ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static cl::opt<bool> EnablePPCPreinc("enable-ppc-preinc",
-cl::desc("enable preincrement load/store generation on PPC (experimental)"),
- cl::Hidden);
+static cl::opt<bool> DisablePPCPreinc("disable-ppc-preinc",
+cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
+
+static cl::opt<bool> DisableILPPref("disable-ppc-ilp-pref",
+cl::desc("disable setting the node scheduling preference to ILP on PPC"), cl::Hidden);
static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
if (TM.getSubtargetImpl()->isDarwin())
setMinStackArgumentAlignment(TM.getSubtarget<PPCSubtarget>().isPPC64() ? 8:4);
// Set up the register classes.
- addRegisterClass(MVT::i32, PPC::GPRCRegisterClass);
- addRegisterClass(MVT::f32, PPC::F4RCRegisterClass);
- addRegisterClass(MVT::f64, PPC::F8RCRegisterClass);
+ addRegisterClass(MVT::i32, &PPC::GPRCRegClass);
+ addRegisterClass(MVT::f32, &PPC::F4RCRegClass);
+ addRegisterClass(MVT::f64, &PPC::F8RCRegClass);
// PowerPC has an i16 but no i8 (or i1) SEXTLOAD
setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
if (TM.getSubtarget<PPCSubtarget>().use64BitRegs()) {
// 64-bit PowerPC implementations can support i64 types directly
- addRegisterClass(MVT::i64, PPC::G8RCRegisterClass);
+ addRegisterClass(MVT::i64, &PPC::G8RCRegClass);
// BUILD_PAIR can't be handled natively, and should be expanded to shl/or
setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
// 64-bit PowerPC wants to expand i128 shifts itself.
setOperationAction(ISD::SELECT, MVT::v4i32, Expand);
setOperationAction(ISD::STORE , MVT::v4i32, Legal);
- addRegisterClass(MVT::v4f32, PPC::VRRCRegisterClass);
- addRegisterClass(MVT::v4i32, PPC::VRRCRegisterClass);
- addRegisterClass(MVT::v8i16, PPC::VRRCRegisterClass);
- addRegisterClass(MVT::v16i8, PPC::VRRCRegisterClass);
+ addRegisterClass(MVT::v4f32, &PPC::VRRCRegClass);
+ addRegisterClass(MVT::v4i32, &PPC::VRRCRegClass);
+ addRegisterClass(MVT::v8i16, &PPC::VRRCRegClass);
+ addRegisterClass(MVT::v16i8, &PPC::VRRCRegClass);
setOperationAction(ISD::MUL, MVT::v4f32, Legal);
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
+ if (TM.getSubtarget<PPCSubtarget>().has64BitSupport())
+ setOperationAction(ISD::PREFETCH, MVT::Other, Legal);
+
setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
// Darwin passes everything on 4 byte boundary.
if (TM.getSubtarget<PPCSubtarget>().isDarwin())
return 4;
- // FIXME SVR4 TBD
+
+ // 16byte and wider vectors are passed on 16byte boundary.
+ if (VectorType *VTy = dyn_cast<VectorType>(Ty))
+ if (VTy->getBitWidth() >= 128)
+ return 16;
+
+ // The rest is 8 on PPC64 and 4 on PPC32 boundary.
+ if (PPCSubTarget.isPPC64())
+ return 8;
+
return 4;
}
APInt LHSKnownZero, LHSKnownOne;
APInt RHSKnownZero, RHSKnownOne;
DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
LHSKnownZero, LHSKnownOne);
if (LHSKnownZero.getBoolValue()) {
DAG.ComputeMaskedBits(N.getOperand(1),
- APInt::getAllOnesValue(N.getOperand(1)
- .getValueSizeInBits()),
RHSKnownZero, RHSKnownOne);
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
return true; // [r+i]
} else if (N.getOperand(1).getOpcode() == PPCISD::Lo) {
// Match LOAD (ADD (X, Lo(G))).
- assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getZExtValue()
+ assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getZExtValue()
&& "Cannot handle constant offsets yet!");
Disp = N.getOperand(1).getOperand(0); // The global address.
assert(Disp.getOpcode() == ISD::TargetGlobalAddress ||
+ Disp.getOpcode() == ISD::TargetGlobalTLSAddress ||
Disp.getOpcode() == ISD::TargetConstantPool ||
Disp.getOpcode() == ISD::TargetJumpTable);
Base = N.getOperand(0);
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
APInt LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
- LHSKnownZero, LHSKnownOne);
+ DAG.ComputeMaskedBits(N.getOperand(0), LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
if (N.getOpcode() == ISD::ADD) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) && (imm & 3) == 0) {
- Disp = DAG.getTargetConstant(((int)imm & 0xFFFF) >> 2, MVT::i32);
+ Disp = DAG.getTargetConstant(((int)imm & 0xFFFF) >> 2, MVT::i32);
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0))) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
} else {
return true; // [r+i]
} else if (N.getOperand(1).getOpcode() == PPCISD::Lo) {
// Match LOAD (ADD (X, Lo(G))).
- assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getZExtValue()
+ assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getZExtValue()
&& "Cannot handle constant offsets yet!");
Disp = N.getOperand(1).getOperand(0); // The global address.
assert(Disp.getOpcode() == ISD::TargetGlobalAddress ||
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
APInt LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
- LHSKnownZero, LHSKnownOne);
+ DAG.ComputeMaskedBits(N.getOperand(0), LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
- // Disabled by default for now.
- if (!EnablePPCPreinc) return false;
+ if (DisablePPCPreinc) return false;
SDValue Ptr;
EVT VT;
if (VT.isVector())
return false;
- // TODO: Check reg+reg first.
+ if (SelectAddressRegReg(Ptr, Offset, Base, DAG)) {
+ if (isa<StoreSDNode>(N)) {
+ AM = ISD::PRE_INC;
+ return true;
+ }
+
+ // FIXME: reg+reg preinc loads
+ return false;
+ }
// LDU/STU use reg+imm*4, others use reg+imm.
if (VT != MVT::i64) {
return LowerLabelRef(TgtBAHi, TgtBALo, isPIC, DAG);
}
+SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
+ SelectionDAG &DAG) const {
+
+ GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
+ DebugLoc dl = GA->getDebugLoc();
+ const GlobalValue *GV = GA->getGlobal();
+ EVT PtrVT = getPointerTy();
+ bool is64bit = PPCSubTarget.isPPC64();
+
+ TLSModel::Model model = getTargetMachine().getTLSModel(GV);
+
+ SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+ PPCII::MO_TPREL16_HA);
+ SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+ PPCII::MO_TPREL16_LO);
+
+ if (model != TLSModel::LocalExec)
+ llvm_unreachable("only local-exec TLS mode supported");
+ SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
+ is64bit ? MVT::i64 : MVT::i32);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
+ return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
+}
+
SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
Entry.Node = Nest; Args.push_back(Entry);
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
- std::pair<SDValue, SDValue> CallResult =
- LowerCallTo(Chain, Type::getVoidTy(*DAG.getContext()),
- false, false, false, false, 0, CallingConv::C,
+ TargetLowering::CallLoweringInfo CLI(Chain,
+ Type::getVoidTy(*DAG.getContext()),
+ false, false, false, false, 0,
+ CallingConv::C,
/*isTailCall=*/false,
- /*doesNotRet=*/false, /*isReturnValueUsed=*/true,
+ /*doesNotRet=*/false,
+ /*isReturnValueUsed=*/true,
DAG.getExternalSymbol("__trampoline_setup", PtrVT),
Args, DAG, dl);
+ std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
return CallResult.second;
}
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ArgLocs, *DAG.getContext());
+ getTargetMachine(), ArgLocs, *DAG.getContext());
// Reserve space for the linkage area on the stack.
CCInfo.AllocateStack(PPCFrameLowering::getLinkageSize(false, false), PtrByteSize);
default:
llvm_unreachable("ValVT not supported by formal arguments Lowering");
case MVT::i32:
- RC = PPC::GPRCRegisterClass;
+ RC = &PPC::GPRCRegClass;
break;
case MVT::f32:
- RC = PPC::F4RCRegisterClass;
+ RC = &PPC::F4RCRegClass;
break;
case MVT::f64:
- RC = PPC::F8RCRegisterClass;
+ RC = &PPC::F8RCRegClass;
break;
case MVT::v16i8:
case MVT::v8i16:
case MVT::v4i32:
case MVT::v4f32:
- RC = PPC::VRRCRegisterClass;
+ RC = &PPC::VRRCRegClass;
break;
}
// caller's stack frame, right above the parameter list area.
SmallVector<CCValAssign, 16> ByValArgLocs;
CCState CCByValInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ByValArgLocs, *DAG.getContext());
+ getTargetMachine(), ByValArgLocs, *DAG.getContext());
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
SmallVector<CCValAssign, 16> RVLocs;
CCState CCRetInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext());
+ getTargetMachine(), RVLocs, *DAG.getContext());
CCRetInfo.AnalyzeCallResult(Ins, RetCC_PPC);
// Copy all of the result registers out of their specified physreg.
if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext());
+ getTargetMachine(), RVLocs, *DAG.getContext());
CCInfo.AnalyzeCallResult(Ins, RetCC_PPC);
for (unsigned i = 0; i != RVLocs.size(); ++i)
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
SDValue
-PPCTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
- CallingConv::ID CallConv, bool isVarArg,
- bool doesNotRet, bool &isTailCall,
- const SmallVectorImpl<ISD::OutputArg> &Outs,
- const SmallVectorImpl<SDValue> &OutVals,
- const SmallVectorImpl<ISD::InputArg> &Ins,
- DebugLoc dl, SelectionDAG &DAG,
+PPCTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
+ SelectionDAG &DAG = CLI.DAG;
+ DebugLoc &dl = CLI.DL;
+ SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
+ SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
+ SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
+ SDValue Chain = CLI.Chain;
+ SDValue Callee = CLI.Callee;
+ bool &isTailCall = CLI.IsTailCall;
+ CallingConv::ID CallConv = CLI.CallConv;
+ bool isVarArg = CLI.IsVarArg;
+
if (isTailCall)
isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv, isVarArg,
Ins, DAG);
// Assign locations to all of the outgoing arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ArgLocs, *DAG.getContext());
+ getTargetMachine(), ArgLocs, *DAG.getContext());
// Reserve space for the linkage area on the stack.
CCInfo.AllocateStack(PPCFrameLowering::getLinkageSize(false, false), PtrByteSize);
// Assign locations to all of the outgoing aggregate by value arguments.
SmallVector<CCValAssign, 16> ByValArgLocs;
CCState CCByValInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ByValArgLocs, *DAG.getContext());
+ getTargetMachine(), ByValArgLocs, *DAG.getContext());
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext());
+ getTargetMachine(), RVLocs, *DAG.getContext());
CCInfo.AnalyzeReturn(Outs, RetCC_PPC);
// If this is the first return lowered for this function, add the regs to the
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
- case ISD::GlobalTLSAddress: llvm_unreachable("TLS not implemented for PPC");
+ case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SETCC: return LowerSETCC(Op, DAG);
case ISD::INIT_TRAMPOLINE: return LowerINIT_TRAMPOLINE(Op, DAG);
//===----------------------------------------------------------------------===//
void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
- const APInt &Mask,
APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
- KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
+ KnownZero = KnownOne = APInt(KnownZero.getBitWidth(), 0);
switch (Op.getOpcode()) {
default: break;
case PPCISD::LBRX: {
case 'b': // R1-R31
case 'r': // R0-R31
if (VT == MVT::i64 && PPCSubTarget.isPPC64())
- return std::make_pair(0U, PPC::G8RCRegisterClass);
- return std::make_pair(0U, PPC::GPRCRegisterClass);
+ return std::make_pair(0U, &PPC::G8RCRegClass);
+ return std::make_pair(0U, &PPC::GPRCRegClass);
case 'f':
if (VT == MVT::f32)
- return std::make_pair(0U, PPC::F4RCRegisterClass);
- else if (VT == MVT::f64)
- return std::make_pair(0U, PPC::F8RCRegisterClass);
+ return std::make_pair(0U, &PPC::F4RCRegClass);
+ if (VT == MVT::f64)
+ return std::make_pair(0U, &PPC::F8RCRegClass);
break;
case 'v':
- return std::make_pair(0U, PPC::VRRCRegisterClass);
+ return std::make_pair(0U, &PPC::VRRCRegClass);
case 'y': // crrc
- return std::make_pair(0U, PPC::CRRCRegisterClass);
+ return std::make_pair(0U, &PPC::CRRCRegClass);
}
}
return MVT::i32;
}
}
+
+Sched::Preference PPCTargetLowering::getSchedulingPreference(SDNode *N) const {
+ if (DisableILPPref)
+ return TargetLowering::getSchedulingPreference(N);
+
+ return Sched::ILP;
+}
+
projects / oota-llvm.git / blobdiff