DAG.getNode(PPCISD::TOC_RESTORE, dl, VTs, Chain, InFlag);
+ InFlag = Chain.getValue(1);
+ } else {
+ // Otherwise insert NOP.
+ InFlag = DAG.getNode(PPCISD::NOP, dl, MVT::Flag, InFlag);
+ }
+ }
+
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(BytesCalleePops, true),
InFlag);
}
SDValue
-PPCTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
- unsigned CallConv, bool isVarArg,
- bool isTailCall,
+PPCTargetLowering::LowerCall(SDValue Chain, SDValue Callee, const Type *RetTy,
+ CallingConv::ID CallConv, bool isVarArg,
+ bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
- if (PPCSubTarget.isSVR4ABI()) {
+ if (isTailCall)
+ isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv, isVarArg,
+ Ins, DAG);
+
+ if (PPCSubTarget.isSVR4ABI() && !PPCSubTarget.isPPC64()) {
return LowerCall_SVR4(Chain, Callee, CallConv, isVarArg,
isTailCall, Outs, Ins,
dl, DAG, InVals);
SDValue
PPCTargetLowering::LowerCall_SVR4(SDValue Chain, SDValue Callee,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
bool isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
// See PPCTargetLowering::LowerFormalArguments_SVR4() for a description
- // of the SVR4 ABI stack frame layout.
-
- assert((!isTailCall ||
- (CallConv == CallingConv::Fast && PerformTailCallOpt)) &&
- "IsEligibleForTailCallOptimization missed a case!");
+ // of the 32-bit SVR4 ABI stack frame layout.
assert((CallConv == CallingConv::C ||
CallConv == CallingConv::Fast) && "Unknown calling convention!");
if (Result) {
#ifndef NDEBUG
- cerr << "Call operand #" << i << " has unhandled type "
+ errs() << "Call operand #" << i << " has unhandled type "
<< ArgVT.getEVTString() << "\n";
#endif
llvm_unreachable(0);
// Set CR6 to true if this is a vararg call.
if (isVarArg) {
- SDValue SetCR(DAG.getTargetNode(PPC::CRSET, dl, MVT::i32), 0);
+ SDValue SetCR(DAG.getMachineNode(PPC::CRSET, dl, MVT::i32), 0);
Chain = DAG.getCopyToReg(Chain, dl, PPC::CR1EQ, SetCR, InFlag);
InFlag = Chain.getValue(1);
}
SDValue
PPCTargetLowering::LowerCall_Darwin(SDValue Chain, SDValue Callee,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
bool isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
- static const unsigned *FPR = GetFPR(PPCSubTarget);
+ static const unsigned *FPR = GetFPR();
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
SmallVector<SDValue, 8> MemOpChains;
for (unsigned i = 0; i != NumOps; ++i) {
- bool inMem = false;
SDValue Arg = Outs[i].Val;
ISD::ArgFlagsTy Flags = Outs[i].Flags;
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
isPPC64, isTailCall, false, MemOpChains,
TailCallArguments, dl);
- inMem = true;
}
ArgOffset += PtrByteSize;
break;
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
isPPC64, isTailCall, false, MemOpChains,
TailCallArguments, dl);
- inMem = true;
}
if (isPPC64)
ArgOffset += 8;
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
+ // Check if this is an indirect call (MTCTR/BCTRL).
+ // See PrepareCall() for more information about calls through function
+ // pointers in the 64-bit SVR4 ABI.
+ if (!isTailCall && isPPC64 && PPCSubTarget.isSVR4ABI() &&
+ !dyn_cast<GlobalAddressSDNode>(Callee) &&
+ !dyn_cast<ExternalSymbolSDNode>(Callee) &&
+ !isBLACompatibleAddress(Callee, DAG)) {
+ // Load r2 into a virtual register and store it to the TOC save area.
+ SDValue Val = DAG.getCopyFromReg(Chain, dl, PPC::X2, MVT::i64);
+ // TOC save area offset.
+ SDValue PtrOff = DAG.getIntPtrConstant(40);
+ SDValue AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
+ Chain = DAG.getStore(Val.getValue(1), dl, Val, AddPtr, NULL, 0);
+ }
+
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDValue InFlag;
SDValue
PPCTargetLowering::LowerReturn(SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
DebugLoc dl, SelectionDAG &DAG) {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Construct the stack pointer operand.
- bool IsPPC64 = Subtarget.isPPC64();
- unsigned SP = IsPPC64 ? PPC::X1 : PPC::R1;
+ bool isPPC64 = Subtarget.isPPC64();
+ unsigned SP = isPPC64 ? PPC::X1 : PPC::R1;
SDValue StackPtr = DAG.getRegister(SP, PtrVT);
// Get the operands for the STACKRESTORE.
SDValue
PPCTargetLowering::getReturnAddrFrameIndex(SelectionDAG & DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
- bool IsPPC64 = PPCSubTarget.isPPC64();
+ bool isPPC64 = PPCSubTarget.isPPC64();
bool isDarwinABI = PPCSubTarget.isDarwinABI();
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// If the frame pointer save index hasn't been defined yet.
if (!RASI) {
// Find out what the fix offset of the frame pointer save area.
- int LROffset = PPCFrameInfo::getReturnSaveOffset(IsPPC64, isDarwinABI);
+ int LROffset = PPCFrameInfo::getReturnSaveOffset(isPPC64, isDarwinABI);
// Allocate the frame index for frame pointer save area.
- RASI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, LROffset);
+ RASI = MF.getFrameInfo()->CreateFixedObject(isPPC64? 8 : 4, LROffset,
+ true, false);
// Save the result.
FI->setReturnAddrSaveIndex(RASI);
}
SDValue
PPCTargetLowering::getFramePointerFrameIndex(SelectionDAG & DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
- bool IsPPC64 = PPCSubTarget.isPPC64();
+ bool isPPC64 = PPCSubTarget.isPPC64();
bool isDarwinABI = PPCSubTarget.isDarwinABI();
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// If the frame pointer save index hasn't been defined yet.
if (!FPSI) {
// Find out what the fix offset of the frame pointer save area.
- int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64,
+ int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(isPPC64,
isDarwinABI);
// Allocate the frame index for frame pointer save area.
- FPSI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, FPOffset);
+ FPSI = MF.getFrameInfo()->CreateFixedObject(isPPC64? 8 : 4, FPOffset,
+ true, false);
// Save the result.
FI->setFramePointerSaveIndex(FPSI);
}
// 64-bit registers. In particular, sign extend the input value into the
// 64-bit register with extsw, store the WHOLE 64-bit value into the stack
// then lfd it and fcfid it.
- MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
- int FrameIdx = FrameInfo->CreateStackObject(8, 8);
+ MachineFunction &MF = DAG.getMachineFunction();
+ MachineFrameInfo *FrameInfo = MF.getFrameInfo();
+ int FrameIdx = FrameInfo->CreateStackObject(8, 8, false);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
Op.getOperand(0));
// STD the extended value into the stack slot.
- MachineMemOperand MO(PseudoSourceValue::getFixedStack(FrameIdx),
- MachineMemOperand::MOStore, 0, 8, 8);
- SDValue Store = DAG.getNode(PPCISD::STD_32, dl, MVT::Other,
- DAG.getEntryNode(), Ext64, FIdx,
- DAG.getMemOperand(MO));
+ MachineMemOperand *MMO =
+ MF.getMachineMemOperand(PseudoSourceValue::getFixedStack(FrameIdx),
+ MachineMemOperand::MOStore, 0, 8, 8);
+ SDValue Ops[] = { DAG.getEntryNode(), Ext64, FIdx };
+ SDValue Store =
+ DAG.getMemIntrinsicNode(PPCISD::STD_32, dl, DAG.getVTList(MVT::Other),
+ Ops, 4, MVT::i64, MMO);
// Load the value as a double.
SDValue Ld = DAG.getLoad(MVT::f64, dl, Store, FIdx, NULL, 0);
SDValue Chain = DAG.getNode(PPCISD::MFFS, dl, NodeTys, &InFlag, 0);
// Save FP register to stack slot
- int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
+ int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8, false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Chain,
StackSlot, NULL, 0);
unsigned SplatBitSize;
bool HasAnyUndefs;
if (! BVN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
- HasAnyUndefs) || SplatBitSize > 32)
+ HasAnyUndefs, 0, true) || SplatBitSize > 32)
return SDValue();
unsigned SplatBits = APSplatBits.getZExtValue();
DebugLoc dl = Op.getDebugLoc();
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
- int FrameIdx = FrameInfo->CreateStackObject(16, 16);
+ int FrameIdx = FrameInfo->CreateStackObject(16, 16, false);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
switch (Op.getOpcode()) {
default: llvm_unreachable("Wasn't expecting to be able to lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
+ case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
MachineBasicBlock *
PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
- MachineBasicBlock *BB) const {
+ MachineBasicBlock *BB,
+ DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) const {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
// To "insert" these instructions we actually have to insert their
.addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB);
F->insert(It, copy0MBB);
F->insert(It, sinkMBB);
- // Update machine-CFG edges by transferring all successors of the current
+ // Update machine-CFG edges by first adding all successors of the current
// block to the new block which will contain the Phi node for the select.
- sinkMBB->transferSuccessors(BB);
+ // Also inform sdisel of the edge changes.
+ for (MachineBasicBlock::succ_iterator I = BB->succ_begin(),
+ E = BB->succ_end(); I != E; ++I) {
+ EM->insert(std::make_pair(*I, sinkMBB));
+ sinkMBB->addSuccessor(*I);
+ }
+ // Next, remove all successors of the current block, and add the true
+ // and fallthrough blocks as its successors.
+ while (!BB->succ_empty())
+ BB->removeSuccessor(BB->succ_begin());
// Next, add the true and fallthrough blocks as its successors.
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
}
// Turn STORE (BSWAP) -> sthbrx/stwbrx.
- if (N->getOperand(1).getOpcode() == ISD::BSWAP &&
+ if (cast<StoreSDNode>(N)->isUnindexed() &&
+ N->getOperand(1).getOpcode() == ISD::BSWAP &&
N->getOperand(1).getNode()->hasOneUse() &&
(N->getOperand(1).getValueType() == MVT::i32 ||
N->getOperand(1).getValueType() == MVT::i16)) {
if (BSwapOp.getValueType() == MVT::i16)
BSwapOp = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, BSwapOp);
- return DAG.getNode(PPCISD::STBRX, dl, MVT::Other, N->getOperand(0),
- BSwapOp, N->getOperand(2), N->getOperand(3),
- DAG.getValueType(N->getOperand(1).getValueType()));
+ SDValue Ops[] = {
+ N->getOperand(0), BSwapOp, N->getOperand(2),
+ DAG.getValueType(N->getOperand(1).getValueType())
+ };
+ return
+ DAG.getMemIntrinsicNode(PPCISD::STBRX, dl, DAG.getVTList(MVT::Other),
+ Ops, array_lengthof(Ops),
+ cast<StoreSDNode>(N)->getMemoryVT(),
+ cast<StoreSDNode>(N)->getMemOperand());
}
break;
case ISD::BSWAP:
SDValue Load = N->getOperand(0);
LoadSDNode *LD = cast<LoadSDNode>(Load);
// Create the byte-swapping load.
- std::vector<EVT> VTs;
- VTs.push_back(MVT::i32);
- VTs.push_back(MVT::Other);
- SDValue MO = DAG.getMemOperand(LD->getMemOperand());
SDValue Ops[] = {
LD->getChain(), // Chain
LD->getBasePtr(), // Ptr
- MO, // MemOperand
DAG.getValueType(N->getValueType(0)) // VT
};
- SDValue BSLoad = DAG.getNode(PPCISD::LBRX, dl, VTs, Ops, 4);
+ SDValue BSLoad =
+ DAG.getMemIntrinsicNode(PPCISD::LBRX, dl,
+ DAG.getVTList(MVT::i32, MVT::Other), Ops, 3,
+ LD->getMemoryVT(), LD->getMemOperand());
// If this is an i16 load, insert the truncate.
SDValue ResVal = BSLoad;
default: break;
case PPCISD::LBRX: {
// lhbrx is known to have the top bits cleared out.
- if (cast<VTSDNode>(Op.getOperand(3))->getVT() == MVT::i16)
+ if (cast<VTSDNode>(Op.getOperand(2))->getVT() == MVT::i16)
KnownZero = 0xFFFF0000;
break;
}
projects / oota-llvm.git / blobdiff