}
void ARMTargetLowering::addQRTypeForNEON(MVT VT) {
- addRegisterClass(VT, &ARM::QPRRegClass);
+ addRegisterClass(VT, &ARM::DPairRegClass);
addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}
static TargetLoweringObjectFile *createTLOF(TargetMachine &TM) {
- if (TM.getSubtarget<ARMSubtarget>().isTargetDarwin())
+ if (TM.getSubtarget<ARMSubtarget>().isTargetMachO())
return new TargetLoweringObjectFileMachO();
return new ARMElfTargetObjectFile();
setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
- if (Subtarget->isTargetIOS()) {
+ if (Subtarget->isTargetMachO()) {
// Uses VFP for Thumb libfuncs if available.
if (Subtarget->isThumb() && Subtarget->hasVFP2() &&
Subtarget->hasARMOps()) {
setLibcallName(RTLIB::SRL_I128, 0);
setLibcallName(RTLIB::SRA_I128, 0);
- if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetDarwin()) {
+ if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetMachO()) {
// Double-precision floating-point arithmetic helper functions
// RTABI chapter 4.1.2, Table 2
setLibcallName(RTLIB::ADD_F64, "__aeabi_dadd");
setOperationAction(ISD::FP_ROUND, MVT::v2f32, Expand);
setOperationAction(ISD::FP_EXTEND, MVT::v2f64, Expand);
- // Custom expand long extensions to vectors.
- setOperationAction(ISD::SIGN_EXTEND, MVT::v8i32, Custom);
- setOperationAction(ISD::ZERO_EXTEND, MVT::v8i32, Custom);
- setOperationAction(ISD::SIGN_EXTEND, MVT::v4i64, Custom);
- setOperationAction(ISD::ZERO_EXTEND, MVT::v4i64, Custom);
- setOperationAction(ISD::SIGN_EXTEND, MVT::v16i32, Custom);
- setOperationAction(ISD::ZERO_EXTEND, MVT::v16i32, Custom);
- setOperationAction(ISD::SIGN_EXTEND, MVT::v8i64, Custom);
- setOperationAction(ISD::ZERO_EXTEND, MVT::v8i64, Custom);
-
// NEON does not have single instruction CTPOP for vectors with element
// types wider than 8-bits. However, custom lowering can leverage the
// v8i8/v16i8 vcnt instruction.
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
- if (!Subtarget->isTargetDarwin()) {
- // Non-Darwin platforms may return values in these registers via the
+ if (!Subtarget->isTargetMachO()) {
+ // Non-MachO platforms may return values in these registers via the
// personality function.
setExceptionPointerRegister(ARM::R0);
setExceptionSelectorRegister(ARM::R1);
setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand);
}
}
+
+ // Combine sin / cos into one node or libcall if possible.
+ if (Subtarget->hasSinCos()) {
+ setLibcallName(RTLIB::SINCOS_F32, "sincosf");
+ setLibcallName(RTLIB::SINCOS_F64, "sincos");
+ if (Subtarget->getTargetTriple().getOS() == Triple::IOS) {
+ // For iOS, we don't want to the normal expansion of a libcall to
+ // sincos. We want to issue a libcall to __sincos_stret.
+ setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
+ setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
+ }
+ }
// We have target-specific dag combine patterns for the following nodes:
// ARMISD::VMOVRRD - No need to call setTargetDAGCombine
switch (Opcode) {
default: return 0;
case ARMISD::Wrapper: return "ARMISD::Wrapper";
- case ARMISD::WrapperDYN: return "ARMISD::WrapperDYN";
case ARMISD::WrapperPIC: return "ARMISD::WrapperPIC";
case ARMISD::WrapperJT: return "ARMISD::WrapperJT";
case ARMISD::CALL: return "ARMISD::CALL";
case ARMISD::VSHL: return "ARMISD::VSHL";
case ARMISD::VSHRs: return "ARMISD::VSHRs";
case ARMISD::VSHRu: return "ARMISD::VSHRu";
- case ARMISD::VSHLLs: return "ARMISD::VSHLLs";
- case ARMISD::VSHLLu: return "ARMISD::VSHLLu";
- case ARMISD::VSHLLi: return "ARMISD::VSHLLi";
- case ARMISD::VSHRN: return "ARMISD::VSHRN";
case ARMISD::VRSHRs: return "ARMISD::VRSHRs";
case ARMISD::VRSHRu: return "ARMISD::VRSHRu";
case ARMISD::VRSHRN: return "ARMISD::VRSHRN";
const GlobalValue *GV = G->getGlobal();
isDirect = true;
bool isExt = GV->isDeclaration() || GV->isWeakForLinker();
- bool isStub = (isExt && Subtarget->isTargetDarwin()) &&
+ bool isStub = (isExt && Subtarget->isTargetMachO()) &&
getTargetMachine().getRelocationModel() != Reloc::Static;
isARMFunc = !Subtarget->isThumb() || isStub;
// ARM call to a local ARM function is predicable.
isLocalARMFunc = !Subtarget->isThumb() && (!isExt || !ARMInterworking);
// tBX takes a register source operand.
- if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
- unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV =
- ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 4);
- SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
- CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
- Callee = DAG.getLoad(getPointerTy(), dl,
- DAG.getEntryNode(), CPAddr,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 0);
- SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
- Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
- getPointerTy(), Callee, PICLabel);
+ if (isStub && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
+ assert(Subtarget->isTargetMachO() && "WrapperPIC use on non-MachO?");
+ Callee = DAG.getNode(ARMISD::WrapperPIC, dl, getPointerTy(),
+ DAG.getTargetGlobalAddress(GV, dl, getPointerTy()));
} else {
// On ELF targets for PIC code, direct calls should go through the PLT
unsigned OpFlags = 0;
}
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
isDirect = true;
- bool isStub = Subtarget->isTargetDarwin() &&
+ bool isStub = Subtarget->isTargetMachO() &&
getTargetMachine().getRelocationModel() != Reloc::Static;
isARMFunc = !Subtarget->isThumb() || isStub;
// tBX takes a register source operand.
// FIXME: handle tail calls differently.
unsigned CallOpc;
- bool HasMinSizeAttr = MF.getFunction()->getAttributes().
- hasAttribute(AttributeSet::FunctionIndex, Attribute::MinSize);
+ bool HasMinSizeAttr = Subtarget->isMinSize();
if (Subtarget->isThumb()) {
if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
CallOpc = ARMISD::CALL_NOLINK;
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
- // FIXME: Enable this for static codegen when tool issues are fixed. Also
- // update ARMFastISel::ARMMaterializeGV.
- if (Subtarget->useMovt() && RelocM != Reloc::Static) {
+ if (Subtarget->useMovt())
++NumMovwMovt;
- // FIXME: Once remat is capable of dealing with instructions with register
- // operands, expand this into two nodes.
- if (RelocM == Reloc::Static)
- return DAG.getNode(ARMISD::Wrapper, dl, PtrVT,
- DAG.getTargetGlobalAddress(GV, dl, PtrVT));
-
- unsigned Wrapper = (RelocM == Reloc::PIC_)
- ? ARMISD::WrapperPIC : ARMISD::WrapperDYN;
- SDValue Result = DAG.getNode(Wrapper, dl, PtrVT,
- DAG.getTargetGlobalAddress(GV, dl, PtrVT));
- if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
- Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
- MachinePointerInfo::getGOT(),
- false, false, false, 0);
- return Result;
- }
- unsigned ARMPCLabelIndex = 0;
- SDValue CPAddr;
- if (RelocM == Reloc::Static) {
- CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
- } else {
- ARMFunctionInfo *AFI = DAG.getMachineFunction().getInfo<ARMFunctionInfo>();
- ARMPCLabelIndex = AFI->createPICLabelUId();
- unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8);
- ARMConstantPoolValue *CPV =
- ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue,
- PCAdj);
- CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
- }
- CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
+ // FIXME: Once remat is capable of dealing with instructions with register
+ // operands, expand this into multiple nodes
+ unsigned Wrapper =
+ RelocM == Reloc::PIC_ ? ARMISD::WrapperPIC : ARMISD::Wrapper;
- SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 0);
- SDValue Chain = Result.getValue(1);
-
- if (RelocM == Reloc::PIC_) {
- SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
- Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
- }
+ SDValue G = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, ARMII::MO_NONLAZY);
+ SDValue Result = DAG.getNode(Wrapper, dl, PtrVT, G);
if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
- Result = DAG.getLoad(PtrVT, dl, Chain, Result, MachinePointerInfo::getGOT(),
- false, false, false, 0);
-
+ Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
+ MachinePointerInfo::getGOT(), false, false, false, 0);
return Result;
}
ArgRegsSize = NumGPRs * 4;
// If parameter is split between stack and GPRs...
- if (NumGPRs && Align == 8 &&
+ if (NumGPRs && Align > 4 &&
(ArgRegsSize < ArgSize ||
InRegsParamRecordIdx >= CCInfo.getInRegsParamsCount())) {
- // Add padding for part of param recovered from GPRs, so
- // its last byte must be at address K*8 - 1.
+ // Add padding for part of param recovered from GPRs. For example,
+ // if Align == 8, its last byte must be at address K*8 - 1.
// We need to do it, since remained (stack) part of parameter has
// stack alignment, and we need to "attach" "GPRs head" without gaps
// to it:
//
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned Padding =
- ((ArgRegsSize + AFI->getArgRegsSaveSize() + Align - 1) & ~(Align-1)) -
- (ArgRegsSize + AFI->getArgRegsSaveSize());
+ OffsetToAlignment(ArgRegsSize + AFI->getArgRegsSaveSize(), Align);
ArgRegsSaveSize = ArgRegsSize + Padding;
} else
// We don't need to extend regs save size for byval parameters if they
bool ForceMutable) const {
// Currently, two use-cases possible:
- // Case #1. Non var-args function, and we meet first byval parameter.
+ // Case #1. Non-var-args function, and we meet first byval parameter.
// Setup first unallocated register as first byval register;
// eat all remained registers
// (these two actions are performed by HandleByVal method).
static ISD::CondCode getInverseCCForVSEL(ISD::CondCode CC) {
if (CC == ISD::SETNE)
return ISD::SETEQ;
- return ISD::getSetCCSwappedOperands(CC);
+ return ISD::getSetCCInverse(CC, true);
}
static void checkVSELConstraints(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setReturnAddressIsTaken(true);
+ if (verifyReturnAddressArgumentIsConstant(Op, DAG))
+ return SDValue();
+
EVT VT = Op.getValueType();
SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
EVT VT = Op.getValueType();
SDLoc dl(Op); // FIXME probably not meaningful
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
- unsigned FrameReg = (Subtarget->isThumb() || Subtarget->isTargetDarwin())
+ unsigned FrameReg = (Subtarget->isThumb() || Subtarget->isTargetMachO())
? ARM::R7 : ARM::R11;
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, VT);
while (Depth--)
return FrameAddr;
}
-/// Custom Expand long vector extensions, where size(DestVec) > 2*size(SrcVec),
-/// and size(DestVec) > 128-bits.
-/// This is achieved by doing the one extension from the SrcVec, splitting the
-/// result, extending these parts, and then concatenating these into the
-/// destination.
-static SDValue ExpandVectorExtension(SDNode *N, SelectionDAG &DAG) {
- SDValue Op = N->getOperand(0);
- EVT SrcVT = Op.getValueType();
- EVT DestVT = N->getValueType(0);
-
- assert(DestVT.getSizeInBits() > 128 &&
- "Custom sext/zext expansion needs >128-bit vector.");
- // If this is a normal length extension, use the default expansion.
- if (SrcVT.getSizeInBits()*4 != DestVT.getSizeInBits() &&
- SrcVT.getSizeInBits()*8 != DestVT.getSizeInBits())
- return SDValue();
-
- SDLoc dl(N);
- unsigned SrcEltSize = SrcVT.getVectorElementType().getSizeInBits();
- unsigned DestEltSize = DestVT.getVectorElementType().getSizeInBits();
- unsigned NumElts = SrcVT.getVectorNumElements();
- LLVMContext &Ctx = *DAG.getContext();
- SDValue Mid, SplitLo, SplitHi, ExtLo, ExtHi;
-
- EVT MidVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, SrcEltSize*2),
- NumElts);
- EVT SplitVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, SrcEltSize*2),
- NumElts/2);
- EVT ExtVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, DestEltSize),
- NumElts/2);
-
- Mid = DAG.getNode(N->getOpcode(), dl, MidVT, Op);
- SplitLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitVT, Mid,
- DAG.getIntPtrConstant(0));
- SplitHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitVT, Mid,
- DAG.getIntPtrConstant(NumElts/2));
- ExtLo = DAG.getNode(N->getOpcode(), dl, ExtVT, SplitLo);
- ExtHi = DAG.getNode(N->getOpcode(), dl, ExtVT, SplitHi);
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, DestVT, ExtLo, ExtHi);
-}
-
/// ExpandBITCAST - If the target supports VFP, this function is called to
/// expand a bit convert where either the source or destination type is i64 to
/// use a VMOVDRR or VMOVRRD node. This should not be done when the non-i64
Op.getOperand(1), Op.getOperand(2));
}
+SDValue ARMTargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
+ assert(Subtarget->isTargetDarwin());
+
+ // For iOS, we want to call an alternative entry point: __sincos_stret,
+ // return values are passed via sret.
+ SDLoc dl(Op);
+ SDValue Arg = Op.getOperand(0);
+ EVT ArgVT = Arg.getValueType();
+ Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+
+ MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+ // Pair of floats / doubles used to pass the result.
+ StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+
+ // Create stack object for sret.
+ const uint64_t ByteSize = TLI.getDataLayout()->getTypeAllocSize(RetTy);
+ const unsigned StackAlign = TLI.getDataLayout()->getPrefTypeAlignment(RetTy);
+ int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
+ SDValue SRet = DAG.getFrameIndex(FrameIdx, TLI.getPointerTy());
+
+ ArgListTy Args;
+ ArgListEntry Entry;
+
+ Entry.Node = SRet;
+ Entry.Ty = RetTy->getPointerTo();
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Entry.isSRet = true;
+ Args.push_back(Entry);
+
+ Entry.Node = Arg;
+ Entry.Ty = ArgTy;
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Args.push_back(Entry);
+
+ const char *LibcallName = (ArgVT == MVT::f64)
+ ? "__sincos_stret" : "__sincosf_stret";
+ SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());
+
+ TargetLowering::
+ CallLoweringInfo CLI(DAG.getEntryNode(), Type::getVoidTy(*DAG.getContext()),
+ false, false, false, false, 0,
+ CallingConv::C, /*isTaillCall=*/false,
+ /*doesNotRet=*/false, /*isReturnValueUsed*/false,
+ Callee, Args, DAG, dl);
+ std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
+
+ SDValue LoadSin = DAG.getLoad(ArgVT, dl, CallResult.second, SRet,
+ MachinePointerInfo(), false, false, false, 0);
+
+ // Address of cos field.
+ SDValue Add = DAG.getNode(ISD::ADD, dl, getPointerTy(), SRet,
+ DAG.getIntPtrConstant(ArgVT.getStoreSize()));
+ SDValue LoadCos = DAG.getLoad(ArgVT, dl, LoadSin.getValue(1), Add,
+ MachinePointerInfo(), false, false, false, 0);
+
+ SDVTList Tys = DAG.getVTList(ArgVT, ArgVT);
+ return DAG.getNode(ISD::MERGE_VALUES, dl, Tys,
+ LoadSin.getValue(0), LoadCos.getValue(0));
+}
+
static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) {
// Monotonic load/store is legal for all targets
if (cast<AtomicSDNode>(Op)->getOrdering() <= Monotonic)
return Op;
- // Aquire/Release load/store is not legal for targets without a
+ // Acquire/Release load/store is not legal for targets without a
// dmb or equivalent available.
return SDValue();
}
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
case ISD::GlobalAddress:
- return Subtarget->isTargetDarwin() ? LowerGlobalAddressDarwin(Op, DAG) :
+ return Subtarget->isTargetMachO() ? LowerGlobalAddressDarwin(Op, DAG) :
LowerGlobalAddressELF(Op, DAG);
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::SELECT: return LowerSELECT(Op, DAG);
case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
case ISD::ATOMIC_LOAD:
case ISD::ATOMIC_STORE: return LowerAtomicLoadStore(Op, DAG);
+ case ISD::FSINCOS: return LowerFSINCOS(Op, DAG);
case ISD::SDIVREM:
case ISD::UDIVREM: return LowerDivRem(Op, DAG);
}
case ISD::BITCAST:
Res = ExpandBITCAST(N, DAG);
break;
- case ISD::SIGN_EXTEND:
- case ISD::ZERO_EXTEND:
- Res = ExpandVectorExtension(N, DAG);
- break;
case ISD::SRL:
case ISD::SRA:
Res = Expand64BitShift(N, DAG, Subtarget);
llvm_unreachable("Expecting a BB with two successors!");
}
+/// Return the load opcode for a given load size. If load size >= 8,
+/// neon opcode will be returned.
+static unsigned getLdOpcode(unsigned LdSize, bool IsThumb1, bool IsThumb2) {
+ if (LdSize >= 8)
+ return LdSize == 16 ? ARM::VLD1q32wb_fixed
+ : LdSize == 8 ? ARM::VLD1d32wb_fixed : 0;
+ if (IsThumb1)
+ return LdSize == 4 ? ARM::tLDRi
+ : LdSize == 2 ? ARM::tLDRHi
+ : LdSize == 1 ? ARM::tLDRBi : 0;
+ if (IsThumb2)
+ return LdSize == 4 ? ARM::t2LDR_POST
+ : LdSize == 2 ? ARM::t2LDRH_POST
+ : LdSize == 1 ? ARM::t2LDRB_POST : 0;
+ return LdSize == 4 ? ARM::LDR_POST_IMM
+ : LdSize == 2 ? ARM::LDRH_POST
+ : LdSize == 1 ? ARM::LDRB_POST_IMM : 0;
+}
+
+/// Return the store opcode for a given store size. If store size >= 8,
+/// neon opcode will be returned.
+static unsigned getStOpcode(unsigned StSize, bool IsThumb1, bool IsThumb2) {
+ if (StSize >= 8)
+ return StSize == 16 ? ARM::VST1q32wb_fixed
+ : StSize == 8 ? ARM::VST1d32wb_fixed : 0;
+ if (IsThumb1)
+ return StSize == 4 ? ARM::tSTRi
+ : StSize == 2 ? ARM::tSTRHi
+ : StSize == 1 ? ARM::tSTRBi : 0;
+ if (IsThumb2)
+ return StSize == 4 ? ARM::t2STR_POST
+ : StSize == 2 ? ARM::t2STRH_POST
+ : StSize == 1 ? ARM::t2STRB_POST : 0;
+ return StSize == 4 ? ARM::STR_POST_IMM
+ : StSize == 2 ? ARM::STRH_POST
+ : StSize == 1 ? ARM::STRB_POST_IMM : 0;
+}
+
+/// Emit a post-increment load operation with given size. The instructions
+/// will be added to BB at Pos.
+static void emitPostLd(MachineBasicBlock *BB, MachineInstr *Pos,
+ const TargetInstrInfo *TII, DebugLoc dl,
+ unsigned LdSize, unsigned Data, unsigned AddrIn,
+ unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
+ unsigned LdOpc = getLdOpcode(LdSize, IsThumb1, IsThumb2);
+ assert(LdOpc != 0 && "Should have a load opcode");
+ if (LdSize >= 8) {
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
+ .addReg(AddrOut, RegState::Define).addReg(AddrIn)
+ .addImm(0));
+ } else if (IsThumb1) {
+ // load + update AddrIn
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
+ .addReg(AddrIn).addImm(0));
+ MachineInstrBuilder MIB =
+ BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
+ MIB = AddDefaultT1CC(MIB);
+ MIB.addReg(AddrIn).addImm(LdSize);
+ AddDefaultPred(MIB);
+ } else if (IsThumb2) {
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
+ .addReg(AddrOut, RegState::Define).addReg(AddrIn)
+ .addImm(LdSize));
+ } else { // arm
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
+ .addReg(AddrOut, RegState::Define).addReg(AddrIn)
+ .addReg(0).addImm(LdSize));
+ }
+}
+
+/// Emit a post-increment store operation with given size. The instructions
+/// will be added to BB at Pos.
+static void emitPostSt(MachineBasicBlock *BB, MachineInstr *Pos,
+ const TargetInstrInfo *TII, DebugLoc dl,
+ unsigned StSize, unsigned Data, unsigned AddrIn,
+ unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
+ unsigned StOpc = getStOpcode(StSize, IsThumb1, IsThumb2);
+ assert(StOpc != 0 && "Should have a store opcode");
+ if (StSize >= 8) {
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
+ .addReg(AddrIn).addImm(0).addReg(Data));
+ } else if (IsThumb1) {
+ // store + update AddrIn
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc)).addReg(Data)
+ .addReg(AddrIn).addImm(0));
+ MachineInstrBuilder MIB =
+ BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
+ MIB = AddDefaultT1CC(MIB);
+ MIB.addReg(AddrIn).addImm(StSize);
+ AddDefaultPred(MIB);
+ } else if (IsThumb2) {
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
+ .addReg(Data).addReg(AddrIn).addImm(StSize));
+ } else { // arm
+ AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
+ .addReg(Data).addReg(AddrIn).addReg(0)
+ .addImm(StSize));
+ }
+}
+
MachineBasicBlock *
ARMTargetLowering::EmitStructByval(MachineInstr *MI,
MachineBasicBlock *BB) const {
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
unsigned UnitSize = 0;
- unsigned UnitLdOpc = 0;
- unsigned UnitStOpc = 0;
const TargetRegisterClass *TRC = 0;
const TargetRegisterClass *VecTRC = 0;
bool IsNeon = UnitSize >= 8;
TRC = (IsThumb1 || IsThumb2) ? (const TargetRegisterClass *)&ARM::tGPRRegClass
: (const TargetRegisterClass *)&ARM::GPRRegClass;
- if (IsNeon) {
- UnitLdOpc = UnitSize == 16 ? ARM::VLD1q32wb_fixed
- : UnitSize == 8 ? ARM::VLD1d32wb_fixed : 0;
- UnitStOpc = UnitSize == 16 ? ARM::VST1q32wb_fixed
- : UnitSize == 8 ? ARM::VST1d32wb_fixed : 0;
+ if (IsNeon)
VecTRC = UnitSize == 16
? (const TargetRegisterClass *)&ARM::DPairRegClass
: UnitSize == 8
? (const TargetRegisterClass *)&ARM::DPRRegClass
: 0;
- } else if (IsThumb1) {
- UnitLdOpc = UnitSize == 4 ? ARM::tLDRi
- : UnitSize == 2 ? ARM::tLDRHi
- : UnitSize == 1 ? ARM::tLDRBi : 0;
- UnitStOpc = UnitSize == 4 ? ARM::tSTRi
- : UnitSize == 2 ? ARM::tSTRHi
- : UnitSize == 1 ? ARM::tSTRBi : 0;
- } else if (IsThumb2) {
- UnitLdOpc = UnitSize == 4
- ? ARM::t2LDR_POST
- : UnitSize == 2 ? ARM::t2LDRH_POST
- : UnitSize == 1 ? ARM::t2LDRB_POST : 0;
- UnitStOpc = UnitSize == 4
- ? ARM::t2STR_POST
- : UnitSize == 2 ? ARM::t2STRH_POST
- : UnitSize == 1 ? ARM::t2STRB_POST : 0;
- } else {
- UnitLdOpc = UnitSize == 4
- ? ARM::LDR_POST_IMM
- : UnitSize == 2 ? ARM::LDRH_POST
- : UnitSize == 1 ? ARM::LDRB_POST_IMM : 0;
- UnitStOpc = UnitSize == 4
- ? ARM::STR_POST_IMM
- : UnitSize == 2 ? ARM::STRH_POST
- : UnitSize == 1 ? ARM::STRB_POST_IMM : 0;
- }
- assert(UnitLdOpc != 0 && UnitStOpc != 0 && "Should have unit opcodes");
unsigned BytesLeft = SizeVal % UnitSize;
unsigned LoopSize = SizeVal - BytesLeft;
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
- if (IsNeon) {
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcOut, RegState::Define).addReg(srcIn)
- .addImm(0));
-
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitStOpc), destOut)
- .addReg(destIn).addImm(0).addReg(scratch));
- } else if (IsThumb1) {
- // load + update srcIn
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcIn).addImm(0));
- MachineInstrBuilder MIB =
- BuildMI(*BB, MI, dl, TII->get(ARM::tADDi8), srcOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(srcIn).addImm(UnitSize);
- AddDefaultPred(MIB);
-
- // store + update destIn
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitStOpc)).addReg(scratch)
- .addReg(destIn).addImm(0));
- MIB = BuildMI(*BB, MI, dl, TII->get(ARM::tADDi8), destOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(destIn).addImm(UnitSize);
- AddDefaultPred(MIB);
- } else if (IsThumb2) {
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcOut, RegState::Define).addReg(srcIn)
- .addImm(UnitSize));
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitStOpc), destOut)
- .addReg(scratch).addReg(destIn).addImm(UnitSize));
- } else { // arm
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcOut, RegState::Define).addReg(srcIn)
- .addReg(0).addImm(UnitSize));
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(UnitStOpc), destOut)
- .addReg(scratch).addReg(destIn).addReg(0)
- .addImm(UnitSize));
- }
+ emitPostLd(BB, MI, TII, dl, UnitSize, scratch, srcIn, srcOut,
+ IsThumb1, IsThumb2);
+ emitPostSt(BB, MI, TII, dl, UnitSize, scratch, destIn, destOut,
+ IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(TRC);
- if (IsThumb1) {
- // load into scratch
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::tLDRBi), scratch)
- .addReg(srcIn).addImm(0));
-
- // update base pointer
- MachineInstrBuilder MIB =
- BuildMI(*BB, MI, dl, TII->get(ARM::tADDi8), srcOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(srcIn).addImm(1);
- AddDefaultPred(MIB);
-
- // store
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::tSTRBi))
- .addReg(scratch).addReg(destIn).addImm(0));
-
- // update base pointer
- MIB = BuildMI(*BB, MI, dl, TII->get(ARM::tADDi8), destOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(destIn).addImm(1);
- AddDefaultPred(MIB);
- } else if (IsThumb2) {
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::t2LDRB_POST), scratch)
- .addReg(srcOut, RegState::Define).addReg(srcIn)
- .addImm(1));
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::t2STRB_POST), destOut)
- .addReg(scratch).addReg(destIn).addImm(1));
- } else { // arm
- AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::LDRB_POST_IMM),
- scratch).addReg(srcOut, RegState::Define)
- .addReg(srcIn).addReg(0).addImm(1));
- AddDefaultPred(
- BuildMI(*BB, MI, dl, TII->get(ARM::STRB_POST_IMM), destOut)
- .addReg(scratch).addReg(destIn).addReg(0).addImm(1));
- }
+ emitPostLd(BB, MI, TII, dl, 1, scratch, srcIn, srcOut,
+ IsThumb1, IsThumb2);
+ emitPostSt(BB, MI, TII, dl, 1, scratch, destIn, destOut,
+ IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
// [scratch, srcLoop] = LDR_POST(srcPhi, UnitSize)
// [destLoop] = STR_POST(scratch, destPhi, UnitSiz)
unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
- if (IsNeon) {
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcLoop, RegState::Define).addReg(srcPhi)
- .addImm(0));
-
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitStOpc), destLoop)
- .addReg(destPhi).addImm(0).addReg(scratch));
- } else if (IsThumb1) {
- // load + update srcIn
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcPhi).addImm(0));
- MachineInstrBuilder MIB =
- BuildMI(*BB, BB->end(), dl, TII->get(ARM::tADDi8), srcLoop);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(srcPhi).addImm(UnitSize);
- AddDefaultPred(MIB);
-
- // store + update destIn
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitStOpc))
- .addReg(scratch).addReg(destPhi).addImm(0));
- MIB = BuildMI(*BB, BB->end(), dl, TII->get(ARM::tADDi8), destLoop);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(destPhi).addImm(UnitSize);
- AddDefaultPred(MIB);
- } else if (IsThumb2) {
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcLoop, RegState::Define).addReg(srcPhi)
- .addImm(UnitSize));
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitStOpc), destLoop)
- .addReg(scratch).addReg(destPhi).addImm(UnitSize));
- } else { // arm
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitLdOpc), scratch)
- .addReg(srcLoop, RegState::Define).addReg(srcPhi)
- .addReg(0).addImm(UnitSize));
- AddDefaultPred(BuildMI(*BB, BB->end(), dl, TII->get(UnitStOpc), destLoop)
- .addReg(scratch).addReg(destPhi).addReg(0)
- .addImm(UnitSize));
- }
+ emitPostLd(BB, BB->end(), TII, dl, UnitSize, scratch, srcPhi, srcLoop,
+ IsThumb1, IsThumb2);
+ emitPostSt(BB, BB->end(), TII, dl, UnitSize, scratch, destPhi, destLoop,
+ IsThumb1, IsThumb2);
// Decrement loop variable by UnitSize.
if (IsThumb1) {
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(TRC);
- if (IsThumb1) {
- // load into scratch
- AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(ARM::tLDRBi),
- scratch).addReg(srcIn).addImm(0));
-
- // update base pointer
- MachineInstrBuilder MIB =
- BuildMI(*BB, StartOfExit, dl, TII->get(ARM::tADDi8), srcOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(srcIn).addImm(1);
- AddDefaultPred(MIB);
-
- // store
- AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(ARM::tSTRBi))
- .addReg(scratch).addReg(destIn).addImm(0));
-
- // update base pointer
- MIB = BuildMI(*BB, StartOfExit, dl, TII->get(ARM::tADDi8), destOut);
- MIB = AddDefaultT1CC(MIB);
- MIB.addReg(destIn).addImm(1);
- AddDefaultPred(MIB);
- } else if (IsThumb2) {
- AddDefaultPred(
- BuildMI(*BB, StartOfExit, dl, TII->get(ARM::t2LDRB_POST), scratch)
- .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(1));
- AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(ARM::t2STRB_POST),
- destOut).addReg(scratch).addReg(destIn).addImm(1));
- } else { // arm
- AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(ARM::LDRB_POST_IMM),
- scratch).addReg(srcOut, RegState::Define)
- .addReg(srcIn).addReg(0).addImm(1));
- AddDefaultPred(
- BuildMI(*BB, StartOfExit, dl, TII->get(ARM::STRB_POST_IMM), destOut)
- .addReg(scratch).addReg(destIn).addReg(0).addImm(1));
- }
+ emitPostLd(BB, StartOfExit, TII, dl, 1, scratch, srcIn, srcOut,
+ IsThumb1, IsThumb2);
+ emitPostSt(BB, StartOfExit, TII, dl, 1, scratch, destIn, destOut,
+ IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
// loads from a constant pool.
case Intrinsic::arm_neon_vshifts:
case Intrinsic::arm_neon_vshiftu:
- case Intrinsic::arm_neon_vshiftls:
- case Intrinsic::arm_neon_vshiftlu:
- case Intrinsic::arm_neon_vshiftn:
case Intrinsic::arm_neon_vrshifts:
case Intrinsic::arm_neon_vrshiftu:
case Intrinsic::arm_neon_vrshiftn:
}
return SDValue();
- case Intrinsic::arm_neon_vshiftls:
- case Intrinsic::arm_neon_vshiftlu:
- if (isVShiftLImm(N->getOperand(2), VT, true, Cnt))
- break;
- llvm_unreachable("invalid shift count for vshll intrinsic");
-
case Intrinsic::arm_neon_vrshifts:
case Intrinsic::arm_neon_vrshiftu:
if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt))
break;
llvm_unreachable("invalid shift count for vqshlu intrinsic");
- case Intrinsic::arm_neon_vshiftn:
case Intrinsic::arm_neon_vrshiftn:
case Intrinsic::arm_neon_vqshiftns:
case Intrinsic::arm_neon_vqshiftnu:
case Intrinsic::arm_neon_vshiftu:
// Opcode already set above.
break;
- case Intrinsic::arm_neon_vshiftls:
- case Intrinsic::arm_neon_vshiftlu:
- if (Cnt == VT.getVectorElementType().getSizeInBits())
- VShiftOpc = ARMISD::VSHLLi;
- else
- VShiftOpc = (IntNo == Intrinsic::arm_neon_vshiftls ?
- ARMISD::VSHLLs : ARMISD::VSHLLu);
- break;
- case Intrinsic::arm_neon_vshiftn:
- VShiftOpc = ARMISD::VSHRN; break;
case Intrinsic::arm_neon_vrshifts:
VShiftOpc = ARMISD::VRSHRs; break;
case Intrinsic::arm_neon_vrshiftu:
return (VT == MVT::f32) && (Opc == ISD::LOAD || Opc == ISD::STORE);
}
-bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const {
+bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT, unsigned,
+ bool *Fast) const {
// The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus
bool AllowsUnaligned = Subtarget->allowsUnalignedMem();
case MVT::v2f64: {
// For any little-endian targets with neon, we can support unaligned ld/st
// of D and Q (e.g. {D0,D1}) registers by using vld1.i8/vst1.i8.
- // A big-endian target may also explictly support unaligned accesses
+ // A big-endian target may also explicitly support unaligned accesses
if (Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian())) {
if (Fast)
*Fast = true;
bool Fast;
if (Size >= 16 &&
(memOpAlign(SrcAlign, DstAlign, 16) ||
- (allowsUnalignedMemoryAccesses(MVT::v2f64, &Fast) && Fast))) {
+ (allowsUnalignedMemoryAccesses(MVT::v2f64, 0, &Fast) && Fast))) {
return MVT::v2f64;
} else if (Size >= 8 &&
(memOpAlign(SrcAlign, DstAlign, 8) ||
- (allowsUnalignedMemoryAccesses(MVT::f64, &Fast) && Fast))) {
+ (allowsUnalignedMemoryAccesses(MVT::f64, 0, &Fast) && Fast))) {
return MVT::f64;
}
}
case 'r':
return RCPair(0U, &ARM::GPRRegClass);
case 'w':
+ if (VT == MVT::Other)
+ break;
if (VT == MVT::f32)
return RCPair(0U, &ARM::SPRRegClass);
if (VT.getSizeInBits() == 64)
return RCPair(0U, &ARM::QPRRegClass);
break;
case 'x':
+ if (VT == MVT::Other)
+ break;
if (VT == MVT::f32)
return RCPair(0U, &ARM::SPR_8RegClass);
if (VT.getSizeInBits() == 64)
return false;
}
+
+/// \brief Returns true if it is beneficial to convert a load of a constant
+/// to just the constant itself.
+bool ARMTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
+ Type *Ty) const {
+ assert(Ty->isIntegerTy());
+
+ unsigned Bits = Ty->getPrimitiveSizeInBits();
+ if (Bits == 0 || Bits > 32)
+ return false;
+ return true;
+}
projects / oota-llvm.git / blobdiff