disas/libvixl: Update to upstream VIXL 1.12

Update our copy of libvixl to upstream's 1.12 release.
The major benefit from QEMU's point of view is that some instructions
previously disassembled as "unimplemented (System)" are now displayed
as something more useful. It also fixes some warnings about format
strings that newer w64-mingw32 compilers were emitting.

We didn't have any local changes to libvixl so nothing needed
to be forward-ported.

Although this is a large commit (due to upstream renaming most
of the files), only a few of the files changed in this commit
are not just straight copies of upstream libvixl files:
 disas/arm-a64.cc
 disas/libvixl/Makefile.objs
 disas/libvixl/README

Note that this commit introduces some signed-unsigned comparison
warnings on the old mingw compilers. Those compilers have broken
TLS support anyway so have only ever been much use for compile tests;
anybody still using them should add -Wno-sign-compare to their
--extra-cflags.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

24 files changed, 11989 insertions, 5188 deletions

diff --git a/disas/arm-a64.cc b/disas/arm-a64.cc
index b57256b267..d4d46d5ff3 100644
--- a/disas/arm-a64.cc
+++ b/disas/arm-a64.cc
@@ -17,7 +17,7 @@
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "a64/disasm-a64.h"
+#include "vixl/a64/disasm-a64.h"
 
 extern "C" {
 #include "disas/bfd.h"
diff --git a/disas/libvixl/Makefile.objs b/disas/libvixl/Makefile.objs
index 17e6565d10..e373cf02d3 100644
--- a/disas/libvixl/Makefile.objs
+++ b/disas/libvixl/Makefile.objs
@@ -1,7 +1,8 @@
-libvixl_OBJS = utils.o \
-               a64/instructions-a64.o \
-               a64/decoder-a64.o \
-               a64/disasm-a64.o
+libvixl_OBJS = vixl/utils.o \
+               vixl/compiler-intrinsics.o \
+               vixl/a64/instructions-a64.o \
+               vixl/a64/decoder-a64.o \
+               vixl/a64/disasm-a64.o
 
 $(addprefix $(obj)/,$(libvixl_OBJS)): QEMU_CFLAGS := -I$(SRC_PATH)/disas/libvixl $(QEMU_CFLAGS)
 
diff --git a/disas/libvixl/README b/disas/libvixl/README
index 58db41c67c..932a41adf7 100644
--- a/disas/libvixl/README
+++ b/disas/libvixl/README
@@ -2,11 +2,10 @@
 The code in this directory is a subset of libvixl:
  https://github.com/armvixl/vixl
 (specifically, it is the set of files needed for disassembly only,
-taken from libvixl 1.7).
+taken from libvixl 1.12).
 Bugfixes should preferably be sent upstream initially.
 
 The disassembler does not currently support the entire A64 instruction
 set. Notably:
- * No Advanced SIMD support.
  * Limited support for system instructions.
  * A few miscellaneous integer and floating point instructions are missing.
diff --git a/disas/libvixl/a64/assembler-a64.h b/disas/libvixl/a64/assembler-a64.h
deleted file mode 100644
index 35aaf20f72..0000000000
--- a/disas/libvixl/a64/assembler-a64.h
+++ /dev/null
@@ -1,2353 +0,0 @@
-// Copyright 2013, 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 <list>
-#include <stack>
-
-#include "globals.h"
-#include "utils.h"
-#include "code-buffer.h"
-#include "a64/instructions-a64.h"
-
-namespace vixl {
-
-typedef uint64_t RegList;
-static const int kRegListSizeInBits = sizeof(RegList) * 8;
-
-
-// Registers.
-
-// Some CPURegister methods can return Register and FPRegister types, so we
-// need to declare them in advance.
-class Register;
-class FPRegister;
-
-
-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,
-    kFPRegister,
-    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 Is32Bits() const {
-    VIXL_ASSERT(IsValid());
-    return size_ == 32;
-  }
-
-  bool Is64Bits() const {
-    VIXL_ASSERT(IsValid());
-    return size_ == 64;
-  }
-
-  bool IsValid() const {
-    if (IsValidRegister() || IsValidFPRegister()) {
-      VIXL_ASSERT(!IsNone());
-      return true;
-    } else {
-      VIXL_ASSERT(IsNone());
-      return false;
-    }
-  }
-
-  bool IsValidRegister() const {
-    return IsRegister() &&
-           ((size_ == kWRegSize) || (size_ == kXRegSize)) &&
-           ((code_ < kNumberOfRegisters) || (code_ == kSPRegInternalCode));
-  }
-
-  bool IsValidFPRegister() const {
-    return IsFPRegister() &&
-           ((size_ == kSRegSize) || (size_ == kDRegSize)) &&
-           (code_ < kNumberOfFPRegisters);
-  }
-
-  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 IsFPRegister() const {
-    return type_ == kFPRegister;
-  }
-
-  bool IsW() const { return IsValidRegister() && Is32Bits(); }
-  bool IsX() const { return IsValidRegister() && Is64Bits(); }
-  bool IsS() const { return IsValidFPRegister() && Is32Bits(); }
-  bool IsD() const { return IsValidFPRegister() && Is64Bits(); }
-
-  const Register& W() const;
-  const Register& X() const;
-  const FPRegister& S() const;
-  const FPRegister& D() 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 FPRegister : public CPURegister {
- public:
-  FPRegister() : CPURegister() {}
-  explicit FPRegister(const CPURegister& other)
-      : CPURegister(other.code(), other.size(), other.type()) {
-    VIXL_ASSERT(IsValidFPRegister());
-  }
-  FPRegister(unsigned code, unsigned size)
-      : CPURegister(code, size, kFPRegister) {}
-
-  bool IsValid() const {
-    VIXL_ASSERT(IsFPRegister() || IsNone());
-    return IsValidFPRegister();
-  }
-
-  static const FPRegister& SRegFromCode(unsigned code);
-  static const FPRegister& DRegFromCode(unsigned code);
-
- private:
-  static const FPRegister sregisters[];
-  static const FPRegister dregisters[];
-};
-
-
-// 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 FPRegister
-// variants are provided for convenience.
-const Register NoReg;
-const FPRegister NoFPReg;
-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_FPREGISTERS(N)  \
-const FPRegister s##N(N, kSRegSize);  \
-const FPRegister d##N(N, kDRegSize);
-REGISTER_CODE_LIST(DEFINE_FPREGISTERS)
-#undef DEFINE_FPREGISTERS
-
-
-// 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);
-
-
-// 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::kFPRegister) &&
-                 (last_reg < kNumberOfFPRegisters)));
-    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);
-
-  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 GetCalleeSavedFP(unsigned size = kDRegSize);
-
-  // AAPCS64 caller-saved registers. Note that this includes lr.
-  static CPURegList GetCallerSaved(unsigned size = kXRegSize);
-  static CPURegList GetCallerSavedFP(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_, kRegListSizeInBits);
-  }
-
-  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 kCalleeSavedFP;
-
-
-// AAPCS64 caller-saved registers. Note that this includes lr.
-extern const CPURegList kCallerSaved;
-extern const CPURegList kCallerSavedFP;
-
-
-// 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);           // 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);
-  explicit MemOperand(Register base,
-                      Register regoffset,
-                      Shift shift = LSL,
-                      unsigned shift_amount = 0);
-  explicit MemOperand(Register base,
-                      Register regoffset,
-                      Extend extend,
-                      unsigned shift_amount = 0);
-  explicit 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;
-
- private:
-  Register base_;
-  Register regoffset_;
-  int64_t offset_;
-  AddrMode addrmode_;
-  Shift shift_;
-  Extend extend_;
-  unsigned shift_amount_;
-};
-
-
-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_; }
-
- private:
-  // The list of linked instructions is stored in a stack-like structure. We
-  // don't use std::stack directly because it's slow for the common case where
-  // only one or two instructions refer to a label, and labels themselves are
-  // short-lived. This class behaves like std::stack, but the first few links
-  // are preallocated (configured by kPreallocatedLinks).
-  //
-  // If more than N links are required, this falls back to std::stack.
-  class LinksStack {
-   public:
-    LinksStack() : size_(0), links_extended_(NULL) {}
-    ~LinksStack() {
-      delete links_extended_;
-    }
-
-    size_t size() const {
-      return size_;
-    }
-
-    bool empty() const {
-      return size_ == 0;
-    }
-
-    void push(ptrdiff_t value) {
-      if (size_ < kPreallocatedLinks) {
-        links_[size_] = value;
-      } else {
-        if (links_extended_ == NULL) {
-          links_extended_ = new std::stack<ptrdiff_t>();
-        }
-        VIXL_ASSERT(size_ == (links_extended_->size() + kPreallocatedLinks));
-        links_extended_->push(value);
-      }
-      size_++;
-    }
-
-    ptrdiff_t top() const {
-      return (size_ <= kPreallocatedLinks) ? links_[size_ - 1]
-                                           : links_extended_->top();
-    }
-
-    void pop() {
-      size_--;
-      if (size_ >= kPreallocatedLinks) {
-        links_extended_->pop();
-        VIXL_ASSERT(size_ == (links_extended_->size() + kPreallocatedLinks));
-      }
-    }
-
-   private:
-    static const size_t kPreallocatedLinks = 4;
-
-    size_t size_;
-    ptrdiff_t links_[kPreallocatedLinks];
-    std::stack<ptrdiff_t> * links_extended_;
-  };
-
-  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_.push(instruction);
-  }
-
-  ptrdiff_t GetAndRemoveNextLink() {
-    VIXL_ASSERT(IsLinked());
-    ptrdiff_t link = links_.top();
-    links_.pop();
-    return link;
-  }
-
-  // The offsets of the instructions that have linked to this label.
-  LinksStack 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 by declaring
-  // it private (without an implementation).
-  Label(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;
-};
-
-
-// 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:
-  RawLiteral() : size_(0), offset_(0), raw_value_(0) {}
-
-  size_t size() {
-    VIXL_STATIC_ASSERT(kDRegSizeInBytes == kXRegSizeInBytes);
-    VIXL_STATIC_ASSERT(kSRegSizeInBytes == kWRegSizeInBytes);
-    VIXL_ASSERT((size_ == kXRegSizeInBytes) || (size_ == kWRegSizeInBytes));
-    return size_;
-  }
-  uint64_t raw_value64() {
-    VIXL_ASSERT(size_ == kXRegSizeInBytes);
-    return raw_value_;
-  }
-  uint32_t raw_value32() {
-    VIXL_ASSERT(size_ == kWRegSizeInBytes);
-    VIXL_ASSERT(is_uint32(raw_value_) || is_int32(raw_value_));
-    return static_cast<uint32_t>(raw_value_);
-  }
-  bool IsUsed() { return offset_ < 0; }
-  bool IsPlaced() { return offset_ > 0; }
-
- protected:
-  ptrdiff_t offset() {
-    VIXL_ASSERT(IsPlaced());
-    return offset_ - 1;
-  }
-  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 raw_value_;
-
-  friend class Assembler;
-};
-
-
-template <typename T>
-class Literal : public RawLiteral {
- public:
-  explicit Literal(T value) {
-    size_ = sizeof(value);
-    memcpy(&raw_value_, &value, sizeof(value));
-  }
-};
-
-
-// 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) {
-    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) {
-    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() {
-    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() {
-    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
-    return GetOffsetAddress<T>(0);
-  }
-
-  // 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);
-
-  // 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);
-
-  // 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);
-
-  // 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);
-
-  // 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 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 instructions.
-  // Move double precision immediate to FP register.
-  void fmov(const FPRegister& fd, double imm);
-
-  // Move single precision immediate to FP register.
-  void fmov(const FPRegister& fd, float imm);
-
-  // Move FP register to register.
-  void fmov(const Register& rd, const FPRegister& fn);
-
-  // Move register to FP register.
-  void fmov(const FPRegister& fd, const Register& rn);
-
-  // Move FP register to FP register.
-  void fmov(const FPRegister& fd, const FPRegister& fn);
-
-  // FP add.
-  void fadd(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP subtract.
-  void fsub(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP multiply.
-  void fmul(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP fused multiply and add.
-  void fmadd(const FPRegister& fd,
-             const FPRegister& fn,
-             const FPRegister& fm,
-             const FPRegister& fa);
-
-  // FP fused multiply and subtract.
-  void fmsub(const FPRegister& fd,
-             const FPRegister& fn,
-             const FPRegister& fm,
-             const FPRegister& fa);
-
-  // FP fused multiply, add and negate.
-  void fnmadd(const FPRegister& fd,
-              const FPRegister& fn,
-              const FPRegister& fm,
-              const FPRegister& fa);
-
-  // FP fused multiply, subtract and negate.
-  void fnmsub(const FPRegister& fd,
-              const FPRegister& fn,
-              const FPRegister& fm,
-              const FPRegister& fa);
-
-  // FP divide.
-  void fdiv(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP maximum.
-  void fmax(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP minimum.
-  void fmin(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP maximum number.
-  void fmaxnm(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP minimum number.
-  void fminnm(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm);
-
-  // FP absolute.
-  void fabs(const FPRegister& fd, const FPRegister& fn);
-
-  // FP negate.
-  void fneg(const FPRegister& fd, const FPRegister& fn);
-
-  // FP square root.
-  void fsqrt(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (nearest with ties to away).
-  void frinta(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (implicit rounding).
-  void frinti(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (toward minus infinity).
-  void frintm(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (nearest with ties to even).
-  void frintn(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (toward plus infinity).
-  void frintp(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (exact, implicit rounding).
-  void frintx(const FPRegister& fd, const FPRegister& fn);
-
-  // FP round to integer (towards zero).
-  void frintz(const FPRegister& fd, const FPRegister& fn);
-
-  // FP compare registers.
-  void fcmp(const FPRegister& fn, const FPRegister& fm);
-
-  // FP compare immediate.
-  void fcmp(const FPRegister& fn, double value);
-
-  // FP conditional compare.
-  void fccmp(const FPRegister& fn,
-             const FPRegister& fm,
-             StatusFlags nzcv,
-             Condition cond);
-
-  // FP conditional select.
-  void fcsel(const FPRegister& fd,
-             const FPRegister& fn,
-             const FPRegister& fm,
-             Condition cond);
-
-  // Common FP Convert function.
-  void FPConvertToInt(const Register& rd,
-                      const FPRegister& fn,
-                      FPIntegerConvertOp op);
-
-  // FP convert between single and double precision.
-  void fcvt(const FPRegister& fd, const FPRegister& fn);
-
-  // Convert FP to signed integer (nearest with ties to away).
-  void fcvtas(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to unsigned integer (nearest with ties to away).
-  void fcvtau(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to signed integer (round towards -infinity).
-  void fcvtms(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to unsigned integer (round towards -infinity).
-  void fcvtmu(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to signed integer (nearest with ties to even).
-  void fcvtns(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to unsigned integer (nearest with ties to even).
-  void fcvtnu(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to signed integer (round towards zero).
-  void fcvtzs(const Register& rd, const FPRegister& fn);
-
-  // Convert FP to unsigned integer (round towards zero).
-  void fcvtzu(const Register& rd, const FPRegister& fn);
-
-  // Convert signed integer or fixed point to FP.
-  void scvtf(const FPRegister& fd, const Register& rn, unsigned fbits = 0);
-
-  // Convert unsigned integer or fixed point to FP.
-  void ucvtf(const FPRegister& fd, const Register& rn, unsigned fbits = 0);
-
-  // Emit generic instructions.
-  // Emit raw instructions into the instruction stream.
-  void dci(Instr raw_inst) { Emit(raw_inst); }
-
-  // Emit 32 bits of data into the instruction stream.
-  void dc32(uint32_t data) {
-    VIXL_ASSERT(buffer_monitor_ > 0);
-    buffer_->Emit32(data);
-  }
-
-  // Emit 64 bits of data into the instruction stream.
-  void dc64(uint64_t data) {
-    VIXL_ASSERT(buffer_monitor_ > 0);
-    buffer_->Emit64(data);
-  }
-
-  // Copy a string into the instruction stream, including the terminating NULL
-  // character. The instruction pointer is then aligned correctly for
-  // subsequent instructions.
-  void EmitString(const char * string) {
-    VIXL_ASSERT(string != NULL);
-    VIXL_ASSERT(buffer_monitor_ > 0);
-
-    buffer_->EmitString(string);
-    buffer_->Align();
-  }
-
-  // Code generation helpers.
-
-  // Register encoding.
-  static Instr Rd(CPURegister rd) {
-    VIXL_ASSERT(rd.code() != kSPRegInternalCode);
-    return rd.code() << Rd_offset;
-  }
-
-  static Instr Rn(CPURegister rn) {
-    VIXL_ASSERT(rn.code() != kSPRegInternalCode);
-    return rn.code() << Rn_offset;
-  }
-
-  static Instr Rm(CPURegister rm) {
-    VIXL_ASSERT(rm.code() != kSPRegInternalCode);
-    return rm.code() << Rm_offset;
-  }
-
-  static Instr Ra(CPURegister ra) {
-    VIXL_ASSERT(ra.code() != kSPRegInternalCode);
-    return ra.code() << Ra_offset;
-  }
-
-  static Instr Rt(CPURegister rt) {
-    VIXL_ASSERT(rt.code() != kSPRegInternalCode);
-    return rt.code() << Rt_offset;
-  }
-
-  static Instr Rt2(CPURegister rt2) {
-    VIXL_ASSERT(rt2.code() != kSPRegInternalCode);
-    return rt2.code() << Rt2_offset;
-  }
-
-  static Instr Rs(CPURegister rs) {
-    VIXL_ASSERT(rs.code() != kSPRegInternalCode);
-    return rs.code() << Rs_offset;
-  }
-
-  // These encoding functions allow the stack pointer to be encoded, and
-  // disallow the zero register.
-  static Instr RdSP(Register rd) {
-    VIXL_ASSERT(!rd.IsZero());
-    return (rd.code() & kRegCodeMask) << Rd_offset;
-  }
-
-  static Instr RnSP(Register rn) {
-    VIXL_ASSERT(!rn.IsZero());
-    return (rn.code() & kRegCodeMask) << Rn_offset;
-  }
-
-  // Flags encoding.
-  static Instr Flags(FlagsUpdate S) {
-    if (S == SetFlags) {
-      return 1 << FlagsUpdate_offset;
-    } else if (S == LeaveFlags) {
-      return 0 << FlagsUpdate_offset;
-    }
-    VIXL_UNREACHABLE();
-    return 0;
-  }
-
-  static Instr Cond(Condition cond) {
-    return cond << Condition_offset;
-  }
-
-  // PC-relative address encoding.
-  static Instr ImmPCRelAddress(int imm21) {
-    VIXL_ASSERT(is_int21(imm21));
-    Instr imm = static_cast<Instr>(truncate_to_int21(imm21));
-    Instr immhi = (imm >> ImmPCRelLo_width) << ImmPCRelHi_offset;
-    Instr immlo = imm << ImmPCRelLo_offset;
-    return (immhi & ImmPCRelHi_mask) | (immlo & ImmPCRelLo_mask);
-  }
-
-  // Branch encoding.
-  static Instr ImmUncondBranch(int imm26) {
-    VIXL_ASSERT(is_int26(imm26));
-    return truncate_to_int26(imm26) << ImmUncondBranch_offset;
-  }
-
-  static Instr ImmCondBranch(int imm19) {
-    VIXL_ASSERT(is_int19(imm19));
-    return truncate_to_int19(imm19) << ImmCondBranch_offset;
-  }
-
-  static Instr ImmCmpBranch(int imm19) {
-    VIXL_ASSERT(is_int19(imm19));
-    return truncate_to_int19(imm19) << ImmCmpBranch_offset;
-  }
-
-  static Instr ImmTestBranch(int imm14) {
-    VIXL_ASSERT(is_int14(imm14));
-    return truncate_to_int14(imm14) << ImmTestBranch_offset;
-  }
-
-  static Instr ImmTestBranchBit(unsigned bit_pos) {
-    VIXL_ASSERT(is_uint6(bit_pos));
-    // Subtract five from the shift offset, as we need bit 5 from bit_pos.
-    unsigned b5 = bit_pos << (ImmTestBranchBit5_offset - 5);
-    unsigned b40 = bit_pos << ImmTestBranchBit40_offset;
-    b5 &= ImmTestBranchBit5_mask;
-    b40 &= ImmTestBranchBit40_mask;
-    return b5 | b40;
-  }
-
-  // Data Processing encoding.
-  static Instr SF(Register rd) {
-      return rd.Is64Bits() ? SixtyFourBits : ThirtyTwoBits;
-  }
-
-  static Instr ImmAddSub(int64_t imm) {
-    VIXL_ASSERT(IsImmAddSub(imm));
-    if (is_uint12(imm)) {  // No shift required.
-      return imm << ImmAddSub_offset;
-    } else {
-      return ((imm >> 12) << ImmAddSub_offset) | (1 << ShiftAddSub_offset);
-    }
-  }
-
-  static Instr ImmS(unsigned imms, unsigned reg_size) {
-    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(imms)) ||
-           ((reg_size == kWRegSize) && is_uint5(imms)));
-    USE(reg_size);
-    return imms << ImmS_offset;
-  }
-
-  static Instr ImmR(unsigned immr, unsigned reg_size) {
-    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(immr)) ||
-           ((reg_size == kWRegSize) && is_uint5(immr)));
-    USE(reg_size);
-    VIXL_ASSERT(is_uint6(immr));
-    return immr << ImmR_offset;
-  }
-
-  static Instr ImmSetBits(unsigned imms, unsigned reg_size) {
-    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
-    VIXL_ASSERT(is_uint6(imms));
-    VIXL_ASSERT((reg_size == kXRegSize) || is_uint6(imms + 3));
-    USE(reg_size);
-    return imms << ImmSetBits_offset;
-  }
-
-  static Instr ImmRotate(unsigned immr, unsigned reg_size) {
-    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
-    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(immr)) ||
-           ((reg_size == kWRegSize) && is_uint5(immr)));
-    USE(reg_size);
-    return immr << ImmRotate_offset;
-  }
-
-  static Instr ImmLLiteral(int imm19) {
-    VIXL_ASSERT(is_int19(imm19));
-    return truncate_to_int19(imm19) << ImmLLiteral_offset;
-  }
-
-  static Instr BitN(unsigned bitn, unsigned reg_size) {
-    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
-    VIXL_ASSERT((reg_size == kXRegSize) || (bitn == 0));
-    USE(reg_size);
-    return bitn << BitN_offset;
-  }
-
-  static Instr ShiftDP(Shift shift) {
-    VIXL_ASSERT(shift == LSL || shift == LSR || shift == ASR || shift == ROR);
-    return shift << ShiftDP_offset;
-  }
-
-  static Instr ImmDPShift(unsigned amount) {
-    VIXL_ASSERT(is_uint6(amount));
-    return amount << ImmDPShift_offset;
-  }
-
-  static Instr ExtendMode(Extend extend) {
-    return extend << ExtendMode_offset;
-  }
-
-  static Instr ImmExtendShift(unsigned left_shift) {
-    VIXL_ASSERT(left_shift <= 4);
-    return left_shift << ImmExtendShift_offset;
-  }
-
-  static Instr ImmCondCmp(unsigned imm) {
-    VIXL_ASSERT(is_uint5(imm));
-    return imm << ImmCondCmp_offset;
-  }
-
-  static Instr Nzcv(StatusFlags nzcv) {
-    return ((nzcv >> Flags_offset) & 0xf) << Nzcv_offset;
-  }
-
-  // MemOperand offset encoding.
-  static Instr ImmLSUnsigned(int imm12) {
-    VIXL_ASSERT(is_uint12(imm12));
-    return imm12 << ImmLSUnsigned_offset;
-  }
-
-  static Instr ImmLS(int imm9) {
-    VIXL_ASSERT(is_int9(imm9));
-    return truncate_to_int9(imm9) << ImmLS_offset;
-  }
-
-  static Instr ImmLSPair(int imm7, LSDataSize size) {
-    VIXL_ASSERT(((imm7 >> size) << size) == imm7);
-    int scaled_imm7 = imm7 >> size;
-    VIXL_ASSERT(is_int7(scaled_imm7));
-    return truncate_to_int7(scaled_imm7) << ImmLSPair_offset;
-  }
-
-  static Instr ImmShiftLS(unsigned shift_amount) {
-    VIXL_ASSERT(is_uint1(shift_amount));
-    return shift_amount << ImmShiftLS_offset;
-  }
-
-  static Instr ImmPrefetchOperation(int imm5) {
-    VIXL_ASSERT(is_uint5(imm5));
-    return imm5 << ImmPrefetchOperation_offset;
-  }
-
-  static Instr ImmException(int imm16) {
-    VIXL_ASSERT(is_uint16(imm16));
-    return imm16 << ImmException_offset;
-  }
-
-  static Instr ImmSystemRegister(int imm15) {
-    VIXL_ASSERT(is_uint15(imm15));
-    return imm15 << ImmSystemRegister_offset;
-  }
-
-  static Instr ImmHint(int imm7) {
-    VIXL_ASSERT(is_uint7(imm7));
-    return imm7 << ImmHint_offset;
-  }
-
-  static Instr CRm(int imm4) {
-    VIXL_ASSERT(is_uint4(imm4));
-    return imm4 << CRm_offset;
-  }
-
-  static Instr ImmBarrierDomain(int imm2) {
-    VIXL_ASSERT(is_uint2(imm2));
-    return imm2 << ImmBarrierDomain_offset;
-  }
-
-  static Instr ImmBarrierType(int imm2) {
-    VIXL_ASSERT(is_uint2(imm2));
-    return imm2 << ImmBarrierType_offset;
-  }
-
-  static LSDataSize CalcLSDataSize(LoadStoreOp op) {
-    VIXL_ASSERT((SizeLS_offset + SizeLS_width) == (kInstructionSize * 8));
-    return static_cast<LSDataSize>(op >> SizeLS_offset);
-  }
-
-  // Move immediates encoding.
-  static Instr ImmMoveWide(uint64_t imm) {
-    VIXL_ASSERT(is_uint16(imm));
-    return imm << ImmMoveWide_offset;
-  }
-
-  static Instr ShiftMoveWide(int64_t shift) {
-    VIXL_ASSERT(is_uint2(shift));
-    return shift << ShiftMoveWide_offset;
-  }
-
-  // FP Immediates.
-  static Instr ImmFP32(float imm);
-  static Instr ImmFP64(double imm);
-
-  // FP register type.
-  static Instr FPType(FPRegister fd) {
-    return fd.Is64Bits() ? FP64 : FP32;
-  }
-
-  static Instr FPScale(unsigned scale) {
-    VIXL_ASSERT(is_uint6(scale));
-    return scale << FPScale_offset;
-  }
-
-  // Immediate field checking helpers.
-  static bool IsImmAddSub(int64_t immediate);
-  static bool IsImmConditionalCompare(int64_t immediate);
-  static bool IsImmFP32(float imm);
-  static bool IsImmFP64(double imm);
-  static bool IsImmLogical(uint64_t value,
-                           unsigned width,
-                           unsigned* n = NULL,
-                           unsigned* imm_s = NULL,
-                           unsigned* imm_r = NULL);
-  static bool IsImmLSPair(int64_t offset, LSDataSize size);
-  static bool IsImmLSScaled(int64_t offset, LSDataSize size);
-  static bool IsImmLSUnscaled(int64_t offset);
-  static bool IsImmMovn(uint64_t imm, unsigned reg_size);
-  static bool IsImmMovz(uint64_t imm, unsigned reg_size);
-
-  // Size of the code generated since label to the current position.
-  size_t SizeOfCodeGeneratedSince(Label* label) const {
-    VIXL_ASSERT(label->IsBound());
-    return buffer_->OffsetFrom(label->location());
-  }
-
-  size_t SizeOfCodeGenerated() const {
-    return buffer_->CursorOffset();
-  }
-
-  size_t BufferCapacity() const { return buffer_->capacity(); }
-
-  size_t RemainingBufferSpace() const { return buffer_->RemainingBytes(); }
-
-  void EnsureSpaceFor(size_t amount) {
-    if (buffer_->RemainingBytes() < amount) {
-      size_t capacity = buffer_->capacity();
-      size_t size = buffer_->CursorOffset();
-      do {
-        // TODO(all): refine.
-        capacity *= 2;
-      } while ((capacity - size) <  amount);
-      buffer_->Grow(capacity);
-    }
-  }
-
-#ifdef VIXL_DEBUG
-  void AcquireBuffer() {
-    VIXL_ASSERT(buffer_monitor_ >= 0);
-    buffer_monitor_++;
-  }
-
-  void ReleaseBuffer() {
-    buffer_monitor_--;
-    VIXL_ASSERT(buffer_monitor_ >= 0);
-  }
-#endif
-
-  PositionIndependentCodeOption pic() const {
-    return pic_;
-  }
-
-  bool AllowPageOffsetDependentCode() const {
-    return (pic() == PageOffsetDependentCode) ||
-           (pic() == PositionDependentCode);
-  }
-
-  static const Register& AppropriateZeroRegFor(const CPURegister& reg) {
-    return reg.Is64Bits() ? xzr : wzr;
-  }
-
-
- protected:
-  void LoadStore(const CPURegister& rt,
-                 const MemOperand& addr,
-                 LoadStoreOp op,
-                 LoadStoreScalingOption option = PreferScaledOffset);
-
-  void LoadStorePair(const CPURegister& rt,
-                     const CPURegister& rt2,
-                     const MemOperand& addr,
-                     LoadStorePairOp op);
-
-  void Prefetch(PrefetchOperation op,
-                const MemOperand& addr,
-                LoadStoreScalingOption option = PreferScaledOffset);
-
-  // TODO(all): The third parameter should be passed by reference but gcc 4.8.2
-  // reports a bogus uninitialised warning then.
-  void Logical(const Register& rd,
-               const Register& rn,
-               const Operand operand,
-               LogicalOp op);
-  void LogicalImmediate(const Register& rd,
-                        const Register& rn,
-                        unsigned n,
-                        unsigned imm_s,
-                        unsigned imm_r,
-                        LogicalOp op);
-
-  void ConditionalCompare(const Register& rn,
-                          const Operand& operand,
-                          StatusFlags nzcv,
-                          Condition cond,
-                          ConditionalCompareOp op);
-
-  void AddSubWithCarry(const Register& rd,
-                       const Register& rn,
-                       const Operand& operand,
-                       FlagsUpdate S,
-                       AddSubWithCarryOp op);
-
-
-  // Functions for emulating operands not directly supported by the instruction
-  // set.
-  void EmitShift(const Register& rd,
-                 const Register& rn,
-                 Shift shift,
-                 unsigned amount);
-  void EmitExtendShift(const Register& rd,
-                       const Register& rn,
-                       Extend extend,
-                       unsigned left_shift);
-
-  void AddSub(const Register& rd,
-              const Register& rn,
-              const Operand& operand,
-              FlagsUpdate S,
-              AddSubOp op);
-
-  // Find an appropriate LoadStoreOp or LoadStorePairOp for the specified
-  // registers. Only simple loads are supported; sign- and zero-extension (such
-  // as in LDPSW_x or LDRB_w) are not supported.
-  static LoadStoreOp LoadOpFor(const CPURegister& rt);
-  static LoadStorePairOp LoadPairOpFor(const CPURegister& rt,
-                                       const CPURegister& rt2);
-  static LoadStoreOp StoreOpFor(const CPURegister& rt);
-  static LoadStorePairOp StorePairOpFor(const CPURegister& rt,
-                                        const CPURegister& rt2);
-  static LoadStorePairNonTemporalOp LoadPairNonTemporalOpFor(
-    const CPURegister& rt, const CPURegister& rt2);
-  static LoadStorePairNonTemporalOp StorePairNonTemporalOpFor(
-    const CPURegister& rt, const CPURegister& rt2);
-  static LoadLiteralOp LoadLiteralOpFor(const CPURegister& rt);
-
-
- private:
-  // Instruction helpers.
-  void MoveWide(const Register& rd,
-                uint64_t imm,
-                int shift,
-                MoveWideImmediateOp mov_op);
-  void DataProcShiftedRegister(const Register& rd,
-                               const Register& rn,
-                               const Operand& operand,
-                               FlagsUpdate S,
-                               Instr op);
-  void DataProcExtendedRegister(const Register& rd,
-                                const Register& rn,
-                                const Operand& operand,
-                                FlagsUpdate S,
-                                Instr op);
-  void LoadStorePairNonTemporal(const CPURegister& rt,
-                                const CPURegister& rt2,
-                                const MemOperand& addr,
-                                LoadStorePairNonTemporalOp op);
-  void LoadLiteral(const CPURegister& rt, uint64_t imm, LoadLiteralOp op);
-  void ConditionalSelect(const Register& rd,
-                         const Register& rn,
-                         const Register& rm,
-                         Condition cond,
-                         ConditionalSelectOp op);
-  void DataProcessing1Source(const Register& rd,
-                             const Register& rn,
-                             DataProcessing1SourceOp op);
-  void DataProcessing3Source(const Register& rd,
-                             const Register& rn,
-                             const Register& rm,
-                             const Register& ra,
-                             DataProcessing3SourceOp op);
-  void FPDataProcessing1Source(const FPRegister& fd,
-                               const FPRegister& fn,
-                               FPDataProcessing1SourceOp op);
-  void FPDataProcessing2Source(const FPRegister& fd,
-                               const FPRegister& fn,
-                               const FPRegister& fm,
-                               FPDataProcessing2SourceOp op);
-  void FPDataProcessing3Source(const FPRegister& fd,
-                               const FPRegister& fn,
-                               const FPRegister& fm,
-                               const FPRegister& fa,
-                               FPDataProcessing3SourceOp op);
-
-  // Encode the specified MemOperand for the specified access size and scaling
-  // preference.
-  Instr LoadStoreMemOperand(const MemOperand& addr,
-                            LSDataSize size,
-                            LoadStoreScalingOption option);
-
-  // Link the current (not-yet-emitted) instruction to the specified label, then
-  // return an offset to be encoded in the instruction. If the label is not yet
-  // bound, an offset of 0 is returned.
-  ptrdiff_t LinkAndGetByteOffsetTo(Label * label);
-  ptrdiff_t LinkAndGetInstructionOffsetTo(Label * label);
-  ptrdiff_t LinkAndGetPageOffsetTo(Label * label);
-
-  // A common implementation for the LinkAndGet<Type>OffsetTo helpers.
-  template <int element_shift>
-  ptrdiff_t LinkAndGetOffsetTo(Label* label);
-
-  // Literal load offset are in words (32-bit).
-  ptrdiff_t LinkAndGetWordOffsetTo(RawLiteral* literal);
-
-  // Emit the instruction in buffer_.
-  void Emit(Instr instruction) {
-    VIXL_STATIC_ASSERT(sizeof(instruction) == kInstructionSize);
-    VIXL_ASSERT(buffer_monitor_ > 0);
-    buffer_->Emit32(instruction);
-  }
-
-  // Buffer where the code is emitted.
-  CodeBuffer* buffer_;
-  PositionIndependentCodeOption pic_;
-
-#ifdef VIXL_DEBUG
-  int64_t buffer_monitor_;
-#endif
-};
-
-
-// All Assembler emits MUST acquire/release the underlying code buffer. The
-// helper scope below will do so and optionally ensure the buffer is big enough
-// to receive the emit. It is possible to request the scope not to perform any
-// checks (kNoCheck) if for example it is known in advance the buffer size is
-// adequate or there is some other size checking mechanism in place.
-class CodeBufferCheckScope {
- public:
-  // Tell whether or not the scope needs to ensure the associated CodeBuffer
-  // has enough space for the requested size.
-  enum CheckPolicy {
-    kNoCheck,
-    kCheck
-  };
-
-  // Tell whether or not the scope should assert the amount of code emitted
-  // within the scope is consistent with the requested amount.
-  enum AssertPolicy {
-    kNoAssert,    // No assert required.
-    kExactSize,   // The code emitted must be exactly size bytes.
-    kMaximumSize  // The code emitted must be at most size bytes.
-  };
-
-  CodeBufferCheckScope(Assembler* assm,
-                       size_t size,
-                       CheckPolicy check_policy = kCheck,
-                       AssertPolicy assert_policy = kMaximumSize)
-      : assm_(assm) {
-    if (check_policy == kCheck) assm->EnsureSpaceFor(size);
-#ifdef VIXL_DEBUG
-    assm->bind(&start_);
-    size_ = size;
-    assert_policy_ = assert_policy;
-    assm->AcquireBuffer();
-#else
-    USE(assert_policy);
-#endif
-  }
-
-  // This is a shortcut for CodeBufferCheckScope(assm, 0, kNoCheck, kNoAssert).
-  explicit CodeBufferCheckScope(Assembler* assm) : assm_(assm) {
-#ifdef VIXL_DEBUG
-    size_ = 0;
-    assert_policy_ = kNoAssert;
-    assm->AcquireBuffer();
-#endif
-  }
-
-  ~CodeBufferCheckScope() {
-#ifdef VIXL_DEBUG
-    assm_->ReleaseBuffer();
-    switch (assert_policy_) {
-      case kNoAssert: break;
-      case kExactSize:
-        VIXL_ASSERT(assm_->SizeOfCodeGeneratedSince(&start_) == size_);
-        break;
-      case kMaximumSize:
-        VIXL_ASSERT(assm_->SizeOfCodeGeneratedSince(&start_) <= size_);
-        break;
-      default:
-        VIXL_UNREACHABLE();
-    }
-#endif
-  }
-
- protected:
-  Assembler* assm_;
-#ifdef VIXL_DEBUG
-  Label start_;
-  size_t size_;
-  AssertPolicy assert_policy_;
-#endif
-};
-
-}  // namespace vixl
-
-#endif  // VIXL_A64_ASSEMBLER_A64_H_
diff --git a/disas/libvixl/a64/disasm-a64.cc b/disas/libvixl/a64/disasm-a64.cc
deleted file mode 100644
index f7bc2468bb..0000000000
--- a/disas/libvixl/a64/disasm-a64.cc
+++ /dev/null
@@ -1,1954 +0,0 @@
-// Copyright 2013, 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.
-
-#include <cstdlib>
-#include "a64/disasm-a64.h"
-
-namespace vixl {
-
-Disassembler::Disassembler() {
-  buffer_size_ = 256;
-  buffer_ = reinterpret_cast<char*>(malloc(buffer_size_));
-  buffer_pos_ = 0;
-  own_buffer_ = true;
-  code_address_offset_ = 0;
-}
-
-
-Disassembler::Disassembler(char* text_buffer, int buffer_size) {
-  buffer_size_ = buffer_size;
-  buffer_ = text_buffer;
-  buffer_pos_ = 0;
-  own_buffer_ = false;
-  code_address_offset_ = 0;
-}
-
-
-Disassembler::~Disassembler() {
-  if (own_buffer_) {
-    free(buffer_);
-  }
-}
-
-
-char* Disassembler::GetOutput() {
-  return buffer_;
-}
-
-
-void Disassembler::VisitAddSubImmediate(const Instruction* instr) {
-  bool rd_is_zr = RdIsZROrSP(instr);
-  bool stack_op = (rd_is_zr || RnIsZROrSP(instr)) &&
-                  (instr->ImmAddSub() == 0) ? true : false;
-  const char *mnemonic = "";
-  const char *form = "'Rds, 'Rns, 'IAddSub";
-  const char *form_cmp = "'Rns, 'IAddSub";
-  const char *form_mov = "'Rds, 'Rns";
-
-  switch (instr->Mask(AddSubImmediateMask)) {
-    case ADD_w_imm:
-    case ADD_x_imm: {
-      mnemonic = "add";
-      if (stack_op) {
-        mnemonic = "mov";
-        form = form_mov;
-      }
-      break;
-    }
-    case ADDS_w_imm:
-    case ADDS_x_imm: {
-      mnemonic = "adds";
-      if (rd_is_zr) {
-        mnemonic = "cmn";
-        form = form_cmp;
-      }
-      break;
-    }
-    case SUB_w_imm:
-    case SUB_x_imm: mnemonic = "sub"; break;
-    case SUBS_w_imm:
-    case SUBS_x_imm: {
-      mnemonic = "subs";
-      if (rd_is_zr) {
-        mnemonic = "cmp";
-        form = form_cmp;
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitAddSubShifted(const Instruction* instr) {
-  bool rd_is_zr = RdIsZROrSP(instr);
-  bool rn_is_zr = RnIsZROrSP(instr);
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn, 'Rm'HDP";
-  const char *form_cmp = "'Rn, 'Rm'HDP";
-  const char *form_neg = "'Rd, 'Rm'HDP";
-
-  switch (instr->Mask(AddSubShiftedMask)) {
-    case ADD_w_shift:
-    case ADD_x_shift: mnemonic = "add"; break;
-    case ADDS_w_shift:
-    case ADDS_x_shift: {
-      mnemonic = "adds";
-      if (rd_is_zr) {
-        mnemonic = "cmn";
-        form = form_cmp;
-      }
-      break;
-    }
-    case SUB_w_shift:
-    case SUB_x_shift: {
-      mnemonic = "sub";
-      if (rn_is_zr) {
-        mnemonic = "neg";
-        form = form_neg;
-      }
-      break;
-    }
-    case SUBS_w_shift:
-    case SUBS_x_shift: {
-      mnemonic = "subs";
-      if (rd_is_zr) {
-        mnemonic = "cmp";
-        form = form_cmp;
-      } else if (rn_is_zr) {
-        mnemonic = "negs";
-        form = form_neg;
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitAddSubExtended(const Instruction* instr) {
-  bool rd_is_zr = RdIsZROrSP(instr);
-  const char *mnemonic = "";
-  Extend mode = static_cast<Extend>(instr->ExtendMode());
-  const char *form = ((mode == UXTX) || (mode == SXTX)) ?
-                     "'Rds, 'Rns, 'Xm'Ext" : "'Rds, 'Rns, 'Wm'Ext";
-  const char *form_cmp = ((mode == UXTX) || (mode == SXTX)) ?
-                         "'Rns, 'Xm'Ext" : "'Rns, 'Wm'Ext";
-
-  switch (instr->Mask(AddSubExtendedMask)) {
-    case ADD_w_ext:
-    case ADD_x_ext: mnemonic = "add"; break;
-    case ADDS_w_ext:
-    case ADDS_x_ext: {
-      mnemonic = "adds";
-      if (rd_is_zr) {
-        mnemonic = "cmn";
-        form = form_cmp;
-      }
-      break;
-    }
-    case SUB_w_ext:
-    case SUB_x_ext: mnemonic = "sub"; break;
-    case SUBS_w_ext:
-    case SUBS_x_ext: {
-      mnemonic = "subs";
-      if (rd_is_zr) {
-        mnemonic = "cmp";
-        form = form_cmp;
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitAddSubWithCarry(const Instruction* instr) {
-  bool rn_is_zr = RnIsZROrSP(instr);
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn, 'Rm";
-  const char *form_neg = "'Rd, 'Rm";
-
-  switch (instr->Mask(AddSubWithCarryMask)) {
-    case ADC_w:
-    case ADC_x: mnemonic = "adc"; break;
-    case ADCS_w:
-    case ADCS_x: mnemonic = "adcs"; break;
-    case SBC_w:
-    case SBC_x: {
-      mnemonic = "sbc";
-      if (rn_is_zr) {
-        mnemonic = "ngc";
-        form = form_neg;
-      }
-      break;
-    }
-    case SBCS_w:
-    case SBCS_x: {
-      mnemonic = "sbcs";
-      if (rn_is_zr) {
-        mnemonic = "ngcs";
-        form = form_neg;
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLogicalImmediate(const Instruction* instr) {
-  bool rd_is_zr = RdIsZROrSP(instr);
-  bool rn_is_zr = RnIsZROrSP(instr);
-  const char *mnemonic = "";
-  const char *form = "'Rds, 'Rn, 'ITri";
-
-  if (instr->ImmLogical() == 0) {
-    // The immediate encoded in the instruction is not in the expected format.
-    Format(instr, "unallocated", "(LogicalImmediate)");
-    return;
-  }
-
-  switch (instr->Mask(LogicalImmediateMask)) {
-    case AND_w_imm:
-    case AND_x_imm: mnemonic = "and"; break;
-    case ORR_w_imm:
-    case ORR_x_imm: {
-      mnemonic = "orr";
-      unsigned reg_size = (instr->SixtyFourBits() == 1) ? kXRegSize
-                                                        : kWRegSize;
-      if (rn_is_zr && !IsMovzMovnImm(reg_size, instr->ImmLogical())) {
-        mnemonic = "mov";
-        form = "'Rds, 'ITri";
-      }
-      break;
-    }
-    case EOR_w_imm:
-    case EOR_x_imm: mnemonic = "eor"; break;
-    case ANDS_w_imm:
-    case ANDS_x_imm: {
-      mnemonic = "ands";
-      if (rd_is_zr) {
-        mnemonic = "tst";
-        form = "'Rn, 'ITri";
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-bool Disassembler::IsMovzMovnImm(unsigned reg_size, uint64_t value) {
-  VIXL_ASSERT((reg_size == kXRegSize) ||
-              ((reg_size == kWRegSize) && (value <= 0xffffffff)));
-
-  // Test for movz: 16 bits set at positions 0, 16, 32 or 48.
-  if (((value & UINT64_C(0xffffffffffff0000)) == 0) ||
-      ((value & UINT64_C(0xffffffff0000ffff)) == 0) ||
-      ((value & UINT64_C(0xffff0000ffffffff)) == 0) ||
-      ((value & UINT64_C(0x0000ffffffffffff)) == 0)) {
-    return true;
-  }
-
-  // Test for movn: NOT(16 bits set at positions 0, 16, 32 or 48).
-  if ((reg_size == kXRegSize) &&
-      (((~value & UINT64_C(0xffffffffffff0000)) == 0) ||
-       ((~value & UINT64_C(0xffffffff0000ffff)) == 0) ||
-       ((~value & UINT64_C(0xffff0000ffffffff)) == 0) ||
-       ((~value & UINT64_C(0x0000ffffffffffff)) == 0))) {
-    return true;
-  }
-  if ((reg_size == kWRegSize) &&
-      (((value & 0xffff0000) == 0xffff0000) ||
-       ((value & 0x0000ffff) == 0x0000ffff))) {
-    return true;
-  }
-  return false;
-}
-
-
-void Disassembler::VisitLogicalShifted(const Instruction* instr) {
-  bool rd_is_zr = RdIsZROrSP(instr);
-  bool rn_is_zr = RnIsZROrSP(instr);
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn, 'Rm'HLo";
-
-  switch (instr->Mask(LogicalShiftedMask)) {
-    case AND_w:
-    case AND_x: mnemonic = "and"; break;
-    case BIC_w:
-    case BIC_x: mnemonic = "bic"; break;
-    case EOR_w:
-    case EOR_x: mnemonic = "eor"; break;
-    case EON_w:
-    case EON_x: mnemonic = "eon"; break;
-    case BICS_w:
-    case BICS_x: mnemonic = "bics"; break;
-    case ANDS_w:
-    case ANDS_x: {
-      mnemonic = "ands";
-      if (rd_is_zr) {
-        mnemonic = "tst";
-        form = "'Rn, 'Rm'HLo";
-      }
-      break;
-    }
-    case ORR_w:
-    case ORR_x: {
-      mnemonic = "orr";
-      if (rn_is_zr && (instr->ImmDPShift() == 0) && (instr->ShiftDP() == LSL)) {
-        mnemonic = "mov";
-        form = "'Rd, 'Rm";
-      }
-      break;
-    }
-    case ORN_w:
-    case ORN_x: {
-      mnemonic = "orn";
-      if (rn_is_zr) {
-        mnemonic = "mvn";
-        form = "'Rd, 'Rm'HLo";
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitConditionalCompareRegister(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rn, 'Rm, 'INzcv, 'Cond";
-
-  switch (instr->Mask(ConditionalCompareRegisterMask)) {
-    case CCMN_w:
-    case CCMN_x: mnemonic = "ccmn"; break;
-    case CCMP_w:
-    case CCMP_x: mnemonic = "ccmp"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitConditionalCompareImmediate(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rn, 'IP, 'INzcv, 'Cond";
-
-  switch (instr->Mask(ConditionalCompareImmediateMask)) {
-    case CCMN_w_imm:
-    case CCMN_x_imm: mnemonic = "ccmn"; break;
-    case CCMP_w_imm:
-    case CCMP_x_imm: mnemonic = "ccmp"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitConditionalSelect(const Instruction* instr) {
-  bool rnm_is_zr = (RnIsZROrSP(instr) && RmIsZROrSP(instr));
-  bool rn_is_rm = (instr->Rn() == instr->Rm());
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn, 'Rm, 'Cond";
-  const char *form_test = "'Rd, 'CInv";
-  const char *form_update = "'Rd, 'Rn, 'CInv";
-
-  Condition cond = static_cast<Condition>(instr->Condition());
-  bool invertible_cond = (cond != al) && (cond != nv);
-
-  switch (instr->Mask(ConditionalSelectMask)) {
-    case CSEL_w:
-    case CSEL_x: mnemonic = "csel"; break;
-    case CSINC_w:
-    case CSINC_x: {
-      mnemonic = "csinc";
-      if (rnm_is_zr && invertible_cond) {
-        mnemonic = "cset";
-        form = form_test;
-      } else if (rn_is_rm && invertible_cond) {
-        mnemonic = "cinc";
-        form = form_update;
-      }
-      break;
-    }
-    case CSINV_w:
-    case CSINV_x: {
-      mnemonic = "csinv";
-      if (rnm_is_zr && invertible_cond) {
-        mnemonic = "csetm";
-        form = form_test;
-      } else if (rn_is_rm && invertible_cond) {
-        mnemonic = "cinv";
-        form = form_update;
-      }
-      break;
-    }
-    case CSNEG_w:
-    case CSNEG_x: {
-      mnemonic = "csneg";
-      if (rn_is_rm && invertible_cond) {
-        mnemonic = "cneg";
-        form = form_update;
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitBitfield(const Instruction* instr) {
-  unsigned s = instr->ImmS();
-  unsigned r = instr->ImmR();
-  unsigned rd_size_minus_1 =
-    ((instr->SixtyFourBits() == 1) ? kXRegSize : kWRegSize) - 1;
-  const char *mnemonic = "";
-  const char *form = "";
-  const char *form_shift_right = "'Rd, 'Rn, 'IBr";
-  const char *form_extend = "'Rd, 'Wn";
-  const char *form_bfiz = "'Rd, 'Rn, 'IBZ-r, 'IBs+1";
-  const char *form_bfx = "'Rd, 'Rn, 'IBr, 'IBs-r+1";
-  const char *form_lsl = "'Rd, 'Rn, 'IBZ-r";
-
-  switch (instr->Mask(BitfieldMask)) {
-    case SBFM_w:
-    case SBFM_x: {
-      mnemonic = "sbfx";
-      form = form_bfx;
-      if (r == 0) {
-        form = form_extend;
-        if (s == 7) {
-          mnemonic = "sxtb";
-        } else if (s == 15) {
-          mnemonic = "sxth";
-        } else if ((s == 31) && (instr->SixtyFourBits() == 1)) {
-          mnemonic = "sxtw";
-        } else {
-          form = form_bfx;
-        }
-      } else if (s == rd_size_minus_1) {
-        mnemonic = "asr";
-        form = form_shift_right;
-      } else if (s < r) {
-        mnemonic = "sbfiz";
-        form = form_bfiz;
-      }
-      break;
-    }
-    case UBFM_w:
-    case UBFM_x: {
-      mnemonic = "ubfx";
-      form = form_bfx;
-      if (r == 0) {
-        form = form_extend;
-        if (s == 7) {
-          mnemonic = "uxtb";
-        } else if (s == 15) {
-          mnemonic = "uxth";
-        } else {
-          form = form_bfx;
-        }
-      }
-      if (s == rd_size_minus_1) {
-        mnemonic = "lsr";
-        form = form_shift_right;
-      } else if (r == s + 1) {
-        mnemonic = "lsl";
-        form = form_lsl;
-      } else if (s < r) {
-        mnemonic = "ubfiz";
-        form = form_bfiz;
-      }
-      break;
-    }
-    case BFM_w:
-    case BFM_x: {
-      mnemonic = "bfxil";
-      form = form_bfx;
-      if (s < r) {
-        mnemonic = "bfi";
-        form = form_bfiz;
-      }
-    }
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitExtract(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn, 'Rm, 'IExtract";
-
-  switch (instr->Mask(ExtractMask)) {
-    case EXTR_w:
-    case EXTR_x: {
-      if (instr->Rn() == instr->Rm()) {
-        mnemonic = "ror";
-        form = "'Rd, 'Rn, 'IExtract";
-      } else {
-        mnemonic = "extr";
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitPCRelAddressing(const Instruction* instr) {
-  switch (instr->Mask(PCRelAddressingMask)) {
-    case ADR: Format(instr, "adr", "'Xd, 'AddrPCRelByte"); break;
-    case ADRP: Format(instr, "adrp", "'Xd, 'AddrPCRelPage"); break;
-    default: Format(instr, "unimplemented", "(PCRelAddressing)");
-  }
-}
-
-
-void Disassembler::VisitConditionalBranch(const Instruction* instr) {
-  switch (instr->Mask(ConditionalBranchMask)) {
-    case B_cond: Format(instr, "b.'CBrn", "'BImmCond"); break;
-    default: VIXL_UNREACHABLE();
-  }
-}
-
-
-void Disassembler::VisitUnconditionalBranchToRegister(
-    const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'Xn";
-
-  switch (instr->Mask(UnconditionalBranchToRegisterMask)) {
-    case BR: mnemonic = "br"; break;
-    case BLR: mnemonic = "blr"; break;
-    case RET: {
-      mnemonic = "ret";
-      if (instr->Rn() == kLinkRegCode) {
-        form = NULL;
-      }
-      break;
-    }
-    default: form = "(UnconditionalBranchToRegister)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitUnconditionalBranch(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'BImmUncn";
-
-  switch (instr->Mask(UnconditionalBranchMask)) {
-    case B: mnemonic = "b"; break;
-    case BL: mnemonic = "bl"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitDataProcessing1Source(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Rn";
-
-  switch (instr->Mask(DataProcessing1SourceMask)) {
-    #define FORMAT(A, B)  \
-    case A##_w:           \
-    case A##_x: mnemonic = B; break;
-    FORMAT(RBIT, "rbit");
-    FORMAT(REV16, "rev16");
-    FORMAT(REV, "rev");
-    FORMAT(CLZ, "clz");
-    FORMAT(CLS, "cls");
-    #undef FORMAT
-    case REV32_x: mnemonic = "rev32"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitDataProcessing2Source(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'Rd, 'Rn, 'Rm";
-
-  switch (instr->Mask(DataProcessing2SourceMask)) {
-    #define FORMAT(A, B)  \
-    case A##_w:           \
-    case A##_x: mnemonic = B; break;
-    FORMAT(UDIV, "udiv");
-    FORMAT(SDIV, "sdiv");
-    FORMAT(LSLV, "lsl");
-    FORMAT(LSRV, "lsr");
-    FORMAT(ASRV, "asr");
-    FORMAT(RORV, "ror");
-    #undef FORMAT
-    default: form = "(DataProcessing2Source)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitDataProcessing3Source(const Instruction* instr) {
-  bool ra_is_zr = RaIsZROrSP(instr);
-  const char *mnemonic = "";
-  const char *form = "'Xd, 'Wn, 'Wm, 'Xa";
-  const char *form_rrr = "'Rd, 'Rn, 'Rm";
-  const char *form_rrrr = "'Rd, 'Rn, 'Rm, 'Ra";
-  const char *form_xww = "'Xd, 'Wn, 'Wm";
-  const char *form_xxx = "'Xd, 'Xn, 'Xm";
-
-  switch (instr->Mask(DataProcessing3SourceMask)) {
-    case MADD_w:
-    case MADD_x: {
-      mnemonic = "madd";
-      form = form_rrrr;
-      if (ra_is_zr) {
-        mnemonic = "mul";
-        form = form_rrr;
-      }
-      break;
-    }
-    case MSUB_w:
-    case MSUB_x: {
-      mnemonic = "msub";
-      form = form_rrrr;
-      if (ra_is_zr) {
-        mnemonic = "mneg";
-        form = form_rrr;
-      }
-      break;
-    }
-    case SMADDL_x: {
-      mnemonic = "smaddl";
-      if (ra_is_zr) {
-        mnemonic = "smull";
-        form = form_xww;
-      }
-      break;
-    }
-    case SMSUBL_x: {
-      mnemonic = "smsubl";
-      if (ra_is_zr) {
-        mnemonic = "smnegl";
-        form = form_xww;
-      }
-      break;
-    }
-    case UMADDL_x: {
-      mnemonic = "umaddl";
-      if (ra_is_zr) {
-        mnemonic = "umull";
-        form = form_xww;
-      }
-      break;
-    }
-    case UMSUBL_x: {
-      mnemonic = "umsubl";
-      if (ra_is_zr) {
-        mnemonic = "umnegl";
-        form = form_xww;
-      }
-      break;
-    }
-    case SMULH_x: {
-      mnemonic = "smulh";
-      form = form_xxx;
-      break;
-    }
-    case UMULH_x: {
-      mnemonic = "umulh";
-      form = form_xxx;
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitCompareBranch(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rt, 'BImmCmpa";
-
-  switch (instr->Mask(CompareBranchMask)) {
-    case CBZ_w:
-    case CBZ_x: mnemonic = "cbz"; break;
-    case CBNZ_w:
-    case CBNZ_x: mnemonic = "cbnz"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitTestBranch(const Instruction* instr) {
-  const char *mnemonic = "";
-  // If the top bit of the immediate is clear, the tested register is
-  // disassembled as Wt, otherwise Xt. As the top bit of the immediate is
-  // encoded in bit 31 of the instruction, we can reuse the Rt form, which
-  // uses bit 31 (normally "sf") to choose the register size.
-  const char *form = "'Rt, 'IS, 'BImmTest";
-
-  switch (instr->Mask(TestBranchMask)) {
-    case TBZ: mnemonic = "tbz"; break;
-    case TBNZ: mnemonic = "tbnz"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitMoveWideImmediate(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'IMoveImm";
-
-  // Print the shift separately for movk, to make it clear which half word will
-  // be overwritten. Movn and movz print the computed immediate, which includes
-  // shift calculation.
-  switch (instr->Mask(MoveWideImmediateMask)) {
-    case MOVN_w:
-    case MOVN_x:
-      if ((instr->ImmMoveWide()) || (instr->ShiftMoveWide() == 0)) {
-        if ((instr->SixtyFourBits() == 0) && (instr->ImmMoveWide() == 0xffff)) {
-          mnemonic = "movn";
-        } else {
-          mnemonic = "mov";
-          form = "'Rd, 'IMoveNeg";
-        }
-      } else {
-        mnemonic = "movn";
-      }
-      break;
-    case MOVZ_w:
-    case MOVZ_x:
-      if ((instr->ImmMoveWide()) || (instr->ShiftMoveWide() == 0))
-        mnemonic = "mov";
-      else
-        mnemonic = "movz";
-      break;
-    case MOVK_w:
-    case MOVK_x: mnemonic = "movk"; form = "'Rd, 'IMoveLSL"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-#define LOAD_STORE_LIST(V)    \
-  V(STRB_w, "strb", "'Wt")    \
-  V(STRH_w, "strh", "'Wt")    \
-  V(STR_w, "str", "'Wt")      \
-  V(STR_x, "str", "'Xt")      \
-  V(LDRB_w, "ldrb", "'Wt")    \
-  V(LDRH_w, "ldrh", "'Wt")    \
-  V(LDR_w, "ldr", "'Wt")      \
-  V(LDR_x, "ldr", "'Xt")      \
-  V(LDRSB_x, "ldrsb", "'Xt")  \
-  V(LDRSH_x, "ldrsh", "'Xt")  \
-  V(LDRSW_x, "ldrsw", "'Xt")  \
-  V(LDRSB_w, "ldrsb", "'Wt")  \
-  V(LDRSH_w, "ldrsh", "'Wt")  \
-  V(STR_s, "str", "'St")      \
-  V(STR_d, "str", "'Dt")      \
-  V(LDR_s, "ldr", "'St")      \
-  V(LDR_d, "ldr", "'Dt")
-
-void Disassembler::VisitLoadStorePreIndex(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStorePreIndex)";
-
-  switch (instr->Mask(LoadStorePreIndexMask)) {
-    #define LS_PREINDEX(A, B, C) \
-    case A##_pre: mnemonic = B; form = C ", ['Xns'ILS]!"; break;
-    LOAD_STORE_LIST(LS_PREINDEX)
-    #undef LS_PREINDEX
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStorePostIndex(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStorePostIndex)";
-
-  switch (instr->Mask(LoadStorePostIndexMask)) {
-    #define LS_POSTINDEX(A, B, C) \
-    case A##_post: mnemonic = B; form = C ", ['Xns]'ILS"; break;
-    LOAD_STORE_LIST(LS_POSTINDEX)
-    #undef LS_POSTINDEX
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStoreUnsignedOffset(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStoreUnsignedOffset)";
-
-  switch (instr->Mask(LoadStoreUnsignedOffsetMask)) {
-    #define LS_UNSIGNEDOFFSET(A, B, C) \
-    case A##_unsigned: mnemonic = B; form = C ", ['Xns'ILU]"; break;
-    LOAD_STORE_LIST(LS_UNSIGNEDOFFSET)
-    #undef LS_UNSIGNEDOFFSET
-    case PRFM_unsigned: mnemonic = "prfm"; form = "'PrefOp, ['Xns'ILU]";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStoreRegisterOffset(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStoreRegisterOffset)";
-
-  switch (instr->Mask(LoadStoreRegisterOffsetMask)) {
-    #define LS_REGISTEROFFSET(A, B, C) \
-    case A##_reg: mnemonic = B; form = C ", ['Xns, 'Offsetreg]"; break;
-    LOAD_STORE_LIST(LS_REGISTEROFFSET)
-    #undef LS_REGISTEROFFSET
-    case PRFM_reg: mnemonic = "prfm"; form = "'PrefOp, ['Xns, 'Offsetreg]";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStoreUnscaledOffset(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'Wt, ['Xns'ILS]";
-  const char *form_x = "'Xt, ['Xns'ILS]";
-  const char *form_s = "'St, ['Xns'ILS]";
-  const char *form_d = "'Dt, ['Xns'ILS]";
-  const char *form_prefetch = "'PrefOp, ['Xns'ILS]";
-
-  switch (instr->Mask(LoadStoreUnscaledOffsetMask)) {
-    case STURB_w:  mnemonic = "sturb"; break;
-    case STURH_w:  mnemonic = "sturh"; break;
-    case STUR_w:   mnemonic = "stur"; break;
-    case STUR_x:   mnemonic = "stur"; form = form_x; break;
-    case STUR_s:   mnemonic = "stur"; form = form_s; break;
-    case STUR_d:   mnemonic = "stur"; form = form_d; break;
-    case LDURB_w:  mnemonic = "ldurb"; break;
-    case LDURH_w:  mnemonic = "ldurh"; break;
-    case LDUR_w:   mnemonic = "ldur"; break;
-    case LDUR_x:   mnemonic = "ldur"; form = form_x; break;
-    case LDUR_s:   mnemonic = "ldur"; form = form_s; break;
-    case LDUR_d:   mnemonic = "ldur"; form = form_d; break;
-    case LDURSB_x: form = form_x;  // Fall through.
-    case LDURSB_w: mnemonic = "ldursb"; break;
-    case LDURSH_x: form = form_x;  // Fall through.
-    case LDURSH_w: mnemonic = "ldursh"; break;
-    case LDURSW_x: mnemonic = "ldursw"; form = form_x; break;
-    case PRFUM:    mnemonic = "prfum"; form = form_prefetch; break;
-    default: form = "(LoadStoreUnscaledOffset)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadLiteral(const Instruction* instr) {
-  const char *mnemonic = "ldr";
-  const char *form = "(LoadLiteral)";
-
-  switch (instr->Mask(LoadLiteralMask)) {
-    case LDR_w_lit: form = "'Wt, 'ILLiteral 'LValue"; break;
-    case LDR_x_lit: form = "'Xt, 'ILLiteral 'LValue"; break;
-    case LDR_s_lit: form = "'St, 'ILLiteral 'LValue"; break;
-    case LDR_d_lit: form = "'Dt, 'ILLiteral 'LValue"; break;
-    case LDRSW_x_lit: {
-      mnemonic = "ldrsw";
-      form = "'Xt, 'ILLiteral 'LValue";
-      break;
-    }
-    case PRFM_lit: {
-      mnemonic = "prfm";
-      form = "'PrefOp, 'ILLiteral 'LValue";
-      break;
-    }
-    default: mnemonic = "unimplemented";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-#define LOAD_STORE_PAIR_LIST(V)         \
-  V(STP_w, "stp", "'Wt, 'Wt2", "4")     \
-  V(LDP_w, "ldp", "'Wt, 'Wt2", "4")     \
-  V(LDPSW_x, "ldpsw", "'Xt, 'Xt2", "4") \
-  V(STP_x, "stp", "'Xt, 'Xt2", "8")     \
-  V(LDP_x, "ldp", "'Xt, 'Xt2", "8")     \
-  V(STP_s, "stp", "'St, 'St2", "4")     \
-  V(LDP_s, "ldp", "'St, 'St2", "4")     \
-  V(STP_d, "stp", "'Dt, 'Dt2", "8")     \
-  V(LDP_d, "ldp", "'Dt, 'Dt2", "8")
-
-void Disassembler::VisitLoadStorePairPostIndex(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStorePairPostIndex)";
-
-  switch (instr->Mask(LoadStorePairPostIndexMask)) {
-    #define LSP_POSTINDEX(A, B, C, D) \
-    case A##_post: mnemonic = B; form = C ", ['Xns]'ILP" D; break;
-    LOAD_STORE_PAIR_LIST(LSP_POSTINDEX)
-    #undef LSP_POSTINDEX
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStorePairPreIndex(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStorePairPreIndex)";
-
-  switch (instr->Mask(LoadStorePairPreIndexMask)) {
-    #define LSP_PREINDEX(A, B, C, D) \
-    case A##_pre: mnemonic = B; form = C ", ['Xns'ILP" D "]!"; break;
-    LOAD_STORE_PAIR_LIST(LSP_PREINDEX)
-    #undef LSP_PREINDEX
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStorePairOffset(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(LoadStorePairOffset)";
-
-  switch (instr->Mask(LoadStorePairOffsetMask)) {
-    #define LSP_OFFSET(A, B, C, D) \
-    case A##_off: mnemonic = B; form = C ", ['Xns'ILP" D "]"; break;
-    LOAD_STORE_PAIR_LIST(LSP_OFFSET)
-    #undef LSP_OFFSET
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStorePairNonTemporal(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form;
-
-  switch (instr->Mask(LoadStorePairNonTemporalMask)) {
-    case STNP_w: mnemonic = "stnp"; form = "'Wt, 'Wt2, ['Xns'ILP4]"; break;
-    case LDNP_w: mnemonic = "ldnp"; form = "'Wt, 'Wt2, ['Xns'ILP4]"; break;
-    case STNP_x: mnemonic = "stnp"; form = "'Xt, 'Xt2, ['Xns'ILP8]"; break;
-    case LDNP_x: mnemonic = "ldnp"; form = "'Xt, 'Xt2, ['Xns'ILP8]"; break;
-    case STNP_s: mnemonic = "stnp"; form = "'St, 'St2, ['Xns'ILP4]"; break;
-    case LDNP_s: mnemonic = "ldnp"; form = "'St, 'St2, ['Xns'ILP4]"; break;
-    case STNP_d: mnemonic = "stnp"; form = "'Dt, 'Dt2, ['Xns'ILP8]"; break;
-    case LDNP_d: mnemonic = "ldnp"; form = "'Dt, 'Dt2, ['Xns'ILP8]"; break;
-    default: form = "(LoadStorePairNonTemporal)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitLoadStoreExclusive(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form;
-
-  switch (instr->Mask(LoadStoreExclusiveMask)) {
-    case STXRB_w: mnemonic = "stxrb"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STXRH_w: mnemonic = "stxrh"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STXR_w: mnemonic = "stxr"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STXR_x: mnemonic = "stxr"; form = "'Ws, 'Xt, ['Xns]"; break;
-    case LDXRB_w: mnemonic = "ldxrb"; form = "'Wt, ['Xns]"; break;
-    case LDXRH_w: mnemonic = "ldxrh"; form = "'Wt, ['Xns]"; break;
-    case LDXR_w: mnemonic = "ldxr"; form = "'Wt, ['Xns]"; break;
-    case LDXR_x: mnemonic = "ldxr"; form = "'Xt, ['Xns]"; break;
-    case STXP_w: mnemonic = "stxp"; form = "'Ws, 'Wt, 'Wt2, ['Xns]"; break;
-    case STXP_x: mnemonic = "stxp"; form = "'Ws, 'Xt, 'Xt2, ['Xns]"; break;
-    case LDXP_w: mnemonic = "ldxp"; form = "'Wt, 'Wt2, ['Xns]"; break;
-    case LDXP_x: mnemonic = "ldxp"; form = "'Xt, 'Xt2, ['Xns]"; break;
-    case STLXRB_w: mnemonic = "stlxrb"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STLXRH_w: mnemonic = "stlxrh"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STLXR_w: mnemonic = "stlxr"; form = "'Ws, 'Wt, ['Xns]"; break;
-    case STLXR_x: mnemonic = "stlxr"; form = "'Ws, 'Xt, ['Xns]"; break;
-    case LDAXRB_w: mnemonic = "ldaxrb"; form = "'Wt, ['Xns]"; break;
-    case LDAXRH_w: mnemonic = "ldaxrh"; form = "'Wt, ['Xns]"; break;
-    case LDAXR_w: mnemonic = "ldaxr"; form = "'Wt, ['Xns]"; break;
-    case LDAXR_x: mnemonic = "ldaxr"; form = "'Xt, ['Xns]"; break;
-    case STLXP_w: mnemonic = "stlxp"; form = "'Ws, 'Wt, 'Wt2, ['Xns]"; break;
-    case STLXP_x: mnemonic = "stlxp"; form = "'Ws, 'Xt, 'Xt2, ['Xns]"; break;
-    case LDAXP_w: mnemonic = "ldaxp"; form = "'Wt, 'Wt2, ['Xns]"; break;
-    case LDAXP_x: mnemonic = "ldaxp"; form = "'Xt, 'Xt2, ['Xns]"; break;
-    case STLRB_w: mnemonic = "stlrb"; form = "'Wt, ['Xns]"; break;
-    case STLRH_w: mnemonic = "stlrh"; form = "'Wt, ['Xns]"; break;
-    case STLR_w: mnemonic = "stlr"; form = "'Wt, ['Xns]"; break;
-    case STLR_x: mnemonic = "stlr"; form = "'Xt, ['Xns]"; break;
-    case LDARB_w: mnemonic = "ldarb"; form = "'Wt, ['Xns]"; break;
-    case LDARH_w: mnemonic = "ldarh"; form = "'Wt, ['Xns]"; break;
-    case LDAR_w: mnemonic = "ldar"; form = "'Wt, ['Xns]"; break;
-    case LDAR_x: mnemonic = "ldar"; form = "'Xt, ['Xns]"; break;
-    default: form = "(LoadStoreExclusive)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPCompare(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'Fn, 'Fm";
-  const char *form_zero = "'Fn, #0.0";
-
-  switch (instr->Mask(FPCompareMask)) {
-    case FCMP_s_zero:
-    case FCMP_d_zero: form = form_zero;  // Fall through.
-    case FCMP_s:
-    case FCMP_d: mnemonic = "fcmp"; break;
-    default: form = "(FPCompare)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPConditionalCompare(const Instruction* instr) {
-  const char *mnemonic = "unmplemented";
-  const char *form = "'Fn, 'Fm, 'INzcv, 'Cond";
-
-  switch (instr->Mask(FPConditionalCompareMask)) {
-    case FCCMP_s:
-    case FCCMP_d: mnemonic = "fccmp"; break;
-    case FCCMPE_s:
-    case FCCMPE_d: mnemonic = "fccmpe"; break;
-    default: form = "(FPConditionalCompare)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPConditionalSelect(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Fd, 'Fn, 'Fm, 'Cond";
-
-  switch (instr->Mask(FPConditionalSelectMask)) {
-    case FCSEL_s:
-    case FCSEL_d: mnemonic = "fcsel"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPDataProcessing1Source(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'Fd, 'Fn";
-
-  switch (instr->Mask(FPDataProcessing1SourceMask)) {
-    #define FORMAT(A, B)  \
-    case A##_s:           \
-    case A##_d: mnemonic = B; break;
-    FORMAT(FMOV, "fmov");
-    FORMAT(FABS, "fabs");
-    FORMAT(FNEG, "fneg");
-    FORMAT(FSQRT, "fsqrt");
-    FORMAT(FRINTN, "frintn");
-    FORMAT(FRINTP, "frintp");
-    FORMAT(FRINTM, "frintm");
-    FORMAT(FRINTZ, "frintz");
-    FORMAT(FRINTA, "frinta");
-    FORMAT(FRINTX, "frintx");
-    FORMAT(FRINTI, "frinti");
-    #undef FORMAT
-    case FCVT_ds: mnemonic = "fcvt"; form = "'Dd, 'Sn"; break;
-    case FCVT_sd: mnemonic = "fcvt"; form = "'Sd, 'Dn"; break;
-    default: form = "(FPDataProcessing1Source)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPDataProcessing2Source(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Fd, 'Fn, 'Fm";
-
-  switch (instr->Mask(FPDataProcessing2SourceMask)) {
-    #define FORMAT(A, B)  \
-    case A##_s:           \
-    case A##_d: mnemonic = B; break;
-    FORMAT(FMUL, "fmul");
-    FORMAT(FDIV, "fdiv");
-    FORMAT(FADD, "fadd");
-    FORMAT(FSUB, "fsub");
-    FORMAT(FMAX, "fmax");
-    FORMAT(FMIN, "fmin");
-    FORMAT(FMAXNM, "fmaxnm");
-    FORMAT(FMINNM, "fminnm");
-    FORMAT(FNMUL, "fnmul");
-    #undef FORMAT
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPDataProcessing3Source(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Fd, 'Fn, 'Fm, 'Fa";
-
-  switch (instr->Mask(FPDataProcessing3SourceMask)) {
-    #define FORMAT(A, B)  \
-    case A##_s:           \
-    case A##_d: mnemonic = B; break;
-    FORMAT(FMADD, "fmadd");
-    FORMAT(FMSUB, "fmsub");
-    FORMAT(FNMADD, "fnmadd");
-    FORMAT(FNMSUB, "fnmsub");
-    #undef FORMAT
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPImmediate(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "(FPImmediate)";
-
-  switch (instr->Mask(FPImmediateMask)) {
-    case FMOV_s_imm: mnemonic = "fmov"; form = "'Sd, 'IFPSingle"; break;
-    case FMOV_d_imm: mnemonic = "fmov"; form = "'Dd, 'IFPDouble"; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPIntegerConvert(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "(FPIntegerConvert)";
-  const char *form_rf = "'Rd, 'Fn";
-  const char *form_fr = "'Fd, 'Rn";
-
-  switch (instr->Mask(FPIntegerConvertMask)) {
-    case FMOV_ws:
-    case FMOV_xd: mnemonic = "fmov"; form = form_rf; break;
-    case FMOV_sw:
-    case FMOV_dx: mnemonic = "fmov"; form = form_fr; break;
-    case FCVTAS_ws:
-    case FCVTAS_xs:
-    case FCVTAS_wd:
-    case FCVTAS_xd: mnemonic = "fcvtas"; form = form_rf; break;
-    case FCVTAU_ws:
-    case FCVTAU_xs:
-    case FCVTAU_wd:
-    case FCVTAU_xd: mnemonic = "fcvtau"; form = form_rf; break;
-    case FCVTMS_ws:
-    case FCVTMS_xs:
-    case FCVTMS_wd:
-    case FCVTMS_xd: mnemonic = "fcvtms"; form = form_rf; break;
-    case FCVTMU_ws:
-    case FCVTMU_xs:
-    case FCVTMU_wd:
-    case FCVTMU_xd: mnemonic = "fcvtmu"; form = form_rf; break;
-    case FCVTNS_ws:
-    case FCVTNS_xs:
-    case FCVTNS_wd:
-    case FCVTNS_xd: mnemonic = "fcvtns"; form = form_rf; break;
-    case FCVTNU_ws:
-    case FCVTNU_xs:
-    case FCVTNU_wd:
-    case FCVTNU_xd: mnemonic = "fcvtnu"; form = form_rf; break;
-    case FCVTZU_xd:
-    case FCVTZU_ws:
-    case FCVTZU_wd:
-    case FCVTZU_xs: mnemonic = "fcvtzu"; form = form_rf; break;
-    case FCVTZS_xd:
-    case FCVTZS_wd:
-    case FCVTZS_xs:
-    case FCVTZS_ws: mnemonic = "fcvtzs"; form = form_rf; break;
-    case SCVTF_sw:
-    case SCVTF_sx:
-    case SCVTF_dw:
-    case SCVTF_dx: mnemonic = "scvtf"; form = form_fr; break;
-    case UCVTF_sw:
-    case UCVTF_sx:
-    case UCVTF_dw:
-    case UCVTF_dx: mnemonic = "ucvtf"; form = form_fr; break;
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitFPFixedPointConvert(const Instruction* instr) {
-  const char *mnemonic = "";
-  const char *form = "'Rd, 'Fn, 'IFPFBits";
-  const char *form_fr = "'Fd, 'Rn, 'IFPFBits";
-
-  switch (instr->Mask(FPFixedPointConvertMask)) {
-    case FCVTZS_ws_fixed:
-    case FCVTZS_xs_fixed:
-    case FCVTZS_wd_fixed:
-    case FCVTZS_xd_fixed: mnemonic = "fcvtzs"; break;
-    case FCVTZU_ws_fixed:
-    case FCVTZU_xs_fixed:
-    case FCVTZU_wd_fixed:
-    case FCVTZU_xd_fixed: mnemonic = "fcvtzu"; break;
-    case SCVTF_sw_fixed:
-    case SCVTF_sx_fixed:
-    case SCVTF_dw_fixed:
-    case SCVTF_dx_fixed: mnemonic = "scvtf"; form = form_fr; break;
-    case UCVTF_sw_fixed:
-    case UCVTF_sx_fixed:
-    case UCVTF_dw_fixed:
-    case UCVTF_dx_fixed: mnemonic = "ucvtf"; form = form_fr; break;
-    default: VIXL_UNREACHABLE();
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitSystem(const Instruction* instr) {
-  // Some system instructions hijack their Op and Cp fields to represent a
-  // range of immediates instead of indicating a different instruction. This
-  // makes the decoding tricky.
-  const char *mnemonic = "unimplemented";
-  const char *form = "(System)";
-
-  if (instr->Mask(SystemExclusiveMonitorFMask) == SystemExclusiveMonitorFixed) {
-    switch (instr->Mask(SystemExclusiveMonitorMask)) {
-      case CLREX: {
-        mnemonic = "clrex";
-        form = (instr->CRm() == 0xf) ? NULL : "'IX";
-        break;
-      }
-    }
-  } else if (instr->Mask(SystemSysRegFMask) == SystemSysRegFixed) {
-    switch (instr->Mask(SystemSysRegMask)) {
-      case MRS: {
-        mnemonic = "mrs";
-        switch (instr->ImmSystemRegister()) {
-          case NZCV: form = "'Xt, nzcv"; break;
-          case FPCR: form = "'Xt, fpcr"; break;
-          default: form = "'Xt, (unknown)"; break;
-        }
-        break;
-      }
-      case MSR: {
-        mnemonic = "msr";
-        switch (instr->ImmSystemRegister()) {
-          case NZCV: form = "nzcv, 'Xt"; break;
-          case FPCR: form = "fpcr, 'Xt"; break;
-          default: form = "(unknown), 'Xt"; break;
-        }
-        break;
-      }
-    }
-  } else if (instr->Mask(SystemHintFMask) == SystemHintFixed) {
-    switch (instr->ImmHint()) {
-      case NOP: {
-        mnemonic = "nop";
-        form = NULL;
-        break;
-      }
-    }
-  } else if (instr->Mask(MemBarrierFMask) == MemBarrierFixed) {
-    switch (instr->Mask(MemBarrierMask)) {
-      case DMB: {
-        mnemonic = "dmb";
-        form = "'M";
-        break;
-      }
-      case DSB: {
-        mnemonic = "dsb";
-        form = "'M";
-        break;
-      }
-      case ISB: {
-        mnemonic = "isb";
-        form = NULL;
-        break;
-      }
-    }
-  }
-
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitException(const Instruction* instr) {
-  const char *mnemonic = "unimplemented";
-  const char *form = "'IDebug";
-
-  switch (instr->Mask(ExceptionMask)) {
-    case HLT: mnemonic = "hlt"; break;
-    case BRK: mnemonic = "brk"; break;
-    case SVC: mnemonic = "svc"; break;
-    case HVC: mnemonic = "hvc"; break;
-    case SMC: mnemonic = "smc"; break;
-    case DCPS1: mnemonic = "dcps1"; form = "{'IDebug}"; break;
-    case DCPS2: mnemonic = "dcps2"; form = "{'IDebug}"; break;
-    case DCPS3: mnemonic = "dcps3"; form = "{'IDebug}"; break;
-    default: form = "(Exception)";
-  }
-  Format(instr, mnemonic, form);
-}
-
-
-void Disassembler::VisitUnimplemented(const Instruction* instr) {
-  Format(instr, "unimplemented", "(Unimplemented)");
-}
-
-
-void Disassembler::VisitUnallocated(const Instruction* instr) {
-  Format(instr, "unallocated", "(Unallocated)");
-}
-
-
-void Disassembler::ProcessOutput(const Instruction* /*instr*/) {
-  // The base disasm does nothing more than disassembling into a buffer.
-}
-
-
-void Disassembler::AppendRegisterNameToOutput(const Instruction* instr,
-                                              const CPURegister& reg) {
-  USE(instr);
-  VIXL_ASSERT(reg.IsValid());
-  char reg_char;
-
-  if (reg.IsRegister()) {
-    reg_char = reg.Is64Bits() ? 'x' : 'w';
-  } else {
-    VIXL_ASSERT(reg.IsFPRegister());
-    reg_char = reg.Is64Bits() ? 'd' : 's';
-  }
-
-  if (reg.IsFPRegister() || !(reg.Aliases(sp) || reg.Aliases(xzr))) {
-    // A normal register: w0 - w30, x0 - x30, s0 - s31, d0 - d31.
-    AppendToOutput("%c%d", reg_char, reg.code());
-  } else if (reg.Aliases(sp)) {
-    // Disassemble w31/x31 as stack pointer wsp/sp.
-    AppendToOutput("%s", reg.Is64Bits() ? "sp" : "wsp");
-  } else {
-    // Disassemble w31/x31 as zero register wzr/xzr.
-    AppendToOutput("%czr", reg_char);
-  }
-}
-
-
-void Disassembler::AppendPCRelativeOffsetToOutput(const Instruction* instr,
-                                                  int64_t offset) {
-  USE(instr);
-  char sign = (offset < 0) ? '-' : '+';
-  AppendToOutput("#%c0x%" PRIx64, sign, std::abs(offset));
-}
-
-
-void Disassembler::AppendAddressToOutput(const Instruction* instr,
-                                         const void* addr) {
-  USE(instr);
-  AppendToOutput("(addr 0x%" PRIxPTR ")", reinterpret_cast<uintptr_t>(addr));
-}
-
-
-void Disassembler::AppendCodeAddressToOutput(const Instruction* instr,
-                                             const void* addr) {
-  AppendAddressToOutput(instr, addr);
-}
-
-
-void Disassembler::AppendDataAddressToOutput(const Instruction* instr,
-                                             const void* addr) {
-  AppendAddressToOutput(instr, addr);
-}
-
-
-void Disassembler::AppendCodeRelativeAddressToOutput(const Instruction* instr,
-                                                     const void* addr) {
-  USE(instr);
-  int64_t rel_addr = CodeRelativeAddress(addr);
-  if (rel_addr >= 0) {
-    AppendToOutput("(addr 0x%" PRIx64 ")", rel_addr);
-  } else {
-    AppendToOutput("(addr -0x%" PRIx64 ")", -rel_addr);
-  }
-}
-
-
-void Disassembler::AppendCodeRelativeCodeAddressToOutput(
-    const Instruction* instr, const void* addr) {
-  AppendCodeRelativeAddressToOutput(instr, addr);
-}
-
-
-void Disassembler::AppendCodeRelativeDataAddressToOutput(
-    const Instruction* instr, const void* addr) {
-  AppendCodeRelativeAddressToOutput(instr, addr);
-}
-
-
-void Disassembler::MapCodeAddress(int64_t base_address,
-                                  const Instruction* instr_address) {
-  set_code_address_offset(
-      base_address - reinterpret_cast<intptr_t>(instr_address));
-}
-int64_t Disassembler::CodeRelativeAddress(const void* addr) {
-  return reinterpret_cast<intptr_t>(addr) + code_address_offset();
-}
-
-
-void Disassembler::Format(const Instruction* instr, const char* mnemonic,
-                          const char* format) {
-  VIXL_ASSERT(mnemonic != NULL);
-  ResetOutput();
-  Substitute(instr, mnemonic);
-  if (format != NULL) {
-    buffer_[buffer_pos_++] = ' ';
-    Substitute(instr, format);
-  }
-  buffer_[buffer_pos_] = 0;
-  ProcessOutput(instr);
-}
-
-
-void Disassembler::Substitute(const Instruction* instr, const char* string) {
-  char chr = *string++;
-  while (chr != '\0') {
-    if (chr == '\'') {
-      string += SubstituteField(instr, string);
-    } else {
-      buffer_[buffer_pos_++] = chr;
-    }
-    chr = *string++;
-  }
-}
-
-
-int Disassembler::SubstituteField(const Instruction* instr,
-                                  const char* format) {
-  switch (format[0]) {
-    case 'R':  // Register. X or W, selected by sf bit.
-    case 'F':  // FP Register. S or D, selected by type field.
-    case 'W':
-    case 'X':
-    case 'S':
-    case 'D': return SubstituteRegisterField(instr, format);
-    case 'I': return SubstituteImmediateField(instr, format);
-    case 'L': return SubstituteLiteralField(instr, format);
-    case 'H': return SubstituteShiftField(instr, format);
-    case 'P': return SubstitutePrefetchField(instr, format);
-    case 'C': return SubstituteConditionField(instr, format);
-    case 'E': return SubstituteExtendField(instr, format);
-    case 'A': return SubstitutePCRelAddressField(instr, format);
-    case 'B': return SubstituteBranchTargetField(instr, format);
-    case 'O': return SubstituteLSRegOffsetField(instr, format);
-    case 'M': return SubstituteBarrierField(instr, format);
-    default: {
-      VIXL_UNREACHABLE();
-      return 1;
-    }
-  }
-}
-
-
-int Disassembler::SubstituteRegisterField(const Instruction* instr,
-                                          const char* format) {
-  unsigned reg_num = 0;
-  unsigned field_len = 2;
-  switch (format[1]) {
-    case 'd': reg_num = instr->Rd(); break;
-    case 'n': reg_num = instr->Rn(); break;
-    case 'm': reg_num = instr->Rm(); break;
-    case 'a': reg_num = instr->Ra(); break;
-    case 's': reg_num = instr->Rs(); break;
-    case 't': {
-      if (format[2] == '2') {
-        reg_num = instr->Rt2();
-        field_len = 3;
-      } else {
-        reg_num = instr->Rt();
-      }
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-
-  // Increase field length for registers tagged as stack.
-  if (format[2] == 's') {
-    field_len = 3;
-  }
-
-  CPURegister::RegisterType reg_type;
-  unsigned reg_size;
-
-  if (format[0] == 'R') {
-    // Register type is R: use sf bit to choose X and W.
-    reg_type = CPURegister::kRegister;
-    reg_size = instr->SixtyFourBits() ? kXRegSize : kWRegSize;
-  } else if (format[0] == 'F') {
-    // Floating-point register: use type field to choose S or D.
-    reg_type = CPURegister::kFPRegister;
-    reg_size = ((instr->FPType() & 1) == 0) ? kSRegSize : kDRegSize;
-  } else {
-    // The register type is specified.
-    switch (format[0]) {
-      case 'W':
-        reg_type = CPURegister::kRegister; reg_size = kWRegSize; break;
-      case 'X':
-        reg_type = CPURegister::kRegister; reg_size = kXRegSize; break;
-      case 'S':
-        reg_type = CPURegister::kFPRegister; reg_size = kSRegSize; break;
-      case 'D':
-        reg_type = CPURegister::kFPRegister; reg_size = kDRegSize; break;
-      default:
-        VIXL_UNREACHABLE();
-        reg_type = CPURegister::kRegister;
-        reg_size = kXRegSize;
-    }
-  }
-
-  if ((reg_type == CPURegister::kRegister) &&
-      (reg_num == kZeroRegCode) && (format[2] == 's')) {
-    reg_num = kSPRegInternalCode;
-  }
-
-  AppendRegisterNameToOutput(instr, CPURegister(reg_num, reg_size, reg_type));
-
-  return field_len;
-}
-
-
-int Disassembler::SubstituteImmediateField(const Instruction* instr,
-                                           const char* format) {
-  VIXL_ASSERT(format[0] == 'I');
-
-  switch (format[1]) {
-    case 'M': {  // IMoveImm, IMoveNeg or IMoveLSL.
-      if (format[5] == 'L') {
-        AppendToOutput("#0x%" PRIx64, instr->ImmMoveWide());
-        if (instr->ShiftMoveWide() > 0) {
-          AppendToOutput(", lsl #%" PRId64, 16 * instr->ShiftMoveWide());
-        }
-      } else {
-        VIXL_ASSERT((format[5] == 'I') || (format[5] == 'N'));
-        uint64_t imm = instr->ImmMoveWide() << (16 * instr->ShiftMoveWide());
-        if (format[5] == 'N')
-          imm = ~imm;
-        if (!instr->SixtyFourBits())
-          imm &= UINT64_C(0xffffffff);
-        AppendToOutput("#0x%" PRIx64, imm);
-      }
-      return 8;
-    }
-    case 'L': {
-      switch (format[2]) {
-        case 'L': {  // ILLiteral - Immediate Load Literal.
-          AppendToOutput("pc%+" PRId64,
-                         instr->ImmLLiteral() << kLiteralEntrySizeLog2);
-          return 9;
-        }
-        case 'S': {  // ILS - Immediate Load/Store.
-          if (instr->ImmLS() != 0) {
-            AppendToOutput(", #%" PRId64, instr->ImmLS());
-          }
-          return 3;
-        }
-        case 'P': {  // ILPx - Immediate Load/Store Pair, x = access size.
-          if (instr->ImmLSPair() != 0) {
-            // format[3] is the scale value. Convert to a number.
-            int scale = format[3] - 0x30;
-            AppendToOutput(", #%" PRId64, instr->ImmLSPair() * scale);
-          }
-          return 4;
-        }
-        case 'U': {  // ILU - Immediate Load/Store Unsigned.
-          if (instr->ImmLSUnsigned() != 0) {
-            AppendToOutput(", #%" PRIu64,
-                           instr->ImmLSUnsigned() << instr->SizeLS());
-          }
-          return 3;
-        }
-      }
-    }
-    case 'C': {  // ICondB - Immediate Conditional Branch.
-      int64_t offset = instr->ImmCondBranch() << 2;
-      AppendPCRelativeOffsetToOutput(instr, offset);
-      return 6;
-    }
-    case 'A': {  // IAddSub.
-      VIXL_ASSERT(instr->ShiftAddSub() <= 1);
-      int64_t imm = instr->ImmAddSub() << (12 * instr->ShiftAddSub());
-      AppendToOutput("#0x%" PRIx64 " (%" PRId64 ")", imm, imm);
-      return 7;
-    }
-    case 'F': {  // IFPSingle, IFPDouble or IFPFBits.
-      if (format[3] == 'F') {  // IFPFbits.
-        AppendToOutput("#%" PRId64, 64 - instr->FPScale());
-        return 8;
-      } else {
-        AppendToOutput("#0x%" PRIx64 " (%.4f)", instr->ImmFP(),
-                       format[3] == 'S' ? instr->ImmFP32() : instr->ImmFP64());
-        return 9;
-      }
-    }
-    case 'T': {  // ITri - Immediate Triangular Encoded.
-      AppendToOutput("#0x%" PRIx64, instr->ImmLogical());
-      return 4;
-    }
-    case 'N': {  // INzcv.
-      int nzcv = (instr->Nzcv() << Flags_offset);
-      AppendToOutput("#%c%c%c%c", ((nzcv & NFlag) == 0) ? 'n' : 'N',
-                                  ((nzcv & ZFlag) == 0) ? 'z' : 'Z',
-                                  ((nzcv & CFlag) == 0) ? 'c' : 'C',
-                                  ((nzcv & VFlag) == 0) ? 'v' : 'V');
-      return 5;
-    }
-    case 'P': {  // IP - Conditional compare.
-      AppendToOutput("#%" PRId64, instr->ImmCondCmp());
-      return 2;
-    }
-    case 'B': {  // Bitfields.
-      return SubstituteBitfieldImmediateField(instr, format);
-    }
-    case 'E': {  // IExtract.
-      AppendToOutput("#%" PRId64, instr->ImmS());
-      return 8;
-    }
-    case 'S': {  // IS - Test and branch bit.
-      AppendToOutput("#%" PRId64, (instr->ImmTestBranchBit5() << 5) |
-                                  instr->ImmTestBranchBit40());
-      return 2;
-    }
-    case 'D': {  // IDebug - HLT and BRK instructions.
-      AppendToOutput("#0x%" PRIx64, instr->ImmException());
-      return 6;
-    }
-    case 'X': {  // IX - CLREX instruction.
-      AppendToOutput("#0x%" PRIx64, instr->CRm());
-      return 2;
-    }
-    default: {
-      VIXL_UNIMPLEMENTED();
-      return 0;
-    }
-  }
-}
-
-
-int Disassembler::SubstituteBitfieldImmediateField(const Instruction* instr,
-                                                   const char* format) {
-  VIXL_ASSERT((format[0] == 'I') && (format[1] == 'B'));
-  unsigned r = instr->ImmR();
-  unsigned s = instr->ImmS();
-
-  switch (format[2]) {
-    case 'r': {  // IBr.
-      AppendToOutput("#%d", r);
-      return 3;
-    }
-    case 's': {  // IBs+1 or IBs-r+1.
-      if (format[3] == '+') {
-        AppendToOutput("#%d", s + 1);
-        return 5;
-      } else {
-        VIXL_ASSERT(format[3] == '-');
-        AppendToOutput("#%d", s - r + 1);
-        return 7;
-      }
-    }
-    case 'Z': {  // IBZ-r.
-      VIXL_ASSERT((format[3] == '-') && (format[4] == 'r'));
-      unsigned reg_size = (instr->SixtyFourBits() == 1) ? kXRegSize : kWRegSize;
-      AppendToOutput("#%d", reg_size - r);
-      return 5;
-    }
-    default: {
-      VIXL_UNREACHABLE();
-      return 0;
-    }
-  }
-}
-
-
-int Disassembler::SubstituteLiteralField(const Instruction* instr,
-                                         const char* format) {
-  VIXL_ASSERT(strncmp(format, "LValue", 6) == 0);
-  USE(format);
-
-  const void * address = instr->LiteralAddress<const void *>();
-  switch (instr->Mask(LoadLiteralMask)) {
-    case LDR_w_lit:
-    case LDR_x_lit:
-    case LDRSW_x_lit:
-    case LDR_s_lit:
-    case LDR_d_lit:
-      AppendCodeRelativeDataAddressToOutput(instr, address);
-      break;
-    case PRFM_lit: {
-      // Use the prefetch hint to decide how to print the address.
-      switch (instr->PrefetchHint()) {
-        case 0x0:     // PLD: prefetch for load.
-        case 0x2:     // PST: prepare for store.
-          AppendCodeRelativeDataAddressToOutput(instr, address);
-          break;
-        case 0x1:     // PLI: preload instructions.
-          AppendCodeRelativeCodeAddressToOutput(instr, address);
-          break;
-        case 0x3:     // Unallocated hint.
-          AppendCodeRelativeAddressToOutput(instr, address);
-          break;
-      }
-      break;
-    }
-    default:
-      VIXL_UNREACHABLE();
-  }
-
-  return 6;
-}
-
-
-int Disassembler::SubstituteShiftField(const Instruction* instr,
-                                       const char* format) {
-  VIXL_ASSERT(format[0] == 'H');
-  VIXL_ASSERT(instr->ShiftDP() <= 0x3);
-
-  switch (format[1]) {
-    case 'D': {  // HDP.
-      VIXL_ASSERT(instr->ShiftDP() != ROR);
-    }  // Fall through.
-    case 'L': {  // HLo.
-      if (instr->ImmDPShift() != 0) {
-        const char* shift_type[] = {"lsl", "lsr", "asr", "ror"};
-        AppendToOutput(", %s #%" PRId64, shift_type[instr->ShiftDP()],
-                       instr->ImmDPShift());
-      }
-      return 3;
-    }
-    default:
-      VIXL_UNIMPLEMENTED();
-      return 0;
-  }
-}
-
-
-int Disassembler::SubstituteConditionField(const Instruction* instr,
-                                           const char* format) {
-  VIXL_ASSERT(format[0] == 'C');
-  const char* condition_code[] = { "eq", "ne", "hs", "lo",
-                                   "mi", "pl", "vs", "vc",
-                                   "hi", "ls", "ge", "lt",
-                                   "gt", "le", "al", "nv" };
-  int cond;
-  switch (format[1]) {
-    case 'B': cond = instr->ConditionBranch(); break;
-    case 'I': {
-      cond = InvertCondition(static_cast<Condition>(instr->Condition()));
-      break;
-    }
-    default: cond = instr->Condition();
-  }
-  AppendToOutput("%s", condition_code[cond]);
-  return 4;
-}
-
-
-int Disassembler::SubstitutePCRelAddressField(const Instruction* instr,
-                                              const char* format) {
-  VIXL_ASSERT((strcmp(format, "AddrPCRelByte") == 0) ||   // Used by `adr`.
-              (strcmp(format, "AddrPCRelPage") == 0));    // Used by `adrp`.
-
-  int64_t offset = instr->ImmPCRel();
-
-  // Compute the target address based on the effective address (after applying
-  // code_address_offset). This is required for correct behaviour of adrp.
-  const Instruction* base = instr + code_address_offset();
-  if (format[9] == 'P') {
-    offset *= kPageSize;
-    base = AlignDown(base, kPageSize);
-  }
-  // Strip code_address_offset before printing, so we can use the
-  // semantically-correct AppendCodeRelativeAddressToOutput.
-  const void* target =
-      reinterpret_cast<const void*>(base + offset - code_address_offset());
-
-  AppendPCRelativeOffsetToOutput(instr, offset);
-  AppendToOutput(" ");
-  AppendCodeRelativeAddressToOutput(instr, target);
-  return 13;
-}
-
-
-int Disassembler::SubstituteBranchTargetField(const Instruction* instr,
-                                              const char* format) {
-  VIXL_ASSERT(strncmp(format, "BImm", 4) == 0);
-
-  int64_t offset = 0;
-  switch (format[5]) {
-    // BImmUncn - unconditional branch immediate.
-    case 'n': offset = instr->ImmUncondBranch(); break;
-    // BImmCond - conditional branch immediate.
-    case 'o': offset = instr->ImmCondBranch(); break;
-    // BImmCmpa - compare and branch immediate.
-    case 'm': offset = instr->ImmCmpBranch(); break;
-    // BImmTest - test and branch immediate.
-    case 'e': offset = instr->ImmTestBranch(); break;
-    default: VIXL_UNIMPLEMENTED();
-  }
-  offset <<= kInstructionSizeLog2;
-  const void* target_address = reinterpret_cast<const void*>(instr + offset);
-  VIXL_STATIC_ASSERT(sizeof(*instr) == 1);
-
-  AppendPCRelativeOffsetToOutput(instr, offset);
-  AppendToOutput(" ");
-  AppendCodeRelativeCodeAddressToOutput(instr, target_address);
-
-  return 8;
-}
-
-
-int Disassembler::SubstituteExtendField(const Instruction* instr,
-                                        const char* format) {
-  VIXL_ASSERT(strncmp(format, "Ext", 3) == 0);
-  VIXL_ASSERT(instr->ExtendMode() <= 7);
-  USE(format);
-
-  const char* extend_mode[] = { "uxtb", "uxth", "uxtw", "uxtx",
-                                "sxtb", "sxth", "sxtw", "sxtx" };
-
-  // If rd or rn is SP, uxtw on 32-bit registers and uxtx on 64-bit
-  // registers becomes lsl.
-  if (((instr->Rd() == kZeroRegCode) || (instr->Rn() == kZeroRegCode)) &&
-      (((instr->ExtendMode() == UXTW) && (instr->SixtyFourBits() == 0)) ||
-       (instr->ExtendMode() == UXTX))) {
-    if (instr->ImmExtendShift() > 0) {
-      AppendToOutput(", lsl #%" PRId64, instr->ImmExtendShift());
-    }
-  } else {
-    AppendToOutput(", %s", extend_mode[instr->ExtendMode()]);
-    if (instr->ImmExtendShift() > 0) {
-      AppendToOutput(" #%" PRId64, instr->ImmExtendShift());
-    }
-  }
-  return 3;
-}
-
-
-int Disassembler::SubstituteLSRegOffsetField(const Instruction* instr,
-                                             const char* format) {
-  VIXL_ASSERT(strncmp(format, "Offsetreg", 9) == 0);
-  const char* extend_mode[] = { "undefined", "undefined", "uxtw", "lsl",
-                                "undefined", "undefined", "sxtw", "sxtx" };
-  USE(format);
-
-  unsigned shift = instr->ImmShiftLS();
-  Extend ext = static_cast<Extend>(instr->ExtendMode());
-  char reg_type = ((ext == UXTW) || (ext == SXTW)) ? 'w' : 'x';
-
-  unsigned rm = instr->Rm();
-  if (rm == kZeroRegCode) {
-    AppendToOutput("%czr", reg_type);
-  } else {
-    AppendToOutput("%c%d", reg_type, rm);
-  }
-
-  // Extend mode UXTX is an alias for shift mode LSL here.
-  if (!((ext == UXTX) && (shift == 0))) {
-    AppendToOutput(", %s", extend_mode[ext]);
-    if (shift != 0) {
-      AppendToOutput(" #%" PRId64, instr->SizeLS());
-    }
-  }
-  return 9;
-}
-
-
-int Disassembler::SubstitutePrefetchField(const Instruction* instr,
-                                          const char* format) {
-  VIXL_ASSERT(format[0] == 'P');
-  USE(format);
-
-  static const char* hints[] = {"ld", "li", "st"};
-  static const char* stream_options[] = {"keep", "strm"};
-
-  unsigned hint = instr->PrefetchHint();
-  unsigned target = instr->PrefetchTarget() + 1;
-  unsigned stream = instr->PrefetchStream();
-
-  if ((hint >= (sizeof(hints) / sizeof(hints[0]))) || (target > 3)) {
-    // Unallocated prefetch operations.
-    int prefetch_mode = instr->ImmPrefetchOperation();
-    AppendToOutput("#0b%c%c%c%c%c",
-                   (prefetch_mode & (1 << 4)) ? '1' : '0',
-                   (prefetch_mode & (1 << 3)) ? '1' : '0',
-                   (prefetch_mode & (1 << 2)) ? '1' : '0',
-                   (prefetch_mode & (1 << 1)) ? '1' : '0',
-                   (prefetch_mode & (1 << 0)) ? '1' : '0');
-  } else {
-    VIXL_ASSERT(stream < (sizeof(stream_options) / sizeof(stream_options[0])));
-    AppendToOutput("p%sl%d%s", hints[hint], target, stream_options[stream]);
-  }
-  return 6;
-}
-
-int Disassembler::SubstituteBarrierField(const Instruction* instr,
-                                         const char* format) {
-  VIXL_ASSERT(format[0] == 'M');
-  USE(format);
-
-  static const char* options[4][4] = {
-    { "sy (0b0000)", "oshld", "oshst", "osh" },
-    { "sy (0b0100)", "nshld", "nshst", "nsh" },
-    { "sy (0b1000)", "ishld", "ishst", "ish" },
-    { "sy (0b1100)", "ld", "st", "sy" }
-  };
-  int domain = instr->ImmBarrierDomain();
-  int type = instr->ImmBarrierType();
-
-  AppendToOutput("%s", options[domain][type]);
-  return 1;
-}
-
-void Disassembler::ResetOutput() {
-  buffer_pos_ = 0;
-  buffer_[buffer_pos_] = 0;
-}
-
-
-void Disassembler::AppendToOutput(const char* format, ...) {
-  va_list args;
-  va_start(args, format);
-  buffer_pos_ += vsnprintf(&buffer_[buffer_pos_], buffer_size_, format, args);
-  va_end(args);
-}
-
-
-void PrintDisassembler::ProcessOutput(const Instruction* instr) {
-  fprintf(stream_, "0x%016" PRIx64 "  %08" PRIx32 "\t\t%s\n",
-          reinterpret_cast<uint64_t>(instr),
-          instr->InstructionBits(),
-          GetOutput());
-}
-}  // namespace vixl
diff --git a/disas/libvixl/a64/instructions-a64.cc b/disas/libvixl/a64/instructions-a64.cc
deleted file mode 100644
index b091886838..0000000000
--- a/disas/libvixl/a64/instructions-a64.cc
+++ /dev/null
@@ -1,314 +0,0 @@
-// Copyright 2013, 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.
-
-#include "a64/instructions-a64.h"
-#include "a64/assembler-a64.h"
-
-namespace vixl {
-
-
-// Floating-point infinity values.
-const float kFP32PositiveInfinity = rawbits_to_float(0x7f800000);
-const float kFP32NegativeInfinity = rawbits_to_float(0xff800000);
-const double kFP64PositiveInfinity =
-    rawbits_to_double(UINT64_C(0x7ff0000000000000));
-const double kFP64NegativeInfinity =
-    rawbits_to_double(UINT64_C(0xfff0000000000000));
-
-
-// The default NaN values (for FPCR.DN=1).
-const double kFP64DefaultNaN = rawbits_to_double(UINT64_C(0x7ff8000000000000));
-const float kFP32DefaultNaN = rawbits_to_float(0x7fc00000);
-
-
-static uint64_t RotateRight(uint64_t value,
-                            unsigned int rotate,
-                            unsigned int width) {
-  VIXL_ASSERT(width <= 64);
-  rotate &= 63;
-  return ((value & ((UINT64_C(1) << rotate) - 1)) <<
-          (width - rotate)) | (value >> rotate);
-}
-
-
-static uint64_t RepeatBitsAcrossReg(unsigned reg_size,
-                                    uint64_t value,
-                                    unsigned width) {
-  VIXL_ASSERT((width == 2) || (width == 4) || (width == 8) || (width == 16) ||
-              (width == 32));
-  VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
-  uint64_t result = value & ((UINT64_C(1) << width) - 1);
-  for (unsigned i = width; i < reg_size; i *= 2) {
-    result |= (result << i);
-  }
-  return result;
-}
-
-
-bool Instruction::IsLoad() const {
-  if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) {
-    return false;
-  }
-
-  if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) {
-    return Mask(LoadStorePairLBit) != 0;
-  } else {
-    LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreOpMask));
-    switch (op) {
-      case LDRB_w:
-      case LDRH_w:
-      case LDR_w:
-      case LDR_x:
-      case LDRSB_w:
-      case LDRSB_x:
-      case LDRSH_w:
-      case LDRSH_x:
-      case LDRSW_x:
-      case LDR_s:
-      case LDR_d: return true;
-      default: return false;
-    }
-  }
-}
-
-
-bool Instruction::IsStore() const {
-  if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) {
-    return false;
-  }
-
-  if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) {
-    return Mask(LoadStorePairLBit) == 0;
-  } else {
-    LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreOpMask));
-    switch (op) {
-      case STRB_w:
-      case STRH_w:
-      case STR_w:
-      case STR_x:
-      case STR_s:
-      case STR_d: return true;
-      default: return false;
-    }
-  }
-}
-
-
-// Logical immediates can't encode zero, so a return value of zero is used to
-// indicate a failure case. Specifically, where the constraints on imm_s are
-// not met.
-uint64_t Instruction::ImmLogical() const {
-  unsigned reg_size = SixtyFourBits() ? kXRegSize : kWRegSize;
-  int64_t n = BitN();
-  int64_t imm_s = ImmSetBits();
-  int64_t imm_r = ImmRotate();
-
-  // An integer is constructed from the n, imm_s and imm_r bits according to
-  // the following table:
-  //
-  //  N   imms    immr    size        S             R
-  //  1  ssssss  rrrrrr    64    UInt(ssssss)  UInt(rrrrrr)
-  //  0  0sssss  xrrrrr    32    UInt(sssss)   UInt(rrrrr)
-  //  0  10ssss  xxrrrr    16    UInt(ssss)    UInt(rrrr)
-  //  0  110sss  xxxrrr     8    UInt(sss)     UInt(rrr)
-  //  0  1110ss  xxxxrr     4    UInt(ss)      UInt(rr)
-  //  0  11110s  xxxxxr     2    UInt(s)       UInt(r)
-  // (s bits must not be all set)
-  //
-  // A pattern is constructed of size bits, where the least significant S+1
-  // bits are set. The pattern is rotated right by R, and repeated across a
-  // 32 or 64-bit value, depending on destination register width.
-  //
-
-  if (n == 1) {
-    if (imm_s == 0x3F) {
-      return 0;
-    }
-    uint64_t bits = (UINT64_C(1) << (imm_s + 1)) - 1;
-    return RotateRight(bits, imm_r, 64);
-  } else {
-    if ((imm_s >> 1) == 0x1F) {
-      return 0;
-    }
-    for (int width = 0x20; width >= 0x2; width >>= 1) {
-      if ((imm_s & width) == 0) {
-        int mask = width - 1;
-        if ((imm_s & mask) == mask) {
-          return 0;
-        }
-        uint64_t bits = (UINT64_C(1) << ((imm_s & mask) + 1)) - 1;
-        return RepeatBitsAcrossReg(reg_size,
-                                   RotateRight(bits, imm_r & mask, width),
-                                   width);
-      }
-    }
-  }
-  VIXL_UNREACHABLE();
-  return 0;
-}
-
-
-float Instruction::ImmFP32() const {
-  //  ImmFP: abcdefgh (8 bits)
-  // Single: aBbb.bbbc.defg.h000.0000.0000.0000.0000 (32 bits)
-  // where B is b ^ 1
-  uint32_t bits = ImmFP();
-  uint32_t bit7 = (bits >> 7) & 0x1;
-  uint32_t bit6 = (bits >> 6) & 0x1;
-  uint32_t bit5_to_0 = bits & 0x3f;
-  uint32_t result = (bit7 << 31) | ((32 - bit6) << 25) | (bit5_to_0 << 19);
-
-  return rawbits_to_float(result);
-}
-
-
-double Instruction::ImmFP64() const {
-  //  ImmFP: abcdefgh (8 bits)
-  // Double: aBbb.bbbb.bbcd.efgh.0000.0000.0000.0000
-  //         0000.0000.0000.0000.0000.0000.0000.0000 (64 bits)
-  // where B is b ^ 1
-  uint32_t bits = ImmFP();
-  uint64_t bit7 = (bits >> 7) & 0x1;
-  uint64_t bit6 = (bits >> 6) & 0x1;
-  uint64_t bit5_to_0 = bits & 0x3f;
-  uint64_t result = (bit7 << 63) | ((256 - bit6) << 54) | (bit5_to_0 << 48);
-
-  return rawbits_to_double(result);
-}
-
-
-LSDataSize CalcLSPairDataSize(LoadStorePairOp op) {
-  switch (op) {
-    case STP_x:
-    case LDP_x:
-    case STP_d:
-    case LDP_d: return LSDoubleWord;
-    default: return LSWord;
-  }
-}
-
-
-const Instruction* Instruction::ImmPCOffsetTarget() const {
-  const Instruction * base = this;
-  ptrdiff_t offset;
-  if (IsPCRelAddressing()) {
-    // ADR and ADRP.
-    offset = ImmPCRel();
-    if (Mask(PCRelAddressingMask) == ADRP) {
-      base = AlignDown(base, kPageSize);
-      offset *= kPageSize;
-    } else {
-      VIXL_ASSERT(Mask(PCRelAddressingMask) == ADR);
-    }
-  } else {
-    // All PC-relative branches.
-    VIXL_ASSERT(BranchType() != UnknownBranchType);
-    // Relative branch offsets are instruction-size-aligned.
-    offset = ImmBranch() << kInstructionSizeLog2;
-  }
-  return base + offset;
-}
-
-
-inline int Instruction::ImmBranch() const {
-  switch (BranchType()) {
-    case CondBranchType: return ImmCondBranch();
-    case UncondBranchType: return ImmUncondBranch();
-    case CompareBranchType: return ImmCmpBranch();
-    case TestBranchType: return ImmTestBranch();
-    default: VIXL_UNREACHABLE();
-  }
-  return 0;
-}
-
-
-void Instruction::SetImmPCOffsetTarget(const Instruction* target) {
-  if (IsPCRelAddressing()) {
-    SetPCRelImmTarget(target);
-  } else {
-    SetBranchImmTarget(target);
-  }
-}
-
-
-void Instruction::SetPCRelImmTarget(const Instruction* target) {
-  int32_t imm21;
-  if ((Mask(PCRelAddressingMask) == ADR)) {
-    imm21 = target - this;
-  } else {
-    VIXL_ASSERT(Mask(PCRelAddressingMask) == ADRP);
-    uintptr_t this_page = reinterpret_cast<uintptr_t>(this) / kPageSize;
-    uintptr_t target_page = reinterpret_cast<uintptr_t>(target) / kPageSize;
-    imm21 = target_page - this_page;
-  }
-  Instr imm = Assembler::ImmPCRelAddress(imm21);
-
-  SetInstructionBits(Mask(~ImmPCRel_mask) | imm);
-}
-
-
-void Instruction::SetBranchImmTarget(const Instruction* target) {
-  VIXL_ASSERT(((target - this) & 3) == 0);
-  Instr branch_imm = 0;
-  uint32_t imm_mask = 0;
-  int offset = (target - this) >> kInstructionSizeLog2;
-  switch (BranchType()) {
-    case CondBranchType: {
-      branch_imm = Assembler::ImmCondBranch(offset);
-      imm_mask = ImmCondBranch_mask;
-      break;
-    }
-    case UncondBranchType: {
-      branch_imm = Assembler::ImmUncondBranch(offset);
-      imm_mask = ImmUncondBranch_mask;
-      break;
-    }
-    case CompareBranchType: {
-      branch_imm = Assembler::ImmCmpBranch(offset);
-      imm_mask = ImmCmpBranch_mask;
-      break;
-    }
-    case TestBranchType: {
-      branch_imm = Assembler::ImmTestBranch(offset);
-      imm_mask = ImmTestBranch_mask;
-      break;
-    }
-    default: VIXL_UNREACHABLE();
-  }
-  SetInstructionBits(Mask(~imm_mask) | branch_imm);
-}
-
-
-void Instruction::SetImmLLiteral(const Instruction* source) {
-  VIXL_ASSERT(IsWordAligned(source));
-  ptrdiff_t offset = (source - this) >> kLiteralEntrySizeLog2;
-  Instr imm = Assembler::ImmLLiteral(offset);
-  Instr mask = ImmLLiteral_mask;
-
-  SetInstructionBits(Mask(~mask) | imm);
-}
-}  // namespace vixl
-
diff --git a/disas/libvixl/a64/instructions-a64.h b/disas/libvixl/a64/instructions-a64.h
deleted file mode 100644
index f1d883ccc7..0000000000
--- a/disas/libvixl/a64/instructions-a64.h
+++ /dev/null
@@ -1,384 +0,0 @@
-// Copyright 2013, 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_INSTRUCTIONS_A64_H_
-#define VIXL_A64_INSTRUCTIONS_A64_H_
-
-#include "globals.h"
-#include "utils.h"
-#include "a64/constants-a64.h"
-
-namespace vixl {
-// ISA constants. --------------------------------------------------------------
-
-typedef uint32_t Instr;
-const unsigned kInstructionSize = 4;
-const unsigned kInstructionSizeLog2 = 2;
-const unsigned kLiteralEntrySize = 4;
-const unsigned kLiteralEntrySizeLog2 = 2;
-const unsigned kMaxLoadLiteralRange = 1 * MBytes;
-
-// This is the nominal page size (as used by the adrp instruction); the actual
-// size of the memory pages allocated by the kernel is likely to differ.
-const unsigned kPageSize = 4 * KBytes;
-const unsigned kPageSizeLog2 = 12;
-
-const unsigned kWRegSize = 32;
-const unsigned kWRegSizeLog2 = 5;
-const unsigned kWRegSizeInBytes = kWRegSize / 8;
-const unsigned kWRegSizeInBytesLog2 = kWRegSizeLog2 - 3;
-const unsigned kXRegSize = 64;
-const unsigned kXRegSizeLog2 = 6;
-const unsigned kXRegSizeInBytes = kXRegSize / 8;
-const unsigned kXRegSizeInBytesLog2 = kXRegSizeLog2 - 3;
-const unsigned kSRegSize = 32;
-const unsigned kSRegSizeLog2 = 5;
-const unsigned kSRegSizeInBytes = kSRegSize / 8;
-const unsigned kSRegSizeInBytesLog2 = kSRegSizeLog2 - 3;
-const unsigned kDRegSize = 64;
-const unsigned kDRegSizeLog2 = 6;
-const unsigned kDRegSizeInBytes = kDRegSize / 8;
-const unsigned kDRegSizeInBytesLog2 = kDRegSizeLog2 - 3;
-const uint64_t kWRegMask = UINT64_C(0xffffffff);
-const uint64_t kXRegMask = UINT64_C(0xffffffffffffffff);
-const uint64_t kSRegMask = UINT64_C(0xffffffff);
-const uint64_t kDRegMask = UINT64_C(0xffffffffffffffff);
-const uint64_t kSSignMask = UINT64_C(0x80000000);
-const uint64_t kDSignMask = UINT64_C(0x8000000000000000);
-const uint64_t kWSignMask = UINT64_C(0x80000000);
-const uint64_t kXSignMask = UINT64_C(0x8000000000000000);
-const uint64_t kByteMask = UINT64_C(0xff);
-const uint64_t kHalfWordMask = UINT64_C(0xffff);
-const uint64_t kWordMask = UINT64_C(0xffffffff);
-const uint64_t kXMaxUInt = UINT64_C(0xffffffffffffffff);
-const uint64_t kWMaxUInt = UINT64_C(0xffffffff);
-const int64_t kXMaxInt = INT64_C(0x7fffffffffffffff);
-const int64_t kXMinInt = INT64_C(0x8000000000000000);
-const int32_t kWMaxInt = INT32_C(0x7fffffff);
-const int32_t kWMinInt = INT32_C(0x80000000);
-const unsigned kLinkRegCode = 30;
-const unsigned kZeroRegCode = 31;
-const unsigned kSPRegInternalCode = 63;
-const unsigned kRegCodeMask = 0x1f;
-
-const unsigned kAddressTagOffset = 56;
-const unsigned kAddressTagWidth = 8;
-const uint64_t kAddressTagMask =
-    ((UINT64_C(1) << kAddressTagWidth) - 1) << kAddressTagOffset;
-VIXL_STATIC_ASSERT(kAddressTagMask == UINT64_C(0xff00000000000000));
-
-// AArch64 floating-point specifics. These match IEEE-754.
-const unsigned kDoubleMantissaBits = 52;
-const unsigned kDoubleExponentBits = 11;
-const unsigned kFloatMantissaBits = 23;
-const unsigned kFloatExponentBits = 8;
-
-// Floating-point infinity values.
-extern const float kFP32PositiveInfinity;
-extern const float kFP32NegativeInfinity;
-extern const double kFP64PositiveInfinity;
-extern const double kFP64NegativeInfinity;
-
-// The default NaN values (for FPCR.DN=1).
-extern const double kFP64DefaultNaN;
-extern const float kFP32DefaultNaN;
-
-
-enum LSDataSize {
-  LSByte        = 0,
-  LSHalfword    = 1,
-  LSWord        = 2,
-  LSDoubleWord  = 3
-};
-
-LSDataSize CalcLSPairDataSize(LoadStorePairOp op);
-
-enum ImmBranchType {
-  UnknownBranchType = 0,
-  CondBranchType    = 1,
-  UncondBranchType  = 2,
-  CompareBranchType = 3,
-  TestBranchType    = 4
-};
-
-enum AddrMode {
-  Offset,
-  PreIndex,
-  PostIndex
-};
-
-enum FPRounding {
-  // The first four values are encodable directly by FPCR<RMode>.
-  FPTieEven = 0x0,
-  FPPositiveInfinity = 0x1,
-  FPNegativeInfinity = 0x2,
-  FPZero = 0x3,
-
-  // The final rounding mode is only available when explicitly specified by the
-  // instruction (such as with fcvta). It cannot be set in FPCR.
-  FPTieAway
-};
-
-enum Reg31Mode {
-  Reg31IsStackPointer,
-  Reg31IsZeroRegister
-};
-
-// Instructions. ---------------------------------------------------------------
-
-class Instruction {
- public:
-  Instr InstructionBits() const {
-    return *(reinterpret_cast<const Instr*>(this));
-  }
-
-  void SetInstructionBits(Instr new_instr) {
-    *(reinterpret_cast<Instr*>(this)) = new_instr;
-  }
-
-  int Bit(int pos) const {
-    return (InstructionBits() >> pos) & 1;
-  }
-
-  uint32_t Bits(int msb, int lsb) const {
-    return unsigned_bitextract_32(msb, lsb, InstructionBits());
-  }
-
-  int32_t SignedBits(int msb, int lsb) const {
-    int32_t bits = *(reinterpret_cast<const int32_t*>(this));
-    return signed_bitextract_32(msb, lsb, bits);
-  }
-
-  Instr Mask(uint32_t mask) const {
-    return InstructionBits() & mask;
-  }
-
-  #define DEFINE_GETTER(Name, HighBit, LowBit, Func)             \
-  int64_t Name() const { return Func(HighBit, LowBit); }
-  INSTRUCTION_FIELDS_LIST(DEFINE_GETTER)
-  #undef DEFINE_GETTER
-
-  // ImmPCRel is a compound field (not present in INSTRUCTION_FIELDS_LIST),
-  // formed from ImmPCRelLo and ImmPCRelHi.
-  int ImmPCRel() const {
-    int const offset = ((ImmPCRelHi() << ImmPCRelLo_width) | ImmPCRelLo());
-    int const width = ImmPCRelLo_width + ImmPCRelHi_width;
-    return signed_bitextract_32(width-1, 0, offset);
-  }
-
-  uint64_t ImmLogical() const;
-  float ImmFP32() const;
-  double ImmFP64() const;
-
-  LSDataSize SizeLSPair() const {
-    return CalcLSPairDataSize(
-             static_cast<LoadStorePairOp>(Mask(LoadStorePairMask)));
-  }
-
-  // Helpers.
-  bool IsCondBranchImm() const {
-    return Mask(ConditionalBranchFMask) == ConditionalBranchFixed;
-  }
-
-  bool IsUncondBranchImm() const {
-    return Mask(UnconditionalBranchFMask) == UnconditionalBranchFixed;
-  }
-
-  bool IsCompareBranch() const {
-    return Mask(CompareBranchFMask) == CompareBranchFixed;
-  }
-
-  bool IsTestBranch() const {
-    return Mask(TestBranchFMask) == TestBranchFixed;
-  }
-
-  bool IsPCRelAddressing() const {
-    return Mask(PCRelAddressingFMask) == PCRelAddressingFixed;
-  }
-
-  bool IsLogicalImmediate() const {
-    return Mask(LogicalImmediateFMask) == LogicalImmediateFixed;
-  }
-
-  bool IsAddSubImmediate() const {
-    return Mask(AddSubImmediateFMask) == AddSubImmediateFixed;
-  }
-
-  bool IsAddSubExtended() const {
-    return Mask(AddSubExtendedFMask) == AddSubExtendedFixed;
-  }
-
-  bool IsLoadOrStore() const {
-    return Mask(LoadStoreAnyFMask) == LoadStoreAnyFixed;
-  }
-
-  bool IsLoad() const;
-  bool IsStore() const;
-
-  bool IsLoadLiteral() const {
-    // This includes PRFM_lit.
-    return Mask(LoadLiteralFMask) == LoadLiteralFixed;
-  }
-
-  bool IsMovn() const {
-    return (Mask(MoveWideImmediateMask) == MOVN_x) ||
-           (Mask(MoveWideImmediateMask) == MOVN_w);
-  }
-
-  // Indicate whether Rd can be the stack pointer or the zero register. This
-  // does not check that the instruction actually has an Rd field.
-  Reg31Mode RdMode() const {
-    // The following instructions use sp or wsp as Rd:
-    //  Add/sub (immediate) when not setting the flags.
-    //  Add/sub (extended) when not setting the flags.
-    //  Logical (immediate) when not setting the flags.
-    // Otherwise, r31 is the zero register.
-    if (IsAddSubImmediate() || IsAddSubExtended()) {
-      if (Mask(AddSubSetFlagsBit)) {
-        return Reg31IsZeroRegister;
-      } else {
-        return Reg31IsStackPointer;
-      }
-    }
-    if (IsLogicalImmediate()) {
-      // Of the logical (immediate) instructions, only ANDS (and its aliases)
-      // can set the flags. The others can all write into sp.
-      // Note that some logical operations are not available to
-      // immediate-operand instructions, so we have to combine two masks here.
-      if (Mask(LogicalImmediateMask & LogicalOpMask) == ANDS) {
-        return Reg31IsZeroRegister;
-      } else {
-        return Reg31IsStackPointer;
-      }
-    }
-    return Reg31IsZeroRegister;
-  }
-
-  // Indicate whether Rn can be the stack pointer or the zero register. This
-  // does not check that the instruction actually has an Rn field.
-  Reg31Mode RnMode() const {
-    // The following instructions use sp or wsp as Rn:
-    //  All loads and stores.
-    //  Add/sub (immediate).
-    //  Add/sub (extended).
-    // Otherwise, r31 is the zero register.
-    if (IsLoadOrStore() || IsAddSubImmediate() || IsAddSubExtended()) {
-      return Reg31IsStackPointer;
-    }
-    return Reg31IsZeroRegister;
-  }
-
-  ImmBranchType BranchType() const {
-    if (IsCondBranchImm()) {
-      return CondBranchType;
-    } else if (IsUncondBranchImm()) {
-      return UncondBranchType;
-    } else if (IsCompareBranch()) {
-      return CompareBranchType;
-    } else if (IsTestBranch()) {
-      return TestBranchType;
-    } else {
-      return UnknownBranchType;
-    }
-  }
-
-  // Find the target of this instruction. 'this' may be a branch or a
-  // PC-relative addressing instruction.
-  const Instruction* ImmPCOffsetTarget() const;
-
-  // Patch a PC-relative offset to refer to 'target'. 'this' may be a branch or
-  // a PC-relative addressing instruction.
-  void SetImmPCOffsetTarget(const Instruction* target);
-  // Patch a literal load instruction to load from 'source'.
-  void SetImmLLiteral(const Instruction* source);
-
-  // Calculate the address of a literal referred to by a load-literal
-  // instruction, and return it as the specified type.
-  //
-  // The literal itself is safely mutable only if the backing buffer is safely
-  // mutable.
-  template <typename T>
-  T LiteralAddress() const {
-    uint64_t base_raw = reinterpret_cast<uintptr_t>(this);
-    ptrdiff_t offset = ImmLLiteral() << kLiteralEntrySizeLog2;
-    uint64_t address_raw = base_raw + offset;
-
-    // Cast the address using a C-style cast. A reinterpret_cast would be
-    // appropriate, but it can't cast one integral type to another.
-    T address = (T)(address_raw);
-
-    // Assert that the address can be represented by the specified type.
-    VIXL_ASSERT((uint64_t)(address) == address_raw);
-
-    return address;
-  }
-
-  uint32_t Literal32() const {
-    uint32_t literal;
-    memcpy(&literal, LiteralAddress<const void*>(), sizeof(literal));
-    return literal;
-  }
-
-  uint64_t Literal64() const {
-    uint64_t literal;
-    memcpy(&literal, LiteralAddress<const void*>(), sizeof(literal));
-    return literal;
-  }
-
-  float LiteralFP32() const {
-    return rawbits_to_float(Literal32());
-  }
-
-  double LiteralFP64() const {
-    return rawbits_to_double(Literal64());
-  }
-
-  const Instruction* NextInstruction() const {
-    return this + kInstructionSize;
-  }
-
-  const Instruction* InstructionAtOffset(int64_t offset) const {
-    VIXL_ASSERT(IsWordAligned(this + offset));
-    return this + offset;
-  }
-
-  template<typename T> static Instruction* Cast(T src) {
-    return reinterpret_cast<Instruction*>(src);
-  }
-
-  template<typename T> static const Instruction* CastConst(T src) {
-    return reinterpret_cast<const Instruction*>(src);
-  }
-
- private:
-  int ImmBranch() const;
-
-  void SetPCRelImmTarget(const Instruction* target);
-  void SetBranchImmTarget(const Instruction* target);
-};
-}  // namespace vixl
-
-#endif  // VIXL_A64_INSTRUCTIONS_A64_H_
diff --git a/disas/libvixl/vixl/a64/assembler-a64.h b/disas/libvixl/vixl/a64/assembler-a64.h
new file mode 100644
index 0000000000..fda5ccc6c7
--- /dev/null
+++ b/disas/libvixl/vixl/a64/assembler-a64.h
@@ -0,0 +1,4624 @@
+// 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.
+  void sadalp(const VRegister& vd,
+              const VRegister& vn);
+
+  // Unsigned pairwise long add and accumulate.
+  void uadalp(const VRegister& vd,
+              const VRegister& vn);
+
+  // Shift left by immediate.
+  void shl(const VRegister& vd,
+           const VRegister& vn,
+           int shift);
+
+  // Signed saturating shift left by immediate.
+  void sqshl(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Signed saturating shift left unsigned by immediate.
+  void sqshlu(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Unsigned saturating shift left by immediate.
+  void uqshl(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Signed shift left long by immediate.
+  void sshll(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Signed shift left long by immediate (second part).
+  void sshll2(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Signed extend long.
+  void sxtl(const VRegister& vd,
+            const VRegister& vn);
+
+  // Signed extend long (second part).
+  void sxtl2(const VRegister& vd,
+             const VRegister& vn);
+
+  // Unsigned shift left long by immediate.
+  void ushll(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Unsigned shift left long by immediate (second part).
+  void ushll2(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Shift left long by element size.
+  void shll(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Shift left long by element size (second part).
+  void shll2(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Unsigned extend long.
+  void uxtl(const VRegister& vd,
+            const VRegister& vn);
+
+  // Unsigned extend long (second part).
+  void uxtl2(const VRegister& vd,
+             const VRegister& vn);
+
+  // Shift left by immediate and insert.
+  void sli(const VRegister& vd,
+           const VRegister& vn,
+           int shift);
+
+  // Shift right by immediate and insert.
+  void sri(const VRegister& vd,
+           const VRegister& vn,
+           int shift);
+
+  // Signed maximum.
+  void smax(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Signed pairwise maximum.
+  void smaxp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Add across vector.
+  void addv(const VRegister& vd,
+            const VRegister& vn);
+
+  // Signed add long across vector.
+  void saddlv(const VRegister& vd,
+              const VRegister& vn);
+
+  // Unsigned add long across vector.
+  void uaddlv(const VRegister& vd,
+              const VRegister& vn);
+
+  // FP maximum number across vector.
+  void fmaxnmv(const VRegister& vd,
+               const VRegister& vn);
+
+  // FP maximum across vector.
+  void fmaxv(const VRegister& vd,
+             const VRegister& vn);
+
+  // FP minimum number across vector.
+  void fminnmv(const VRegister& vd,
+               const VRegister& vn);
+
+  // FP minimum across vector.
+  void fminv(const VRegister& vd,
+             const VRegister& vn);
+
+  // Signed maximum across vector.
+  void smaxv(const VRegister& vd,
+             const VRegister& vn);
+
+  // Signed minimum.
+  void smin(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Signed minimum pairwise.
+  void sminp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed minimum across vector.
+  void sminv(const VRegister& vd,
+             const VRegister& vn);
+
+  // One-element structure store from one register.
+  void st1(const VRegister& vt,
+           const MemOperand& src);
+
+  // One-element structure store from two registers.
+  void st1(const VRegister& vt,
+           const VRegister& vt2,
+           const MemOperand& src);
+
+  // One-element structure store from three registers.
+  void st1(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           const MemOperand& src);
+
+  // One-element structure store from four registers.
+  void st1(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           const VRegister& vt4,
+           const MemOperand& src);
+
+  // One-element single structure store from one lane.
+  void st1(const VRegister& vt,
+           int lane,
+           const MemOperand& src);
+
+  // Two-element structure store from two registers.
+  void st2(const VRegister& vt,
+           const VRegister& vt2,
+           const MemOperand& src);
+
+  // Two-element single structure store from two lanes.
+  void st2(const VRegister& vt,
+           const VRegister& vt2,
+           int lane,
+           const MemOperand& src);
+
+  // Three-element structure store from three registers.
+  void st3(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           const MemOperand& src);
+
+  // Three-element single structure store from three lanes.
+  void st3(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           int lane,
+           const MemOperand& src);
+
+  // Four-element structure store from four registers.
+  void st4(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           const VRegister& vt4,
+           const MemOperand& src);
+
+  // Four-element single structure store from four lanes.
+  void st4(const VRegister& vt,
+           const VRegister& vt2,
+           const VRegister& vt3,
+           const VRegister& vt4,
+           int lane,
+           const MemOperand& src);
+
+  // Unsigned add long.
+  void uaddl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned add long (second part).
+  void uaddl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned add wide.
+  void uaddw(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned add wide (second part).
+  void uaddw2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed add long.
+  void saddl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed add long (second part).
+  void saddl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed add wide.
+  void saddw(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed add wide (second part).
+  void saddw2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned subtract long.
+  void usubl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned subtract long (second part).
+  void usubl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned subtract wide.
+  void usubw(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned subtract wide (second part).
+  void usubw2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed subtract long.
+  void ssubl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed subtract long (second part).
+  void ssubl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed integer subtract wide.
+  void ssubw(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed integer subtract wide (second part).
+  void ssubw2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned maximum.
+  void umax(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Unsigned pairwise maximum.
+  void umaxp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned maximum across vector.
+  void umaxv(const VRegister& vd,
+             const VRegister& vn);
+
+  // Unsigned minimum.
+  void umin(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Unsigned pairwise minimum.
+  void uminp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned minimum across vector.
+  void uminv(const VRegister& vd,
+             const VRegister& vn);
+
+  // Transpose vectors (primary).
+  void trn1(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Transpose vectors (secondary).
+  void trn2(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Unzip vectors (primary).
+  void uzp1(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Unzip vectors (secondary).
+  void uzp2(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Zip vectors (primary).
+  void zip1(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Zip vectors (secondary).
+  void zip2(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // Signed shift right by immediate.
+  void sshr(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Unsigned shift right by immediate.
+  void ushr(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Signed rounding shift right by immediate.
+  void srshr(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Unsigned rounding shift right by immediate.
+  void urshr(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Signed shift right by immediate and accumulate.
+  void ssra(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Unsigned shift right by immediate and accumulate.
+  void usra(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Signed rounding shift right by immediate and accumulate.
+  void srsra(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Unsigned rounding shift right by immediate and accumulate.
+  void ursra(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Shift right narrow by immediate.
+  void shrn(const VRegister& vd,
+            const VRegister& vn,
+            int shift);
+
+  // Shift right narrow by immediate (second part).
+  void shrn2(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Rounding shift right narrow by immediate.
+  void rshrn(const VRegister& vd,
+             const VRegister& vn,
+             int shift);
+
+  // Rounding shift right narrow by immediate (second part).
+  void rshrn2(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Unsigned saturating shift right narrow by immediate.
+  void uqshrn(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Unsigned saturating shift right narrow by immediate (second part).
+  void uqshrn2(const VRegister& vd,
+               const VRegister& vn,
+               int shift);
+
+  // Unsigned saturating rounding shift right narrow by immediate.
+  void uqrshrn(const VRegister& vd,
+               const VRegister& vn,
+               int shift);
+
+  // Unsigned saturating rounding shift right narrow by immediate (second part).
+  void uqrshrn2(const VRegister& vd,
+                const VRegister& vn,
+                int shift);
+
+  // Signed saturating shift right narrow by immediate.
+  void sqshrn(const VRegister& vd,
+              const VRegister& vn,
+              int shift);
+
+  // Signed saturating shift right narrow by immediate (second part).
+  void sqshrn2(const VRegister& vd,
+               const VRegister& vn,
+               int shift);
+
+  // Signed saturating rounded shift right narrow by immediate.
+  void sqrshrn(const VRegister& vd,
+               const VRegister& vn,
+               int shift);
+
+  // Signed saturating rounded shift right narrow by immediate (second part).
+  void sqrshrn2(const VRegister& vd,
+                const VRegister& vn,
+                int shift);
+
+  // Signed saturating shift right unsigned narrow by immediate.
+  void sqshrun(const VRegister& vd,
+               const VRegister& vn,
+               int shift);
+
+  // Signed saturating shift right unsigned narrow by immediate (second part).
+  void sqshrun2(const VRegister& vd,
+                const VRegister& vn,
+                int shift);
+
+  // Signed sat rounded shift right unsigned narrow by immediate.
+  void sqrshrun(const VRegister& vd,
+                const VRegister& vn,
+                int shift);
+
+  // Signed sat rounded shift right unsigned narrow by immediate (second part).
+  void sqrshrun2(const VRegister& vd,
+                 const VRegister& vn,
+                 int shift);
+
+  // FP reciprocal step.
+  void frecps(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // FP reciprocal estimate.
+  void frecpe(const VRegister& vd,
+              const VRegister& vn);
+
+  // FP reciprocal square root estimate.
+  void frsqrte(const VRegister& vd,
+               const VRegister& vn);
+
+  // FP reciprocal square root step.
+  void frsqrts(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Signed absolute difference and accumulate long.
+  void sabal(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed absolute difference and accumulate long (second part).
+  void sabal2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned absolute difference and accumulate long.
+  void uabal(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned absolute difference and accumulate long (second part).
+  void uabal2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed absolute difference long.
+  void sabdl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed absolute difference long (second part).
+  void sabdl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned absolute difference long.
+  void uabdl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned absolute difference long (second part).
+  void uabdl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Polynomial multiply long.
+  void pmull(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Polynomial multiply long (second part).
+  void pmull2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed long multiply-add.
+  void smlal(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed long multiply-add (second part).
+  void smlal2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned long multiply-add.
+  void umlal(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned long multiply-add (second part).
+  void umlal2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed long multiply-sub.
+  void smlsl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed long multiply-sub (second part).
+  void smlsl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Unsigned long multiply-sub.
+  void umlsl(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned long multiply-sub (second part).
+  void umlsl2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed long multiply.
+  void smull(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Signed long multiply (second part).
+  void smull2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Signed saturating doubling long multiply-add.
+  void sqdmlal(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Signed saturating doubling long multiply-add (second part).
+  void sqdmlal2(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm);
+
+  // Signed saturating doubling long multiply-subtract.
+  void sqdmlsl(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Signed saturating doubling long multiply-subtract (second part).
+  void sqdmlsl2(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm);
+
+  // Signed saturating doubling long multiply.
+  void sqdmull(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Signed saturating doubling long multiply (second part).
+  void sqdmull2(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm);
+
+  // Signed saturating doubling multiply returning high half.
+  void sqdmulh(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Signed saturating rounding doubling multiply returning high half.
+  void sqrdmulh(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm);
+
+  // Signed saturating doubling multiply element returning high half.
+  void sqdmulh(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm,
+               int vm_index);
+
+  // Signed saturating rounding doubling multiply element returning high half.
+  void sqrdmulh(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm,
+                int vm_index);
+
+  // Unsigned long multiply long.
+  void umull(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Unsigned long multiply (second part).
+  void umull2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Add narrow returning high half.
+  void addhn(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Add narrow returning high half (second part).
+  void addhn2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Rounding add narrow returning high half.
+  void raddhn(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Rounding add narrow returning high half (second part).
+  void raddhn2(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // Subtract narrow returning high half.
+  void subhn(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // Subtract narrow returning high half (second part).
+  void subhn2(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Rounding subtract narrow returning high half.
+  void rsubhn(const VRegister& vd,
+              const VRegister& vn,
+              const VRegister& vm);
+
+  // Rounding subtract narrow returning high half (second part).
+  void rsubhn2(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // FP vector multiply accumulate.
+  void fmla(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // FP vector multiply subtract.
+  void fmls(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // FP vector multiply extended.
+  void fmulx(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP absolute greater than or equal.
+  void facge(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP absolute greater than.
+  void facgt(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP multiply by element.
+  void fmul(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm,
+            int vm_index);
+
+  // FP fused multiply-add to accumulator by element.
+  void fmla(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm,
+            int vm_index);
+
+  // FP fused multiply-sub from accumulator by element.
+  void fmls(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm,
+            int vm_index);
+
+  // FP multiply extended by element.
+  void fmulx(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm,
+             int vm_index);
+
+  // FP compare equal.
+  void fcmeq(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP greater than.
+  void fcmgt(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP greater than or equal.
+  void fcmge(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP compare equal to zero.
+  void fcmeq(const VRegister& vd,
+             const VRegister& vn,
+             double imm);
+
+  // FP greater than zero.
+  void fcmgt(const VRegister& vd,
+             const VRegister& vn,
+             double imm);
+
+  // FP greater than or equal to zero.
+  void fcmge(const VRegister& vd,
+             const VRegister& vn,
+             double imm);
+
+  // FP less than or equal to zero.
+  void fcmle(const VRegister& vd,
+             const VRegister& vn,
+             double imm);
+
+  // FP less than to zero.
+  void fcmlt(const VRegister& vd,
+             const VRegister& vn,
+             double imm);
+
+  // FP absolute difference.
+  void fabd(const VRegister& vd,
+            const VRegister& vn,
+            const VRegister& vm);
+
+  // FP pairwise add vector.
+  void faddp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP pairwise add scalar.
+  void faddp(const VRegister& vd,
+             const VRegister& vn);
+
+  // FP pairwise maximum vector.
+  void fmaxp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP pairwise maximum scalar.
+  void fmaxp(const VRegister& vd,
+             const VRegister& vn);
+
+  // FP pairwise minimum vector.
+  void fminp(const VRegister& vd,
+             const VRegister& vn,
+             const VRegister& vm);
+
+  // FP pairwise minimum scalar.
+  void fminp(const VRegister& vd,
+             const VRegister& vn);
+
+  // FP pairwise maximum number vector.
+  void fmaxnmp(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // FP pairwise maximum number scalar.
+  void fmaxnmp(const VRegister& vd,
+               const VRegister& vn);
+
+  // FP pairwise minimum number vector.
+  void fminnmp(const VRegister& vd,
+               const VRegister& vn,
+               const VRegister& vm);
+
+  // FP pairwise minimum number scalar.
+  void fminnmp(const VRegister& vd,
+               const VRegister& vn);
+
+  // Emit generic instructions.
+  // Emit raw instructions into the instruction stream.
+  void dci(Instr raw_inst) { Emit(raw_inst); }
+
+  // Emit 32 bits of data into the instruction stream.
+  void dc32(uint32_t data) {
+    VIXL_ASSERT(buffer_monitor_ > 0);
+    buffer_->Emit32(data);
+  }
+
+  // Emit 64 bits of data into the instruction stream.
+  void dc64(uint64_t data) {
+    VIXL_ASSERT(buffer_monitor_ > 0);
+    buffer_->Emit64(data);
+  }
+
+  // Copy a string into the instruction stream, including the terminating NULL
+  // character. The instruction pointer is then aligned correctly for
+  // subsequent instructions.
+  void EmitString(const char * string) {
+    VIXL_ASSERT(string != NULL);
+    VIXL_ASSERT(buffer_monitor_ > 0);
+
+    buffer_->EmitString(string);
+    buffer_->Align();
+  }
+
+  // Code generation helpers.
+
+  // Register encoding.
+  static Instr Rd(CPURegister rd) {
+    VIXL_ASSERT(rd.code() != kSPRegInternalCode);
+    return rd.code() << Rd_offset;
+  }
+
+  static Instr Rn(CPURegister rn) {
+    VIXL_ASSERT(rn.code() != kSPRegInternalCode);
+    return rn.code() << Rn_offset;
+  }
+
+  static Instr Rm(CPURegister rm) {
+    VIXL_ASSERT(rm.code() != kSPRegInternalCode);
+    return rm.code() << Rm_offset;
+  }
+
+  static Instr RmNot31(CPURegister rm) {
+    VIXL_ASSERT(rm.code() != kSPRegInternalCode);
+    VIXL_ASSERT(!rm.IsZero());
+    return Rm(rm);
+  }
+
+  static Instr Ra(CPURegister ra) {
+    VIXL_ASSERT(ra.code() != kSPRegInternalCode);
+    return ra.code() << Ra_offset;
+  }
+
+  static Instr Rt(CPURegister rt) {
+    VIXL_ASSERT(rt.code() != kSPRegInternalCode);
+    return rt.code() << Rt_offset;
+  }
+
+  static Instr Rt2(CPURegister rt2) {
+    VIXL_ASSERT(rt2.code() != kSPRegInternalCode);
+    return rt2.code() << Rt2_offset;
+  }
+
+  static Instr Rs(CPURegister rs) {
+    VIXL_ASSERT(rs.code() != kSPRegInternalCode);
+    return rs.code() << Rs_offset;
+  }
+
+  // These encoding functions allow the stack pointer to be encoded, and
+  // disallow the zero register.
+  static Instr RdSP(Register rd) {
+    VIXL_ASSERT(!rd.IsZero());
+    return (rd.code() & kRegCodeMask) << Rd_offset;
+  }
+
+  static Instr RnSP(Register rn) {
+    VIXL_ASSERT(!rn.IsZero());
+    return (rn.code() & kRegCodeMask) << Rn_offset;
+  }
+
+  // Flags encoding.
+  static Instr Flags(FlagsUpdate S) {
+    if (S == SetFlags) {
+      return 1 << FlagsUpdate_offset;
+    } else if (S == LeaveFlags) {
+      return 0 << FlagsUpdate_offset;
+    }
+    VIXL_UNREACHABLE();
+    return 0;
+  }
+
+  static Instr Cond(Condition cond) {
+    return cond << Condition_offset;
+  }
+
+  // PC-relative address encoding.
+  static Instr ImmPCRelAddress(int imm21) {
+    VIXL_ASSERT(is_int21(imm21));
+    Instr imm = static_cast<Instr>(truncate_to_int21(imm21));
+    Instr immhi = (imm >> ImmPCRelLo_width) << ImmPCRelHi_offset;
+    Instr immlo = imm << ImmPCRelLo_offset;
+    return (immhi & ImmPCRelHi_mask) | (immlo & ImmPCRelLo_mask);
+  }
+
+  // Branch encoding.
+  static Instr ImmUncondBranch(int imm26) {
+    VIXL_ASSERT(is_int26(imm26));
+    return truncate_to_int26(imm26) << ImmUncondBranch_offset;
+  }
+
+  static Instr ImmCondBranch(int imm19) {
+    VIXL_ASSERT(is_int19(imm19));
+    return truncate_to_int19(imm19) << ImmCondBranch_offset;
+  }
+
+  static Instr ImmCmpBranch(int imm19) {
+    VIXL_ASSERT(is_int19(imm19));
+    return truncate_to_int19(imm19) << ImmCmpBranch_offset;
+  }
+
+  static Instr ImmTestBranch(int imm14) {
+    VIXL_ASSERT(is_int14(imm14));
+    return truncate_to_int14(imm14) << ImmTestBranch_offset;
+  }
+
+  static Instr ImmTestBranchBit(unsigned bit_pos) {
+    VIXL_ASSERT(is_uint6(bit_pos));
+    // Subtract five from the shift offset, as we need bit 5 from bit_pos.
+    unsigned b5 = bit_pos << (ImmTestBranchBit5_offset - 5);
+    unsigned b40 = bit_pos << ImmTestBranchBit40_offset;
+    b5 &= ImmTestBranchBit5_mask;
+    b40 &= ImmTestBranchBit40_mask;
+    return b5 | b40;
+  }
+
+  // Data Processing encoding.
+  static Instr SF(Register rd) {
+      return rd.Is64Bits() ? SixtyFourBits : ThirtyTwoBits;
+  }
+
+  static Instr ImmAddSub(int imm) {
+    VIXL_ASSERT(IsImmAddSub(imm));
+    if (is_uint12(imm)) {  // No shift required.
+      imm <<= ImmAddSub_offset;
+    } else {
+      imm = ((imm >> 12) << ImmAddSub_offset) | (1 << ShiftAddSub_offset);
+    }
+    return imm;
+  }
+
+  static Instr ImmS(unsigned imms, unsigned reg_size) {
+    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(imms)) ||
+           ((reg_size == kWRegSize) && is_uint5(imms)));
+    USE(reg_size);
+    return imms << ImmS_offset;
+  }
+
+  static Instr ImmR(unsigned immr, unsigned reg_size) {
+    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(immr)) ||
+           ((reg_size == kWRegSize) && is_uint5(immr)));
+    USE(reg_size);
+    VIXL_ASSERT(is_uint6(immr));
+    return immr << ImmR_offset;
+  }
+
+  static Instr ImmSetBits(unsigned imms, unsigned reg_size) {
+    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
+    VIXL_ASSERT(is_uint6(imms));
+    VIXL_ASSERT((reg_size == kXRegSize) || is_uint6(imms + 3));
+    USE(reg_size);
+    return imms << ImmSetBits_offset;
+  }
+
+  static Instr ImmRotate(unsigned immr, unsigned reg_size) {
+    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
+    VIXL_ASSERT(((reg_size == kXRegSize) && is_uint6(immr)) ||
+           ((reg_size == kWRegSize) && is_uint5(immr)));
+    USE(reg_size);
+    return immr << ImmRotate_offset;
+  }
+
+  static Instr ImmLLiteral(int imm19) {
+    VIXL_ASSERT(is_int19(imm19));
+    return truncate_to_int19(imm19) << ImmLLiteral_offset;
+  }
+
+  static Instr BitN(unsigned bitn, unsigned reg_size) {
+    VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
+    VIXL_ASSERT((reg_size == kXRegSize) || (bitn == 0));
+    USE(reg_size);
+    return bitn << BitN_offset;
+  }
+
+  static Instr ShiftDP(Shift shift) {
+    VIXL_ASSERT(shift == LSL || shift == LSR || shift == ASR || shift == ROR);
+    return shift << ShiftDP_offset;
+  }
+
+  static Instr ImmDPShift(unsigned amount) {
+    VIXL_ASSERT(is_uint6(amount));
+    return amount << ImmDPShift_offset;
+  }
+
+  static Instr ExtendMode(Extend extend) {
+    return extend << ExtendMode_offset;
+  }
+
+  static Instr ImmExtendShift(unsigned left_shift) {
+    VIXL_ASSERT(left_shift <= 4);
+    return left_shift << ImmExtendShift_offset;
+  }
+
+  static Instr ImmCondCmp(unsigned imm) {
+    VIXL_ASSERT(is_uint5(imm));
+    return imm << ImmCondCmp_offset;
+  }
+
+  static Instr Nzcv(StatusFlags nzcv) {
+    return ((nzcv >> Flags_offset) & 0xf) << Nzcv_offset;
+  }
+
+  // MemOperand offset encoding.
+  static Instr ImmLSUnsigned(int imm12) {
+    VIXL_ASSERT(is_uint12(imm12));
+    return imm12 << ImmLSUnsigned_offset;
+  }
+
+  static Instr ImmLS(int imm9) {
+    VIXL_ASSERT(is_int9(imm9));
+    return truncate_to_int9(imm9) << ImmLS_offset;
+  }
+
+  static Instr ImmLSPair(int imm7, unsigned access_size) {
+    VIXL_ASSERT(((imm7 >> access_size) << access_size) == imm7);
+    int scaled_imm7 = imm7 >> access_size;
+    VIXL_ASSERT(is_int7(scaled_imm7));
+    return truncate_to_int7(scaled_imm7) << ImmLSPair_offset;
+  }
+
+  static Instr ImmShiftLS(unsigned shift_amount) {
+    VIXL_ASSERT(is_uint1(shift_amount));
+    return shift_amount << ImmShiftLS_offset;
+  }
+
+  static Instr ImmPrefetchOperation(int imm5) {
+    VIXL_ASSERT(is_uint5(imm5));
+    return imm5 << ImmPrefetchOperation_offset;
+  }
+
+  static Instr ImmException(int imm16) {
+    VIXL_ASSERT(is_uint16(imm16));
+    return imm16 << ImmException_offset;
+  }
+
+  static Instr ImmSystemRegister(int imm15) {
+    VIXL_ASSERT(is_uint15(imm15));
+    return imm15 << ImmSystemRegister_offset;
+  }
+
+  static Instr ImmHint(int imm7) {
+    VIXL_ASSERT(is_uint7(imm7));
+    return imm7 << ImmHint_offset;
+  }
+
+  static Instr CRm(int imm4) {
+    VIXL_ASSERT(is_uint4(imm4));
+    return imm4 << CRm_offset;
+  }
+
+  static Instr CRn(int imm4) {
+    VIXL_ASSERT(is_uint4(imm4));
+    return imm4 << CRn_offset;
+  }
+
+  static Instr SysOp(int imm14) {
+    VIXL_ASSERT(is_uint14(imm14));
+    return imm14 << SysOp_offset;
+  }
+
+  static Instr ImmSysOp1(int imm3) {
+    VIXL_ASSERT(is_uint3(imm3));
+    return imm3 << SysOp1_offset;
+  }
+
+  static Instr ImmSysOp2(int imm3) {
+    VIXL_ASSERT(is_uint3(imm3));
+    return imm3 << SysOp2_offset;
+  }
+
+  static Instr ImmBarrierDomain(int imm2) {
+    VIXL_ASSERT(is_uint2(imm2));
+    return imm2 << ImmBarrierDomain_offset;
+  }
+
+  static Instr ImmBarrierType(int imm2) {
+    VIXL_ASSERT(is_uint2(imm2));
+    return imm2 << ImmBarrierType_offset;
+  }
+
+  // Move immediates encoding.
+  static Instr ImmMoveWide(uint64_t imm) {
+    VIXL_ASSERT(is_uint16(imm));
+    return static_cast<Instr>(imm << ImmMoveWide_offset);
+  }
+
+  static Instr ShiftMoveWide(int64_t shift) {
+    VIXL_ASSERT(is_uint2(shift));
+    return static_cast<Instr>(shift << ShiftMoveWide_offset);
+  }
+
+  // FP Immediates.
+  static Instr ImmFP32(float imm);
+  static Instr ImmFP64(double imm);
+
+  // FP register type.
+  static Instr FPType(FPRegister fd) {
+    return fd.Is64Bits() ? FP64 : FP32;
+  }
+
+  static Instr FPScale(unsigned scale) {
+    VIXL_ASSERT(is_uint6(scale));
+    return scale << FPScale_offset;
+  }
+
+  // Immediate field checking helpers.
+  static bool IsImmAddSub(int64_t immediate);
+  static bool IsImmConditionalCompare(int64_t immediate);
+  static bool IsImmFP32(float imm);
+  static bool IsImmFP64(double imm);
+  static bool IsImmLogical(uint64_t value,
+                           unsigned width,
+                           unsigned* n = NULL,
+                           unsigned* imm_s = NULL,
+                           unsigned* imm_r = NULL);
+  static bool IsImmLSPair(int64_t offset, unsigned access_size);
+  static bool IsImmLSScaled(int64_t offset, unsigned access_size);
+  static bool IsImmLSUnscaled(int64_t offset);
+  static bool IsImmMovn(uint64_t imm, unsigned reg_size);
+  static bool IsImmMovz(uint64_t imm, unsigned reg_size);
+
+  // Instruction bits for vector format in data processing operations.
+  static Instr VFormat(VRegister vd) {
+    if (vd.Is64Bits()) {
+      switch (vd.lanes()) {
+        case 2: return NEON_2S;
+        case 4: return NEON_4H;
+        case 8: return NEON_8B;
+        default: return 0xffffffff;
+      }
+    } else {
+      VIXL_ASSERT(vd.Is128Bits());
+      switch (vd.lanes()) {
+        case 2: return NEON_2D;
+        case 4: return NEON_4S;
+        case 8: return NEON_8H;
+        case 16: return NEON_16B;
+        default: return 0xffffffff;
+      }
+    }
+  }
+
+  // Instruction bits for vector format in floating point data processing
+  // operations.
+  static Instr FPFormat(VRegister vd) {
+    if (vd.lanes() == 1) {
+      // Floating point scalar formats.
+      VIXL_ASSERT(vd.Is32Bits() || vd.Is64Bits());
+      return vd.Is64Bits() ? FP64 : FP32;
+    }
+
+    // Two lane floating point vector formats.
+    if (vd.lanes() == 2) {
+      VIXL_ASSERT(vd.Is64Bits() || vd.Is128Bits());
+      return vd.Is128Bits() ? NEON_FP_2D : NEON_FP_2S;
+    }
+
+    // Four lane floating point vector format.
+    VIXL_ASSERT((vd.lanes() == 4) && vd.Is128Bits());
+    return NEON_FP_4S;
+  }
+
+  // Instruction bits for vector format in load and store operations.
+  static Instr LSVFormat(VRegister vd) {
+    if (vd.Is64Bits()) {
+      switch (vd.lanes()) {
+        case 1: return LS_NEON_1D;
+        case 2: return LS_NEON_2S;
+        case 4: return LS_NEON_4H;
+        case 8: return LS_NEON_8B;
+        default: return 0xffffffff;
+      }
+    } else {
+      VIXL_ASSERT(vd.Is128Bits());
+      switch (vd.lanes()) {
+        case 2: return LS_NEON_2D;
+        case 4: return LS_NEON_4S;
+        case 8: return LS_NEON_8H;
+        case 16: return LS_NEON_16B;
+        default: return 0xffffffff;
+      }
+    }
+  }
+
+  // Instruction bits for scalar format in data processing operations.
+  static Instr SFormat(VRegister vd) {
+    VIXL_ASSERT(vd.lanes() == 1);
+    switch (vd.SizeInBytes()) {
+      case 1: return NEON_B;
+      case 2: return NEON_H;
+      case 4: return NEON_S;
+      case 8: return NEON_D;
+      default: return 0xffffffff;
+    }
+  }
+
+  static Instr ImmNEONHLM(int index, int num_bits) {
+    int h, l, m;
+    if (num_bits == 3) {
+      VIXL_ASSERT(is_uint3(index));
+      h  = (index >> 2) & 1;
+      l  = (index >> 1) & 1;
+      m  = (index >> 0) & 1;
+    } else if (num_bits == 2) {
+      VIXL_ASSERT(is_uint2(index));
+      h  = (index >> 1) & 1;
+      l  = (index >> 0) & 1;
+      m  = 0;
+    } else {
+      VIXL_ASSERT(is_uint1(index) && (num_bits == 1));
+      h  = (index >> 0) & 1;
+      l  = 0;
+      m  = 0;
+    }
+    return (h << NEONH_offset) | (l << NEONL_offset) | (m << NEONM_offset);
+  }
+
+  static Instr ImmNEONExt(int imm4) {
+    VIXL_ASSERT(is_uint4(imm4));
+    return imm4 << ImmNEONExt_offset;
+  }
+
+  static Instr ImmNEON5(Instr format, int index) {
+    VIXL_ASSERT(is_uint4(index));
+    int s = LaneSizeInBytesLog2FromFormat(static_cast<VectorFormat>(format));
+    int imm5 = (index << (s + 1)) | (1 << s);
+    return imm5 << ImmNEON5_offset;
+  }
+
+  static Instr ImmNEON4(Instr format, int index) {
+    VIXL_ASSERT(is_uint4(index));
+    int s = LaneSizeInBytesLog2FromFormat(static_cast<VectorFormat>(format));
+    int imm4 = index << s;
+    return imm4 << ImmNEON4_offset;
+  }
+
+  static Instr ImmNEONabcdefgh(int imm8) {
+    VIXL_ASSERT(is_uint8(imm8));
+    Instr instr;
+    instr  = ((imm8 >> 5) & 7) << ImmNEONabc_offset;
+    instr |= (imm8 & 0x1f) << ImmNEONdefgh_offset;
+    return instr;
+  }
+
+  static Instr NEONCmode(int cmode) {
+    VIXL_ASSERT(is_uint4(cmode));
+    return cmode << NEONCmode_offset;
+  }
+
+  static Instr NEONModImmOp(int op) {
+    VIXL_ASSERT(is_uint1(op));
+    return op << NEONModImmOp_offset;
+  }
+
+  // Size of the code generated since label to the current position.
+  size_t SizeOfCodeGeneratedSince(Label* label) const {
+    VIXL_ASSERT(label->IsBound());
+    return buffer_->OffsetFrom(label->location());
+  }
+
+  size_t SizeOfCodeGenerated() const {
+    return buffer_->CursorOffset();
+  }
+
+  size_t BufferCapacity() const { return buffer_->capacity(); }
+
+  size_t RemainingBufferSpace() const { return buffer_->RemainingBytes(); }
+
+  void EnsureSpaceFor(size_t amount) {
+    if (buffer_->RemainingBytes() < amount) {
+      size_t capacity = buffer_->capacity();
+      size_t size = buffer_->CursorOffset();
+      do {
+        // TODO(all): refine.
+        capacity *= 2;
+      } while ((capacity - size) <  amount);
+      buffer_->Grow(capacity);
+    }
+  }
+
+#ifdef VIXL_DEBUG
+  void AcquireBuffer() {
+    VIXL_ASSERT(buffer_monitor_ >= 0);
+    buffer_monitor_++;
+  }
+
+  void ReleaseBuffer() {
+    buffer_monitor_--;
+    VIXL_ASSERT(buffer_monitor_ >= 0);
+  }
+#endif
+
+  PositionIndependentCodeOption pic() const {
+    return pic_;
+  }
+
+  bool AllowPageOffsetDependentCode() const {
+    return (pic() == PageOffsetDependentCode) ||
+           (pic() == PositionDependentCode);
+  }
+
+  static const Register& AppropriateZeroRegFor(const CPURegister& reg) {
+    return reg.Is64Bits() ? xzr : wzr;
+  }
+
+
+ protected:
+  void LoadStore(const CPURegister& rt,
+                 const MemOperand& addr,
+                 LoadStoreOp op,
+                 LoadStoreScalingOption option = PreferScaledOffset);
+
+  void LoadStorePair(const CPURegister& rt,
+                     const CPURegister& rt2,
+                     const MemOperand& addr,
+                     LoadStorePairOp op);
+  void LoadStoreStruct(const VRegister& vt,
+                       const MemOperand& addr,
+                       NEONLoadStoreMultiStructOp op);
+  void LoadStoreStruct1(const VRegister& vt,
+                        int reg_count,
+                        const MemOperand& addr);
+  void LoadStoreStructSingle(const VRegister& vt,
+                             uint32_t lane,
+                             const MemOperand& addr,
+                             NEONLoadStoreSingleStructOp op);
+  void LoadStoreStructSingleAllLanes(const VRegister& vt,
+                                     const MemOperand& addr,
+                                     NEONLoadStoreSingleStructOp op);
+  void LoadStoreStructVerify(const VRegister& vt,
+                             const MemOperand& addr,
+                             Instr op);
+
+  void Prefetch(PrefetchOperation op,
+                const MemOperand& addr,
+                LoadStoreScalingOption option = PreferScaledOffset);
+
+  // TODO(all): The third parameter should be passed by reference but gcc 4.8.2
+  // reports a bogus uninitialised warning then.
+  void Logical(const Register& rd,
+               const Register& rn,
+               const Operand operand,
+               LogicalOp op);
+  void LogicalImmediate(const Register& rd,
+                        const Register& rn,
+                        unsigned n,
+                        unsigned imm_s,
+                        unsigned imm_r,
+                        LogicalOp op);
+
+  void ConditionalCompare(const Register& rn,
+                          const Operand& operand,
+                          StatusFlags nzcv,
+                          Condition cond,
+                          ConditionalCompareOp op);
+
+  void AddSubWithCarry(const Register& rd,
+                       const Register& rn,
+                       const Operand& operand,
+                       FlagsUpdate S,
+                       AddSubWithCarryOp op);
+
+
+  // Functions for emulating operands not directly supported by the instruction
+  // set.
+  void EmitShift(const Register& rd,
+                 const Register& rn,
+                 Shift shift,
+                 unsigned amount);
+  void EmitExtendShift(const Register& rd,
+                       const Register& rn,
+                       Extend extend,
+                       unsigned left_shift);
+
+  void AddSub(const Register& rd,
+              const Register& rn,
+              const Operand& operand,
+              FlagsUpdate S,
+              AddSubOp op);
+
+  void NEONTable(const VRegister& vd,
+                 const VRegister& vn,
+                 const VRegister& vm,
+                 NEONTableOp op);
+
+  // Find an appropriate LoadStoreOp or LoadStorePairOp for the specified
+  // registers. Only simple loads are supported; sign- and zero-extension (such
+  // as in LDPSW_x or LDRB_w) are not supported.
+  static LoadStoreOp LoadOpFor(const CPURegister& rt);
+  static LoadStorePairOp LoadPairOpFor(const CPURegister& rt,
+                                       const CPURegister& rt2);
+  static LoadStoreOp StoreOpFor(const CPURegister& rt);
+  static LoadStorePairOp StorePairOpFor(const CPURegister& rt,
+                                        const CPURegister& rt2);
+  static LoadStorePairNonTemporalOp LoadPairNonTemporalOpFor(
+    const CPURegister& rt, const CPURegister& rt2);
+  static LoadStorePairNonTemporalOp StorePairNonTemporalOpFor(
+    const CPURegister& rt, const CPURegister& rt2);
+  static LoadLiteralOp LoadLiteralOpFor(const CPURegister& rt);
+
+
+ private:
+  static uint32_t FP32ToImm8(float imm);
+  static uint32_t FP64ToImm8(double imm);
+
+  // Instruction helpers.
+  void MoveWide(const Register& rd,
+                uint64_t imm,
+                int shift,
+                MoveWideImmediateOp mov_op);
+  void DataProcShiftedRegister(const Register& rd,
+                               const Register& rn,
+                               const Operand& operand,
+                               FlagsUpdate S,
+                               Instr op);
+  void DataProcExtendedRegister(const Register& rd,
+                                const Register& rn,
+                                const Operand& operand,
+                                FlagsUpdate S,
+                                Instr op);
+  void LoadStorePairNonTemporal(const CPURegister& rt,
+                                const CPURegister& rt2,
+                                const MemOperand& addr,
+                                LoadStorePairNonTemporalOp op);
+  void LoadLiteral(const CPURegister& rt, uint64_t imm, LoadLiteralOp op);
+  void ConditionalSelect(const Register& rd,
+                         const Register& rn,
+                         const Register& rm,
+                         Condition cond,
+                         ConditionalSelectOp op);
+  void DataProcessing1Source(const Register& rd,
+                             const Register& rn,
+                             DataProcessing1SourceOp op);
+  void DataProcessing3Source(const Register& rd,
+                             const Register& rn,
+                             const Register& rm,
+                             const Register& ra,
+                             DataProcessing3SourceOp op);
+  void FPDataProcessing1Source(const VRegister& fd,
+                               const VRegister& fn,
+                               FPDataProcessing1SourceOp op);
+  void FPDataProcessing3Source(const VRegister& fd,
+                               const VRegister& fn,
+                               const VRegister& fm,
+                               const VRegister& fa,
+                               FPDataProcessing3SourceOp op);
+  void NEONAcrossLanesL(const VRegister& vd,
+                        const VRegister& vn,
+                        NEONAcrossLanesOp op);
+  void NEONAcrossLanes(const VRegister& vd,
+                       const VRegister& vn,
+                       NEONAcrossLanesOp op);
+  void NEONModifiedImmShiftLsl(const VRegister& vd,
+                               const int imm8,
+                               const int left_shift,
+                               NEONModifiedImmediateOp op);
+  void NEONModifiedImmShiftMsl(const VRegister& vd,
+                               const int imm8,
+                               const int shift_amount,
+                               NEONModifiedImmediateOp op);
+  void NEONFP2Same(const VRegister& vd,
+                   const VRegister& vn,
+                   Instr vop);
+  void NEON3Same(const VRegister& vd,
+                 const VRegister& vn,
+                 const VRegister& vm,
+                 NEON3SameOp vop);
+  void NEONFP3Same(const VRegister& vd,
+                   const VRegister& vn,
+                   const VRegister& vm,
+                   Instr op);
+  void NEON3DifferentL(const VRegister& vd,
+                       const VRegister& vn,
+                       const VRegister& vm,
+                       NEON3DifferentOp vop);
+  void NEON3DifferentW(const VRegister& vd,
+                       const VRegister& vn,
+                       const VRegister& vm,
+                       NEON3DifferentOp vop);
+  void NEON3DifferentHN(const VRegister& vd,
+                        const VRegister& vn,
+                        const VRegister& vm,
+                        NEON3DifferentOp vop);
+  void NEONFP2RegMisc(const VRegister& vd,
+                      const VRegister& vn,
+                      NEON2RegMiscOp vop,
+                      double value = 0.0);
+  void NEON2RegMisc(const VRegister& vd,
+                    const VRegister& vn,
+                    NEON2RegMiscOp vop,
+                    int value = 0);
+  void NEONFP2RegMisc(const VRegister& vd,
+                      const VRegister& vn,
+                      Instr op);
+  void NEONAddlp(const VRegister& vd,
+                 const VRegister& vn,
+                 NEON2RegMiscOp op);
+  void NEONPerm(const VRegister& vd,
+                const VRegister& vn,
+                const VRegister& vm,
+                NEONPermOp op);
+  void NEONFPByElement(const VRegister& vd,
+                       const VRegister& vn,
+                       const VRegister& vm,
+                       int vm_index,
+                       NEONByIndexedElementOp op);
+  void NEONByElement(const VRegister& vd,
+                     const VRegister& vn,
+                     const VRegister& vm,
+                     int vm_index,
+                     NEONByIndexedElementOp op);
+  void NEONByElementL(const VRegister& vd,
+                      const VRegister& vn,
+                      const VRegister& vm,
+                      int vm_index,
+                      NEONByIndexedElementOp op);
+  void NEONShiftImmediate(const VRegister& vd,
+                          const VRegister& vn,
+                          NEONShiftImmediateOp op,
+                          int immh_immb);
+  void NEONShiftLeftImmediate(const VRegister& vd,
+                              const VRegister& vn,
+                              int shift,
+                              NEONShiftImmediateOp op);
+  void NEONShiftRightImmediate(const VRegister& vd,
+                               const VRegister& vn,
+                               int shift,
+                               NEONShiftImmediateOp op);
+  void NEONShiftImmediateL(const VRegister& vd,
+                           const VRegister& vn,
+                           int shift,
+                           NEONShiftImmediateOp op);
+  void NEONShiftImmediateN(const VRegister& vd,
+                           const VRegister& vn,
+                           int shift,
+                           NEONShiftImmediateOp op);
+  void NEONXtn(const VRegister& vd,
+               const VRegister& vn,
+               NEON2RegMiscOp vop);
+
+  Instr LoadStoreStructAddrModeField(const MemOperand& addr);
+
+  // Encode the specified MemOperand for the specified access size and scaling
+  // preference.
+  Instr LoadStoreMemOperand(const MemOperand& addr,
+                            unsigned access_size,
+                            LoadStoreScalingOption option);
+
+  // Link the current (not-yet-emitted) instruction to the specified label, then
+  // return an offset to be encoded in the instruction. If the label is not yet
+  // bound, an offset of 0 is returned.
+  ptrdiff_t LinkAndGetByteOffsetTo(Label * label);
+  ptrdiff_t LinkAndGetInstructionOffsetTo(Label * label);
+  ptrdiff_t LinkAndGetPageOffsetTo(Label * label);
+
+  // A common implementation for the LinkAndGet<Type>OffsetTo helpers.
+  template <int element_shift>
+  ptrdiff_t LinkAndGetOffsetTo(Label* label);
+
+  // Literal load offset are in words (32-bit).
+  ptrdiff_t LinkAndGetWordOffsetTo(RawLiteral* literal);
+
+  // Emit the instruction in buffer_.
+  void Emit(Instr instruction) {
+    VIXL_STATIC_ASSERT(sizeof(instruction) == kInstructionSize);
+    VIXL_ASSERT(buffer_monitor_ > 0);
+    buffer_->Emit32(instruction);
+  }
+
+  // Buffer where the code is emitted.
+  CodeBuffer* buffer_;
+  PositionIndependentCodeOption pic_;
+
+#ifdef VIXL_DEBUG
+  int64_t buffer_monitor_;
+#endif
+};
+
+
+// All Assembler emits MUST acquire/release the underlying code buffer. The
+// helper scope below will do so and optionally ensure the buffer is big enough
+// to receive the emit. It is possible to request the scope not to perform any
+// checks (kNoCheck) if for example it is known in advance the buffer size is
+// adequate or there is some other size checking mechanism in place.
+class CodeBufferCheckScope {
+ public:
+  // Tell whether or not the scope needs to ensure the associated CodeBuffer
+  // has enough space for the requested size.
+  enum CheckPolicy {
+    kNoCheck,
+    kCheck
+  };
+
+  // Tell whether or not the scope should assert the amount of code emitted
+  // within the scope is consistent with the requested amount.
+  enum AssertPolicy {
+    kNoAssert,    // No assert required.
+    kExactSize,   // The code emitted must be exactly size bytes.
+    kMaximumSize  // The code emitted must be at most size bytes.
+  };
+
+  CodeBufferCheckScope(Assembler* assm,
+                       size_t size,
+                       CheckPolicy check_policy = kCheck,
+                       AssertPolicy assert_policy = kMaximumSize)
+      : assm_(assm) {
+    if (check_policy == kCheck) assm->EnsureSpaceFor(size);
+#ifdef VIXL_DEBUG
+    assm->bind(&start_);
+    size_ = size;
+    assert_policy_ = assert_policy;
+    assm->AcquireBuffer();
+#else
+    USE(assert_policy);
+#endif
+  }
+
+  // This is a shortcut for CodeBufferCheckScope(assm, 0, kNoCheck, kNoAssert).
+  explicit CodeBufferCheckScope(Assembler* assm) : assm_(assm) {
+#ifdef VIXL_DEBUG
+    size_ = 0;
+    assert_policy_ = kNoAssert;
+    assm->AcquireBuffer();
+#endif
+  }
+
+  ~CodeBufferCheckScope() {
+#ifdef VIXL_DEBUG
+    assm_->ReleaseBuffer();
+    switch (assert_policy_) {
+      case kNoAssert: break;
+      case kExactSize:
+        VIXL_ASSERT(assm_->SizeOfCodeGeneratedSince(&start_) == size_);
+        break;
+      case kMaximumSize:
+        VIXL_ASSERT(assm_->SizeOfCodeGeneratedSince(&start_) <= size_);
+        break;
+      default:
+        VIXL_UNREACHABLE();
+    }
+#endif
+  }
+
+ protected:
+  Assembler* assm_;
+#ifdef VIXL_DEBUG
+  Label start_;
+  size_t size_;
+  AssertPolicy assert_policy_;
+#endif
+};
+
+
+template <typename T>
+void Literal<T>::UpdateValue(T new_value, const Assembler* assembler) {
+  return UpdateValue(new_value, assembler->GetStartAddress<uint8_t*>());
+}
+
+
+template <typename T>
+void Literal<T>::UpdateValue(T high64, T low64, const Assembler* assembler) {
+  return UpdateValue(high64, low64, assembler->GetStartAddress<uint8_t*>());
+}
+
+
+}  // namespace vixl
+
+#endif  // VIXL_A64_ASSEMBLER_A64_H_
diff --git a/disas/libvixl/a64/constants-a64.h b/disas/libvixl/vixl/a64/constants-a64.h
index bc1a2c4b9b..2caa73af87 100644
--- a/disas/libvixl/a64/constants-a64.h
+++ b/disas/libvixl/vixl/a64/constants-a64.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2015, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -30,7 +30,14 @@
 namespace vixl {
 
 const unsigned kNumberOfRegisters = 32;
-const unsigned kNumberOfFPRegisters = 32;
+const unsigned kNumberOfVRegisters = 32;
+const unsigned kNumberOfFPRegisters = kNumberOfVRegisters;
+// Callee saved registers are x21-x30(lr).
+const int kNumberOfCalleeSavedRegisters = 10;
+const int kFirstCalleeSavedRegisterIndex = 21;
+// Callee saved FP registers are d8-d15.
+const int kNumberOfCalleeSavedFPRegisters = 8;
+const int kFirstCalleeSavedFPRegisterIndex = 8;
 
 #define REGISTER_CODE_LIST(R)                                                  \
 R(0)  R(1)  R(2)  R(3)  R(4)  R(5)  R(6)  R(7)                                 \
@@ -100,8 +107,10 @@ V_(FPScale, 15, 10, Bits)                                                      \
 V_(ImmLS, 20, 12, SignedBits)                                                  \
 V_(ImmLSUnsigned, 21, 10, Bits)                                                \
 V_(ImmLSPair, 21, 15, SignedBits)                                              \
-V_(SizeLS, 31, 30, Bits)                                                       \
 V_(ImmShiftLS, 12, 12, Bits)                                                   \
+V_(LSOpc, 23, 22, Bits)                                                        \
+V_(LSVector, 26, 26, Bits)                                                     \
+V_(LSSize, 31, 30, Bits)                                                       \
 V_(ImmPrefetchOperation, 4, 0, Bits)                                           \
 V_(PrefetchHint, 4, 3, Bits)                                                   \
 V_(PrefetchTarget, 2, 1, Bits)                                                 \
@@ -116,9 +125,10 @@ V_(ImmHint, 11, 5, Bits)                                                       \
 V_(ImmBarrierDomain, 11, 10, Bits)                                             \
 V_(ImmBarrierType, 9, 8, Bits)                                                 \
                                                                                \
-/* System (MRS, MSR) */                                                        \
+/* System (MRS, MSR, SYS) */                                                   \
 V_(ImmSystemRegister, 19, 5, Bits)                                             \
 V_(SysO0, 19, 19, Bits)                                                        \
+V_(SysOp, 18, 5, Bits)                                                         \
 V_(SysOp1, 18, 16, Bits)                                                       \
 V_(SysOp2, 7, 5, Bits)                                                         \
 V_(CRn, 15, 12, Bits)                                                          \
@@ -130,7 +140,29 @@ V_(LdStXNotExclusive, 23, 23, Bits)                                            \
 V_(LdStXAcquireRelease, 15, 15, Bits)                                          \
 V_(LdStXSizeLog2, 31, 30, Bits)                                                \
 V_(LdStXPair, 21, 21, Bits)                                                    \
-
+                                                                               \
+/* NEON generic fields */                                                      \
+V_(NEONQ, 30, 30, Bits)                                                        \
+V_(NEONSize, 23, 22, Bits)                                                     \
+V_(NEONLSSize, 11, 10, Bits)                                                   \
+V_(NEONS, 12, 12, Bits)                                                        \
+V_(NEONL, 21, 21, Bits)                                                        \
+V_(NEONM, 20, 20, Bits)                                                        \
+V_(NEONH, 11, 11, Bits)                                                        \
+V_(ImmNEONExt, 14, 11, Bits)                                                   \
+V_(ImmNEON5, 20, 16, Bits)                                                     \
+V_(ImmNEON4, 14, 11, Bits)                                                     \
+                                                                               \
+/* NEON Modified Immediate fields */                                           \
+V_(ImmNEONabc, 18, 16, Bits)                                                   \
+V_(ImmNEONdefgh, 9, 5, Bits)                                                   \
+V_(NEONModImmOp, 29, 29, Bits)                                                 \
+V_(NEONCmode, 15, 12, Bits)                                                    \
+                                                                               \
+/* NEON Shift Immediate fields */                                              \
+V_(ImmNEONImmhImmb, 22, 16, Bits)                                              \
+V_(ImmNEONImmh, 22, 19, Bits)                                                  \
+V_(ImmNEONImmb, 18, 16, Bits)
 
 #define SYSTEM_REGISTER_FIELDS_LIST(V_, M_)                                    \
 /* NZCV */                                                                     \
@@ -140,7 +172,6 @@ V_(Z, 30, 30, Bits)                                                            \
 V_(C, 29, 29, Bits)                                                            \
 V_(V, 28, 28, Bits)                                                            \
 M_(NZCV, Flags_mask)                                                           \
-                                                                               \
 /* FPCR */                                                                     \
 V_(AHP, 26, 26, Bits)                                                          \
 V_(DN, 25, 25, Bits)                                                           \
@@ -148,7 +179,6 @@ V_(FZ, 24, 24, Bits)                                                           \
 V_(RMode, 23, 22, Bits)                                                        \
 M_(FPCR, AHP_mask | DN_mask | FZ_mask | RMode_mask)
 
-
 // Fields offsets.
 #define DECLARE_FIELDS_OFFSETS(Name, HighBit, LowBit, X)                       \
 const int Name##_offset = LowBit;                                              \
@@ -166,22 +196,26 @@ const int ImmPCRel_mask = ImmPCRelLo_mask | ImmPCRelHi_mask;
 
 // Condition codes.
 enum Condition {
-  eq = 0,
-  ne = 1,
-  hs = 2,
-  lo = 3,
-  mi = 4,
-  pl = 5,
-  vs = 6,
-  vc = 7,
-  hi = 8,
-  ls = 9,
-  ge = 10,
-  lt = 11,
-  gt = 12,
-  le = 13,
-  al = 14,
-  nv = 15  // Behaves as always/al.
+  eq = 0,   // Z set            Equal.
+  ne = 1,   // Z clear          Not equal.
+  cs = 2,   // C set            Carry set.
+  cc = 3,   // C clear          Carry clear.
+  mi = 4,   // N set            Negative.
+  pl = 5,   // N clear          Positive or zero.
+  vs = 6,   // V set            Overflow.
+  vc = 7,   // V clear          No overflow.
+  hi = 8,   // C set, Z clear   Unsigned higher.
+  ls = 9,   // C clear or Z set Unsigned lower or same.
+  ge = 10,  // N == V           Greater or equal.
+  lt = 11,  // N != V           Less than.
+  gt = 12,  // Z clear, N == V  Greater than.
+  le = 13,  // Z set or N != V  Less then or equal
+  al = 14,  //                  Always.
+  nv = 15,  // Behaves as always/al.
+
+  // Aliases.
+  hs = cs,  // C set            Unsigned higher or same.
+  lo = cc   // C clear          Unsigned lower.
 };
 
 inline Condition InvertCondition(Condition cond) {
@@ -191,6 +225,11 @@ inline Condition InvertCondition(Condition cond) {
   return static_cast<Condition>(cond ^ 1);
 }
 
+enum FPTrapFlags {
+  EnableTrap   = 1,
+  DisableTrap = 0
+};
+
 enum FlagsUpdate {
   SetFlags   = 1,
   LeaveFlags = 0
@@ -228,7 +267,8 @@ enum Shift {
   LSL = 0x0,
   LSR = 0x1,
   ASR = 0x2,
-  ROR = 0x3
+  ROR = 0x3,
+  MSL = 0x4
 };
 
 enum Extend {
@@ -305,6 +345,32 @@ enum SystemRegister {
           (0x0 << SysOp2_offset)) >> ImmSystemRegister_offset
 };
 
+enum InstructionCacheOp {
+  IVAU = ((0x3 << SysOp1_offset) |
+          (0x7 << CRn_offset) |
+          (0x5 << CRm_offset) |
+          (0x1 << SysOp2_offset)) >> SysOp_offset
+};
+
+enum DataCacheOp {
+  CVAC  = ((0x3 << SysOp1_offset) |
+           (0x7 << CRn_offset) |
+           (0xa << CRm_offset) |
+           (0x1 << SysOp2_offset)) >> SysOp_offset,
+  CVAU  = ((0x3 << SysOp1_offset) |
+           (0x7 << CRn_offset) |
+           (0xb << CRm_offset) |
+           (0x1 << SysOp2_offset)) >> SysOp_offset,
+  CIVAC = ((0x3 << SysOp1_offset) |
+           (0x7 << CRn_offset) |
+           (0xe << CRm_offset) |
+           (0x1 << SysOp2_offset)) >> SysOp_offset,
+  ZVA   = ((0x3 << SysOp1_offset) |
+           (0x7 << CRn_offset) |
+           (0x4 << CRm_offset) |
+           (0x1 << SysOp2_offset)) >> SysOp_offset
+};
+
 // Instruction enumerations.
 //
 // These are the masks that define a class of instructions, and the list of
@@ -333,6 +399,47 @@ enum GenericInstrField {
   FP64                 = 0x00400000
 };
 
+enum NEONFormatField {
+  NEONFormatFieldMask   = 0x40C00000,
+  NEON_Q                = 0x40000000,
+  NEON_8B               = 0x00000000,
+  NEON_16B              = NEON_8B | NEON_Q,
+  NEON_4H               = 0x00400000,
+  NEON_8H               = NEON_4H | NEON_Q,
+  NEON_2S               = 0x00800000,
+  NEON_4S               = NEON_2S | NEON_Q,
+  NEON_1D               = 0x00C00000,
+  NEON_2D               = 0x00C00000 | NEON_Q
+};
+
+enum NEONFPFormatField {
+  NEONFPFormatFieldMask = 0x40400000,
+  NEON_FP_2S            = FP32,
+  NEON_FP_4S            = FP32 | NEON_Q,
+  NEON_FP_2D            = FP64 | NEON_Q
+};
+
+enum NEONLSFormatField {
+  NEONLSFormatFieldMask = 0x40000C00,
+  LS_NEON_8B            = 0x00000000,
+  LS_NEON_16B           = LS_NEON_8B | NEON_Q,
+  LS_NEON_4H            = 0x00000400,
+  LS_NEON_8H            = LS_NEON_4H | NEON_Q,
+  LS_NEON_2S            = 0x00000800,
+  LS_NEON_4S            = LS_NEON_2S | NEON_Q,
+  LS_NEON_1D            = 0x00000C00,
+  LS_NEON_2D            = LS_NEON_1D | NEON_Q
+};
+
+enum NEONScalarFormatField {
+  NEONScalarFormatFieldMask = 0x00C00000,
+  NEONScalar                = 0x10000000,
+  NEON_B                    = 0x00000000,
+  NEON_H                    = 0x00400000,
+  NEON_S                    = 0x00800000,
+  NEON_D                    = 0x00C00000
+};
+
 // PC relative addressing.
 enum PCRelAddressingOp {
   PCRelAddressingFixed = 0x10000000,
@@ -588,6 +695,13 @@ enum SystemHintOp {
   HINT            = SystemHintFixed | 0x00000000
 };
 
+enum SystemSysOp {
+  SystemSysFixed  = 0xD5080000,
+  SystemSysFMask  = 0xFFF80000,
+  SystemSysMask   = 0xFFF80000,
+  SYS             = SystemSysFixed | 0x00000000
+};
+
 // Exception.
 enum ExceptionOp {
   ExceptionFixed = 0xD4000000,
@@ -640,7 +754,9 @@ enum LoadStorePairAnyOp {
   V(STP, s,   0x04000000),          \
   V(LDP, s,   0x04400000),          \
   V(STP, d,   0x44000000),          \
-  V(LDP, d,   0x44400000)
+  V(LDP, d,   0x44400000),          \
+  V(STP, q,   0x84000000),          \
+  V(LDP, q,   0x84400000)
 
 // Load/store pair (post, pre and offset.)
 enum LoadStorePairOp {
@@ -686,6 +802,7 @@ enum LoadStorePairNonTemporalOp {
   LoadStorePairNonTemporalFixed = 0x28000000,
   LoadStorePairNonTemporalFMask = 0x3B800000,
   LoadStorePairNonTemporalMask  = 0xFFC00000,
+  LoadStorePairNonTemporalLBit = 1 << 22,
   STNP_w = LoadStorePairNonTemporalFixed | STP_w,
   LDNP_w = LoadStorePairNonTemporalFixed | LDP_w,
   STNP_x = LoadStorePairNonTemporalFixed | STP_x,
@@ -693,7 +810,9 @@ enum LoadStorePairNonTemporalOp {
   STNP_s = LoadStorePairNonTemporalFixed | STP_s,
   LDNP_s = LoadStorePairNonTemporalFixed | LDP_s,
   STNP_d = LoadStorePairNonTemporalFixed | STP_d,
-  LDNP_d = LoadStorePairNonTemporalFixed | LDP_d
+  LDNP_d = LoadStorePairNonTemporalFixed | LDP_d,
+  STNP_q = LoadStorePairNonTemporalFixed | STP_q,
+  LDNP_q = LoadStorePairNonTemporalFixed | LDP_q
 };
 
 // Load literal.
@@ -706,7 +825,8 @@ enum LoadLiteralOp {
   LDRSW_x_lit      = LoadLiteralFixed | 0x80000000,
   PRFM_lit         = LoadLiteralFixed | 0xC0000000,
   LDR_s_lit        = LoadLiteralFixed | 0x04000000,
-  LDR_d_lit        = LoadLiteralFixed | 0x44000000
+  LDR_d_lit        = LoadLiteralFixed | 0x44000000,
+  LDR_q_lit        = LoadLiteralFixed | 0x84000000
 };
 
 #define LOAD_STORE_OP_LIST(V)     \
@@ -723,15 +843,21 @@ enum LoadLiteralOp {
   V(LD, RSW, x, 0x80800000),  \
   V(LD, RSB, w, 0x00C00000),  \
   V(LD, RSH, w, 0x40C00000),  \
+  V(ST, R, b,   0x04000000),  \
+  V(ST, R, h,   0x44000000),  \
   V(ST, R, s,   0x84000000),  \
   V(ST, R, d,   0xC4000000),  \
+  V(ST, R, q,   0x04800000),  \
+  V(LD, R, b,   0x04400000),  \
+  V(LD, R, h,   0x44400000),  \
   V(LD, R, s,   0x84400000),  \
-  V(LD, R, d,   0xC4400000)
-
+  V(LD, R, d,   0xC4400000),  \
+  V(LD, R, q,   0x04C00000)
 
 // Load/store (post, pre, offset and unsigned.)
 enum LoadStoreOp {
-  LoadStoreOpMask = 0xC4C00000,
+  LoadStoreMask = 0xC4C00000,
+  LoadStoreVMask = 0x04000000,
   #define LOAD_STORE(A, B, C, D)  \
   A##B##_##C = D
   LOAD_STORE_OP_LIST(LOAD_STORE),
@@ -971,8 +1097,10 @@ enum FPCompareOp {
   FCMP_zero      = FCMP_s_zero,
   FCMPE_s        = FPCompareFixed | 0x00000010,
   FCMPE_d        = FPCompareFixed | FP64 | 0x00000010,
+  FCMPE          = FCMPE_s,
   FCMPE_s_zero   = FPCompareFixed | 0x00000018,
-  FCMPE_d_zero   = FPCompareFixed | FP64 | 0x00000018
+  FCMPE_d_zero   = FPCompareFixed | FP64 | 0x00000018,
+  FCMPE_zero     = FCMPE_s_zero
 };
 
 // Floating point conditional compare.
@@ -1026,6 +1154,10 @@ enum FPDataProcessing1SourceOp {
   FSQRT    = FSQRT_s,
   FCVT_ds  = FPDataProcessing1SourceFixed | 0x00028000,
   FCVT_sd  = FPDataProcessing1SourceFixed | FP64 | 0x00020000,
+  FCVT_hs  = FPDataProcessing1SourceFixed | 0x00038000,
+  FCVT_hd  = FPDataProcessing1SourceFixed | FP64 | 0x00038000,
+  FCVT_sh  = FPDataProcessing1SourceFixed | 0x00C20000,
+  FCVT_dh  = FPDataProcessing1SourceFixed | 0x00C28000,
   FRINTN_s = FPDataProcessing1SourceFixed | 0x00040000,
   FRINTN_d = FPDataProcessing1SourceFixed | FP64 | 0x00040000,
   FRINTN   = FRINTN_s,
@@ -1166,7 +1298,9 @@ enum FPIntegerConvertOp {
   FMOV_ws   = FPIntegerConvertFixed | 0x00060000,
   FMOV_sw   = FPIntegerConvertFixed | 0x00070000,
   FMOV_xd   = FMOV_ws | SixtyFourBits | FP64,
-  FMOV_dx   = FMOV_sw | SixtyFourBits | FP64
+  FMOV_dx   = FMOV_sw | SixtyFourBits | FP64,
+  FMOV_d1_x = FPIntegerConvertFixed | SixtyFourBits | 0x008F0000,
+  FMOV_x_d1 = FPIntegerConvertFixed | SixtyFourBits | 0x008E0000
 };
 
 // Conversion between fixed point and floating point.
@@ -1196,6 +1330,775 @@ enum FPFixedPointConvertOp {
   UCVTF_dx_fixed  = UCVTF_fixed | SixtyFourBits | FP64
 };
 
+// Crypto - two register SHA.
+enum Crypto2RegSHAOp {
+  Crypto2RegSHAFixed = 0x5E280800,
+  Crypto2RegSHAFMask = 0xFF3E0C00
+};
+
+// Crypto - three register SHA.
+enum Crypto3RegSHAOp {
+  Crypto3RegSHAFixed = 0x5E000000,
+  Crypto3RegSHAFMask = 0xFF208C00
+};
+
+// Crypto - AES.
+enum CryptoAESOp {
+  CryptoAESFixed = 0x4E280800,
+  CryptoAESFMask = 0xFF3E0C00
+};
+
+// NEON instructions with two register operands.
+enum NEON2RegMiscOp {
+  NEON2RegMiscFixed = 0x0E200800,
+  NEON2RegMiscFMask = 0x9F3E0C00,
+  NEON2RegMiscMask  = 0xBF3FFC00,
+  NEON2RegMiscUBit  = 0x20000000,
+  NEON_REV64     = NEON2RegMiscFixed | 0x00000000,
+  NEON_REV32     = NEON2RegMiscFixed | 0x20000000,
+  NEON_REV16     = NEON2RegMiscFixed | 0x00001000,
+  NEON_SADDLP    = NEON2RegMiscFixed | 0x00002000,
+  NEON_UADDLP    = NEON_SADDLP | NEON2RegMiscUBit,
+  NEON_SUQADD    = NEON2RegMiscFixed | 0x00003000,
+  NEON_USQADD    = NEON_SUQADD | NEON2RegMiscUBit,
+  NEON_CLS       = NEON2RegMiscFixed | 0x00004000,
+  NEON_CLZ       = NEON2RegMiscFixed | 0x20004000,
+  NEON_CNT       = NEON2RegMiscFixed | 0x00005000,
+  NEON_RBIT_NOT  = NEON2RegMiscFixed | 0x20005000,
+  NEON_SADALP    = NEON2RegMiscFixed | 0x00006000,
+  NEON_UADALP    = NEON_SADALP | NEON2RegMiscUBit,
+  NEON_SQABS     = NEON2RegMiscFixed | 0x00007000,
+  NEON_SQNEG     = NEON2RegMiscFixed | 0x20007000,
+  NEON_CMGT_zero = NEON2RegMiscFixed | 0x00008000,
+  NEON_CMGE_zero = NEON2RegMiscFixed | 0x20008000,
+  NEON_CMEQ_zero = NEON2RegMiscFixed | 0x00009000,
+  NEON_CMLE_zero = NEON2RegMiscFixed | 0x20009000,
+  NEON_CMLT_zero = NEON2RegMiscFixed | 0x0000A000,
+  NEON_ABS       = NEON2RegMiscFixed | 0x0000B000,
+  NEON_NEG       = NEON2RegMiscFixed | 0x2000B000,
+  NEON_XTN       = NEON2RegMiscFixed | 0x00012000,
+  NEON_SQXTUN    = NEON2RegMiscFixed | 0x20012000,
+  NEON_SHLL      = NEON2RegMiscFixed | 0x20013000,
+  NEON_SQXTN     = NEON2RegMiscFixed | 0x00014000,
+  NEON_UQXTN     = NEON_SQXTN | NEON2RegMiscUBit,
+
+  NEON2RegMiscOpcode = 0x0001F000,
+  NEON_RBIT_NOT_opcode = NEON_RBIT_NOT & NEON2RegMiscOpcode,
+  NEON_NEG_opcode = NEON_NEG & NEON2RegMiscOpcode,
+  NEON_XTN_opcode = NEON_XTN & NEON2RegMiscOpcode,
+  NEON_UQXTN_opcode = NEON_UQXTN & NEON2RegMiscOpcode,
+
+  // These instructions use only one bit of the size field. The other bit is
+  // used to distinguish between instructions.
+  NEON2RegMiscFPMask = NEON2RegMiscMask | 0x00800000,
+  NEON_FABS   = NEON2RegMiscFixed | 0x0080F000,
+  NEON_FNEG   = NEON2RegMiscFixed | 0x2080F000,
+  NEON_FCVTN  = NEON2RegMiscFixed | 0x00016000,
+  NEON_FCVTXN = NEON2RegMiscFixed | 0x20016000,
+  NEON_FCVTL  = NEON2RegMiscFixed | 0x00017000,
+  NEON_FRINTN = NEON2RegMiscFixed | 0x00018000,
+  NEON_FRINTA = NEON2RegMiscFixed | 0x20018000,
+  NEON_FRINTP = NEON2RegMiscFixed | 0x00818000,
+  NEON_FRINTM = NEON2RegMiscFixed | 0x00019000,
+  NEON_FRINTX = NEON2RegMiscFixed | 0x20019000,
+  NEON_FRINTZ = NEON2RegMiscFixed | 0x00819000,
+  NEON_FRINTI = NEON2RegMiscFixed | 0x20819000,
+  NEON_FCVTNS = NEON2RegMiscFixed | 0x0001A000,
+  NEON_FCVTNU = NEON_FCVTNS | NEON2RegMiscUBit,
+  NEON_FCVTPS = NEON2RegMiscFixed | 0x0081A000,
+  NEON_FCVTPU = NEON_FCVTPS | NEON2RegMiscUBit,
+  NEON_FCVTMS = NEON2RegMiscFixed | 0x0001B000,
+  NEON_FCVTMU = NEON_FCVTMS | NEON2RegMiscUBit,
+  NEON_FCVTZS = NEON2RegMiscFixed | 0x0081B000,
+  NEON_FCVTZU = NEON_FCVTZS | NEON2RegMiscUBit,
+  NEON_FCVTAS = NEON2RegMiscFixed | 0x0001C000,
+  NEON_FCVTAU = NEON_FCVTAS | NEON2RegMiscUBit,
+  NEON_FSQRT  = NEON2RegMiscFixed | 0x2081F000,
+  NEON_SCVTF  = NEON2RegMiscFixed | 0x0001D000,
+  NEON_UCVTF  = NEON_SCVTF | NEON2RegMiscUBit,
+  NEON_URSQRTE = NEON2RegMiscFixed | 0x2081C000,
+  NEON_URECPE  = NEON2RegMiscFixed | 0x0081C000,
+  NEON_FRSQRTE = NEON2RegMiscFixed | 0x2081D000,
+  NEON_FRECPE  = NEON2RegMiscFixed | 0x0081D000,
+  NEON_FCMGT_zero = NEON2RegMiscFixed | 0x0080C000,
+  NEON_FCMGE_zero = NEON2RegMiscFixed | 0x2080C000,
+  NEON_FCMEQ_zero = NEON2RegMiscFixed | 0x0080D000,
+  NEON_FCMLE_zero = NEON2RegMiscFixed | 0x2080D000,
+  NEON_FCMLT_zero = NEON2RegMiscFixed | 0x0080E000,
+
+  NEON_FCVTL_opcode = NEON_FCVTL & NEON2RegMiscOpcode,
+  NEON_FCVTN_opcode = NEON_FCVTN & NEON2RegMiscOpcode
+};
+
+// NEON instructions with three same-type operands.
+enum NEON3SameOp {
+  NEON3SameFixed = 0x0E200400,
+  NEON3SameFMask = 0x9F200400,
+  NEON3SameMask = 0xBF20FC00,
+  NEON3SameUBit = 0x20000000,
+  NEON_ADD    = NEON3SameFixed | 0x00008000,
+  NEON_ADDP   = NEON3SameFixed | 0x0000B800,
+  NEON_SHADD  = NEON3SameFixed | 0x00000000,
+  NEON_SHSUB  = NEON3SameFixed | 0x00002000,
+  NEON_SRHADD = NEON3SameFixed | 0x00001000,
+  NEON_CMEQ   = NEON3SameFixed | NEON3SameUBit | 0x00008800,
+  NEON_CMGE   = NEON3SameFixed | 0x00003800,
+  NEON_CMGT   = NEON3SameFixed | 0x00003000,
+  NEON_CMHI   = NEON3SameFixed | NEON3SameUBit | NEON_CMGT,
+  NEON_CMHS   = NEON3SameFixed | NEON3SameUBit | NEON_CMGE,
+  NEON_CMTST  = NEON3SameFixed | 0x00008800,
+  NEON_MLA    = NEON3SameFixed | 0x00009000,
+  NEON_MLS    = NEON3SameFixed | 0x20009000,
+  NEON_MUL    = NEON3SameFixed | 0x00009800,
+  NEON_PMUL   = NEON3SameFixed | 0x20009800,
+  NEON_SRSHL  = NEON3SameFixed | 0x00005000,
+  NEON_SQSHL  = NEON3SameFixed | 0x00004800,
+  NEON_SQRSHL = NEON3SameFixed | 0x00005800,
+  NEON_SSHL   = NEON3SameFixed | 0x00004000,
+  NEON_SMAX   = NEON3SameFixed | 0x00006000,
+  NEON_SMAXP  = NEON3SameFixed | 0x0000A000,
+  NEON_SMIN   = NEON3SameFixed | 0x00006800,
+  NEON_SMINP  = NEON3SameFixed | 0x0000A800,
+  NEON_SABD   = NEON3SameFixed | 0x00007000,
+  NEON_SABA   = NEON3SameFixed | 0x00007800,
+  NEON_UABD   = NEON3SameFixed | NEON3SameUBit | NEON_SABD,
+  NEON_UABA   = NEON3SameFixed | NEON3SameUBit | NEON_SABA,
+  NEON_SQADD  = NEON3SameFixed | 0x00000800,
+  NEON_SQSUB  = NEON3SameFixed | 0x00002800,
+  NEON_SUB    = NEON3SameFixed | NEON3SameUBit | 0x00008000,
+  NEON_UHADD  = NEON3SameFixed | NEON3SameUBit | NEON_SHADD,
+  NEON_UHSUB  = NEON3SameFixed | NEON3SameUBit | NEON_SHSUB,
+  NEON_URHADD = NEON3SameFixed | NEON3SameUBit | NEON_SRHADD,
+  NEON_UMAX   = NEON3SameFixed | NEON3SameUBit | NEON_SMAX,
+  NEON_UMAXP  = NEON3SameFixed | NEON3SameUBit | NEON_SMAXP,
+  NEON_UMIN   = NEON3SameFixed | NEON3SameUBit | NEON_SMIN,
+  NEON_UMINP  = NEON3SameFixed | NEON3SameUBit | NEON_SMINP,
+  NEON_URSHL  = NEON3SameFixed | NEON3SameUBit | NEON_SRSHL,
+  NEON_UQADD  = NEON3SameFixed | NEON3SameUBit | NEON_SQADD,
+  NEON_UQRSHL = NEON3SameFixed | NEON3SameUBit | NEON_SQRSHL,
+  NEON_UQSHL  = NEON3SameFixed | NEON3SameUBit | NEON_SQSHL,
+  NEON_UQSUB  = NEON3SameFixed | NEON3SameUBit | NEON_SQSUB,
+  NEON_USHL   = NEON3SameFixed | NEON3SameUBit | NEON_SSHL,
+  NEON_SQDMULH  = NEON3SameFixed | 0x0000B000,
+  NEON_SQRDMULH = NEON3SameFixed | 0x2000B000,
+
+  // NEON floating point instructions with three same-type operands.
+  NEON3SameFPFixed = NEON3SameFixed | 0x0000C000,
+  NEON3SameFPFMask = NEON3SameFMask | 0x0000C000,
+  NEON3SameFPMask = NEON3SameMask | 0x00800000,
+  NEON_FADD    = NEON3SameFixed | 0x0000D000,
+  NEON_FSUB    = NEON3SameFixed | 0x0080D000,
+  NEON_FMUL    = NEON3SameFixed | 0x2000D800,
+  NEON_FDIV    = NEON3SameFixed | 0x2000F800,
+  NEON_FMAX    = NEON3SameFixed | 0x0000F000,
+  NEON_FMAXNM  = NEON3SameFixed | 0x0000C000,
+  NEON_FMAXP   = NEON3SameFixed | 0x2000F000,
+  NEON_FMAXNMP = NEON3SameFixed | 0x2000C000,
+  NEON_FMIN    = NEON3SameFixed | 0x0080F000,
+  NEON_FMINNM  = NEON3SameFixed | 0x0080C000,
+  NEON_FMINP   = NEON3SameFixed | 0x2080F000,
+  NEON_FMINNMP = NEON3SameFixed | 0x2080C000,
+  NEON_FMLA    = NEON3SameFixed | 0x0000C800,
+  NEON_FMLS    = NEON3SameFixed | 0x0080C800,
+  NEON_FMULX   = NEON3SameFixed | 0x0000D800,
+  NEON_FRECPS  = NEON3SameFixed | 0x0000F800,
+  NEON_FRSQRTS = NEON3SameFixed | 0x0080F800,
+  NEON_FABD    = NEON3SameFixed | 0x2080D000,
+  NEON_FADDP   = NEON3SameFixed | 0x2000D000,
+  NEON_FCMEQ   = NEON3SameFixed | 0x0000E000,
+  NEON_FCMGE   = NEON3SameFixed | 0x2000E000,
+  NEON_FCMGT   = NEON3SameFixed | 0x2080E000,
+  NEON_FACGE   = NEON3SameFixed | 0x2000E800,
+  NEON_FACGT   = NEON3SameFixed | 0x2080E800,
+
+  // NEON logical instructions with three same-type operands.
+  NEON3SameLogicalFixed = NEON3SameFixed | 0x00001800,
+  NEON3SameLogicalFMask = NEON3SameFMask | 0x0000F800,
+  NEON3SameLogicalMask = 0xBFE0FC00,
+  NEON3SameLogicalFormatMask = NEON_Q,
+  NEON_AND = NEON3SameLogicalFixed | 0x00000000,
+  NEON_ORR = NEON3SameLogicalFixed | 0x00A00000,
+  NEON_ORN = NEON3SameLogicalFixed | 0x00C00000,
+  NEON_EOR = NEON3SameLogicalFixed | 0x20000000,
+  NEON_BIC = NEON3SameLogicalFixed | 0x00400000,
+  NEON_BIF = NEON3SameLogicalFixed | 0x20C00000,
+  NEON_BIT = NEON3SameLogicalFixed | 0x20800000,
+  NEON_BSL = NEON3SameLogicalFixed | 0x20400000
+};
+
+// NEON instructions with three different-type operands.
+enum NEON3DifferentOp {
+  NEON3DifferentFixed = 0x0E200000,
+  NEON3DifferentFMask = 0x9F200C00,
+  NEON3DifferentMask  = 0xFF20FC00,
+  NEON_ADDHN    = NEON3DifferentFixed | 0x00004000,
+  NEON_ADDHN2   = NEON_ADDHN | NEON_Q,
+  NEON_PMULL    = NEON3DifferentFixed | 0x0000E000,
+  NEON_PMULL2   = NEON_PMULL | NEON_Q,
+  NEON_RADDHN   = NEON3DifferentFixed | 0x20004000,
+  NEON_RADDHN2  = NEON_RADDHN | NEON_Q,
+  NEON_RSUBHN   = NEON3DifferentFixed | 0x20006000,
+  NEON_RSUBHN2  = NEON_RSUBHN | NEON_Q,
+  NEON_SABAL    = NEON3DifferentFixed | 0x00005000,
+  NEON_SABAL2   = NEON_SABAL | NEON_Q,
+  NEON_SABDL    = NEON3DifferentFixed | 0x00007000,
+  NEON_SABDL2   = NEON_SABDL | NEON_Q,
+  NEON_SADDL    = NEON3DifferentFixed | 0x00000000,
+  NEON_SADDL2   = NEON_SADDL | NEON_Q,
+  NEON_SADDW    = NEON3DifferentFixed | 0x00001000,
+  NEON_SADDW2   = NEON_SADDW | NEON_Q,
+  NEON_SMLAL    = NEON3DifferentFixed | 0x00008000,
+  NEON_SMLAL2   = NEON_SMLAL | NEON_Q,
+  NEON_SMLSL    = NEON3DifferentFixed | 0x0000A000,
+  NEON_SMLSL2   = NEON_SMLSL | NEON_Q,
+  NEON_SMULL    = NEON3DifferentFixed | 0x0000C000,
+  NEON_SMULL2   = NEON_SMULL | NEON_Q,
+  NEON_SSUBL    = NEON3DifferentFixed | 0x00002000,
+  NEON_SSUBL2   = NEON_SSUBL | NEON_Q,
+  NEON_SSUBW    = NEON3DifferentFixed | 0x00003000,
+  NEON_SSUBW2   = NEON_SSUBW | NEON_Q,
+  NEON_SQDMLAL  = NEON3DifferentFixed | 0x00009000,
+  NEON_SQDMLAL2 = NEON_SQDMLAL | NEON_Q,
+  NEON_SQDMLSL  = NEON3DifferentFixed | 0x0000B000,
+  NEON_SQDMLSL2 = NEON_SQDMLSL | NEON_Q,
+  NEON_SQDMULL  = NEON3DifferentFixed | 0x0000D000,
+  NEON_SQDMULL2 = NEON_SQDMULL | NEON_Q,
+  NEON_SUBHN    = NEON3DifferentFixed | 0x00006000,
+  NEON_SUBHN2   = NEON_SUBHN | NEON_Q,
+  NEON_UABAL    = NEON_SABAL | NEON3SameUBit,
+  NEON_UABAL2   = NEON_UABAL | NEON_Q,
+  NEON_UABDL    = NEON_SABDL | NEON3SameUBit,
+  NEON_UABDL2   = NEON_UABDL | NEON_Q,
+  NEON_UADDL    = NEON_SADDL | NEON3SameUBit,
+  NEON_UADDL2   = NEON_UADDL | NEON_Q,
+  NEON_UADDW    = NEON_SADDW | NEON3SameUBit,
+  NEON_UADDW2   = NEON_UADDW | NEON_Q,
+  NEON_UMLAL    = NEON_SMLAL | NEON3SameUBit,
+  NEON_UMLAL2   = NEON_UMLAL | NEON_Q,
+  NEON_UMLSL    = NEON_SMLSL | NEON3SameUBit,
+  NEON_UMLSL2   = NEON_UMLSL | NEON_Q,
+  NEON_UMULL    = NEON_SMULL | NEON3SameUBit,
+  NEON_UMULL2   = NEON_UMULL | NEON_Q,
+  NEON_USUBL    = NEON_SSUBL | NEON3SameUBit,
+  NEON_USUBL2   = NEON_USUBL | NEON_Q,
+  NEON_USUBW    = NEON_SSUBW | NEON3SameUBit,
+  NEON_USUBW2   = NEON_USUBW | NEON_Q
+};
+
+// NEON instructions operating across vectors.
+enum NEONAcrossLanesOp {
+  NEONAcrossLanesFixed = 0x0E300800,
+  NEONAcrossLanesFMask = 0x9F3E0C00,
+  NEONAcrossLanesMask  = 0xBF3FFC00,
+  NEON_ADDV   = NEONAcrossLanesFixed | 0x0001B000,
+  NEON_SADDLV = NEONAcrossLanesFixed | 0x00003000,
+  NEON_UADDLV = NEONAcrossLanesFixed | 0x20003000,
+  NEON_SMAXV  = NEONAcrossLanesFixed | 0x0000A000,
+  NEON_SMINV  = NEONAcrossLanesFixed | 0x0001A000,
+  NEON_UMAXV  = NEONAcrossLanesFixed | 0x2000A000,
+  NEON_UMINV  = NEONAcrossLanesFixed | 0x2001A000,
+
+  // NEON floating point across instructions.
+  NEONAcrossLanesFPFixed = NEONAcrossLanesFixed | 0x0000C000,
+  NEONAcrossLanesFPFMask = NEONAcrossLanesFMask | 0x0000C000,
+  NEONAcrossLanesFPMask  = NEONAcrossLanesMask  | 0x00800000,
+
+  NEON_FMAXV   = NEONAcrossLanesFPFixed | 0x2000F000,
+  NEON_FMINV   = NEONAcrossLanesFPFixed | 0x2080F000,
+  NEON_FMAXNMV = NEONAcrossLanesFPFixed | 0x2000C000,
+  NEON_FMINNMV = NEONAcrossLanesFPFixed | 0x2080C000
+};
+
+// NEON instructions with indexed element operand.
+enum NEONByIndexedElementOp {
+  NEONByIndexedElementFixed = 0x0F000000,
+  NEONByIndexedElementFMask = 0x9F000400,
+  NEONByIndexedElementMask  = 0xBF00F400,
+  NEON_MUL_byelement   = NEONByIndexedElementFixed | 0x00008000,
+  NEON_MLA_byelement   = NEONByIndexedElementFixed | 0x20000000,
+  NEON_MLS_byelement   = NEONByIndexedElementFixed | 0x20004000,
+  NEON_SMULL_byelement = NEONByIndexedElementFixed | 0x0000A000,
+  NEON_SMLAL_byelement = NEONByIndexedElementFixed | 0x00002000,
+  NEON_SMLSL_byelement = NEONByIndexedElementFixed | 0x00006000,
+  NEON_UMULL_byelement = NEONByIndexedElementFixed | 0x2000A000,
+  NEON_UMLAL_byelement = NEONByIndexedElementFixed | 0x20002000,
+  NEON_UMLSL_byelement = NEONByIndexedElementFixed | 0x20006000,
+  NEON_SQDMULL_byelement = NEONByIndexedElementFixed | 0x0000B000,
+  NEON_SQDMLAL_byelement = NEONByIndexedElementFixed | 0x00003000,
+  NEON_SQDMLSL_byelement = NEONByIndexedElementFixed | 0x00007000,
+  NEON_SQDMULH_byelement  = NEONByIndexedElementFixed | 0x0000C000,
+  NEON_SQRDMULH_byelement = NEONByIndexedElementFixed | 0x0000D000,
+
+  // Floating point instructions.
+  NEONByIndexedElementFPFixed = NEONByIndexedElementFixed | 0x00800000,
+  NEONByIndexedElementFPMask = NEONByIndexedElementMask | 0x00800000,
+  NEON_FMLA_byelement  = NEONByIndexedElementFPFixed | 0x00001000,
+  NEON_FMLS_byelement  = NEONByIndexedElementFPFixed | 0x00005000,
+  NEON_FMUL_byelement  = NEONByIndexedElementFPFixed | 0x00009000,
+  NEON_FMULX_byelement = NEONByIndexedElementFPFixed | 0x20009000
+};
+
+// NEON register copy.
+enum NEONCopyOp {
+  NEONCopyFixed = 0x0E000400,
+  NEONCopyFMask = 0x9FE08400,
+  NEONCopyMask  = 0x3FE08400,
+  NEONCopyInsElementMask = NEONCopyMask | 0x40000000,
+  NEONCopyInsGeneralMask = NEONCopyMask | 0x40007800,
+  NEONCopyDupElementMask = NEONCopyMask | 0x20007800,
+  NEONCopyDupGeneralMask = NEONCopyDupElementMask,
+  NEONCopyUmovMask       = NEONCopyMask | 0x20007800,
+  NEONCopySmovMask       = NEONCopyMask | 0x20007800,
+  NEON_INS_ELEMENT       = NEONCopyFixed | 0x60000000,
+  NEON_INS_GENERAL       = NEONCopyFixed | 0x40001800,
+  NEON_DUP_ELEMENT       = NEONCopyFixed | 0x00000000,
+  NEON_DUP_GENERAL       = NEONCopyFixed | 0x00000800,
+  NEON_SMOV              = NEONCopyFixed | 0x00002800,
+  NEON_UMOV              = NEONCopyFixed | 0x00003800
+};
+
+// NEON extract.
+enum NEONExtractOp {
+  NEONExtractFixed = 0x2E000000,
+  NEONExtractFMask = 0xBF208400,
+  NEONExtractMask = 0xBFE08400,
+  NEON_EXT = NEONExtractFixed | 0x00000000
+};
+
+enum NEONLoadStoreMultiOp {
+  NEONLoadStoreMultiL    = 0x00400000,
+  NEONLoadStoreMulti1_1v = 0x00007000,
+  NEONLoadStoreMulti1_2v = 0x0000A000,
+  NEONLoadStoreMulti1_3v = 0x00006000,
+  NEONLoadStoreMulti1_4v = 0x00002000,
+  NEONLoadStoreMulti2    = 0x00008000,
+  NEONLoadStoreMulti3    = 0x00004000,
+  NEONLoadStoreMulti4    = 0x00000000
+};
+
+// NEON load/store multiple structures.
+enum NEONLoadStoreMultiStructOp {
+  NEONLoadStoreMultiStructFixed = 0x0C000000,
+  NEONLoadStoreMultiStructFMask = 0xBFBF0000,
+  NEONLoadStoreMultiStructMask  = 0xBFFFF000,
+  NEONLoadStoreMultiStructStore = NEONLoadStoreMultiStructFixed,
+  NEONLoadStoreMultiStructLoad  = NEONLoadStoreMultiStructFixed |
+                                  NEONLoadStoreMultiL,
+  NEON_LD1_1v = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti1_1v,
+  NEON_LD1_2v = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti1_2v,
+  NEON_LD1_3v = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti1_3v,
+  NEON_LD1_4v = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti1_4v,
+  NEON_LD2    = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti2,
+  NEON_LD3    = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti3,
+  NEON_LD4    = NEONLoadStoreMultiStructLoad | NEONLoadStoreMulti4,
+  NEON_ST1_1v = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti1_1v,
+  NEON_ST1_2v = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti1_2v,
+  NEON_ST1_3v = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti1_3v,
+  NEON_ST1_4v = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti1_4v,
+  NEON_ST2    = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti2,
+  NEON_ST3    = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti3,
+  NEON_ST4    = NEONLoadStoreMultiStructStore | NEONLoadStoreMulti4
+};
+
+// NEON load/store multiple structures with post-index addressing.
+enum NEONLoadStoreMultiStructPostIndexOp {
+  NEONLoadStoreMultiStructPostIndexFixed = 0x0C800000,
+  NEONLoadStoreMultiStructPostIndexFMask = 0xBFA00000,
+  NEONLoadStoreMultiStructPostIndexMask  = 0xBFE0F000,
+  NEONLoadStoreMultiStructPostIndex = 0x00800000,
+  NEON_LD1_1v_post = NEON_LD1_1v | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD1_2v_post = NEON_LD1_2v | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD1_3v_post = NEON_LD1_3v | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD1_4v_post = NEON_LD1_4v | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD2_post = NEON_LD2 | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD3_post = NEON_LD3 | NEONLoadStoreMultiStructPostIndex,
+  NEON_LD4_post = NEON_LD4 | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST1_1v_post = NEON_ST1_1v | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST1_2v_post = NEON_ST1_2v | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST1_3v_post = NEON_ST1_3v | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST1_4v_post = NEON_ST1_4v | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST2_post = NEON_ST2 | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST3_post = NEON_ST3 | NEONLoadStoreMultiStructPostIndex,
+  NEON_ST4_post = NEON_ST4 | NEONLoadStoreMultiStructPostIndex
+};
+
+enum NEONLoadStoreSingleOp {
+  NEONLoadStoreSingle1        = 0x00000000,
+  NEONLoadStoreSingle2        = 0x00200000,
+  NEONLoadStoreSingle3        = 0x00002000,
+  NEONLoadStoreSingle4        = 0x00202000,
+  NEONLoadStoreSingleL        = 0x00400000,
+  NEONLoadStoreSingle_b       = 0x00000000,
+  NEONLoadStoreSingle_h       = 0x00004000,
+  NEONLoadStoreSingle_s       = 0x00008000,
+  NEONLoadStoreSingle_d       = 0x00008400,
+  NEONLoadStoreSingleAllLanes = 0x0000C000,
+  NEONLoadStoreSingleLenMask  = 0x00202000
+};
+
+// NEON load/store single structure.
+enum NEONLoadStoreSingleStructOp {
+  NEONLoadStoreSingleStructFixed = 0x0D000000,
+  NEONLoadStoreSingleStructFMask = 0xBF9F0000,
+  NEONLoadStoreSingleStructMask  = 0xBFFFE000,
+  NEONLoadStoreSingleStructStore = NEONLoadStoreSingleStructFixed,
+  NEONLoadStoreSingleStructLoad  = NEONLoadStoreSingleStructFixed |
+                                   NEONLoadStoreSingleL,
+  NEONLoadStoreSingleStructLoad1 = NEONLoadStoreSingle1 |
+                                   NEONLoadStoreSingleStructLoad,
+  NEONLoadStoreSingleStructLoad2 = NEONLoadStoreSingle2 |
+                                   NEONLoadStoreSingleStructLoad,
+  NEONLoadStoreSingleStructLoad3 = NEONLoadStoreSingle3 |
+                                   NEONLoadStoreSingleStructLoad,
+  NEONLoadStoreSingleStructLoad4 = NEONLoadStoreSingle4 |
+                                   NEONLoadStoreSingleStructLoad,
+  NEONLoadStoreSingleStructStore1 = NEONLoadStoreSingle1 |
+                                    NEONLoadStoreSingleStructFixed,
+  NEONLoadStoreSingleStructStore2 = NEONLoadStoreSingle2 |
+                                    NEONLoadStoreSingleStructFixed,
+  NEONLoadStoreSingleStructStore3 = NEONLoadStoreSingle3 |
+                                    NEONLoadStoreSingleStructFixed,
+  NEONLoadStoreSingleStructStore4 = NEONLoadStoreSingle4 |
+                                    NEONLoadStoreSingleStructFixed,
+  NEON_LD1_b = NEONLoadStoreSingleStructLoad1 | NEONLoadStoreSingle_b,
+  NEON_LD1_h = NEONLoadStoreSingleStructLoad1 | NEONLoadStoreSingle_h,
+  NEON_LD1_s = NEONLoadStoreSingleStructLoad1 | NEONLoadStoreSingle_s,
+  NEON_LD1_d = NEONLoadStoreSingleStructLoad1 | NEONLoadStoreSingle_d,
+  NEON_LD1R  = NEONLoadStoreSingleStructLoad1 | NEONLoadStoreSingleAllLanes,
+  NEON_ST1_b = NEONLoadStoreSingleStructStore1 | NEONLoadStoreSingle_b,
+  NEON_ST1_h = NEONLoadStoreSingleStructStore1 | NEONLoadStoreSingle_h,
+  NEON_ST1_s = NEONLoadStoreSingleStructStore1 | NEONLoadStoreSingle_s,
+  NEON_ST1_d = NEONLoadStoreSingleStructStore1 | NEONLoadStoreSingle_d,
+
+  NEON_LD2_b = NEONLoadStoreSingleStructLoad2 | NEONLoadStoreSingle_b,
+  NEON_LD2_h = NEONLoadStoreSingleStructLoad2 | NEONLoadStoreSingle_h,
+  NEON_LD2_s = NEONLoadStoreSingleStructLoad2 | NEONLoadStoreSingle_s,
+  NEON_LD2_d = NEONLoadStoreSingleStructLoad2 | NEONLoadStoreSingle_d,
+  NEON_LD2R  = NEONLoadStoreSingleStructLoad2 | NEONLoadStoreSingleAllLanes,
+  NEON_ST2_b = NEONLoadStoreSingleStructStore2 | NEONLoadStoreSingle_b,
+  NEON_ST2_h = NEONLoadStoreSingleStructStore2 | NEONLoadStoreSingle_h,
+  NEON_ST2_s = NEONLoadStoreSingleStructStore2 | NEONLoadStoreSingle_s,
+  NEON_ST2_d = NEONLoadStoreSingleStructStore2 | NEONLoadStoreSingle_d,
+
+  NEON_LD3_b = NEONLoadStoreSingleStructLoad3 | NEONLoadStoreSingle_b,
+  NEON_LD3_h = NEONLoadStoreSingleStructLoad3 | NEONLoadStoreSingle_h,
+  NEON_LD3_s = NEONLoadStoreSingleStructLoad3 | NEONLoadStoreSingle_s,
+  NEON_LD3_d = NEONLoadStoreSingleStructLoad3 | NEONLoadStoreSingle_d,
+  NEON_LD3R  = NEONLoadStoreSingleStructLoad3 | NEONLoadStoreSingleAllLanes,
+  NEON_ST3_b = NEONLoadStoreSingleStructStore3 | NEONLoadStoreSingle_b,
+  NEON_ST3_h = NEONLoadStoreSingleStructStore3 | NEONLoadStoreSingle_h,
+  NEON_ST3_s = NEONLoadStoreSingleStructStore3 | NEONLoadStoreSingle_s,
+  NEON_ST3_d = NEONLoadStoreSingleStructStore3 | NEONLoadStoreSingle_d,
+
+  NEON_LD4_b = NEONLoadStoreSingleStructLoad4 | NEONLoadStoreSingle_b,
+  NEON_LD4_h = NEONLoadStoreSingleStructLoad4 | NEONLoadStoreSingle_h,
+  NEON_LD4_s = NEONLoadStoreSingleStructLoad4 | NEONLoadStoreSingle_s,
+  NEON_LD4_d = NEONLoadStoreSingleStructLoad4 | NEONLoadStoreSingle_d,
+  NEON_LD4R  = NEONLoadStoreSingleStructLoad4 | NEONLoadStoreSingleAllLanes,
+  NEON_ST4_b = NEONLoadStoreSingleStructStore4 | NEONLoadStoreSingle_b,
+  NEON_ST4_h = NEONLoadStoreSingleStructStore4 | NEONLoadStoreSingle_h,
+  NEON_ST4_s = NEONLoadStoreSingleStructStore4 | NEONLoadStoreSingle_s,
+  NEON_ST4_d = NEONLoadStoreSingleStructStore4 | NEONLoadStoreSingle_d
+};
+
+// NEON load/store single structure with post-index addressing.
+enum NEONLoadStoreSingleStructPostIndexOp {
+  NEONLoadStoreSingleStructPostIndexFixed = 0x0D800000,
+  NEONLoadStoreSingleStructPostIndexFMask = 0xBF800000,
+  NEONLoadStoreSingleStructPostIndexMask  = 0xBFE0E000,
+  NEONLoadStoreSingleStructPostIndex = 0x00800000,
+  NEON_LD1_b_post = NEON_LD1_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD1_h_post = NEON_LD1_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD1_s_post = NEON_LD1_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD1_d_post = NEON_LD1_d | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD1R_post  = NEON_LD1R | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST1_b_post = NEON_ST1_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST1_h_post = NEON_ST1_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST1_s_post = NEON_ST1_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST1_d_post = NEON_ST1_d | NEONLoadStoreSingleStructPostIndex,
+
+  NEON_LD2_b_post = NEON_LD2_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD2_h_post = NEON_LD2_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD2_s_post = NEON_LD2_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD2_d_post = NEON_LD2_d | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD2R_post  = NEON_LD2R | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST2_b_post = NEON_ST2_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST2_h_post = NEON_ST2_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST2_s_post = NEON_ST2_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST2_d_post = NEON_ST2_d | NEONLoadStoreSingleStructPostIndex,
+
+  NEON_LD3_b_post = NEON_LD3_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD3_h_post = NEON_LD3_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD3_s_post = NEON_LD3_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD3_d_post = NEON_LD3_d | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD3R_post  = NEON_LD3R | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST3_b_post = NEON_ST3_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST3_h_post = NEON_ST3_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST3_s_post = NEON_ST3_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST3_d_post = NEON_ST3_d | NEONLoadStoreSingleStructPostIndex,
+
+  NEON_LD4_b_post = NEON_LD4_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD4_h_post = NEON_LD4_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD4_s_post = NEON_LD4_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD4_d_post = NEON_LD4_d | NEONLoadStoreSingleStructPostIndex,
+  NEON_LD4R_post  = NEON_LD4R | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST4_b_post = NEON_ST4_b | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST4_h_post = NEON_ST4_h | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST4_s_post = NEON_ST4_s | NEONLoadStoreSingleStructPostIndex,
+  NEON_ST4_d_post = NEON_ST4_d | NEONLoadStoreSingleStructPostIndex
+};
+
+// NEON modified immediate.
+enum NEONModifiedImmediateOp {
+  NEONModifiedImmediateFixed = 0x0F000400,
+  NEONModifiedImmediateFMask = 0x9FF80400,
+  NEONModifiedImmediateOpBit = 0x20000000,
+  NEONModifiedImmediate_MOVI = NEONModifiedImmediateFixed | 0x00000000,
+  NEONModifiedImmediate_MVNI = NEONModifiedImmediateFixed | 0x20000000,
+  NEONModifiedImmediate_ORR  = NEONModifiedImmediateFixed | 0x00001000,
+  NEONModifiedImmediate_BIC  = NEONModifiedImmediateFixed | 0x20001000
+};
+
+// NEON shift immediate.
+enum NEONShiftImmediateOp {
+  NEONShiftImmediateFixed = 0x0F000400,
+  NEONShiftImmediateFMask = 0x9F800400,
+  NEONShiftImmediateMask  = 0xBF80FC00,
+  NEONShiftImmediateUBit  = 0x20000000,
+  NEON_SHL      = NEONShiftImmediateFixed | 0x00005000,
+  NEON_SSHLL    = NEONShiftImmediateFixed | 0x0000A000,
+  NEON_USHLL    = NEONShiftImmediateFixed | 0x2000A000,
+  NEON_SLI      = NEONShiftImmediateFixed | 0x20005000,
+  NEON_SRI      = NEONShiftImmediateFixed | 0x20004000,
+  NEON_SHRN     = NEONShiftImmediateFixed | 0x00008000,
+  NEON_RSHRN    = NEONShiftImmediateFixed | 0x00008800,
+  NEON_UQSHRN   = NEONShiftImmediateFixed | 0x20009000,
+  NEON_UQRSHRN  = NEONShiftImmediateFixed | 0x20009800,
+  NEON_SQSHRN   = NEONShiftImmediateFixed | 0x00009000,
+  NEON_SQRSHRN  = NEONShiftImmediateFixed | 0x00009800,
+  NEON_SQSHRUN  = NEONShiftImmediateFixed | 0x20008000,
+  NEON_SQRSHRUN = NEONShiftImmediateFixed | 0x20008800,
+  NEON_SSHR     = NEONShiftImmediateFixed | 0x00000000,
+  NEON_SRSHR    = NEONShiftImmediateFixed | 0x00002000,
+  NEON_USHR     = NEONShiftImmediateFixed | 0x20000000,
+  NEON_URSHR    = NEONShiftImmediateFixed | 0x20002000,
+  NEON_SSRA     = NEONShiftImmediateFixed | 0x00001000,
+  NEON_SRSRA    = NEONShiftImmediateFixed | 0x00003000,
+  NEON_USRA     = NEONShiftImmediateFixed | 0x20001000,
+  NEON_URSRA    = NEONShiftImmediateFixed | 0x20003000,
+  NEON_SQSHLU   = NEONShiftImmediateFixed | 0x20006000,
+  NEON_SCVTF_imm = NEONShiftImmediateFixed | 0x0000E000,
+  NEON_UCVTF_imm = NEONShiftImmediateFixed | 0x2000E000,
+  NEON_FCVTZS_imm = NEONShiftImmediateFixed | 0x0000F800,
+  NEON_FCVTZU_imm = NEONShiftImmediateFixed | 0x2000F800,
+  NEON_SQSHL_imm = NEONShiftImmediateFixed | 0x00007000,
+  NEON_UQSHL_imm = NEONShiftImmediateFixed | 0x20007000
+};
+
+// NEON table.
+enum NEONTableOp {
+  NEONTableFixed = 0x0E000000,
+  NEONTableFMask = 0xBF208C00,
+  NEONTableExt   = 0x00001000,
+  NEONTableMask  = 0xBF20FC00,
+  NEON_TBL_1v    = NEONTableFixed | 0x00000000,
+  NEON_TBL_2v    = NEONTableFixed | 0x00002000,
+  NEON_TBL_3v    = NEONTableFixed | 0x00004000,
+  NEON_TBL_4v    = NEONTableFixed | 0x00006000,
+  NEON_TBX_1v    = NEON_TBL_1v | NEONTableExt,
+  NEON_TBX_2v    = NEON_TBL_2v | NEONTableExt,
+  NEON_TBX_3v    = NEON_TBL_3v | NEONTableExt,
+  NEON_TBX_4v    = NEON_TBL_4v | NEONTableExt
+};
+
+// NEON perm.
+enum NEONPermOp {
+  NEONPermFixed = 0x0E000800,
+  NEONPermFMask = 0xBF208C00,
+  NEONPermMask  = 0x3F20FC00,
+  NEON_UZP1 = NEONPermFixed | 0x00001000,
+  NEON_TRN1 = NEONPermFixed | 0x00002000,
+  NEON_ZIP1 = NEONPermFixed | 0x00003000,
+  NEON_UZP2 = NEONPermFixed | 0x00005000,
+  NEON_TRN2 = NEONPermFixed | 0x00006000,
+  NEON_ZIP2 = NEONPermFixed | 0x00007000
+};
+
+// NEON scalar instructions with two register operands.
+enum NEONScalar2RegMiscOp {
+  NEONScalar2RegMiscFixed = 0x5E200800,
+  NEONScalar2RegMiscFMask = 0xDF3E0C00,
+  NEONScalar2RegMiscMask = NEON_Q | NEONScalar | NEON2RegMiscMask,
+  NEON_CMGT_zero_scalar = NEON_Q | NEONScalar | NEON_CMGT_zero,
+  NEON_CMEQ_zero_scalar = NEON_Q | NEONScalar | NEON_CMEQ_zero,
+  NEON_CMLT_zero_scalar = NEON_Q | NEONScalar | NEON_CMLT_zero,
+  NEON_CMGE_zero_scalar = NEON_Q | NEONScalar | NEON_CMGE_zero,
+  NEON_CMLE_zero_scalar = NEON_Q | NEONScalar | NEON_CMLE_zero,
+  NEON_ABS_scalar       = NEON_Q | NEONScalar | NEON_ABS,
+  NEON_SQABS_scalar     = NEON_Q | NEONScalar | NEON_SQABS,
+  NEON_NEG_scalar       = NEON_Q | NEONScalar | NEON_NEG,
+  NEON_SQNEG_scalar     = NEON_Q | NEONScalar | NEON_SQNEG,
+  NEON_SQXTN_scalar     = NEON_Q | NEONScalar | NEON_SQXTN,
+  NEON_UQXTN_scalar     = NEON_Q | NEONScalar | NEON_UQXTN,
+  NEON_SQXTUN_scalar    = NEON_Q | NEONScalar | NEON_SQXTUN,
+  NEON_SUQADD_scalar    = NEON_Q | NEONScalar | NEON_SUQADD,
+  NEON_USQADD_scalar    = NEON_Q | NEONScalar | NEON_USQADD,
+
+  NEONScalar2RegMiscOpcode = NEON2RegMiscOpcode,
+  NEON_NEG_scalar_opcode = NEON_NEG_scalar & NEONScalar2RegMiscOpcode,
+
+  NEONScalar2RegMiscFPMask  = NEONScalar2RegMiscMask | 0x00800000,
+  NEON_FRSQRTE_scalar    = NEON_Q | NEONScalar | NEON_FRSQRTE,
+  NEON_FRECPE_scalar     = NEON_Q | NEONScalar | NEON_FRECPE,
+  NEON_SCVTF_scalar      = NEON_Q | NEONScalar | NEON_SCVTF,
+  NEON_UCVTF_scalar      = NEON_Q | NEONScalar | NEON_UCVTF,
+  NEON_FCMGT_zero_scalar = NEON_Q | NEONScalar | NEON_FCMGT_zero,
+  NEON_FCMEQ_zero_scalar = NEON_Q | NEONScalar | NEON_FCMEQ_zero,
+  NEON_FCMLT_zero_scalar = NEON_Q | NEONScalar | NEON_FCMLT_zero,
+  NEON_FCMGE_zero_scalar = NEON_Q | NEONScalar | NEON_FCMGE_zero,
+  NEON_FCMLE_zero_scalar = NEON_Q | NEONScalar | NEON_FCMLE_zero,
+  NEON_FRECPX_scalar     = NEONScalar2RegMiscFixed | 0x0081F000,
+  NEON_FCVTNS_scalar     = NEON_Q | NEONScalar | NEON_FCVTNS,
+  NEON_FCVTNU_scalar     = NEON_Q | NEONScalar | NEON_FCVTNU,
+  NEON_FCVTPS_scalar     = NEON_Q | NEONScalar | NEON_FCVTPS,
+  NEON_FCVTPU_scalar     = NEON_Q | NEONScalar | NEON_FCVTPU,
+  NEON_FCVTMS_scalar     = NEON_Q | NEONScalar | NEON_FCVTMS,
+  NEON_FCVTMU_scalar     = NEON_Q | NEONScalar | NEON_FCVTMU,
+  NEON_FCVTZS_scalar     = NEON_Q | NEONScalar | NEON_FCVTZS,
+  NEON_FCVTZU_scalar     = NEON_Q | NEONScalar | NEON_FCVTZU,
+  NEON_FCVTAS_scalar     = NEON_Q | NEONScalar | NEON_FCVTAS,
+  NEON_FCVTAU_scalar     = NEON_Q | NEONScalar | NEON_FCVTAU,
+  NEON_FCVTXN_scalar     = NEON_Q | NEONScalar | NEON_FCVTXN
+};
+
+// NEON scalar instructions with three same-type operands.
+enum NEONScalar3SameOp {
+  NEONScalar3SameFixed = 0x5E200400,
+  NEONScalar3SameFMask = 0xDF200400,
+  NEONScalar3SameMask  = 0xFF20FC00,
+  NEON_ADD_scalar    = NEON_Q | NEONScalar | NEON_ADD,
+  NEON_CMEQ_scalar   = NEON_Q | NEONScalar | NEON_CMEQ,
+  NEON_CMGE_scalar   = NEON_Q | NEONScalar | NEON_CMGE,
+  NEON_CMGT_scalar   = NEON_Q | NEONScalar | NEON_CMGT,
+  NEON_CMHI_scalar   = NEON_Q | NEONScalar | NEON_CMHI,
+  NEON_CMHS_scalar   = NEON_Q | NEONScalar | NEON_CMHS,
+  NEON_CMTST_scalar  = NEON_Q | NEONScalar | NEON_CMTST,
+  NEON_SUB_scalar    = NEON_Q | NEONScalar | NEON_SUB,
+  NEON_UQADD_scalar  = NEON_Q | NEONScalar | NEON_UQADD,
+  NEON_SQADD_scalar  = NEON_Q | NEONScalar | NEON_SQADD,
+  NEON_UQSUB_scalar  = NEON_Q | NEONScalar | NEON_UQSUB,
+  NEON_SQSUB_scalar  = NEON_Q | NEONScalar | NEON_SQSUB,
+  NEON_USHL_scalar   = NEON_Q | NEONScalar | NEON_USHL,
+  NEON_SSHL_scalar   = NEON_Q | NEONScalar | NEON_SSHL,
+  NEON_UQSHL_scalar  = NEON_Q | NEONScalar | NEON_UQSHL,
+  NEON_SQSHL_scalar  = NEON_Q | NEONScalar | NEON_SQSHL,
+  NEON_URSHL_scalar  = NEON_Q | NEONScalar | NEON_URSHL,
+  NEON_SRSHL_scalar  = NEON_Q | NEONScalar | NEON_SRSHL,
+  NEON_UQRSHL_scalar = NEON_Q | NEONScalar | NEON_UQRSHL,
+  NEON_SQRSHL_scalar = NEON_Q | NEONScalar | NEON_SQRSHL,
+  NEON_SQDMULH_scalar = NEON_Q | NEONScalar | NEON_SQDMULH,
+  NEON_SQRDMULH_scalar = NEON_Q | NEONScalar | NEON_SQRDMULH,
+
+  // NEON floating point scalar instructions with three same-type operands.
+  NEONScalar3SameFPFixed = NEONScalar3SameFixed | 0x0000C000,
+  NEONScalar3SameFPFMask = NEONScalar3SameFMask | 0x0000C000,
+  NEONScalar3SameFPMask  = NEONScalar3SameMask | 0x00800000,
+  NEON_FACGE_scalar   = NEON_Q | NEONScalar | NEON_FACGE,
+  NEON_FACGT_scalar   = NEON_Q | NEONScalar | NEON_FACGT,
+  NEON_FCMEQ_scalar   = NEON_Q | NEONScalar | NEON_FCMEQ,
+  NEON_FCMGE_scalar   = NEON_Q | NEONScalar | NEON_FCMGE,
+  NEON_FCMGT_scalar   = NEON_Q | NEONScalar | NEON_FCMGT,
+  NEON_FMULX_scalar   = NEON_Q | NEONScalar | NEON_FMULX,
+  NEON_FRECPS_scalar  = NEON_Q | NEONScalar | NEON_FRECPS,
+  NEON_FRSQRTS_scalar = NEON_Q | NEONScalar | NEON_FRSQRTS,
+  NEON_FABD_scalar    = NEON_Q | NEONScalar | NEON_FABD
+};
+
+// NEON scalar instructions with three different-type operands.
+enum NEONScalar3DiffOp {
+  NEONScalar3DiffFixed = 0x5E200000,
+  NEONScalar3DiffFMask = 0xDF200C00,
+  NEONScalar3DiffMask  = NEON_Q | NEONScalar | NEON3DifferentMask,
+  NEON_SQDMLAL_scalar  = NEON_Q | NEONScalar | NEON_SQDMLAL,
+  NEON_SQDMLSL_scalar  = NEON_Q | NEONScalar | NEON_SQDMLSL,
+  NEON_SQDMULL_scalar  = NEON_Q | NEONScalar | NEON_SQDMULL
+};
+
+// NEON scalar instructions with indexed element operand.
+enum NEONScalarByIndexedElementOp {
+  NEONScalarByIndexedElementFixed = 0x5F000000,
+  NEONScalarByIndexedElementFMask = 0xDF000400,
+  NEONScalarByIndexedElementMask  = 0xFF00F400,
+  NEON_SQDMLAL_byelement_scalar  = NEON_Q | NEONScalar | NEON_SQDMLAL_byelement,
+  NEON_SQDMLSL_byelement_scalar  = NEON_Q | NEONScalar | NEON_SQDMLSL_byelement,
+  NEON_SQDMULL_byelement_scalar  = NEON_Q | NEONScalar | NEON_SQDMULL_byelement,
+  NEON_SQDMULH_byelement_scalar  = NEON_Q | NEONScalar | NEON_SQDMULH_byelement,
+  NEON_SQRDMULH_byelement_scalar
+    = NEON_Q | NEONScalar | NEON_SQRDMULH_byelement,
+
+  // Floating point instructions.
+  NEONScalarByIndexedElementFPFixed
+    = NEONScalarByIndexedElementFixed | 0x00800000,
+  NEONScalarByIndexedElementFPMask
+    = NEONScalarByIndexedElementMask | 0x00800000,
+  NEON_FMLA_byelement_scalar  = NEON_Q | NEONScalar | NEON_FMLA_byelement,
+  NEON_FMLS_byelement_scalar  = NEON_Q | NEONScalar | NEON_FMLS_byelement,
+  NEON_FMUL_byelement_scalar  = NEON_Q | NEONScalar | NEON_FMUL_byelement,
+  NEON_FMULX_byelement_scalar = NEON_Q | NEONScalar | NEON_FMULX_byelement
+};
+
+// NEON scalar register copy.
+enum NEONScalarCopyOp {
+  NEONScalarCopyFixed = 0x5E000400,
+  NEONScalarCopyFMask = 0xDFE08400,
+  NEONScalarCopyMask  = 0xFFE0FC00,
+  NEON_DUP_ELEMENT_scalar = NEON_Q | NEONScalar | NEON_DUP_ELEMENT
+};
+
+// NEON scalar pairwise instructions.
+enum NEONScalarPairwiseOp {
+  NEONScalarPairwiseFixed = 0x5E300800,
+  NEONScalarPairwiseFMask = 0xDF3E0C00,
+  NEONScalarPairwiseMask  = 0xFFB1F800,
+  NEON_ADDP_scalar    = NEONScalarPairwiseFixed | 0x0081B000,
+  NEON_FMAXNMP_scalar = NEONScalarPairwiseFixed | 0x2000C000,
+  NEON_FMINNMP_scalar = NEONScalarPairwiseFixed | 0x2080C000,
+  NEON_FADDP_scalar   = NEONScalarPairwiseFixed | 0x2000D000,
+  NEON_FMAXP_scalar   = NEONScalarPairwiseFixed | 0x2000F000,
+  NEON_FMINP_scalar   = NEONScalarPairwiseFixed | 0x2080F000
+};
+
+// NEON scalar shift immediate.
+enum NEONScalarShiftImmediateOp {
+  NEONScalarShiftImmediateFixed = 0x5F000400,
+  NEONScalarShiftImmediateFMask = 0xDF800400,
+  NEONScalarShiftImmediateMask  = 0xFF80FC00,
+  NEON_SHL_scalar  = NEON_Q | NEONScalar | NEON_SHL,
+  NEON_SLI_scalar  = NEON_Q | NEONScalar | NEON_SLI,
+  NEON_SRI_scalar  = NEON_Q | NEONScalar | NEON_SRI,
+  NEON_SSHR_scalar = NEON_Q | NEONScalar | NEON_SSHR,
+  NEON_USHR_scalar = NEON_Q | NEONScalar | NEON_USHR,
+  NEON_SRSHR_scalar = NEON_Q | NEONScalar | NEON_SRSHR,
+  NEON_URSHR_scalar = NEON_Q | NEONScalar | NEON_URSHR,
+  NEON_SSRA_scalar = NEON_Q | NEONScalar | NEON_SSRA,
+  NEON_USRA_scalar = NEON_Q | NEONScalar | NEON_USRA,
+  NEON_SRSRA_scalar = NEON_Q | NEONScalar | NEON_SRSRA,
+  NEON_URSRA_scalar = NEON_Q | NEONScalar | NEON_URSRA,
+  NEON_UQSHRN_scalar = NEON_Q | NEONScalar | NEON_UQSHRN,
+  NEON_UQRSHRN_scalar = NEON_Q | NEONScalar | NEON_UQRSHRN,
+  NEON_SQSHRN_scalar = NEON_Q | NEONScalar | NEON_SQSHRN,
+  NEON_SQRSHRN_scalar = NEON_Q | NEONScalar | NEON_SQRSHRN,
+  NEON_SQSHRUN_scalar = NEON_Q | NEONScalar | NEON_SQSHRUN,
+  NEON_SQRSHRUN_scalar = NEON_Q | NEONScalar | NEON_SQRSHRUN,
+  NEON_SQSHLU_scalar = NEON_Q | NEONScalar | NEON_SQSHLU,
+  NEON_SQSHL_imm_scalar  = NEON_Q | NEONScalar | NEON_SQSHL_imm,
+  NEON_UQSHL_imm_scalar  = NEON_Q | NEONScalar | NEON_UQSHL_imm,
+  NEON_SCVTF_imm_scalar = NEON_Q | NEONScalar | NEON_SCVTF_imm,
+  NEON_UCVTF_imm_scalar = NEON_Q | NEONScalar | NEON_UCVTF_imm,
+  NEON_FCVTZS_imm_scalar = NEON_Q | NEONScalar | NEON_FCVTZS_imm,
+  NEON_FCVTZU_imm_scalar = NEON_Q | NEONScalar | NEON_FCVTZU_imm
+};
+
 // Unimplemented and unallocated instructions. These are defined to make fixed
 // bit assertion easier.
 enum UnimplementedOp {
diff --git a/disas/libvixl/a64/cpu-a64.h b/disas/libvixl/vixl/a64/cpu-a64.h
index 59b7974a19..cdf09a6af1 100644
--- a/disas/libvixl/a64/cpu-a64.h
+++ b/disas/libvixl/vixl/a64/cpu-a64.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2014, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -27,8 +27,8 @@
 #ifndef VIXL_CPU_A64_H
 #define VIXL_CPU_A64_H
 
-#include "globals.h"
-#include "instructions-a64.h"
+#include "vixl/globals.h"
+#include "vixl/a64/instructions-a64.h"
 
 namespace vixl {
 
diff --git a/disas/libvixl/a64/decoder-a64.cc b/disas/libvixl/vixl/a64/decoder-a64.cc
index 82591ca309..5ba2d3ce04 100644
--- a/disas/libvixl/a64/decoder-a64.cc
+++ b/disas/libvixl/vixl/a64/decoder-a64.cc
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2014, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -24,9 +24,9 @@
 // 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.
 
-#include "globals.h"
-#include "utils.h"
-#include "a64/decoder-a64.h"
+#include "vixl/globals.h"
+#include "vixl/utils.h"
+#include "vixl/a64/decoder-a64.h"
 
 namespace vixl {
 
@@ -271,6 +271,11 @@ void Decoder::DecodeLoadStore(const Instruction* instr) {
               (instr->Bits(27, 24) == 0x9) ||
               (instr->Bits(27, 24) == 0xC) ||
               (instr->Bits(27, 24) == 0xD) );
+  // TODO(all): rearrange the tree to integrate this branch.
+  if ((instr->Bit(28) == 0) && (instr->Bit(29) == 0) && (instr->Bit(26) == 1)) {
+    DecodeNEONLoadStore(instr);
+    return;
+  }
 
   if (instr->Bit(24) == 0) {
     if (instr->Bit(28) == 0) {
@@ -278,7 +283,7 @@ void Decoder::DecodeLoadStore(const Instruction* instr) {
         if (instr->Bit(26) == 0) {
           VisitLoadStoreExclusive(instr);
         } else {
-          DecodeAdvSIMDLoadStore(instr);
+          VIXL_UNREACHABLE();
         }
       } else {
         if ((instr->Bits(31, 30) == 0x3) ||
@@ -483,6 +488,7 @@ void Decoder::DecodeDataProcessing(const Instruction* instr) {
         case 6: {
           if (instr->Bit(29) == 0x1) {
             VisitUnallocated(instr);
+            VIXL_FALLTHROUGH();
           } else {
             if (instr->Bit(30) == 0) {
               if ((instr->Bit(15) == 0x1) ||
@@ -556,18 +562,15 @@ void Decoder::DecodeDataProcessing(const Instruction* instr) {
 void Decoder::DecodeFP(const Instruction* instr) {
   VIXL_ASSERT((instr->Bits(27, 24) == 0xE) ||
               (instr->Bits(27, 24) == 0xF));
-
   if (instr->Bit(28) == 0) {
-    DecodeAdvSIMDDataProcessing(instr);
+    DecodeNEONVectorDataProcessing(instr);
   } else {
-    if (instr->Bit(29) == 1) {
+    if (instr->Bits(31, 30) == 0x3) {
       VisitUnallocated(instr);
+    } else if (instr->Bits(31, 30) == 0x1) {
+      DecodeNEONScalarDataProcessing(instr);
     } else {
-      if (instr->Bits(31, 30) == 0x3) {
-        VisitUnallocated(instr);
-      } else if (instr->Bits(31, 30) == 0x1) {
-        DecodeAdvSIMDDataProcessing(instr);
-      } else {
+      if (instr->Bit(29) == 0) {
         if (instr->Bit(24) == 0) {
           if (instr->Bit(21) == 0) {
             if ((instr->Bit(23) == 1) ||
@@ -674,23 +677,190 @@ void Decoder::DecodeFP(const Instruction* instr) {
             VisitFPDataProcessing3Source(instr);
           }
         }
+      } else {
+        VisitUnallocated(instr);
       }
     }
   }
 }
 
 
-void Decoder::DecodeAdvSIMDLoadStore(const Instruction* instr) {
-  // TODO: Implement Advanced SIMD load/store instruction decode.
+void Decoder::DecodeNEONLoadStore(const Instruction* instr) {
   VIXL_ASSERT(instr->Bits(29, 25) == 0x6);
-  VisitUnimplemented(instr);
+  if (instr->Bit(31) == 0) {
+    if ((instr->Bit(24) == 0) && (instr->Bit(21) == 1)) {
+      VisitUnallocated(instr);
+      return;
+    }
+
+    if (instr->Bit(23) == 0) {
+      if (instr->Bits(20, 16) == 0) {
+        if (instr->Bit(24) == 0) {
+          VisitNEONLoadStoreMultiStruct(instr);
+        } else {
+          VisitNEONLoadStoreSingleStruct(instr);
+        }
+      } else {
+        VisitUnallocated(instr);
+      }
+    } else {
+      if (instr->Bit(24) == 0) {
+        VisitNEONLoadStoreMultiStructPostIndex(instr);
+      } else {
+        VisitNEONLoadStoreSingleStructPostIndex(instr);
+      }
+    }
+  } else {
+    VisitUnallocated(instr);
+  }
+}
+
+
+void Decoder::DecodeNEONVectorDataProcessing(const Instruction* instr) {
+  VIXL_ASSERT(instr->Bits(28, 25) == 0x7);
+  if (instr->Bit(31) == 0) {
+    if (instr->Bit(24) == 0) {
+      if (instr->Bit(21) == 0) {
+        if (instr->Bit(15) == 0) {
+          if (instr->Bit(10) == 0) {
+            if (instr->Bit(29) == 0) {
+              if (instr->Bit(11) == 0) {
+                VisitNEONTable(instr);
+              } else {
+                VisitNEONPerm(instr);
+              }
+            } else {
+              VisitNEONExtract(instr);
+            }
+          } else {
+            if (instr->Bits(23, 22) == 0) {
+              VisitNEONCopy(instr);
+            } else {
+              VisitUnallocated(instr);
+            }
+          }
+        } else {
+          VisitUnallocated(instr);
+        }
+      } else {
+        if (instr->Bit(10) == 0) {
+          if (instr->Bit(11) == 0) {
+            VisitNEON3Different(instr);
+          } else {
+            if (instr->Bits(18, 17) == 0) {
+              if (instr->Bit(20) == 0) {
+                if (instr->Bit(19) == 0) {
+                  VisitNEON2RegMisc(instr);
+                } else {
+                  if (instr->Bits(30, 29) == 0x2) {
+                    VisitCryptoAES(instr);
+                  } else {
+                    VisitUnallocated(instr);
+                  }
+                }
+              } else {
+                if (instr->Bit(19) == 0) {
+                  VisitNEONAcrossLanes(instr);
+                } else {
+                  VisitUnallocated(instr);
+                }
+              }
+            } else {
+              VisitUnallocated(instr);
+            }
+          }
+        } else {
+          VisitNEON3Same(instr);
+        }
+      }
+    } else {
+      if (instr->Bit(10) == 0) {
+        VisitNEONByIndexedElement(instr);
+      } else {
+        if (instr->Bit(23) == 0) {
+          if (instr->Bits(22, 19) == 0) {
+            VisitNEONModifiedImmediate(instr);
+          } else {
+            VisitNEONShiftImmediate(instr);
+          }
+        } else {
+          VisitUnallocated(instr);
+        }
+      }
+    }
+  } else {
+    VisitUnallocated(instr);
+  }
 }
 
 
-void Decoder::DecodeAdvSIMDDataProcessing(const Instruction* instr) {
-  // TODO: Implement Advanced SIMD data processing instruction decode.
-  VIXL_ASSERT(instr->Bits(27, 25) == 0x7);
-  VisitUnimplemented(instr);
+void Decoder::DecodeNEONScalarDataProcessing(const Instruction* instr) {
+  VIXL_ASSERT(instr->Bits(28, 25) == 0xF);
+  if (instr->Bit(24) == 0) {
+    if (instr->Bit(21) == 0) {
+      if (instr->Bit(15) == 0) {
+        if (instr->Bit(10) == 0) {
+          if (instr->Bit(29) == 0) {
+            if (instr->Bit(11) == 0) {
+              VisitCrypto3RegSHA(instr);
+            } else {
+              VisitUnallocated(instr);
+            }
+          } else {
+            VisitUnallocated(instr);
+          }
+        } else {
+          if (instr->Bits(23, 22) == 0) {
+            VisitNEONScalarCopy(instr);
+          } else {
+            VisitUnallocated(instr);
+          }
+        }
+      } else {
+        VisitUnallocated(instr);
+      }
+    } else {
+      if (instr->Bit(10) == 0) {
+        if (instr->Bit(11) == 0) {
+          VisitNEONScalar3Diff(instr);
+        } else {
+          if (instr->Bits(18, 17) == 0) {
+            if (instr->Bit(20) == 0) {
+              if (instr->Bit(19) == 0) {
+                VisitNEONScalar2RegMisc(instr);
+              } else {
+                if (instr->Bit(29) == 0) {
+                  VisitCrypto2RegSHA(instr);
+                } else {
+                  VisitUnallocated(instr);
+                }
+              }
+            } else {
+              if (instr->Bit(19) == 0) {
+                VisitNEONScalarPairwise(instr);
+              } else {
+                VisitUnallocated(instr);
+              }
+            }
+          } else {
+            VisitUnallocated(instr);
+          }
+        }
+      } else {
+        VisitNEONScalar3Same(instr);
+      }
+    }
+  } else {
+    if (instr->Bit(10) == 0) {
+      VisitNEONScalarByIndexedElement(instr);
+    } else {
+      if (instr->Bit(23) == 0) {
+        VisitNEONScalarShiftImmediate(instr);
+      } else {
+        VisitUnallocated(instr);
+      }
+    }
+  }
 }
 
 
diff --git a/disas/libvixl/a64/decoder-a64.h b/disas/libvixl/vixl/a64/decoder-a64.h
index fd08d6c1f4..b3f04f68fc 100644
--- a/disas/libvixl/a64/decoder-a64.h
+++ b/disas/libvixl/vixl/a64/decoder-a64.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2014, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -29,13 +29,13 @@
 
 #include <list>
 
-#include "globals.h"
-#include "a64/instructions-a64.h"
+#include "vixl/globals.h"
+#include "vixl/a64/instructions-a64.h"
 
 
 // List macro containing all visitors needed by the decoder class.
 
-#define VISITOR_LIST(V)             \
+#define VISITOR_LIST_THAT_RETURN(V) \
   V(PCRelAddressing)                \
   V(AddSubImmediate)                \
   V(LogicalImmediate)               \
@@ -79,8 +79,39 @@
   V(FPDataProcessing3Source)        \
   V(FPIntegerConvert)               \
   V(FPFixedPointConvert)            \
-  V(Unallocated)                    \
-  V(Unimplemented)
+  V(Crypto2RegSHA)                  \
+  V(Crypto3RegSHA)                  \
+  V(CryptoAES)                      \
+  V(NEON2RegMisc)                   \
+  V(NEON3Different)                 \
+  V(NEON3Same)                      \
+  V(NEONAcrossLanes)                \
+  V(NEONByIndexedElement)           \
+  V(NEONCopy)                       \
+  V(NEONExtract)                    \
+  V(NEONLoadStoreMultiStruct)       \
+  V(NEONLoadStoreMultiStructPostIndex)  \
+  V(NEONLoadStoreSingleStruct)      \
+  V(NEONLoadStoreSingleStructPostIndex) \
+  V(NEONModifiedImmediate)          \
+  V(NEONScalar2RegMisc)             \
+  V(NEONScalar3Diff)                \
+  V(NEONScalar3Same)                \
+  V(NEONScalarByIndexedElement)     \
+  V(NEONScalarCopy)                 \
+  V(NEONScalarPairwise)             \
+  V(NEONScalarShiftImmediate)       \
+  V(NEONShiftImmediate)             \
+  V(NEONTable)                      \
+  V(NEONPerm)                       \
+
+#define VISITOR_LIST_THAT_DONT_RETURN(V)  \
+  V(Unallocated)                          \
+  V(Unimplemented)                        \
+
+#define VISITOR_LIST(V)             \
+  VISITOR_LIST_THAT_RETURN(V)       \
+  VISITOR_LIST_THAT_DONT_RETURN(V)  \
 
 namespace vixl {
 
@@ -222,12 +253,17 @@ class Decoder {
   // Decode the Advanced SIMD (NEON) load/store part of the instruction tree,
   // and call the corresponding visitors.
   // On entry, instruction bits 29:25 = 0x6.
-  void DecodeAdvSIMDLoadStore(const Instruction* instr);
+  void DecodeNEONLoadStore(const Instruction* instr);
 
-  // Decode the Advanced SIMD (NEON) data processing part of the instruction
-  // tree, and call the corresponding visitors.
-  // On entry, instruction bits 27:25 = 0x7.
-  void DecodeAdvSIMDDataProcessing(const Instruction* instr);
+  // Decode the Advanced SIMD (NEON) vector data processing part of the
+  // instruction tree, and call the corresponding visitors.
+  // On entry, instruction bits 28:25 = 0x7.
+  void DecodeNEONVectorDataProcessing(const Instruction* instr);
+
+  // Decode the Advanced SIMD (NEON) scalar data processing part of the
+  // instruction tree, and call the corresponding visitors.
+  // On entry, instruction bits 28:25 = 0xF.
+  void DecodeNEONScalarDataProcessing(const Instruction* instr);
 
  private:
   // Visitors are registered in a list.
diff --git a/disas/libvixl/vixl/a64/disasm-a64.cc b/disas/libvixl/vixl/a64/disasm-a64.cc
new file mode 100644
index 0000000000..20caba4317
--- /dev/null
+++ b/disas/libvixl/vixl/a64/disasm-a64.cc
@@ -0,0 +1,3487 @@
+// 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.
+
+#include <cstdlib>
+#include "vixl/a64/disasm-a64.h"
+
+namespace vixl {
+
+Disassembler::Disassembler() {
+  buffer_size_ = 256;
+  buffer_ = reinterpret_cast<char*>(malloc(buffer_size_));
+  buffer_pos_ = 0;
+  own_buffer_ = true;
+  code_address_offset_ = 0;
+}
+
+
+Disassembler::Disassembler(char* text_buffer, int buffer_size) {
+  buffer_size_ = buffer_size;
+  buffer_ = text_buffer;
+  buffer_pos_ = 0;
+  own_buffer_ = false;
+  code_address_offset_ = 0;
+}
+
+
+Disassembler::~Disassembler() {
+  if (own_buffer_) {
+    free(buffer_);
+  }
+}
+
+
+char* Disassembler::GetOutput() {
+  return buffer_;
+}
+
+
+void Disassembler::VisitAddSubImmediate(const Instruction* instr) {
+  bool rd_is_zr = RdIsZROrSP(instr);
+  bool stack_op = (rd_is_zr || RnIsZROrSP(instr)) &&
+                  (instr->ImmAddSub() == 0) ? true : false;
+  const char *mnemonic = "";
+  const char *form = "'Rds, 'Rns, 'IAddSub";
+  const char *form_cmp = "'Rns, 'IAddSub";
+  const char *form_mov = "'Rds, 'Rns";
+
+  switch (instr->Mask(AddSubImmediateMask)) {
+    case ADD_w_imm:
+    case ADD_x_imm: {
+      mnemonic = "add";
+      if (stack_op) {
+        mnemonic = "mov";
+        form = form_mov;
+      }
+      break;
+    }
+    case ADDS_w_imm:
+    case ADDS_x_imm: {
+      mnemonic = "adds";
+      if (rd_is_zr) {
+        mnemonic = "cmn";
+        form = form_cmp;
+      }
+      break;
+    }
+    case SUB_w_imm:
+    case SUB_x_imm: mnemonic = "sub"; break;
+    case SUBS_w_imm:
+    case SUBS_x_imm: {
+      mnemonic = "subs";
+      if (rd_is_zr) {
+        mnemonic = "cmp";
+        form = form_cmp;
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitAddSubShifted(const Instruction* instr) {
+  bool rd_is_zr = RdIsZROrSP(instr);
+  bool rn_is_zr = RnIsZROrSP(instr);
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn, 'Rm'NDP";
+  const char *form_cmp = "'Rn, 'Rm'NDP";
+  const char *form_neg = "'Rd, 'Rm'NDP";
+
+  switch (instr->Mask(AddSubShiftedMask)) {
+    case ADD_w_shift:
+    case ADD_x_shift: mnemonic = "add"; break;
+    case ADDS_w_shift:
+    case ADDS_x_shift: {
+      mnemonic = "adds";
+      if (rd_is_zr) {
+        mnemonic = "cmn";
+        form = form_cmp;
+      }
+      break;
+    }
+    case SUB_w_shift:
+    case SUB_x_shift: {
+      mnemonic = "sub";
+      if (rn_is_zr) {
+        mnemonic = "neg";
+        form = form_neg;
+      }
+      break;
+    }
+    case SUBS_w_shift:
+    case SUBS_x_shift: {
+      mnemonic = "subs";
+      if (rd_is_zr) {
+        mnemonic = "cmp";
+        form = form_cmp;
+      } else if (rn_is_zr) {
+        mnemonic = "negs";
+        form = form_neg;
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitAddSubExtended(const Instruction* instr) {
+  bool rd_is_zr = RdIsZROrSP(instr);
+  const char *mnemonic = "";
+  Extend mode = static_cast<Extend>(instr->ExtendMode());
+  const char *form = ((mode == UXTX) || (mode == SXTX)) ?
+                     "'Rds, 'Rns, 'Xm'Ext" : "'Rds, 'Rns, 'Wm'Ext";
+  const char *form_cmp = ((mode == UXTX) || (mode == SXTX)) ?
+                         "'Rns, 'Xm'Ext" : "'Rns, 'Wm'Ext";
+
+  switch (instr->Mask(AddSubExtendedMask)) {
+    case ADD_w_ext:
+    case ADD_x_ext: mnemonic = "add"; break;
+    case ADDS_w_ext:
+    case ADDS_x_ext: {
+      mnemonic = "adds";
+      if (rd_is_zr) {
+        mnemonic = "cmn";
+        form = form_cmp;
+      }
+      break;
+    }
+    case SUB_w_ext:
+    case SUB_x_ext: mnemonic = "sub"; break;
+    case SUBS_w_ext:
+    case SUBS_x_ext: {
+      mnemonic = "subs";
+      if (rd_is_zr) {
+        mnemonic = "cmp";
+        form = form_cmp;
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitAddSubWithCarry(const Instruction* instr) {
+  bool rn_is_zr = RnIsZROrSP(instr);
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn, 'Rm";
+  const char *form_neg = "'Rd, 'Rm";
+
+  switch (instr->Mask(AddSubWithCarryMask)) {
+    case ADC_w:
+    case ADC_x: mnemonic = "adc"; break;
+    case ADCS_w:
+    case ADCS_x: mnemonic = "adcs"; break;
+    case SBC_w:
+    case SBC_x: {
+      mnemonic = "sbc";
+      if (rn_is_zr) {
+        mnemonic = "ngc";
+        form = form_neg;
+      }
+      break;
+    }
+    case SBCS_w:
+    case SBCS_x: {
+      mnemonic = "sbcs";
+      if (rn_is_zr) {
+        mnemonic = "ngcs";
+        form = form_neg;
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLogicalImmediate(const Instruction* instr) {
+  bool rd_is_zr = RdIsZROrSP(instr);
+  bool rn_is_zr = RnIsZROrSP(instr);
+  const char *mnemonic = "";
+  const char *form = "'Rds, 'Rn, 'ITri";
+
+  if (instr->ImmLogical() == 0) {
+    // The immediate encoded in the instruction is not in the expected format.
+    Format(instr, "unallocated", "(LogicalImmediate)");
+    return;
+  }
+
+  switch (instr->Mask(LogicalImmediateMask)) {
+    case AND_w_imm:
+    case AND_x_imm: mnemonic = "and"; break;
+    case ORR_w_imm:
+    case ORR_x_imm: {
+      mnemonic = "orr";
+      unsigned reg_size = (instr->SixtyFourBits() == 1) ? kXRegSize
+                                                        : kWRegSize;
+      if (rn_is_zr && !IsMovzMovnImm(reg_size, instr->ImmLogical())) {
+        mnemonic = "mov";
+        form = "'Rds, 'ITri";
+      }
+      break;
+    }
+    case EOR_w_imm:
+    case EOR_x_imm: mnemonic = "eor"; break;
+    case ANDS_w_imm:
+    case ANDS_x_imm: {
+      mnemonic = "ands";
+      if (rd_is_zr) {
+        mnemonic = "tst";
+        form = "'Rn, 'ITri";
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+bool Disassembler::IsMovzMovnImm(unsigned reg_size, uint64_t value) {
+  VIXL_ASSERT((reg_size == kXRegSize) ||
+              ((reg_size == kWRegSize) && (value <= 0xffffffff)));
+
+  // Test for movz: 16 bits set at positions 0, 16, 32 or 48.
+  if (((value & UINT64_C(0xffffffffffff0000)) == 0) ||
+      ((value & UINT64_C(0xffffffff0000ffff)) == 0) ||
+      ((value & UINT64_C(0xffff0000ffffffff)) == 0) ||
+      ((value & UINT64_C(0x0000ffffffffffff)) == 0)) {
+    return true;
+  }
+
+  // Test for movn: NOT(16 bits set at positions 0, 16, 32 or 48).
+  if ((reg_size == kXRegSize) &&
+      (((~value & UINT64_C(0xffffffffffff0000)) == 0) ||
+       ((~value & UINT64_C(0xffffffff0000ffff)) == 0) ||
+       ((~value & UINT64_C(0xffff0000ffffffff)) == 0) ||
+       ((~value & UINT64_C(0x0000ffffffffffff)) == 0))) {
+    return true;
+  }
+  if ((reg_size == kWRegSize) &&
+      (((value & 0xffff0000) == 0xffff0000) ||
+       ((value & 0x0000ffff) == 0x0000ffff))) {
+    return true;
+  }
+  return false;
+}
+
+
+void Disassembler::VisitLogicalShifted(const Instruction* instr) {
+  bool rd_is_zr = RdIsZROrSP(instr);
+  bool rn_is_zr = RnIsZROrSP(instr);
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn, 'Rm'NLo";
+
+  switch (instr->Mask(LogicalShiftedMask)) {
+    case AND_w:
+    case AND_x: mnemonic = "and"; break;
+    case BIC_w:
+    case BIC_x: mnemonic = "bic"; break;
+    case EOR_w:
+    case EOR_x: mnemonic = "eor"; break;
+    case EON_w:
+    case EON_x: mnemonic = "eon"; break;
+    case BICS_w:
+    case BICS_x: mnemonic = "bics"; break;
+    case ANDS_w:
+    case ANDS_x: {
+      mnemonic = "ands";
+      if (rd_is_zr) {
+        mnemonic = "tst";
+        form = "'Rn, 'Rm'NLo";
+      }
+      break;
+    }
+    case ORR_w:
+    case ORR_x: {
+      mnemonic = "orr";
+      if (rn_is_zr && (instr->ImmDPShift() == 0) && (instr->ShiftDP() == LSL)) {
+        mnemonic = "mov";
+        form = "'Rd, 'Rm";
+      }
+      break;
+    }
+    case ORN_w:
+    case ORN_x: {
+      mnemonic = "orn";
+      if (rn_is_zr) {
+        mnemonic = "mvn";
+        form = "'Rd, 'Rm'NLo";
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitConditionalCompareRegister(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rn, 'Rm, 'INzcv, 'Cond";
+
+  switch (instr->Mask(ConditionalCompareRegisterMask)) {
+    case CCMN_w:
+    case CCMN_x: mnemonic = "ccmn"; break;
+    case CCMP_w:
+    case CCMP_x: mnemonic = "ccmp"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitConditionalCompareImmediate(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rn, 'IP, 'INzcv, 'Cond";
+
+  switch (instr->Mask(ConditionalCompareImmediateMask)) {
+    case CCMN_w_imm:
+    case CCMN_x_imm: mnemonic = "ccmn"; break;
+    case CCMP_w_imm:
+    case CCMP_x_imm: mnemonic = "ccmp"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitConditionalSelect(const Instruction* instr) {
+  bool rnm_is_zr = (RnIsZROrSP(instr) && RmIsZROrSP(instr));
+  bool rn_is_rm = (instr->Rn() == instr->Rm());
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn, 'Rm, 'Cond";
+  const char *form_test = "'Rd, 'CInv";
+  const char *form_update = "'Rd, 'Rn, 'CInv";
+
+  Condition cond = static_cast<Condition>(instr->Condition());
+  bool invertible_cond = (cond != al) && (cond != nv);
+
+  switch (instr->Mask(ConditionalSelectMask)) {
+    case CSEL_w:
+    case CSEL_x: mnemonic = "csel"; break;
+    case CSINC_w:
+    case CSINC_x: {
+      mnemonic = "csinc";
+      if (rnm_is_zr && invertible_cond) {
+        mnemonic = "cset";
+        form = form_test;
+      } else if (rn_is_rm && invertible_cond) {
+        mnemonic = "cinc";
+        form = form_update;
+      }
+      break;
+    }
+    case CSINV_w:
+    case CSINV_x: {
+      mnemonic = "csinv";
+      if (rnm_is_zr && invertible_cond) {
+        mnemonic = "csetm";
+        form = form_test;
+      } else if (rn_is_rm && invertible_cond) {
+        mnemonic = "cinv";
+        form = form_update;
+      }
+      break;
+    }
+    case CSNEG_w:
+    case CSNEG_x: {
+      mnemonic = "csneg";
+      if (rn_is_rm && invertible_cond) {
+        mnemonic = "cneg";
+        form = form_update;
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitBitfield(const Instruction* instr) {
+  unsigned s = instr->ImmS();
+  unsigned r = instr->ImmR();
+  unsigned rd_size_minus_1 =
+    ((instr->SixtyFourBits() == 1) ? kXRegSize : kWRegSize) - 1;
+  const char *mnemonic = "";
+  const char *form = "";
+  const char *form_shift_right = "'Rd, 'Rn, 'IBr";
+  const char *form_extend = "'Rd, 'Wn";
+  const char *form_bfiz = "'Rd, 'Rn, 'IBZ-r, 'IBs+1";
+  const char *form_bfx = "'Rd, 'Rn, 'IBr, 'IBs-r+1";
+  const char *form_lsl = "'Rd, 'Rn, 'IBZ-r";
+
+  switch (instr->Mask(BitfieldMask)) {
+    case SBFM_w:
+    case SBFM_x: {
+      mnemonic = "sbfx";
+      form = form_bfx;
+      if (r == 0) {
+        form = form_extend;
+        if (s == 7) {
+          mnemonic = "sxtb";
+        } else if (s == 15) {
+          mnemonic = "sxth";
+        } else if ((s == 31) && (instr->SixtyFourBits() == 1)) {
+          mnemonic = "sxtw";
+        } else {
+          form = form_bfx;
+        }
+      } else if (s == rd_size_minus_1) {
+        mnemonic = "asr";
+        form = form_shift_right;
+      } else if (s < r) {
+        mnemonic = "sbfiz";
+        form = form_bfiz;
+      }
+      break;
+    }
+    case UBFM_w:
+    case UBFM_x: {
+      mnemonic = "ubfx";
+      form = form_bfx;
+      if (r == 0) {
+        form = form_extend;
+        if (s == 7) {
+          mnemonic = "uxtb";
+        } else if (s == 15) {
+          mnemonic = "uxth";
+        } else {
+          form = form_bfx;
+        }
+      }
+      if (s == rd_size_minus_1) {
+        mnemonic = "lsr";
+        form = form_shift_right;
+      } else if (r == s + 1) {
+        mnemonic = "lsl";
+        form = form_lsl;
+      } else if (s < r) {
+        mnemonic = "ubfiz";
+        form = form_bfiz;
+      }
+      break;
+    }
+    case BFM_w:
+    case BFM_x: {
+      mnemonic = "bfxil";
+      form = form_bfx;
+      if (s < r) {
+        mnemonic = "bfi";
+        form = form_bfiz;
+      }
+    }
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitExtract(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn, 'Rm, 'IExtract";
+
+  switch (instr->Mask(ExtractMask)) {
+    case EXTR_w:
+    case EXTR_x: {
+      if (instr->Rn() == instr->Rm()) {
+        mnemonic = "ror";
+        form = "'Rd, 'Rn, 'IExtract";
+      } else {
+        mnemonic = "extr";
+      }
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitPCRelAddressing(const Instruction* instr) {
+  switch (instr->Mask(PCRelAddressingMask)) {
+    case ADR: Format(instr, "adr", "'Xd, 'AddrPCRelByte"); break;
+    case ADRP: Format(instr, "adrp", "'Xd, 'AddrPCRelPage"); break;
+    default: Format(instr, "unimplemented", "(PCRelAddressing)");
+  }
+}
+
+
+void Disassembler::VisitConditionalBranch(const Instruction* instr) {
+  switch (instr->Mask(ConditionalBranchMask)) {
+    case B_cond: Format(instr, "b.'CBrn", "'TImmCond"); break;
+    default: VIXL_UNREACHABLE();
+  }
+}
+
+
+void Disassembler::VisitUnconditionalBranchToRegister(
+    const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Xn";
+
+  switch (instr->Mask(UnconditionalBranchToRegisterMask)) {
+    case BR: mnemonic = "br"; break;
+    case BLR: mnemonic = "blr"; break;
+    case RET: {
+      mnemonic = "ret";
+      if (instr->Rn() == kLinkRegCode) {
+        form = NULL;
+      }
+      break;
+    }
+    default: form = "(UnconditionalBranchToRegister)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitUnconditionalBranch(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'TImmUncn";
+
+  switch (instr->Mask(UnconditionalBranchMask)) {
+    case B: mnemonic = "b"; break;
+    case BL: mnemonic = "bl"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitDataProcessing1Source(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Rn";
+
+  switch (instr->Mask(DataProcessing1SourceMask)) {
+    #define FORMAT(A, B)  \
+    case A##_w:           \
+    case A##_x: mnemonic = B; break;
+    FORMAT(RBIT, "rbit");
+    FORMAT(REV16, "rev16");
+    FORMAT(REV, "rev");
+    FORMAT(CLZ, "clz");
+    FORMAT(CLS, "cls");
+    #undef FORMAT
+    case REV32_x: mnemonic = "rev32"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitDataProcessing2Source(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Rd, 'Rn, 'Rm";
+  const char *form_wwx = "'Wd, 'Wn, 'Xm";
+
+  switch (instr->Mask(DataProcessing2SourceMask)) {
+    #define FORMAT(A, B)  \
+    case A##_w:           \
+    case A##_x: mnemonic = B; break;
+    FORMAT(UDIV, "udiv");
+    FORMAT(SDIV, "sdiv");
+    FORMAT(LSLV, "lsl");
+    FORMAT(LSRV, "lsr");
+    FORMAT(ASRV, "asr");
+    FORMAT(RORV, "ror");
+    #undef FORMAT
+    case CRC32B: mnemonic = "crc32b"; break;
+    case CRC32H: mnemonic = "crc32h"; break;
+    case CRC32W: mnemonic = "crc32w"; break;
+    case CRC32X: mnemonic = "crc32x"; form = form_wwx; break;
+    case CRC32CB: mnemonic = "crc32cb"; break;
+    case CRC32CH: mnemonic = "crc32ch"; break;
+    case CRC32CW: mnemonic = "crc32cw"; break;
+    case CRC32CX: mnemonic = "crc32cx"; form = form_wwx; break;
+    default: form = "(DataProcessing2Source)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitDataProcessing3Source(const Instruction* instr) {
+  bool ra_is_zr = RaIsZROrSP(instr);
+  const char *mnemonic = "";
+  const char *form = "'Xd, 'Wn, 'Wm, 'Xa";
+  const char *form_rrr = "'Rd, 'Rn, 'Rm";
+  const char *form_rrrr = "'Rd, 'Rn, 'Rm, 'Ra";
+  const char *form_xww = "'Xd, 'Wn, 'Wm";
+  const char *form_xxx = "'Xd, 'Xn, 'Xm";
+
+  switch (instr->Mask(DataProcessing3SourceMask)) {
+    case MADD_w:
+    case MADD_x: {
+      mnemonic = "madd";
+      form = form_rrrr;
+      if (ra_is_zr) {
+        mnemonic = "mul";
+        form = form_rrr;
+      }
+      break;
+    }
+    case MSUB_w:
+    case MSUB_x: {
+      mnemonic = "msub";
+      form = form_rrrr;
+      if (ra_is_zr) {
+        mnemonic = "mneg";
+        form = form_rrr;
+      }
+      break;
+    }
+    case SMADDL_x: {
+      mnemonic = "smaddl";
+      if (ra_is_zr) {
+        mnemonic = "smull";
+        form = form_xww;
+      }
+      break;
+    }
+    case SMSUBL_x: {
+      mnemonic = "smsubl";
+      if (ra_is_zr) {
+        mnemonic = "smnegl";
+        form = form_xww;
+      }
+      break;
+    }
+    case UMADDL_x: {
+      mnemonic = "umaddl";
+      if (ra_is_zr) {
+        mnemonic = "umull";
+        form = form_xww;
+      }
+      break;
+    }
+    case UMSUBL_x: {
+      mnemonic = "umsubl";
+      if (ra_is_zr) {
+        mnemonic = "umnegl";
+        form = form_xww;
+      }
+      break;
+    }
+    case SMULH_x: {
+      mnemonic = "smulh";
+      form = form_xxx;
+      break;
+    }
+    case UMULH_x: {
+      mnemonic = "umulh";
+      form = form_xxx;
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitCompareBranch(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rt, 'TImmCmpa";
+
+  switch (instr->Mask(CompareBranchMask)) {
+    case CBZ_w:
+    case CBZ_x: mnemonic = "cbz"; break;
+    case CBNZ_w:
+    case CBNZ_x: mnemonic = "cbnz"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitTestBranch(const Instruction* instr) {
+  const char *mnemonic = "";
+  // If the top bit of the immediate is clear, the tested register is
+  // disassembled as Wt, otherwise Xt. As the top bit of the immediate is
+  // encoded in bit 31 of the instruction, we can reuse the Rt form, which
+  // uses bit 31 (normally "sf") to choose the register size.
+  const char *form = "'Rt, 'IS, 'TImmTest";
+
+  switch (instr->Mask(TestBranchMask)) {
+    case TBZ: mnemonic = "tbz"; break;
+    case TBNZ: mnemonic = "tbnz"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitMoveWideImmediate(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'IMoveImm";
+
+  // Print the shift separately for movk, to make it clear which half word will
+  // be overwritten. Movn and movz print the computed immediate, which includes
+  // shift calculation.
+  switch (instr->Mask(MoveWideImmediateMask)) {
+    case MOVN_w:
+    case MOVN_x:
+      if ((instr->ImmMoveWide()) || (instr->ShiftMoveWide() == 0)) {
+        if ((instr->SixtyFourBits() == 0) && (instr->ImmMoveWide() == 0xffff)) {
+          mnemonic = "movn";
+        } else {
+          mnemonic = "mov";
+          form = "'Rd, 'IMoveNeg";
+        }
+      } else {
+        mnemonic = "movn";
+      }
+      break;
+    case MOVZ_w:
+    case MOVZ_x:
+      if ((instr->ImmMoveWide()) || (instr->ShiftMoveWide() == 0))
+        mnemonic = "mov";
+      else
+        mnemonic = "movz";
+      break;
+    case MOVK_w:
+    case MOVK_x: mnemonic = "movk"; form = "'Rd, 'IMoveLSL"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+#define LOAD_STORE_LIST(V)    \
+  V(STRB_w, "strb", "'Wt")    \
+  V(STRH_w, "strh", "'Wt")    \
+  V(STR_w, "str", "'Wt")      \
+  V(STR_x, "str", "'Xt")      \
+  V(LDRB_w, "ldrb", "'Wt")    \
+  V(LDRH_w, "ldrh", "'Wt")    \
+  V(LDR_w, "ldr", "'Wt")      \
+  V(LDR_x, "ldr", "'Xt")      \
+  V(LDRSB_x, "ldrsb", "'Xt")  \
+  V(LDRSH_x, "ldrsh", "'Xt")  \
+  V(LDRSW_x, "ldrsw", "'Xt")  \
+  V(LDRSB_w, "ldrsb", "'Wt")  \
+  V(LDRSH_w, "ldrsh", "'Wt")  \
+  V(STR_b, "str", "'Bt")      \
+  V(STR_h, "str", "'Ht")      \
+  V(STR_s, "str", "'St")      \
+  V(STR_d, "str", "'Dt")      \
+  V(LDR_b, "ldr", "'Bt")      \
+  V(LDR_h, "ldr", "'Ht")      \
+  V(LDR_s, "ldr", "'St")      \
+  V(LDR_d, "ldr", "'Dt")      \
+  V(STR_q, "str", "'Qt")      \
+  V(LDR_q, "ldr", "'Qt")
+
+void Disassembler::VisitLoadStorePreIndex(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStorePreIndex)";
+
+  switch (instr->Mask(LoadStorePreIndexMask)) {
+    #define LS_PREINDEX(A, B, C) \
+    case A##_pre: mnemonic = B; form = C ", ['Xns'ILS]!"; break;
+    LOAD_STORE_LIST(LS_PREINDEX)
+    #undef LS_PREINDEX
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStorePostIndex(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStorePostIndex)";
+
+  switch (instr->Mask(LoadStorePostIndexMask)) {
+    #define LS_POSTINDEX(A, B, C) \
+    case A##_post: mnemonic = B; form = C ", ['Xns]'ILS"; break;
+    LOAD_STORE_LIST(LS_POSTINDEX)
+    #undef LS_POSTINDEX
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStoreUnsignedOffset(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStoreUnsignedOffset)";
+
+  switch (instr->Mask(LoadStoreUnsignedOffsetMask)) {
+    #define LS_UNSIGNEDOFFSET(A, B, C) \
+    case A##_unsigned: mnemonic = B; form = C ", ['Xns'ILU]"; break;
+    LOAD_STORE_LIST(LS_UNSIGNEDOFFSET)
+    #undef LS_UNSIGNEDOFFSET
+    case PRFM_unsigned: mnemonic = "prfm"; form = "'PrefOp, ['Xns'ILU]";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStoreRegisterOffset(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStoreRegisterOffset)";
+
+  switch (instr->Mask(LoadStoreRegisterOffsetMask)) {
+    #define LS_REGISTEROFFSET(A, B, C) \
+    case A##_reg: mnemonic = B; form = C ", ['Xns, 'Offsetreg]"; break;
+    LOAD_STORE_LIST(LS_REGISTEROFFSET)
+    #undef LS_REGISTEROFFSET
+    case PRFM_reg: mnemonic = "prfm"; form = "'PrefOp, ['Xns, 'Offsetreg]";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStoreUnscaledOffset(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Wt, ['Xns'ILS]";
+  const char *form_x = "'Xt, ['Xns'ILS]";
+  const char *form_b = "'Bt, ['Xns'ILS]";
+  const char *form_h = "'Ht, ['Xns'ILS]";
+  const char *form_s = "'St, ['Xns'ILS]";
+  const char *form_d = "'Dt, ['Xns'ILS]";
+  const char *form_q = "'Qt, ['Xns'ILS]";
+  const char *form_prefetch = "'PrefOp, ['Xns'ILS]";
+
+  switch (instr->Mask(LoadStoreUnscaledOffsetMask)) {
+    case STURB_w:  mnemonic = "sturb"; break;
+    case STURH_w:  mnemonic = "sturh"; break;
+    case STUR_w:   mnemonic = "stur"; break;
+    case STUR_x:   mnemonic = "stur"; form = form_x; break;
+    case STUR_b:   mnemonic = "stur"; form = form_b; break;
+    case STUR_h:   mnemonic = "stur"; form = form_h; break;
+    case STUR_s:   mnemonic = "stur"; form = form_s; break;
+    case STUR_d:   mnemonic = "stur"; form = form_d; break;
+    case STUR_q:   mnemonic = "stur"; form = form_q; break;
+    case LDURB_w:  mnemonic = "ldurb"; break;
+    case LDURH_w:  mnemonic = "ldurh"; break;
+    case LDUR_w:   mnemonic = "ldur"; break;
+    case LDUR_x:   mnemonic = "ldur"; form = form_x; break;
+    case LDUR_b:   mnemonic = "ldur"; form = form_b; break;
+    case LDUR_h:   mnemonic = "ldur"; form = form_h; break;
+    case LDUR_s:   mnemonic = "ldur"; form = form_s; break;
+    case LDUR_d:   mnemonic = "ldur"; form = form_d; break;
+    case LDUR_q:   mnemonic = "ldur"; form = form_q; break;
+    case LDURSB_x: form = form_x; VIXL_FALLTHROUGH();
+    case LDURSB_w: mnemonic = "ldursb"; break;
+    case LDURSH_x: form = form_x; VIXL_FALLTHROUGH();
+    case LDURSH_w: mnemonic = "ldursh"; break;
+    case LDURSW_x: mnemonic = "ldursw"; form = form_x; break;
+    case PRFUM:    mnemonic = "prfum"; form = form_prefetch; break;
+    default: form = "(LoadStoreUnscaledOffset)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadLiteral(const Instruction* instr) {
+  const char *mnemonic = "ldr";
+  const char *form = "(LoadLiteral)";
+
+  switch (instr->Mask(LoadLiteralMask)) {
+    case LDR_w_lit: form = "'Wt, 'ILLiteral 'LValue"; break;
+    case LDR_x_lit: form = "'Xt, 'ILLiteral 'LValue"; break;
+    case LDR_s_lit: form = "'St, 'ILLiteral 'LValue"; break;
+    case LDR_d_lit: form = "'Dt, 'ILLiteral 'LValue"; break;
+    case LDR_q_lit: form = "'Qt, 'ILLiteral 'LValue"; break;
+    case LDRSW_x_lit: {
+      mnemonic = "ldrsw";
+      form = "'Xt, 'ILLiteral 'LValue";
+      break;
+    }
+    case PRFM_lit: {
+      mnemonic = "prfm";
+      form = "'PrefOp, 'ILLiteral 'LValue";
+      break;
+    }
+    default: mnemonic = "unimplemented";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+#define LOAD_STORE_PAIR_LIST(V)         \
+  V(STP_w, "stp", "'Wt, 'Wt2", "2")     \
+  V(LDP_w, "ldp", "'Wt, 'Wt2", "2")     \
+  V(LDPSW_x, "ldpsw", "'Xt, 'Xt2", "2") \
+  V(STP_x, "stp", "'Xt, 'Xt2", "3")     \
+  V(LDP_x, "ldp", "'Xt, 'Xt2", "3")     \
+  V(STP_s, "stp", "'St, 'St2", "2")     \
+  V(LDP_s, "ldp", "'St, 'St2", "2")     \
+  V(STP_d, "stp", "'Dt, 'Dt2", "3")     \
+  V(LDP_d, "ldp", "'Dt, 'Dt2", "3")     \
+  V(LDP_q, "ldp", "'Qt, 'Qt2", "4")     \
+  V(STP_q, "stp", "'Qt, 'Qt2", "4")
+
+void Disassembler::VisitLoadStorePairPostIndex(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStorePairPostIndex)";
+
+  switch (instr->Mask(LoadStorePairPostIndexMask)) {
+    #define LSP_POSTINDEX(A, B, C, D) \
+    case A##_post: mnemonic = B; form = C ", ['Xns]'ILP" D; break;
+    LOAD_STORE_PAIR_LIST(LSP_POSTINDEX)
+    #undef LSP_POSTINDEX
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStorePairPreIndex(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStorePairPreIndex)";
+
+  switch (instr->Mask(LoadStorePairPreIndexMask)) {
+    #define LSP_PREINDEX(A, B, C, D) \
+    case A##_pre: mnemonic = B; form = C ", ['Xns'ILP" D "]!"; break;
+    LOAD_STORE_PAIR_LIST(LSP_PREINDEX)
+    #undef LSP_PREINDEX
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStorePairOffset(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(LoadStorePairOffset)";
+
+  switch (instr->Mask(LoadStorePairOffsetMask)) {
+    #define LSP_OFFSET(A, B, C, D) \
+    case A##_off: mnemonic = B; form = C ", ['Xns'ILP" D "]"; break;
+    LOAD_STORE_PAIR_LIST(LSP_OFFSET)
+    #undef LSP_OFFSET
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStorePairNonTemporal(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form;
+
+  switch (instr->Mask(LoadStorePairNonTemporalMask)) {
+    case STNP_w: mnemonic = "stnp"; form = "'Wt, 'Wt2, ['Xns'ILP2]"; break;
+    case LDNP_w: mnemonic = "ldnp"; form = "'Wt, 'Wt2, ['Xns'ILP2]"; break;
+    case STNP_x: mnemonic = "stnp"; form = "'Xt, 'Xt2, ['Xns'ILP3]"; break;
+    case LDNP_x: mnemonic = "ldnp"; form = "'Xt, 'Xt2, ['Xns'ILP3]"; break;
+    case STNP_s: mnemonic = "stnp"; form = "'St, 'St2, ['Xns'ILP2]"; break;
+    case LDNP_s: mnemonic = "ldnp"; form = "'St, 'St2, ['Xns'ILP2]"; break;
+    case STNP_d: mnemonic = "stnp"; form = "'Dt, 'Dt2, ['Xns'ILP3]"; break;
+    case LDNP_d: mnemonic = "ldnp"; form = "'Dt, 'Dt2, ['Xns'ILP3]"; break;
+    case STNP_q: mnemonic = "stnp"; form = "'Qt, 'Qt2, ['Xns'ILP4]"; break;
+    case LDNP_q: mnemonic = "ldnp"; form = "'Qt, 'Qt2, ['Xns'ILP4]"; break;
+    default: form = "(LoadStorePairNonTemporal)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitLoadStoreExclusive(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form;
+
+  switch (instr->Mask(LoadStoreExclusiveMask)) {
+    case STXRB_w: mnemonic = "stxrb"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STXRH_w: mnemonic = "stxrh"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STXR_w: mnemonic = "stxr"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STXR_x: mnemonic = "stxr"; form = "'Ws, 'Xt, ['Xns]"; break;
+    case LDXRB_w: mnemonic = "ldxrb"; form = "'Wt, ['Xns]"; break;
+    case LDXRH_w: mnemonic = "ldxrh"; form = "'Wt, ['Xns]"; break;
+    case LDXR_w: mnemonic = "ldxr"; form = "'Wt, ['Xns]"; break;
+    case LDXR_x: mnemonic = "ldxr"; form = "'Xt, ['Xns]"; break;
+    case STXP_w: mnemonic = "stxp"; form = "'Ws, 'Wt, 'Wt2, ['Xns]"; break;
+    case STXP_x: mnemonic = "stxp"; form = "'Ws, 'Xt, 'Xt2, ['Xns]"; break;
+    case LDXP_w: mnemonic = "ldxp"; form = "'Wt, 'Wt2, ['Xns]"; break;
+    case LDXP_x: mnemonic = "ldxp"; form = "'Xt, 'Xt2, ['Xns]"; break;
+    case STLXRB_w: mnemonic = "stlxrb"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STLXRH_w: mnemonic = "stlxrh"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STLXR_w: mnemonic = "stlxr"; form = "'Ws, 'Wt, ['Xns]"; break;
+    case STLXR_x: mnemonic = "stlxr"; form = "'Ws, 'Xt, ['Xns]"; break;
+    case LDAXRB_w: mnemonic = "ldaxrb"; form = "'Wt, ['Xns]"; break;
+    case LDAXRH_w: mnemonic = "ldaxrh"; form = "'Wt, ['Xns]"; break;
+    case LDAXR_w: mnemonic = "ldaxr"; form = "'Wt, ['Xns]"; break;
+    case LDAXR_x: mnemonic = "ldaxr"; form = "'Xt, ['Xns]"; break;
+    case STLXP_w: mnemonic = "stlxp"; form = "'Ws, 'Wt, 'Wt2, ['Xns]"; break;
+    case STLXP_x: mnemonic = "stlxp"; form = "'Ws, 'Xt, 'Xt2, ['Xns]"; break;
+    case LDAXP_w: mnemonic = "ldaxp"; form = "'Wt, 'Wt2, ['Xns]"; break;
+    case LDAXP_x: mnemonic = "ldaxp"; form = "'Xt, 'Xt2, ['Xns]"; break;
+    case STLRB_w: mnemonic = "stlrb"; form = "'Wt, ['Xns]"; break;
+    case STLRH_w: mnemonic = "stlrh"; form = "'Wt, ['Xns]"; break;
+    case STLR_w: mnemonic = "stlr"; form = "'Wt, ['Xns]"; break;
+    case STLR_x: mnemonic = "stlr"; form = "'Xt, ['Xns]"; break;
+    case LDARB_w: mnemonic = "ldarb"; form = "'Wt, ['Xns]"; break;
+    case LDARH_w: mnemonic = "ldarh"; form = "'Wt, ['Xns]"; break;
+    case LDAR_w: mnemonic = "ldar"; form = "'Wt, ['Xns]"; break;
+    case LDAR_x: mnemonic = "ldar"; form = "'Xt, ['Xns]"; break;
+    default: form = "(LoadStoreExclusive)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPCompare(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Fn, 'Fm";
+  const char *form_zero = "'Fn, #0.0";
+
+  switch (instr->Mask(FPCompareMask)) {
+    case FCMP_s_zero:
+    case FCMP_d_zero: form = form_zero; VIXL_FALLTHROUGH();
+    case FCMP_s:
+    case FCMP_d: mnemonic = "fcmp"; break;
+    case FCMPE_s_zero:
+    case FCMPE_d_zero: form = form_zero; VIXL_FALLTHROUGH();
+    case FCMPE_s:
+    case FCMPE_d: mnemonic = "fcmpe"; break;
+    default: form = "(FPCompare)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPConditionalCompare(const Instruction* instr) {
+  const char *mnemonic = "unmplemented";
+  const char *form = "'Fn, 'Fm, 'INzcv, 'Cond";
+
+  switch (instr->Mask(FPConditionalCompareMask)) {
+    case FCCMP_s:
+    case FCCMP_d: mnemonic = "fccmp"; break;
+    case FCCMPE_s:
+    case FCCMPE_d: mnemonic = "fccmpe"; break;
+    default: form = "(FPConditionalCompare)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPConditionalSelect(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Fd, 'Fn, 'Fm, 'Cond";
+
+  switch (instr->Mask(FPConditionalSelectMask)) {
+    case FCSEL_s:
+    case FCSEL_d: mnemonic = "fcsel"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPDataProcessing1Source(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Fd, 'Fn";
+
+  switch (instr->Mask(FPDataProcessing1SourceMask)) {
+    #define FORMAT(A, B)  \
+    case A##_s:           \
+    case A##_d: mnemonic = B; break;
+    FORMAT(FMOV, "fmov");
+    FORMAT(FABS, "fabs");
+    FORMAT(FNEG, "fneg");
+    FORMAT(FSQRT, "fsqrt");
+    FORMAT(FRINTN, "frintn");
+    FORMAT(FRINTP, "frintp");
+    FORMAT(FRINTM, "frintm");
+    FORMAT(FRINTZ, "frintz");
+    FORMAT(FRINTA, "frinta");
+    FORMAT(FRINTX, "frintx");
+    FORMAT(FRINTI, "frinti");
+    #undef FORMAT
+    case FCVT_ds: mnemonic = "fcvt"; form = "'Dd, 'Sn"; break;
+    case FCVT_sd: mnemonic = "fcvt"; form = "'Sd, 'Dn"; break;
+    case FCVT_hs: mnemonic = "fcvt"; form = "'Hd, 'Sn"; break;
+    case FCVT_sh: mnemonic = "fcvt"; form = "'Sd, 'Hn"; break;
+    case FCVT_dh: mnemonic = "fcvt"; form = "'Dd, 'Hn"; break;
+    case FCVT_hd: mnemonic = "fcvt"; form = "'Hd, 'Dn"; break;
+    default: form = "(FPDataProcessing1Source)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPDataProcessing2Source(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Fd, 'Fn, 'Fm";
+
+  switch (instr->Mask(FPDataProcessing2SourceMask)) {
+    #define FORMAT(A, B)  \
+    case A##_s:           \
+    case A##_d: mnemonic = B; break;
+    FORMAT(FMUL, "fmul");
+    FORMAT(FDIV, "fdiv");
+    FORMAT(FADD, "fadd");
+    FORMAT(FSUB, "fsub");
+    FORMAT(FMAX, "fmax");
+    FORMAT(FMIN, "fmin");
+    FORMAT(FMAXNM, "fmaxnm");
+    FORMAT(FMINNM, "fminnm");
+    FORMAT(FNMUL, "fnmul");
+    #undef FORMAT
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPDataProcessing3Source(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Fd, 'Fn, 'Fm, 'Fa";
+
+  switch (instr->Mask(FPDataProcessing3SourceMask)) {
+    #define FORMAT(A, B)  \
+    case A##_s:           \
+    case A##_d: mnemonic = B; break;
+    FORMAT(FMADD, "fmadd");
+    FORMAT(FMSUB, "fmsub");
+    FORMAT(FNMADD, "fnmadd");
+    FORMAT(FNMSUB, "fnmsub");
+    #undef FORMAT
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPImmediate(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "(FPImmediate)";
+
+  switch (instr->Mask(FPImmediateMask)) {
+    case FMOV_s_imm: mnemonic = "fmov"; form = "'Sd, 'IFPSingle"; break;
+    case FMOV_d_imm: mnemonic = "fmov"; form = "'Dd, 'IFPDouble"; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPIntegerConvert(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(FPIntegerConvert)";
+  const char *form_rf = "'Rd, 'Fn";
+  const char *form_fr = "'Fd, 'Rn";
+
+  switch (instr->Mask(FPIntegerConvertMask)) {
+    case FMOV_ws:
+    case FMOV_xd: mnemonic = "fmov"; form = form_rf; break;
+    case FMOV_sw:
+    case FMOV_dx: mnemonic = "fmov"; form = form_fr; break;
+    case FMOV_d1_x: mnemonic = "fmov"; form = "'Vd.D[1], 'Rn"; break;
+    case FMOV_x_d1: mnemonic = "fmov"; form = "'Rd, 'Vn.D[1]"; break;
+    case FCVTAS_ws:
+    case FCVTAS_xs:
+    case FCVTAS_wd:
+    case FCVTAS_xd: mnemonic = "fcvtas"; form = form_rf; break;
+    case FCVTAU_ws:
+    case FCVTAU_xs:
+    case FCVTAU_wd:
+    case FCVTAU_xd: mnemonic = "fcvtau"; form = form_rf; break;
+    case FCVTMS_ws:
+    case FCVTMS_xs:
+    case FCVTMS_wd:
+    case FCVTMS_xd: mnemonic = "fcvtms"; form = form_rf; break;
+    case FCVTMU_ws:
+    case FCVTMU_xs:
+    case FCVTMU_wd:
+    case FCVTMU_xd: mnemonic = "fcvtmu"; form = form_rf; break;
+    case FCVTNS_ws:
+    case FCVTNS_xs:
+    case FCVTNS_wd:
+    case FCVTNS_xd: mnemonic = "fcvtns"; form = form_rf; break;
+    case FCVTNU_ws:
+    case FCVTNU_xs:
+    case FCVTNU_wd:
+    case FCVTNU_xd: mnemonic = "fcvtnu"; form = form_rf; break;
+    case FCVTZU_xd:
+    case FCVTZU_ws:
+    case FCVTZU_wd:
+    case FCVTZU_xs: mnemonic = "fcvtzu"; form = form_rf; break;
+    case FCVTZS_xd:
+    case FCVTZS_wd:
+    case FCVTZS_xs:
+    case FCVTZS_ws: mnemonic = "fcvtzs"; form = form_rf; break;
+    case FCVTPU_xd:
+    case FCVTPU_ws:
+    case FCVTPU_wd:
+    case FCVTPU_xs: mnemonic = "fcvtpu"; form = form_rf; break;
+    case FCVTPS_xd:
+    case FCVTPS_wd:
+    case FCVTPS_xs:
+    case FCVTPS_ws: mnemonic = "fcvtps"; form = form_rf; break;
+    case SCVTF_sw:
+    case SCVTF_sx:
+    case SCVTF_dw:
+    case SCVTF_dx: mnemonic = "scvtf"; form = form_fr; break;
+    case UCVTF_sw:
+    case UCVTF_sx:
+    case UCVTF_dw:
+    case UCVTF_dx: mnemonic = "ucvtf"; form = form_fr; break;
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitFPFixedPointConvert(const Instruction* instr) {
+  const char *mnemonic = "";
+  const char *form = "'Rd, 'Fn, 'IFPFBits";
+  const char *form_fr = "'Fd, 'Rn, 'IFPFBits";
+
+  switch (instr->Mask(FPFixedPointConvertMask)) {
+    case FCVTZS_ws_fixed:
+    case FCVTZS_xs_fixed:
+    case FCVTZS_wd_fixed:
+    case FCVTZS_xd_fixed: mnemonic = "fcvtzs"; break;
+    case FCVTZU_ws_fixed:
+    case FCVTZU_xs_fixed:
+    case FCVTZU_wd_fixed:
+    case FCVTZU_xd_fixed: mnemonic = "fcvtzu"; break;
+    case SCVTF_sw_fixed:
+    case SCVTF_sx_fixed:
+    case SCVTF_dw_fixed:
+    case SCVTF_dx_fixed: mnemonic = "scvtf"; form = form_fr; break;
+    case UCVTF_sw_fixed:
+    case UCVTF_sx_fixed:
+    case UCVTF_dw_fixed:
+    case UCVTF_dx_fixed: mnemonic = "ucvtf"; form = form_fr; break;
+    default: VIXL_UNREACHABLE();
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitSystem(const Instruction* instr) {
+  // Some system instructions hijack their Op and Cp fields to represent a
+  // range of immediates instead of indicating a different instruction. This
+  // makes the decoding tricky.
+  const char *mnemonic = "unimplemented";
+  const char *form = "(System)";
+
+  if (instr->Mask(SystemExclusiveMonitorFMask) == SystemExclusiveMonitorFixed) {
+    switch (instr->Mask(SystemExclusiveMonitorMask)) {
+      case CLREX: {
+        mnemonic = "clrex";
+        form = (instr->CRm() == 0xf) ? NULL : "'IX";
+        break;
+      }
+    }
+  } else if (instr->Mask(SystemSysRegFMask) == SystemSysRegFixed) {
+    switch (instr->Mask(SystemSysRegMask)) {
+      case MRS: {
+        mnemonic = "mrs";
+        switch (instr->ImmSystemRegister()) {
+          case NZCV: form = "'Xt, nzcv"; break;
+          case FPCR: form = "'Xt, fpcr"; break;
+          default: form = "'Xt, (unknown)"; break;
+        }
+        break;
+      }
+      case MSR: {
+        mnemonic = "msr";
+        switch (instr->ImmSystemRegister()) {
+          case NZCV: form = "nzcv, 'Xt"; break;
+          case FPCR: form = "fpcr, 'Xt"; break;
+          default: form = "(unknown), 'Xt"; break;
+        }
+        break;
+      }
+    }
+  } else if (instr->Mask(SystemHintFMask) == SystemHintFixed) {
+    switch (instr->ImmHint()) {
+      case NOP: {
+        mnemonic = "nop";
+        form = NULL;
+        break;
+      }
+    }
+  } else if (instr->Mask(MemBarrierFMask) == MemBarrierFixed) {
+    switch (instr->Mask(MemBarrierMask)) {
+      case DMB: {
+        mnemonic = "dmb";
+        form = "'M";
+        break;
+      }
+      case DSB: {
+        mnemonic = "dsb";
+        form = "'M";
+        break;
+      }
+      case ISB: {
+        mnemonic = "isb";
+        form = NULL;
+        break;
+      }
+    }
+  } else if (instr->Mask(SystemSysFMask) == SystemSysFixed) {
+    switch (instr->SysOp()) {
+      case IVAU:
+        mnemonic = "ic";
+        form = "ivau, 'Xt";
+        break;
+      case CVAC:
+        mnemonic = "dc";
+        form = "cvac, 'Xt";
+        break;
+      case CVAU:
+        mnemonic = "dc";
+        form = "cvau, 'Xt";
+        break;
+      case CIVAC:
+        mnemonic = "dc";
+        form = "civac, 'Xt";
+        break;
+      case ZVA:
+        mnemonic = "dc";
+        form = "zva, 'Xt";
+        break;
+      default:
+        mnemonic = "sys";
+        if (instr->Rt() == 31) {
+          form = "'G1, 'Kn, 'Km, 'G2";
+        } else {
+          form = "'G1, 'Kn, 'Km, 'G2, 'Xt";
+        }
+        break;
+      }
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitException(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'IDebug";
+
+  switch (instr->Mask(ExceptionMask)) {
+    case HLT: mnemonic = "hlt"; break;
+    case BRK: mnemonic = "brk"; break;
+    case SVC: mnemonic = "svc"; break;
+    case HVC: mnemonic = "hvc"; break;
+    case SMC: mnemonic = "smc"; break;
+    case DCPS1: mnemonic = "dcps1"; form = "{'IDebug}"; break;
+    case DCPS2: mnemonic = "dcps2"; form = "{'IDebug}"; break;
+    case DCPS3: mnemonic = "dcps3"; form = "{'IDebug}"; break;
+    default: form = "(Exception)";
+  }
+  Format(instr, mnemonic, form);
+}
+
+
+void Disassembler::VisitCrypto2RegSHA(const Instruction* instr) {
+  VisitUnimplemented(instr);
+}
+
+
+void Disassembler::VisitCrypto3RegSHA(const Instruction* instr) {
+  VisitUnimplemented(instr);
+}
+
+
+void Disassembler::VisitCryptoAES(const Instruction* instr) {
+  VisitUnimplemented(instr);
+}
+
+
+void Disassembler::VisitNEON2RegMisc(const Instruction* instr) {
+  const char *mnemonic       = "unimplemented";
+  const char *form           = "'Vd.%s, 'Vn.%s";
+  const char *form_cmp_zero  = "'Vd.%s, 'Vn.%s, #0";
+  const char *form_fcmp_zero = "'Vd.%s, 'Vn.%s, #0.0";
+  NEONFormatDecoder nfd(instr);
+
+  static const NEONFormatMap map_lp_ta = {
+    {23, 22, 30}, {NF_4H, NF_8H, NF_2S, NF_4S, NF_1D, NF_2D}
+  };
+
+  static const NEONFormatMap map_cvt_ta = {
+    {22}, {NF_4S, NF_2D}
+  };
+
+  static const NEONFormatMap map_cvt_tb = {
+    {22, 30}, {NF_4H, NF_8H, NF_2S, NF_4S}
+  };
+
+  if (instr->Mask(NEON2RegMiscOpcode) <= NEON_NEG_opcode) {
+    // These instructions all use a two bit size field, except NOT and RBIT,
+    // which use the field to encode the operation.
+    switch (instr->Mask(NEON2RegMiscMask)) {
+      case NEON_REV64:     mnemonic = "rev64"; break;
+      case NEON_REV32:     mnemonic = "rev32"; break;
+      case NEON_REV16:     mnemonic = "rev16"; break;
+      case NEON_SADDLP:
+        mnemonic = "saddlp";
+        nfd.SetFormatMap(0, &map_lp_ta);
+        break;
+      case NEON_UADDLP:
+        mnemonic = "uaddlp";
+        nfd.SetFormatMap(0, &map_lp_ta);
+        break;
+      case NEON_SUQADD:    mnemonic = "suqadd"; break;
+      case NEON_USQADD:    mnemonic = "usqadd"; break;
+      case NEON_CLS:       mnemonic = "cls"; break;
+      case NEON_CLZ:       mnemonic = "clz"; break;
+      case NEON_CNT:       mnemonic = "cnt"; break;
+      case NEON_SADALP:
+        mnemonic = "sadalp";
+        nfd.SetFormatMap(0, &map_lp_ta);
+        break;
+      case NEON_UADALP:
+        mnemonic = "uadalp";
+        nfd.SetFormatMap(0, &map_lp_ta);
+        break;
+      case NEON_SQABS:     mnemonic = "sqabs"; break;
+      case NEON_SQNEG:     mnemonic = "sqneg"; break;
+      case NEON_CMGT_zero: mnemonic = "cmgt"; form = form_cmp_zero; break;
+      case NEON_CMGE_zero: mnemonic = "cmge"; form = form_cmp_zero; break;
+      case NEON_CMEQ_zero: mnemonic = "cmeq"; form = form_cmp_zero; break;
+      case NEON_CMLE_zero: mnemonic = "cmle"; form = form_cmp_zero; break;
+      case NEON_CMLT_zero: mnemonic = "cmlt"; form = form_cmp_zero; break;
+      case NEON_ABS:       mnemonic = "abs"; break;
+      case NEON_NEG:       mnemonic = "neg"; break;
+      case NEON_RBIT_NOT:
+        switch (instr->FPType()) {
+          case 0: mnemonic = "mvn"; break;
+          case 1: mnemonic = "rbit"; break;
+          default: form = "(NEON2RegMisc)";
+        }
+        nfd.SetFormatMaps(nfd.LogicalFormatMap());
+        break;
+    }
+  } else {
+    // These instructions all use a one bit size field, except XTN, SQXTUN,
+    // SHLL, SQXTN and UQXTN, which use a two bit size field.
+    nfd.SetFormatMaps(nfd.FPFormatMap());
+    switch (instr->Mask(NEON2RegMiscFPMask)) {
+      case NEON_FABS:   mnemonic = "fabs"; break;
+      case NEON_FNEG:   mnemonic = "fneg"; break;
+      case NEON_FCVTN:
+        mnemonic = instr->Mask(NEON_Q) ? "fcvtn2" : "fcvtn";
+        nfd.SetFormatMap(0, &map_cvt_tb);
+        nfd.SetFormatMap(1, &map_cvt_ta);
+        break;
+      case NEON_FCVTXN:
+        mnemonic = instr->Mask(NEON_Q) ? "fcvtxn2" : "fcvtxn";
+        nfd.SetFormatMap(0, &map_cvt_tb);
+        nfd.SetFormatMap(1, &map_cvt_ta);
+        break;
+      case NEON_FCVTL:
+        mnemonic = instr->Mask(NEON_Q) ? "fcvtl2" : "fcvtl";
+        nfd.SetFormatMap(0, &map_cvt_ta);
+        nfd.SetFormatMap(1, &map_cvt_tb);
+        break;
+      case NEON_FRINTN: mnemonic = "frintn"; break;
+      case NEON_FRINTA: mnemonic = "frinta"; break;
+      case NEON_FRINTP: mnemonic = "frintp"; break;
+      case NEON_FRINTM: mnemonic = "frintm"; break;
+      case NEON_FRINTX: mnemonic = "frintx"; break;
+      case NEON_FRINTZ: mnemonic = "frintz"; break;
+      case NEON_FRINTI: mnemonic = "frinti"; break;
+      case NEON_FCVTNS: mnemonic = "fcvtns"; break;
+      case NEON_FCVTNU: mnemonic = "fcvtnu"; break;
+      case NEON_FCVTPS: mnemonic = "fcvtps"; break;
+      case NEON_FCVTPU: mnemonic = "fcvtpu"; break;
+      case NEON_FCVTMS: mnemonic = "fcvtms"; break;
+      case NEON_FCVTMU: mnemonic = "fcvtmu"; break;
+      case NEON_FCVTZS: mnemonic = "fcvtzs"; break;
+      case NEON_FCVTZU: mnemonic = "fcvtzu"; break;
+      case NEON_FCVTAS: mnemonic = "fcvtas"; break;
+      case NEON_FCVTAU: mnemonic = "fcvtau"; break;
+      case NEON_FSQRT:  mnemonic = "fsqrt"; break;
+      case NEON_SCVTF:  mnemonic = "scvtf"; break;
+      case NEON_UCVTF:  mnemonic = "ucvtf"; break;
+      case NEON_URSQRTE: mnemonic = "ursqrte"; break;
+      case NEON_URECPE:  mnemonic = "urecpe";  break;
+      case NEON_FRSQRTE: mnemonic = "frsqrte"; break;
+      case NEON_FRECPE:  mnemonic = "frecpe";  break;
+      case NEON_FCMGT_zero: mnemonic = "fcmgt"; form = form_fcmp_zero; break;
+      case NEON_FCMGE_zero: mnemonic = "fcmge"; form = form_fcmp_zero; break;
+      case NEON_FCMEQ_zero: mnemonic = "fcmeq"; form = form_fcmp_zero; break;
+      case NEON_FCMLE_zero: mnemonic = "fcmle"; form = form_fcmp_zero; break;
+      case NEON_FCMLT_zero: mnemonic = "fcmlt"; form = form_fcmp_zero; break;
+      default:
+        if ((NEON_XTN_opcode <= instr->Mask(NEON2RegMiscOpcode)) &&
+            (instr->Mask(NEON2RegMiscOpcode) <= NEON_UQXTN_opcode)) {
+          nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+          nfd.SetFormatMap(1, nfd.LongIntegerFormatMap());
+
+          switch (instr->Mask(NEON2RegMiscMask)) {
+            case NEON_XTN:    mnemonic = "xtn"; break;
+            case NEON_SQXTN:  mnemonic = "sqxtn"; break;
+            case NEON_UQXTN:  mnemonic = "uqxtn"; break;
+            case NEON_SQXTUN: mnemonic = "sqxtun"; break;
+            case NEON_SHLL:
+              mnemonic = "shll";
+              nfd.SetFormatMap(0, nfd.LongIntegerFormatMap());
+              nfd.SetFormatMap(1, nfd.IntegerFormatMap());
+              switch (instr->NEONSize()) {
+                case 0: form = "'Vd.%s, 'Vn.%s, #8"; break;
+                case 1: form = "'Vd.%s, 'Vn.%s, #16"; break;
+                case 2: form = "'Vd.%s, 'Vn.%s, #32"; break;
+                default: form = "(NEON2RegMisc)";
+              }
+          }
+          Format(instr, nfd.Mnemonic(mnemonic), nfd.Substitute(form));
+          return;
+        } else {
+          form = "(NEON2RegMisc)";
+        }
+    }
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEON3Same(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Vd.%s, 'Vn.%s, 'Vm.%s";
+  NEONFormatDecoder nfd(instr);
+
+  if (instr->Mask(NEON3SameLogicalFMask) == NEON3SameLogicalFixed) {
+    switch (instr->Mask(NEON3SameLogicalMask)) {
+      case NEON_AND: mnemonic = "and"; break;
+      case NEON_ORR:
+        mnemonic = "orr";
+        if (instr->Rm() == instr->Rn()) {
+          mnemonic = "mov";
+          form = "'Vd.%s, 'Vn.%s";
+        }
+        break;
+      case NEON_ORN: mnemonic = "orn"; break;
+      case NEON_EOR: mnemonic = "eor"; break;
+      case NEON_BIC: mnemonic = "bic"; break;
+      case NEON_BIF: mnemonic = "bif"; break;
+      case NEON_BIT: mnemonic = "bit"; break;
+      case NEON_BSL: mnemonic = "bsl"; break;
+      default: form = "(NEON3Same)";
+    }
+    nfd.SetFormatMaps(nfd.LogicalFormatMap());
+  } else {
+    static const char *mnemonics[] = {
+        "shadd", "uhadd", "shadd", "uhadd",
+        "sqadd", "uqadd", "sqadd", "uqadd",
+        "srhadd", "urhadd", "srhadd", "urhadd",
+        NULL, NULL, NULL, NULL,  // Handled by logical cases above.
+        "shsub", "uhsub", "shsub", "uhsub",
+        "sqsub", "uqsub", "sqsub", "uqsub",
+        "cmgt", "cmhi", "cmgt", "cmhi",
+        "cmge", "cmhs", "cmge", "cmhs",
+        "sshl", "ushl", "sshl", "ushl",
+        "sqshl", "uqshl", "sqshl", "uqshl",
+        "srshl", "urshl", "srshl", "urshl",
+        "sqrshl", "uqrshl", "sqrshl", "uqrshl",
+        "smax", "umax", "smax", "umax",
+        "smin", "umin", "smin", "umin",
+        "sabd", "uabd", "sabd", "uabd",
+        "saba", "uaba", "saba", "uaba",
+        "add", "sub", "add", "sub",
+        "cmtst", "cmeq", "cmtst", "cmeq",
+        "mla", "mls", "mla", "mls",
+        "mul", "pmul", "mul", "pmul",
+        "smaxp", "umaxp", "smaxp", "umaxp",
+        "sminp", "uminp", "sminp", "uminp",
+        "sqdmulh", "sqrdmulh", "sqdmulh", "sqrdmulh",
+        "addp", "unallocated", "addp", "unallocated",
+        "fmaxnm", "fmaxnmp", "fminnm", "fminnmp",
+        "fmla", "unallocated", "fmls", "unallocated",
+        "fadd", "faddp", "fsub", "fabd",
+        "fmulx", "fmul", "unallocated", "unallocated",
+        "fcmeq", "fcmge", "unallocated", "fcmgt",
+        "unallocated", "facge", "unallocated", "facgt",
+        "fmax", "fmaxp", "fmin", "fminp",
+        "frecps", "fdiv", "frsqrts", "unallocated"};
+
+    // Operation is determined by the opcode bits (15-11), the top bit of
+    // size (23) and the U bit (29).
+    unsigned index = (instr->Bits(15, 11) << 2) | (instr->Bit(23) << 1) |
+                     instr->Bit(29);
+    VIXL_ASSERT(index < (sizeof(mnemonics) / sizeof(mnemonics[0])));
+    mnemonic = mnemonics[index];
+    // Assert that index is not one of the previously handled logical
+    // instructions.
+    VIXL_ASSERT(mnemonic != NULL);
+
+    if (instr->Mask(NEON3SameFPFMask) == NEON3SameFPFixed) {
+      nfd.SetFormatMaps(nfd.FPFormatMap());
+    }
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEON3Different(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Vd.%s, 'Vn.%s, 'Vm.%s";
+
+  NEONFormatDecoder nfd(instr);
+  nfd.SetFormatMap(0, nfd.LongIntegerFormatMap());
+
+  // Ignore the Q bit. Appending a "2" suffix is handled later.
+  switch (instr->Mask(NEON3DifferentMask) & ~NEON_Q) {
+    case NEON_PMULL:    mnemonic = "pmull";   break;
+    case NEON_SABAL:    mnemonic = "sabal";   break;
+    case NEON_SABDL:    mnemonic = "sabdl";   break;
+    case NEON_SADDL:    mnemonic = "saddl";   break;
+    case NEON_SMLAL:    mnemonic = "smlal";   break;
+    case NEON_SMLSL:    mnemonic = "smlsl";   break;
+    case NEON_SMULL:    mnemonic = "smull";   break;
+    case NEON_SSUBL:    mnemonic = "ssubl";   break;
+    case NEON_SQDMLAL:  mnemonic = "sqdmlal"; break;
+    case NEON_SQDMLSL:  mnemonic = "sqdmlsl"; break;
+    case NEON_SQDMULL:  mnemonic = "sqdmull"; break;
+    case NEON_UABAL:    mnemonic = "uabal";   break;
+    case NEON_UABDL:    mnemonic = "uabdl";   break;
+    case NEON_UADDL:    mnemonic = "uaddl";   break;
+    case NEON_UMLAL:    mnemonic = "umlal";   break;
+    case NEON_UMLSL:    mnemonic = "umlsl";   break;
+    case NEON_UMULL:    mnemonic = "umull";   break;
+    case NEON_USUBL:    mnemonic = "usubl";   break;
+    case NEON_SADDW:
+      mnemonic = "saddw";
+      nfd.SetFormatMap(1, nfd.LongIntegerFormatMap());
+      break;
+    case NEON_SSUBW:
+      mnemonic = "ssubw";
+      nfd.SetFormatMap(1, nfd.LongIntegerFormatMap());
+      break;
+    case NEON_UADDW:
+      mnemonic = "uaddw";
+      nfd.SetFormatMap(1, nfd.LongIntegerFormatMap());
+      break;
+    case NEON_USUBW:
+      mnemonic = "usubw";
+      nfd.SetFormatMap(1, nfd.LongIntegerFormatMap());
+      break;
+    case NEON_ADDHN:
+      mnemonic = "addhn";
+      nfd.SetFormatMaps(nfd.LongIntegerFormatMap());
+      nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+      break;
+    case NEON_RADDHN:
+      mnemonic = "raddhn";
+      nfd.SetFormatMaps(nfd.LongIntegerFormatMap());
+      nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+      break;
+    case NEON_RSUBHN:
+      mnemonic = "rsubhn";
+      nfd.SetFormatMaps(nfd.LongIntegerFormatMap());
+      nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+      break;
+    case NEON_SUBHN:
+      mnemonic = "subhn";
+      nfd.SetFormatMaps(nfd.LongIntegerFormatMap());
+      nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+      break;
+    default: form = "(NEON3Different)";
+  }
+  Format(instr, nfd.Mnemonic(mnemonic), nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONAcrossLanes(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "%sd, 'Vn.%s";
+
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::ScalarFormatMap(),
+                               NEONFormatDecoder::IntegerFormatMap());
+
+  if (instr->Mask(NEONAcrossLanesFPFMask) == NEONAcrossLanesFPFixed) {
+    nfd.SetFormatMap(0, nfd.FPScalarFormatMap());
+    nfd.SetFormatMap(1, nfd.FPFormatMap());
+    switch (instr->Mask(NEONAcrossLanesFPMask)) {
+      case NEON_FMAXV: mnemonic = "fmaxv"; break;
+      case NEON_FMINV: mnemonic = "fminv"; break;
+      case NEON_FMAXNMV: mnemonic = "fmaxnmv"; break;
+      case NEON_FMINNMV: mnemonic = "fminnmv"; break;
+      default: form = "(NEONAcrossLanes)"; break;
+    }
+  } else if (instr->Mask(NEONAcrossLanesFMask) == NEONAcrossLanesFixed) {
+    switch (instr->Mask(NEONAcrossLanesMask)) {
+      case NEON_ADDV:  mnemonic = "addv"; break;
+      case NEON_SMAXV: mnemonic = "smaxv"; break;
+      case NEON_SMINV: mnemonic = "sminv"; break;
+      case NEON_UMAXV: mnemonic = "umaxv"; break;
+      case NEON_UMINV: mnemonic = "uminv"; break;
+      case NEON_SADDLV:
+        mnemonic = "saddlv";
+        nfd.SetFormatMap(0, nfd.LongScalarFormatMap());
+        break;
+      case NEON_UADDLV:
+        mnemonic = "uaddlv";
+        nfd.SetFormatMap(0, nfd.LongScalarFormatMap());
+        break;
+      default: form = "(NEONAcrossLanes)"; break;
+    }
+  }
+  Format(instr, mnemonic, nfd.Substitute(form,
+      NEONFormatDecoder::kPlaceholder, NEONFormatDecoder::kFormat));
+}
+
+
+void Disassembler::VisitNEONByIndexedElement(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  bool l_instr = false;
+  bool fp_instr = false;
+
+  const char *form = "'Vd.%s, 'Vn.%s, 'Ve.%s['IVByElemIndex]";
+
+  static const NEONFormatMap map_ta = {
+    {23, 22}, {NF_UNDEF, NF_4S, NF_2D}
+  };
+  NEONFormatDecoder nfd(instr, &map_ta,
+                        NEONFormatDecoder::IntegerFormatMap(),
+                        NEONFormatDecoder::ScalarFormatMap());
+
+  switch (instr->Mask(NEONByIndexedElementMask)) {
+    case NEON_SMULL_byelement:    mnemonic = "smull"; l_instr = true; break;
+    case NEON_UMULL_byelement:    mnemonic = "umull"; l_instr = true; break;
+    case NEON_SMLAL_byelement:    mnemonic = "smlal"; l_instr = true; break;
+    case NEON_UMLAL_byelement:    mnemonic = "umlal"; l_instr = true; break;
+    case NEON_SMLSL_byelement:    mnemonic = "smlsl"; l_instr = true; break;
+    case NEON_UMLSL_byelement:    mnemonic = "umlsl"; l_instr = true; break;
+    case NEON_SQDMULL_byelement:  mnemonic = "sqdmull"; l_instr = true; break;
+    case NEON_SQDMLAL_byelement:  mnemonic = "sqdmlal"; l_instr = true; break;
+    case NEON_SQDMLSL_byelement:  mnemonic = "sqdmlsl"; l_instr = true; break;
+    case NEON_MUL_byelement:      mnemonic = "mul"; break;
+    case NEON_MLA_byelement:      mnemonic = "mla"; break;
+    case NEON_MLS_byelement:      mnemonic = "mls"; break;
+    case NEON_SQDMULH_byelement:  mnemonic = "sqdmulh";  break;
+    case NEON_SQRDMULH_byelement: mnemonic = "sqrdmulh"; break;
+    default:
+      switch (instr->Mask(NEONByIndexedElementFPMask)) {
+        case NEON_FMUL_byelement:  mnemonic = "fmul";  fp_instr = true; break;
+        case NEON_FMLA_byelement:  mnemonic = "fmla";  fp_instr = true; break;
+        case NEON_FMLS_byelement:  mnemonic = "fmls";  fp_instr = true; break;
+        case NEON_FMULX_byelement: mnemonic = "fmulx"; fp_instr = true; break;
+      }
+  }
+
+  if (l_instr) {
+    Format(instr, nfd.Mnemonic(mnemonic), nfd.Substitute(form));
+  } else if (fp_instr) {
+    nfd.SetFormatMap(0, nfd.FPFormatMap());
+    Format(instr, mnemonic, nfd.Substitute(form));
+  } else {
+    nfd.SetFormatMap(0, nfd.IntegerFormatMap());
+    Format(instr, mnemonic, nfd.Substitute(form));
+  }
+}
+
+
+void Disassembler::VisitNEONCopy(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONCopy)";
+
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::TriangularFormatMap(),
+                               NEONFormatDecoder::TriangularScalarFormatMap());
+
+  if (instr->Mask(NEONCopyInsElementMask) == NEON_INS_ELEMENT) {
+    mnemonic = "mov";
+    nfd.SetFormatMap(0, nfd.TriangularScalarFormatMap());
+    form = "'Vd.%s['IVInsIndex1], 'Vn.%s['IVInsIndex2]";
+  } else if (instr->Mask(NEONCopyInsGeneralMask) == NEON_INS_GENERAL) {
+    mnemonic = "mov";
+    nfd.SetFormatMap(0, nfd.TriangularScalarFormatMap());
+    if (nfd.GetVectorFormat() == kFormatD) {
+      form = "'Vd.%s['IVInsIndex1], 'Xn";
+    } else {
+      form = "'Vd.%s['IVInsIndex1], 'Wn";
+    }
+  } else if (instr->Mask(NEONCopyUmovMask) == NEON_UMOV) {
+    if (instr->Mask(NEON_Q) || ((instr->ImmNEON5() & 7) == 4)) {
+      mnemonic = "mov";
+    } else {
+      mnemonic = "umov";
+    }
+    nfd.SetFormatMap(0, nfd.TriangularScalarFormatMap());
+    if (nfd.GetVectorFormat() == kFormatD) {
+      form = "'Xd, 'Vn.%s['IVInsIndex1]";
+    } else {
+      form = "'Wd, 'Vn.%s['IVInsIndex1]";
+    }
+  } else if (instr->Mask(NEONCopySmovMask) == NEON_SMOV) {
+    mnemonic = "smov";
+    nfd.SetFormatMap(0, nfd.TriangularScalarFormatMap());
+    form = "'Rdq, 'Vn.%s['IVInsIndex1]";
+  } else if (instr->Mask(NEONCopyDupElementMask) == NEON_DUP_ELEMENT) {
+    mnemonic = "dup";
+    form = "'Vd.%s, 'Vn.%s['IVInsIndex1]";
+  } else if (instr->Mask(NEONCopyDupGeneralMask) == NEON_DUP_GENERAL) {
+    mnemonic = "dup";
+    if (nfd.GetVectorFormat() == kFormat2D) {
+      form = "'Vd.%s, 'Xn";
+    } else {
+      form = "'Vd.%s, 'Wn";
+    }
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONExtract(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONExtract)";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LogicalFormatMap());
+  if (instr->Mask(NEONExtractMask) == NEON_EXT) {
+    mnemonic = "ext";
+    form = "'Vd.%s, 'Vn.%s, 'Vm.%s, 'IVExtract";
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONLoadStoreMultiStruct(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONLoadStoreMultiStruct)";
+  const char *form_1v = "{'Vt.%1$s}, ['Xns]";
+  const char *form_2v = "{'Vt.%1$s, 'Vt2.%1$s}, ['Xns]";
+  const char *form_3v = "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s}, ['Xns]";
+  const char *form_4v = "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s, 'Vt4.%1$s}, ['Xns]";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LoadStoreFormatMap());
+
+  switch (instr->Mask(NEONLoadStoreMultiStructMask)) {
+    case NEON_LD1_1v: mnemonic = "ld1"; form = form_1v; break;
+    case NEON_LD1_2v: mnemonic = "ld1"; form = form_2v; break;
+    case NEON_LD1_3v: mnemonic = "ld1"; form = form_3v; break;
+    case NEON_LD1_4v: mnemonic = "ld1"; form = form_4v; break;
+    case NEON_LD2: mnemonic = "ld2"; form = form_2v; break;
+    case NEON_LD3: mnemonic = "ld3"; form = form_3v; break;
+    case NEON_LD4: mnemonic = "ld4"; form = form_4v; break;
+    case NEON_ST1_1v: mnemonic = "st1"; form = form_1v; break;
+    case NEON_ST1_2v: mnemonic = "st1"; form = form_2v; break;
+    case NEON_ST1_3v: mnemonic = "st1"; form = form_3v; break;
+    case NEON_ST1_4v: mnemonic = "st1"; form = form_4v; break;
+    case NEON_ST2: mnemonic = "st2"; form = form_2v; break;
+    case NEON_ST3: mnemonic = "st3"; form = form_3v; break;
+    case NEON_ST4: mnemonic = "st4"; form = form_4v; break;
+    default: break;
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONLoadStoreMultiStructPostIndex(
+    const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONLoadStoreMultiStructPostIndex)";
+  const char *form_1v = "{'Vt.%1$s}, ['Xns], 'Xmr1";
+  const char *form_2v = "{'Vt.%1$s, 'Vt2.%1$s}, ['Xns], 'Xmr2";
+  const char *form_3v = "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s}, ['Xns], 'Xmr3";
+  const char *form_4v =
+      "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s, 'Vt4.%1$s}, ['Xns], 'Xmr4";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LoadStoreFormatMap());
+
+  switch (instr->Mask(NEONLoadStoreMultiStructPostIndexMask)) {
+    case NEON_LD1_1v_post: mnemonic = "ld1"; form = form_1v; break;
+    case NEON_LD1_2v_post: mnemonic = "ld1"; form = form_2v; break;
+    case NEON_LD1_3v_post: mnemonic = "ld1"; form = form_3v; break;
+    case NEON_LD1_4v_post: mnemonic = "ld1"; form = form_4v; break;
+    case NEON_LD2_post: mnemonic = "ld2"; form = form_2v; break;
+    case NEON_LD3_post: mnemonic = "ld3"; form = form_3v; break;
+    case NEON_LD4_post: mnemonic = "ld4"; form = form_4v; break;
+    case NEON_ST1_1v_post: mnemonic = "st1"; form = form_1v; break;
+    case NEON_ST1_2v_post: mnemonic = "st1"; form = form_2v; break;
+    case NEON_ST1_3v_post: mnemonic = "st1"; form = form_3v; break;
+    case NEON_ST1_4v_post: mnemonic = "st1"; form = form_4v; break;
+    case NEON_ST2_post: mnemonic = "st2"; form = form_2v; break;
+    case NEON_ST3_post: mnemonic = "st3"; form = form_3v; break;
+    case NEON_ST4_post: mnemonic = "st4"; form = form_4v; break;
+    default: break;
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONLoadStoreSingleStruct(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONLoadStoreSingleStruct)";
+
+  const char *form_1b = "{'Vt.b}['IVLSLane0], ['Xns]";
+  const char *form_1h = "{'Vt.h}['IVLSLane1], ['Xns]";
+  const char *form_1s = "{'Vt.s}['IVLSLane2], ['Xns]";
+  const char *form_1d = "{'Vt.d}['IVLSLane3], ['Xns]";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LoadStoreFormatMap());
+
+  switch (instr->Mask(NEONLoadStoreSingleStructMask)) {
+    case NEON_LD1_b: mnemonic = "ld1"; form = form_1b; break;
+    case NEON_LD1_h: mnemonic = "ld1"; form = form_1h; break;
+    case NEON_LD1_s:
+      mnemonic = "ld1";
+      VIXL_STATIC_ASSERT((NEON_LD1_s | (1 << NEONLSSize_offset)) == NEON_LD1_d);
+      form = ((instr->NEONLSSize() & 1) == 0) ? form_1s : form_1d;
+      break;
+    case NEON_ST1_b: mnemonic = "st1"; form = form_1b; break;
+    case NEON_ST1_h: mnemonic = "st1"; form = form_1h; break;
+    case NEON_ST1_s:
+      mnemonic = "st1";
+      VIXL_STATIC_ASSERT((NEON_ST1_s | (1 << NEONLSSize_offset)) == NEON_ST1_d);
+      form = ((instr->NEONLSSize() & 1) == 0) ? form_1s : form_1d;
+      break;
+    case NEON_LD1R:
+      mnemonic = "ld1r";
+      form = "{'Vt.%s}, ['Xns]";
+      break;
+    case NEON_LD2_b:
+    case NEON_ST2_b:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      form = "{'Vt.b, 'Vt2.b}['IVLSLane0], ['Xns]";
+      break;
+    case NEON_LD2_h:
+    case NEON_ST2_h:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      form = "{'Vt.h, 'Vt2.h}['IVLSLane1], ['Xns]";
+      break;
+    case NEON_LD2_s:
+    case NEON_ST2_s:
+      VIXL_STATIC_ASSERT((NEON_ST2_s | (1 << NEONLSSize_offset)) == NEON_ST2_d);
+      VIXL_STATIC_ASSERT((NEON_LD2_s | (1 << NEONLSSize_offset)) == NEON_LD2_d);
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s}['IVLSLane2], ['Xns]";
+      else
+        form = "{'Vt.d, 'Vt2.d}['IVLSLane3], ['Xns]";
+      break;
+    case NEON_LD2R:
+      mnemonic = "ld2r";
+      form = "{'Vt.%s, 'Vt2.%s}, ['Xns]";
+      break;
+    case NEON_LD3_b:
+    case NEON_ST3_b:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      form = "{'Vt.b, 'Vt2.b, 'Vt3.b}['IVLSLane0], ['Xns]";
+      break;
+    case NEON_LD3_h:
+    case NEON_ST3_h:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      form = "{'Vt.h, 'Vt2.h, 'Vt3.h}['IVLSLane1], ['Xns]";
+      break;
+    case NEON_LD3_s:
+    case NEON_ST3_s:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s, 'Vt3.s}['IVLSLane2], ['Xns]";
+      else
+        form = "{'Vt.d, 'Vt2.d, 'Vt3.d}['IVLSLane3], ['Xns]";
+      break;
+    case NEON_LD3R:
+      mnemonic = "ld3r";
+      form = "{'Vt.%s, 'Vt2.%s, 'Vt3.%s}, ['Xns]";
+      break;
+    case NEON_LD4_b:
+     case NEON_ST4_b:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld4" : "st4";
+      form = "{'Vt.b, 'Vt2.b, 'Vt3.b, 'Vt4.b}['IVLSLane0], ['Xns]";
+      break;
+    case NEON_LD4_h:
+    case NEON_ST4_h:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld4" : "st4";
+      form = "{'Vt.h, 'Vt2.h, 'Vt3.h, 'Vt4.h}['IVLSLane1], ['Xns]";
+      break;
+    case NEON_LD4_s:
+    case NEON_ST4_s:
+      VIXL_STATIC_ASSERT((NEON_LD4_s | (1 << NEONLSSize_offset)) == NEON_LD4_d);
+      VIXL_STATIC_ASSERT((NEON_ST4_s | (1 << NEONLSSize_offset)) == NEON_ST4_d);
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld4" : "st4";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s, 'Vt3.s, 'Vt4.s}['IVLSLane2], ['Xns]";
+      else
+        form = "{'Vt.d, 'Vt2.d, 'Vt3.d, 'Vt4.d}['IVLSLane3], ['Xns]";
+      break;
+    case NEON_LD4R:
+      mnemonic = "ld4r";
+      form = "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s, 'Vt4.%1$s}, ['Xns]";
+      break;
+    default: break;
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONLoadStoreSingleStructPostIndex(
+    const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONLoadStoreSingleStructPostIndex)";
+
+  const char *form_1b = "{'Vt.b}['IVLSLane0], ['Xns], 'Xmb1";
+  const char *form_1h = "{'Vt.h}['IVLSLane1], ['Xns], 'Xmb2";
+  const char *form_1s = "{'Vt.s}['IVLSLane2], ['Xns], 'Xmb4";
+  const char *form_1d = "{'Vt.d}['IVLSLane3], ['Xns], 'Xmb8";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LoadStoreFormatMap());
+
+  switch (instr->Mask(NEONLoadStoreSingleStructPostIndexMask)) {
+    case NEON_LD1_b_post: mnemonic = "ld1"; form = form_1b; break;
+    case NEON_LD1_h_post: mnemonic = "ld1"; form = form_1h; break;
+    case NEON_LD1_s_post:
+      mnemonic = "ld1";
+      VIXL_STATIC_ASSERT((NEON_LD1_s | (1 << NEONLSSize_offset)) == NEON_LD1_d);
+      form = ((instr->NEONLSSize() & 1) == 0) ? form_1s : form_1d;
+      break;
+    case NEON_ST1_b_post: mnemonic = "st1"; form = form_1b; break;
+    case NEON_ST1_h_post: mnemonic = "st1"; form = form_1h; break;
+    case NEON_ST1_s_post:
+      mnemonic = "st1";
+      VIXL_STATIC_ASSERT((NEON_ST1_s | (1 << NEONLSSize_offset)) == NEON_ST1_d);
+      form = ((instr->NEONLSSize() & 1) == 0) ? form_1s : form_1d;
+      break;
+    case NEON_LD1R_post:
+      mnemonic = "ld1r";
+      form = "{'Vt.%s}, ['Xns], 'Xmz1";
+      break;
+    case NEON_LD2_b_post:
+    case NEON_ST2_b_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      form = "{'Vt.b, 'Vt2.b}['IVLSLane0], ['Xns], 'Xmb2";
+      break;
+    case NEON_ST2_h_post:
+    case NEON_LD2_h_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      form = "{'Vt.h, 'Vt2.h}['IVLSLane1], ['Xns], 'Xmb4";
+      break;
+    case NEON_LD2_s_post:
+    case NEON_ST2_s_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld2" : "st2";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s}['IVLSLane2], ['Xns], 'Xmb8";
+      else
+        form = "{'Vt.d, 'Vt2.d}['IVLSLane3], ['Xns], 'Xmb16";
+      break;
+    case NEON_LD2R_post:
+      mnemonic = "ld2r";
+      form = "{'Vt.%s, 'Vt2.%s}, ['Xns], 'Xmz2";
+      break;
+    case NEON_LD3_b_post:
+    case NEON_ST3_b_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      form = "{'Vt.b, 'Vt2.b, 'Vt3.b}['IVLSLane0], ['Xns], 'Xmb3";
+      break;
+    case NEON_LD3_h_post:
+    case NEON_ST3_h_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      form = "{'Vt.h, 'Vt2.h, 'Vt3.h}['IVLSLane1], ['Xns], 'Xmb6";
+      break;
+    case NEON_LD3_s_post:
+    case NEON_ST3_s_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld3" : "st3";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s, 'Vt3.s}['IVLSLane2], ['Xns], 'Xmb12";
+      else
+        form = "{'Vt.d, 'Vt2.d, 'Vt3.d}['IVLSLane3], ['Xns], 'Xmr3";
+      break;
+    case NEON_LD3R_post:
+      mnemonic = "ld3r";
+      form = "{'Vt.%s, 'Vt2.%s, 'Vt3.%s}, ['Xns], 'Xmz3";
+      break;
+    case NEON_LD4_b_post:
+    case NEON_ST4_b_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld4" : "st4";
+      form = "{'Vt.b, 'Vt2.b, 'Vt3.b, 'Vt4.b}['IVLSLane0], ['Xns], 'Xmb4";
+      break;
+    case NEON_LD4_h_post:
+    case NEON_ST4_h_post:
+      mnemonic = (instr->LdStXLoad()) == 1 ? "ld4" : "st4";
+      form = "{'Vt.h, 'Vt2.h, 'Vt3.h, 'Vt4.h}['IVLSLane1], ['Xns], 'Xmb8";
+      break;
+    case NEON_LD4_s_post:
+    case NEON_ST4_s_post:
+      mnemonic = (instr->LdStXLoad() == 1) ? "ld4" : "st4";
+      if ((instr->NEONLSSize() & 1) == 0)
+        form = "{'Vt.s, 'Vt2.s, 'Vt3.s, 'Vt4.s}['IVLSLane2], ['Xns], 'Xmb16";
+      else
+        form = "{'Vt.d, 'Vt2.d, 'Vt3.d, 'Vt4.d}['IVLSLane3], ['Xns], 'Xmb32";
+      break;
+    case NEON_LD4R_post:
+      mnemonic = "ld4r";
+      form = "{'Vt.%1$s, 'Vt2.%1$s, 'Vt3.%1$s, 'Vt4.%1$s}, ['Xns], 'Xmz4";
+      break;
+    default: break;
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONModifiedImmediate(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Vt.%s, 'IVMIImm8, lsl 'IVMIShiftAmt1";
+
+  int cmode   = instr->NEONCmode();
+  int cmode_3 = (cmode >> 3) & 1;
+  int cmode_2 = (cmode >> 2) & 1;
+  int cmode_1 = (cmode >> 1) & 1;
+  int cmode_0 = cmode & 1;
+  int q = instr->NEONQ();
+  int op = instr->NEONModImmOp();
+
+  static const NEONFormatMap map_b = { {30}, {NF_8B, NF_16B} };
+  static const NEONFormatMap map_h = { {30}, {NF_4H, NF_8H} };
+  static const NEONFormatMap map_s = { {30}, {NF_2S, NF_4S} };
+  NEONFormatDecoder nfd(instr, &map_b);
+
+  if (cmode_3 == 0) {
+    if (cmode_0 == 0) {
+      mnemonic = (op == 1) ? "mvni" : "movi";
+    } else {  // cmode<0> == '1'.
+      mnemonic = (op == 1) ? "bic" : "orr";
+    }
+    nfd.SetFormatMap(0, &map_s);
+  } else {  // cmode<3> == '1'.
+    if (cmode_2 == 0) {
+      if (cmode_0 == 0) {
+        mnemonic = (op == 1) ? "mvni" : "movi";
+      } else {  // cmode<0> == '1'.
+        mnemonic = (op == 1) ? "bic" : "orr";
+      }
+      nfd.SetFormatMap(0, &map_h);
+    } else {  // cmode<2> == '1'.
+      if (cmode_1 == 0) {
+        mnemonic = (op == 1) ? "mvni" : "movi";
+        form = "'Vt.%s, 'IVMIImm8, msl 'IVMIShiftAmt2";
+        nfd.SetFormatMap(0, &map_s);
+      } else {   // cmode<1> == '1'.
+        if (cmode_0 == 0) {
+          mnemonic = "movi";
+          if (op == 0) {
+            form = "'Vt.%s, 'IVMIImm8";
+          } else {
+            form = (q == 0) ? "'Dd, 'IVMIImm" : "'Vt.2d, 'IVMIImm";
+          }
+        } else {  // cmode<0> == '1'
+          mnemonic = "fmov";
+          if (op == 0) {
+            form  = "'Vt.%s, 'IVMIImmFPSingle";
+            nfd.SetFormatMap(0, &map_s);
+          } else {
+            if (q == 1) {
+              form = "'Vt.2d, 'IVMIImmFPDouble";
+            }
+          }
+        }
+      }
+    }
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONScalar2RegMisc(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form     = "%sd, %sn";
+  const char *form_0   = "%sd, %sn, #0";
+  const char *form_fp0 = "%sd, %sn, #0.0";
+
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::ScalarFormatMap());
+
+  if (instr->Mask(NEON2RegMiscOpcode) <= NEON_NEG_scalar_opcode) {
+    // These instructions all use a two bit size field, except NOT and RBIT,
+    // which use the field to encode the operation.
+    switch (instr->Mask(NEONScalar2RegMiscMask)) {
+      case NEON_CMGT_zero_scalar: mnemonic = "cmgt"; form = form_0; break;
+      case NEON_CMGE_zero_scalar: mnemonic = "cmge"; form = form_0; break;
+      case NEON_CMLE_zero_scalar: mnemonic = "cmle"; form = form_0; break;
+      case NEON_CMLT_zero_scalar: mnemonic = "cmlt"; form = form_0; break;
+      case NEON_CMEQ_zero_scalar: mnemonic = "cmeq"; form = form_0; break;
+      case NEON_NEG_scalar:       mnemonic = "neg";   break;
+      case NEON_SQNEG_scalar:     mnemonic = "sqneg"; break;
+      case NEON_ABS_scalar:       mnemonic = "abs";   break;
+      case NEON_SQABS_scalar:     mnemonic = "sqabs"; break;
+      case NEON_SUQADD_scalar:    mnemonic = "suqadd"; break;
+      case NEON_USQADD_scalar:    mnemonic = "usqadd"; break;
+      default: form = "(NEONScalar2RegMisc)";
+    }
+  } else {
+    // These instructions all use a one bit size field, except SQXTUN, SQXTN
+    // and UQXTN, which use a two bit size field.
+    nfd.SetFormatMaps(nfd.FPScalarFormatMap());
+    switch (instr->Mask(NEONScalar2RegMiscFPMask)) {
+      case NEON_FRSQRTE_scalar:    mnemonic = "frsqrte"; break;
+      case NEON_FRECPE_scalar:     mnemonic = "frecpe";  break;
+      case NEON_SCVTF_scalar:      mnemonic = "scvtf"; break;
+      case NEON_UCVTF_scalar:      mnemonic = "ucvtf"; break;
+      case NEON_FCMGT_zero_scalar: mnemonic = "fcmgt"; form = form_fp0; break;
+      case NEON_FCMGE_zero_scalar: mnemonic = "fcmge"; form = form_fp0; break;
+      case NEON_FCMLE_zero_scalar: mnemonic = "fcmle"; form = form_fp0; break;
+      case NEON_FCMLT_zero_scalar: mnemonic = "fcmlt"; form = form_fp0; break;
+      case NEON_FCMEQ_zero_scalar: mnemonic = "fcmeq"; form = form_fp0; break;
+      case NEON_FRECPX_scalar:     mnemonic = "frecpx"; break;
+      case NEON_FCVTNS_scalar:     mnemonic = "fcvtns"; break;
+      case NEON_FCVTNU_scalar:     mnemonic = "fcvtnu"; break;
+      case NEON_FCVTPS_scalar:     mnemonic = "fcvtps"; break;
+      case NEON_FCVTPU_scalar:     mnemonic = "fcvtpu"; break;
+      case NEON_FCVTMS_scalar:     mnemonic = "fcvtms"; break;
+      case NEON_FCVTMU_scalar:     mnemonic = "fcvtmu"; break;
+      case NEON_FCVTZS_scalar:     mnemonic = "fcvtzs"; break;
+      case NEON_FCVTZU_scalar:     mnemonic = "fcvtzu"; break;
+      case NEON_FCVTAS_scalar:     mnemonic = "fcvtas"; break;
+      case NEON_FCVTAU_scalar:     mnemonic = "fcvtau"; break;
+      case NEON_FCVTXN_scalar:
+        nfd.SetFormatMap(0, nfd.LongScalarFormatMap());
+        mnemonic = "fcvtxn";
+        break;
+      default:
+        nfd.SetFormatMap(0, nfd.ScalarFormatMap());
+        nfd.SetFormatMap(1, nfd.LongScalarFormatMap());
+        switch (instr->Mask(NEONScalar2RegMiscMask)) {
+          case NEON_SQXTN_scalar:  mnemonic = "sqxtn"; break;
+          case NEON_UQXTN_scalar:  mnemonic = "uqxtn"; break;
+          case NEON_SQXTUN_scalar: mnemonic = "sqxtun"; break;
+          default: form = "(NEONScalar2RegMisc)";
+        }
+    }
+  }
+  Format(instr, mnemonic, nfd.SubstitutePlaceholders(form));
+}
+
+
+void Disassembler::VisitNEONScalar3Diff(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "%sd, %sn, %sm";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::LongScalarFormatMap(),
+                               NEONFormatDecoder::ScalarFormatMap());
+
+  switch (instr->Mask(NEONScalar3DiffMask)) {
+    case NEON_SQDMLAL_scalar  : mnemonic = "sqdmlal"; break;
+    case NEON_SQDMLSL_scalar  : mnemonic = "sqdmlsl"; break;
+    case NEON_SQDMULL_scalar  : mnemonic = "sqdmull"; break;
+    default: form = "(NEONScalar3Diff)";
+  }
+  Format(instr, mnemonic, nfd.SubstitutePlaceholders(form));
+}
+
+
+void Disassembler::VisitNEONScalar3Same(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "%sd, %sn, %sm";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::ScalarFormatMap());
+
+  if (instr->Mask(NEONScalar3SameFPFMask) == NEONScalar3SameFPFixed) {
+    nfd.SetFormatMaps(nfd.FPScalarFormatMap());
+    switch (instr->Mask(NEONScalar3SameFPMask)) {
+      case NEON_FACGE_scalar:   mnemonic = "facge"; break;
+      case NEON_FACGT_scalar:   mnemonic = "facgt"; break;
+      case NEON_FCMEQ_scalar:   mnemonic = "fcmeq"; break;
+      case NEON_FCMGE_scalar:   mnemonic = "fcmge"; break;
+      case NEON_FCMGT_scalar:   mnemonic = "fcmgt"; break;
+      case NEON_FMULX_scalar:   mnemonic = "fmulx"; break;
+      case NEON_FRECPS_scalar:  mnemonic = "frecps"; break;
+      case NEON_FRSQRTS_scalar: mnemonic = "frsqrts"; break;
+      case NEON_FABD_scalar:    mnemonic = "fabd"; break;
+      default: form = "(NEONScalar3Same)";
+    }
+  } else {
+    switch (instr->Mask(NEONScalar3SameMask)) {
+      case NEON_ADD_scalar:    mnemonic = "add";    break;
+      case NEON_SUB_scalar:    mnemonic = "sub";    break;
+      case NEON_CMEQ_scalar:   mnemonic = "cmeq";   break;
+      case NEON_CMGE_scalar:   mnemonic = "cmge";   break;
+      case NEON_CMGT_scalar:   mnemonic = "cmgt";   break;
+      case NEON_CMHI_scalar:   mnemonic = "cmhi";   break;
+      case NEON_CMHS_scalar:   mnemonic = "cmhs";   break;
+      case NEON_CMTST_scalar:  mnemonic = "cmtst";  break;
+      case NEON_UQADD_scalar:  mnemonic = "uqadd";  break;
+      case NEON_SQADD_scalar:  mnemonic = "sqadd";  break;
+      case NEON_UQSUB_scalar:  mnemonic = "uqsub";  break;
+      case NEON_SQSUB_scalar:  mnemonic = "sqsub";  break;
+      case NEON_USHL_scalar:   mnemonic = "ushl";   break;
+      case NEON_SSHL_scalar:   mnemonic = "sshl";   break;
+      case NEON_UQSHL_scalar:  mnemonic = "uqshl";  break;
+      case NEON_SQSHL_scalar:  mnemonic = "sqshl";  break;
+      case NEON_URSHL_scalar:  mnemonic = "urshl";  break;
+      case NEON_SRSHL_scalar:  mnemonic = "srshl";  break;
+      case NEON_UQRSHL_scalar: mnemonic = "uqrshl"; break;
+      case NEON_SQRSHL_scalar: mnemonic = "sqrshl"; break;
+      case NEON_SQDMULH_scalar:  mnemonic = "sqdmulh";  break;
+      case NEON_SQRDMULH_scalar: mnemonic = "sqrdmulh"; break;
+      default: form = "(NEONScalar3Same)";
+    }
+  }
+  Format(instr, mnemonic, nfd.SubstitutePlaceholders(form));
+}
+
+
+void Disassembler::VisitNEONScalarByIndexedElement(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "%sd, %sn, 'Ve.%s['IVByElemIndex]";
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::ScalarFormatMap());
+  bool long_instr = false;
+
+  switch (instr->Mask(NEONScalarByIndexedElementMask)) {
+    case NEON_SQDMULL_byelement_scalar:
+      mnemonic = "sqdmull";
+      long_instr = true;
+      break;
+    case NEON_SQDMLAL_byelement_scalar:
+      mnemonic = "sqdmlal";
+      long_instr = true;
+      break;
+    case NEON_SQDMLSL_byelement_scalar:
+      mnemonic = "sqdmlsl";
+      long_instr = true;
+      break;
+    case NEON_SQDMULH_byelement_scalar:
+      mnemonic = "sqdmulh";
+      break;
+    case NEON_SQRDMULH_byelement_scalar:
+      mnemonic = "sqrdmulh";
+      break;
+    default:
+      nfd.SetFormatMap(0, nfd.FPScalarFormatMap());
+      switch (instr->Mask(NEONScalarByIndexedElementFPMask)) {
+        case NEON_FMUL_byelement_scalar: mnemonic = "fmul"; break;
+        case NEON_FMLA_byelement_scalar: mnemonic = "fmla"; break;
+        case NEON_FMLS_byelement_scalar: mnemonic = "fmls"; break;
+        case NEON_FMULX_byelement_scalar: mnemonic = "fmulx"; break;
+        default: form = "(NEONScalarByIndexedElement)";
+      }
+  }
+
+  if (long_instr) {
+    nfd.SetFormatMap(0, nfd.LongScalarFormatMap());
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(
+    form, nfd.kPlaceholder, nfd.kPlaceholder, nfd.kFormat));
+}
+
+
+void Disassembler::VisitNEONScalarCopy(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONScalarCopy)";
+
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::TriangularScalarFormatMap());
+
+  if (instr->Mask(NEONScalarCopyMask) == NEON_DUP_ELEMENT_scalar) {
+    mnemonic = "mov";
+    form = "%sd, 'Vn.%s['IVInsIndex1]";
+  }
+
+  Format(instr, mnemonic, nfd.Substitute(form, nfd.kPlaceholder, nfd.kFormat));
+}
+
+
+void Disassembler::VisitNEONScalarPairwise(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "%sd, 'Vn.%s";
+  NEONFormatMap map = { {22}, {NF_2S, NF_2D} };
+  NEONFormatDecoder nfd(instr, NEONFormatDecoder::FPScalarFormatMap(), &map);
+
+  switch (instr->Mask(NEONScalarPairwiseMask)) {
+    case NEON_ADDP_scalar:    mnemonic = "addp"; break;
+    case NEON_FADDP_scalar:   mnemonic = "faddp"; break;
+    case NEON_FMAXP_scalar:   mnemonic = "fmaxp"; break;
+    case NEON_FMAXNMP_scalar: mnemonic = "fmaxnmp"; break;
+    case NEON_FMINP_scalar:   mnemonic = "fminp"; break;
+    case NEON_FMINNMP_scalar: mnemonic = "fminnmp"; break;
+    default: form = "(NEONScalarPairwise)";
+  }
+  Format(instr, mnemonic, nfd.Substitute(form,
+      NEONFormatDecoder::kPlaceholder, NEONFormatDecoder::kFormat));
+}
+
+
+void Disassembler::VisitNEONScalarShiftImmediate(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form   = "%sd, %sn, 'Is1";
+  const char *form_2 = "%sd, %sn, 'Is2";
+
+  static const NEONFormatMap map_shift = {
+    {22, 21, 20, 19},
+    {NF_UNDEF, NF_B, NF_H, NF_H, NF_S, NF_S, NF_S, NF_S,
+     NF_D,     NF_D, NF_D, NF_D, NF_D, NF_D, NF_D, NF_D}
+  };
+  static const NEONFormatMap map_shift_narrow = {
+    {21, 20, 19},
+    {NF_UNDEF, NF_H, NF_S, NF_S, NF_D, NF_D, NF_D, NF_D}
+  };
+  NEONFormatDecoder nfd(instr, &map_shift);
+
+  if (instr->ImmNEONImmh()) {  // immh has to be non-zero.
+    switch (instr->Mask(NEONScalarShiftImmediateMask)) {
+      case NEON_FCVTZU_imm_scalar: mnemonic = "fcvtzu";    break;
+      case NEON_FCVTZS_imm_scalar: mnemonic = "fcvtzs";   break;
+      case NEON_SCVTF_imm_scalar: mnemonic = "scvtf";    break;
+      case NEON_UCVTF_imm_scalar: mnemonic = "ucvtf";   break;
+      case NEON_SRI_scalar:       mnemonic = "sri";    break;
+      case NEON_SSHR_scalar:      mnemonic = "sshr";   break;
+      case NEON_USHR_scalar:      mnemonic = "ushr";   break;
+      case NEON_SRSHR_scalar:     mnemonic = "srshr";  break;
+      case NEON_URSHR_scalar:     mnemonic = "urshr";  break;
+      case NEON_SSRA_scalar:      mnemonic = "ssra";   break;
+      case NEON_USRA_scalar:      mnemonic = "usra";   break;
+      case NEON_SRSRA_scalar:     mnemonic = "srsra";  break;
+      case NEON_URSRA_scalar:     mnemonic = "ursra";  break;
+      case NEON_SHL_scalar:       mnemonic = "shl";    form = form_2; break;
+      case NEON_SLI_scalar:       mnemonic = "sli";    form = form_2; break;
+      case NEON_SQSHLU_scalar:    mnemonic = "sqshlu"; form = form_2; break;
+      case NEON_SQSHL_imm_scalar: mnemonic = "sqshl";  form = form_2; break;
+      case NEON_UQSHL_imm_scalar: mnemonic = "uqshl";  form = form_2; break;
+      case NEON_UQSHRN_scalar:
+        mnemonic = "uqshrn";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      case NEON_UQRSHRN_scalar:
+        mnemonic = "uqrshrn";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      case NEON_SQSHRN_scalar:
+        mnemonic = "sqshrn";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      case NEON_SQRSHRN_scalar:
+        mnemonic = "sqrshrn";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      case NEON_SQSHRUN_scalar:
+        mnemonic = "sqshrun";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      case NEON_SQRSHRUN_scalar:
+        mnemonic = "sqrshrun";
+        nfd.SetFormatMap(1, &map_shift_narrow);
+        break;
+      default:
+        form = "(NEONScalarShiftImmediate)";
+    }
+  } else {
+    form = "(NEONScalarShiftImmediate)";
+  }
+  Format(instr, mnemonic, nfd.SubstitutePlaceholders(form));
+}
+
+
+void Disassembler::VisitNEONShiftImmediate(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form         = "'Vd.%s, 'Vn.%s, 'Is1";
+  const char *form_shift_2 = "'Vd.%s, 'Vn.%s, 'Is2";
+  const char *form_xtl     = "'Vd.%s, 'Vn.%s";
+
+  // 0001->8H, 001x->4S, 01xx->2D, all others undefined.
+  static const NEONFormatMap map_shift_ta = {
+    {22, 21, 20, 19},
+    {NF_UNDEF, NF_8H, NF_4S, NF_4S, NF_2D, NF_2D, NF_2D, NF_2D}
+  };
+
+  // 00010->8B, 00011->16B, 001x0->4H, 001x1->8H,
+  // 01xx0->2S, 01xx1->4S, 1xxx1->2D, all others undefined.
+  static const NEONFormatMap map_shift_tb = {
+    {22, 21, 20, 19, 30},
+    {NF_UNDEF, NF_UNDEF, NF_8B,    NF_16B, NF_4H,    NF_8H, NF_4H,    NF_8H,
+     NF_2S,    NF_4S,    NF_2S,    NF_4S,  NF_2S,    NF_4S, NF_2S,    NF_4S,
+     NF_UNDEF, NF_2D,    NF_UNDEF, NF_2D,  NF_UNDEF, NF_2D, NF_UNDEF, NF_2D,
+     NF_UNDEF, NF_2D,    NF_UNDEF, NF_2D,  NF_UNDEF, NF_2D, NF_UNDEF, NF_2D}
+  };
+
+  NEONFormatDecoder nfd(instr, &map_shift_tb);
+
+  if (instr->ImmNEONImmh()) {  // immh has to be non-zero.
+    switch (instr->Mask(NEONShiftImmediateMask)) {
+      case NEON_SQSHLU:     mnemonic = "sqshlu"; form = form_shift_2; break;
+      case NEON_SQSHL_imm:  mnemonic = "sqshl";  form = form_shift_2; break;
+      case NEON_UQSHL_imm:  mnemonic = "uqshl";  form = form_shift_2; break;
+      case NEON_SHL:        mnemonic = "shl";    form = form_shift_2; break;
+      case NEON_SLI:        mnemonic = "sli";    form = form_shift_2; break;
+      case NEON_SCVTF_imm:  mnemonic = "scvtf";  break;
+      case NEON_UCVTF_imm:  mnemonic = "ucvtf";  break;
+      case NEON_FCVTZU_imm: mnemonic = "fcvtzu"; break;
+      case NEON_FCVTZS_imm: mnemonic = "fcvtzs"; break;
+      case NEON_SRI:        mnemonic = "sri";    break;
+      case NEON_SSHR:       mnemonic = "sshr";   break;
+      case NEON_USHR:       mnemonic = "ushr";   break;
+      case NEON_SRSHR:      mnemonic = "srshr";  break;
+      case NEON_URSHR:      mnemonic = "urshr";  break;
+      case NEON_SSRA:       mnemonic = "ssra";   break;
+      case NEON_USRA:       mnemonic = "usra";   break;
+      case NEON_SRSRA:      mnemonic = "srsra";  break;
+      case NEON_URSRA:      mnemonic = "ursra";  break;
+      case NEON_SHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "shrn2" : "shrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_RSHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "rshrn2" : "rshrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_UQSHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "uqshrn2" : "uqshrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_UQRSHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "uqrshrn2" : "uqrshrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_SQSHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "sqshrn2" : "sqshrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_SQRSHRN:
+        mnemonic = instr->Mask(NEON_Q) ? "sqrshrn2" : "sqrshrn";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_SQSHRUN:
+        mnemonic = instr->Mask(NEON_Q) ? "sqshrun2" : "sqshrun";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_SQRSHRUN:
+        mnemonic = instr->Mask(NEON_Q) ? "sqrshrun2" : "sqrshrun";
+        nfd.SetFormatMap(1, &map_shift_ta);
+        break;
+      case NEON_SSHLL:
+        nfd.SetFormatMap(0, &map_shift_ta);
+        if (instr->ImmNEONImmb() == 0 &&
+            CountSetBits(instr->ImmNEONImmh(), 32) == 1) {  // sxtl variant.
+          form = form_xtl;
+          mnemonic = instr->Mask(NEON_Q) ? "sxtl2" : "sxtl";
+        } else {  // sshll variant.
+          form = form_shift_2;
+          mnemonic = instr->Mask(NEON_Q) ? "sshll2" : "sshll";
+        }
+        break;
+      case NEON_USHLL:
+        nfd.SetFormatMap(0, &map_shift_ta);
+        if (instr->ImmNEONImmb() == 0 &&
+            CountSetBits(instr->ImmNEONImmh(), 32) == 1) {  // uxtl variant.
+          form = form_xtl;
+          mnemonic = instr->Mask(NEON_Q) ? "uxtl2" : "uxtl";
+        } else {  // ushll variant.
+          form = form_shift_2;
+          mnemonic = instr->Mask(NEON_Q) ? "ushll2" : "ushll";
+        }
+        break;
+      default: form = "(NEONShiftImmediate)";
+    }
+  } else {
+    form = "(NEONShiftImmediate)";
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitNEONTable(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "(NEONTable)";
+  const char form_1v[] = "'Vd.%%s, {'Vn.16b}, 'Vm.%%s";
+  const char form_2v[] = "'Vd.%%s, {'Vn.16b, v%d.16b}, 'Vm.%%s";
+  const char form_3v[] = "'Vd.%%s, {'Vn.16b, v%d.16b, v%d.16b}, 'Vm.%%s";
+  const char form_4v[] =
+      "'Vd.%%s, {'Vn.16b, v%d.16b, v%d.16b, v%d.16b}, 'Vm.%%s";
+  static const NEONFormatMap map_b = { {30}, {NF_8B, NF_16B} };
+  NEONFormatDecoder nfd(instr, &map_b);
+
+  switch (instr->Mask(NEONTableMask)) {
+    case NEON_TBL_1v: mnemonic = "tbl"; form = form_1v; break;
+    case NEON_TBL_2v: mnemonic = "tbl"; form = form_2v; break;
+    case NEON_TBL_3v: mnemonic = "tbl"; form = form_3v; break;
+    case NEON_TBL_4v: mnemonic = "tbl"; form = form_4v; break;
+    case NEON_TBX_1v: mnemonic = "tbx"; form = form_1v; break;
+    case NEON_TBX_2v: mnemonic = "tbx"; form = form_2v; break;
+    case NEON_TBX_3v: mnemonic = "tbx"; form = form_3v; break;
+    case NEON_TBX_4v: mnemonic = "tbx"; form = form_4v; break;
+    default: break;
+  }
+
+  char re_form[sizeof(form_4v) + 6];
+  int reg_num = instr->Rn();
+  snprintf(re_form, sizeof(re_form), form,
+           (reg_num + 1) % kNumberOfVRegisters,
+           (reg_num + 2) % kNumberOfVRegisters,
+           (reg_num + 3) % kNumberOfVRegisters);
+
+  Format(instr, mnemonic, nfd.Substitute(re_form));
+}
+
+
+void Disassembler::VisitNEONPerm(const Instruction* instr) {
+  const char *mnemonic = "unimplemented";
+  const char *form = "'Vd.%s, 'Vn.%s, 'Vm.%s";
+  NEONFormatDecoder nfd(instr);
+
+  switch (instr->Mask(NEONPermMask)) {
+    case NEON_TRN1: mnemonic = "trn1";   break;
+    case NEON_TRN2: mnemonic = "trn2";  break;
+    case NEON_UZP1: mnemonic = "uzp1"; break;
+    case NEON_UZP2: mnemonic = "uzp2";  break;
+    case NEON_ZIP1: mnemonic = "zip1"; break;
+    case NEON_ZIP2: mnemonic = "zip2"; break;
+    default: form = "(NEONPerm)";
+  }
+  Format(instr, mnemonic, nfd.Substitute(form));
+}
+
+
+void Disassembler::VisitUnimplemented(const Instruction* instr) {
+  Format(instr, "unimplemented", "(Unimplemented)");
+}
+
+
+void Disassembler::VisitUnallocated(const Instruction* instr) {
+  Format(instr, "unallocated", "(Unallocated)");
+}
+
+
+void Disassembler::ProcessOutput(const Instruction* /*instr*/) {
+  // The base disasm does nothing more than disassembling into a buffer.
+}
+
+
+void Disassembler::AppendRegisterNameToOutput(const Instruction* instr,
+                                              const CPURegister& reg) {
+  USE(instr);
+  VIXL_ASSERT(reg.IsValid());
+  char reg_char;
+
+  if (reg.IsRegister()) {
+    reg_char = reg.Is64Bits() ? 'x' : 'w';
+  } else {
+    VIXL_ASSERT(reg.IsVRegister());
+    switch (reg.SizeInBits()) {
+      case kBRegSize: reg_char = 'b'; break;
+      case kHRegSize: reg_char = 'h'; break;
+      case kSRegSize: reg_char = 's'; break;
+      case kDRegSize: reg_char = 'd'; break;
+      default:
+        VIXL_ASSERT(reg.Is128Bits());
+        reg_char = 'q';
+    }
+  }
+
+  if (reg.IsVRegister() || !(reg.Aliases(sp) || reg.Aliases(xzr))) {
+    // A core or scalar/vector register: [wx]0 - 30, [bhsdq]0 - 31.
+    AppendToOutput("%c%d", reg_char, reg.code());
+  } else if (reg.Aliases(sp)) {
+    // Disassemble w31/x31 as stack pointer wsp/sp.
+    AppendToOutput("%s", reg.Is64Bits() ? "sp" : "wsp");
+  } else {
+    // Disassemble w31/x31 as zero register wzr/xzr.
+    AppendToOutput("%czr", reg_char);
+  }
+}
+
+
+void Disassembler::AppendPCRelativeOffsetToOutput(const Instruction* instr,
+                                                  int64_t offset) {
+  USE(instr);
+  char sign = (offset < 0) ? '-' : '+';
+  AppendToOutput("#%c0x%" PRIx64, sign, std::abs(offset));
+}
+
+
+void Disassembler::AppendAddressToOutput(const Instruction* instr,
+                                         const void* addr) {
+  USE(instr);
+  AppendToOutput("(addr 0x%" PRIxPTR ")", reinterpret_cast<uintptr_t>(addr));
+}
+
+
+void Disassembler::AppendCodeAddressToOutput(const Instruction* instr,
+                                             const void* addr) {
+  AppendAddressToOutput(instr, addr);
+}
+
+
+void Disassembler::AppendDataAddressToOutput(const Instruction* instr,
+                                             const void* addr) {
+  AppendAddressToOutput(instr, addr);
+}
+
+
+void Disassembler::AppendCodeRelativeAddressToOutput(const Instruction* instr,
+                                                     const void* addr) {
+  USE(instr);
+  int64_t rel_addr = CodeRelativeAddress(addr);
+  if (rel_addr >= 0) {
+    AppendToOutput("(addr 0x%" PRIx64 ")", rel_addr);
+  } else {
+    AppendToOutput("(addr -0x%" PRIx64 ")", -rel_addr);
+  }
+}
+
+
+void Disassembler::AppendCodeRelativeCodeAddressToOutput(
+    const Instruction* instr, const void* addr) {
+  AppendCodeRelativeAddressToOutput(instr, addr);
+}
+
+
+void Disassembler::AppendCodeRelativeDataAddressToOutput(
+    const Instruction* instr, const void* addr) {
+  AppendCodeRelativeAddressToOutput(instr, addr);
+}
+
+
+void Disassembler::MapCodeAddress(int64_t base_address,
+                                  const Instruction* instr_address) {
+  set_code_address_offset(
+      base_address - reinterpret_cast<intptr_t>(instr_address));
+}
+int64_t Disassembler::CodeRelativeAddress(const void* addr) {
+  return reinterpret_cast<intptr_t>(addr) + code_address_offset();
+}
+
+
+void Disassembler::Format(const Instruction* instr, const char* mnemonic,
+                          const char* format) {
+  VIXL_ASSERT(mnemonic != NULL);
+  ResetOutput();
+  Substitute(instr, mnemonic);
+  if (format != NULL) {
+    VIXL_ASSERT(buffer_pos_ < buffer_size_);
+    buffer_[buffer_pos_++] = ' ';
+    Substitute(instr, format);
+  }
+  VIXL_ASSERT(buffer_pos_ < buffer_size_);
+  buffer_[buffer_pos_] = 0;
+  ProcessOutput(instr);
+}
+
+
+void Disassembler::Substitute(const Instruction* instr, const char* string) {
+  char chr = *string++;
+  while (chr != '\0') {
+    if (chr == '\'') {
+      string += SubstituteField(instr, string);
+    } else {
+      VIXL_ASSERT(buffer_pos_ < buffer_size_);
+      buffer_[buffer_pos_++] = chr;
+    }
+    chr = *string++;
+  }
+}
+
+
+int Disassembler::SubstituteField(const Instruction* instr,
+                                  const char* format) {
+  switch (format[0]) {
+    // NB. The remaining substitution prefix characters are: GJKUZ.
+    case 'R':  // Register. X or W, selected by sf bit.
+    case 'F':  // FP register. S or D, selected by type field.
+    case 'V':  // Vector register, V, vector format.
+    case 'W':
+    case 'X':
+    case 'B':
+    case 'H':
+    case 'S':
+    case 'D':
+    case 'Q': return SubstituteRegisterField(instr, format);
+    case 'I': return SubstituteImmediateField(instr, format);
+    case 'L': return SubstituteLiteralField(instr, format);
+    case 'N': return SubstituteShiftField(instr, format);
+    case 'P': return SubstitutePrefetchField(instr, format);
+    case 'C': return SubstituteConditionField(instr, format);
+    case 'E': return SubstituteExtendField(instr, format);
+    case 'A': return SubstitutePCRelAddressField(instr, format);
+    case 'T': return SubstituteBranchTargetField(instr, format);
+    case 'O': return SubstituteLSRegOffsetField(instr, format);
+    case 'M': return SubstituteBarrierField(instr, format);
+    case 'K': return SubstituteCrField(instr, format);
+    case 'G': return SubstituteSysOpField(instr, format);
+    default: {
+      VIXL_UNREACHABLE();
+      return 1;
+    }
+  }
+}
+
+
+int Disassembler::SubstituteRegisterField(const Instruction* instr,
+                                          const char* format) {
+  char reg_prefix = format[0];
+  unsigned reg_num = 0;
+  unsigned field_len = 2;
+
+  switch (format[1]) {
+    case 'd':
+      reg_num = instr->Rd();
+      if (format[2] == 'q') {
+        reg_prefix = instr->NEONQ() ? 'X' : 'W';
+        field_len = 3;
+      }
+      break;
+    case 'n': reg_num = instr->Rn(); break;
+    case 'm':
+      reg_num = instr->Rm();
+      switch (format[2]) {
+          // Handle registers tagged with b (bytes), z (instruction), or
+          // r (registers), used for address updates in
+          // NEON load/store instructions.
+        case 'r':
+        case 'b':
+        case 'z': {
+          field_len = 3;
+          char* eimm;
+          int imm = static_cast<int>(strtol(&format[3], &eimm, 10));
+          field_len += eimm - &format[3];
+          if (reg_num == 31) {
+            switch (format[2]) {
+              case 'z':
+                imm *= (1 << instr->NEONLSSize());
+                break;
+              case 'r':
+                imm *= (instr->NEONQ() == 0) ? kDRegSizeInBytes
+                                             : kQRegSizeInBytes;
+                break;
+              case 'b':
+                break;
+            }
+            AppendToOutput("#%d", imm);
+            return field_len;
+          }
+          break;
+        }
+      }
+      break;
+    case 'e':
+      // This is register Rm, but using a 4-bit specifier. Used in NEON
+      // by-element instructions.
+      reg_num = (instr->Rm() & 0xf);
+      break;
+    case 'a': reg_num = instr->Ra(); break;
+    case 's': reg_num = instr->Rs(); break;
+    case 't':
+      reg_num = instr->Rt();
+      if (format[0] == 'V') {
+        if ((format[2] >= '2') && (format[2] <= '4')) {
+          // Handle consecutive vector register specifiers Vt2, Vt3 and Vt4.
+          reg_num = (reg_num + format[2] - '1') % 32;
+          field_len = 3;
+        }
+      } else {
+        if (format[2] == '2') {
+        // Handle register specifier Rt2.
+          reg_num = instr->Rt2();
+          field_len = 3;
+        }
+      }
+      break;
+    default: VIXL_UNREACHABLE();
+  }
+
+  // Increase field length for registers tagged as stack.
+  if (format[2] == 's') {
+    field_len = 3;
+  }
+
+  CPURegister::RegisterType reg_type = CPURegister::kRegister;
+  unsigned reg_size = kXRegSize;
+
+  if (reg_prefix == 'R') {
+    reg_prefix = instr->SixtyFourBits() ? 'X' : 'W';
+  } else if (reg_prefix == 'F') {
+    reg_prefix = ((instr->FPType() & 1) == 0) ? 'S' : 'D';
+  }
+
+  switch (reg_prefix) {
+    case 'W':
+      reg_type = CPURegister::kRegister; reg_size = kWRegSize; break;
+    case 'X':
+      reg_type = CPURegister::kRegister; reg_size = kXRegSize; break;
+    case 'B':
+      reg_type = CPURegister::kVRegister; reg_size = kBRegSize; break;
+    case 'H':
+      reg_type = CPURegister::kVRegister; reg_size = kHRegSize; break;
+    case 'S':
+      reg_type = CPURegister::kVRegister; reg_size = kSRegSize; break;
+    case 'D':
+      reg_type = CPURegister::kVRegister; reg_size = kDRegSize; break;
+    case 'Q':
+      reg_type = CPURegister::kVRegister; reg_size = kQRegSize; break;
+    case 'V':
+      AppendToOutput("v%d", reg_num);
+      return field_len;
+    default:
+      VIXL_UNREACHABLE();
+  }
+
+  if ((reg_type == CPURegister::kRegister) &&
+      (reg_num == kZeroRegCode) && (format[2] == 's')) {
+    reg_num = kSPRegInternalCode;
+  }
+
+  AppendRegisterNameToOutput(instr, CPURegister(reg_num, reg_size, reg_type));
+
+  return field_len;
+}
+
+
+int Disassembler::SubstituteImmediateField(const Instruction* instr,
+                                           const char* format) {
+  VIXL_ASSERT(format[0] == 'I');
+
+  switch (format[1]) {
+    case 'M': {  // IMoveImm, IMoveNeg or IMoveLSL.
+      if (format[5] == 'L') {
+        AppendToOutput("#0x%" PRIx32, instr->ImmMoveWide());
+        if (instr->ShiftMoveWide() > 0) {
+          AppendToOutput(", lsl #%" PRId32, 16 * instr->ShiftMoveWide());
+        }
+      } else {
+        VIXL_ASSERT((format[5] == 'I') || (format[5] == 'N'));
+        uint64_t imm = static_cast<uint64_t>(instr->ImmMoveWide()) <<
+            (16 * instr->ShiftMoveWide());
+        if (format[5] == 'N')
+          imm = ~imm;
+        if (!instr->SixtyFourBits())
+          imm &= UINT64_C(0xffffffff);
+        AppendToOutput("#0x%" PRIx64, imm);
+      }
+      return 8;
+    }
+    case 'L': {
+      switch (format[2]) {
+        case 'L': {  // ILLiteral - Immediate Load Literal.
+          AppendToOutput("pc%+" PRId32,
+                         instr->ImmLLiteral() << kLiteralEntrySizeLog2);
+          return 9;
+        }
+        case 'S': {  // ILS - Immediate Load/Store.
+          if (instr->ImmLS() != 0) {
+            AppendToOutput(", #%" PRId32, instr->ImmLS());
+          }
+          return 3;
+        }
+        case 'P': {  // ILPx - Immediate Load/Store Pair, x = access size.
+          if (instr->ImmLSPair() != 0) {
+            // format[3] is the scale value. Convert to a number.
+            int scale = 1 << (format[3] - '0');
+            AppendToOutput(", #%" PRId32, instr->ImmLSPair() * scale);
+          }
+          return 4;
+        }
+        case 'U': {  // ILU - Immediate Load/Store Unsigned.
+          if (instr->ImmLSUnsigned() != 0) {
+            int shift = instr->SizeLS();
+            AppendToOutput(", #%" PRId32, instr->ImmLSUnsigned() << shift);
+          }
+          return 3;
+        }
+      }
+    }
+    case 'C': {  // ICondB - Immediate Conditional Branch.
+      int64_t offset = instr->ImmCondBranch() << 2;
+      AppendPCRelativeOffsetToOutput(instr, offset);
+      return 6;
+    }
+    case 'A': {  // IAddSub.
+      VIXL_ASSERT(instr->ShiftAddSub() <= 1);
+      int64_t imm = instr->ImmAddSub() << (12 * instr->ShiftAddSub());
+      AppendToOutput("#0x%" PRIx64 " (%" PRId64 ")", imm, imm);
+      return 7;
+    }
+    case 'F': {  // IFPSingle, IFPDouble or IFPFBits.
+      if (format[3] == 'F') {  // IFPFbits.
+        AppendToOutput("#%" PRId32, 64 - instr->FPScale());
+        return 8;
+      } else {
+        AppendToOutput("#0x%" PRIx32 " (%.4f)", instr->ImmFP(),
+                       format[3] == 'S' ? instr->ImmFP32() : instr->ImmFP64());
+        return 9;
+      }
+    }
+    case 'T': {  // ITri - Immediate Triangular Encoded.
+      AppendToOutput("#0x%" PRIx64, instr->ImmLogical());
+      return 4;
+    }
+    case 'N': {  // INzcv.
+      int nzcv = (instr->Nzcv() << Flags_offset);
+      AppendToOutput("#%c%c%c%c", ((nzcv & NFlag) == 0) ? 'n' : 'N',
+                                  ((nzcv & ZFlag) == 0) ? 'z' : 'Z',
+                                  ((nzcv & CFlag) == 0) ? 'c' : 'C',
+                                  ((nzcv & VFlag) == 0) ? 'v' : 'V');
+      return 5;
+    }
+    case 'P': {  // IP - Conditional compare.
+      AppendToOutput("#%" PRId32, instr->ImmCondCmp());
+      return 2;
+    }
+    case 'B': {  // Bitfields.
+      return SubstituteBitfieldImmediateField(instr, format);
+    }
+    case 'E': {  // IExtract.
+      AppendToOutput("#%" PRId32, instr->ImmS());
+      return 8;
+    }
+    case 'S': {  // IS - Test and branch bit.
+      AppendToOutput("#%" PRId32, (instr->ImmTestBranchBit5() << 5) |
+                                  instr->ImmTestBranchBit40());
+      return 2;
+    }
+    case 's': {  // Is - Shift (immediate).
+      switch (format[2]) {
+        case '1': {  // Is1 - SSHR.
+          int shift = 16 << HighestSetBitPosition(instr->ImmNEONImmh());
+          shift -= instr->ImmNEONImmhImmb();
+          AppendToOutput("#%d", shift);
+          return 3;
+        }
+        case '2': {  // Is2 - SLI.
+          int shift = instr->ImmNEONImmhImmb();
+          shift -= 8 << HighestSetBitPosition(instr->ImmNEONImmh());
+          AppendToOutput("#%d", shift);
+          return 3;
+        }
+        default: {
+          VIXL_UNIMPLEMENTED();
+          return 0;
+        }
+      }
+    }
+    case 'D': {  // IDebug - HLT and BRK instructions.
+      AppendToOutput("#0x%" PRIx32, instr->ImmException());
+      return 6;
+    }
+    case 'V': {  // Immediate Vector.
+      switch (format[2]) {
+        case 'E': {  // IVExtract.
+          AppendToOutput("#%" PRId32, instr->ImmNEONExt());
+          return 9;
+        }
+        case 'B': {  // IVByElemIndex.
+          int vm_index = (instr->NEONH() << 1) | instr->NEONL();
+          if (instr->NEONSize() == 1) {
+            vm_index = (vm_index << 1) | instr->NEONM();
+          }
+          AppendToOutput("%d", vm_index);
+          return strlen("IVByElemIndex");
+        }
+        case 'I': {  // INS element.
+          if (strncmp(format, "IVInsIndex", strlen("IVInsIndex")) == 0) {
+            int rd_index, rn_index;
+            int imm5 = instr->ImmNEON5();
+            int imm4 = instr->ImmNEON4();
+            int tz = CountTrailingZeros(imm5, 32);
+            rd_index = imm5 >> (tz + 1);
+            rn_index = imm4 >> tz;
+            if (strncmp(format, "IVInsIndex1", strlen("IVInsIndex1")) == 0) {
+              AppendToOutput("%d", rd_index);
+              return strlen("IVInsIndex1");
+            } else if (strncmp(format, "IVInsIndex2",
+                       strlen("IVInsIndex2")) == 0) {
+              AppendToOutput("%d", rn_index);
+              return strlen("IVInsIndex2");
+            } else {
+              VIXL_UNIMPLEMENTED();
+              return 0;
+            }
+          }
+          VIXL_FALLTHROUGH();
+        }
+        case 'L': {  // IVLSLane[0123] - suffix indicates access size shift.
+          AppendToOutput("%d", instr->NEONLSIndex(format[8] - '0'));
+          return 9;
+        }
+        case 'M': {  // Modified Immediate cases.
+          if (strncmp(format,
+                      "IVMIImmFPSingle",
+                      strlen("IVMIImmFPSingle")) == 0) {
+            AppendToOutput("#0x%" PRIx32 " (%.4f)", instr->ImmNEONabcdefgh(),
+                           instr->ImmNEONFP32());
+            return strlen("IVMIImmFPSingle");
+          } else if (strncmp(format,
+                             "IVMIImmFPDouble",
+                             strlen("IVMIImmFPDouble")) == 0) {
+            AppendToOutput("#0x%" PRIx32 " (%.4f)", instr->ImmNEONabcdefgh(),
+                           instr->ImmNEONFP64());
+            return strlen("IVMIImmFPDouble");
+          } else if (strncmp(format, "IVMIImm8", strlen("IVMIImm8")) == 0) {
+            uint64_t imm8 = instr->ImmNEONabcdefgh();
+            AppendToOutput("#0x%" PRIx64, imm8);
+            return strlen("IVMIImm8");
+          } else if (strncmp(format, "IVMIImm", strlen("IVMIImm")) == 0) {
+            uint64_t imm8 = instr->ImmNEONabcdefgh();
+            uint64_t imm = 0;
+            for (int i = 0; i < 8; ++i) {
+              if (imm8 & (1 << i)) {
+                imm |= (UINT64_C(0xff) << (8 * i));
+              }
+            }
+            AppendToOutput("#0x%" PRIx64, imm);
+            return strlen("IVMIImm");
+          } else if (strncmp(format, "IVMIShiftAmt1",
+                             strlen("IVMIShiftAmt1")) == 0) {
+            int cmode = instr->NEONCmode();
+            int shift_amount = 8 * ((cmode >> 1) & 3);
+            AppendToOutput("#%d", shift_amount);
+            return strlen("IVMIShiftAmt1");
+          } else if (strncmp(format, "IVMIShiftAmt2",
+                             strlen("IVMIShiftAmt2")) == 0) {
+            int cmode = instr->NEONCmode();
+            int shift_amount = 8 << (cmode & 1);
+            AppendToOutput("#%d", shift_amount);
+            return strlen("IVMIShiftAmt2");
+          } else {
+            VIXL_UNIMPLEMENTED();
+            return 0;
+          }
+        }
+        default: {
+          VIXL_UNIMPLEMENTED();
+          return 0;
+        }
+      }
+    }
+    case 'X': {  // IX - CLREX instruction.
+      AppendToOutput("#0x%" PRIx32, instr->CRm());
+      return 2;
+    }
+    default: {
+      VIXL_UNIMPLEMENTED();
+      return 0;
+    }
+  }
+}
+
+
+int Disassembler::SubstituteBitfieldImmediateField(const Instruction* instr,
+                                                   const char* format) {
+  VIXL_ASSERT((format[0] == 'I') && (format[1] == 'B'));
+  unsigned r = instr->ImmR();
+  unsigned s = instr->ImmS();
+
+  switch (format[2]) {
+    case 'r': {  // IBr.
+      AppendToOutput("#%d", r);
+      return 3;
+    }
+    case 's': {  // IBs+1 or IBs-r+1.
+      if (format[3] == '+') {
+        AppendToOutput("#%d", s + 1);
+        return 5;
+      } else {
+        VIXL_ASSERT(format[3] == '-');
+        AppendToOutput("#%d", s - r + 1);
+        return 7;
+      }
+    }
+    case 'Z': {  // IBZ-r.
+      VIXL_ASSERT((format[3] == '-') && (format[4] == 'r'));
+      unsigned reg_size = (instr->SixtyFourBits() == 1) ? kXRegSize : kWRegSize;
+      AppendToOutput("#%d", reg_size - r);
+      return 5;
+    }
+    default: {
+      VIXL_UNREACHABLE();
+      return 0;
+    }
+  }
+}
+
+
+int Disassembler::SubstituteLiteralField(const Instruction* instr,
+                                         const char* format) {
+  VIXL_ASSERT(strncmp(format, "LValue", 6) == 0);
+  USE(format);
+
+  const void * address = instr->LiteralAddress<const void *>();
+  switch (instr->Mask(LoadLiteralMask)) {
+    case LDR_w_lit:
+    case LDR_x_lit:
+    case LDRSW_x_lit:
+    case LDR_s_lit:
+    case LDR_d_lit:
+    case LDR_q_lit:
+      AppendCodeRelativeDataAddressToOutput(instr, address);
+      break;
+    case PRFM_lit: {
+      // Use the prefetch hint to decide how to print the address.
+      switch (instr->PrefetchHint()) {
+        case 0x0:     // PLD: prefetch for load.
+        case 0x2:     // PST: prepare for store.
+          AppendCodeRelativeDataAddressToOutput(instr, address);
+          break;
+        case 0x1:     // PLI: preload instructions.
+          AppendCodeRelativeCodeAddressToOutput(instr, address);
+          break;
+        case 0x3:     // Unallocated hint.
+          AppendCodeRelativeAddressToOutput(instr, address);
+          break;
+      }
+      break;
+    }
+    default:
+      VIXL_UNREACHABLE();
+  }
+
+  return 6;
+}
+
+
+int Disassembler::SubstituteShiftField(const Instruction* instr,
+                                       const char* format) {
+  VIXL_ASSERT(format[0] == 'N');
+  VIXL_ASSERT(instr->ShiftDP() <= 0x3);
+
+  switch (format[1]) {
+    case 'D': {  // HDP.
+      VIXL_ASSERT(instr->ShiftDP() != ROR);
+      VIXL_FALLTHROUGH();
+    }
+    case 'L': {  // HLo.
+      if (instr->ImmDPShift() != 0) {
+        const char* shift_type[] = {"lsl", "lsr", "asr", "ror"};
+        AppendToOutput(", %s #%" PRId32, shift_type[instr->ShiftDP()],
+                       instr->ImmDPShift());
+      }
+      return 3;
+    }
+    default:
+      VIXL_UNIMPLEMENTED();
+      return 0;
+  }
+}
+
+
+int Disassembler::SubstituteConditionField(const Instruction* instr,
+                                           const char* format) {
+  VIXL_ASSERT(format[0] == 'C');
+  const char* condition_code[] = { "eq", "ne", "hs", "lo",
+                                   "mi", "pl", "vs", "vc",
+                                   "hi", "ls", "ge", "lt",
+                                   "gt", "le", "al", "nv" };
+  int cond;
+  switch (format[1]) {
+    case 'B': cond = instr->ConditionBranch(); break;
+    case 'I': {
+      cond = InvertCondition(static_cast<Condition>(instr->Condition()));
+      break;
+    }
+    default: cond = instr->Condition();
+  }
+  AppendToOutput("%s", condition_code[cond]);
+  return 4;
+}
+
+
+int Disassembler::SubstitutePCRelAddressField(const Instruction* instr,
+                                              const char* format) {
+  VIXL_ASSERT((strcmp(format, "AddrPCRelByte") == 0) ||   // Used by `adr`.
+              (strcmp(format, "AddrPCRelPage") == 0));    // Used by `adrp`.
+
+  int64_t offset = instr->ImmPCRel();
+
+  // Compute the target address based on the effective address (after applying
+  // code_address_offset). This is required for correct behaviour of adrp.
+  const Instruction* base = instr + code_address_offset();
+  if (format[9] == 'P') {
+    offset *= kPageSize;
+    base = AlignDown(base, kPageSize);
+  }
+  // Strip code_address_offset before printing, so we can use the
+  // semantically-correct AppendCodeRelativeAddressToOutput.
+  const void* target =
+      reinterpret_cast<const void*>(base + offset - code_address_offset());
+
+  AppendPCRelativeOffsetToOutput(instr, offset);
+  AppendToOutput(" ");
+  AppendCodeRelativeAddressToOutput(instr, target);
+  return 13;
+}
+
+
+int Disassembler::SubstituteBranchTargetField(const Instruction* instr,
+                                              const char* format) {
+  VIXL_ASSERT(strncmp(format, "TImm", 4) == 0);
+
+  int64_t offset = 0;
+  switch (format[5]) {
+    // BImmUncn - unconditional branch immediate.
+    case 'n': offset = instr->ImmUncondBranch(); break;
+    // BImmCond - conditional branch immediate.
+    case 'o': offset = instr->ImmCondBranch(); break;
+    // BImmCmpa - compare and branch immediate.
+    case 'm': offset = instr->ImmCmpBranch(); break;
+    // BImmTest - test and branch immediate.
+    case 'e': offset = instr->ImmTestBranch(); break;
+    default: VIXL_UNIMPLEMENTED();
+  }
+  offset <<= kInstructionSizeLog2;
+  const void* target_address = reinterpret_cast<const void*>(instr + offset);
+  VIXL_STATIC_ASSERT(sizeof(*instr) == 1);
+
+  AppendPCRelativeOffsetToOutput(instr, offset);
+  AppendToOutput(" ");
+  AppendCodeRelativeCodeAddressToOutput(instr, target_address);
+
+  return 8;
+}
+
+
+int Disassembler::SubstituteExtendField(const Instruction* instr,
+                                        const char* format) {
+  VIXL_ASSERT(strncmp(format, "Ext", 3) == 0);
+  VIXL_ASSERT(instr->ExtendMode() <= 7);
+  USE(format);
+
+  const char* extend_mode[] = { "uxtb", "uxth", "uxtw", "uxtx",
+                                "sxtb", "sxth", "sxtw", "sxtx" };
+
+  // If rd or rn is SP, uxtw on 32-bit registers and uxtx on 64-bit
+  // registers becomes lsl.
+  if (((instr->Rd() == kZeroRegCode) || (instr->Rn() == kZeroRegCode)) &&
+      (((instr->ExtendMode() == UXTW) && (instr->SixtyFourBits() == 0)) ||
+       (instr->ExtendMode() == UXTX))) {
+    if (instr->ImmExtendShift() > 0) {
+      AppendToOutput(", lsl #%" PRId32, instr->ImmExtendShift());
+    }
+  } else {
+    AppendToOutput(", %s", extend_mode[instr->ExtendMode()]);
+    if (instr->ImmExtendShift() > 0) {
+      AppendToOutput(" #%" PRId32, instr->ImmExtendShift());
+    }
+  }
+  return 3;
+}
+
+
+int Disassembler::SubstituteLSRegOffsetField(const Instruction* instr,
+                                             const char* format) {
+  VIXL_ASSERT(strncmp(format, "Offsetreg", 9) == 0);
+  const char* extend_mode[] = { "undefined", "undefined", "uxtw", "lsl",
+                                "undefined", "undefined", "sxtw", "sxtx" };
+  USE(format);
+
+  unsigned shift = instr->ImmShiftLS();
+  Extend ext = static_cast<Extend>(instr->ExtendMode());
+  char reg_type = ((ext == UXTW) || (ext == SXTW)) ? 'w' : 'x';
+
+  unsigned rm = instr->Rm();
+  if (rm == kZeroRegCode) {
+    AppendToOutput("%czr", reg_type);
+  } else {
+    AppendToOutput("%c%d", reg_type, rm);
+  }
+
+  // Extend mode UXTX is an alias for shift mode LSL here.
+  if (!((ext == UXTX) && (shift == 0))) {
+    AppendToOutput(", %s", extend_mode[ext]);
+    if (shift != 0) {
+      AppendToOutput(" #%d", instr->SizeLS());
+    }
+  }
+  return 9;
+}
+
+
+int Disassembler::SubstitutePrefetchField(const Instruction* instr,
+                                          const char* format) {
+  VIXL_ASSERT(format[0] == 'P');
+  USE(format);
+
+  static const char* hints[] = {"ld", "li", "st"};
+  static const char* stream_options[] = {"keep", "strm"};
+
+  unsigned hint = instr->PrefetchHint();
+  unsigned target = instr->PrefetchTarget() + 1;
+  unsigned stream = instr->PrefetchStream();
+
+  if ((hint >= (sizeof(hints) / sizeof(hints[0]))) || (target > 3)) {
+    // Unallocated prefetch operations.
+    int prefetch_mode = instr->ImmPrefetchOperation();
+    AppendToOutput("#0b%c%c%c%c%c",
+                   (prefetch_mode & (1 << 4)) ? '1' : '0',
+                   (prefetch_mode & (1 << 3)) ? '1' : '0',
+                   (prefetch_mode & (1 << 2)) ? '1' : '0',
+                   (prefetch_mode & (1 << 1)) ? '1' : '0',
+                   (prefetch_mode & (1 << 0)) ? '1' : '0');
+  } else {
+    VIXL_ASSERT(stream < (sizeof(stream_options) / sizeof(stream_options[0])));
+    AppendToOutput("p%sl%d%s", hints[hint], target, stream_options[stream]);
+  }
+  return 6;
+}
+
+int Disassembler::SubstituteBarrierField(const Instruction* instr,
+                                         const char* format) {
+  VIXL_ASSERT(format[0] == 'M');
+  USE(format);
+
+  static const char* options[4][4] = {
+    { "sy (0b0000)", "oshld", "oshst", "osh" },
+    { "sy (0b0100)", "nshld", "nshst", "nsh" },
+    { "sy (0b1000)", "ishld", "ishst", "ish" },
+    { "sy (0b1100)", "ld", "st", "sy" }
+  };
+  int domain = instr->ImmBarrierDomain();
+  int type = instr->ImmBarrierType();
+
+  AppendToOutput("%s", options[domain][type]);
+  return 1;
+}
+
+int Disassembler::SubstituteSysOpField(const Instruction* instr,
+                                       const char* format) {
+  VIXL_ASSERT(format[0] == 'G');
+  int op = -1;
+  switch (format[1]) {
+    case '1': op = instr->SysOp1(); break;
+    case '2': op = instr->SysOp2(); break;
+    default:
+      VIXL_UNREACHABLE();
+  }
+  AppendToOutput("#%d", op);
+  return 2;
+}
+
+int Disassembler::SubstituteCrField(const Instruction* instr,
+                                    const char* format) {
+  VIXL_ASSERT(format[0] == 'K');
+  int cr = -1;
+  switch (format[1]) {
+    case 'n': cr = instr->CRn(); break;
+    case 'm': cr = instr->CRm(); break;
+    default:
+      VIXL_UNREACHABLE();
+  }
+  AppendToOutput("C%d", cr);
+  return 2;
+}
+
+void Disassembler::ResetOutput() {
+  buffer_pos_ = 0;
+  buffer_[buffer_pos_] = 0;
+}
+
+
+void Disassembler::AppendToOutput(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  buffer_pos_ += vsnprintf(&buffer_[buffer_pos_], buffer_size_ - buffer_pos_,
+          format, args);
+  va_end(args);
+}
+
+
+void PrintDisassembler::ProcessOutput(const Instruction* instr) {
+  fprintf(stream_, "0x%016" PRIx64 "  %08" PRIx32 "\t\t%s\n",
+          reinterpret_cast<uint64_t>(instr),
+          instr->InstructionBits(),
+          GetOutput());
+}
+
+}  // namespace vixl
diff --git a/disas/libvixl/a64/disasm-a64.h b/disas/libvixl/vixl/a64/disasm-a64.h
index ddfe98be19..930df6ea6a 100644
--- a/disas/libvixl/a64/disasm-a64.h
+++ b/disas/libvixl/vixl/a64/disasm-a64.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2015, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -27,11 +27,11 @@
 #ifndef VIXL_A64_DISASM_A64_H
 #define VIXL_A64_DISASM_A64_H
 
-#include "globals.h"
-#include "utils.h"
-#include "instructions-a64.h"
-#include "decoder-a64.h"
-#include "assembler-a64.h"
+#include "vixl/globals.h"
+#include "vixl/utils.h"
+#include "vixl/a64/instructions-a64.h"
+#include "vixl/a64/decoder-a64.h"
+#include "vixl/a64/assembler-a64.h"
 
 namespace vixl {
 
@@ -55,6 +55,7 @@ class Disassembler: public DecoderVisitor {
   // customize the disassembly output.
 
   // Prints the name of a register.
+  // TODO: This currently doesn't allow renaming of V registers.
   virtual void AppendRegisterNameToOutput(const Instruction* instr,
                                           const CPURegister& reg);
 
@@ -122,7 +123,8 @@ class Disassembler: public DecoderVisitor {
   int SubstituteLSRegOffsetField(const Instruction* instr, const char* format);
   int SubstitutePrefetchField(const Instruction* instr, const char* format);
   int SubstituteBarrierField(const Instruction* instr, const char* format);
-
+  int SubstituteSysOpField(const Instruction* instr, const char* format);
+  int SubstituteCrField(const Instruction* instr, const char* format);
   bool RdIsZROrSP(const Instruction* instr) const {
     return (instr->Rd() == kZeroRegCode);
   }
@@ -163,7 +165,6 @@ class Disassembler: public DecoderVisitor {
 class PrintDisassembler: public Disassembler {
  public:
   explicit PrintDisassembler(FILE* stream) : stream_(stream) { }
-  virtual ~PrintDisassembler() { }
 
  protected:
   virtual void ProcessOutput(const Instruction* instr);
diff --git a/disas/libvixl/vixl/a64/instructions-a64.cc b/disas/libvixl/vixl/a64/instructions-a64.cc
new file mode 100644
index 0000000000..33992f88a4
--- /dev/null
+++ b/disas/libvixl/vixl/a64/instructions-a64.cc
@@ -0,0 +1,622 @@
+// 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.
+
+#include "vixl/a64/instructions-a64.h"
+#include "vixl/a64/assembler-a64.h"
+
+namespace vixl {
+
+
+// Floating-point infinity values.
+const float16 kFP16PositiveInfinity = 0x7c00;
+const float16 kFP16NegativeInfinity = 0xfc00;
+const float kFP32PositiveInfinity = rawbits_to_float(0x7f800000);
+const float kFP32NegativeInfinity = rawbits_to_float(0xff800000);
+const double kFP64PositiveInfinity =
+    rawbits_to_double(UINT64_C(0x7ff0000000000000));
+const double kFP64NegativeInfinity =
+    rawbits_to_double(UINT64_C(0xfff0000000000000));
+
+
+// The default NaN values (for FPCR.DN=1).
+const double kFP64DefaultNaN = rawbits_to_double(UINT64_C(0x7ff8000000000000));
+const float kFP32DefaultNaN = rawbits_to_float(0x7fc00000);
+const float16 kFP16DefaultNaN = 0x7e00;
+
+
+static uint64_t RotateRight(uint64_t value,
+                            unsigned int rotate,
+                            unsigned int width) {
+  VIXL_ASSERT(width <= 64);
+  rotate &= 63;
+  return ((value & ((UINT64_C(1) << rotate) - 1)) <<
+          (width - rotate)) | (value >> rotate);
+}
+
+
+static uint64_t RepeatBitsAcrossReg(unsigned reg_size,
+                                    uint64_t value,
+                                    unsigned width) {
+  VIXL_ASSERT((width == 2) || (width == 4) || (width == 8) || (width == 16) ||
+              (width == 32));
+  VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize));
+  uint64_t result = value & ((UINT64_C(1) << width) - 1);
+  for (unsigned i = width; i < reg_size; i *= 2) {
+    result |= (result << i);
+  }
+  return result;
+}
+
+
+bool Instruction::IsLoad() const {
+  if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) {
+    return false;
+  }
+
+  if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) {
+    return Mask(LoadStorePairLBit) != 0;
+  } else {
+    LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreMask));
+    switch (op) {
+      case LDRB_w:
+      case LDRH_w:
+      case LDR_w:
+      case LDR_x:
+      case LDRSB_w:
+      case LDRSB_x:
+      case LDRSH_w:
+      case LDRSH_x:
+      case LDRSW_x:
+      case LDR_b:
+      case LDR_h:
+      case LDR_s:
+      case LDR_d:
+      case LDR_q: return true;
+      default: return false;
+    }
+  }
+}
+
+
+bool Instruction::IsStore() const {
+  if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) {
+    return false;
+  }
+
+  if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) {
+    return Mask(LoadStorePairLBit) == 0;
+  } else {
+    LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreMask));
+    switch (op) {
+      case STRB_w:
+      case STRH_w:
+      case STR_w:
+      case STR_x:
+      case STR_b:
+      case STR_h:
+      case STR_s:
+      case STR_d:
+      case STR_q: return true;
+      default: return false;
+    }
+  }
+}
+
+
+// Logical immediates can't encode zero, so a return value of zero is used to
+// indicate a failure case. Specifically, where the constraints on imm_s are
+// not met.
+uint64_t Instruction::ImmLogical() const {
+  unsigned reg_size = SixtyFourBits() ? kXRegSize : kWRegSize;
+  int32_t n = BitN();
+  int32_t imm_s = ImmSetBits();
+  int32_t imm_r = ImmRotate();
+
+  // An integer is constructed from the n, imm_s and imm_r bits according to
+  // the following table:
+  //
+  //  N   imms    immr    size        S             R
+  //  1  ssssss  rrrrrr    64    UInt(ssssss)  UInt(rrrrrr)
+  //  0  0sssss  xrrrrr    32    UInt(sssss)   UInt(rrrrr)
+  //  0  10ssss  xxrrrr    16    UInt(ssss)    UInt(rrrr)
+  //  0  110sss  xxxrrr     8    UInt(sss)     UInt(rrr)
+  //  0  1110ss  xxxxrr     4    UInt(ss)      UInt(rr)
+  //  0  11110s  xxxxxr     2    UInt(s)       UInt(r)
+  // (s bits must not be all set)
+  //
+  // A pattern is constructed of size bits, where the least significant S+1
+  // bits are set. The pattern is rotated right by R, and repeated across a
+  // 32 or 64-bit value, depending on destination register width.
+  //
+
+  if (n == 1) {
+    if (imm_s == 0x3f) {
+      return 0;
+    }
+    uint64_t bits = (UINT64_C(1) << (imm_s + 1)) - 1;
+    return RotateRight(bits, imm_r, 64);
+  } else {
+    if ((imm_s >> 1) == 0x1f) {
+      return 0;
+    }
+    for (int width = 0x20; width >= 0x2; width >>= 1) {
+      if ((imm_s & width) == 0) {
+        int mask = width - 1;
+        if ((imm_s & mask) == mask) {
+          return 0;
+        }
+        uint64_t bits = (UINT64_C(1) << ((imm_s & mask) + 1)) - 1;
+        return RepeatBitsAcrossReg(reg_size,
+                                   RotateRight(bits, imm_r & mask, width),
+                                   width);
+      }
+    }
+  }
+  VIXL_UNREACHABLE();
+  return 0;
+}
+
+
+uint32_t Instruction::ImmNEONabcdefgh() const {
+  return ImmNEONabc() << 5 | ImmNEONdefgh();
+}
+
+
+float Instruction::Imm8ToFP32(uint32_t imm8) {
+  //   Imm8: abcdefgh (8 bits)
+  // Single: aBbb.bbbc.defg.h000.0000.0000.0000.0000 (32 bits)
+  // where B is b ^ 1
+  uint32_t bits = imm8;
+  uint32_t bit7 = (bits >> 7) & 0x1;
+  uint32_t bit6 = (bits >> 6) & 0x1;
+  uint32_t bit5_to_0 = bits & 0x3f;
+  uint32_t result = (bit7 << 31) | ((32 - bit6) << 25) | (bit5_to_0 << 19);
+
+  return rawbits_to_float(result);
+}
+
+
+float Instruction::ImmFP32() const {
+  return Imm8ToFP32(ImmFP());
+}
+
+
+double Instruction::Imm8ToFP64(uint32_t imm8) {
+  //   Imm8: abcdefgh (8 bits)
+  // Double: aBbb.bbbb.bbcd.efgh.0000.0000.0000.0000
+  //         0000.0000.0000.0000.0000.0000.0000.0000 (64 bits)
+  // where B is b ^ 1
+  uint32_t bits = imm8;
+  uint64_t bit7 = (bits >> 7) & 0x1;
+  uint64_t bit6 = (bits >> 6) & 0x1;
+  uint64_t bit5_to_0 = bits & 0x3f;
+  uint64_t result = (bit7 << 63) | ((256 - bit6) << 54) | (bit5_to_0 << 48);
+
+  return rawbits_to_double(result);
+}
+
+
+double Instruction::ImmFP64() const {
+  return Imm8ToFP64(ImmFP());
+}
+
+
+float Instruction::ImmNEONFP32() const {
+  return Imm8ToFP32(ImmNEONabcdefgh());
+}
+
+
+double Instruction::ImmNEONFP64() const {
+  return Imm8ToFP64(ImmNEONabcdefgh());
+}
+
+
+unsigned CalcLSDataSize(LoadStoreOp op) {
+  VIXL_ASSERT((LSSize_offset + LSSize_width) == (kInstructionSize * 8));
+  unsigned size = static_cast<Instr>(op) >> LSSize_offset;
+  if ((op & LSVector_mask) != 0) {
+    // Vector register memory operations encode the access size in the "size"
+    // and "opc" fields.
+    if ((size == 0) && ((op & LSOpc_mask) >> LSOpc_offset) >= 2) {
+      size = kQRegSizeInBytesLog2;
+    }
+  }
+  return size;
+}
+
+
+unsigned CalcLSPairDataSize(LoadStorePairOp op) {
+  VIXL_STATIC_ASSERT(kXRegSizeInBytes == kDRegSizeInBytes);
+  VIXL_STATIC_ASSERT(kWRegSizeInBytes == kSRegSizeInBytes);
+  switch (op) {
+    case STP_q:
+    case LDP_q: return kQRegSizeInBytesLog2;
+    case STP_x:
+    case LDP_x:
+    case STP_d:
+    case LDP_d: return kXRegSizeInBytesLog2;
+    default: return kWRegSizeInBytesLog2;
+  }
+}
+
+
+int Instruction::ImmBranchRangeBitwidth(ImmBranchType branch_type) {
+  switch (branch_type) {
+    case UncondBranchType:
+      return ImmUncondBranch_width;
+    case CondBranchType:
+      return ImmCondBranch_width;
+    case CompareBranchType:
+      return ImmCmpBranch_width;
+    case TestBranchType:
+      return ImmTestBranch_width;
+    default:
+      VIXL_UNREACHABLE();
+      return 0;
+  }
+}
+
+
+int32_t Instruction::ImmBranchForwardRange(ImmBranchType branch_type) {
+  int32_t encoded_max = 1 << (ImmBranchRangeBitwidth(branch_type) - 1);
+  return encoded_max * kInstructionSize;
+}
+
+
+bool Instruction::IsValidImmPCOffset(ImmBranchType branch_type,
+                                     int64_t offset) {
+  return is_intn(ImmBranchRangeBitwidth(branch_type), offset);
+}
+
+
+const Instruction* Instruction::ImmPCOffsetTarget() const {
+  const Instruction * base = this;
+  ptrdiff_t offset;
+  if (IsPCRelAddressing()) {
+    // ADR and ADRP.
+    offset = ImmPCRel();
+    if (Mask(PCRelAddressingMask) == ADRP) {
+      base = AlignDown(base, kPageSize);
+      offset *= kPageSize;
+    } else {
+      VIXL_ASSERT(Mask(PCRelAddressingMask) == ADR);
+    }
+  } else {
+    // All PC-relative branches.
+    VIXL_ASSERT(BranchType() != UnknownBranchType);
+    // Relative branch offsets are instruction-size-aligned.
+    offset = ImmBranch() << kInstructionSizeLog2;
+  }
+  return base + offset;
+}
+
+
+int Instruction::ImmBranch() const {
+  switch (BranchType()) {
+    case CondBranchType: return ImmCondBranch();
+    case UncondBranchType: return ImmUncondBranch();
+    case CompareBranchType: return ImmCmpBranch();
+    case TestBranchType: return ImmTestBranch();
+    default: VIXL_UNREACHABLE();
+  }
+  return 0;
+}
+
+
+void Instruction::SetImmPCOffsetTarget(const Instruction* target) {
+  if (IsPCRelAddressing()) {
+    SetPCRelImmTarget(target);
+  } else {
+    SetBranchImmTarget(target);
+  }
+}
+
+
+void Instruction::SetPCRelImmTarget(const Instruction* target) {
+  ptrdiff_t imm21;
+  if ((Mask(PCRelAddressingMask) == ADR)) {
+    imm21 = target - this;
+  } else {
+    VIXL_ASSERT(Mask(PCRelAddressingMask) == ADRP);
+    uintptr_t this_page = reinterpret_cast<uintptr_t>(this) / kPageSize;
+    uintptr_t target_page = reinterpret_cast<uintptr_t>(target) / kPageSize;
+    imm21 = target_page - this_page;
+  }
+  Instr imm = Assembler::ImmPCRelAddress(static_cast<int32_t>(imm21));
+
+  SetInstructionBits(Mask(~ImmPCRel_mask) | imm);
+}
+
+
+void Instruction::SetBranchImmTarget(const Instruction* target) {
+  VIXL_ASSERT(((target - this) & 3) == 0);
+  Instr branch_imm = 0;
+  uint32_t imm_mask = 0;
+  int offset = static_cast<int>((target - this) >> kInstructionSizeLog2);
+  switch (BranchType()) {
+    case CondBranchType: {
+      branch_imm = Assembler::ImmCondBranch(offset);
+      imm_mask = ImmCondBranch_mask;
+      break;
+    }
+    case UncondBranchType: {
+      branch_imm = Assembler::ImmUncondBranch(offset);
+      imm_mask = ImmUncondBranch_mask;
+      break;
+    }
+    case CompareBranchType: {
+      branch_imm = Assembler::ImmCmpBranch(offset);
+      imm_mask = ImmCmpBranch_mask;
+      break;
+    }
+    case TestBranchType: {
+      branch_imm = Assembler::ImmTestBranch(offset);
+      imm_mask = ImmTestBranch_mask;
+      break;
+    }
+    default: VIXL_UNREACHABLE();
+  }
+  SetInstructionBits(Mask(~imm_mask) | branch_imm);
+}
+
+
+void Instruction::SetImmLLiteral(const Instruction* source) {
+  VIXL_ASSERT(IsWordAligned(source));
+  ptrdiff_t offset = (source - this) >> kLiteralEntrySizeLog2;
+  Instr imm = Assembler::ImmLLiteral(static_cast<int>(offset));
+  Instr mask = ImmLLiteral_mask;
+
+  SetInstructionBits(Mask(~mask) | imm);
+}
+
+
+VectorFormat VectorFormatHalfWidth(const VectorFormat vform) {
+  VIXL_ASSERT(vform == kFormat8H || vform == kFormat4S || vform == kFormat2D ||
+              vform == kFormatH || vform == kFormatS || vform == kFormatD);
+  switch (vform) {
+    case kFormat8H: return kFormat8B;
+    case kFormat4S: return kFormat4H;
+    case kFormat2D: return kFormat2S;
+    case kFormatH:  return kFormatB;
+    case kFormatS:  return kFormatH;
+    case kFormatD:  return kFormatS;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+
+VectorFormat VectorFormatDoubleWidth(const VectorFormat vform) {
+  VIXL_ASSERT(vform == kFormat8B || vform == kFormat4H || vform == kFormat2S ||
+              vform == kFormatB || vform == kFormatH || vform == kFormatS);
+  switch (vform) {
+    case kFormat8B: return kFormat8H;
+    case kFormat4H: return kFormat4S;
+    case kFormat2S: return kFormat2D;
+    case kFormatB:  return kFormatH;
+    case kFormatH:  return kFormatS;
+    case kFormatS:  return kFormatD;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+
+VectorFormat VectorFormatFillQ(const VectorFormat vform) {
+  switch (vform) {
+    case kFormatB:
+    case kFormat8B:
+    case kFormat16B: return kFormat16B;
+    case kFormatH:
+    case kFormat4H:
+    case kFormat8H:  return kFormat8H;
+    case kFormatS:
+    case kFormat2S:
+    case kFormat4S:  return kFormat4S;
+    case kFormatD:
+    case kFormat1D:
+    case kFormat2D:  return kFormat2D;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+VectorFormat VectorFormatHalfWidthDoubleLanes(const VectorFormat vform) {
+  switch (vform) {
+    case kFormat4H: return kFormat8B;
+    case kFormat8H: return kFormat16B;
+    case kFormat2S: return kFormat4H;
+    case kFormat4S: return kFormat8H;
+    case kFormat1D: return kFormat2S;
+    case kFormat2D: return kFormat4S;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+VectorFormat VectorFormatDoubleLanes(const VectorFormat vform) {
+  VIXL_ASSERT(vform == kFormat8B || vform == kFormat4H || vform == kFormat2S);
+  switch (vform) {
+    case kFormat8B: return kFormat16B;
+    case kFormat4H: return kFormat8H;
+    case kFormat2S: return kFormat4S;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+
+VectorFormat VectorFormatHalfLanes(const VectorFormat vform) {
+  VIXL_ASSERT(vform == kFormat16B || vform == kFormat8H || vform == kFormat4S);
+  switch (vform) {
+    case kFormat16B: return kFormat8B;
+    case kFormat8H: return kFormat4H;
+    case kFormat4S: return kFormat2S;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+
+VectorFormat ScalarFormatFromLaneSize(int laneSize) {
+  switch (laneSize) {
+    case 8:  return kFormatB;
+    case 16: return kFormatH;
+    case 32: return kFormatS;
+    case 64: return kFormatD;
+    default: VIXL_UNREACHABLE(); return kFormatUndefined;
+  }
+}
+
+
+unsigned RegisterSizeInBitsFromFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormatB: return kBRegSize;
+    case kFormatH: return kHRegSize;
+    case kFormatS: return kSRegSize;
+    case kFormatD: return kDRegSize;
+    case kFormat8B:
+    case kFormat4H:
+    case kFormat2S:
+    case kFormat1D: return kDRegSize;
+    default: return kQRegSize;
+  }
+}
+
+
+unsigned RegisterSizeInBytesFromFormat(VectorFormat vform) {
+  return RegisterSizeInBitsFromFormat(vform) / 8;
+}
+
+
+unsigned LaneSizeInBitsFromFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormatB:
+    case kFormat8B:
+    case kFormat16B: return 8;
+    case kFormatH:
+    case kFormat4H:
+    case kFormat8H: return 16;
+    case kFormatS:
+    case kFormat2S:
+    case kFormat4S: return 32;
+    case kFormatD:
+    case kFormat1D:
+    case kFormat2D: return 64;
+    default: VIXL_UNREACHABLE(); return 0;
+  }
+}
+
+
+int LaneSizeInBytesFromFormat(VectorFormat vform) {
+  return LaneSizeInBitsFromFormat(vform) / 8;
+}
+
+
+int LaneSizeInBytesLog2FromFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormatB:
+    case kFormat8B:
+    case kFormat16B: return 0;
+    case kFormatH:
+    case kFormat4H:
+    case kFormat8H: return 1;
+    case kFormatS:
+    case kFormat2S:
+    case kFormat4S: return 2;
+    case kFormatD:
+    case kFormat1D:
+    case kFormat2D: return 3;
+    default: VIXL_UNREACHABLE(); return 0;
+  }
+}
+
+
+int LaneCountFromFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormat16B: return 16;
+    case kFormat8B:
+    case kFormat8H: return 8;
+    case kFormat4H:
+    case kFormat4S: return 4;
+    case kFormat2S:
+    case kFormat2D: return 2;
+    case kFormat1D:
+    case kFormatB:
+    case kFormatH:
+    case kFormatS:
+    case kFormatD: return 1;
+    default: VIXL_UNREACHABLE(); return 0;
+  }
+}
+
+
+int MaxLaneCountFromFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormatB:
+    case kFormat8B:
+    case kFormat16B: return 16;
+    case kFormatH:
+    case kFormat4H:
+    case kFormat8H: return 8;
+    case kFormatS:
+    case kFormat2S:
+    case kFormat4S: return 4;
+    case kFormatD:
+    case kFormat1D:
+    case kFormat2D: return 2;
+    default: VIXL_UNREACHABLE(); return 0;
+  }
+}
+
+
+// Does 'vform' indicate a vector format or a scalar format?
+bool IsVectorFormat(VectorFormat vform) {
+  VIXL_ASSERT(vform != kFormatUndefined);
+  switch (vform) {
+    case kFormatB:
+    case kFormatH:
+    case kFormatS:
+    case kFormatD: return false;
+    default: return true;
+  }
+}
+
+
+int64_t MaxIntFromFormat(VectorFormat vform) {
+  return INT64_MAX >> (64 - LaneSizeInBitsFromFormat(vform));
+}
+
+
+int64_t MinIntFromFormat(VectorFormat vform) {
+  return INT64_MIN >> (64 - LaneSizeInBitsFromFormat(vform));
+}
+
+
+uint64_t MaxUintFromFormat(VectorFormat vform) {
+  return UINT64_MAX >> (64 - LaneSizeInBitsFromFormat(vform));
+}
+}  // namespace vixl
+
diff --git a/disas/libvixl/vixl/a64/instructions-a64.h b/disas/libvixl/vixl/a64/instructions-a64.h
new file mode 100644
index 0000000000..7e0dbae36a
--- /dev/null
+++ b/disas/libvixl/vixl/a64/instructions-a64.h
@@ -0,0 +1,757 @@
+// 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_INSTRUCTIONS_A64_H_
+#define VIXL_A64_INSTRUCTIONS_A64_H_
+
+#include "vixl/globals.h"
+#include "vixl/utils.h"
+#include "vixl/a64/constants-a64.h"
+
+namespace vixl {
+// ISA constants. --------------------------------------------------------------
+
+typedef uint32_t Instr;
+const unsigned kInstructionSize = 4;
+const unsigned kInstructionSizeLog2 = 2;
+const unsigned kLiteralEntrySize = 4;
+const unsigned kLiteralEntrySizeLog2 = 2;
+const unsigned kMaxLoadLiteralRange = 1 * MBytes;
+
+// This is the nominal page size (as used by the adrp instruction); the actual
+// size of the memory pages allocated by the kernel is likely to differ.
+const unsigned kPageSize = 4 * KBytes;
+const unsigned kPageSizeLog2 = 12;
+
+const unsigned kBRegSize = 8;
+const unsigned kBRegSizeLog2 = 3;
+const unsigned kBRegSizeInBytes = kBRegSize / 8;
+const unsigned kBRegSizeInBytesLog2 = kBRegSizeLog2 - 3;
+const unsigned kHRegSize = 16;
+const unsigned kHRegSizeLog2 = 4;
+const unsigned kHRegSizeInBytes = kHRegSize / 8;
+const unsigned kHRegSizeInBytesLog2 = kHRegSizeLog2 - 3;
+const unsigned kWRegSize = 32;
+const unsigned kWRegSizeLog2 = 5;
+const unsigned kWRegSizeInBytes = kWRegSize / 8;
+const unsigned kWRegSizeInBytesLog2 = kWRegSizeLog2 - 3;
+const unsigned kXRegSize = 64;
+const unsigned kXRegSizeLog2 = 6;
+const unsigned kXRegSizeInBytes = kXRegSize / 8;
+const unsigned kXRegSizeInBytesLog2 = kXRegSizeLog2 - 3;
+const unsigned kSRegSize = 32;
+const unsigned kSRegSizeLog2 = 5;
+const unsigned kSRegSizeInBytes = kSRegSize / 8;
+const unsigned kSRegSizeInBytesLog2 = kSRegSizeLog2 - 3;
+const unsigned kDRegSize = 64;
+const unsigned kDRegSizeLog2 = 6;
+const unsigned kDRegSizeInBytes = kDRegSize / 8;
+const unsigned kDRegSizeInBytesLog2 = kDRegSizeLog2 - 3;
+const unsigned kQRegSize = 128;
+const unsigned kQRegSizeLog2 = 7;
+const unsigned kQRegSizeInBytes = kQRegSize / 8;
+const unsigned kQRegSizeInBytesLog2 = kQRegSizeLog2 - 3;
+const uint64_t kWRegMask = UINT64_C(0xffffffff);
+const uint64_t kXRegMask = UINT64_C(0xffffffffffffffff);
+const uint64_t kSRegMask = UINT64_C(0xffffffff);
+const uint64_t kDRegMask = UINT64_C(0xffffffffffffffff);
+const uint64_t kSSignMask = UINT64_C(0x80000000);
+const uint64_t kDSignMask = UINT64_C(0x8000000000000000);
+const uint64_t kWSignMask = UINT64_C(0x80000000);
+const uint64_t kXSignMask = UINT64_C(0x8000000000000000);
+const uint64_t kByteMask = UINT64_C(0xff);
+const uint64_t kHalfWordMask = UINT64_C(0xffff);
+const uint64_t kWordMask = UINT64_C(0xffffffff);
+const uint64_t kXMaxUInt = UINT64_C(0xffffffffffffffff);
+const uint64_t kWMaxUInt = UINT64_C(0xffffffff);
+const int64_t kXMaxInt = INT64_C(0x7fffffffffffffff);
+const int64_t kXMinInt = INT64_C(0x8000000000000000);
+const int32_t kWMaxInt = INT32_C(0x7fffffff);
+const int32_t kWMinInt = INT32_C(0x80000000);
+const unsigned kLinkRegCode = 30;
+const unsigned kZeroRegCode = 31;
+const unsigned kSPRegInternalCode = 63;
+const unsigned kRegCodeMask = 0x1f;
+
+const unsigned kAddressTagOffset = 56;
+const unsigned kAddressTagWidth = 8;
+const uint64_t kAddressTagMask =
+    ((UINT64_C(1) << kAddressTagWidth) - 1) << kAddressTagOffset;
+VIXL_STATIC_ASSERT(kAddressTagMask == UINT64_C(0xff00000000000000));
+
+// AArch64 floating-point specifics. These match IEEE-754.
+const unsigned kDoubleMantissaBits = 52;
+const unsigned kDoubleExponentBits = 11;
+const unsigned kFloatMantissaBits = 23;
+const unsigned kFloatExponentBits = 8;
+const unsigned kFloat16MantissaBits = 10;
+const unsigned kFloat16ExponentBits = 5;
+
+// Floating-point infinity values.
+extern const float16 kFP16PositiveInfinity;
+extern const float16 kFP16NegativeInfinity;
+extern const float kFP32PositiveInfinity;
+extern const float kFP32NegativeInfinity;
+extern const double kFP64PositiveInfinity;
+extern const double kFP64NegativeInfinity;
+
+// The default NaN values (for FPCR.DN=1).
+extern const float16 kFP16DefaultNaN;
+extern const float kFP32DefaultNaN;
+extern const double kFP64DefaultNaN;
+
+unsigned CalcLSDataSize(LoadStoreOp op);
+unsigned CalcLSPairDataSize(LoadStorePairOp op);
+
+enum ImmBranchType {
+  UnknownBranchType = 0,
+  CondBranchType    = 1,
+  UncondBranchType  = 2,
+  CompareBranchType = 3,
+  TestBranchType    = 4
+};
+
+enum AddrMode {
+  Offset,
+  PreIndex,
+  PostIndex
+};
+
+enum FPRounding {
+  // The first four values are encodable directly by FPCR<RMode>.
+  FPTieEven = 0x0,
+  FPPositiveInfinity = 0x1,
+  FPNegativeInfinity = 0x2,
+  FPZero = 0x3,
+
+  // The final rounding modes are only available when explicitly specified by
+  // the instruction (such as with fcvta). It cannot be set in FPCR.
+  FPTieAway,
+  FPRoundOdd
+};
+
+enum Reg31Mode {
+  Reg31IsStackPointer,
+  Reg31IsZeroRegister
+};
+
+// Instructions. ---------------------------------------------------------------
+
+class Instruction {
+ public:
+  Instr InstructionBits() const {
+    return *(reinterpret_cast<const Instr*>(this));
+  }
+
+  void SetInstructionBits(Instr new_instr) {
+    *(reinterpret_cast<Instr*>(this)) = new_instr;
+  }
+
+  int Bit(int pos) const {
+    return (InstructionBits() >> pos) & 1;
+  }
+
+  uint32_t Bits(int msb, int lsb) const {
+    return unsigned_bitextract_32(msb, lsb, InstructionBits());
+  }
+
+  int32_t SignedBits(int msb, int lsb) const {
+    int32_t bits = *(reinterpret_cast<const int32_t*>(this));
+    return signed_bitextract_32(msb, lsb, bits);
+  }
+
+  Instr Mask(uint32_t mask) const {
+    return InstructionBits() & mask;
+  }
+
+  #define DEFINE_GETTER(Name, HighBit, LowBit, Func)             \
+  int32_t Name() const { return Func(HighBit, LowBit); }
+  INSTRUCTION_FIELDS_LIST(DEFINE_GETTER)
+  #undef DEFINE_GETTER
+
+  // ImmPCRel is a compound field (not present in INSTRUCTION_FIELDS_LIST),
+  // formed from ImmPCRelLo and ImmPCRelHi.
+  int ImmPCRel() const {
+    int offset =
+        static_cast<int>((ImmPCRelHi() << ImmPCRelLo_width) | ImmPCRelLo());
+    int width = ImmPCRelLo_width + ImmPCRelHi_width;
+    return signed_bitextract_32(width - 1, 0, offset);
+  }
+
+  uint64_t ImmLogical() const;
+  unsigned ImmNEONabcdefgh() const;
+  float ImmFP32() const;
+  double ImmFP64() const;
+  float ImmNEONFP32() const;
+  double ImmNEONFP64() const;
+
+  unsigned SizeLS() const {
+    return CalcLSDataSize(static_cast<LoadStoreOp>(Mask(LoadStoreMask)));
+  }
+
+  unsigned SizeLSPair() const {
+    return CalcLSPairDataSize(
+        static_cast<LoadStorePairOp>(Mask(LoadStorePairMask)));
+  }
+
+  int NEONLSIndex(int access_size_shift) const {
+    int64_t q = NEONQ();
+    int64_t s = NEONS();
+    int64_t size = NEONLSSize();
+    int64_t index = (q << 3) | (s << 2) | size;
+    return static_cast<int>(index >> access_size_shift);
+  }
+
+  // Helpers.
+  bool IsCondBranchImm() const {
+    return Mask(ConditionalBranchFMask) == ConditionalBranchFixed;
+  }
+
+  bool IsUncondBranchImm() const {
+    return Mask(UnconditionalBranchFMask) == UnconditionalBranchFixed;
+  }
+
+  bool IsCompareBranch() const {
+    return Mask(CompareBranchFMask) == CompareBranchFixed;
+  }
+
+  bool IsTestBranch() const {
+    return Mask(TestBranchFMask) == TestBranchFixed;
+  }
+
+  bool IsImmBranch() const {
+    return BranchType() != UnknownBranchType;
+  }
+
+  bool IsPCRelAddressing() const {
+    return Mask(PCRelAddressingFMask) == PCRelAddressingFixed;
+  }
+
+  bool IsLogicalImmediate() const {
+    return Mask(LogicalImmediateFMask) == LogicalImmediateFixed;
+  }
+
+  bool IsAddSubImmediate() const {
+    return Mask(AddSubImmediateFMask) == AddSubImmediateFixed;
+  }
+
+  bool IsAddSubExtended() const {
+    return Mask(AddSubExtendedFMask) == AddSubExtendedFixed;
+  }
+
+  bool IsLoadOrStore() const {
+    return Mask(LoadStoreAnyFMask) == LoadStoreAnyFixed;
+  }
+
+  bool IsLoad() const;
+  bool IsStore() const;
+
+  bool IsLoadLiteral() const {
+    // This includes PRFM_lit.
+    return Mask(LoadLiteralFMask) == LoadLiteralFixed;
+  }
+
+  bool IsMovn() const {
+    return (Mask(MoveWideImmediateMask) == MOVN_x) ||
+           (Mask(MoveWideImmediateMask) == MOVN_w);
+  }
+
+  static int ImmBranchRangeBitwidth(ImmBranchType branch_type);
+  static int32_t ImmBranchForwardRange(ImmBranchType branch_type);
+  static bool IsValidImmPCOffset(ImmBranchType branch_type, int64_t offset);
+
+  // Indicate whether Rd can be the stack pointer or the zero register. This
+  // does not check that the instruction actually has an Rd field.
+  Reg31Mode RdMode() const {
+    // The following instructions use sp or wsp as Rd:
+    //  Add/sub (immediate) when not setting the flags.
+    //  Add/sub (extended) when not setting the flags.
+    //  Logical (immediate) when not setting the flags.
+    // Otherwise, r31 is the zero register.
+    if (IsAddSubImmediate() || IsAddSubExtended()) {
+      if (Mask(AddSubSetFlagsBit)) {
+        return Reg31IsZeroRegister;
+      } else {
+        return Reg31IsStackPointer;
+      }
+    }
+    if (IsLogicalImmediate()) {
+      // Of the logical (immediate) instructions, only ANDS (and its aliases)
+      // can set the flags. The others can all write into sp.
+      // Note that some logical operations are not available to
+      // immediate-operand instructions, so we have to combine two masks here.
+      if (Mask(LogicalImmediateMask & LogicalOpMask) == ANDS) {
+        return Reg31IsZeroRegister;
+      } else {
+        return Reg31IsStackPointer;
+      }
+    }
+    return Reg31IsZeroRegister;
+  }
+
+  // Indicate whether Rn can be the stack pointer or the zero register. This
+  // does not check that the instruction actually has an Rn field.
+  Reg31Mode RnMode() const {
+    // The following instructions use sp or wsp as Rn:
+    //  All loads and stores.
+    //  Add/sub (immediate).
+    //  Add/sub (extended).
+    // Otherwise, r31 is the zero register.
+    if (IsLoadOrStore() || IsAddSubImmediate() || IsAddSubExtended()) {
+      return Reg31IsStackPointer;
+    }
+    return Reg31IsZeroRegister;
+  }
+
+  ImmBranchType BranchType() const {
+    if (IsCondBranchImm()) {
+      return CondBranchType;
+    } else if (IsUncondBranchImm()) {
+      return UncondBranchType;
+    } else if (IsCompareBranch()) {
+      return CompareBranchType;
+    } else if (IsTestBranch()) {
+      return TestBranchType;
+    } else {
+      return UnknownBranchType;
+    }
+  }
+
+  // Find the target of this instruction. 'this' may be a branch or a
+  // PC-relative addressing instruction.
+  const Instruction* ImmPCOffsetTarget() const;
+
+  // Patch a PC-relative offset to refer to 'target'. 'this' may be a branch or
+  // a PC-relative addressing instruction.
+  void SetImmPCOffsetTarget(const Instruction* target);
+  // Patch a literal load instruction to load from 'source'.
+  void SetImmLLiteral(const Instruction* source);
+
+  // The range of a load literal instruction, expressed as 'instr +- range'.
+  // The range is actually the 'positive' range; the branch instruction can
+  // target [instr - range - kInstructionSize, instr + range].
+  static const int kLoadLiteralImmBitwidth = 19;
+  static const int kLoadLiteralRange =
+      (1 << kLoadLiteralImmBitwidth) / 2 - kInstructionSize;
+
+  // Calculate the address of a literal referred to by a load-literal
+  // instruction, and return it as the specified type.
+  //
+  // The literal itself is safely mutable only if the backing buffer is safely
+  // mutable.
+  template <typename T>
+  T LiteralAddress() const {
+    uint64_t base_raw = reinterpret_cast<uint64_t>(this);
+    int64_t offset = ImmLLiteral() << kLiteralEntrySizeLog2;
+    uint64_t address_raw = base_raw + offset;
+
+    // Cast the address using a C-style cast. A reinterpret_cast would be
+    // appropriate, but it can't cast one integral type to another.
+    T address = (T)(address_raw);
+
+    // Assert that the address can be represented by the specified type.
+    VIXL_ASSERT((uint64_t)(address) == address_raw);
+
+    return address;
+  }
+
+  uint32_t Literal32() const {
+    uint32_t literal;
+    memcpy(&literal, LiteralAddress<const void*>(), sizeof(literal));
+    return literal;
+  }
+
+  uint64_t Literal64() const {
+    uint64_t literal;
+    memcpy(&literal, LiteralAddress<const void*>(), sizeof(literal));
+    return literal;
+  }
+
+  float LiteralFP32() const {
+    return rawbits_to_float(Literal32());
+  }
+
+  double LiteralFP64() const {
+    return rawbits_to_double(Literal64());
+  }
+
+  const Instruction* NextInstruction() const {
+    return this + kInstructionSize;
+  }
+
+  const Instruction* InstructionAtOffset(int64_t offset) const {
+    VIXL_ASSERT(IsWordAligned(this + offset));
+    return this + offset;
+  }
+
+  template<typename T> static Instruction* Cast(T src) {
+    return reinterpret_cast<Instruction*>(src);
+  }
+
+  template<typename T> static const Instruction* CastConst(T src) {
+    return reinterpret_cast<const Instruction*>(src);
+  }
+
+ private:
+  int ImmBranch() const;
+
+  static float Imm8ToFP32(uint32_t imm8);
+  static double Imm8ToFP64(uint32_t imm8);
+
+  void SetPCRelImmTarget(const Instruction* target);
+  void SetBranchImmTarget(const Instruction* target);
+};
+
+
+// Functions for handling NEON vector format information.
+enum VectorFormat {
+  kFormatUndefined = 0xffffffff,
+  kFormat8B  = NEON_8B,
+  kFormat16B = NEON_16B,
+  kFormat4H  = NEON_4H,
+  kFormat8H  = NEON_8H,
+  kFormat2S  = NEON_2S,
+  kFormat4S  = NEON_4S,
+  kFormat1D  = NEON_1D,
+  kFormat2D  = NEON_2D,
+
+  // Scalar formats. We add the scalar bit to distinguish between scalar and
+  // vector enumerations; the bit is always set in the encoding of scalar ops
+  // and always clear for vector ops. Although kFormatD and kFormat1D appear
+  // to be the same, their meaning is subtly different. The first is a scalar
+  // operation, the second a vector operation that only affects one lane.
+  kFormatB = NEON_B | NEONScalar,
+  kFormatH = NEON_H | NEONScalar,
+  kFormatS = NEON_S | NEONScalar,
+  kFormatD = NEON_D | NEONScalar
+};
+
+VectorFormat VectorFormatHalfWidth(const VectorFormat vform);
+VectorFormat VectorFormatDoubleWidth(const VectorFormat vform);
+VectorFormat VectorFormatDoubleLanes(const VectorFormat vform);
+VectorFormat VectorFormatHalfLanes(const VectorFormat vform);
+VectorFormat ScalarFormatFromLaneSize(int lanesize);
+VectorFormat VectorFormatHalfWidthDoubleLanes(const VectorFormat vform);
+VectorFormat VectorFormatFillQ(const VectorFormat vform);
+unsigned RegisterSizeInBitsFromFormat(VectorFormat vform);
+unsigned RegisterSizeInBytesFromFormat(VectorFormat vform);
+// TODO: Make the return types of these functions consistent.
+unsigned LaneSizeInBitsFromFormat(VectorFormat vform);
+int LaneSizeInBytesFromFormat(VectorFormat vform);
+int LaneSizeInBytesLog2FromFormat(VectorFormat vform);
+int LaneCountFromFormat(VectorFormat vform);
+int MaxLaneCountFromFormat(VectorFormat vform);
+bool IsVectorFormat(VectorFormat vform);
+int64_t MaxIntFromFormat(VectorFormat vform);
+int64_t MinIntFromFormat(VectorFormat vform);
+uint64_t MaxUintFromFormat(VectorFormat vform);
+
+
+enum NEONFormat {
+  NF_UNDEF = 0,
+  NF_8B    = 1,
+  NF_16B   = 2,
+  NF_4H    = 3,
+  NF_8H    = 4,
+  NF_2S    = 5,
+  NF_4S    = 6,
+  NF_1D    = 7,
+  NF_2D    = 8,
+  NF_B     = 9,
+  NF_H     = 10,
+  NF_S     = 11,
+  NF_D     = 12
+};
+
+static const unsigned kNEONFormatMaxBits = 6;
+
+struct NEONFormatMap {
+  // The bit positions in the instruction to consider.
+  uint8_t bits[kNEONFormatMaxBits];
+
+  // Mapping from concatenated bits to format.
+  NEONFormat map[1 << kNEONFormatMaxBits];
+};
+
+class NEONFormatDecoder {
+ public:
+  enum SubstitutionMode {
+    kPlaceholder,
+    kFormat
+  };
+
+  // Construct a format decoder with increasingly specific format maps for each
+  // subsitution. If no format map is specified, the default is the integer
+  // format map.
+  explicit NEONFormatDecoder(const Instruction* instr) {
+    instrbits_ = instr->InstructionBits();
+    SetFormatMaps(IntegerFormatMap());
+  }
+  NEONFormatDecoder(const Instruction* instr,
+                    const NEONFormatMap* format) {
+    instrbits_ = instr->InstructionBits();
+    SetFormatMaps(format);
+  }
+  NEONFormatDecoder(const Instruction* instr,
+                    const NEONFormatMap* format0,
+                    const NEONFormatMap* format1) {
+    instrbits_ = instr->InstructionBits();
+    SetFormatMaps(format0, format1);
+  }
+  NEONFormatDecoder(const Instruction* instr,
+                    const NEONFormatMap* format0,
+                    const NEONFormatMap* format1,
+                    const NEONFormatMap* format2) {
+    instrbits_ = instr->InstructionBits();
+    SetFormatMaps(format0, format1, format2);
+  }
+
+  // Set the format mapping for all or individual substitutions.
+  void SetFormatMaps(const NEONFormatMap* format0,
+                     const NEONFormatMap* format1 = NULL,
+                     const NEONFormatMap* format2 = NULL) {
+    VIXL_ASSERT(format0 != NULL);
+    formats_[0] = format0;
+    formats_[1] = (format1 == NULL) ? formats_[0] : format1;
+    formats_[2] = (format2 == NULL) ? formats_[1] : format2;
+  }
+  void SetFormatMap(unsigned index, const NEONFormatMap* format) {
+    VIXL_ASSERT(index <= (sizeof(formats_) / sizeof(formats_[0])));
+    VIXL_ASSERT(format != NULL);
+    formats_[index] = format;
+  }
+
+  // Substitute %s in the input string with the placeholder string for each
+  // register, ie. "'B", "'H", etc.
+  const char* SubstitutePlaceholders(const char* string) {
+    return Substitute(string, kPlaceholder, kPlaceholder, kPlaceholder);
+  }
+
+  // Substitute %s in the input string with a new string based on the
+  // substitution mode.
+  const char* Substitute(const char* string,
+                         SubstitutionMode mode0 = kFormat,
+                         SubstitutionMode mode1 = kFormat,
+                         SubstitutionMode mode2 = kFormat) {
+    snprintf(form_buffer_, sizeof(form_buffer_), string,
+             GetSubstitute(0, mode0),
+             GetSubstitute(1, mode1),
+             GetSubstitute(2, mode2));
+    return form_buffer_;
+  }
+
+  // Append a "2" to a mnemonic string based of the state of the Q bit.
+  const char* Mnemonic(const char* mnemonic) {
+    if ((instrbits_ & NEON_Q) != 0) {
+      snprintf(mne_buffer_, sizeof(mne_buffer_), "%s2", mnemonic);
+      return mne_buffer_;
+    }
+    return mnemonic;
+  }
+
+  VectorFormat GetVectorFormat(int format_index = 0) {
+    return GetVectorFormat(formats_[format_index]);
+  }
+
+  VectorFormat GetVectorFormat(const NEONFormatMap* format_map) {
+    static const VectorFormat vform[] = {
+      kFormatUndefined,
+      kFormat8B, kFormat16B, kFormat4H, kFormat8H,
+      kFormat2S, kFormat4S, kFormat1D, kFormat2D,
+      kFormatB, kFormatH, kFormatS, kFormatD
+    };
+    VIXL_ASSERT(GetNEONFormat(format_map) < (sizeof(vform) / sizeof(vform[0])));
+    return vform[GetNEONFormat(format_map)];
+  }
+
+  // Built in mappings for common cases.
+
+  // The integer format map uses three bits (Q, size<1:0>) to encode the
+  // "standard" set of NEON integer vector formats.
+  static const NEONFormatMap* IntegerFormatMap() {
+    static const NEONFormatMap map = {
+      {23, 22, 30},
+      {NF_8B, NF_16B, NF_4H, NF_8H, NF_2S, NF_4S, NF_UNDEF, NF_2D}
+    };
+    return &map;
+  }
+
+  // The long integer format map uses two bits (size<1:0>) to encode the
+  // long set of NEON integer vector formats. These are used in narrow, wide
+  // and long operations.
+  static const NEONFormatMap* LongIntegerFormatMap() {
+    static const NEONFormatMap map = {
+      {23, 22}, {NF_8H, NF_4S, NF_2D}
+    };
+    return &map;
+  }
+
+  // The FP format map uses two bits (Q, size<0>) to encode the NEON FP vector
+  // formats: NF_2S, NF_4S, NF_2D.
+  static const NEONFormatMap* FPFormatMap() {
+    // The FP format map assumes two bits (Q, size<0>) are used to encode the
+    // NEON FP vector formats: NF_2S, NF_4S, NF_2D.
+    static const NEONFormatMap map = {
+      {22, 30}, {NF_2S, NF_4S, NF_UNDEF, NF_2D}
+    };
+    return &map;
+  }
+
+  // The load/store format map uses three bits (Q, 11, 10) to encode the
+  // set of NEON vector formats.
+  static const NEONFormatMap* LoadStoreFormatMap() {
+    static const NEONFormatMap map = {
+      {11, 10, 30},
+      {NF_8B, NF_16B, NF_4H, NF_8H, NF_2S, NF_4S, NF_1D, NF_2D}
+    };
+    return &map;
+  }
+
+  // The logical format map uses one bit (Q) to encode the NEON vector format:
+  // NF_8B, NF_16B.
+  static const NEONFormatMap* LogicalFormatMap() {
+    static const NEONFormatMap map = {
+      {30}, {NF_8B, NF_16B}
+    };
+    return &map;
+  }
+
+  // The triangular format map uses between two and five bits to encode the NEON
+  // vector format:
+  // xxx10->8B, xxx11->16B, xx100->4H, xx101->8H
+  // x1000->2S, x1001->4S,  10001->2D, all others undefined.
+  static const NEONFormatMap* TriangularFormatMap() {
+    static const NEONFormatMap map = {
+      {19, 18, 17, 16, 30},
+      {NF_UNDEF, NF_UNDEF, NF_8B, NF_16B, NF_4H, NF_8H, NF_8B, NF_16B, NF_2S,
+       NF_4S, NF_8B, NF_16B, NF_4H, NF_8H, NF_8B, NF_16B, NF_UNDEF, NF_2D,
+       NF_8B, NF_16B, NF_4H, NF_8H, NF_8B, NF_16B, NF_2S, NF_4S, NF_8B, NF_16B,
+       NF_4H, NF_8H, NF_8B, NF_16B}
+    };
+    return &map;
+  }
+
+  // The scalar format map uses two bits (size<1:0>) to encode the NEON scalar
+  // formats: NF_B, NF_H, NF_S, NF_D.
+  static const NEONFormatMap* ScalarFormatMap() {
+    static const NEONFormatMap map = {
+      {23, 22}, {NF_B, NF_H, NF_S, NF_D}
+    };
+    return &map;
+  }
+
+  // The long scalar format map uses two bits (size<1:0>) to encode the longer
+  // NEON scalar formats: NF_H, NF_S, NF_D.
+  static const NEONFormatMap* LongScalarFormatMap() {
+    static const NEONFormatMap map = {
+      {23, 22}, {NF_H, NF_S, NF_D}
+    };
+    return &map;
+  }
+
+  // The FP scalar format map assumes one bit (size<0>) is used to encode the
+  // NEON FP scalar formats: NF_S, NF_D.
+  static const NEONFormatMap* FPScalarFormatMap() {
+    static const NEONFormatMap map = {
+      {22}, {NF_S, NF_D}
+    };
+    return &map;
+  }
+
+  // The triangular scalar format map uses between one and four bits to encode
+  // the NEON FP scalar formats:
+  // xxx1->B, xx10->H, x100->S, 1000->D, all others undefined.
+  static const NEONFormatMap* TriangularScalarFormatMap() {
+    static const NEONFormatMap map = {
+      {19, 18, 17, 16},
+      {NF_UNDEF, NF_B, NF_H, NF_B, NF_S, NF_B, NF_H, NF_B,
+       NF_D,     NF_B, NF_H, NF_B, NF_S, NF_B, NF_H, NF_B}
+    };
+    return &map;
+  }
+
+ private:
+  // Get a pointer to a string that represents the format or placeholder for
+  // the specified substitution index, based on the format map and instruction.
+  const char* GetSubstitute(int index, SubstitutionMode mode) {
+    if (mode == kFormat) {
+      return NEONFormatAsString(GetNEONFormat(formats_[index]));
+    }
+    VIXL_ASSERT(mode == kPlaceholder);
+    return NEONFormatAsPlaceholder(GetNEONFormat(formats_[index]));
+  }
+
+  // Get the NEONFormat enumerated value for bits obtained from the
+  // instruction based on the specified format mapping.
+  NEONFormat GetNEONFormat(const NEONFormatMap* format_map) {
+    return format_map->map[PickBits(format_map->bits)];
+  }
+
+  // Convert a NEONFormat into a string.
+  static const char* NEONFormatAsString(NEONFormat format) {
+    static const char* formats[] = {
+      "undefined",
+      "8b", "16b", "4h", "8h", "2s", "4s", "1d", "2d",
+      "b", "h", "s", "d"
+    };
+    VIXL_ASSERT(format < (sizeof(formats) / sizeof(formats[0])));
+    return formats[format];
+  }
+
+  // Convert a NEONFormat into a register placeholder string.
+  static const char* NEONFormatAsPlaceholder(NEONFormat format) {
+    VIXL_ASSERT((format == NF_B) || (format == NF_H) ||
+                (format == NF_S) || (format == NF_D) ||
+                (format == NF_UNDEF));
+    static const char* formats[] = {
+      "undefined",
+      "undefined", "undefined", "undefined", "undefined",
+      "undefined", "undefined", "undefined", "undefined",
+      "'B", "'H", "'S", "'D"
+    };
+    return formats[format];
+  }
+
+  // Select bits from instrbits_ defined by the bits array, concatenate them,
+  // and return the value.
+  uint8_t PickBits(const uint8_t bits[]) {
+    uint8_t result = 0;
+    for (unsigned b = 0; b < kNEONFormatMaxBits; b++) {
+      if (bits[b] == 0) break;
+      result <<= 1;
+      result |= ((instrbits_ & (1 << bits[b])) == 0) ? 0 : 1;
+    }
+    return result;
+  }
+
+  Instr instrbits_;
+  const NEONFormatMap* formats_[3];
+  char form_buffer_[64];
+  char mne_buffer_[16];
+};
+}  // namespace vixl
+
+#endif  // VIXL_A64_INSTRUCTIONS_A64_H_
diff --git a/disas/libvixl/code-buffer.h b/disas/libvixl/vixl/code-buffer.h
index da6233dd80..f93ebb6b82 100644
--- a/disas/libvixl/code-buffer.h
+++ b/disas/libvixl/vixl/code-buffer.h
@@ -28,7 +28,7 @@
 #define VIXL_CODE_BUFFER_H
 
 #include <string.h>
-#include "globals.h"
+#include "vixl/globals.h"
 
 namespace vixl {
 
diff --git a/disas/libvixl/utils.cc b/disas/libvixl/vixl/compiler-intrinsics.cc
index 80b132a11e..fd551faeb1 100644
--- a/disas/libvixl/utils.cc
+++ b/disas/libvixl/vixl/compiler-intrinsics.cc
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2015, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -24,53 +24,13 @@
 // 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.
 
-#include "utils.h"
-#include <stdio.h>
+#include "compiler-intrinsics.h"
 
 namespace vixl {
 
-uint32_t float_to_rawbits(float value) {
-  uint32_t bits = 0;
-  memcpy(&bits, &value, 4);
-  return bits;
-}
-
-
-uint64_t double_to_rawbits(double value) {
-  uint64_t bits = 0;
-  memcpy(&bits, &value, 8);
-  return bits;
-}
-
-
-float rawbits_to_float(uint32_t bits) {
-  float value = 0.0;
-  memcpy(&value, &bits, 4);
-  return value;
-}
-
-
-double rawbits_to_double(uint64_t bits) {
-  double value = 0.0;
-  memcpy(&value, &bits, 8);
-  return value;
-}
-
 
-int CountLeadingZeros(uint64_t value, int width) {
-  VIXL_ASSERT((width == 32) || (width == 64));
-  int count = 0;
-  uint64_t bit_test = UINT64_C(1) << (width - 1);
-  while ((count < width) && ((bit_test & value) == 0)) {
-    count++;
-    bit_test >>= 1;
-  }
-  return count;
-}
-
-
-int CountLeadingSignBits(int64_t value, int width) {
-  VIXL_ASSERT((width == 32) || (width == 64));
+int CountLeadingSignBitsFallBack(int64_t value, int width) {
+  VIXL_ASSERT(IsPowerOf2(width) && (width <= 64));
   if (value >= 0) {
     return CountLeadingZeros(value, width) - 1;
   } else {
@@ -79,23 +39,46 @@ int CountLeadingSignBits(int64_t value, int width) {
 }
 
 
-int CountTrailingZeros(uint64_t value, int width) {
-  VIXL_ASSERT((width == 32) || (width == 64));
+int CountLeadingZerosFallBack(uint64_t value, int width) {
+  VIXL_ASSERT(IsPowerOf2(width) && (width <= 64));
+  if (value == 0) {
+    return width;
+  }
   int count = 0;
-  while ((count < width) && (((value >> count) & 1) == 0)) {
-    count++;
+  value = value << (64 - width);
+  if ((value & UINT64_C(0xffffffff00000000)) == 0) {
+    count += 32;
+    value = value << 32;
   }
+  if ((value & UINT64_C(0xffff000000000000)) == 0) {
+    count += 16;
+    value = value << 16;
+  }
+  if ((value & UINT64_C(0xff00000000000000)) == 0) {
+    count += 8;
+    value = value << 8;
+  }
+  if ((value & UINT64_C(0xf000000000000000)) == 0) {
+    count += 4;
+    value = value << 4;
+  }
+  if ((value & UINT64_C(0xc000000000000000)) == 0) {
+    count += 2;
+    value = value << 2;
+  }
+  if ((value & UINT64_C(0x8000000000000000)) == 0) {
+    count += 1;
+  }
+  count += (value == 0);
   return count;
 }
 
 
-int CountSetBits(uint64_t value, int width) {
-  // TODO: Other widths could be added here, as the implementation already
-  // supports them.
-  VIXL_ASSERT((width == 32) || (width == 64));
+int CountSetBitsFallBack(uint64_t value, int width) {
+  VIXL_ASSERT(IsPowerOf2(width) && (width <= 64));
 
   // Mask out unused bits to ensure that they are not counted.
-  value &= (UINT64_C(0xffffffffffffffff) >> (64-width));
+  value &= (UINT64_C(0xffffffffffffffff) >> (64 - width));
 
   // Add up the set bits.
   // The algorithm works by adding pairs of bit fields together iteratively,
@@ -122,30 +105,40 @@ int CountSetBits(uint64_t value, int width) {
     value = ((value >> shift) & kMasks[i]) + (value & kMasks[i]);
   }
 
-  return value;
-}
-
-
-uint64_t LowestSetBit(uint64_t value) {
-  return value & -value;
-}
-
-
-bool IsPowerOf2(int64_t value) {
-  return (value != 0) && ((value & (value - 1)) == 0);
+  return static_cast<int>(value);
 }
 
 
-unsigned CountClearHalfWords(uint64_t imm, unsigned reg_size) {
-  VIXL_ASSERT((reg_size % 8) == 0);
+int CountTrailingZerosFallBack(uint64_t value, int width) {
+  VIXL_ASSERT(IsPowerOf2(width) && (width <= 64));
   int count = 0;
-  for (unsigned i = 0; i < (reg_size / 16); i++) {
-    if ((imm & 0xffff) == 0) {
-      count++;
-    }
-    imm >>= 16;
+  value = value << (64 - width);
+  if ((value & UINT64_C(0xffffffff)) == 0) {
+    count += 32;
+    value = value >> 32;
   }
-  return count;
+  if ((value & 0xffff) == 0) {
+    count += 16;
+    value = value >> 16;
+  }
+  if ((value & 0xff) == 0) {
+    count += 8;
+    value = value >> 8;
+  }
+  if ((value & 0xf) == 0) {
+    count += 4;
+    value = value >> 4;
+  }
+  if ((value & 0x3) == 0) {
+    count += 2;
+    value = value >> 2;
+  }
+  if ((value & 0x1) == 0) {
+    count += 1;
+  }
+  count += (value == 0);
+  return count - (64 - width);
 }
 
+
 }  // namespace vixl
diff --git a/disas/libvixl/vixl/compiler-intrinsics.h b/disas/libvixl/vixl/compiler-intrinsics.h
new file mode 100644
index 0000000000..9431beddb9
--- /dev/null
+++ b/disas/libvixl/vixl/compiler-intrinsics.h
@@ -0,0 +1,155 @@
+// 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_COMPILER_INTRINSICS_H
+#define VIXL_COMPILER_INTRINSICS_H
+
+#include "globals.h"
+
+namespace vixl {
+
+// Helper to check whether the version of GCC used is greater than the specified
+// requirement.
+#define MAJOR 1000000
+#define MINOR 1000
+#if defined(__GNUC__) && defined(__GNUC_MINOR__) && defined(__GNUC_PATCHLEVEL__)
+#define GCC_VERSION_OR_NEWER(major, minor, patchlevel)                         \
+    ((__GNUC__ * MAJOR + __GNUC_MINOR__ * MINOR + __GNUC_PATCHLEVEL__) >=      \
+     ((major) * MAJOR + (minor) * MINOR + (patchlevel)))
+#elif defined(__GNUC__) && defined(__GNUC_MINOR__)
+#define GCC_VERSION_OR_NEWER(major, minor, patchlevel)                         \
+    ((__GNUC__ * MAJOR + __GNUC_MINOR__ * MINOR) >=                            \
+     ((major) * MAJOR + (minor) * MINOR + (patchlevel)))
+#else
+#define GCC_VERSION_OR_NEWER(major, minor, patchlevel) 0
+#endif
+
+
+#if defined(__clang__) && !defined(VIXL_NO_COMPILER_BUILTINS)
+
+#define COMPILER_HAS_BUILTIN_CLRSB    (__has_builtin(__builtin_clrsb))
+#define COMPILER_HAS_BUILTIN_CLZ      (__has_builtin(__builtin_clz))
+#define COMPILER_HAS_BUILTIN_CTZ      (__has_builtin(__builtin_ctz))
+#define COMPILER_HAS_BUILTIN_FFS      (__has_builtin(__builtin_ffs))
+#define COMPILER_HAS_BUILTIN_POPCOUNT (__has_builtin(__builtin_popcount))
+
+#elif defined(__GNUC__) && !defined(VIXL_NO_COMPILER_BUILTINS)
+// The documentation for these builtins is available at:
+// https://gcc.gnu.org/onlinedocs/gcc-$MAJOR.$MINOR.$PATCHLEVEL/gcc//Other-Builtins.html
+
+# define COMPILER_HAS_BUILTIN_CLRSB    (GCC_VERSION_OR_NEWER(4, 7, 0))
+# define COMPILER_HAS_BUILTIN_CLZ      (GCC_VERSION_OR_NEWER(3, 4, 0))
+# define COMPILER_HAS_BUILTIN_CTZ      (GCC_VERSION_OR_NEWER(3, 4, 0))
+# define COMPILER_HAS_BUILTIN_FFS      (GCC_VERSION_OR_NEWER(3, 4, 0))
+# define COMPILER_HAS_BUILTIN_POPCOUNT (GCC_VERSION_OR_NEWER(3, 4, 0))
+
+#else
+// One can define VIXL_NO_COMPILER_BUILTINS to force using the manually
+// implemented C++ methods.
+
+#define COMPILER_HAS_BUILTIN_BSWAP    false
+#define COMPILER_HAS_BUILTIN_CLRSB    false
+#define COMPILER_HAS_BUILTIN_CLZ      false
+#define COMPILER_HAS_BUILTIN_CTZ      false
+#define COMPILER_HAS_BUILTIN_FFS      false
+#define COMPILER_HAS_BUILTIN_POPCOUNT false
+
+#endif
+
+
+template<typename V>
+inline bool IsPowerOf2(V value) {
+  return (value != 0) && ((value & (value - 1)) == 0);
+}
+
+
+// Declaration of fallback functions.
+int CountLeadingSignBitsFallBack(int64_t value, int width);
+int CountLeadingZerosFallBack(uint64_t value, int width);
+int CountSetBitsFallBack(uint64_t value, int width);
+int CountTrailingZerosFallBack(uint64_t value, int width);
+
+
+// Implementation of intrinsics functions.
+// TODO: The implementations could be improved for sizes different from 32bit
+// and 64bit: we could mask the values and call the appropriate builtin.
+
+template<typename V>
+inline int CountLeadingSignBits(V value, int width = (sizeof(V) * 8)) {
+#if COMPILER_HAS_BUILTIN_CLRSB
+  if (width == 32) {
+    return __builtin_clrsb(value);
+  } else if (width == 64) {
+    return __builtin_clrsbll(value);
+  }
+#endif
+  return CountLeadingSignBitsFallBack(value, width);
+}
+
+
+template<typename V>
+inline int CountLeadingZeros(V value, int width = (sizeof(V) * 8)) {
+#if COMPILER_HAS_BUILTIN_CLZ
+  if (width == 32) {
+    return (value == 0) ? 32 : __builtin_clz(static_cast<unsigned>(value));
+  } else if (width == 64) {
+    return (value == 0) ? 64 : __builtin_clzll(value);
+  }
+#endif
+  return CountLeadingZerosFallBack(value, width);
+}
+
+
+template<typename V>
+inline int CountSetBits(V value, int width = (sizeof(V) * 8)) {
+#if COMPILER_HAS_BUILTIN_POPCOUNT
+  if (width == 32) {
+    return __builtin_popcount(static_cast<unsigned>(value));
+  } else if (width == 64) {
+    return __builtin_popcountll(value);
+  }
+#endif
+  return CountSetBitsFallBack(value, width);
+}
+
+
+template<typename V>
+inline int CountTrailingZeros(V value, int width = (sizeof(V) * 8)) {
+#if COMPILER_HAS_BUILTIN_CTZ
+  if (width == 32) {
+    return (value == 0) ? 32 : __builtin_ctz(static_cast<unsigned>(value));
+  } else if (width == 64) {
+    return (value == 0) ? 64 : __builtin_ctzll(value);
+  }
+#endif
+  return CountTrailingZerosFallBack(value, width);
+}
+
+}  // namespace vixl
+
+#endif  // VIXL_COMPILER_INTRINSICS_H
+
diff --git a/disas/libvixl/globals.h b/disas/libvixl/vixl/globals.h
index 0c2493105d..61dc9f7f7e 100644
--- a/disas/libvixl/globals.h
+++ b/disas/libvixl/vixl/globals.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2015, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -49,20 +49,26 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <stddef.h>
-#include "platform.h"
+#include "vixl/platform.h"
 
 
 typedef uint8_t byte;
 
+// Type for half-precision (16 bit) floating point numbers.
+typedef uint16_t float16;
+
 const int KBytes = 1024;
 const int MBytes = 1024 * KBytes;
 
-#define VIXL_ABORT() printf("in %s, line %i", __FILE__, __LINE__); abort()
+#define VIXL_ABORT() \
+    do { printf("in %s, line %i", __FILE__, __LINE__); abort(); } while (false)
 #ifdef VIXL_DEBUG
   #define VIXL_ASSERT(condition) assert(condition)
   #define VIXL_CHECK(condition) VIXL_ASSERT(condition)
-  #define VIXL_UNIMPLEMENTED() printf("UNIMPLEMENTED\t"); VIXL_ABORT()
-  #define VIXL_UNREACHABLE() printf("UNREACHABLE\t"); VIXL_ABORT()
+  #define VIXL_UNIMPLEMENTED() \
+    do { fprintf(stderr, "UNIMPLEMENTED\t"); VIXL_ABORT(); } while (false)
+  #define VIXL_UNREACHABLE() \
+    do { fprintf(stderr, "UNREACHABLE\t"); VIXL_ABORT(); } while (false)
 #else
   #define VIXL_ASSERT(condition) ((void) 0)
   #define VIXL_CHECK(condition) assert(condition)
@@ -76,10 +82,70 @@ const int MBytes = 1024 * KBytes;
 #define VIXL_STATIC_ASSERT_LINE(line, condition) \
   typedef char VIXL_CONCAT(STATIC_ASSERT_LINE_, line)[(condition) ? 1 : -1] \
   __attribute__((unused))
-#define VIXL_STATIC_ASSERT(condition) VIXL_STATIC_ASSERT_LINE(__LINE__, condition) //NOLINT
+#define VIXL_STATIC_ASSERT(condition) \
+    VIXL_STATIC_ASSERT_LINE(__LINE__, condition)
+
+template <typename T1>
+inline void USE(T1) {}
+
+template <typename T1, typename T2>
+inline void USE(T1, T2) {}
+
+template <typename T1, typename T2, typename T3>
+inline void USE(T1, T2, T3) {}
+
+template <typename T1, typename T2, typename T3, typename T4>
+inline void USE(T1, T2, T3, T4) {}
+
+#define VIXL_ALIGNMENT_EXCEPTION() \
+    do { fprintf(stderr, "ALIGNMENT EXCEPTION\t"); VIXL_ABORT(); } while (0)
+
+// The clang::fallthrough attribute is used along with the Wimplicit-fallthrough
+// argument to annotate intentional fall-through between switch labels.
+// For more information please refer to:
+// http://clang.llvm.org/docs/AttributeReference.html#fallthrough-clang-fallthrough
+#ifndef __has_warning
+  #define __has_warning(x)  0
+#endif
+
+// Note: This option is only available for Clang. And will only be enabled for
+// C++11(201103L).
+#if __has_warning("-Wimplicit-fallthrough") && __cplusplus >= 201103L
+  #define VIXL_FALLTHROUGH() [[clang::fallthrough]] //NOLINT
+#else
+  #define VIXL_FALLTHROUGH() do {} while (0)
+#endif
+
+#if __cplusplus >= 201103L
+  #define VIXL_NO_RETURN  [[noreturn]] //NOLINT
+#else
+  #define VIXL_NO_RETURN  __attribute__((noreturn))
+#endif
+
+// Some functions might only be marked as "noreturn" for the DEBUG build. This
+// macro should be used for such cases (for more details see what
+// VIXL_UNREACHABLE expands to).
+#ifdef VIXL_DEBUG
+  #define VIXL_DEBUG_NO_RETURN  VIXL_NO_RETURN
+#else
+  #define VIXL_DEBUG_NO_RETURN
+#endif
 
-template <typename T> inline void USE(T) {}
+#ifdef VIXL_INCLUDE_SIMULATOR
+#ifndef VIXL_GENERATE_SIMULATOR_INSTRUCTIONS_VALUE
+  #define VIXL_GENERATE_SIMULATOR_INSTRUCTIONS_VALUE  1
+#endif
+#else
+#ifndef VIXL_GENERATE_SIMULATOR_INSTRUCTIONS_VALUE
+  #define VIXL_GENERATE_SIMULATOR_INSTRUCTIONS_VALUE  0
+#endif
+#if VIXL_GENERATE_SIMULATOR_INSTRUCTIONS_VALUE
+  #warning "Generating Simulator instructions without Simulator support."
+#endif
+#endif
 
-#define VIXL_ALIGNMENT_EXCEPTION() printf("ALIGNMENT EXCEPTION\t"); VIXL_ABORT()
+#ifdef USE_SIMULATOR
+  #error "Please see the release notes for USE_SIMULATOR."
+#endif
 
 #endif  // VIXL_GLOBALS_H
diff --git a/disas/libvixl/vixl/invalset.h b/disas/libvixl/vixl/invalset.h
new file mode 100644
index 0000000000..ffdc0237b4
--- /dev/null
+++ b/disas/libvixl/vixl/invalset.h
@@ -0,0 +1,775 @@
+// 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_INVALSET_H_
+#define VIXL_INVALSET_H_
+
+#include <string.h>
+
+#include <algorithm>
+#include <vector>
+
+#include "vixl/globals.h"
+
+namespace vixl {
+
+// We define a custom data structure template and its iterator as `std`
+// containers do not fit the performance requirements for some of our use cases.
+//
+// The structure behaves like an iterable unordered set with special properties
+// and restrictions. "InvalSet" stands for "Invalidatable Set".
+//
+// Restrictions and requirements:
+// - Adding an element already present in the set is illegal. In debug mode,
+//   this is checked at insertion time.
+// - The templated class `ElementType` must provide comparison operators so that
+//   `std::sort()` can be used.
+// - A key must be available to represent invalid elements.
+// - Elements with an invalid key must compare higher or equal to any other
+//   element.
+//
+// Use cases and performance considerations:
+// Our use cases present two specificities that allow us to design this
+// structure to provide fast insertion *and* fast search and deletion
+// operations:
+// - Elements are (generally) inserted in order (sorted according to their key).
+// - A key is available to mark elements as invalid (deleted).
+// The backing `std::vector` allows for fast insertions. When
+// searching for an element we ensure the elements are sorted (this is generally
+// the case) and perform a binary search. When deleting an element we do not
+// free the associated memory immediately. Instead, an element to be deleted is
+// marked with the 'invalid' key. Other methods of the container take care of
+// ignoring entries marked as invalid.
+// To avoid the overhead of the `std::vector` container when only few entries
+// are used, a number of elements are preallocated.
+
+// 'ElementType' and 'KeyType' are respectively the types of the elements and
+// their key.  The structure only reclaims memory when safe to do so, if the
+// number of elements that can be reclaimed is greater than `RECLAIM_FROM` and
+// greater than `<total number of elements> / RECLAIM_FACTOR.
+#define TEMPLATE_INVALSET_P_DECL                                               \
+  class ElementType,                                                           \
+  unsigned N_PREALLOCATED_ELEMENTS,                                            \
+  class KeyType,                                                               \
+  KeyType INVALID_KEY,                                                         \
+  size_t RECLAIM_FROM,                                                         \
+  unsigned RECLAIM_FACTOR
+
+#define TEMPLATE_INVALSET_P_DEF                                                \
+ElementType, N_PREALLOCATED_ELEMENTS,                                          \
+KeyType, INVALID_KEY, RECLAIM_FROM, RECLAIM_FACTOR
+
+template<class S> class InvalSetIterator;  // Forward declaration.
+
+template<TEMPLATE_INVALSET_P_DECL> class InvalSet {
+ public:
+  InvalSet();
+  ~InvalSet();
+
+  static const size_t kNPreallocatedElements = N_PREALLOCATED_ELEMENTS;
+  static const KeyType kInvalidKey = INVALID_KEY;
+
+  // It is illegal to insert an element already present in the set.
+  void insert(const ElementType& element);
+
+  // Looks for the specified element in the set and - if found - deletes it.
+  void erase(const ElementType& element);
+
+  // This indicates the number of (valid) elements stored in this set.
+  size_t size() const;
+
+  // Returns true if no elements are stored in the set.
+  // Note that this does not mean the the backing storage is empty: it can still
+  // contain invalid elements.
+  bool empty() const;
+
+  void clear();
+
+  const ElementType min_element();
+
+  // This returns the key of the minimum element in the set.
+  KeyType min_element_key();
+
+  static bool IsValid(const ElementType& element);
+  static KeyType Key(const ElementType& element);
+  static void SetKey(ElementType* element, KeyType key);
+
+ protected:
+  // Returns a pointer to the element in vector_ if it was found, or NULL
+  // otherwise.
+  ElementType* Search(const ElementType& element);
+
+  // The argument *must* point to an element stored in *this* set.
+  // This function is not allowed to move elements in the backing vector
+  // storage.
+  void EraseInternal(ElementType* element);
+
+  // The elements in the range searched must be sorted.
+  ElementType* BinarySearch(const ElementType& element,
+                            ElementType* start,
+                            ElementType* end) const;
+
+  // Sort the elements.
+  enum SortType {
+    // The 'hard' version guarantees that invalid elements are moved to the end
+    // of the container.
+    kHardSort,
+    // The 'soft' version only guarantees that the elements will be sorted.
+    // Invalid elements may still be present anywhere in the set.
+    kSoftSort
+  };
+  void Sort(SortType sort_type);
+
+  // Delete the elements that have an invalid key. The complexity is linear
+  // with the size of the vector.
+  void Clean();
+
+  const ElementType Front() const;
+  const ElementType Back() const;
+
+  // Delete invalid trailing elements and return the last valid element in the
+  // set.
+  const ElementType CleanBack();
+
+  // Returns a pointer to the start or end of the backing storage.
+  const ElementType* StorageBegin() const;
+  const ElementType* StorageEnd() const;
+  ElementType* StorageBegin();
+  ElementType* StorageEnd();
+
+  // Returns the index of the element within the backing storage. The element
+  // must belong to the backing storage.
+  size_t ElementIndex(const ElementType* element) const;
+
+  // Returns the element at the specified index in the backing storage.
+  const ElementType* ElementAt(size_t index) const;
+  ElementType* ElementAt(size_t index);
+
+  static const ElementType* FirstValidElement(const ElementType* from,
+                                              const ElementType* end);
+
+  void CacheMinElement();
+  const ElementType CachedMinElement() const;
+
+  bool ShouldReclaimMemory() const;
+  void ReclaimMemory();
+
+  bool IsUsingVector() const { return vector_ != NULL; }
+  void set_sorted(bool sorted) { sorted_ = sorted; }
+
+  // We cache some data commonly required by users to improve performance.
+  // We cannot cache pointers to elements as we do not control the backing
+  // storage.
+  bool valid_cached_min_;
+  size_t cached_min_index_;  // Valid iff `valid_cached_min_` is true.
+  KeyType cached_min_key_;         // Valid iff `valid_cached_min_` is true.
+
+  // Indicates whether the elements are sorted.
+  bool sorted_;
+
+  // This represents the number of (valid) elements in this set.
+  size_t size_;
+
+  // The backing storage is either the array of preallocated elements or the
+  // vector. The structure starts by using the preallocated elements, and
+  // transitions (permanently) to using the vector once more than
+  // kNPreallocatedElements are used.
+  // Elements are only invalidated when using the vector. The preallocated
+  // storage always only contains valid elements.
+  ElementType preallocated_[kNPreallocatedElements];
+  std::vector<ElementType>* vector_;
+
+#ifdef VIXL_DEBUG
+  // Iterators acquire and release this monitor. While a set is acquired,
+  // certain operations are illegal to ensure that the iterator will
+  // correctly iterate over the elements in the set.
+  int monitor_;
+  int monitor() const { return monitor_; }
+  void Acquire() { monitor_++; }
+  void Release() {
+    monitor_--;
+    VIXL_ASSERT(monitor_ >= 0);
+  }
+#endif
+
+  friend class InvalSetIterator<InvalSet<TEMPLATE_INVALSET_P_DEF> >;
+  typedef ElementType _ElementType;
+  typedef KeyType _KeyType;
+};
+
+
+template<class S> class InvalSetIterator {
+ private:
+  // Redefine types to mirror the associated set types.
+  typedef typename S::_ElementType ElementType;
+  typedef typename S::_KeyType KeyType;
+
+ public:
+  explicit InvalSetIterator(S* inval_set);
+  ~InvalSetIterator();
+
+  ElementType* Current() const;
+  void Advance();
+  bool Done() const;
+
+  // Mark this iterator as 'done'.
+  void Finish();
+
+  // Delete the current element and advance the iterator to point to the next
+  // element.
+  void DeleteCurrentAndAdvance();
+
+  static bool IsValid(const ElementType& element);
+  static KeyType Key(const ElementType& element);
+
+ protected:
+  void MoveToValidElement();
+
+  // Indicates if the iterator is looking at the vector or at the preallocated
+  // elements.
+  const bool using_vector_;
+  // Used when looking at the preallocated elements, or in debug mode when using
+  // the vector to track how many times the iterator has advanced.
+  size_t index_;
+  typename std::vector<ElementType>::iterator iterator_;
+  S* inval_set_;
+};
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+InvalSet<TEMPLATE_INVALSET_P_DEF>::InvalSet()
+  : valid_cached_min_(false),
+    sorted_(true), size_(0), vector_(NULL) {
+#ifdef VIXL_DEBUG
+  monitor_ = 0;
+#endif
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+InvalSet<TEMPLATE_INVALSET_P_DEF>::~InvalSet() {
+  VIXL_ASSERT(monitor_ == 0);
+  delete vector_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::insert(const ElementType& element) {
+  VIXL_ASSERT(monitor() == 0);
+  VIXL_ASSERT(IsValid(element));
+  VIXL_ASSERT(Search(element) == NULL);
+  set_sorted(empty() || (sorted_ && (element > CleanBack())));
+  if (IsUsingVector()) {
+    vector_->push_back(element);
+  } else {
+    if (size_ < kNPreallocatedElements) {
+      preallocated_[size_] = element;
+    } else {
+      // Transition to using the vector.
+      vector_ = new std::vector<ElementType>(preallocated_,
+                                             preallocated_ + size_);
+      vector_->push_back(element);
+    }
+  }
+  size_++;
+
+  if (valid_cached_min_ && (element < min_element())) {
+    cached_min_index_ = IsUsingVector() ? vector_->size() - 1 : size_ - 1;
+    cached_min_key_ = Key(element);
+    valid_cached_min_ = true;
+  }
+
+  if (ShouldReclaimMemory()) {
+    ReclaimMemory();
+  }
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::erase(const ElementType& element) {
+  VIXL_ASSERT(monitor() == 0);
+  VIXL_ASSERT(IsValid(element));
+  ElementType* local_element = Search(element);
+  if (local_element != NULL) {
+    EraseInternal(local_element);
+  }
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::Search(
+    const ElementType& element) {
+  VIXL_ASSERT(monitor() == 0);
+  if (empty()) {
+    return NULL;
+  }
+  if (ShouldReclaimMemory()) {
+    ReclaimMemory();
+  }
+  if (!sorted_) {
+    Sort(kHardSort);
+  }
+  if (!valid_cached_min_) {
+    CacheMinElement();
+  }
+  return BinarySearch(element, ElementAt(cached_min_index_), StorageEnd());
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+size_t InvalSet<TEMPLATE_INVALSET_P_DEF>::size() const {
+  return size_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+bool InvalSet<TEMPLATE_INVALSET_P_DEF>::empty() const {
+  return size_ == 0;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::clear() {
+  VIXL_ASSERT(monitor() == 0);
+  size_ = 0;
+  if (IsUsingVector()) {
+    vector_->clear();
+  }
+  set_sorted(true);
+  valid_cached_min_ = false;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType InvalSet<TEMPLATE_INVALSET_P_DEF>::min_element() {
+  VIXL_ASSERT(monitor() == 0);
+  VIXL_ASSERT(!empty());
+  CacheMinElement();
+  return *ElementAt(cached_min_index_);
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+KeyType InvalSet<TEMPLATE_INVALSET_P_DEF>::min_element_key() {
+  VIXL_ASSERT(monitor() == 0);
+  if (valid_cached_min_) {
+    return cached_min_key_;
+  } else {
+    return Key(min_element());
+  }
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+bool InvalSet<TEMPLATE_INVALSET_P_DEF>::IsValid(const ElementType& element) {
+  return Key(element) != kInvalidKey;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::EraseInternal(ElementType* element) {
+  // Note that this function must be safe even while an iterator has acquired
+  // this set.
+  VIXL_ASSERT(element != NULL);
+  size_t deleted_index = ElementIndex(element);
+  if (IsUsingVector()) {
+    VIXL_ASSERT((&(vector_->front()) <= element) &&
+                (element <= &(vector_->back())));
+    SetKey(element, kInvalidKey);
+  } else {
+    VIXL_ASSERT((preallocated_ <= element) &&
+                (element < (preallocated_ + kNPreallocatedElements)));
+    ElementType* end = preallocated_ + kNPreallocatedElements;
+    size_t copy_size = sizeof(*element) * (end - element - 1);
+    memmove(element, element + 1, copy_size);
+  }
+  size_--;
+
+  if (valid_cached_min_ &&
+      (deleted_index == cached_min_index_)) {
+    if (sorted_ && !empty()) {
+      const ElementType* min = FirstValidElement(element, StorageEnd());
+      cached_min_index_ = ElementIndex(min);
+      cached_min_key_ = Key(*min);
+      valid_cached_min_ = true;
+    } else {
+      valid_cached_min_ = false;
+    }
+  }
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::BinarySearch(
+    const ElementType& element, ElementType* start, ElementType* end) const {
+  if (start == end) {
+    return NULL;
+  }
+  VIXL_ASSERT(sorted_);
+  VIXL_ASSERT(start < end);
+  VIXL_ASSERT(!empty());
+
+  // Perform a binary search through the elements while ignoring invalid
+  // elements.
+  ElementType* elements = start;
+  size_t low = 0;
+  size_t high = (end - start) - 1;
+  while (low < high) {
+    // Find valid bounds.
+    while (!IsValid(elements[low]) && (low < high)) ++low;
+    while (!IsValid(elements[high]) && (low < high)) --high;
+    VIXL_ASSERT(low <= high);
+    // Avoid overflow when computing the middle index.
+    size_t middle = low / 2 + high / 2 + (low & high & 1);
+    if ((middle == low) || (middle == high)) {
+      break;
+    }
+    while (!IsValid(elements[middle]) && (middle < high - 1)) ++middle;
+    while (!IsValid(elements[middle]) && (low + 1 < middle)) --middle;
+    if (!IsValid(elements[middle])) {
+      break;
+    }
+    if (elements[middle] < element) {
+      low = middle;
+    } else {
+      high = middle;
+    }
+  }
+
+  if (elements[low] == element) return &elements[low];
+  if (elements[high] == element) return &elements[high];
+  return NULL;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::Sort(SortType sort_type) {
+  VIXL_ASSERT(monitor() == 0);
+  if (sort_type == kSoftSort) {
+    if (sorted_) {
+      return;
+    }
+  }
+  if (empty()) {
+    return;
+  }
+
+  Clean();
+  std::sort(StorageBegin(), StorageEnd());
+
+  set_sorted(true);
+  cached_min_index_ = 0;
+  cached_min_key_ = Key(Front());
+  valid_cached_min_ = true;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::Clean() {
+  VIXL_ASSERT(monitor() == 0);
+  if (empty() || !IsUsingVector()) {
+    return;
+  }
+  // Manually iterate through the vector storage to discard invalid elements.
+  ElementType* start = &(vector_->front());
+  ElementType* end = start + vector_->size();
+  ElementType* c = start;
+  ElementType* first_invalid;
+  ElementType* first_valid;
+  ElementType* next_invalid;
+
+  while (c < end && IsValid(*c)) { c++; }
+  first_invalid = c;
+
+  while (c < end) {
+    while (c < end && !IsValid(*c)) { c++; }
+    first_valid = c;
+    while (c < end && IsValid(*c)) { c++; }
+    next_invalid = c;
+
+    ptrdiff_t n_moved_elements = (next_invalid - first_valid);
+    memmove(first_invalid, first_valid,  n_moved_elements * sizeof(*c));
+    first_invalid = first_invalid + n_moved_elements;
+    c = next_invalid;
+  }
+
+  // Delete the trailing invalid elements.
+  vector_->erase(vector_->begin() + (first_invalid - start), vector_->end());
+  VIXL_ASSERT(vector_->size() == size_);
+
+  if (sorted_) {
+    valid_cached_min_ = true;
+    cached_min_index_ = 0;
+    cached_min_key_ = Key(*ElementAt(0));
+  } else {
+    valid_cached_min_ = false;
+  }
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType InvalSet<TEMPLATE_INVALSET_P_DEF>::Front() const {
+  VIXL_ASSERT(!empty());
+  return IsUsingVector() ? vector_->front() : preallocated_[0];
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType InvalSet<TEMPLATE_INVALSET_P_DEF>::Back() const {
+  VIXL_ASSERT(!empty());
+  return IsUsingVector() ? vector_->back() : preallocated_[size_ - 1];
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType InvalSet<TEMPLATE_INVALSET_P_DEF>::CleanBack() {
+  VIXL_ASSERT(monitor() == 0);
+  if (IsUsingVector()) {
+    // Delete the invalid trailing elements.
+    typename std::vector<ElementType>::reverse_iterator it = vector_->rbegin();
+    while (!IsValid(*it)) {
+      it++;
+    }
+    vector_->erase(it.base(), vector_->end());
+  }
+  return Back();
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::StorageBegin() const {
+  return IsUsingVector() ? &(vector_->front()) : preallocated_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::StorageEnd() const {
+  return IsUsingVector() ? &(vector_->back()) + 1 : preallocated_ + size_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::StorageBegin() {
+  return IsUsingVector() ? &(vector_->front()) : preallocated_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::StorageEnd() {
+  return IsUsingVector() ? &(vector_->back()) + 1 : preallocated_ + size_;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+size_t InvalSet<TEMPLATE_INVALSET_P_DEF>::ElementIndex(
+    const ElementType* element) const {
+  VIXL_ASSERT((StorageBegin() <= element) && (element < StorageEnd()));
+  return element - StorageBegin();
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::ElementAt(
+    size_t index) const {
+  VIXL_ASSERT(
+      (IsUsingVector() && (index < vector_->size())) || (index < size_));
+  return StorageBegin() + index;
+}
+
+template<TEMPLATE_INVALSET_P_DECL>
+ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::ElementAt(size_t index) {
+  VIXL_ASSERT(
+      (IsUsingVector() && (index < vector_->size())) || (index < size_));
+  return StorageBegin() + index;
+}
+
+template<TEMPLATE_INVALSET_P_DECL>
+const ElementType* InvalSet<TEMPLATE_INVALSET_P_DEF>::FirstValidElement(
+    const ElementType* from, const ElementType* end) {
+  while ((from < end) && !IsValid(*from)) {
+    from++;
+  }
+  return from;
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::CacheMinElement() {
+  VIXL_ASSERT(monitor() == 0);
+  VIXL_ASSERT(!empty());
+
+  if (valid_cached_min_) {
+    return;
+  }
+
+  if (sorted_) {
+    const ElementType* min = FirstValidElement(StorageBegin(), StorageEnd());
+    cached_min_index_ = ElementIndex(min);
+    cached_min_key_ = Key(*min);
+    valid_cached_min_ = true;
+  } else {
+    Sort(kHardSort);
+  }
+  VIXL_ASSERT(valid_cached_min_);
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+bool InvalSet<TEMPLATE_INVALSET_P_DEF>::ShouldReclaimMemory() const {
+  if (!IsUsingVector()) {
+    return false;
+  }
+  size_t n_invalid_elements = vector_->size() - size_;
+  return (n_invalid_elements > RECLAIM_FROM) &&
+         (n_invalid_elements > vector_->size() / RECLAIM_FACTOR);
+}
+
+
+template<TEMPLATE_INVALSET_P_DECL>
+void InvalSet<TEMPLATE_INVALSET_P_DEF>::ReclaimMemory() {
+  VIXL_ASSERT(monitor() == 0);
+  Clean();
+}
+
+
+template<class S>
+InvalSetIterator<S>::InvalSetIterator(S* inval_set)
+    : using_vector_((inval_set != NULL) && inval_set->IsUsingVector()),
+      index_(0),
+      inval_set_(inval_set) {
+  if (inval_set != NULL) {
+    inval_set->Sort(S::kSoftSort);
+#ifdef VIXL_DEBUG
+    inval_set->Acquire();
+#endif
+    if (using_vector_) {
+      iterator_ = typename std::vector<ElementType>::iterator(
+          inval_set_->vector_->begin());
+    }
+    MoveToValidElement();
+  }
+}
+
+
+template<class S>
+InvalSetIterator<S>::~InvalSetIterator() {
+#ifdef VIXL_DEBUG
+  if (inval_set_ != NULL) {
+    inval_set_->Release();
+  }
+#endif
+}
+
+
+template<class S>
+typename S::_ElementType* InvalSetIterator<S>::Current() const {
+  VIXL_ASSERT(!Done());
+  if (using_vector_) {
+    return &(*iterator_);
+  } else {
+    return &(inval_set_->preallocated_[index_]);
+  }
+}
+
+
+template<class S>
+void InvalSetIterator<S>::Advance() {
+  VIXL_ASSERT(!Done());
+  if (using_vector_) {
+    iterator_++;
+#ifdef VIXL_DEBUG
+    index_++;
+#endif
+    MoveToValidElement();
+  } else {
+    index_++;
+  }
+}
+
+
+template<class S>
+bool InvalSetIterator<S>::Done() const {
+  if (using_vector_) {
+    bool done = (iterator_ == inval_set_->vector_->end());
+    VIXL_ASSERT(done == (index_ == inval_set_->size()));
+    return done;
+  } else {
+    return index_ == inval_set_->size();
+  }
+}
+
+
+template<class S>
+void InvalSetIterator<S>::Finish() {
+  VIXL_ASSERT(inval_set_->sorted_);
+  if (using_vector_) {
+    iterator_ = inval_set_->vector_->end();
+  }
+  index_ = inval_set_->size();
+}
+
+
+template<class S>
+void InvalSetIterator<S>::DeleteCurrentAndAdvance() {
+  if (using_vector_) {
+    inval_set_->EraseInternal(&(*iterator_));
+    MoveToValidElement();
+  } else {
+    inval_set_->EraseInternal(inval_set_->preallocated_ + index_);
+  }
+}
+
+
+template<class S>
+bool InvalSetIterator<S>::IsValid(const ElementType& element) {
+  return S::IsValid(element);
+}
+
+
+template<class S>
+typename S::_KeyType InvalSetIterator<S>::Key(const ElementType& element) {
+  return S::Key(element);
+}
+
+
+template<class S>
+void InvalSetIterator<S>::MoveToValidElement() {
+  if (using_vector_) {
+    while ((iterator_ != inval_set_->vector_->end()) && !IsValid(*iterator_)) {
+      iterator_++;
+    }
+  } else {
+    VIXL_ASSERT(inval_set_->empty() || IsValid(inval_set_->preallocated_[0]));
+    // Nothing to do.
+  }
+}
+
+#undef TEMPLATE_INVALSET_P_DECL
+#undef TEMPLATE_INVALSET_P_DEF
+
+}  // namespace vixl
+
+#endif  // VIXL_INVALSET_H_
diff --git a/disas/libvixl/platform.h b/disas/libvixl/vixl/platform.h
index de2b110cca..ab588f07f5 100644
--- a/disas/libvixl/platform.h
+++ b/disas/libvixl/vixl/platform.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2014, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
diff --git a/disas/libvixl/vixl/utils.cc b/disas/libvixl/vixl/utils.cc
new file mode 100644
index 0000000000..3b8bd75fba
--- /dev/null
+++ b/disas/libvixl/vixl/utils.cc
@@ -0,0 +1,142 @@
+// 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.
+
+#include "vixl/utils.h"
+#include <stdio.h>
+
+namespace vixl {
+
+uint32_t float_to_rawbits(float value) {
+  uint32_t bits = 0;
+  memcpy(&bits, &value, 4);
+  return bits;
+}
+
+
+uint64_t double_to_rawbits(double value) {
+  uint64_t bits = 0;
+  memcpy(&bits, &value, 8);
+  return bits;
+}
+
+
+float rawbits_to_float(uint32_t bits) {
+  float value = 0.0;
+  memcpy(&value, &bits, 4);
+  return value;
+}
+
+
+double rawbits_to_double(uint64_t bits) {
+  double value = 0.0;
+  memcpy(&value, &bits, 8);
+  return value;
+}
+
+
+uint32_t float_sign(float val) {
+  uint32_t rawbits = float_to_rawbits(val);
+  return unsigned_bitextract_32(31, 31, rawbits);
+}
+
+
+uint32_t float_exp(float val) {
+  uint32_t rawbits = float_to_rawbits(val);
+  return unsigned_bitextract_32(30, 23, rawbits);
+}
+
+
+uint32_t float_mantissa(float val) {
+  uint32_t rawbits = float_to_rawbits(val);
+  return unsigned_bitextract_32(22, 0, rawbits);
+}
+
+
+uint32_t double_sign(double val) {
+  uint64_t rawbits = double_to_rawbits(val);
+  return static_cast<uint32_t>(unsigned_bitextract_64(63, 63, rawbits));
+}
+
+
+uint32_t double_exp(double val) {
+  uint64_t rawbits = double_to_rawbits(val);
+  return static_cast<uint32_t>(unsigned_bitextract_64(62, 52, rawbits));
+}
+
+
+uint64_t double_mantissa(double val) {
+  uint64_t rawbits = double_to_rawbits(val);
+  return unsigned_bitextract_64(51, 0, rawbits);
+}
+
+
+float float_pack(uint32_t sign, uint32_t exp, uint32_t mantissa) {
+  uint32_t bits = (sign << 31) | (exp << 23) | mantissa;
+  return rawbits_to_float(bits);
+}
+
+
+double double_pack(uint64_t sign, uint64_t exp, uint64_t mantissa) {
+  uint64_t bits = (sign << 63) | (exp << 52) | mantissa;
+  return rawbits_to_double(bits);
+}
+
+
+int float16classify(float16 value) {
+  uint16_t exponent_max = (1 << 5) - 1;
+  uint16_t exponent_mask = exponent_max << 10;
+  uint16_t mantissa_mask = (1 << 10) - 1;
+
+  uint16_t exponent = (value & exponent_mask) >> 10;
+  uint16_t mantissa = value & mantissa_mask;
+  if (exponent == 0) {
+    if (mantissa == 0) {
+      return FP_ZERO;
+    }
+    return FP_SUBNORMAL;
+  } else if (exponent == exponent_max) {
+    if (mantissa == 0) {
+      return FP_INFINITE;
+    }
+    return FP_NAN;
+  }
+  return FP_NORMAL;
+}
+
+
+unsigned CountClearHalfWords(uint64_t imm, unsigned reg_size) {
+  VIXL_ASSERT((reg_size % 8) == 0);
+  int count = 0;
+  for (unsigned i = 0; i < (reg_size / 16); i++) {
+    if ((imm & 0xffff) == 0) {
+      count++;
+    }
+    imm >>= 16;
+  }
+  return count;
+}
+
+}  // namespace vixl
diff --git a/disas/libvixl/utils.h b/disas/libvixl/vixl/utils.h
index b4406263ac..5ab134e240 100644
--- a/disas/libvixl/utils.h
+++ b/disas/libvixl/vixl/utils.h
@@ -1,4 +1,4 @@
-// Copyright 2013, ARM Limited
+// Copyright 2015, ARM Limited
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
@@ -27,16 +27,17 @@
 #ifndef VIXL_UTILS_H
 #define VIXL_UTILS_H
 
-#include <math.h>
 #include <string.h>
-#include "globals.h"
+#include <cmath>
+#include "vixl/globals.h"
+#include "vixl/compiler-intrinsics.h"
 
 namespace vixl {
 
 // Macros for compile-time format checking.
-#if defined(__GNUC__)
+#if GCC_VERSION_OR_NEWER(4, 4, 0)
 #define PRINTF_CHECK(format_index, varargs_index) \
-  __attribute__((format(printf, format_index, varargs_index)))
+  __attribute__((format(gnu_printf, format_index, varargs_index)))
 #else
 #define PRINTF_CHECK(format_index, varargs_index)
 #endif
@@ -53,9 +54,9 @@ inline bool is_uintn(unsigned n, int64_t x) {
   return !(x >> n);
 }
 
-inline unsigned truncate_to_intn(unsigned n, int64_t x) {
+inline uint32_t truncate_to_intn(unsigned n, int64_t x) {
   VIXL_ASSERT((0 < n) && (n < 64));
-  return (x & ((INT64_C(1) << n) - 1));
+  return static_cast<uint32_t>(x & ((INT64_C(1) << n) - 1));
 }
 
 #define INT_1_TO_63_LIST(V)                                                    \
@@ -73,7 +74,7 @@ inline bool is_int##N(int64_t x) { return is_intn(N, x); }
 #define DECLARE_IS_UINT_N(N)                                                   \
 inline bool is_uint##N(int64_t x) { return is_uintn(N, x); }
 #define DECLARE_TRUNCATE_TO_INT_N(N)                                           \
-inline int truncate_to_int##N(int x) { return truncate_to_intn(N, x); }
+inline uint32_t truncate_to_int##N(int x) { return truncate_to_intn(N, x); }
 INT_1_TO_63_LIST(DECLARE_IS_INT_N)
 INT_1_TO_63_LIST(DECLARE_IS_UINT_N)
 INT_1_TO_63_LIST(DECLARE_TRUNCATE_TO_INT_N)
@@ -104,12 +105,24 @@ uint64_t double_to_rawbits(double value);
 float rawbits_to_float(uint32_t bits);
 double rawbits_to_double(uint64_t bits);
 
+uint32_t float_sign(float val);
+uint32_t float_exp(float val);
+uint32_t float_mantissa(float val);
+uint32_t double_sign(double val);
+uint32_t double_exp(double val);
+uint64_t double_mantissa(double val);
+
+float float_pack(uint32_t sign, uint32_t exp, uint32_t mantissa);
+double double_pack(uint64_t sign, uint64_t exp, uint64_t mantissa);
+
+// An fpclassify() function for 16-bit half-precision floats.
+int float16classify(float16 value);
 
 // NaN tests.
 inline bool IsSignallingNaN(double num) {
   const uint64_t kFP64QuietNaNMask = UINT64_C(0x0008000000000000);
   uint64_t raw = double_to_rawbits(num);
-  if (isnan(num) && ((raw & kFP64QuietNaNMask) == 0)) {
+  if (std::isnan(num) && ((raw & kFP64QuietNaNMask) == 0)) {
     return true;
   }
   return false;
@@ -119,30 +132,37 @@ inline bool IsSignallingNaN(double num) {
 inline bool IsSignallingNaN(float num) {
   const uint32_t kFP32QuietNaNMask = 0x00400000;
   uint32_t raw = float_to_rawbits(num);
-  if (isnan(num) && ((raw & kFP32QuietNaNMask) == 0)) {
+  if (std::isnan(num) && ((raw & kFP32QuietNaNMask) == 0)) {
     return true;
   }
   return false;
 }
 
 
+inline bool IsSignallingNaN(float16 num) {
+  const uint16_t kFP16QuietNaNMask = 0x0200;
+  return (float16classify(num) == FP_NAN) &&
+         ((num & kFP16QuietNaNMask) == 0);
+}
+
+
 template <typename T>
 inline bool IsQuietNaN(T num) {
-  return isnan(num) && !IsSignallingNaN(num);
+  return std::isnan(num) && !IsSignallingNaN(num);
 }
 
 
 // Convert the NaN in 'num' to a quiet NaN.
 inline double ToQuietNaN(double num) {
   const uint64_t kFP64QuietNaNMask = UINT64_C(0x0008000000000000);
-  VIXL_ASSERT(isnan(num));
+  VIXL_ASSERT(std::isnan(num));
   return rawbits_to_double(double_to_rawbits(num) | kFP64QuietNaNMask);
 }
 
 
 inline float ToQuietNaN(float num) {
   const uint32_t kFP32QuietNaNMask = 0x00400000;
-  VIXL_ASSERT(isnan(num));
+  VIXL_ASSERT(std::isnan(num));
   return rawbits_to_float(float_to_rawbits(num) | kFP32QuietNaNMask);
 }
 
@@ -158,16 +178,71 @@ inline float FusedMultiplyAdd(float op1, float op2, float a) {
 }
 
 
-// Bit counting.
-int CountLeadingZeros(uint64_t value, int width);
-int CountLeadingSignBits(int64_t value, int width);
-int CountTrailingZeros(uint64_t value, int width);
-int CountSetBits(uint64_t value, int width);
-uint64_t LowestSetBit(uint64_t value);
-bool IsPowerOf2(int64_t value);
+inline uint64_t LowestSetBit(uint64_t value) {
+  return value & -value;
+}
+
+
+template<typename T>
+inline int HighestSetBitPosition(T value) {
+  VIXL_ASSERT(value != 0);
+  return (sizeof(value) * 8 - 1) - CountLeadingZeros(value);
+}
+
+
+template<typename V>
+inline int WhichPowerOf2(V value) {
+  VIXL_ASSERT(IsPowerOf2(value));
+  return CountTrailingZeros(value);
+}
+
 
 unsigned CountClearHalfWords(uint64_t imm, unsigned reg_size);
 
+
+template <typename T>
+T ReverseBits(T value) {
+  VIXL_ASSERT((sizeof(value) == 1) || (sizeof(value) == 2) ||
+              (sizeof(value) == 4) || (sizeof(value) == 8));
+  T result = 0;
+  for (unsigned i = 0; i < (sizeof(value) * 8); i++) {
+    result = (result << 1) | (value & 1);
+    value >>= 1;
+  }
+  return result;
+}
+
+
+template <typename T>
+T ReverseBytes(T value, int block_bytes_log2) {
+  VIXL_ASSERT((sizeof(value) == 4) || (sizeof(value) == 8));
+  VIXL_ASSERT((1U << block_bytes_log2) <= sizeof(value));
+  // Split the 64-bit value into an 8-bit array, where b[0] is the least
+  // significant byte, and b[7] is the most significant.
+  uint8_t bytes[8];
+  uint64_t mask = UINT64_C(0xff00000000000000);
+  for (int i = 7; i >= 0; i--) {
+    bytes[i] = (static_cast<uint64_t>(value) & mask) >> (i * 8);
+    mask >>= 8;
+  }
+
+  // Permutation tables for REV instructions.
+  //  permute_table[0] is used by REV16_x, REV16_w
+  //  permute_table[1] is used by REV32_x, REV_w
+  //  permute_table[2] is used by REV_x
+  VIXL_ASSERT((0 < block_bytes_log2) && (block_bytes_log2 < 4));
+  static const uint8_t permute_table[3][8] = { {6, 7, 4, 5, 2, 3, 0, 1},
+                                               {4, 5, 6, 7, 0, 1, 2, 3},
+                                               {0, 1, 2, 3, 4, 5, 6, 7} };
+  T result = 0;
+  for (int i = 0; i < 8; i++) {
+    result <<= 8;
+    result |= bytes[permute_table[block_bytes_log2 - 1][i]];
+  }
+  return result;
+}
+
+
 // Pointer alignment
 // TODO: rename/refactor to make it specific to instructions.
 template<typename T>