disas/libvixl/vixl/a64/assembler-a64.h

// Copyright 2015, ARM Limited
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
//   * Redistributions of source code must retain the above copyright notice,
//     this list of conditions and the following disclaimer.
//   * Redistributions in binary form must reproduce the above copyright notice,
//     this list of conditions and the following disclaimer in the documentation
//     and/or other materials provided with the distribution.
//   * Neither the name of ARM Limited nor the names of its contributors may be
//     used to endorse or promote products derived from this software without
//     specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS CONTRIBUTORS "AS IS" AND
// ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#ifndef VIXL_A64_ASSEMBLER_A64_H_
#define VIXL_A64_ASSEMBLER_A64_H_


#include "vixl/globals.h"
#include "vixl/invalset.h"
#include "vixl/utils.h"
#include "vixl/code-buffer.h"
#include "vixl/a64/instructions-a64.h"

namespace vixl {

typedef uint64_t RegList;
static const int kRegListSizeInBits = sizeof(RegList) * 8;


// Registers.

// Some CPURegister methods can return Register or VRegister types, so we need
// to declare them in advance.
class Register;
class VRegister;

class CPURegister {
 public:
  enum RegisterType {
    // The kInvalid value is used to detect uninitialized static instances,
    // which are always zero-initialized before any constructors are called.
    kInvalid = 0,
    kRegister,
    kVRegister,
    kFPRegister = kVRegister,
    kNoRegister
  };

  CPURegister() : code_(0), size_(0), type_(kNoRegister) {
    VIXL_ASSERT(!IsValid());
    VIXL_ASSERT(IsNone());
  }

  CPURegister(unsigned code, unsigned size, RegisterType type)
      : code_(code), size_(size), type_(type) {
    VIXL_ASSERT(IsValidOrNone());
  }

  unsigned code() const {
    VIXL_ASSERT(IsValid());
    return code_;
  }

  RegisterType type() const {
    VIXL_ASSERT(IsValidOrNone());
    return type_;
  }

  RegList Bit() const {
    VIXL_ASSERT(code_ < (sizeof(RegList) * 8));
    return IsValid() ? (static_cast<RegList>(1) << code_) : 0;
  }

  unsigned size() const {
    VIXL_ASSERT(IsValid());
    return size_;
  }

  int SizeInBytes() const {
    VIXL_ASSERT(IsValid());
    VIXL_ASSERT(size() % 8 == 0);
    return size_ / 8;
  }

  int SizeInBits() const {
    VIXL_ASSERT(IsValid());
    return size_;
  }

  bool Is8Bits() const {
    VIXL_ASSERT(IsValid());
    return size_ == 8;
  }

  bool Is16Bits() const {
    VIXL_ASSERT(IsValid());
    return size_ == 16;
  }

  bool Is32Bits() const {
    VIXL_ASSERT(IsValid());
    return size_ == 32;
  }

  bool Is64Bits() const {
    VIXL_ASSERT(IsValid());
    return size_ == 64;
  }

  bool Is128Bits() const {
    VIXL_ASSERT(IsValid());
    return size_ == 128;
  }

  bool IsValid() const {
    if (IsValidRegister() || IsValidVRegister()) {
      VIXL_ASSERT(!IsNone());
      return true;
    } else {
      // This assert is hit when the register has not been properly initialized.
      // One cause for this can be an initialisation order fiasco. See
      // https://isocpp.org/wiki/faq/ctors#static-init-order for some details.
      VIXL_ASSERT(IsNone());
      return false;
    }
  }

  bool IsValidRegister() const {
    return IsRegister() &&
           ((size_ == kWRegSize) || (size_ == kXRegSize)) &&
           ((code_ < kNumberOfRegisters) || (code_ == kSPRegInternalCode));
  }

  bool IsValidVRegister() const {
    return IsVRegister() &&
           ((size_ == kBRegSize) || (size_ == kHRegSize) ||
            (size_ == kSRegSize) || (size_ == kDRegSize) ||
            (size_ == kQRegSize)) &&
           (code_ < kNumberOfVRegisters);
  }

  bool IsValidFPRegister() const {
    return IsFPRegister() && (code_ < kNumberOfVRegisters);
  }

  bool IsNone() const {
    // kNoRegister types should always have size 0 and code 0.
    VIXL_ASSERT((type_ != kNoRegister) || (code_ == 0));
    VIXL_ASSERT((type_ != kNoRegister) || (size_ == 0));

    return type_ == kNoRegister;
  }

  bool Aliases(const CPURegister& other) const {
    VIXL_ASSERT(IsValidOrNone() && other.IsValidOrNone());
    return (code_ == other.code_) && (type_ == other.type_);
  }

  bool Is(const CPURegister& other) const {
    VIXL_ASSERT(IsValidOrNone() && other.IsValidOrNone());
    return Aliases(other) && (size_ == other.size_);
  }

  bool IsZero() const {
    VIXL_ASSERT(IsValid());
    return IsRegister() && (code_ == kZeroRegCode);
  }

  bool IsSP() const {
    VIXL_ASSERT(IsValid());
    return IsRegister() && (code_ == kSPRegInternalCode);
  }

  bool IsRegister() const {
    return type_ == kRegister;
  }

  bool IsVRegister() const {
    return type_ == kVRegister;
  }

  bool IsFPRegister() const {
    return IsS() || IsD();
  }

  bool IsW() const { return IsValidRegister() && Is32Bits(); }
  bool IsX() const { return IsValidRegister() && Is64Bits(); }

  // These assertions ensure that the size and type of the register are as
  // described. They do not consider the number of lanes that make up a vector.
  // So, for example, Is8B() implies IsD(), and Is1D() implies IsD, but IsD()
  // does not imply Is1D() or Is8B().
  // Check the number of lanes, ie. the format of the vector, using methods such
  // as Is8B(), Is1D(), etc. in the VRegister class.
  bool IsV() const { return IsVRegister(); }
  bool IsB() const { return IsV() && Is8Bits(); }
  bool IsH() const { return IsV() && Is16Bits(); }
  bool IsS() const { return IsV() && Is32Bits(); }
  bool IsD() const { return IsV() && Is64Bits(); }
  bool IsQ() const { return IsV() && Is128Bits(); }

  const Register& W() const;
  const Register& X() const;
  const VRegister& V() const;
  const VRegister& B() const;
  const VRegister& H() const;
  const VRegister& S() const;
  const VRegister& D() const;
  const VRegister& Q() const;

  bool IsSameSizeAndType(const CPURegister& other) const {
    return (size_ == other.size_) && (type_ == other.type_);
  }

 protected:
  unsigned code_;
  unsigned size_;
  RegisterType type_;

 private:
  bool IsValidOrNone() const {
    return IsValid() || IsNone();
  }
};


class Register : public CPURegister {
 public:
  Register() : CPURegister() {}
  explicit Register(const CPURegister& other)
      : CPURegister(other.code(), other.size(), other.type()) {
    VIXL_ASSERT(IsValidRegister());
  }
  Register(unsigned code, unsigned size)
      : CPURegister(code, size, kRegister) {}

  bool IsValid() const {
    VIXL_ASSERT(IsRegister() || IsNone());
    return IsValidRegister();
  }

  static const Register& WRegFromCode(unsigned code);
  static const Register& XRegFromCode(unsigned code);

 private:
  static const Register wregisters[];
  static const Register xregisters[];
};


class VRegister : public CPURegister {
 public:
  VRegister() : CPURegister(), lanes_(1) {}
  explicit VRegister(const CPURegister& other)
      : CPURegister(other.code(), other.size(), other.type()), lanes_(1) {
    VIXL_ASSERT(IsValidVRegister());
    VIXL_ASSERT(IsPowerOf2(lanes_) && (lanes_ <= 16));
  }
  VRegister(unsigned code, unsigned size, unsigned lanes = 1)
      : CPURegister(code, size, kVRegister), lanes_(lanes) {
    VIXL_ASSERT(IsPowerOf2(lanes_) && (lanes_ <= 16));
  }
  VRegister(unsigned code, VectorFormat format)
      : CPURegister(code, RegisterSizeInBitsFromFormat(format), kVRegister),
        lanes_(IsVectorFormat(format) ? LaneCountFromFormat(format) : 1) {
    VIXL_ASSERT(IsPowerOf2(lanes_) && (lanes_ <= 16));
  }

  bool IsValid() const {
    VIXL_ASSERT(IsVRegister() || IsNone());
    return IsValidVRegister();
  }

  static const VRegister& BRegFromCode(unsigned code);
  static const VRegister& HRegFromCode(unsigned code);
  static const VRegister& SRegFromCode(unsigned code);
  static const VRegister& DRegFromCode(unsigned code);
  static const VRegister& QRegFromCode(unsigned code);
  static const VRegister& VRegFromCode(unsigned code);

  VRegister V8B() const { return VRegister(code_, kDRegSize, 8); }
  VRegister V16B() const { return VRegister(code_, kQRegSize, 16); }
  VRegister V4H() const { return VRegister(code_, kDRegSize, 4); }
  VRegister V8H() const { return VRegister(code_, kQRegSize, 8); }
  VRegister V2S() const { return VRegister(code_, kDRegSize, 2); }
  VRegister V4S() const { return VRegister(code_, kQRegSize, 4); }
  VRegister V2D() const { return VRegister(code_, kQRegSize, 2); }
  VRegister V1D() const { return VRegister(code_, kDRegSize, 1); }

  bool Is8B() const { return (Is64Bits() && (lanes_ == 8)); }
  bool Is16B() const { return (Is128Bits() && (lanes_ == 16)); }
  bool Is4H() const { return (Is64Bits() && (lanes_ == 4)); }
  bool Is8H() const { return (Is128Bits() && (lanes_ == 8)); }
  bool Is2S() const { return (Is64Bits() && (lanes_ == 2)); }
  bool Is4S() const { return (Is128Bits() && (lanes_ == 4)); }
  bool Is1D() const { return (Is64Bits() && (lanes_ == 1)); }
  bool Is2D() const { return (Is128Bits() && (lanes_ == 2)); }

  // For consistency, we assert the number of lanes of these scalar registers,
  // even though there are no vectors of equivalent total size with which they
  // could alias.
  bool Is1B() const {
    VIXL_ASSERT(!(Is8Bits() && IsVector()));
    return Is8Bits();
  }
  bool Is1H() const {
    VIXL_ASSERT(!(Is16Bits() && IsVector()));
    return Is16Bits();
  }
  bool Is1S() const {
    VIXL_ASSERT(!(Is32Bits() && IsVector()));
    return Is32Bits();
  }

  bool IsLaneSizeB() const { return LaneSizeInBits() == kBRegSize; }
  bool IsLaneSizeH() const { return LaneSizeInBits() == kHRegSize; }
  bool IsLaneSizeS() const { return LaneSizeInBits() == kSRegSize; }
  bool IsLaneSizeD() const { return LaneSizeInBits() == kDRegSize; }

  int lanes() const {
    return lanes_;
  }

  bool IsScalar() const {
    return lanes_ == 1;
  }

  bool IsVector() const {
    return lanes_ > 1;
  }

  bool IsSameFormat(const VRegister& other) const {
    return (size_ == other.size_) && (lanes_ == other.lanes_);
  }

  unsigned LaneSizeInBytes() const {
    return SizeInBytes() / lanes_;
  }

  unsigned LaneSizeInBits() const {
    return LaneSizeInBytes() * 8;
  }

 private:
  static const VRegister bregisters[];
  static const VRegister hregisters[];
  static const VRegister sregisters[];
  static const VRegister dregisters[];
  static const VRegister qregisters[];
  static const VRegister vregisters[];
  int lanes_;
};


// Backward compatibility for FPRegisters.
typedef VRegister FPRegister;

// No*Reg is used to indicate an unused argument, or an error case. Note that
// these all compare equal (using the Is() method). The Register and VRegister
// variants are provided for convenience.
const Register NoReg;
const VRegister NoVReg;
const FPRegister NoFPReg;  // For backward compatibility.
const CPURegister NoCPUReg;


#define DEFINE_REGISTERS(N)  \
const Register w##N(N, kWRegSize);  \
const Register x##N(N, kXRegSize);
REGISTER_CODE_LIST(DEFINE_REGISTERS)
#undef DEFINE_REGISTERS
const Register wsp(kSPRegInternalCode, kWRegSize);
const Register sp(kSPRegInternalCode, kXRegSize);


#define DEFINE_VREGISTERS(N)  \
const VRegister b##N(N, kBRegSize);  \
const VRegister h##N(N, kHRegSize);  \
const VRegister s##N(N, kSRegSize);  \
const VRegister d##N(N, kDRegSize);  \
const VRegister q##N(N, kQRegSize);  \
const VRegister v##N(N, kQRegSize);
REGISTER_CODE_LIST(DEFINE_VREGISTERS)
#undef DEFINE_VREGISTERS


// Registers aliases.
const Register ip0 = x16;
const Register ip1 = x17;
const Register lr = x30;
const Register xzr = x31;
const Register wzr = w31;


// AreAliased returns true if any of the named registers overlap. Arguments
// set to NoReg are ignored. The system stack pointer may be specified.
bool AreAliased(const CPURegister& reg1,
                const CPURegister& reg2,
                const CPURegister& reg3 = NoReg,
                const CPURegister& reg4 = NoReg,
                const CPURegister& reg5 = NoReg,
                const CPURegister& reg6 = NoReg,
                const CPURegister& reg7 = NoReg,
                const CPURegister& reg8 = NoReg);


// AreSameSizeAndType returns true if all of the specified registers have the
// same size, and are of the same type. The system stack pointer may be
// specified. Arguments set to NoReg are ignored, as are any subsequent
// arguments. At least one argument (reg1) must be valid (not NoCPUReg).
bool AreSameSizeAndType(const CPURegister& reg1,
                        const CPURegister& reg2,
                        const CPURegister& reg3 = NoCPUReg,
                        const CPURegister& reg4 = NoCPUReg,
                        const CPURegister& reg5 = NoCPUReg,
                        const CPURegister& reg6 = NoCPUReg,
                        const CPURegister& reg7 = NoCPUReg,
                        const CPURegister& reg8 = NoCPUReg);


// AreSameFormat returns true if all of the specified VRegisters have the same
// vector format. Arguments set to NoReg are ignored, as are any subsequent
// arguments. At least one argument (reg1) must be valid (not NoVReg).
bool AreSameFormat(const VRegister& reg1,
                   const VRegister& reg2,
                   const VRegister& reg3 = NoVReg,
                   const VRegister& reg4 = NoVReg);


// AreConsecutive returns true if all of the specified VRegisters are
// consecutive in the register file. Arguments set to NoReg are ignored, as are
// any subsequent arguments. At least one argument (reg1) must be valid
// (not NoVReg).
bool AreConsecutive(const VRegister& reg1,
                    const VRegister& reg2,
                    const VRegister& reg3 = NoVReg,
                    const VRegister& reg4 = NoVReg);


// Lists of registers.
class CPURegList {
 public:
  explicit CPURegList(CPURegister reg1,
                      CPURegister reg2 = NoCPUReg,
                      CPURegister reg3 = NoCPUReg,
                      CPURegister reg4 = NoCPUReg)
      : list_(reg1.Bit() | reg2.Bit() | reg3.Bit() | reg4.Bit()),
        size_(reg1.size()), type_(reg1.type()) {
    VIXL_ASSERT(AreSameSizeAndType(reg1, reg2, reg3, reg4));
    VIXL_ASSERT(IsValid());
  }

  CPURegList(CPURegister::RegisterType type, unsigned size, RegList list)
      : list_(list), size_(size), type_(type) {
    VIXL_ASSERT(IsValid());
  }

  CPURegList(CPURegister::RegisterType type, unsigned size,
             unsigned first_reg, unsigned last_reg)
      : size_(size), type_(type) {
    VIXL_ASSERT(((type == CPURegister::kRegister) &&
                 (last_reg < kNumberOfRegisters)) ||
                ((type == CPURegister::kVRegister) &&
                 (last_reg < kNumberOfVRegisters)));
    VIXL_ASSERT(last_reg >= first_reg);
    list_ = (UINT64_C(1) << (last_reg + 1)) - 1;
    list_ &= ~((UINT64_C(1) << first_reg) - 1);
    VIXL_ASSERT(IsValid());
  }

  CPURegister::RegisterType type() const {
    VIXL_ASSERT(IsValid());
    return type_;
  }

  // Combine another CPURegList into this one. Registers that already exist in
  // this list are left unchanged. The type and size of the registers in the
  // 'other' list must match those in this list.
  void Combine(const CPURegList& other) {
    VIXL_ASSERT(IsValid());
    VIXL_ASSERT(other.type() == type_);
    VIXL_ASSERT(other.RegisterSizeInBits() == size_);
    list_ |= other.list();
  }

  // Remove every register in the other CPURegList from this one. Registers that
  // do not exist in this list are ignored. The type and size of the registers
  // in the 'other' list must match those in this list.
  void Remove(const CPURegList& other) {
    VIXL_ASSERT(IsValid());
    VIXL_ASSERT(other.type() == type_);
    VIXL_ASSERT(other.RegisterSizeInBits() == size_);
    list_ &= ~other.list();
  }

  // Variants of Combine and Remove which take a single register.
  void Combine(const CPURegister& other) {
    VIXL_ASSERT(other.type() == type_);
    VIXL_ASSERT(other.size() == size_);
    Combine(other.code());
  }

  void Remove(const CPURegister& other) {
    VIXL_ASSERT(other.type() == type_);
    VIXL_ASSERT(other.size() == size_);
    Remove(other.code());
  }

  // Variants of Combine and Remove which take a single register by its code;
  // the type and size of the register is inferred from this list.
  void Combine(int code) {
    VIXL_ASSERT(IsValid());
    VIXL_ASSERT(CPURegister(code, size_, type_).IsValid());
    list_ |= (UINT64_C(1) << code);
  }

  void Remove(int code) {
    VIXL_ASSERT(IsValid());
    VIXL_ASSERT(CPURegister(code, size_, type_).IsValid());
    list_ &= ~(UINT64_C(1) << code);
  }

  static CPURegList Union(const CPURegList& list_1, const CPURegList& list_2) {
    VIXL_ASSERT(list_1.type_ == list_2.type_);
    VIXL_ASSERT(list_1.size_ == list_2.size_);
    return CPURegList(list_1.type_, list_1.size_, list_1.list_ | list_2.list_);
  }
  static CPURegList Union(const CPURegList& list_1,
                          const CPURegList& list_2,
                          const CPURegList& list_3);
  static CPURegList Union(const CPURegList& list_1,
                          const CPURegList& list_2,
                          const CPURegList& list_3,
                          const CPURegList& list_4);

  static CPURegList Intersection(const CPURegList& list_1,
                                 const CPURegList& list_2) {
    VIXL_ASSERT(list_1.type_ == list_2.type_);
    VIXL_ASSERT(list_1.size_ == list_2.size_);
    return CPURegList(list_1.type_, list_1.size_, list_1.list_ & list_2.list_);
  }
  static CPURegList Intersection(const CPURegList& list_1,
                                 const CPURegList& list_2,
                                 const CPURegList& list_3);
  static CPURegList Intersection(const CPURegList& list_1,
                                 const CPURegList& list_2,
                                 const CPURegList& list_3,
                                 const CPURegList& list_4);

  bool Overlaps(const CPURegList& other) const {
    return (type_ == other.type_) && ((list_ & other.list_) != 0);
  }

  RegList list() const {
    VIXL_ASSERT(IsValid());
    return list_;
  }

  void set_list(RegList new_list) {
    VIXL_ASSERT(IsValid());
    list_ = new_list;
  }

  // Remove all callee-saved registers from the list. This can be useful when
  // preparing registers for an AAPCS64 function call, for example.
  void RemoveCalleeSaved();

  CPURegister PopLowestIndex();
  CPURegister PopHighestIndex();

  // AAPCS64 callee-saved registers.
  static CPURegList GetCalleeSaved(unsigned size = kXRegSize);
  static CPURegList GetCalleeSavedV(unsigned size = kDRegSize);

  // AAPCS64 caller-saved registers. Note that this includes lr.
  // TODO(all): Determine how we handle d8-d15 being callee-saved, but the top
  // 64-bits being caller-saved.
  static CPURegList GetCallerSaved(unsigned size = kXRegSize);
  static CPURegList GetCallerSavedV(unsigned size = kDRegSize);

  bool IsEmpty() const {
    VIXL_ASSERT(IsValid());
    return list_ == 0;
  }

  bool IncludesAliasOf(const CPURegister& other) const {
    VIXL_ASSERT(IsValid());
    return (type_ == other.type()) && ((other.Bit() & list_) != 0);
  }

  bool IncludesAliasOf(int code) const {
    VIXL_ASSERT(IsValid());
    return ((code & list_) != 0);
  }

  int Count() const {
    VIXL_ASSERT(IsValid());
    return CountSetBits(list_);
  }

  unsigned RegisterSizeInBits() const {
    VIXL_ASSERT(IsValid());
    return size_;
  }

  unsigned RegisterSizeInBytes() const {
    int size_in_bits = RegisterSizeInBits();
    VIXL_ASSERT((size_in_bits % 8) == 0);
    return size_in_bits / 8;
  }

  unsigned TotalSizeInBytes() const {
    VIXL_ASSERT(IsValid());
    return RegisterSizeInBytes() * Count();
  }

 private:
  RegList list_;
  unsigned size_;
  CPURegister::RegisterType type_;

  bool IsValid() const;
};


// AAPCS64 callee-saved registers.
extern const CPURegList kCalleeSaved;
extern const CPURegList kCalleeSavedV;


// AAPCS64 caller-saved registers. Note that this includes lr.
extern const CPURegList kCallerSaved;
extern const CPURegList kCallerSavedV;


// Operand.
class Operand {
 public:
  // #<immediate>
  // where <immediate> is int64_t.
  // This is allowed to be an implicit constructor because Operand is
  // a wrapper class that doesn't normally perform any type conversion.
  Operand(int64_t immediate = 0);           // NOLINT(runtime/explicit)

  // rm, {<shift> #<shift_amount>}
  // where <shift> is one of {LSL, LSR, ASR, ROR}.
  //       <shift_amount> is uint6_t.
  // This is allowed to be an implicit constructor because Operand is
  // a wrapper class that doesn't normally perform any type conversion.
  Operand(Register reg,
          Shift shift = LSL,
          unsigned shift_amount = 0);   // NOLINT(runtime/explicit)

  // rm, {<extend> {#<shift_amount>}}
  // where <extend> is one of {UXTB, UXTH, UXTW, UXTX, SXTB, SXTH, SXTW, SXTX}.
  //       <shift_amount> is uint2_t.
  explicit Operand(Register reg, Extend extend, unsigned shift_amount = 0);

  bool IsImmediate() const;
  bool IsShiftedRegister() const;
  bool IsExtendedRegister() const;
  bool IsZero() const;

  // This returns an LSL shift (<= 4) operand as an equivalent extend operand,
  // which helps in the encoding of instructions that use the stack pointer.
  Operand ToExtendedRegister() const;

  int64_t immediate() const {
    VIXL_ASSERT(IsImmediate());
    return immediate_;
  }

  Register reg() const {
    VIXL_ASSERT(IsShiftedRegister() || IsExtendedRegister());
    return reg_;
  }

  Shift shift() const {
    VIXL_ASSERT(IsShiftedRegister());
    return shift_;
  }

  Extend extend() const {
    VIXL_ASSERT(IsExtendedRegister());
    return extend_;
  }

  unsigned shift_amount() const {
    VIXL_ASSERT(IsShiftedRegister() || IsExtendedRegister());
    return shift_amount_;
  }

 private:
  int64_t immediate_;
  Register reg_;
  Shift shift_;
  Extend extend_;
  unsigned shift_amount_;
};


// MemOperand represents the addressing mode of a load or store instruction.
class MemOperand {
 public:
  explicit MemOperand(Register base,
                      int64_t offset = 0,
                      AddrMode addrmode = Offset);
  MemOperand(Register base,
             Register regoffset,
             Shift shift = LSL,
             unsigned shift_amount = 0);
  MemOperand(Register base,
             Register regoffset,
             Extend extend,
             unsigned shift_amount = 0);
  MemOperand(Register base,
             const Operand& offset,
             AddrMode addrmode = Offset);

  const Register& base() const { return base_; }
  const Register& regoffset() const { return regoffset_; }
  int64_t offset() const { return offset_; }
  AddrMode addrmode() const { return addrmode_; }
  Shift shift() const { return shift_; }
  Extend extend() const { return extend_; }
  unsigned shift_amount() const { return shift_amount_; }
  bool IsImmediateOffset() const;
  bool IsRegisterOffset() const;
  bool IsPreIndex() const;
  bool IsPostIndex() const;

  void AddOffset(int64_t offset);

 private:
  Register base_;
  Register regoffset_;
  int64_t offset_;
  AddrMode addrmode_;
  Shift shift_;
  Extend extend_;
  unsigned shift_amount_;
};


class LabelTestHelper;  // Forward declaration.


class Label {
 public:
  Label() : location_(kLocationUnbound) {}
  ~Label() {
    // If the label has been linked to, it needs to be bound to a target.
    VIXL_ASSERT(!IsLinked() || IsBound());
  }

  bool IsBound() const { return location_ >= 0; }
  bool IsLinked() const { return !links_.empty(); }

  ptrdiff_t location() const { return location_; }

  static const int kNPreallocatedLinks = 4;
  static const ptrdiff_t kInvalidLinkKey = PTRDIFF_MAX;
  static const size_t kReclaimFrom = 512;
  static const size_t kReclaimFactor = 2;

  typedef InvalSet<ptrdiff_t,
                   kNPreallocatedLinks,
                   ptrdiff_t,
                   kInvalidLinkKey,
                   kReclaimFrom,
                   kReclaimFactor> LinksSetBase;
  typedef InvalSetIterator<LinksSetBase> LabelLinksIteratorBase;

 private:
  class LinksSet : public LinksSetBase {
   public:
    LinksSet() : LinksSetBase() {}
  };

  // Allows iterating over the links of a label. The behaviour is undefined if
  // the list of links is modified in any way while iterating.
  class LabelLinksIterator : public LabelLinksIteratorBase {
   public:
    explicit LabelLinksIterator(Label* label)
        : LabelLinksIteratorBase(&label->links_) {}
  };

  void Bind(ptrdiff_t location) {
    // Labels can only be bound once.
    VIXL_ASSERT(!IsBound());
    location_ = location;
  }

  void AddLink(ptrdiff_t instruction) {
    // If a label is bound, the assembler already has the information it needs
    // to write the instruction, so there is no need to add it to links_.
    VIXL_ASSERT(!IsBound());
    links_.insert(instruction);
  }

  void DeleteLink(ptrdiff_t instruction) {
    links_.erase(instruction);
  }

  void ClearAllLinks() {
    links_.clear();
  }

  // TODO: The comment below considers average case complexity for our
  // usual use-cases. The elements of interest are:
  // - Branches to a label are emitted in order: branch instructions to a label
  // are generated at an offset in the code generation buffer greater than any
  // other branch to that same label already generated. As an example, this can
  // be broken when an instruction is patched to become a branch. Note that the
  // code will still work, but the complexity considerations below may locally
  // not apply any more.
  // - Veneers are generated in order: for multiple branches of the same type
  // branching to the same unbound label going out of range, veneers are
  // generated in growing order of the branch instruction offset from the start
  // of the buffer.
  //
  // When creating a veneer for a branch going out of range, the link for this
  // branch needs to be removed from this `links_`. Since all branches are
  // tracked in one underlying InvalSet, the complexity for this deletion is the
  // same as for finding the element, ie. O(n), where n is the number of links
  // in the set.
  // This could be reduced to O(1) by using the same trick as used when tracking
  // branch information for veneers: split the container to use one set per type
  // of branch. With that setup, when a veneer is created and the link needs to
  // be deleted, if the two points above hold, it must be the minimum element of
  // the set for its type of branch, and that minimum element will be accessible
  // in O(1).

  // The offsets of the instructions that have linked to this label.
  LinksSet links_;
  // The label location.
  ptrdiff_t location_;

  static const ptrdiff_t kLocationUnbound = -1;

  // It is not safe to copy labels, so disable the copy constructor and operator
  // by declaring them private (without an implementation).
  Label(const Label&);
  void operator=(const Label&);

  // The Assembler class is responsible for binding and linking labels, since
  // the stored offsets need to be consistent with the Assembler's buffer.
  friend class Assembler;
  // The MacroAssembler and VeneerPool handle resolution of branches to distant
  // targets.
  friend class MacroAssembler;
  friend class VeneerPool;
};


// Required InvalSet template specialisations.
#define INVAL_SET_TEMPLATE_PARAMETERS \
    ptrdiff_t,                        \
    Label::kNPreallocatedLinks,       \
    ptrdiff_t,                        \
    Label::kInvalidLinkKey,           \
    Label::kReclaimFrom,              \
    Label::kReclaimFactor
template<>
inline ptrdiff_t InvalSet<INVAL_SET_TEMPLATE_PARAMETERS>::Key(
    const ptrdiff_t& element) {
  return element;
}
template<>
inline void InvalSet<INVAL_SET_TEMPLATE_PARAMETERS>::SetKey(
              ptrdiff_t* element, ptrdiff_t key) {
  *element = key;
}
#undef INVAL_SET_TEMPLATE_PARAMETERS


class Assembler;
class LiteralPool;

// A literal is a 32-bit or 64-bit piece of data stored in the instruction
// stream and loaded through a pc relative load. The same literal can be
// referred to by multiple instructions but a literal can only reside at one
// place in memory. A literal can be used by a load before or after being
// placed in memory.
//
// Internally an offset of 0 is associated with a literal which has been
// neither used nor placed. Then two possibilities arise:
//  1) the label is placed, the offset (stored as offset + 1) is used to
//     resolve any subsequent load using the label.
//  2) the label is not placed and offset is the offset of the last load using
//     the literal (stored as -offset -1). If multiple loads refer to this
//     literal then the last load holds the offset of the preceding load and
//     all loads form a chain. Once the offset is placed all the loads in the
//     chain are resolved and future loads fall back to possibility 1.
class RawLiteral {
 public:
  enum DeletionPolicy {
    kDeletedOnPlacementByPool,
    kDeletedOnPoolDestruction,
    kManuallyDeleted
  };

  RawLiteral(size_t size,
             LiteralPool* literal_pool,
             DeletionPolicy deletion_policy = kManuallyDeleted);

  // The literal pool only sees and deletes `RawLiteral*` pointers, but they are
  // actually pointing to `Literal<T>` objects.
  virtual ~RawLiteral() {}

  size_t size() {
    VIXL_STATIC_ASSERT(kDRegSizeInBytes == kXRegSizeInBytes);
    VIXL_STATIC_ASSERT(kSRegSizeInBytes == kWRegSizeInBytes);
    VIXL_ASSERT((size_ == kXRegSizeInBytes) ||
                (size_ == kWRegSizeInBytes) ||
                (size_ == kQRegSizeInBytes));
    return size_;
  }
  uint64_t raw_value128_low64() {
    VIXL_ASSERT(size_ == kQRegSizeInBytes);
    return low64_;
  }
  uint64_t raw_value128_high64() {
    VIXL_ASSERT(size_ == kQRegSizeInBytes);
    return high64_;
  }
  uint64_t raw_value64() {
    VIXL_ASSERT(size_ == kXRegSizeInBytes);
    VIXL_ASSERT(high64_ == 0);
    return low64_;
  }
  uint32_t raw_value32() {
    VIXL_ASSERT(size_ == kWRegSizeInBytes);
    VIXL_ASSERT(high64_ == 0);
    VIXL_ASSERT(is_uint32(low64_) || is_int32(low64_));
    return static_cast<uint32_t>(low64_);
  }
  bool IsUsed() { return offset_ < 0; }
  bool IsPlaced() { return offset_ > 0; }

  LiteralPool* GetLiteralPool() const {
    return literal_pool_;
  }

  ptrdiff_t offset() {
    VIXL_ASSERT(IsPlaced());
    return offset_ - 1;
  }

 protected:
  void set_offset(ptrdiff_t offset) {
    VIXL_ASSERT(offset >= 0);
    VIXL_ASSERT(IsWordAligned(offset));
    VIXL_ASSERT(!IsPlaced());
    offset_ = offset + 1;
  }
  ptrdiff_t last_use() {
    VIXL_ASSERT(IsUsed());
    return -offset_ - 1;
  }
  void set_last_use(ptrdiff_t offset) {
    VIXL_ASSERT(offset >= 0);
    VIXL_ASSERT(IsWordAligned(offset));
    VIXL_ASSERT(!IsPlaced());
    offset_ = -offset - 1;
  }

  size_t size_;
  ptrdiff_t offset_;
  uint64_t low64_;
  uint64_t high64_;

 private:
  LiteralPool* literal_pool_;
  DeletionPolicy deletion_policy_;

  friend class Assembler;
  friend class LiteralPool;
};


template <typename T>
class Literal : public RawLiteral {
 public:
  explicit Literal(T value,
                   LiteralPool* literal_pool = NULL,
                   RawLiteral::DeletionPolicy ownership = kManuallyDeleted)
      : RawLiteral(sizeof(value), literal_pool, ownership) {
    VIXL_STATIC_ASSERT(sizeof(value) <= kXRegSizeInBytes);
    UpdateValue(value);
  }

  Literal(T high64, T low64,
          LiteralPool* literal_pool = NULL,
          RawLiteral::DeletionPolicy ownership = kManuallyDeleted)
      : RawLiteral(kQRegSizeInBytes, literal_pool, ownership) {
    VIXL_STATIC_ASSERT(sizeof(low64) == (kQRegSizeInBytes / 2));
    UpdateValue(high64, low64);
  }

  virtual ~Literal() {}

  // Update the value of this literal, if necessary by rewriting the value in
  // the pool.
  // If the literal has already been placed in a literal pool, the address of
  // the start of the code buffer must be provided, as the literal only knows it
  // offset from there.  This also allows patching the value after the code has
  // been moved in memory.
  void UpdateValue(T new_value, uint8_t* code_buffer = NULL) {
    VIXL_ASSERT(sizeof(new_value) == size_);
    memcpy(&low64_, &new_value, sizeof(new_value));
    if (IsPlaced()) {
      VIXL_ASSERT(code_buffer != NULL);
      RewriteValueInCode(code_buffer);
    }
  }

  void UpdateValue(T high64, T low64, uint8_t* code_buffer = NULL) {
    VIXL_ASSERT(sizeof(low64) == size_ / 2);
    memcpy(&low64_, &low64, sizeof(low64));
    memcpy(&high64_, &high64, sizeof(high64));
    if (IsPlaced()) {
      VIXL_ASSERT(code_buffer != NULL);
      RewriteValueInCode(code_buffer);
    }
  }

  void UpdateValue(T new_value, const Assembler* assembler);
  void UpdateValue(T high64, T low64, const Assembler* assembler);

 private:
  void RewriteValueInCode(uint8_t* code_buffer) {
    VIXL_ASSERT(IsPlaced());
    VIXL_STATIC_ASSERT(sizeof(T) <= kXRegSizeInBytes);
    switch (size()) {
      case kSRegSizeInBytes:
        *reinterpret_cast<uint32_t*>(code_buffer + offset()) = raw_value32();
        break;
      case kDRegSizeInBytes:
        *reinterpret_cast<uint64_t*>(code_buffer + offset()) = raw_value64();
        break;
      default:
        VIXL_ASSERT(size() == kQRegSizeInBytes);
        uint64_t* base_address =
            reinterpret_cast<uint64_t*>(code_buffer + offset());
        *base_address = raw_value128_low64();
        *(base_address + 1) = raw_value128_high64();
    }
  }
};


// Control whether or not position-independent code should be emitted.
enum PositionIndependentCodeOption {
  // All code generated will be position-independent; all branches and
  // references to labels generated with the Label class will use PC-relative
  // addressing.
  PositionIndependentCode,

  // Allow VIXL to generate code that refers to absolute addresses. With this
  // option, it will not be possible to copy the code buffer and run it from a
  // different address; code must be generated in its final location.
  PositionDependentCode,

  // Allow VIXL to assume that the bottom 12 bits of the address will be
  // constant, but that the top 48 bits may change. This allows `adrp` to
  // function in systems which copy code between pages, but otherwise maintain
  // 4KB page alignment.
  PageOffsetDependentCode
};


// Control how scaled- and unscaled-offset loads and stores are generated.
enum LoadStoreScalingOption {
  // Prefer scaled-immediate-offset instructions, but emit unscaled-offset,
  // register-offset, pre-index or post-index instructions if necessary.
  PreferScaledOffset,

  // Prefer unscaled-immediate-offset instructions, but emit scaled-offset,
  // register-offset, pre-index or post-index instructions if necessary.
  PreferUnscaledOffset,

  // Require scaled-immediate-offset instructions.
  RequireScaledOffset,

  // Require unscaled-immediate-offset instructions.
  RequireUnscaledOffset
};


// Assembler.
class Assembler {
 public:
  Assembler(size_t capacity,
            PositionIndependentCodeOption pic = PositionIndependentCode);
  Assembler(byte* buffer, size_t capacity,
            PositionIndependentCodeOption pic = PositionIndependentCode);

  // The destructor asserts that one of the following is true:
  //  * The Assembler object has not been used.
  //  * Nothing has been emitted since the last Reset() call.
  //  * Nothing has been emitted since the last FinalizeCode() call.
  ~Assembler();

  // System functions.

  // Start generating code from the beginning of the buffer, discarding any code
  // and data that has already been emitted into the buffer.
  void Reset();

  // Finalize a code buffer of generated instructions. This function must be
  // called before executing or copying code from the buffer.
  void FinalizeCode();

  // Label.
  // Bind a label to the current PC.
  void bind(Label* label);

  // Bind a label to a specified offset from the start of the buffer.
  void BindToOffset(Label* label, ptrdiff_t offset);

  // Place a literal at the current PC.
  void place(RawLiteral* literal);

  ptrdiff_t CursorOffset() const {
    return buffer_->CursorOffset();
  }

  ptrdiff_t BufferEndOffset() const {
    return static_cast<ptrdiff_t>(buffer_->capacity());
  }

  // Return the address of an offset in the buffer.
  template <typename T>
  T GetOffsetAddress(ptrdiff_t offset) const {
    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
    return buffer_->GetOffsetAddress<T>(offset);
  }

  // Return the address of a bound label.
  template <typename T>
  T GetLabelAddress(const Label * label) const {
    VIXL_ASSERT(label->IsBound());
    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
    return GetOffsetAddress<T>(label->location());
  }

  // Return the address of the cursor.
  template <typename T>
  T GetCursorAddress() const {
    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
    return GetOffsetAddress<T>(CursorOffset());
  }

  // Return the address of the start of the buffer.
  template <typename T>
  T GetStartAddress() const {
    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
    return GetOffsetAddress<T>(0);
  }

  Instruction* InstructionAt(ptrdiff_t instruction_offset) {
    return GetOffsetAddress<Instruction*>(instruction_offset);
  }

  ptrdiff_t InstructionOffset(Instruction* instruction) {
    VIXL_STATIC_ASSERT(sizeof(*instruction) == 1);
    ptrdiff_t offset = instruction - GetStartAddress<Instruction*>();
    VIXL_ASSERT((0 <= offset) &&
                (offset < static_cast<ptrdiff_t>(BufferCapacity())));
    return offset;
  }

  // Instruction set functions.

  // Branch / Jump instructions.
  // Branch to register.
  void br(const Register& xn);

  // Branch with link to register.
  void blr(const Register& xn);

  // Branch to register with return hint.
  void ret(const Register& xn = lr);

  // Unconditional branch to label.
  void b(Label* label);

  // Conditional branch to label.
  void b(Label* label, Condition cond);

  // Unconditional branch to PC offset.
  void b(int imm26);

  // Conditional branch to PC offset.
  void b(int imm19, Condition cond);

  // Branch with link to label.
  void bl(Label* label);

  // Branch with link to PC offset.
  void bl(int imm26);

  // Compare and branch to label if zero.
  void cbz(const Register& rt, Label* label);

  // Compare and branch to PC offset if zero.
  void cbz(const Register& rt, int imm19);

  // Compare and branch to label if not zero.
  void cbnz(const Register& rt, Label* label);

  // Compare and branch to PC offset if not zero.
  void cbnz(const Register& rt, int imm19);

  // Table lookup from one register.
  void tbl(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Table lookup from two registers.
  void tbl(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vm);

  // Table lookup from three registers.
  void tbl(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vn3,
           const VRegister& vm);

  // Table lookup from four registers.
  void tbl(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vn3,
           const VRegister& vn4,
           const VRegister& vm);

  // Table lookup extension from one register.
  void tbx(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Table lookup extension from two registers.
  void tbx(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vm);

  // Table lookup extension from three registers.
  void tbx(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vn3,
           const VRegister& vm);

  // Table lookup extension from four registers.
  void tbx(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vn2,
           const VRegister& vn3,
           const VRegister& vn4,
           const VRegister& vm);

  // Test bit and branch to label if zero.
  void tbz(const Register& rt, unsigned bit_pos, Label* label);

  // Test bit and branch to PC offset if zero.
  void tbz(const Register& rt, unsigned bit_pos, int imm14);

  // Test bit and branch to label if not zero.
  void tbnz(const Register& rt, unsigned bit_pos, Label* label);

  // Test bit and branch to PC offset if not zero.
  void tbnz(const Register& rt, unsigned bit_pos, int imm14);

  // Address calculation instructions.
  // Calculate a PC-relative address. Unlike for branches the offset in adr is
  // unscaled (i.e. the result can be unaligned).

  // Calculate the address of a label.
  void adr(const Register& rd, Label* label);

  // Calculate the address of a PC offset.
  void adr(const Register& rd, int imm21);

  // Calculate the page address of a label.
  void adrp(const Register& rd, Label* label);

  // Calculate the page address of a PC offset.
  void adrp(const Register& rd, int imm21);

  // Data Processing instructions.
  // Add.
  void add(const Register& rd,
           const Register& rn,
           const Operand& operand);

  // Add and update status flags.
  void adds(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Compare negative.
  void cmn(const Register& rn, const Operand& operand);

  // Subtract.
  void sub(const Register& rd,
           const Register& rn,
           const Operand& operand);

  // Subtract and update status flags.
  void subs(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Compare.
  void cmp(const Register& rn, const Operand& operand);

  // Negate.
  void neg(const Register& rd,
           const Operand& operand);

  // Negate and update status flags.
  void negs(const Register& rd,
            const Operand& operand);

  // Add with carry bit.
  void adc(const Register& rd,
           const Register& rn,
           const Operand& operand);

  // Add with carry bit and update status flags.
  void adcs(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Subtract with carry bit.
  void sbc(const Register& rd,
           const Register& rn,
           const Operand& operand);

  // Subtract with carry bit and update status flags.
  void sbcs(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Negate with carry bit.
  void ngc(const Register& rd,
           const Operand& operand);

  // Negate with carry bit and update status flags.
  void ngcs(const Register& rd,
            const Operand& operand);

  // Logical instructions.
  // Bitwise and (A & B).
  void and_(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Bitwise and (A & B) and update status flags.
  void ands(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Bit test and set flags.
  void tst(const Register& rn, const Operand& operand);

  // Bit clear (A & ~B).
  void bic(const Register& rd,
           const Register& rn,
           const Operand& operand);

  // Bit clear (A & ~B) and update status flags.
  void bics(const Register& rd,
            const Register& rn,
            const Operand& operand);

  // Bitwise or (A | B).
  void orr(const Register& rd, const Register& rn, const Operand& operand);

  // Bitwise nor (A | ~B).
  void orn(const Register& rd, const Register& rn, const Operand& operand);

  // Bitwise eor/xor (A ^ B).
  void eor(const Register& rd, const Register& rn, const Operand& operand);

  // Bitwise enor/xnor (A ^ ~B).
  void eon(const Register& rd, const Register& rn, const Operand& operand);

  // Logical shift left by variable.
  void lslv(const Register& rd, const Register& rn, const Register& rm);

  // Logical shift right by variable.
  void lsrv(const Register& rd, const Register& rn, const Register& rm);

  // Arithmetic shift right by variable.
  void asrv(const Register& rd, const Register& rn, const Register& rm);

  // Rotate right by variable.
  void rorv(const Register& rd, const Register& rn, const Register& rm);

  // Bitfield instructions.
  // Bitfield move.
  void bfm(const Register& rd,
           const Register& rn,
           unsigned immr,
           unsigned imms);

  // Signed bitfield move.
  void sbfm(const Register& rd,
            const Register& rn,
            unsigned immr,
            unsigned imms);

  // Unsigned bitfield move.
  void ubfm(const Register& rd,
            const Register& rn,
            unsigned immr,
            unsigned imms);

  // Bfm aliases.
  // Bitfield insert.
  void bfi(const Register& rd,
           const Register& rn,
           unsigned lsb,
           unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    bfm(rd, rn, (rd.size() - lsb) & (rd.size() - 1), width - 1);
  }

  // Bitfield extract and insert low.
  void bfxil(const Register& rd,
             const Register& rn,
             unsigned lsb,
             unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    bfm(rd, rn, lsb, lsb + width - 1);
  }

  // Sbfm aliases.
  // Arithmetic shift right.
  void asr(const Register& rd, const Register& rn, unsigned shift) {
    VIXL_ASSERT(shift < rd.size());
    sbfm(rd, rn, shift, rd.size() - 1);
  }

  // Signed bitfield insert with zero at right.
  void sbfiz(const Register& rd,
             const Register& rn,
             unsigned lsb,
             unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    sbfm(rd, rn, (rd.size() - lsb) & (rd.size() - 1), width - 1);
  }

  // Signed bitfield extract.
  void sbfx(const Register& rd,
            const Register& rn,
            unsigned lsb,
            unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    sbfm(rd, rn, lsb, lsb + width - 1);
  }

  // Signed extend byte.
  void sxtb(const Register& rd, const Register& rn) {
    sbfm(rd, rn, 0, 7);
  }

  // Signed extend halfword.
  void sxth(const Register& rd, const Register& rn) {
    sbfm(rd, rn, 0, 15);
  }

  // Signed extend word.
  void sxtw(const Register& rd, const Register& rn) {
    sbfm(rd, rn, 0, 31);
  }

  // Ubfm aliases.
  // Logical shift left.
  void lsl(const Register& rd, const Register& rn, unsigned shift) {
    unsigned reg_size = rd.size();
    VIXL_ASSERT(shift < reg_size);
    ubfm(rd, rn, (reg_size - shift) % reg_size, reg_size - shift - 1);
  }

  // Logical shift right.
  void lsr(const Register& rd, const Register& rn, unsigned shift) {
    VIXL_ASSERT(shift < rd.size());
    ubfm(rd, rn, shift, rd.size() - 1);
  }

  // Unsigned bitfield insert with zero at right.
  void ubfiz(const Register& rd,
             const Register& rn,
             unsigned lsb,
             unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    ubfm(rd, rn, (rd.size() - lsb) & (rd.size() - 1), width - 1);
  }

  // Unsigned bitfield extract.
  void ubfx(const Register& rd,
            const Register& rn,
            unsigned lsb,
            unsigned width) {
    VIXL_ASSERT(width >= 1);
    VIXL_ASSERT(lsb + width <= rn.size());
    ubfm(rd, rn, lsb, lsb + width - 1);
  }

  // Unsigned extend byte.
  void uxtb(const Register& rd, const Register& rn) {
    ubfm(rd, rn, 0, 7);
  }

  // Unsigned extend halfword.
  void uxth(const Register& rd, const Register& rn) {
    ubfm(rd, rn, 0, 15);
  }

  // Unsigned extend word.
  void uxtw(const Register& rd, const Register& rn) {
    ubfm(rd, rn, 0, 31);
  }

  // Extract.
  void extr(const Register& rd,
            const Register& rn,
            const Register& rm,
            unsigned lsb);

  // Conditional select: rd = cond ? rn : rm.
  void csel(const Register& rd,
            const Register& rn,
            const Register& rm,
            Condition cond);

  // Conditional select increment: rd = cond ? rn : rm + 1.
  void csinc(const Register& rd,
             const Register& rn,
             const Register& rm,
             Condition cond);

  // Conditional select inversion: rd = cond ? rn : ~rm.
  void csinv(const Register& rd,
             const Register& rn,
             const Register& rm,
             Condition cond);

  // Conditional select negation: rd = cond ? rn : -rm.
  void csneg(const Register& rd,
             const Register& rn,
             const Register& rm,
             Condition cond);

  // Conditional set: rd = cond ? 1 : 0.
  void cset(const Register& rd, Condition cond);

  // Conditional set mask: rd = cond ? -1 : 0.
  void csetm(const Register& rd, Condition cond);

  // Conditional increment: rd = cond ? rn + 1 : rn.
  void cinc(const Register& rd, const Register& rn, Condition cond);

  // Conditional invert: rd = cond ? ~rn : rn.
  void cinv(const Register& rd, const Register& rn, Condition cond);

  // Conditional negate: rd = cond ? -rn : rn.
  void cneg(const Register& rd, const Register& rn, Condition cond);

  // Rotate right.
  void ror(const Register& rd, const Register& rs, unsigned shift) {
    extr(rd, rs, rs, shift);
  }

  // Conditional comparison.
  // Conditional compare negative.
  void ccmn(const Register& rn,
            const Operand& operand,
            StatusFlags nzcv,
            Condition cond);

  // Conditional compare.
  void ccmp(const Register& rn,
            const Operand& operand,
            StatusFlags nzcv,
            Condition cond);

  // CRC-32 checksum from byte.
  void crc32b(const Register& rd,
              const Register& rn,
              const Register& rm);

  // CRC-32 checksum from half-word.
  void crc32h(const Register& rd,
              const Register& rn,
              const Register& rm);

  // CRC-32 checksum from word.
  void crc32w(const Register& rd,
              const Register& rn,
              const Register& rm);

  // CRC-32 checksum from double word.
  void crc32x(const Register& rd,
              const Register& rn,
              const Register& rm);

  // CRC-32 C checksum from byte.
  void crc32cb(const Register& rd,
               const Register& rn,
               const Register& rm);

  // CRC-32 C checksum from half-word.
  void crc32ch(const Register& rd,
               const Register& rn,
               const Register& rm);

  // CRC-32 C checksum from word.
  void crc32cw(const Register& rd,
               const Register& rn,
               const Register& rm);

  // CRC-32C checksum from double word.
  void crc32cx(const Register& rd,
               const Register& rn,
               const Register& rm);

  // Multiply.
  void mul(const Register& rd, const Register& rn, const Register& rm);

  // Negated multiply.
  void mneg(const Register& rd, const Register& rn, const Register& rm);

  // Signed long multiply: 32 x 32 -> 64-bit.
  void smull(const Register& rd, const Register& rn, const Register& rm);

  // Signed multiply high: 64 x 64 -> 64-bit <127:64>.
  void smulh(const Register& xd, const Register& xn, const Register& xm);

  // Multiply and accumulate.
  void madd(const Register& rd,
            const Register& rn,
            const Register& rm,
            const Register& ra);

  // Multiply and subtract.
  void msub(const Register& rd,
            const Register& rn,
            const Register& rm,
            const Register& ra);

  // Signed long multiply and accumulate: 32 x 32 + 64 -> 64-bit.
  void smaddl(const Register& rd,
              const Register& rn,
              const Register& rm,
              const Register& ra);

  // Unsigned long multiply and accumulate: 32 x 32 + 64 -> 64-bit.
  void umaddl(const Register& rd,
              const Register& rn,
              const Register& rm,
              const Register& ra);

  // Unsigned long multiply: 32 x 32 -> 64-bit.
  void umull(const Register& rd,
             const Register& rn,
             const Register& rm) {
    umaddl(rd, rn, rm, xzr);
  }

  // Unsigned multiply high: 64 x 64 -> 64-bit <127:64>.
  void umulh(const Register& xd,
             const Register& xn,
             const Register& xm);

  // Signed long multiply and subtract: 64 - (32 x 32) -> 64-bit.
  void smsubl(const Register& rd,
              const Register& rn,
              const Register& rm,
              const Register& ra);

  // Unsigned long multiply and subtract: 64 - (32 x 32) -> 64-bit.
  void umsubl(const Register& rd,
              const Register& rn,
              const Register& rm,
              const Register& ra);

  // Signed integer divide.
  void sdiv(const Register& rd, const Register& rn, const Register& rm);

  // Unsigned integer divide.
  void udiv(const Register& rd, const Register& rn, const Register& rm);

  // Bit reverse.
  void rbit(const Register& rd, const Register& rn);

  // Reverse bytes in 16-bit half words.
  void rev16(const Register& rd, const Register& rn);

  // Reverse bytes in 32-bit words.
  void rev32(const Register& rd, const Register& rn);

  // Reverse bytes.
  void rev(const Register& rd, const Register& rn);

  // Count leading zeroes.
  void clz(const Register& rd, const Register& rn);

  // Count leading sign bits.
  void cls(const Register& rd, const Register& rn);

  // Memory instructions.
  // Load integer or FP register.
  void ldr(const CPURegister& rt, const MemOperand& src,
           LoadStoreScalingOption option = PreferScaledOffset);

  // Store integer or FP register.
  void str(const CPURegister& rt, const MemOperand& dst,
           LoadStoreScalingOption option = PreferScaledOffset);

  // Load word with sign extension.
  void ldrsw(const Register& rt, const MemOperand& src,
             LoadStoreScalingOption option = PreferScaledOffset);

  // Load byte.
  void ldrb(const Register& rt, const MemOperand& src,
            LoadStoreScalingOption option = PreferScaledOffset);

  // Store byte.
  void strb(const Register& rt, const MemOperand& dst,
            LoadStoreScalingOption option = PreferScaledOffset);

  // Load byte with sign extension.
  void ldrsb(const Register& rt, const MemOperand& src,
             LoadStoreScalingOption option = PreferScaledOffset);

  // Load half-word.
  void ldrh(const Register& rt, const MemOperand& src,
            LoadStoreScalingOption option = PreferScaledOffset);

  // Store half-word.
  void strh(const Register& rt, const MemOperand& dst,
            LoadStoreScalingOption option = PreferScaledOffset);

  // Load half-word with sign extension.
  void ldrsh(const Register& rt, const MemOperand& src,
             LoadStoreScalingOption option = PreferScaledOffset);

  // Load integer or FP register (with unscaled offset).
  void ldur(const CPURegister& rt, const MemOperand& src,
            LoadStoreScalingOption option = PreferUnscaledOffset);

  // Store integer or FP register (with unscaled offset).
  void stur(const CPURegister& rt, const MemOperand& src,
            LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load word with sign extension.
  void ldursw(const Register& rt, const MemOperand& src,
              LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load byte (with unscaled offset).
  void ldurb(const Register& rt, const MemOperand& src,
             LoadStoreScalingOption option = PreferUnscaledOffset);

  // Store byte (with unscaled offset).
  void sturb(const Register& rt, const MemOperand& dst,
             LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load byte with sign extension (and unscaled offset).
  void ldursb(const Register& rt, const MemOperand& src,
              LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load half-word (with unscaled offset).
  void ldurh(const Register& rt, const MemOperand& src,
             LoadStoreScalingOption option = PreferUnscaledOffset);

  // Store half-word (with unscaled offset).
  void sturh(const Register& rt, const MemOperand& dst,
             LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load half-word with sign extension (and unscaled offset).
  void ldursh(const Register& rt, const MemOperand& src,
              LoadStoreScalingOption option = PreferUnscaledOffset);

  // Load integer or FP register pair.
  void ldp(const CPURegister& rt, const CPURegister& rt2,
           const MemOperand& src);

  // Store integer or FP register pair.
  void stp(const CPURegister& rt, const CPURegister& rt2,
           const MemOperand& dst);

  // Load word pair with sign extension.
  void ldpsw(const Register& rt, const Register& rt2, const MemOperand& src);

  // Load integer or FP register pair, non-temporal.
  void ldnp(const CPURegister& rt, const CPURegister& rt2,
            const MemOperand& src);

  // Store integer or FP register pair, non-temporal.
  void stnp(const CPURegister& rt, const CPURegister& rt2,
            const MemOperand& dst);

  // Load integer or FP register from literal pool.
  void ldr(const CPURegister& rt, RawLiteral* literal);

  // Load word with sign extension from literal pool.
  void ldrsw(const Register& rt, RawLiteral* literal);

  // Load integer or FP register from pc + imm19 << 2.
  void ldr(const CPURegister& rt, int imm19);

  // Load word with sign extension from pc + imm19 << 2.
  void ldrsw(const Register& rt, int imm19);

  // Store exclusive byte.
  void stxrb(const Register& rs, const Register& rt, const MemOperand& dst);

  // Store exclusive half-word.
  void stxrh(const Register& rs, const Register& rt, const MemOperand& dst);

  // Store exclusive register.
  void stxr(const Register& rs, const Register& rt, const MemOperand& dst);

  // Load exclusive byte.
  void ldxrb(const Register& rt, const MemOperand& src);

  // Load exclusive half-word.
  void ldxrh(const Register& rt, const MemOperand& src);

  // Load exclusive register.
  void ldxr(const Register& rt, const MemOperand& src);

  // Store exclusive register pair.
  void stxp(const Register& rs,
            const Register& rt,
            const Register& rt2,
            const MemOperand& dst);

  // Load exclusive register pair.
  void ldxp(const Register& rt, const Register& rt2, const MemOperand& src);

  // Store-release exclusive byte.
  void stlxrb(const Register& rs, const Register& rt, const MemOperand& dst);

  // Store-release exclusive half-word.
  void stlxrh(const Register& rs, const Register& rt, const MemOperand& dst);

  // Store-release exclusive register.
  void stlxr(const Register& rs, const Register& rt, const MemOperand& dst);

  // Load-acquire exclusive byte.
  void ldaxrb(const Register& rt, const MemOperand& src);

  // Load-acquire exclusive half-word.
  void ldaxrh(const Register& rt, const MemOperand& src);

  // Load-acquire exclusive register.
  void ldaxr(const Register& rt, const MemOperand& src);

  // Store-release exclusive register pair.
  void stlxp(const Register& rs,
             const Register& rt,
             const Register& rt2,
             const MemOperand& dst);

  // Load-acquire exclusive register pair.
  void ldaxp(const Register& rt, const Register& rt2, const MemOperand& src);

  // Store-release byte.
  void stlrb(const Register& rt, const MemOperand& dst);

  // Store-release half-word.
  void stlrh(const Register& rt, const MemOperand& dst);

  // Store-release register.
  void stlr(const Register& rt, const MemOperand& dst);

  // Load-acquire byte.
  void ldarb(const Register& rt, const MemOperand& src);

  // Load-acquire half-word.
  void ldarh(const Register& rt, const MemOperand& src);

  // Load-acquire register.
  void ldar(const Register& rt, const MemOperand& src);

  // Prefetch memory.
  void prfm(PrefetchOperation op, const MemOperand& addr,
            LoadStoreScalingOption option = PreferScaledOffset);

  // Prefetch memory (with unscaled offset).
  void prfum(PrefetchOperation op, const MemOperand& addr,
             LoadStoreScalingOption option = PreferUnscaledOffset);

  // Prefetch memory in the literal pool.
  void prfm(PrefetchOperation op, RawLiteral* literal);

  // Prefetch from pc + imm19 << 2.
  void prfm(PrefetchOperation op, int imm19);

  // Move instructions. The default shift of -1 indicates that the move
  // instruction will calculate an appropriate 16-bit immediate and left shift
  // that is equal to the 64-bit immediate argument. If an explicit left shift
  // is specified (0, 16, 32 or 48), the immediate must be a 16-bit value.
  //
  // For movk, an explicit shift can be used to indicate which half word should
  // be overwritten, eg. movk(x0, 0, 0) will overwrite the least-significant
  // half word with zero, whereas movk(x0, 0, 48) will overwrite the
  // most-significant.

  // Move immediate and keep.
  void movk(const Register& rd, uint64_t imm, int shift = -1) {
    MoveWide(rd, imm, shift, MOVK);
  }

  // Move inverted immediate.
  void movn(const Register& rd, uint64_t imm, int shift = -1) {
    MoveWide(rd, imm, shift, MOVN);
  }

  // Move immediate.
  void movz(const Register& rd, uint64_t imm, int shift = -1) {
    MoveWide(rd, imm, shift, MOVZ);
  }

  // Misc instructions.
  // Monitor debug-mode breakpoint.
  void brk(int code);

  // Halting debug-mode breakpoint.
  void hlt(int code);

  // Generate exception targeting EL1.
  void svc(int code);

  // Move register to register.
  void mov(const Register& rd, const Register& rn);

  // Move inverted operand to register.
  void mvn(const Register& rd, const Operand& operand);

  // System instructions.
  // Move to register from system register.
  void mrs(const Register& rt, SystemRegister sysreg);

  // Move from register to system register.
  void msr(SystemRegister sysreg, const Register& rt);

  // System instruction.
  void sys(int op1, int crn, int crm, int op2, const Register& rt = xzr);

  // System instruction with pre-encoded op (op1:crn:crm:op2).
  void sys(int op, const Register& rt = xzr);

  // System data cache operation.
  void dc(DataCacheOp op, const Register& rt);

  // System instruction cache operation.
  void ic(InstructionCacheOp op, const Register& rt);

  // System hint.
  void hint(SystemHint code);

  // Clear exclusive monitor.
  void clrex(int imm4 = 0xf);

  // Data memory barrier.
  void dmb(BarrierDomain domain, BarrierType type);

  // Data synchronization barrier.
  void dsb(BarrierDomain domain, BarrierType type);

  // Instruction synchronization barrier.
  void isb();

  // Alias for system instructions.
  // No-op.
  void nop() {
    hint(NOP);
  }

  // FP and NEON instructions.
  // Move double precision immediate to FP register.
  void fmov(const VRegister& vd, double imm);

  // Move single precision immediate to FP register.
  void fmov(const VRegister& vd, float imm);

  // Move FP register to register.
  void fmov(const Register& rd, const VRegister& fn);

  // Move register to FP register.
  void fmov(const VRegister& vd, const Register& rn);

  // Move FP register to FP register.
  void fmov(const VRegister& vd, const VRegister& fn);

  // Move 64-bit register to top half of 128-bit FP register.
  void fmov(const VRegister& vd, int index, const Register& rn);

  // Move top half of 128-bit FP register to 64-bit register.
  void fmov(const Register& rd, const VRegister& vn, int index);

  // FP add.
  void fadd(const VRegister& vd, const VRegister& vn, const VRegister& vm);

  // FP subtract.
  void fsub(const VRegister& vd, const VRegister& vn, const VRegister& vm);

  // FP multiply.
  void fmul(const VRegister& vd, const VRegister& vn, const VRegister& vm);

  // FP fused multiply-add.
  void fmadd(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             const VRegister& va);

  // FP fused multiply-subtract.
  void fmsub(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             const VRegister& va);

  // FP fused multiply-add and negate.
  void fnmadd(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              const VRegister& va);

  // FP fused multiply-subtract and negate.
  void fnmsub(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              const VRegister& va);

  // FP multiply-negate scalar.
  void fnmul(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // FP reciprocal exponent scalar.
  void frecpx(const VRegister& vd,
              const VRegister& vn);

  // FP divide.
  void fdiv(const VRegister& vd, const VRegister& fn, const VRegister& vm);

  // FP maximum.
  void fmax(const VRegister& vd, const VRegister& fn, const VRegister& vm);

  // FP minimum.
  void fmin(const VRegister& vd, const VRegister& fn, const VRegister& vm);

  // FP maximum number.
  void fmaxnm(const VRegister& vd, const VRegister& fn, const VRegister& vm);

  // FP minimum number.
  void fminnm(const VRegister& vd, const VRegister& fn, const VRegister& vm);

  // FP absolute.
  void fabs(const VRegister& vd, const VRegister& vn);

  // FP negate.
  void fneg(const VRegister& vd, const VRegister& vn);

  // FP square root.
  void fsqrt(const VRegister& vd, const VRegister& vn);

  // FP round to integer, nearest with ties to away.
  void frinta(const VRegister& vd, const VRegister& vn);

  // FP round to integer, implicit rounding.
  void frinti(const VRegister& vd, const VRegister& vn);

  // FP round to integer, toward minus infinity.
  void frintm(const VRegister& vd, const VRegister& vn);

  // FP round to integer, nearest with ties to even.
  void frintn(const VRegister& vd, const VRegister& vn);

  // FP round to integer, toward plus infinity.
  void frintp(const VRegister& vd, const VRegister& vn);

  // FP round to integer, exact, implicit rounding.
  void frintx(const VRegister& vd, const VRegister& vn);

  // FP round to integer, towards zero.
  void frintz(const VRegister& vd, const VRegister& vn);

  void FPCompareMacro(const VRegister& vn,
                      double value,
                      FPTrapFlags trap);

  void FPCompareMacro(const VRegister& vn,
                      const VRegister& vm,
                      FPTrapFlags trap);

  // FP compare registers.
  void fcmp(const VRegister& vn, const VRegister& vm);

  // FP compare immediate.
  void fcmp(const VRegister& vn, double value);

  void FPCCompareMacro(const VRegister& vn,
                       const VRegister& vm,
                       StatusFlags nzcv,
                       Condition cond,
                       FPTrapFlags trap);

  // FP conditional compare.
  void fccmp(const VRegister& vn,
             const VRegister& vm,
             StatusFlags nzcv,
             Condition cond);

  // FP signaling compare registers.
  void fcmpe(const VRegister& vn, const VRegister& vm);

  // FP signaling compare immediate.
  void fcmpe(const VRegister& vn, double value);

  // FP conditional signaling compare.
  void fccmpe(const VRegister& vn,
              const VRegister& vm,
              StatusFlags nzcv,
              Condition cond);

  // FP conditional select.
  void fcsel(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             Condition cond);

  // Common FP Convert functions.
  void NEONFPConvertToInt(const Register& rd,
                          const VRegister& vn,
                          Instr op);
  void NEONFPConvertToInt(const VRegister& vd,
                          const VRegister& vn,
                          Instr op);

  // FP convert between precisions.
  void fcvt(const VRegister& vd, const VRegister& vn);

  // FP convert to higher precision.
  void fcvtl(const VRegister& vd, const VRegister& vn);

  // FP convert to higher precision (second part).
  void fcvtl2(const VRegister& vd, const VRegister& vn);

  // FP convert to lower precision.
  void fcvtn(const VRegister& vd, const VRegister& vn);

  // FP convert to lower prevision (second part).
  void fcvtn2(const VRegister& vd, const VRegister& vn);

  // FP convert to lower precision, rounding to odd.
  void fcvtxn(const VRegister& vd, const VRegister& vn);

  // FP convert to lower precision, rounding to odd (second part).
  void fcvtxn2(const VRegister& vd, const VRegister& vn);

  // FP convert to signed integer, nearest with ties to away.
  void fcvtas(const Register& rd, const VRegister& vn);

  // FP convert to unsigned integer, nearest with ties to away.
  void fcvtau(const Register& rd, const VRegister& vn);

  // FP convert to signed integer, nearest with ties to away.
  void fcvtas(const VRegister& vd, const VRegister& vn);

  // FP convert to unsigned integer, nearest with ties to away.
  void fcvtau(const VRegister& vd, const VRegister& vn);

  // FP convert to signed integer, round towards -infinity.
  void fcvtms(const Register& rd, const VRegister& vn);

  // FP convert to unsigned integer, round towards -infinity.
  void fcvtmu(const Register& rd, const VRegister& vn);

  // FP convert to signed integer, round towards -infinity.
  void fcvtms(const VRegister& vd, const VRegister& vn);

  // FP convert to unsigned integer, round towards -infinity.
  void fcvtmu(const VRegister& vd, const VRegister& vn);

  // FP convert to signed integer, nearest with ties to even.
  void fcvtns(const Register& rd, const VRegister& vn);

  // FP convert to unsigned integer, nearest with ties to even.
  void fcvtnu(const Register& rd, const VRegister& vn);

  // FP convert to signed integer, nearest with ties to even.
  void fcvtns(const VRegister& rd, const VRegister& vn);

  // FP convert to unsigned integer, nearest with ties to even.
  void fcvtnu(const VRegister& rd, const VRegister& vn);

  // FP convert to signed integer or fixed-point, round towards zero.
  void fcvtzs(const Register& rd, const VRegister& vn, int fbits = 0);

  // FP convert to unsigned integer or fixed-point, round towards zero.
  void fcvtzu(const Register& rd, const VRegister& vn, int fbits = 0);

  // FP convert to signed integer or fixed-point, round towards zero.
  void fcvtzs(const VRegister& vd, const VRegister& vn, int fbits = 0);

  // FP convert to unsigned integer or fixed-point, round towards zero.
  void fcvtzu(const VRegister& vd, const VRegister& vn, int fbits = 0);

  // FP convert to signed integer, round towards +infinity.
  void fcvtps(const Register& rd, const VRegister& vn);

  // FP convert to unsigned integer, round towards +infinity.
  void fcvtpu(const Register& rd, const VRegister& vn);

  // FP convert to signed integer, round towards +infinity.
  void fcvtps(const VRegister& vd, const VRegister& vn);

  // FP convert to unsigned integer, round towards +infinity.
  void fcvtpu(const VRegister& vd, const VRegister& vn);

  // Convert signed integer or fixed point to FP.
  void scvtf(const VRegister& fd, const Register& rn, int fbits = 0);

  // Convert unsigned integer or fixed point to FP.
  void ucvtf(const VRegister& fd, const Register& rn, int fbits = 0);

  // Convert signed integer or fixed-point to FP.
  void scvtf(const VRegister& fd, const VRegister& vn, int fbits = 0);

  // Convert unsigned integer or fixed-point to FP.
  void ucvtf(const VRegister& fd, const VRegister& vn, int fbits = 0);

  // Unsigned absolute difference.
  void uabd(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Signed absolute difference.
  void sabd(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Unsigned absolute difference and accumulate.
  void uaba(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Signed absolute difference and accumulate.
  void saba(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Add.
  void add(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Subtract.
  void sub(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Unsigned halving add.
  void uhadd(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed halving add.
  void shadd(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Unsigned rounding halving add.
  void urhadd(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm);

  // Signed rounding halving add.
  void srhadd(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm);

  // Unsigned halving sub.
  void uhsub(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed halving sub.
  void shsub(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Unsigned saturating add.
  void uqadd(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed saturating add.
  void sqadd(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Unsigned saturating subtract.
  void uqsub(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed saturating subtract.
  void sqsub(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Add pairwise.
  void addp(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Add pair of elements scalar.
  void addp(const VRegister& vd,
            const VRegister& vn);

  // Multiply-add to accumulator.
  void mla(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Multiply-subtract to accumulator.
  void mls(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Multiply.
  void mul(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Multiply by scalar element.
  void mul(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm,
           int vm_index);

  // Multiply-add by scalar element.
  void mla(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm,
           int vm_index);

  // Multiply-subtract by scalar element.
  void mls(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm,
           int vm_index);

  // Signed long multiply-add by scalar element.
  void smlal(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Signed long multiply-add by scalar element (second part).
  void smlal2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Unsigned long multiply-add by scalar element.
  void umlal(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Unsigned long multiply-add by scalar element (second part).
  void umlal2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Signed long multiply-sub by scalar element.
  void smlsl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Signed long multiply-sub by scalar element (second part).
  void smlsl2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Unsigned long multiply-sub by scalar element.
  void umlsl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Unsigned long multiply-sub by scalar element (second part).
  void umlsl2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Signed long multiply by scalar element.
  void smull(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Signed long multiply by scalar element (second part).
  void smull2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Unsigned long multiply by scalar element.
  void umull(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm,
             int vm_index);

  // Unsigned long multiply by scalar element (second part).
  void umull2(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm,
              int vm_index);

  // Signed saturating double long multiply by element.
  void sqdmull(const VRegister& vd,
               const VRegister& vn,
               const VRegister& vm,
               int vm_index);

  // Signed saturating double long multiply by element (second part).
  void sqdmull2(const VRegister& vd,
                const VRegister& vn,
                const VRegister& vm,
                int vm_index);

  // Signed saturating doubling long multiply-add by element.
  void sqdmlal(const VRegister& vd,
               const VRegister& vn,
               const VRegister& vm,
               int vm_index);

  // Signed saturating doubling long multiply-add by element (second part).
  void sqdmlal2(const VRegister& vd,
                const VRegister& vn,
                const VRegister& vm,
                int vm_index);

  // Signed saturating doubling long multiply-sub by element.
  void sqdmlsl(const VRegister& vd,
               const VRegister& vn,
               const VRegister& vm,
               int vm_index);

  // Signed saturating doubling long multiply-sub by element (second part).
  void sqdmlsl2(const VRegister& vd,
                const VRegister& vn,
                const VRegister& vm,
                int vm_index);

  // Compare equal.
  void cmeq(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Compare signed greater than or equal.
  void cmge(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Compare signed greater than.
  void cmgt(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Compare unsigned higher.
  void cmhi(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Compare unsigned higher or same.
  void cmhs(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Compare bitwise test bits nonzero.
  void cmtst(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Compare bitwise to zero.
  void cmeq(const VRegister& vd,
            const VRegister& vn,
            int value);

  // Compare signed greater than or equal to zero.
  void cmge(const VRegister& vd,
            const VRegister& vn,
            int value);

  // Compare signed greater than zero.
  void cmgt(const VRegister& vd,
            const VRegister& vn,
            int value);

  // Compare signed less than or equal to zero.
  void cmle(const VRegister& vd,
            const VRegister& vn,
            int value);

  // Compare signed less than zero.
  void cmlt(const VRegister& vd,
            const VRegister& vn,
            int value);

  // Signed shift left by register.
  void sshl(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Unsigned shift left by register.
  void ushl(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Signed saturating shift left by register.
  void sqshl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Unsigned saturating shift left by register.
  void uqshl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed rounding shift left by register.
  void srshl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Unsigned rounding shift left by register.
  void urshl(const VRegister& vd,
             const VRegister& vn,
             const VRegister& vm);

  // Signed saturating rounding shift left by register.
  void sqrshl(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm);

  // Unsigned saturating rounding shift left by register.
  void uqrshl(const VRegister& vd,
              const VRegister& vn,
              const VRegister& vm);

  // Bitwise and.
  void and_(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Bitwise or.
  void orr(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bitwise or immediate.
  void orr(const VRegister& vd,
           const int imm8,
           const int left_shift = 0);

  // Move register to register.
  void mov(const VRegister& vd,
           const VRegister& vn);

  // Bitwise orn.
  void orn(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bitwise eor.
  void eor(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bit clear immediate.
  void bic(const VRegister& vd,
           const int imm8,
           const int left_shift = 0);

  // Bit clear.
  void bic(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bitwise insert if false.
  void bif(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bitwise insert if true.
  void bit(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Bitwise select.
  void bsl(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm);

  // Polynomial multiply.
  void pmul(const VRegister& vd,
            const VRegister& vn,
            const VRegister& vm);

  // Vector move immediate.
  void movi(const VRegister& vd,
            const uint64_t imm,
            Shift shift = LSL,
            const int shift_amount = 0);

  // Bitwise not.
  void mvn(const VRegister& vd,
           const VRegister& vn);

  // Vector move inverted immediate.
  void mvni(const VRegister& vd,
            const int imm8,
            Shift shift = LSL,
            const int shift_amount = 0);

  // Signed saturating accumulate of unsigned value.
  void suqadd(const VRegister& vd,
              const VRegister& vn);

  // Unsigned saturating accumulate of signed value.
  void usqadd(const VRegister& vd,
              const VRegister& vn);

  // Absolute value.
  void abs(const VRegister& vd,
           const VRegister& vn);

  // Signed saturating absolute value.
  void sqabs(const VRegister& vd,
             const VRegister& vn);

  // Negate.
  void neg(const VRegister& vd,
           const VRegister& vn);

  // Signed saturating negate.
  void sqneg(const VRegister& vd,
             const VRegister& vn);

  // Bitwise not.
  void not_(const VRegister& vd,
            const VRegister& vn);

  // Extract narrow.
  void xtn(const VRegister& vd,
           const VRegister& vn);

  // Extract narrow (second part).
  void xtn2(const VRegister& vd,
            const VRegister& vn);

  // Signed saturating extract narrow.
  void sqxtn(const VRegister& vd,
             const VRegister& vn);

  // Signed saturating extract narrow (second part).
  void sqxtn2(const VRegister& vd,
              const VRegister& vn);

  // Unsigned saturating extract narrow.
  void uqxtn(const VRegister& vd,
             const VRegister& vn);

  // Unsigned saturating extract narrow (second part).
  void uqxtn2(const VRegister& vd,
              const VRegister& vn);

  // Signed saturating extract unsigned narrow.
  void sqxtun(const VRegister& vd,
              const VRegister& vn);

  // Signed saturating extract unsigned narrow (second part).
  void sqxtun2(const VRegister& vd,
               const VRegister& vn);

  // Extract vector from pair of vectors.
  void ext(const VRegister& vd,
           const VRegister& vn,
           const VRegister& vm,
           int index);

  // Duplicate vector element to vector or scalar.
  void dup(const VRegister& vd,
           const VRegister& vn,
           int vn_index);

  // Move vector element to scalar.
  void mov(const VRegister& vd,
           const VRegister& vn,
           int vn_index);

  // Duplicate general-purpose register to vector.
  void dup(const VRegister& vd,
           const Register& rn);

  // Insert vector element from another vector element.
  void ins(const VRegister& vd,
           int vd_index,
           const VRegister& vn,
           int vn_index);

  // Move vector element to another vector element.
  void mov(const VRegister& vd,
           int vd_index,
           const VRegister& vn,
           int vn_index);

  // Insert vector element from general-purpose register.
  void ins(const VRegister& vd,
           int vd_index,
           const Register& rn);

  // Move general-purpose register to a vector element.
  void mov(const VRegister& vd,
           int vd_index,
           const Register& rn);

  // Unsigned move vector element to general-purpose register.
  void umov(const Register& rd,
            const VRegister& vn,
            int vn_index);

  // Move vector element to general-purpose register.
  void mov(const Register& rd,
           const VRegister& vn,
           int vn_index);

  // Signed move vector element to general-purpose register.
  void smov(const Register& rd,
            const VRegister& vn,
            int vn_index);

  // One-element structure load to one register.
  void ld1(const VRegister& vt,
           const MemOperand& src);

  // One-element structure load to two registers.
  void ld1(const VRegister& vt,
           const VRegister& vt2,
           const MemOperand& src);

  // One-element structure load to three registers.
  void ld1(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           const MemOperand& src);

  // One-element structure load to four registers.
  void ld1(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           const VRegister& vt4,
           const MemOperand& src);

  // One-element single structure load to one lane.
  void ld1(const VRegister& vt,
           int lane,
           const MemOperand& src);

  // One-element single structure load to all lanes.
  void ld1r(const VRegister& vt,
            const MemOperand& src);

  // Two-element structure load.
  void ld2(const VRegister& vt,
           const VRegister& vt2,
           const MemOperand& src);

  // Two-element single structure load to one lane.
  void ld2(const VRegister& vt,
           const VRegister& vt2,
           int lane,
           const MemOperand& src);

  // Two-element single structure load to all lanes.
  void ld2r(const VRegister& vt,
            const VRegister& vt2,
            const MemOperand& src);

  // Three-element structure load.
  void ld3(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           const MemOperand& src);

  // Three-element single structure load to one lane.
  void ld3(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           int lane,
           const MemOperand& src);

  // Three-element single structure load to all lanes.
  void ld3r(const VRegister& vt,
            const VRegister& vt2,
            const VRegister& vt3,
            const MemOperand& src);

  // Four-element structure load.
  void ld4(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           const VRegister& vt4,
           const MemOperand& src);

  // Four-element single structure load to one lane.
  void ld4(const VRegister& vt,
           const VRegister& vt2,
           const VRegister& vt3,
           const VRegister& vt4,
           int lane,
           const MemOperand& src);

  // Four-element single structure load to all lanes.
  void ld4r(const VRegister& vt,
            const VRegister& vt2,
            const VRegister& vt3,
            const VRegister& vt4,
            const MemOperand& src);

  // Count leading sign bits.
  void cls(const VRegister& vd,
           const VRegister& vn);

  // Count leading zero bits (vector).
  void clz(const VRegister& vd,
           const VRegister& vn);

  // Population count per byte.
  void cnt(const VRegister& vd,
           const VRegister& vn);

  // Reverse bit order.
  void rbit(const VRegister& vd,
            const VRegister& vn);

  // Reverse elements in 16-bit halfwords.
  void rev16(const VRegister& vd,
             const VRegister& vn);

  // Reverse elements in 32-bit words.
  void rev32(const VRegister& vd,
             const VRegister& vn);

  // Reverse elements in 64-bit doublewords.
  void rev64(const VRegister& vd,
             const VRegister& vn);

  // Unsigned reciprocal square root estimate.
  void ursqrte(const VRegister& vd,
               const VRegister& vn);

  // Unsigned reciprocal estimate.
  void urecpe(const VRegister& vd,
              const VRegister& vn);

  // Signed pairwise long add.
  void saddlp(const VRegister& vd,
              const VRegister& vn);

  // Unsigned pairwise long add.
  void uaddlp(const VRegister& vd,
              const VRegister& vn);

  // Signed pairwise long add and accumulate.