setExceptionSelectorRegister(ARM::R1);
}
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
+ if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
+ else
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
+
// ARMv6 Thumb1 (except for CPUs that support dmb / dsb) and earlier use
// the default expansion.
if (Subtarget->hasAnyDataBarrier() && !Subtarget->isThumb1Only()) {
case ARMISD::PRELOAD: return "ARMISD::PRELOAD";
+ case ARMISD::WIN__CHKSTK: return "ARMISD:::WIN__CHKSTK";
+
case ARMISD::VCEQ: return "ARMISD::VCEQ";
case ARMISD::VCEQZ: return "ARMISD::VCEQZ";
case ARMISD::VCGE: return "ARMISD::VCGE";
case CallingConv::C:
if (!Subtarget->isAAPCS_ABI())
return CallingConv::ARM_APCS;
- else if (Subtarget->hasVFP2() &&
+ else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() &&
getTargetMachine().Options.FloatABIType == FloatABI::Hard &&
!isVarArg)
return CallingConv::ARM_AAPCS_VFP;
return CallingConv::ARM_AAPCS;
case CallingConv::Fast:
if (!Subtarget->isAAPCS_ABI()) {
- if (Subtarget->hasVFP2() && !isVarArg)
+ if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::Fast;
return CallingConv::ARM_APCS;
- } else if (Subtarget->hasVFP2() && !isVarArg)
+ } else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::ARM_AAPCS_VFP;
else
return CallingConv::ARM_AAPCS;
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
if (EnableARMLongCalls) {
- assert (getTargetMachine().getRelocationModel() == Reloc::Static
- && "long-calls with non-static relocation model!");
+ assert((Subtarget->isTargetWindows() ||
+ getTargetMachine().getRelocationModel() == Reloc::Static) &&
+ "long-calls with non-static relocation model!");
// Handle a global address or an external symbol. If it's not one of
// those, the target's already in a register, so we don't need to do
// anything extra.
SDLoc dl(Op);
switch (IntNo) {
default: return SDValue(); // Don't custom lower most intrinsics.
+ case Intrinsic::arm_rbit: {
+ assert(Op.getOperand(0).getValueType() == MVT::i32 &&
+ "RBIT intrinsic must have i32 type!");
+ return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(0));
+ }
case Intrinsic::arm_thread_pointer: {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
return DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
BitMask <<= 8;
ImmMask <<= 1;
}
+
+ if (DAG.getTargetLoweringInfo().isBigEndian())
+ // swap higher and lower 32 bit word
+ Imm = ((Imm & 0xf) << 4) | ((Imm & 0xf0) >> 4);
+
// Op=1, Cmode=1110.
OpCmode = 0x1e;
VT = is128Bits ? MVT::v2i64 : MVT::v1i64;
case ISD::FSINCOS: return LowerFSINCOS(Op, DAG);
case ISD::SDIVREM:
case ISD::UDIVREM: return LowerDivRem(Op, DAG);
+ case ISD::DYNAMIC_STACKALLOC:
+ if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
+ return LowerDYNAMIC_STACKALLOC(Op, DAG);
+ llvm_unreachable("Don't know how to custom lower this!");
}
}
return BB;
}
+MachineBasicBlock *
+ARMTargetLowering::EmitLowered__chkstk(MachineInstr *MI,
+ MachineBasicBlock *MBB) const {
+ const TargetMachine &TM = getTargetMachine();
+ const TargetInstrInfo &TII = *TM.getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+
+ assert(Subtarget->isTargetWindows() &&
+ "__chkstk is only supported on Windows");
+ assert(Subtarget->isThumb2() && "Windows on ARM requires Thumb-2 mode");
+
+ // __chkstk takes the number of words to allocate on the stack in R4, and
+ // returns the stack adjustment in number of bytes in R4. This will not
+ // clober any other registers (other than the obvious lr).
+ //
+ // Although, technically, IP should be considered a register which may be
+ // clobbered, the call itself will not touch it. Windows on ARM is a pure
+ // thumb-2 environment, so there is no interworking required. As a result, we
+ // do not expect a veneer to be emitted by the linker, clobbering IP.
+ //
+ // Each module receives its own copy of __chkstk, so no import thunk is
+ // required, again, ensuring that IP is not clobbered.
+ //
+ // Finally, although some linkers may theoretically provide a trampoline for
+ // out of range calls (which is quite common due to a 32M range limitation of
+ // branches for Thumb), we can generate the long-call version via
+ // -mcmodel=large, alleviating the need for the trampoline which may clobber
+ // IP.
+
+ switch (TM.getCodeModel()) {
+ case CodeModel::Small:
+ case CodeModel::Medium:
+ case CodeModel::Default:
+ case CodeModel::Kernel:
+ BuildMI(*MBB, MI, DL, TII.get(ARM::tBL))
+ .addImm((unsigned)ARMCC::AL).addReg(0)
+ .addExternalSymbol("__chkstk")
+ .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Define)
+ .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
+ break;
+ case CodeModel::Large:
+ case CodeModel::JITDefault: {
+ MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+ unsigned Reg = MRI.createVirtualRegister(&ARM::rGPRRegClass);
+
+ BuildMI(*MBB, MI, DL, TII.get(ARM::t2MOVi32imm), Reg)
+ .addExternalSymbol("__chkstk");
+ BuildMI(*MBB, MI, DL, TII.get(ARM::tBLXr))
+ .addImm((unsigned)ARMCC::AL).addReg(0)
+ .addReg(Reg, RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Define)
+ .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
+ break;
+ }
+ }
+
+ AddDefaultCC(AddDefaultPred(BuildMI(*MBB, MI, DL, TII.get(ARM::t2SUBrr),
+ ARM::SP)
+ .addReg(ARM::SP, RegState::Define)
+ .addReg(ARM::R4, RegState::Kill)));
+
+ MI->eraseFromParent();
+ return MBB;
+}
+
MachineBasicBlock *
ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) const {
case ARM::COPY_STRUCT_BYVAL_I32:
++NumLoopByVals;
return EmitStructByval(MI, BB);
+ case ARM::WIN__CHKSTK:
+ return EmitLowered__chkstk(MI, BB);
}
}
SDLoc DL(St);
SDValue WideVec = DAG.getNode(ISD::BITCAST, DL, WideVecVT, StVal);
SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
- for (unsigned i = 0; i < NumElems; ++i) ShuffleVec[i] = i * SizeRatio;
+ for (unsigned i = 0; i < NumElems; ++i)
+ ShuffleVec[i] = TLI.isBigEndian() ? (i+1) * SizeRatio - 1 : i * SizeRatio;
// Can't shuffle using an illegal type.
if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
return CallInfo.first;
}
+SDValue
+ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const {
+ assert(Subtarget->isTargetWindows() && "unsupported target platform");
+ SDLoc DL(Op);
+
+ // Get the inputs.
+ SDValue Chain = Op.getOperand(0);
+ SDValue Size = Op.getOperand(1);
+
+ SDValue Words = DAG.getNode(ISD::SRL, DL, MVT::i32, Size,
+ DAG.getConstant(2, MVT::i32));
+
+ SDValue Flag;
+ Chain = DAG.getCopyToReg(Chain, DL, ARM::R4, Words, Flag);
+ Flag = Chain.getValue(1);
+
+ SDVTList NodeTys = DAG.getVTList(MVT::i32, MVT::Glue);
+ Chain = DAG.getNode(ARMISD::WIN__CHKSTK, DL, NodeTys, Chain, Flag);
+
+ SDValue NewSP = DAG.getCopyFromReg(Chain, DL, ARM::SP, MVT::i32);
+ Chain = NewSP.getValue(1);
+
+ SDValue Ops[2] = { NewSP, Chain };
+ return DAG.getMergeValues(Ops, DL);
+}
+
bool
ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
// The ARM target isn't yet aware of offsets.
bool ARMTargetLowering::shouldExpandAtomicInIR(Instruction *Inst) const {
// Loads and stores less than 64-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
- // things go wrong:
- if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
- return SI->getValueOperand()->getType()->getPrimitiveSizeInBits() == 64;
- else if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
- return LI->getType()->getPrimitiveSizeInBits() == 64;
-
- // For the real atomic operations, we have ldrex/strex up to 64 bits.
- return Inst->getType()->getPrimitiveSizeInBits() <= 64;
+ // things go wrong. Cortex M doesn't have ldrexd/strexd though, so don't emit
+ // anything for those.
+ bool IsMClass = Subtarget->isMClass();
+ if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+ unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
+ return Size == 64 && !IsMClass;
+ } else if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+ return LI->getType()->getPrimitiveSizeInBits() == 64 && !IsMClass;
+ }
+
+ // For the real atomic operations, we have ldrex/strex up to 32 bits,
+ // and up to 64 bits on the non-M profiles
+ unsigned AtomicLimit = IsMClass ? 32 : 64;
+ return Inst->getType()->getPrimitiveSizeInBits() <= AtomicLimit;
}
Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,