AArch64Subtarget.cpp

Go to the documentation of this file.

//===-- AArch64Subtarget.cpp - AArch64 Subtarget Information ----*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements the AArch64 specific subclass of TargetSubtarget.
//
//===----------------------------------------------------------------------===//
 
#include "AArch64Subtarget.h"
 
#include "AArch64.h"
#include "AArch64InstrInfo.h"
#include "AArch64PBQPRegAlloc.h"
#include "AArch64TargetMachine.h"
#include "GISel/AArch64CallLowering.h"
#include "GISel/AArch64LegalizerInfo.h"
#include "GISel/AArch64RegisterBankInfo.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
#include "llvm/CodeGen/GlobalISel/InstructionSelect.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineScheduler.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/Option/LibraryOptions.h"
#include "llvm/Support/SipHash.h"
#include "llvm/TargetParser/AArch64TargetParser.h"
 
using namespace llvm;
 
#define DEBUG_TYPE "aarch64-subtarget"
 
#define GET_SUBTARGETINFO_CTOR
#define GET_SUBTARGETINFO_TARGET_DESC
#include "AArch64GenSubtargetInfo.inc"
 
#define OPTIONS_STRUCT_DEFS
#include "AArch64Options.inc"
 
// Reserve a list of X# registers, so they are unavailable for register
// allocator, but can still be used as ABI requests, such as passing arguments
// to function call.
static cl::list<std::string>
ReservedRegsForRA("reserve-regs-for-regalloc", cl::desc("Reserve physical "
                  "registers, so they can't be used by register allocator. "
                  "Should only be used for testing register allocator."),
                  cl::CommaSeparated, cl::Hidden);
 
unsigned AArch64Subtarget::getVectorInsertExtractBaseCost() const {
  return CLOpts.insert_extract_base_cost.value_or(VectorInsertExtractBaseCost);
}
 
AArch64Subtarget &AArch64Subtarget::initializeSubtargetDependencies(
    StringRef FS, StringRef CPUString, StringRef TuneCPUString,
    bool HasMinSize) {
  // Determine default and user-specified characteristics
 
  if (CPUString.empty())
    CPUString = "generic";
 
  if (TuneCPUString.empty())
    TuneCPUString = CPUString;
 
  ParseSubtargetFeatures(CPUString, TuneCPUString, FS);
  initializeProperties(HasMinSize);
 
  return *this;
}
 
void AArch64Subtarget::initializeProperties(bool HasMinSize) {
  // Initialize CPU specific properties. We should add a tablegen feature for
  // this in the future so we can specify it together with the subtarget
  // features.
  switch (ARMProcFamily) {
  case Generic:
    // Using TuneCPU=generic we avoid ldapur instructions to line up with the
    // cpus that use the AvoidLDAPUR feature. We don't want this to be on
    // forever, so it is enabled between armv8.4 and armv8.7/armv9.2.
    if (hasV8_4aOps() && !hasV8_8aOps())
      AvoidLDAPUR = true;
    break;
  case Carmel:
    break;
  case CortexA35:
  case CortexA53:
  case CortexA55:
  case CortexR82:
  case CortexR82AE:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(16);
    MaxBytesForLoopAlignment = 8;
    break;
  case CortexA57:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(16);
    MaxBytesForLoopAlignment = 8;
    break;
  case CortexA65:
    PrefFunctionAlignment = Align(8);
    break;
  case CortexA72:
  case CortexA73:
  case CortexA75:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(16);
    MaxBytesForLoopAlignment = 8;
    break;
  case CortexA76:
  case CortexA77:
  case CortexA78:
  case CortexA78AE:
  case CortexA78C:
  case CortexX1:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    break;
  case CortexA320:
  case CortexA510:
  case CortexA520:
  case C1Nano:
    PrefFunctionAlignment = Align(16);
    VScaleForTuning = 1;
    PrefLoopAlignment = Align(16);
    MaxBytesForLoopAlignment = 8;
    break;
  case CortexA710:
  case CortexA715:
  case CortexA720:
  case CortexA725:
  case C1Pro:
  case CortexX2:
  case CortexX3:
  case CortexX4:
  case CortexX925:
  case C1Premium:
  case C1Ultra:
  case C2Ultra:
    PrefFunctionAlignment = Align(16);
    VScaleForTuning = 1;
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    break;
  case A64FX:
    CacheLineSize = 256;
    PrefFunctionAlignment = Align(8);
    PrefLoopAlignment = Align(4);
    PrefetchDistance = 128;
    MinPrefetchStride = 1024;
    MaxPrefetchIterationsAhead = 4;
    VScaleForTuning = 4;
    break;
  case MONAKA:
    VScaleForTuning = 2;
    break;
  case AppleA7:
  case AppleA10:
  case AppleA11:
  case AppleA12:
  case AppleA13:
  case AppleA14:
  case AppleA15:
  case AppleA16:
  case AppleA17:
  case AppleM4:
  case AppleM5:
  case AppleM6:
    PrefetchDistance = 280;
    MinPrefetchStride = 2048;
    MaxPrefetchIterationsAhead = 3;
    break;
  case ExynosM3:
    MaxJumpTableSize = 20;
    PrefFunctionAlignment = Align(32);
    PrefLoopAlignment = Align(16);
    break;
  case Falkor:
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    CacheLineSize = 128;
    PrefetchDistance = 820;
    MinPrefetchStride = 2048;
    MaxPrefetchIterationsAhead = 8;
    break;
  case Kryo:
    VectorInsertExtractBaseCost = 2;
    CacheLineSize = 128;
    PrefetchDistance = 740;
    MinPrefetchStride = 1024;
    MaxPrefetchIterationsAhead = 11;
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    break;
  case NeoverseE1:
    PrefFunctionAlignment = Align(8);
    break;
  case NeoverseN1:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    break;
  case NeoverseV2:
  case NeoverseV3:
    EpilogueVectorizationMinVF = 8;
    ScatterOverhead = 13;
    [[fallthrough]];
  case NeoverseN2:
  case NeoverseN3:
  case NeoverseV3AE:
    PrefFunctionAlignment = Align(32);
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    VScaleForTuning = 1;
    break;
  case NeoverseV1:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    VScaleForTuning = 2;
    DefaultSVETFOpts = TailFoldingOpts::Simple;
    break;
  case Neoverse512TVB:
    PrefFunctionAlignment = Align(16);
    VScaleForTuning = 1;
    break;
  case Saphira:
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    break;
  case ThunderX2T99:
    PrefFunctionAlignment = Align(8);
    PrefLoopAlignment = Align(4);
    PrefetchDistance = 128;
    MinPrefetchStride = 1024;
    MaxPrefetchIterationsAhead = 4;
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    break;
  case ThunderX:
  case ThunderXT88:
  case ThunderXT81:
  case ThunderXT83:
    CacheLineSize = 128;
    PrefFunctionAlignment = Align(8);
    PrefLoopAlignment = Align(4);
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    break;
  case TSV110:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(4);
    break;
  case HIP12:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(4);
    VScaleForTuning = 2;
    DefaultSVETFOpts = TailFoldingOpts::Simple;
    break;
  case ThunderX3T110:
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(4);
    PrefetchDistance = 128;
    MinPrefetchStride = 1024;
    MaxPrefetchIterationsAhead = 4;
    // FIXME: remove this to enable 64-bit SLP if performance looks good.
    MinVectorRegisterBitWidth = 128;
    break;
  case Ampere1:
  case Ampere1A:
  case Ampere1B:
  case Ampere1C:
    PrefFunctionAlignment = Align(64);
    PrefLoopAlignment = Align(64);
    break;
  case Oryon:
    PrefFunctionAlignment = Align(16);
    PrefetchDistance = 128;
    MinPrefetchStride = 1024;
    break;
  case Olympus:
    EpilogueVectorizationMinVF = 8;
    ScatterOverhead = 13;
    PrefFunctionAlignment = Align(16);
    PrefLoopAlignment = Align(32);
    MaxBytesForLoopAlignment = 16;
    VScaleForTuning = 1;
    break;
  }
 
  if (CLOpts.min_jump_table_entries || !HasMinSize)
    MinimumJumpTableEntries = CLOpts.min_jump_table_entries.value_or(10);
  if (CLOpts.sve_vscale_for_tuning)
    VScaleForTuning = *CLOpts.sve_vscale_for_tuning;
}
 
AArch64Subtarget::AArch64Subtarget(const Triple &TT, StringRef CPU,
                                   StringRef TuneCPU, StringRef FS,
                                   const TargetMachine &TM, bool LittleEndian,
                                   unsigned MinSVEVectorSizeInBitsOverride,
                                   unsigned MaxSVEVectorSizeInBitsOverride,
                                   bool IsStreaming, bool IsStreamingCompatible,
                                   bool HasMinSize,
                                   bool EnableSRLTSubregToRegMitigation)
    : AArch64GenSubtargetInfo(TT, CPU, TuneCPU, FS),
      CLOpts(static_cast<const AArch64TargetMachine &>(TM).getCLOpts()),
      ReserveXRegister(AArch64::GPR64commonRegClass.getNumRegs()),
      ReserveXRegisterForRA(AArch64::GPR64commonRegClass.getNumRegs()),
      CustomCallSavedXRegs(AArch64::GPR64commonRegClass.getNumRegs()),
      IsLittle(LittleEndian), IsStreaming(IsStreaming),
      IsStreamingCompatible(IsStreamingCompatible),
      MinSVEVectorSizeInBits(MinSVEVectorSizeInBitsOverride),
      MaxSVEVectorSizeInBits(MaxSVEVectorSizeInBitsOverride),
      EnableSRLTSubregToRegMitigation(EnableSRLTSubregToRegMitigation),
      // To benefit from SME2's strided-register multi-vector load/store
      // instructions we'll need to enable subreg liveness. Our longer
      // term aim is to make this the default, regardless of streaming
      // mode, but there are still some outstanding issues, see:
      //  https://github.com/llvm/llvm-project/pull/174188
      // and:
      //  https://github.com/llvm/llvm-project/pull/168353
      EnableSubregLiveness(IsStreaming ||
                           CLOpts.enable_subreg_liveness_tracking),
      TargetTriple(TT),
      InstrInfo(initializeSubtargetDependencies(FS, CPU, TuneCPU, HasMinSize)),
      TLInfo(TM, *this) {
  if (AArch64::isX18ReservedByDefault(TT))
    ReserveXRegister.set(18);
 
  CallLoweringInfo.reset(new AArch64CallLowering(*getTargetLowering()));
  InlineAsmLoweringInfo.reset(new InlineAsmLowering(getTargetLowering()));
  Legalizer.reset(new AArch64LegalizerInfo(*this));
 
  auto *RBI = new AArch64RegisterBankInfo(*getRegisterInfo());
 
  // FIXME: At this point, we can't rely on Subtarget having RBI.
  // It's awkward to mix passing RBI and the Subtarget; should we pass
  // TII/TRI as well?
  InstSelector.reset(createAArch64InstructionSelector(
      *static_cast<const AArch64TargetMachine *>(&TM), *this, *RBI));
 
  RegBankInfo.reset(RBI);
 
  auto TRI = getRegisterInfo();
  StringSet<> ReservedRegNames(llvm::from_range, ReservedRegsForRA);
  for (unsigned i = 0; i < 29; ++i) {
    if (ReservedRegNames.count(TRI->getName(AArch64::X0 + i)))
      ReserveXRegisterForRA.set(i);
  }
  // X30 is named LR, so we can't use TRI->getName to check X30.
  if (ReservedRegNames.count("X30") || ReservedRegNames.count("LR"))
    ReserveXRegisterForRA.set(30);
  // X29 is named FP, so we can't use TRI->getName to check X29.
  if (ReservedRegNames.count("X29") || ReservedRegNames.count("FP"))
    ReserveXRegisterForRA.set(29);
}
 
const CallLowering *AArch64Subtarget::getCallLowering() const {
  return CallLoweringInfo.get();
}
 
const InlineAsmLowering *AArch64Subtarget::getInlineAsmLowering() const {
  return InlineAsmLoweringInfo.get();
}
 
InstructionSelector *AArch64Subtarget::getInstructionSelector() const {
  return InstSelector.get();
}
 
const LegalizerInfo *AArch64Subtarget::getLegalizerInfo() const {
  return Legalizer.get();
}
 
const RegisterBankInfo *AArch64Subtarget::getRegBankInfo() const {
  return RegBankInfo.get();
}
 
/// Find the target operand flags that describe how a global value should be
/// referenced for the current subtarget.
unsigned
AArch64Subtarget::ClassifyGlobalReference(const GlobalValue *GV,
                                          const TargetMachine &TM) const {
  // MachO large model always goes via a GOT, simply to get a single 8-byte
  // absolute relocation on all global addresses.
  if (TM.getCodeModel() == CodeModel::Large && isTargetMachO())
    return AArch64II::MO_GOT;
 
  // All globals dynamically protected by MTE must have their address tags
  // synthesized. This is done by having the loader stash the tag in the GOT
  // entry. Force all tagged globals (even ones with internal linkage) through
  // the GOT.
  if (GV->isTagged())
    return AArch64II::MO_GOT;
 
  if (!TM.shouldAssumeDSOLocal(GV)) {
    if (GV->hasDLLImportStorageClass()) {
      return AArch64II::MO_GOT | AArch64II::MO_DLLIMPORT;
    }
    if (getTargetTriple().isOSWindowsOrUEFI())
      return AArch64II::MO_GOT | AArch64II::MO_COFFSTUB;
    return AArch64II::MO_GOT;
  }
 
  // The small code model's direct accesses use ADRP, which cannot
  // necessarily produce the value 0 (if the code is above 4GB).
  // Same for the tiny code model, where we have a pc relative LDR.
  if ((useSmallAddressing() || TM.getCodeModel() == CodeModel::Tiny) &&
      GV->hasExternalWeakLinkage())
    return AArch64II::MO_GOT;
 
  // References to tagged globals are marked with MO_NC | MO_TAGGED to indicate
  // that their nominal addresses are tagged and outside of the code model. In
  // AArch64ExpandPseudo::expandMI we emit an additional instruction to set the
  // tag if necessary based on MO_TAGGED.
  if (AllowTaggedGlobals && !isa<FunctionType>(GV->getValueType()))
    return AArch64II::MO_NC | AArch64II::MO_TAGGED;
 
  return AArch64II::MO_NO_FLAG;
}
 
unsigned AArch64Subtarget::classifyGlobalFunctionReference(
    const GlobalValue *GV, const TargetMachine &TM) const {
  // MachO large model always goes via a GOT, because we don't have the
  // relocations available to do anything else..
  if (TM.getCodeModel() == CodeModel::Large && isTargetMachO() &&
      !GV->hasInternalLinkage())
    return AArch64II::MO_GOT;
 
  // NonLazyBind goes via GOT unless we know it's available locally.
  auto *F = dyn_cast<Function>(GV);
  if ((!isTargetMachO() || CLOpts.macho_enable_nonlazybind) && F &&
      F->hasFnAttribute(Attribute::NonLazyBind) && !TM.shouldAssumeDSOLocal(GV))
    return AArch64II::MO_GOT;
 
  if (getTargetTriple().isOSWindowsOrUEFI()) {
    if (isWindowsArm64EC() && GV->getValueType()->isFunctionTy()) {
      if (GV->hasDLLImportStorageClass()) {
        // On Arm64EC, if we're calling a symbol from the import table
        // directly, use MO_ARM64EC_CALLMANGLE.
        return AArch64II::MO_GOT | AArch64II::MO_DLLIMPORT |
               AArch64II::MO_ARM64EC_CALLMANGLE;
      }
      if (GV->hasExternalLinkage()) {
        // If we're calling a symbol directly, use the mangled form in the
        // call instruction.
        return AArch64II::MO_ARM64EC_CALLMANGLE;
      }
    }
 
    // Use ClassifyGlobalReference for setting MO_DLLIMPORT/MO_COFFSTUB.
    return ClassifyGlobalReference(GV, TM);
  }
 
  return AArch64II::MO_NO_FLAG;
}
 
void AArch64Subtarget::overrideSchedPolicy(MachineSchedPolicy &Policy,
                                           const SchedRegion &Region) const {
  // LNT run (at least on Cyclone) showed reasonably significant gains for
  // bi-directional scheduling. 253.perlbmk.
  Policy.OnlyTopDown = false;
  Policy.OnlyBottomUp = false;
  // Enabling or Disabling the latency heuristic is a close call: It seems to
  // help nearly no benchmark on out-of-order architectures, on the other hand
  // it regresses register pressure on a few benchmarking.
  Policy.DisableLatencyHeuristic = DisableLatencySchedHeuristic;
}
 
void AArch64Subtarget::adjustSchedDependency(
    SUnit *Def, int DefOpIdx, SUnit *Use, int UseOpIdx, SDep &Dep,
    const TargetSchedModel *SchedModel) const {
  if (!SchedModel || Dep.getKind() != SDep::Kind::Data || !Dep.getReg() ||
      !Def->isInstr() || !Use->isInstr() ||
      (Def->getInstr()->getOpcode() != TargetOpcode::BUNDLE &&
       Use->getInstr()->getOpcode() != TargetOpcode::BUNDLE))
    return;
 
  // If the Def is a BUNDLE, find the last instruction in the bundle that defs
  // the register.
  const MachineInstr *DefMI = Def->getInstr();
  if (DefMI->getOpcode() == TargetOpcode::BUNDLE) {
    Register Reg = DefMI->getOperand(DefOpIdx).getReg();
    for (const auto &Op : const_mi_bundle_ops(*DefMI)) {
      if (Op.isReg() && Op.isDef() && Op.getReg() == Reg) {
        DefMI = Op.getParent();
        DefOpIdx = Op.getOperandNo();
      }
    }
  }
 
  // If the Use is a BUNDLE, find the first instruction that uses the Reg.
  const MachineInstr *UseMI = Use->getInstr();
  if (UseMI->getOpcode() == TargetOpcode::BUNDLE) {
    Register Reg = UseMI->getOperand(UseOpIdx).getReg();
    for (const auto &Op : const_mi_bundle_ops(*UseMI)) {
      if (Op.isReg() && Op.isUse() && Op.getReg() == Reg) {
        UseMI = Op.getParent();
        UseOpIdx = Op.getOperandNo();
        break;
      }
    }
  }
 
  Dep.setLatency(
      SchedModel->computeOperandLatency(DefMI, DefOpIdx, UseMI, UseOpIdx));
}
 
bool AArch64Subtarget::enableEarlyIfConversion() const {
  return CLOpts.early_ifcvt;
}
 
bool AArch64Subtarget::supportsAddressTopByteIgnored() const {
  if (!CLOpts.use_tbi)
    return false;
 
  if (TargetTriple.isDriverKit())
    return true;
  if (TargetTriple.isiOS()) {
    return TargetTriple.getiOSVersion() >= VersionTuple(8);
  }
 
  return false;
}
 
std::unique_ptr<PBQPRAConstraint>
AArch64Subtarget::getCustomPBQPConstraints() const {
  return balanceFPOps() ? std::make_unique<A57ChainingConstraint>() : nullptr;
}
 
void AArch64Subtarget::mirFileLoaded(MachineFunction &MF) const {
  // We usually compute max call frame size after ISel. Do the computation now
  // if the .mir file didn't specify it. Note that this will probably give you
  // bogus values after PEI has eliminated the callframe setup/destroy pseudo
  // instructions, specify explicitly if you need it to be correct.
  MachineFrameInfo &MFI = MF.getFrameInfo();
  if (!MFI.isMaxCallFrameSizeComputed())
    MFI.computeMaxCallFrameSize(MF);
}
 
bool AArch64Subtarget::useAA() const { return CLOpts.use_aa; }
 
bool AArch64Subtarget::useScalarIncVL() const {
  // If SVE2 or SME is present (we are not SVE-1 only) and
  // -sve-use-scalar-inc-vl is not otherwise set, enable it by default.
  return valueOr(CLOpts.sve_use_scalar_inc_vl, hasSVE2() || hasSME());
}
 
// If return address signing is enabled, tail calls are emitted as follows:
//
// ```
//   <authenticate LR>
//   <check LR>
//   TCRETURN          ; the callee may sign and spill the LR in its prologue
// ```
//
// LR may require explicit checking because if FEAT_FPAC is not implemented
// and LR was tampered with, then `<authenticate LR>` will not generate an
// exception on its own. Later, if the callee spills the signed LR value and
// neither FEAT_PAuth2 nor FEAT_EPAC are implemented, the valid PAC replaces
// the higher bits of LR thus hiding the authentication failure.
AArch64PAuth::AuthCheckMethod AArch64Subtarget::getAuthenticatedLRCheckMethod(
    const MachineFunction &MF) const {
  // TODO: Check subtarget for the scheme. Present variant is a default for
  // pauthtest ABI.
  if (MF.getFunction().hasFnAttribute("ptrauth-returns") &&
      MF.getFunction().hasFnAttribute("ptrauth-auth-traps"))
    return AArch64PAuth::AuthCheckMethod::HighBitsNoTBI;
  // At now, use None by default because checks may introduce an unexpected
  // performance regression or incompatibility with execute-only mappings.
  return CLOpts.authenticated_lr_check_method.value_or(
      AArch64PAuth::AuthCheckMethod::None);
}
 
std::optional<uint16_t>
AArch64Subtarget::getPtrAuthBlockAddressDiscriminatorIfEnabled(
    const Function &ParentFn) const {
  if (!ParentFn.hasFnAttribute("ptrauth-indirect-gotos"))
    return std::nullopt;
  // We currently have one simple mechanism for all targets.
  // This isn't ABI, so we can always do better in the future.
  return getPointerAuthStableSipHash(
      (Twine(ParentFn.getName()) + " blockaddress").str());
}
 
bool AArch64Subtarget::isX16X17Safer() const {
  // The Darwin kernel implements special protections for x16 and x17 so we
  // should prefer to use those registers on that platform.
  return isTargetDarwin();
}
 
bool AArch64Subtarget::enableMachinePipeliner() const {
  return getSchedModel().hasInstrSchedModel();
}
 
/// Returns a MOVK's shifter operand, or 0 otherwise.
static unsigned getMOVKShiftImm(const MachineInstr &MI) {
  unsigned Opc = MI.getOpcode();
  if (Opc != AArch64::MOVKWi && Opc != AArch64::MOVKXi)
    return 0;
  return MI.getOperand(3).getImm();
}
 
/// \p HasFirst is false when the 1st instruction is a wildcard.
static bool fusesMOVImmPairImpl(const AArch64Subtarget &ST, bool HasFirst,
                                unsigned FirstOpc, unsigned FirstShift,
                                unsigned SecondOpc, unsigned SecondShift) {
  assert(ST.hasFuseLiterals() && "the subtarget doesn't fuse move immediate");
 
  // 32 bit immediate.
  if ((!HasFirst || FirstOpc == AArch64::MOVZWi) &&
      SecondOpc == AArch64::MOVKWi && SecondShift == 16)
    return true;
 
  // Lower half of 64 bit immediate.
  if ((!HasFirst || FirstOpc == AArch64::MOVZXi) &&
      SecondOpc == AArch64::MOVKXi && SecondShift == 16)
    return true;
 
  // Upper half of 64 bit immediate.
  if ((!HasFirst || (FirstOpc == AArch64::MOVKXi && FirstShift == 32)) &&
      SecondOpc == AArch64::MOVKXi && SecondShift == 48)
    return true;
 
  return false;
}
 
bool AArch64Subtarget::fusesMOVImmPair(unsigned FirstOpc, unsigned FirstShift,
                                       unsigned SecondOpc,
                                       unsigned SecondShift) const {
  return fusesMOVImmPairImpl(*this, /*HasFirst=*/true, FirstOpc, FirstShift,
                             SecondOpc, SecondShift);
}
 
bool AArch64Subtarget::fusesMOVImmPair(const MachineInstr *FirstMI,
                                       const MachineInstr &SecondMI) const {
  return fusesMOVImmPairImpl(*this, FirstMI != nullptr,
                             FirstMI ? FirstMI->getOpcode() : 0,
                             FirstMI ? getMOVKShiftImm(*FirstMI) : 0,
                             SecondMI.getOpcode(), getMOVKShiftImm(SecondMI));
}