switch (TM.getSubtarget<X86Subtarget>().TargetType) {
default: llvm_unreachable("unknown subtarget type");
case X86Subtarget::isDarwin:
- return new TargetLoweringObjectFileMachO(TM);
+ return new TargetLoweringObjectFileMachO();
case X86Subtarget::isELF:
return new TargetLoweringObjectFileELF();
case X86Subtarget::isMingw:
setOperationAction(ISD::SELECT , MVT::i64 , Custom);
setOperationAction(ISD::SETCC , MVT::i64 , Custom);
}
- // X86 ret instruction may pop stack.
- setOperationAction(ISD::RET , MVT::Other, Custom);
setOperationAction(ISD::EH_RETURN , MVT::Other, Custom);
// Darwin ABI issue.
#include "X86GenCallingConv.inc"
-/// LowerRET - Lower an ISD::RET node.
-SDValue X86TargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG) {
- DebugLoc dl = Op.getDebugLoc();
- assert((Op.getNumOperands() & 1) == 1 && "ISD::RET should have odd # args");
+SDValue
+X86TargetLowering::LowerReturn(SDValue Chain,
+ unsigned CallConv, bool isVarArg,
+ const SmallVectorImpl<ISD::OutputArg> &Outs,
+ DebugLoc dl, SelectionDAG &DAG) {
SmallVector<CCValAssign, 16> RVLocs;
- unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
- bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
- CCState CCInfo(CC, isVarArg, getTargetMachine(), RVLocs, *DAG.getContext());
- CCInfo.AnalyzeReturn(Op.getNode(), RetCC_X86);
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ RVLocs, *DAG.getContext());
+ CCInfo.AnalyzeReturn(Outs, RetCC_X86);
// If this is the first return lowered for this function, add the regs to the
// liveout set for the function.
if (RVLocs[i].isRegLoc())
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
- SDValue Chain = Op.getOperand(0);
-
- // Handle tail call return.
- Chain = GetPossiblePreceedingTailCall(Chain, X86ISD::TAILCALL);
- if (Chain.getOpcode() == X86ISD::TAILCALL) {
- SDValue TailCall = Chain;
- SDValue TargetAddress = TailCall.getOperand(1);
- SDValue StackAdjustment = TailCall.getOperand(2);
- assert(((TargetAddress.getOpcode() == ISD::Register &&
- (cast<RegisterSDNode>(TargetAddress)->getReg() == X86::EAX ||
- cast<RegisterSDNode>(TargetAddress)->getReg() == X86::R11)) ||
- TargetAddress.getOpcode() == ISD::TargetExternalSymbol ||
- TargetAddress.getOpcode() == ISD::TargetGlobalAddress) &&
- "Expecting an global address, external symbol, or register");
- assert(StackAdjustment.getOpcode() == ISD::Constant &&
- "Expecting a const value");
-
- SmallVector<SDValue,8> Operands;
- Operands.push_back(Chain.getOperand(0));
- Operands.push_back(TargetAddress);
- Operands.push_back(StackAdjustment);
- // Copy registers used by the call. Last operand is a flag so it is not
- // copied.
- for (unsigned i=3; i < TailCall.getNumOperands()-1; i++) {
- Operands.push_back(Chain.getOperand(i));
- }
- return DAG.getNode(X86ISD::TC_RETURN, dl, MVT::Other, &Operands[0],
- Operands.size());
- }
- // Regular return.
SDValue Flag;
SmallVector<SDValue, 6> RetOps;
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
- SDValue ValToCopy = Op.getOperand(i*2+1);
+ SDValue ValToCopy = Outs[i].Val;
// Returns in ST0/ST1 are handled specially: these are pushed as operands to
// the RET instruction and handled by the FP Stackifier.
MVT::Other, &RetOps[0], RetOps.size());
}
+/// LowerCallResult - Lower the result values of a call into the
+/// appropriate copies out of appropriate physical registers.
+///
+SDValue
+X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
+ unsigned CallConv, bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
-/// LowerCallResult - Lower the result values of an ISD::CALL into the
-/// appropriate copies out of appropriate physical registers. This assumes that
-/// Chain/InFlag are the input chain/flag to use, and that TheCall is the call
-/// being lowered. The returns a SDNode with the same number of values as the
-/// ISD::CALL.
-SDNode *X86TargetLowering::
-LowerCallResult(SDValue Chain, SDValue InFlag, CallSDNode *TheCall,
- unsigned CallingConv, SelectionDAG &DAG) {
-
- DebugLoc dl = TheCall->getDebugLoc();
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
- bool isVarArg = TheCall->isVarArg();
bool Is64Bit = Subtarget->is64Bit();
- CCState CCInfo(CallingConv, isVarArg, getTargetMachine(),
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
RVLocs, *DAG.getContext());
- CCInfo.AnalyzeCallResult(TheCall, RetCC_X86);
-
- SmallVector<SDValue, 8> ResultVals;
+ CCInfo.AnalyzeCallResult(Ins, RetCC_X86);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
// If this is x86-64, and we disabled SSE, we can't return FP values
if ((CopyVT == MVT::f32 || CopyVT == MVT::f64) &&
- ((Is64Bit || TheCall->isInreg()) && !Subtarget->hasSSE1())) {
+ ((Is64Bit || Ins[i].Flags.isInReg()) && !Subtarget->hasSSE1())) {
llvm_report_error("SSE register return with SSE disabled");
}
MVT::v2i64, InFlag).getValue(1);
Val = Chain.getValue(0);
Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
- Val, DAG.getConstant(0, MVT::i64));
+ Val, DAG.getConstant(0, MVT::i64));
} else {
Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
MVT::i64, InFlag).getValue(1);
DAG.getIntPtrConstant(1));
}
- ResultVals.push_back(Val);
+ InVals.push_back(Val);
}
- // Merge everything together with a MERGE_VALUES node.
- ResultVals.push_back(Chain);
- return DAG.getNode(ISD::MERGE_VALUES, dl, TheCall->getVTList(),
- &ResultVals[0], ResultVals.size()).getNode();
+ return Chain;
}
// For info on fast calling convention see Fast Calling Convention (tail call)
// implementation LowerX86_32FastCCCallTo.
-/// CallIsStructReturn - Determines whether a CALL node uses struct return
+/// CallIsStructReturn - Determines whether a call uses struct return
/// semantics.
-static bool CallIsStructReturn(CallSDNode *TheCall) {
- unsigned NumOps = TheCall->getNumArgs();
- if (!NumOps)
+static bool CallIsStructReturn(const SmallVectorImpl<ISD::OutputArg> &Outs) {
+ if (Outs.empty())
return false;
- return TheCall->getArgFlags(0).isSRet();
+ return Outs[0].Flags.isSRet();
}
-/// ArgsAreStructReturn - Determines whether a FORMAL_ARGUMENTS node uses struct
+/// ArgsAreStructReturn - Determines whether a function uses struct
/// return semantics.
-static bool ArgsAreStructReturn(SDValue Op) {
- unsigned NumArgs = Op.getNode()->getNumValues() - 1;
- if (!NumArgs)
+static bool
+ArgsAreStructReturn(const SmallVectorImpl<ISD::InputArg> &Ins) {
+ if (Ins.empty())
return false;
- return cast<ARG_FLAGSSDNode>(Op.getOperand(3))->getArgFlags().isSRet();
+ return Ins[0].Flags.isSRet();
}
-/// IsCalleePop - Determines whether a CALL or FORMAL_ARGUMENTS node requires
-/// the callee to pop its own arguments. Callee pop is necessary to support tail
-/// calls.
+/// IsCalleePop - Determines whether the callee is required to pop its
+/// own arguments. Callee pop is necessary to support tail calls.
bool X86TargetLowering::IsCalleePop(bool IsVarArg, unsigned CallingConv) {
if (IsVarArg)
return false;
return CC_X86_32_C;
}
-/// NameDecorationForFORMAL_ARGUMENTS - Selects the appropriate decoration to
-/// apply to a MachineFunction containing a given FORMAL_ARGUMENTS node.
+/// NameDecorationForCallConv - Selects the appropriate decoration to
+/// apply to a MachineFunction containing a given calling convention.
NameDecorationStyle
-X86TargetLowering::NameDecorationForFORMAL_ARGUMENTS(SDValue Op) {
- unsigned CC = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
- if (CC == CallingConv::X86_FastCall)
+X86TargetLowering::NameDecorationForCallConv(unsigned CallConv) {
+ if (CallConv == CallingConv::X86_FastCall)
return FastCall;
- else if (CC == CallingConv::X86_StdCall)
+ else if (CallConv == CallingConv::X86_StdCall)
return StdCall;
return None;
}
/*AlwaysInline=*/true, NULL, 0, NULL, 0);
}
-SDValue X86TargetLowering::LowerMemArgument(SDValue Op, SelectionDAG &DAG,
- const CCValAssign &VA,
- MachineFrameInfo *MFI,
- unsigned CC,
- SDValue Root, unsigned i) {
+SDValue
+X86TargetLowering::LowerMemArgument(SDValue Chain,
+ unsigned CallConv,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ const CCValAssign &VA,
+ MachineFrameInfo *MFI,
+ unsigned i) {
+
// Create the nodes corresponding to a load from this parameter slot.
- ISD::ArgFlagsTy Flags =
- cast<ARG_FLAGSSDNode>(Op.getOperand(3 + i))->getArgFlags();
- bool AlwaysUseMutable = (CC==CallingConv::Fast) && PerformTailCallOpt;
+ ISD::ArgFlagsTy Flags = Ins[i].Flags;
+ bool AlwaysUseMutable = (CallConv==CallingConv::Fast) && PerformTailCallOpt;
bool isImmutable = !AlwaysUseMutable && !Flags.isByVal();
// FIXME: For now, all byval parameter objects are marked mutable. This can be
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
if (Flags.isByVal())
return FIN;
- return DAG.getLoad(VA.getValVT(), Op.getDebugLoc(), Root, FIN,
+ return DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
PseudoSourceValue::getFixedStack(FI), 0);
}
SDValue
-X86TargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG) {
+X86TargetLowering::LowerFormalArguments(SDValue Chain,
+ unsigned CallConv,
+ bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl,
+ SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+
MachineFunction &MF = DAG.getMachineFunction();
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
- DebugLoc dl = Op.getDebugLoc();
const Function* Fn = MF.getFunction();
if (Fn->hasExternalLinkage() &&
FuncInfo->setForceFramePointer(true);
// Decorate the function name.
- FuncInfo->setDecorationStyle(NameDecorationForFORMAL_ARGUMENTS(Op));
+ FuncInfo->setDecorationStyle(NameDecorationForCallConv(CallConv));
MachineFrameInfo *MFI = MF.getFrameInfo();
- SDValue Root = Op.getOperand(0);
- bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() != 0;
- unsigned CC = MF.getFunction()->getCallingConv();
bool Is64Bit = Subtarget->is64Bit();
bool IsWin64 = Subtarget->isTargetWin64();
- assert(!(isVarArg && CC == CallingConv::Fast) &&
+ assert(!(isVarArg && CallConv == CallingConv::Fast) &&
"Var args not supported with calling convention fastcc");
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeFormalArguments(Op.getNode(), CCAssignFnForNode(CC));
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ ArgLocs, *DAG.getContext());
+ CCInfo.AnalyzeFormalArguments(Ins, CCAssignFnForNode(CallConv));
- SmallVector<SDValue, 8> ArgValues;
unsigned LastVal = ~0U;
+ SDValue ArgValue;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
// TODO: If an arg is passed in two places (e.g. reg and stack), skip later
RC = X86::FR64RegisterClass;
else if (RegVT.isVector() && RegVT.getSizeInBits() == 128)
RC = X86::VR128RegisterClass;
- else if (RegVT.isVector()) {
- assert(RegVT.getSizeInBits() == 64);
- if (!Is64Bit)
- RC = X86::VR64RegisterClass; // MMX values are passed in MMXs.
- else {
- // Darwin calling convention passes MMX values in either GPRs or
- // XMMs in x86-64. Other targets pass them in memory.
- if (RegVT != MVT::v1i64 && Subtarget->hasSSE2()) {
- RC = X86::VR128RegisterClass; // MMX values are passed in XMMs.
- RegVT = MVT::v2i64;
- } else {
- RC = X86::GR64RegisterClass; // v1i64 values are passed in GPRs.
- RegVT = MVT::i64;
- }
- }
- } else {
+ else if (RegVT.isVector() && RegVT.getSizeInBits() == 64)
+ RC = X86::VR64RegisterClass;
+ else
llvm_unreachable("Unknown argument type!");
- }
- unsigned Reg = DAG.getMachineFunction().addLiveIn(VA.getLocReg(), RC);
- SDValue ArgValue = DAG.getCopyFromReg(Root, dl, Reg, RegVT);
+ unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
+ 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
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
+ else if (VA.getLocInfo() == CCValAssign::BCvt)
+ ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
- if (VA.getLocInfo() != CCValAssign::Full)
- ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
-
- // Handle MMX values passed in GPRs.
- if (Is64Bit && RegVT != VA.getLocVT()) {
- if (RegVT.getSizeInBits() == 64 && RC == X86::GR64RegisterClass)
- ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getLocVT(), ArgValue);
- else if (RC == X86::VR128RegisterClass) {
+ if (VA.isExtInLoc()) {
+ // Handle MMX values passed in XMM regs.
+ if (RegVT.isVector()) {
ArgValue = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
ArgValue, DAG.getConstant(0, MVT::i64));
- ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getLocVT(), ArgValue);
- }
+ ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
+ } else
+ ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
}
-
- ArgValues.push_back(ArgValue);
} else {
assert(VA.isMemLoc());
- ArgValues.push_back(LowerMemArgument(Op, DAG, VA, MFI, CC, Root, i));
+ ArgValue = LowerMemArgument(Chain, CallConv, Ins, dl, DAG, VA, MFI, i);
}
+
+ // If value is passed via pointer - do a load.
+ if (VA.getLocInfo() == CCValAssign::Indirect)
+ ArgValue = DAG.getLoad(VA.getValVT(), dl, Chain, ArgValue, NULL, 0);
+
+ InVals.push_back(ArgValue);
}
// The x86-64 ABI for returning structs by value requires that we copy
// the sret argument into %rax for the return. Save the argument into
// a virtual register so that we can access it from the return points.
- if (Is64Bit && DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
- MachineFunction &MF = DAG.getMachineFunction();
+ if (Is64Bit && MF.getFunction()->hasStructRetAttr()) {
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
unsigned Reg = FuncInfo->getSRetReturnReg();
if (!Reg) {
Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
FuncInfo->setSRetReturnReg(Reg);
}
- SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, ArgValues[0]);
- Root = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Root);
+ SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]);
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Chain);
}
unsigned StackSize = CCInfo.getNextStackOffset();
// align stack specially for tail calls
- if (PerformTailCallOpt && CC == CallingConv::Fast)
+ if (PerformTailCallOpt && CallConv == CallingConv::Fast)
StackSize = GetAlignedArgumentStackSize(StackSize, DAG);
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
if (isVarArg) {
- if (Is64Bit || CC != CallingConv::X86_FastCall) {
+ if (Is64Bit || CallConv != CallingConv::X86_FastCall) {
VarArgsFrameIndex = MFI->CreateFixedObject(1, StackSize);
}
if (Is64Bit) {
for (; NumIntRegs != TotalNumIntRegs; ++NumIntRegs) {
unsigned VReg = MF.addLiveIn(GPR64ArgRegs[NumIntRegs],
X86::GR64RegisterClass);
- SDValue Val = DAG.getCopyFromReg(Root, dl, VReg, MVT::i64);
+ SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i64);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN,
PseudoSourceValue::getFixedStack(RegSaveFrameIndex), 0);
for (; NumXMMRegs != TotalNumXMMRegs; ++NumXMMRegs) {
unsigned VReg = MF.addLiveIn(XMMArgRegs[NumXMMRegs],
X86::VR128RegisterClass);
- SDValue Val = DAG.getCopyFromReg(Root, dl, VReg, MVT::v4f32);
+ SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::v4f32);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN,
PseudoSourceValue::getFixedStack(RegSaveFrameIndex), 0);
DAG.getIntPtrConstant(16));
}
if (!MemOps.empty())
- Root = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOps[0], MemOps.size());
}
}
- ArgValues.push_back(Root);
-
// Some CCs need callee pop.
- if (IsCalleePop(isVarArg, CC)) {
+ if (IsCalleePop(isVarArg, CallConv)) {
BytesToPopOnReturn = StackSize; // Callee pops everything.
BytesCallerReserves = 0;
} else {
BytesToPopOnReturn = 0; // Callee pops nothing.
// If this is an sret function, the return should pop the hidden pointer.
- if (!Is64Bit && CC != CallingConv::Fast && ArgsAreStructReturn(Op))
+ if (!Is64Bit && CallConv != CallingConv::Fast && ArgsAreStructReturn(Ins))
BytesToPopOnReturn = 4;
BytesCallerReserves = StackSize;
}
if (!Is64Bit) {
RegSaveFrameIndex = 0xAAAAAAA; // RegSaveFrameIndex is X86-64 only.
- if (CC == CallingConv::X86_FastCall)
+ if (CallConv == CallingConv::X86_FastCall)
VarArgsFrameIndex = 0xAAAAAAA; // fastcc functions can't have varargs.
}
FuncInfo->setBytesToPopOnReturn(BytesToPopOnReturn);
- // Return the new list of results.
- return DAG.getNode(ISD::MERGE_VALUES, dl, Op.getNode()->getVTList(),
- &ArgValues[0], ArgValues.size()).getValue(Op.getResNo());
+ return Chain;
}
SDValue
-X86TargetLowering::LowerMemOpCallTo(CallSDNode *TheCall, SelectionDAG &DAG,
- const SDValue &StackPtr,
+X86TargetLowering::LowerMemOpCallTo(SDValue Chain,
+ SDValue StackPtr, SDValue Arg,
+ DebugLoc dl, SelectionDAG &DAG,
const CCValAssign &VA,
- SDValue Chain,
- SDValue Arg, ISD::ArgFlagsTy Flags) {
- DebugLoc dl = TheCall->getDebugLoc();
- unsigned LocMemOffset = VA.getLocMemOffset();
+ ISD::ArgFlagsTy Flags) {
+ const unsigned FirstStackArgOffset = (Subtarget->isTargetWin64() ? 32 : 0);
+ unsigned LocMemOffset = FirstStackArgOffset + VA.getLocMemOffset();
SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset);
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
if (Flags.isByVal()) {
return Chain;
}
-SDValue X86TargetLowering::LowerCALL(SDValue Op, SelectionDAG &DAG) {
+SDValue
+X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
+ unsigned CallConv, bool isVarArg, bool isTailCall,
+ const SmallVectorImpl<ISD::OutputArg> &Outs,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+
MachineFunction &MF = DAG.getMachineFunction();
- CallSDNode *TheCall = cast<CallSDNode>(Op.getNode());
- SDValue Chain = TheCall->getChain();
- unsigned CC = TheCall->getCallingConv();
- bool isVarArg = TheCall->isVarArg();
- bool IsTailCall = TheCall->isTailCall() &&
- CC == CallingConv::Fast && PerformTailCallOpt;
- SDValue Callee = TheCall->getCallee();
bool Is64Bit = Subtarget->is64Bit();
- bool IsStructRet = CallIsStructReturn(TheCall);
- DebugLoc dl = TheCall->getDebugLoc();
+ bool IsStructRet = CallIsStructReturn(Outs);
- assert(!(isVarArg && CC == CallingConv::Fast) &&
+ assert((!isTailCall ||
+ (CallConv == CallingConv::Fast && PerformTailCallOpt)) &&
+ "IsEligibleForTailCallOptimization missed a case!");
+ assert(!(isVarArg && CallConv == CallingConv::Fast) &&
"Var args not supported with calling convention fastcc");
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(TheCall, CCAssignFnForNode(CC));
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ ArgLocs, *DAG.getContext());
+ CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CallConv));
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
- if (PerformTailCallOpt && CC == CallingConv::Fast)
+ if (PerformTailCallOpt && CallConv == CallingConv::Fast)
NumBytes = GetAlignedArgumentStackSize(NumBytes, DAG);
int FPDiff = 0;
- if (IsTailCall) {
+ if (isTailCall) {
// Lower arguments at fp - stackoffset + fpdiff.
unsigned NumBytesCallerPushed =
MF.getInfo<X86MachineFunctionInfo>()->getBytesToPopOnReturn();
SDValue RetAddrFrIdx;
// Load return adress for tail calls.
- Chain = EmitTailCallLoadRetAddr(DAG, RetAddrFrIdx, Chain, IsTailCall, Is64Bit,
+ Chain = EmitTailCallLoadRetAddr(DAG, RetAddrFrIdx, Chain, isTailCall, Is64Bit,
FPDiff, dl);
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
// of tail call optimization arguments are handle later.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
- SDValue Arg = TheCall->getArg(i);
- ISD::ArgFlagsTy Flags = TheCall->getArgFlags(i);
+ MVT RegVT = VA.getLocVT();
+ SDValue Arg = Outs[i].Val;
+ ISD::ArgFlagsTy Flags = Outs[i].Flags;
bool isByVal = Flags.isByVal();
// Promote the value if needed.
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
- Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
+ Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::ZExt:
- Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
+ Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::AExt:
- Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
+ if (RegVT.isVector() && RegVT.getSizeInBits() == 128) {
+ // Special case: passing MMX values in XMM registers.
+ Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
+ Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
+ Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
+ } else
+ Arg = DAG.getNode(ISD::ANY_EXTEND, dl, RegVT, Arg);
+ break;
+ case CCValAssign::BCvt:
+ Arg = DAG.getNode(ISD::BIT_CONVERT, dl, RegVT, Arg);
break;
+ case CCValAssign::Indirect: {
+ // Store the argument.
+ SDValue SpillSlot = DAG.CreateStackTemporary(VA.getValVT());
+ int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
+ Chain = DAG.getStore(Chain, dl, Arg, SpillSlot,
+ PseudoSourceValue::getFixedStack(FI), 0);
+ Arg = SpillSlot;
+ break;
+ }
}
if (VA.isRegLoc()) {
- if (Is64Bit) {
- MVT RegVT = VA.getLocVT();
- if (RegVT.isVector() && RegVT.getSizeInBits() == 64)
- switch (VA.getLocReg()) {
- default:
- break;
- case X86::RDI: case X86::RSI: case X86::RDX: case X86::RCX:
- case X86::R8: {
- // Special case: passing MMX values in GPR registers.
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
- break;
- }
- case X86::XMM0: case X86::XMM1: case X86::XMM2: case X86::XMM3:
- case X86::XMM4: case X86::XMM5: case X86::XMM6: case X86::XMM7: {
- // Special case: passing MMX values in XMM registers.
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
- Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
- Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
- break;
- }
- }
- }
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
- if (!IsTailCall || (IsTailCall && isByVal)) {
+ if (!isTailCall || (isTailCall && isByVal)) {
assert(VA.isMemLoc());
if (StackPtr.getNode() == 0)
StackPtr = DAG.getCopyFromReg(Chain, dl, X86StackPtr, getPointerTy());
- MemOpChains.push_back(LowerMemOpCallTo(TheCall, DAG, StackPtr, VA,
- Chain, Arg, Flags));
+ MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
+ dl, DAG, VA, Flags));
}
}
}
SDValue InFlag;
// Tail call byval lowering might overwrite argument registers so in case of
// tail call optimization the copies to registers are lowered later.
- if (!IsTailCall)
+ if (!isTailCall)
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
if (Subtarget->isPICStyleGOT()) {
// ELF / PIC requires GOT in the EBX register before function calls via PLT
// GOT pointer.
- if (!IsTailCall) {
+ if (!isTailCall) {
Chain = DAG.getCopyToReg(Chain, dl, X86::EBX,
DAG.getNode(X86ISD::GlobalBaseReg,
DebugLoc::getUnknownLoc(),
// For tail calls lower the arguments to the 'real' stack slot.
- if (IsTailCall) {
+ if (isTailCall) {
+ // Force all the incoming stack arguments to be loaded from the stack
+ // before any new outgoing arguments are stored to the stack, because the
+ // outgoing stack slots may alias the incoming argument stack slots, and
+ // the alias isn't otherwise explicit. This is slightly more conservative
+ // than necessary, because it means that each store effectively depends
+ // on every argument instead of just those arguments it would clobber.
+ SDValue ArgChain = DAG.getStackArgumentTokenFactor(Chain);
+
SmallVector<SDValue, 8> MemOpChains2;
SDValue FIN;
int FI = 0;
CCValAssign &VA = ArgLocs[i];
if (!VA.isRegLoc()) {
assert(VA.isMemLoc());
- SDValue Arg = TheCall->getArg(i);
- ISD::ArgFlagsTy Flags = TheCall->getArgFlags(i);
+ SDValue Arg = Outs[i].Val;
+ ISD::ArgFlagsTy Flags = Outs[i].Flags;
// Create frame index.
int32_t Offset = VA.getLocMemOffset()+FPDiff;
uint32_t OpSize = (VA.getLocVT().getSizeInBits()+7)/8;
getPointerTy());
Source = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, Source);
- MemOpChains2.push_back(CreateCopyOfByValArgument(Source, FIN, Chain,
+ MemOpChains2.push_back(CreateCopyOfByValArgument(Source, FIN,
+ ArgChain,
Flags, DAG, dl));
} else {
// Store relative to framepointer.
MemOpChains2.push_back(
- DAG.getStore(Chain, dl, Arg, FIN,
+ DAG.getStore(ArgChain, dl, Arg, FIN,
PseudoSourceValue::getFixedStack(FI), 0));
}
}
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(),
OpFlags);
- } else if (IsTailCall) {
+ } else if (isTailCall) {
unsigned Opc = Is64Bit ? X86::R11 : X86::EAX;
Chain = DAG.getCopyToReg(Chain, dl,
Callee,InFlag);
Callee = DAG.getRegister(Opc, getPointerTy());
// Add register as live out.
- DAG.getMachineFunction().getRegInfo().addLiveOut(Opc);
+ MF.getRegInfo().addLiveOut(Opc);
}
// Returns a chain & a flag for retval copy to use.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
SmallVector<SDValue, 8> Ops;
- if (IsTailCall) {
+ if (isTailCall) {
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(0, true), InFlag);
InFlag = Chain.getValue(1);
-
- // Returns a chain & a flag for retval copy to use.
- NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
- Ops.clear();
}
Ops.push_back(Chain);
Ops.push_back(Callee);
- if (IsTailCall)
+ if (isTailCall)
Ops.push_back(DAG.getConstant(FPDiff, MVT::i32));
// Add argument registers to the end of the list so that they are known live
RegsToPass[i].second.getValueType()));
// Add an implicit use GOT pointer in EBX.
- if (!IsTailCall && Subtarget->isPICStyleGOT())
+ if (!isTailCall && Subtarget->isPICStyleGOT())
Ops.push_back(DAG.getRegister(X86::EBX, getPointerTy()));
// Add an implicit use of AL for x86 vararg functions.
if (InFlag.getNode())
Ops.push_back(InFlag);
- if (IsTailCall) {
- assert(InFlag.getNode() &&
- "Flag must be set. Depend on flag being set in LowerRET");
- Chain = DAG.getNode(X86ISD::TAILCALL, dl,
- TheCall->getVTList(), &Ops[0], Ops.size());
+ if (isTailCall) {
+ // If this is the first return lowered for this function, add the regs
+ // to the liveout set for the function.
+ if (MF.getRegInfo().liveout_empty()) {
+ SmallVector<CCValAssign, 16> RVLocs;
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(), RVLocs,
+ *DAG.getContext());
+ CCInfo.AnalyzeCallResult(Ins, RetCC_X86);
+ for (unsigned i = 0; i != RVLocs.size(); ++i)
+ if (RVLocs[i].isRegLoc())
+ MF.getRegInfo().addLiveOut(RVLocs[i].getLocReg());
+ }
+
+ assert(((Callee.getOpcode() == ISD::Register &&
+ (cast<RegisterSDNode>(Callee)->getReg() == X86::EAX ||
+ cast<RegisterSDNode>(Callee)->getReg() == X86::R9)) ||
+ Callee.getOpcode() == ISD::TargetExternalSymbol ||
+ Callee.getOpcode() == ISD::TargetGlobalAddress) &&
+ "Expecting an global address, external symbol, or register");
- return SDValue(Chain.getNode(), Op.getResNo());
+ return DAG.getNode(X86ISD::TC_RETURN, dl,
+ NodeTys, &Ops[0], Ops.size());
}
Chain = DAG.getNode(X86ISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
// Create the CALLSEQ_END node.
unsigned NumBytesForCalleeToPush;
- if (IsCalleePop(isVarArg, CC))
+ if (IsCalleePop(isVarArg, CallConv))
NumBytesForCalleeToPush = NumBytes; // Callee pops everything
- else if (!Is64Bit && CC != CallingConv::Fast && IsStructRet)
+ else if (!Is64Bit && CallConv != CallingConv::Fast && IsStructRet)
// If this is is a call to a struct-return function, the callee
// pops the hidden struct pointer, so we have to push it back.
// This is common for Darwin/X86, Linux & Mingw32 targets.
// Handle result values, copying them out of physregs into vregs that we
// return.
- return SDValue(LowerCallResult(Chain, InFlag, TheCall, CC, DAG),
- Op.getResNo());
+ return LowerCallResult(Chain, InFlag, CallConv, isVarArg,
+ Ins, dl, DAG, InVals);
}
return Offset;
}
-/// IsEligibleForTailCallElimination - Check to see whether the next instruction
-/// following the call is a return. A function is eligible if caller/callee
-/// calling conventions match, currently only fastcc supports tail calls, and
-/// the function CALL is immediatly followed by a RET.
-bool X86TargetLowering::IsEligibleForTailCallOptimization(CallSDNode *TheCall,
- SDValue Ret,
- SelectionDAG& DAG) const {
- if (!PerformTailCallOpt)
- return false;
-
- if (CheckTailCallReturnConstraints(TheCall, Ret)) {
- unsigned CallerCC =
- DAG.getMachineFunction().getFunction()->getCallingConv();
- unsigned CalleeCC = TheCall->getCallingConv();
- if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC)
- return true;
- }
-
- return false;
+/// IsEligibleForTailCallOptimization - Check whether the call is eligible
+/// for tail call optimization. Targets which want to do tail call
+/// optimization should implement this function.
+bool
+X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
+ unsigned CalleeCC,
+ bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ SelectionDAG& DAG) const {
+ MachineFunction &MF = DAG.getMachineFunction();
+ unsigned CallerCC = MF.getFunction()->getCallingConv();
+ return CalleeCC == CallingConv::Fast && CallerCC == CalleeCC;
}
FastISel *
return Mask;
}
+/// isZeroNode - Returns true if Elt is a constant zero or a floating point
+/// constant +0.0.
+bool X86::isZeroNode(SDValue Elt) {
+ return ((isa<ConstantSDNode>(Elt) &&
+ cast<ConstantSDNode>(Elt)->getZExtValue() == 0) ||
+ (isa<ConstantFPSDNode>(Elt) &&
+ cast<ConstantFPSDNode>(Elt)->getValueAPF().isPosZero()));
+}
+
/// CommuteVectorShuffle - Swap vector_shuffle operands as well as values in
/// their permute mask.
static SDValue CommuteVectorShuffle(ShuffleVectorSDNode *SVOp,
return true;
}
-/// isZeroNode - Returns true if Elt is a constant zero or a floating point
-/// constant +0.0.
-static inline bool isZeroNode(SDValue Elt) {
- return ((isa<ConstantSDNode>(Elt) &&
- cast<ConstantSDNode>(Elt)->getZExtValue() == 0) ||
- (isa<ConstantFPSDNode>(Elt) &&
- cast<ConstantFPSDNode>(Elt)->getValueAPF().isPosZero()));
-}
-
/// isZeroShuffle - Returns true if N is a VECTOR_SHUFFLE that can be resolved
/// to an zero vector.
/// FIXME: move to dag combiner / method on ShuffleVectorSDNode
unsigned Opc = V2.getOpcode();
if (Opc == ISD::UNDEF || ISD::isBuildVectorAllZeros(V2.getNode()))
continue;
- if (Opc != ISD::BUILD_VECTOR || !isZeroNode(V2.getOperand(Idx-NumElems)))
+ if (Opc != ISD::BUILD_VECTOR ||
+ !X86::isZeroNode(V2.getOperand(Idx-NumElems)))
return false;
} else if (Idx >= 0) {
unsigned Opc = V1.getOpcode();
if (Opc == ISD::UNDEF || ISD::isBuildVectorAllZeros(V1.getNode()))
continue;
- if (Opc != ISD::BUILD_VECTOR || !isZeroNode(V1.getOperand(Idx)))
+ if (Opc != ISD::BUILD_VECTOR ||
+ !X86::isZeroNode(V1.getOperand(Idx)))
return false;
}
}
continue;
}
SDValue Elt = DAG.getShuffleScalarElt(SVOp, Index);
- if (Elt.getNode() && isZeroNode(Elt))
+ if (Elt.getNode() && X86::isZeroNode(Elt))
++NumZeros;
else
break;
if (Elt.getOpcode() != ISD::Constant &&
Elt.getOpcode() != ISD::ConstantFP)
IsAllConstants = false;
- if (isZeroNode(Elt))
+ if (X86::isZeroNode(Elt))
NumZero++;
else {
NonZeros |= (1 << i);
// Is it a vector logical left shift?
if (NumElems == 2 && Idx == 1 &&
- isZeroNode(Op.getOperand(0)) && !isZeroNode(Op.getOperand(1))) {
+ X86::isZeroNode(Op.getOperand(0)) &&
+ !X86::isZeroNode(Op.getOperand(1))) {
unsigned NumBits = VT.getSizeInBits();
return getVShift(true, VT,
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32,
Op.getOperand(0))));
+ if (Op.getValueType() == MVT::v1i64 && Op.getOperand(0).getValueType() == MVT::i64)
+ return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v1i64, Op.getOperand(0));
+
SDValue AnyExt = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, Op.getOperand(0));
MVT VT = MVT::v2i32;
switch (Op.getValueType().getSimpleVT()) {
Args.push_back(Entry);
std::pair<SDValue,SDValue> CallResult =
LowerCallTo(Chain, Type::VoidTy, false, false, false, false,
- 0, CallingConv::C, false,
+ 0, CallingConv::C, false, /*isReturnValueUsed=*/false,
DAG.getExternalSymbol(bzeroEntry, IntPtr), Args, DAG, dl);
return CallResult.second;
}
case ISD::SELECT: return LowerSELECT(Op, DAG);
case ISD::BRCOND: return LowerBRCOND(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
- case ISD::CALL: return LowerCALL(Op, DAG);
- case ISD::RET: return LowerRET(Op, DAG);
- case ISD::FORMAL_ARGUMENTS: return LowerFORMAL_ARGUMENTS(Op, DAG);
case ISD::VASTART: return LowerVASTART(Op, DAG);
case ISD::VAARG: return LowerVAARG(Op, DAG);
case ISD::VACOPY: return LowerVACOPY(Op, DAG);
case X86ISD::FLD: return "X86ISD::FLD";
case X86ISD::FST: return "X86ISD::FST";
case X86ISD::CALL: return "X86ISD::CALL";
- case X86ISD::TAILCALL: return "X86ISD::TAILCALL";
case X86ISD::RDTSC_DAG: return "X86ISD::RDTSC_DAG";
case X86ISD::BT: return "X86ISD::BT";
case X86ISD::CMP: return "X86ISD::CMP";
// 'A' means EAX + EDX.
if (Constraint == "A") {
Res.first = X86::EAX;
- Res.second = X86::GRADRegisterClass;
+ Res.second = X86::GR32_ADRegisterClass;
}
return Res;
}