#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/DerivedTypes.h"
using namespace llvm;
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, EVT &ValVT,
+ EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, EVT &ValVT,
+ EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
if (TM.getSubtargetImpl()->isDarwin())
return new TargetLoweringObjectFileMachO();
- return new TargetLoweringObjectFileELF(false, 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);
// will selectively turn on ones that can be effectively codegen'd.
for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
- MVT VT = (MVT::SimpleValueType)i;
+ MVT::SimpleValueType VT = (MVT::SimpleValueType)i;
// add/sub are legal for all supported vector VT's.
setOperationAction(ISD::ADD , VT, Legal);
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";
}
}
-MVT PPCTargetLowering::getSetCCResultType(MVT VT) const {
+MVT::SimpleValueType PPCTargetLowering::getSetCCResultType(EVT VT) const {
return MVT::i32;
}
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;
}
}
if (!EnablePPCPreinc) return false;
SDValue Ptr;
- MVT VT;
+ EVT VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
VT = LD->getMemoryVT();
SDValue PPCTargetLowering::LowerConstantPool(SDValue Op,
SelectionDAG &DAG) {
- MVT PtrVT = Op.getValueType();
+ EVT PtrVT = Op.getValueType();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDValue CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
}
SDValue PPCTargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) {
- MVT PtrVT = Op.getValueType();
+ EVT PtrVT = Op.getValueType();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
SDValue Zero = DAG.getConstant(0, PtrVT);
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) {
- MVT PtrVT = Op.getValueType();
+ EVT PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDValue GA = DAG.getTargetGlobalAddress(GV, PtrVT, GSDN->getOffset());
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);
// If the global is weak or external, we have to go through the lazy
// resolution stub.
- return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Lo, NULL, 0);
+ return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Lo, NULL, 0,
+ false, false, 0);
}
SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) {
// fold the new nodes.
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
if (C->isNullValue() && CC == ISD::SETEQ) {
- MVT VT = Op.getOperand(0).getValueType();
+ EVT VT = Op.getOperand(0).getValueType();
SDValue Zext = Op.getOperand(0);
if (VT.bitsLT(MVT::i32)) {
VT = MVT::i32;
// condition register, reading it back out, and masking the correct bit. The
// normal approach here uses sub to do this instead of xor. Using xor exposes
// the result to other bit-twiddling opportunities.
- MVT LHSVT = Op.getOperand(0).getValueType();
+ EVT LHSVT = Op.getOperand(0).getValueType();
if (LHSVT.isInteger() && (CC == ISD::SETEQ || CC == ISD::SETNE)) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
SDValue Sub = DAG.getNode(ISD::XOR, dl, LHSVT, Op.getOperand(0),
Op.getOperand(1));
return DAG.getSetCC(dl, VT, Sub, DAG.getConstant(0, LHSVT), CC);
SDValue Nest = Op.getOperand(3); // 'nest' parameter value
DebugLoc dl = Op.getDebugLoc();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ 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().getTypeForMVT(),
+ 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.
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
- return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0);
+ return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0,
+ false, false, 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 {
SDValue ArgFPR = DAG.getConstant(VarArgsNumFPR, MVT::i32);
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue StackOffsetFI = DAG.getFrameIndex(VarArgsStackOffset, PtrVT);
SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
// Store first byte : number of int regs
SDValue firstStore = DAG.getTruncStore(Op.getOperand(0), dl, ArgGPR,
- Op.getOperand(1), SV, 0, MVT::i8);
+ Op.getOperand(1), SV, 0, MVT::i8,
+ false, false, 0);
uint64_t nextOffset = FPROffset;
SDValue nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, Op.getOperand(1),
ConstFPROffset);
// Store second byte : number of float regs
SDValue secondStore =
- DAG.getTruncStore(firstStore, dl, ArgFPR, nextPtr, SV, nextOffset, MVT::i8);
+ DAG.getTruncStore(firstStore, dl, ArgFPR, nextPtr, SV, nextOffset, MVT::i8,
+ false, false, 0);
nextOffset += StackOffset;
nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, nextPtr, ConstStackOffset);
// Store second word : arguments given on stack
SDValue thirdStore =
- DAG.getStore(secondStore, dl, StackOffsetFI, nextPtr, SV, nextOffset);
+ DAG.getStore(secondStore, dl, StackOffsetFI, nextPtr, SV, nextOffset,
+ false, false, 0);
nextOffset += FrameOffset;
nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, nextPtr, ConstFrameOffset);
// Store third word : arguments given in registers
- return DAG.getStore(thirdStore, dl, FR, nextPtr, SV, nextOffset);
+ return DAG.getStore(thirdStore, dl, FR, nextPtr, SV, nextOffset,
+ false, false, 0);
}
#include "PPCGenCallingConv.inc"
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
return true;
}
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, EVT &ValVT,
+ EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
return false;
}
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
+static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, EVT &ValVT,
+ EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
}
/// 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;
}
/// CalculateStackSlotSize - Calculates the size reserved for this argument on
/// the stack.
-static unsigned CalculateStackSlotSize(MVT ArgVT, ISD::ArgFlagsTy Flags,
+static unsigned CalculateStackSlotSize(EVT ArgVT, ISD::ArgFlagsTy Flags,
unsigned PtrByteSize) {
unsigned ArgSize = ArgVT.getSizeInBits()/8;
if (Flags.isByVal())
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 |
// | +-----------------------------------+
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Potential tail calls could cause overwriting of argument stack slots.
- bool isImmutable = !(PerformTailCallOpt && (CallConv==CallingConv::Fast));
+ bool isImmutable = !(GuaranteedTailCallOpt && (CallConv==CallingConv::Fast));
unsigned PtrByteSize = 4;
// Assign locations to all of the incoming arguments.
// Arguments stored in registers.
if (VA.isRegLoc()) {
TargetRegisterClass *RC;
- MVT ValVT = VA.getValVT();
+ EVT ValVT = VA.getValVT();
- switch (ValVT.getSimpleVT()) {
+ switch (ValVT.getSimpleVT().SimpleTy) {
default:
llvm_unreachable("ValVT not supported by formal arguments Lowering");
case MVT::i32:
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);
- InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN, NULL, 0));
+ InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN, NULL, 0,
+ false, false, 0));
}
}
// Make room for NumGPArgRegs and NumFPArgRegs.
int Depth = NumGPArgRegs * PtrVT.getSizeInBits()/8 +
- NumFPArgRegs * MVT(MVT::f64).getSizeInBits()/8;
+ 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
unsigned GPRIndex = 0;
for (; GPRIndex != VarArgsNumGPR; ++GPRIndex) {
SDValue Val = DAG.getRegister(GPArgRegs[GPRIndex], PtrVT);
- SDValue Store = DAG.getStore(Chain, dl, Val, FIN, NULL, 0);
+ SDValue Store = DAG.getStore(Chain, dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, PtrVT);
unsigned VReg = MF.addLiveIn(GPArgRegs[GPRIndex], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
- SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0);
+ SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, PtrVT);
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.
unsigned FPRIndex = 0;
for (FPRIndex = 0; FPRIndex != VarArgsNumFPR; ++FPRIndex) {
SDValue Val = DAG.getRegister(FPArgRegs[FPRIndex], MVT::f64);
- SDValue Store = DAG.getStore(Chain, dl, Val, FIN, NULL, 0);
+ SDValue Store = DAG.getStore(Chain, dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
- SDValue PtrOff = DAG.getConstant(MVT(MVT::f64).getSizeInBits()/8,
+ SDValue PtrOff = DAG.getConstant(EVT(MVT::f64).getSizeInBits()/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
unsigned VReg = MF.addLiveIn(FPArgRegs[FPRIndex], &PPC::F8RCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::f64);
- SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0);
+ SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
- SDValue PtrOff = DAG.getConstant(MVT(MVT::f64).getSizeInBits()/8,
+ SDValue PtrOff = DAG.getConstant(EVT(MVT::f64).getSizeInBits()/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
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();
MachineFrameInfo *MFI = MF.getFrameInfo();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
// Potential tail calls could cause overwriting of argument stack slots.
- bool isImmutable = !(PerformTailCallOpt && (CallConv==CallingConv::Fast));
+ bool isImmutable = !(GuaranteedTailCallOpt && (CallConv==CallingConv::Fast));
unsigned PtrByteSize = isPPC64 ? 8 : 4;
unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64, true);
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,
if (!isVarArg && !isPPC64) {
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e;
++ArgNo) {
- MVT ObjectVT = Ins[ArgNo].VT;
+ EVT ObjectVT = Ins[ArgNo].VT;
unsigned ObjSize = ObjectVT.getSizeInBits()/8;
ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
continue;
}
- switch(ObjectVT.getSimpleVT()) {
+ switch(ObjectVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unhandled argument type!");
case MVT::i32:
case MVT::f32:
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
SDValue ArgVal;
bool needsLoad = false;
- MVT ObjectVT = Ins[ArgNo].VT;
+ EVT ObjectVT = Ins[ArgNo].VT;
unsigned ObjSize = ObjectVT.getSizeInBits()/8;
unsigned ArgSize = ObjSize;
ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
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) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store = DAG.getTruncStore(Val.getValue(1), dl, Val, FIN,
- NULL, 0, ObjSize==1 ? MVT::i8 : MVT::i16 );
+ NULL, 0,
+ ObjSize==1 ? MVT::i8 : MVT::i16,
+ false, false, 0);
MemOps.push_back(Store);
++GPR_idx;
}
// 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);
+ SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
++GPR_idx;
ArgOffset += PtrByteSize;
continue;
}
- switch (ObjectVT.getSimpleVT()) {
+ switch (ObjectVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unhandled argument type!");
case MVT::i32:
if (!isPPC64) {
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);
+ ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, NULL, 0,
+ false, false, 0);
}
InVals.push_back(ArgVal);
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
VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
- SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0);
+ SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN, NULL, 0,
+ false, false, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, PtrVT);
for (unsigned i = 0; i != NumOps; ++i) {
SDValue Arg = Outs[i].Val;
ISD::ArgFlagsTy Flags = Outs[i].Flags;
- MVT ArgVT = Arg.getValueType();
+ EVT ArgVT = Arg.getValueType();
// Varargs Altivec parameters are padded to a 16 byte boundary.
if (ArgVT==MVT::v4f32 || ArgVT==MVT::v4i32 ||
ArgVT==MVT::v8i16 || ArgVT==MVT::v16i8) {
PPCFrameInfo::getMinCallFrameSize(isPPC64, true));
// Tail call needs the stack to be aligned.
- if (CC==CallingConv::Fast && PerformTailCallOpt) {
+ if (CC==CallingConv::Fast && GuaranteedTailCallOpt) {
unsigned TargetAlign = DAG.getMachineFunction().getTarget().getFrameInfo()->
getStackAlignment();
unsigned AlignMask = TargetAlign-1;
/// 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 (!GuaranteedTailCallOpt)
+ 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++) {
// Store relative to framepointer.
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, FIN,
PseudoSourceValue::getFixedStack(FI),
- 0));
+ 0, false, false, 0));
}
}
int NewRetAddrLoc = SPDiff + PPCFrameInfo::getReturnSaveOffset(isPPC64,
isDarwinABI);
int NewRetAddr = MF.getFrameInfo()->CreateFixedObject(SlotSize,
- NewRetAddrLoc);
- MVT VT = isPPC64 ? MVT::i64 : MVT::i32;
+ 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);
+ PseudoSourceValue::getFixedStack(NewRetAddr), 0,
+ false, false, 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);
+ PseudoSourceValue::getFixedStack(NewFPIdx), 0,
+ false, false, 0);
}
}
return Chain;
SmallVector<TailCallArgumentInfo, 8>& TailCallArguments) {
int Offset = ArgOffset + SPDiff;
uint32_t OpSize = (Arg.getValueType().getSizeInBits()+7)/8;
- int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset);
- MVT VT = isPPC64 ? MVT::i64 : MVT::i32;
+ int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset, true,false);
+ EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
SDValue FIN = DAG.getFrameIndex(FI, VT);
TailCallArgumentInfo Info;
Info.Arg = Arg;
DebugLoc dl) {
if (SPDiff) {
// Load the LR and FP stack slot for later adjusting.
- MVT VT = PPCSubTarget.isPPC64() ? MVT::i64 : MVT::i32;
+ EVT VT = PPCSubTarget.isPPC64() ? MVT::i64 : MVT::i32;
LROpOut = getReturnAddrFrameIndex(DAG);
- LROpOut = DAG.getLoad(VT, dl, Chain, LROpOut, NULL, 0);
+ LROpOut = DAG.getLoad(VT, dl, Chain, LROpOut, NULL, 0,
+ false, false, 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);
+ FPOpOut = DAG.getLoad(VT, dl, Chain, FPOpOut, NULL, 0,
+ false, false, 0);
Chain = SDValue(FPOpOut.getNode(), 1);
}
}
bool isVector, SmallVector<SDValue, 8> &MemOpChains,
SmallVector<TailCallArgumentInfo, 8>& TailCallArguments,
DebugLoc dl) {
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
if (!isTailCall) {
if (isVector) {
SDValue StackPtr;
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
DAG.getConstant(ArgOffset, PtrVT));
}
- MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0));
+ MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0,
+ false, false, 0));
// Calculate and remember argument location.
} else CalculateTailCallArgDest(DAG, MF, isPPC64, Arg, SPDiff, ArgOffset,
TailCallArguments);
unsigned PrepareCall(SelectionDAG &DAG, SDValue &Callee, SDValue &InFlag,
SDValue &Chain, DebugLoc dl, int SPDiff, bool isTailCall,
SmallVector<std::pair<unsigned, SDValue>, 8> &RegsToPass,
- SmallVector<SDValue, 8> &Ops, std::vector<MVT> &NodeTys,
- bool isSVR4ABI) {
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ SmallVector<SDValue, 8> &Ops, std::vector<EVT> &NodeTys,
+ 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) {
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
- MVT VT = VA.getValVT();
+ EVT VT = VA.getValVT();
assert(VA.isRegLoc() && "Can only return in registers!");
Chain = DAG.getCopyFromReg(Chain, dl,
VA.getLocReg(), VT, InFlag).getValue(1);
}
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<MVT> NodeTys;
+ 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
// the stack. Account for this here so these bytes can be pushed back on in
// PPCRegisterInfo::eliminateCallFramePseudoInstr.
int BytesCalleePops =
- (CallConv==CallingConv::Fast && PerformTailCallOpt) ? NumBytes : 0;
+ (CallConv==CallingConv::Fast && GuaranteedTailCallOpt) ? NumBytes : 0;
if (InFlag.getNode())
Ops.push_back(InFlag);
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,
+ 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!");
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
unsigned PtrByteSize = 4;
MachineFunction &MF = DAG.getMachineFunction();
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
- if (PerformTailCallOpt && CallConv==CallingConv::Fast)
+ if (GuaranteedTailCallOpt && CallConv==CallingConv::Fast)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
// Count how many bytes are to be pushed on the stack, including the linkage
unsigned NumArgs = Outs.size();
for (unsigned i = 0; i != NumArgs; ++i) {
- MVT ArgVT = Outs[i].Val.getValueType();
+ EVT ArgVT = Outs[i].Val.getValueType();
ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
bool Result;
if (Result) {
#ifndef NDEBUG
- cerr << "Call operand #" << i << " has unhandled type "
- << ArgVT.getMVTString() << "\n";
+ errs() << "Call operand #" << i << " has unhandled type "
+ << ArgVT.getEVTString() << "\n";
#endif
llvm_unreachable(0);
}
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
- PseudoSourceValue::getStack(), LocMemOffset));
+ PseudoSourceValue::getStack(), LocMemOffset,
+ false, false, 0));
} else {
// Calculate and remember argument location.
CalculateTailCallArgDest(DAG, MF, false, Arg, SPDiff, LocMemOffset,
// 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,
unsigned NumOps = Outs.size();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
unsigned PtrByteSize = isPPC64 ? 8 : 4;
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
- if (PerformTailCallOpt && CallConv==CallingConv::Fast)
+ if (GuaranteedTailCallOpt && CallConv==CallingConv::Fast)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
unsigned nAltivecParamsAtEnd = 0;
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;
if (Size==1 || Size==2) {
// Very small objects are passed right-justified.
// Everything else is passed left-justified.
- MVT VT = (Size==1) ? MVT::i8 : MVT::i16;
+ EVT VT = (Size==1) ? MVT::i8 : MVT::i16;
if (GPR_idx != NumGPRs) {
SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, PtrVT, Chain, Arg,
- NULL, 0, VT);
+ NULL, 0, VT, false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
SDValue Const = DAG.getConstant(j, PtrOff.getValueType());
SDValue AddArg = DAG.getNode(ISD::ADD, dl, PtrVT, Arg, Const);
if (GPR_idx != NumGPRs) {
- SDValue Load = DAG.getLoad(PtrVT, dl, Chain, AddArg, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, dl, Chain, AddArg, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
ArgOffset += PtrByteSize;
continue;
}
- switch (Arg.getValueType().getSimpleVT()) {
+ switch (Arg.getValueType().getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unexpected ValueType for argument!");
case MVT::i32:
case MVT::i64:
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
isPPC64, isTailCall, false, MemOpChains,
TailCallArguments, dl);
- inMem = true;
}
ArgOffset += PtrByteSize;
break;
RegsToPass.push_back(std::make_pair(FPR[FPR_idx++], Arg));
if (isVarArg) {
- SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0);
+ SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Store);
// Float varargs are always shadowed in available integer registers
if (GPR_idx != NumGPRs) {
- SDValue Load = DAG.getLoad(PtrVT, dl, Store, PtrOff, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, dl, Store, PtrOff, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
}
if (GPR_idx != NumGPRs && Arg.getValueType() == MVT::f64 && !isPPC64){
SDValue ConstFour = DAG.getConstant(4, PtrOff.getValueType());
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, ConstFour);
- SDValue Load = DAG.getLoad(PtrVT, dl, Store, PtrOff, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, dl, Store, PtrOff, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
}
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
isPPC64, isTailCall, false, MemOpChains,
TailCallArguments, dl);
- inMem = true;
}
if (isPPC64)
ArgOffset += 8;
// entirely in R registers. Maybe later.
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
DAG.getConstant(ArgOffset, PtrVT));
- SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0);
+ SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Store);
if (VR_idx != NumVRs) {
- SDValue Load = DAG.getLoad(MVT::v4f32, dl, Store, PtrOff, NULL, 0);
+ SDValue Load = DAG.getLoad(MVT::v4f32, dl, Store, PtrOff, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(VR[VR_idx++], Load));
}
break;
SDValue Ix = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff,
DAG.getConstant(i, PtrVT));
- SDValue Load = DAG.getLoad(PtrVT, dl, Store, Ix, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, dl, Store, Ix, NULL, 0,
+ false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
}
ArgOffset += 12*16;
for (unsigned i = 0; i != NumOps; ++i) {
SDValue Arg = Outs[i].Val;
- MVT ArgType = Arg.getValueType();
+ EVT ArgType = Arg.getValueType();
if (ArgType==MVT::v4f32 || ArgType==MVT::v4i32 ||
ArgType==MVT::v8i16 || ArgType==MVT::v16i8) {
if (++j > NumVRs) {
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,
+ false, false, 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) {
DebugLoc dl = Op.getDebugLoc();
// Get the corect type for pointers.
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ 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 SaveSP = Op.getOperand(1);
// Load the old link SP.
- SDValue LoadLinkSP = DAG.getLoad(PtrVT, dl, Chain, StackPtr, NULL, 0);
+ SDValue LoadLinkSP = DAG.getLoad(PtrVT, dl, Chain, StackPtr, NULL, 0,
+ false, false, 0);
// Restore the stack pointer.
Chain = DAG.getCopyToReg(LoadLinkSP.getValue(1), dl, SP, SaveSP);
// Store the old link SP.
- return DAG.getStore(Chain, dl, LoadLinkSP, StackPtr, NULL, 0);
+ return DAG.getStore(Chain, dl, LoadLinkSP, StackPtr, NULL, 0,
+ false, false, 0);
}
SDValue
PPCTargetLowering::getReturnAddrFrameIndex(SelectionDAG & DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
- bool IsPPC64 = PPCSubTarget.isPPC64();
+ bool isPPC64 = PPCSubTarget.isPPC64();
bool isDarwinABI = PPCSubTarget.isDarwinABI();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
// 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();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
// 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);
}
DebugLoc dl = Op.getDebugLoc();
// Get the corect type for pointers.
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Negate the size.
SDValue NegSize = DAG.getNode(ISD::SUB, dl, PtrVT,
DAG.getConstant(0, PtrVT), Size);
// Cannot handle SETEQ/SETNE.
if (CC == ISD::SETEQ || CC == ISD::SETNE) return Op;
- MVT ResVT = Op.getValueType();
- MVT CmpVT = Op.getOperand(0).getValueType();
+ EVT ResVT = Op.getValueType();
+ EVT CmpVT = Op.getOperand(0).getValueType();
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDValue TV = Op.getOperand(2), FV = Op.getOperand(3);
DebugLoc dl = Op.getDebugLoc();
Src = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Src);
SDValue Tmp;
- switch (Op.getValueType().getSimpleVT()) {
+ switch (Op.getValueType().getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unhandled FP_TO_INT type in custom expander!");
case MVT::i32:
Tmp = DAG.getNode(Op.getOpcode()==ISD::FP_TO_SINT ? PPCISD::FCTIWZ :
SDValue FIPtr = DAG.CreateStackTemporary(MVT::f64);
// Emit a store to the stack slot.
- SDValue Chain = DAG.getStore(DAG.getEntryNode(), dl, Tmp, FIPtr, NULL, 0);
+ SDValue Chain = DAG.getStore(DAG.getEntryNode(), dl, Tmp, FIPtr, NULL, 0,
+ false, false, 0);
// Result is a load from the stack slot. If loading 4 bytes, make sure to
// add in a bias.
if (Op.getValueType() == MVT::i32)
FIPtr = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr,
DAG.getConstant(4, FIPtr.getValueType()));
- return DAG.getLoad(Op.getValueType(), dl, Chain, FIPtr, NULL, 0);
+ return DAG.getLoad(Op.getValueType(), dl, Chain, FIPtr, NULL, 0,
+ false, false, 0);
}
SDValue PPCTargetLowering::LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
// 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);
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ 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);
SDValue Ext64 = DAG.getNode(PPCISD::EXTSW_32, dl, MVT::i32,
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 Ld = DAG.getLoad(MVT::f64, dl, Store, FIdx, NULL, 0, false, false, 0);
// FCFID it and return it.
SDValue FP = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Ld);
*/
MachineFunction &MF = DAG.getMachineFunction();
- MVT VT = Op.getValueType();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- std::vector<MVT> NodeTys;
+ EVT VT = Op.getValueType();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ std::vector<EVT> NodeTys;
SDValue MFFSreg, InFlag;
// Save FP Control Word to register
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);
+ StackSlot, NULL, 0, false, false, 0);
// Load FP Control Word from low 32 bits of stack slot.
SDValue Four = DAG.getConstant(4, PtrVT);
SDValue Addr = DAG.getNode(ISD::ADD, dl, PtrVT, StackSlot, Four);
- SDValue CWD = DAG.getLoad(MVT::i32, dl, Store, Addr, NULL, 0);
+ SDValue CWD = DAG.getLoad(MVT::i32, dl, Store, Addr, NULL, 0,
+ false, false, 0);
// Transform as necessary
SDValue CWD1 =
}
SDValue PPCTargetLowering::LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
DebugLoc dl = Op.getDebugLoc();
assert(Op.getNumOperands() == 3 &&
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
- MVT AmtVT = Amt.getValueType();
+ EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
}
SDValue PPCTargetLowering::LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc();
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
- MVT AmtVT = Amt.getValueType();
+ EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
SDValue PPCTargetLowering::LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
- MVT AmtVT = Amt.getValueType();
+ EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
/// BuildSplatI - Build a canonical splati of Val with an element size of
/// SplatSize. Cast the result to VT.
-static SDValue BuildSplatI(int Val, unsigned SplatSize, MVT VT,
+static SDValue BuildSplatI(int Val, unsigned SplatSize, EVT VT,
SelectionDAG &DAG, DebugLoc dl) {
assert(Val >= -16 && Val <= 15 && "vsplti is out of range!");
- static const MVT VTys[] = { // canonical VT to use for each size.
+ static const EVT VTys[] = { // canonical VT to use for each size.
MVT::v16i8, MVT::v8i16, MVT::Other, MVT::v4i32
};
- MVT ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
+ EVT ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
// Force vspltis[hw] -1 to vspltisb -1 to canonicalize.
if (Val == -1)
SplatSize = 1;
- MVT CanonicalVT = VTys[SplatSize-1];
+ EVT CanonicalVT = VTys[SplatSize-1];
// Build a canonical splat for this value.
SDValue Elt = DAG.getConstant(Val, MVT::i32);
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue LHS, SDValue RHS,
SelectionDAG &DAG, DebugLoc dl,
- MVT DestVT = MVT::Other) {
+ EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = LHS.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, MVT::i32), LHS, RHS);
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op0, SDValue Op1,
SDValue Op2, SelectionDAG &DAG,
- DebugLoc dl, MVT DestVT = MVT::Other) {
+ DebugLoc dl, EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op0.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, MVT::i32), Op0, Op1, Op2);
/// BuildVSLDOI - Return a VECTOR_SHUFFLE that is a vsldoi of the specified
/// amount. The result has the specified value type.
static SDValue BuildVSLDOI(SDValue LHS, SDValue RHS, unsigned Amt,
- MVT VT, SelectionDAG &DAG, DebugLoc dl) {
+ EVT VT, SelectionDAG &DAG, DebugLoc dl) {
// Force LHS/RHS to be the right type.
LHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, RHS);
unsigned SplatBitSize;
bool HasAnyUndefs;
if (! BVN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
- HasAnyUndefs) || SplatBitSize > 32)
+ HasAnyUndefs, 0, true) || SplatBitSize > 32)
return SDValue();
unsigned SplatBits = APSplatBits.getZExtValue();
case OP_VSLDOI12:
return BuildVSLDOI(OpLHS, OpRHS, 12, OpLHS.getValueType(), DAG, dl);
}
- MVT VT = OpLHS.getValueType();
+ EVT VT = OpLHS.getValueType();
OpLHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, OpLHS);
OpRHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, OpRHS);
SDValue T = DAG.getVectorShuffle(MVT::v16i8, dl, OpLHS, OpRHS, ShufIdxs);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
// Cases that are handled by instructions that take permute immediates
// (such as vsplt*) should be left as VECTOR_SHUFFLE nodes so they can be
// The SHUFFLE_VECTOR mask is almost exactly what we want for vperm, except
// that it is in input element units, not in bytes. Convert now.
- MVT EltVT = V1.getValueType().getVectorElementType();
+ EVT EltVT = V1.getValueType().getVectorElementType();
unsigned BytesPerElement = EltVT.getSizeInBits()/8;
SmallVector<SDValue, 16> ResultMask;
Op.getOperand(3), // RHS
DAG.getConstant(CompareOpc, MVT::i32)
};
- std::vector<MVT> VTs;
+ std::vector<EVT> VTs;
VTs.push_back(Op.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
DebugLoc dl = Op.getDebugLoc();
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
- int FrameIdx = FrameInfo->CreateStackObject(16, 16);
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ int FrameIdx = FrameInfo->CreateStackObject(16, 16, false);
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
// Store the input value into Value#0 of the stack slot.
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl,
- Op.getOperand(0), FIdx, NULL, 0);
+ Op.getOperand(0), FIdx, NULL, 0,
+ false, false, 0);
// Load it out.
- return DAG.getLoad(Op.getValueType(), dl, Store, FIdx, NULL, 0);
+ return DAG.getLoad(Op.getValueType(), dl, Store, FIdx, NULL, 0,
+ false, false, 0);
}
SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) {
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);
// This sequence changes FPSCR to do round-to-zero, adds the two halves
// of the long double, and puts FPSCR back the way it was. We do not
// actually model FPSCR.
- std::vector<MVT> NodeTys;
+ std::vector<EVT> NodeTys;
SDValue Ops[4], Result, MFFSreg, InFlag, FPreg;
NodeTys.push_back(MVT::f64); // Return register
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<MVT> 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;
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
// Create the PPCISD altivec 'dot' comparison node.
- std::vector<MVT> VTs;
+ std::vector<EVT> VTs;
SDValue Ops[] = {
LHS.getOperand(2), // LHS of compare
LHS.getOperand(3), // RHS of compare
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;
}
std::pair<unsigned, const TargetRegisterClass*>
PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
- MVT VT) const {
+ EVT VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
// to the stack.
FuncInfo->setLRStoreRequired();
return DAG.getLoad(getPointerTy(), dl,
- DAG.getEntryNode(), RetAddrFI, NULL, 0);
+ DAG.getEntryNode(), RetAddrFI, NULL, 0,
+ false, false, 0);
}
SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) {
if (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() > 0)
return SDValue();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
MachineFunction &MF = DAG.getMachineFunction();
return false;
}
-MVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size, unsigned Align,
+EVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size, unsigned Align,
bool isSrcConst, bool isSrcStr,
SelectionDAG &DAG) const {
if (this->PPCSubTarget.isPPC64()) {