disas/libvixl: Update to libvixl 1.6

Update our copy of libvixl to upstream 1.6. There are no
changes of any particular interest to QEMU, so this is simply
keeping up with current upstream.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1412091418-25744-1-git-send-email-peter.maydell@linaro.org

12 files changed, 715 insertions, 332 deletions

diff --git a/disas/arm-a64.cc b/disas/arm-a64.cc
index 162be0c420..ca29f6f253 100644
--- a/disas/arm-a64.cc
+++ b/disas/arm-a64.cc
@@ -39,7 +39,7 @@ public:
     ~QEMUDisassembler() { }
 
 protected:
-    void ProcessOutput(Instruction *instr) {
+    virtual void ProcessOutput(const Instruction *instr) {
         fprintf(stream_, "%08" PRIx32 "      %s",
                 instr->InstructionBits(), GetOutput());
     }
diff --git a/disas/libvixl/README b/disas/libvixl/README
index 8301996e68..cba31b458b 100644
--- a/disas/libvixl/README
+++ b/disas/libvixl/README
@@ -2,7 +2,7 @@
 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 1.5).
+taken from libvixl 1.6).
 Bugfixes should preferably be sent upstream initially.
 
 The disassembler does not currently support the entire A64 instruction
diff --git a/disas/libvixl/a64/assembler-a64.h b/disas/libvixl/a64/assembler-a64.h
index cc0b758dd3..16a704b7d4 100644
--- a/disas/libvixl/a64/assembler-a64.h
+++ b/disas/libvixl/a64/assembler-a64.h
@@ -32,6 +32,7 @@
 
 #include "globals.h"
 #include "utils.h"
+#include "code-buffer.h"
 #include "a64/instructions-a64.h"
 
 namespace vixl {
@@ -168,6 +169,11 @@ class CPURegister {
     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;
@@ -191,12 +197,12 @@ class CPURegister {
 
 class Register : public CPURegister {
  public:
-  explicit Register() : CPURegister() {}
+  Register() : CPURegister() {}
   inline explicit Register(const CPURegister& other)
       : CPURegister(other.code(), other.size(), other.type()) {
     VIXL_ASSERT(IsValidRegister());
   }
-  explicit Register(unsigned code, unsigned size)
+  Register(unsigned code, unsigned size)
       : CPURegister(code, size, kRegister) {}
 
   bool IsValid() const {
@@ -536,7 +542,7 @@ class Operand {
 class MemOperand {
  public:
   explicit MemOperand(Register base,
-                      ptrdiff_t offset = 0,
+                      int64_t offset = 0,
                       AddrMode addrmode = Offset);
   explicit MemOperand(Register base,
                       Register regoffset,
@@ -552,7 +558,7 @@ class MemOperand {
 
   const Register& base() const { return base_; }
   const Register& regoffset() const { return regoffset_; }
-  ptrdiff_t offset() const { return offset_; }
+  int64_t offset() const { return offset_; }
   AddrMode addrmode() const { return addrmode_; }
   Shift shift() const { return shift_; }
   Extend extend() const { return extend_; }
@@ -565,7 +571,7 @@ class MemOperand {
  private:
   Register base_;
   Register regoffset_;
-  ptrdiff_t offset_;
+  int64_t offset_;
   AddrMode addrmode_;
   Shift shift_;
   Extend extend_;
@@ -680,32 +686,80 @@ class Label {
 };
 
 
-// TODO: Obtain better values for these, based on real-world data.
-const int kLiteralPoolCheckInterval = 4 * KBytes;
-const int kRecommendedLiteralPoolRange = 2 * kLiteralPoolCheckInterval;
+// 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; }
 
-// Control whether a branch over the literal pool should also be emitted. This
-// is needed if the literal pool has to be emitted in the middle of the JITted
-// code.
-enum LiteralPoolEmitOption {
-  JumpRequired,
-  NoJumpRequired
-};
+ 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_;
 
-// Literal pool entry.
-class Literal {
- public:
-  Literal(Instruction* pc, uint64_t imm, unsigned size)
-      : pc_(pc), value_(imm), size_(size) {}
+  friend class Assembler;
+};
 
- private:
-  Instruction* pc_;
-  int64_t value_;
-  unsigned size_;
 
-  friend class Assembler;
+template <typename T>
+class Literal : public RawLiteral {
+ public:
+  explicit Literal(T value) {
+    size_ = sizeof(value);
+    memcpy(&raw_value_, &value, sizeof(value));
+  }
 };
 
 
@@ -750,7 +804,9 @@ enum LoadStoreScalingOption {
 // Assembler.
 class Assembler {
  public:
-  Assembler(byte* buffer, unsigned buffer_size,
+  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:
@@ -763,9 +819,6 @@ class Assembler {
 
   // Start generating code from the beginning of the buffer, discarding any code
   // and data that has already been emitted into the buffer.
-  //
-  // In order to avoid any accidental transfer of state, Reset ASSERTs that the
-  // constant pool is not blocked.
   void Reset();
 
   // Finalize a code buffer of generated instructions. This function must be
@@ -776,13 +829,47 @@ class Assembler {
   // 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>
+  inline 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>
   inline T GetLabelAddress(const Label * label) {
     VIXL_ASSERT(label->IsBound());
     VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
-    VIXL_STATIC_ASSERT(sizeof(*buffer_) == 1);
-    return reinterpret_cast<T>(buffer_ + label->location());
+    return GetOffsetAddress<T>(label->location());
+  }
+
+  // Return the address of the cursor.
+  template <typename T>
+  inline 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>
+  inline T GetStartAddress() {
+    VIXL_STATIC_ASSERT(sizeof(T) >= sizeof(uintptr_t));
+    return GetOffsetAddress<T>(0);
   }
 
   // Instruction set functions.
@@ -1324,14 +1411,17 @@ class Assembler {
   void stnp(const CPURegister& rt, const CPURegister& rt2,
             const MemOperand& dst);
 
-  // Load literal to register.
-  void ldr(const Register& rt, uint64_t imm);
+  // 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 double precision floating point literal to FP register.
-  void ldr(const FPRegister& ft, double imm);
+  // Load integer or FP register from pc + imm19 << 2.
+  void ldr(const CPURegister& rt, int imm19);
 
-  // Load single precision floating point literal to FP register.
-  void ldr(const FPRegister& ft, float imm);
+  // 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);
@@ -1618,25 +1708,26 @@ class Assembler {
   inline void dci(Instr raw_inst) { Emit(raw_inst); }
 
   // Emit 32 bits of data into the instruction stream.
-  inline void dc32(uint32_t data) { EmitData(&data, sizeof(data)); }
+  inline void dc32(uint32_t data) {
+    VIXL_ASSERT(buffer_monitor_ > 0);
+    buffer_->Emit32(data);
+  }
 
   // Emit 64 bits of data into the instruction stream.
-  inline void dc64(uint64_t data) { EmitData(&data, sizeof(data)); }
+  inline 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 (pc_) is then aligned correctly for
+  // character. The instruction pointer is then aligned correctly for
   // subsequent instructions.
-  void EmitStringData(const char * string) {
+  void EmitString(const char * string) {
     VIXL_ASSERT(string != NULL);
+    VIXL_ASSERT(buffer_monitor_ > 0);
 
-    size_t len = strlen(string) + 1;
-    EmitData(string, len);
-
-    // Pad with NULL characters until pc_ is aligned.
-    const char pad[] = {'\0', '\0', '\0', '\0'};
-    VIXL_STATIC_ASSERT(sizeof(pad) == kInstructionSize);
-    Instruction* next_pc = AlignUp(pc_, kInstructionSize);
-    EmitData(&pad, next_pc - pc_);
+    buffer_->EmitString(string);
+    buffer_->Align();
   }
 
   // Code generation helpers.
@@ -1912,43 +2003,39 @@ class Assembler {
     return scale << FPScale_offset;
   }
 
-  // Size of the code generated in bytes
-  size_t SizeOfCodeGenerated() const {
-    VIXL_ASSERT((pc_ >= buffer_) && (pc_ < (buffer_ + buffer_size_)));
-    return pc_ - buffer_;
-  }
-
   // Size of the code generated since label to the current position.
   size_t SizeOfCodeGeneratedSince(Label* label) const {
-    size_t pc_offset = SizeOfCodeGenerated();
-
     VIXL_ASSERT(label->IsBound());
-    VIXL_ASSERT(pc_offset >= static_cast<size_t>(label->location()));
-    VIXL_ASSERT(pc_offset < buffer_size_);
-
-    return pc_offset - label->location();
+    return buffer_->OffsetFrom(label->location());
   }
 
+  size_t BufferCapacity() const { return buffer_->capacity(); }
 
-  inline void BlockLiteralPool() {
-    literal_pool_monitor_++;
-  }
+  size_t RemainingBufferSpace() const { return buffer_->RemainingBytes(); }
 
-  inline void ReleaseLiteralPool() {
-    if (--literal_pool_monitor_ == 0) {
-      // Has the literal pool been blocked for too long?
-      VIXL_ASSERT(literals_.empty() ||
-             (pc_ < (literals_.back()->pc_ + kMaxLoadLiteralRange)));
+  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);
     }
   }
 
-  inline bool IsLiteralPoolBlocked() {
-    return literal_pool_monitor_ != 0;
+#ifdef DEBUG
+  void AcquireBuffer() {
+    VIXL_ASSERT(buffer_monitor_ >= 0);
+    buffer_monitor_++;
   }
 
-  void CheckLiteralPool(LiteralPoolEmitOption option = JumpRequired);
-  void EmitLiteralPool(LiteralPoolEmitOption option = NoJumpRequired);
-  size_t LiteralPoolSize();
+  void ReleaseBuffer() {
+    buffer_monitor_--;
+    VIXL_ASSERT(buffer_monitor_ >= 0);
+  }
+#endif
 
   inline PositionIndependentCodeOption pic() {
     return pic_;
@@ -1959,22 +2046,30 @@ class Assembler {
            (pic() == PositionDependentCode);
   }
 
- protected:
-  inline const Register& AppropriateZeroRegFor(const CPURegister& reg) const {
+  static inline 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);
-  static bool IsImmLSUnscaled(ptrdiff_t offset);
-  static bool IsImmLSScaled(ptrdiff_t offset, LSDataSize size);
+  static bool IsImmLSUnscaled(int64_t offset);
+  static bool IsImmLSScaled(int64_t offset, LSDataSize size);
+
+  void LoadStorePair(const CPURegister& rt,
+                     const CPURegister& rt2,
+                     const MemOperand& addr,
+                     LoadStorePairOp op);
+  static bool IsImmLSPair(int64_t offset, LSDataSize size);
 
+  // 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,
+               const Operand operand,
                LogicalOp op);
   void LogicalImmediate(const Register& rd,
                         const Register& rn,
@@ -2035,6 +2130,7 @@ class Assembler {
     const CPURegister& rt, const CPURegister& rt2);
   static LoadStorePairNonTemporalOp StorePairNonTemporalOpFor(
     const CPURegister& rt, const CPURegister& rt2);
+  static LoadLiteralOp LoadLiteralOpFor(const CPURegister& rt);
 
 
  private:
@@ -2053,10 +2149,6 @@ class Assembler {
                                 const Operand& operand,
                                 FlagsUpdate S,
                                 Instr op);
-  void LoadStorePair(const CPURegister& rt,
-                     const CPURegister& rt2,
-                     const MemOperand& addr,
-                     LoadStorePairOp op);
   void LoadStorePairNonTemporal(const CPURegister& rt,
                                 const CPURegister& rt2,
                                 const MemOperand& addr,
@@ -2088,8 +2180,6 @@ class Assembler {
                                const FPRegister& fa,
                                FPDataProcessing3SourceOp op);
 
-  void RecordLiteral(int64_t imm, unsigned size);
-
   // 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.
@@ -2098,79 +2188,102 @@ class Assembler {
   ptrdiff_t LinkAndGetPageOffsetTo(Label * label);
 
   // A common implementation for the LinkAndGet<Type>OffsetTo helpers.
-  template <int element_size>
+  template <int element_shift>
   ptrdiff_t LinkAndGetOffsetTo(Label* label);
 
-  // Emit the instruction at pc_.
+  // 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(*pc_) == 1);
     VIXL_STATIC_ASSERT(sizeof(instruction) == kInstructionSize);
-    VIXL_ASSERT((pc_ + sizeof(instruction)) <= (buffer_ + buffer_size_));
-
-#ifdef DEBUG
-    finalized_ = false;
-#endif
-
-    memcpy(pc_, &instruction, sizeof(instruction));
-    pc_ += sizeof(instruction);
-    CheckBufferSpace();
+    VIXL_ASSERT(buffer_monitor_ > 0);
+    buffer_->Emit32(instruction);
   }
 
-  // Emit data inline in the instruction stream.
-  void EmitData(void const * data, unsigned size) {
-    VIXL_STATIC_ASSERT(sizeof(*pc_) == 1);
-    VIXL_ASSERT((pc_ + size) <= (buffer_ + buffer_size_));
+  // Buffer where the code is emitted.
+  CodeBuffer* buffer_;
+  PositionIndependentCodeOption pic_;
 
 #ifdef DEBUG
-    finalized_ = false;
+  int64_t buffer_monitor_;
 #endif
+};
 
-    // TODO: Record this 'instruction' as data, so that it can be disassembled
-    // correctly.
-    memcpy(pc_, data, size);
-    pc_ += size;
-    CheckBufferSpace();
-  }
-
-  inline void CheckBufferSpace() {
-    VIXL_ASSERT(pc_ < (buffer_ + buffer_size_));
-    if (pc_ > next_literal_pool_check_) {
-      CheckLiteralPool();
-    }
-  }
-
-  // The buffer into which code and relocation info are generated.
-  Instruction* buffer_;
-  // Buffer size, in bytes.
-  size_t buffer_size_;
-  Instruction* pc_;
-  std::list<Literal*> literals_;
-  Instruction* next_literal_pool_check_;
-  unsigned literal_pool_monitor_;
 
-  PositionIndependentCodeOption pic_;
+// 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
+  };
 
-  friend class Label;
-  friend class BlockLiteralPoolScope;
+  // 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 DEBUG
-  bool finalized_;
+    assm->bind(&start_);
+    size_ = size;
+    assert_policy_ = assert_policy;
+    assm->AcquireBuffer();
+#else
+    USE(assert_policy);
 #endif
-};
+  }
 
-class BlockLiteralPoolScope {
- public:
-  explicit BlockLiteralPoolScope(Assembler* assm) : assm_(assm) {
-    assm_->BlockLiteralPool();
+  // This is a shortcut for CodeBufferCheckScope(assm, 0, kNoCheck, kNoAssert).
+  explicit CodeBufferCheckScope(Assembler* assm) : assm_(assm) {
+#ifdef DEBUG
+    size_ = 0;
+    assert_policy_ = kNoAssert;
+    assm->AcquireBuffer();
+#endif
   }
 
-  ~BlockLiteralPoolScope() {
-    assm_->ReleaseLiteralPool();
+  ~CodeBufferCheckScope() {
+#ifdef 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
   }
 
- private:
+ protected:
   Assembler* assm_;
+#ifdef DEBUG
+  Label start_;
+  size_t size_;
+  AssertPolicy assert_policy_;
+#endif
 };
+
 }  // namespace vixl
 
 #endif  // VIXL_A64_ASSEMBLER_A64_H_
diff --git a/disas/libvixl/a64/decoder-a64.cc b/disas/libvixl/a64/decoder-a64.cc
index 5831b73024..82591ca309 100644
--- a/disas/libvixl/a64/decoder-a64.cc
+++ b/disas/libvixl/a64/decoder-a64.cc
@@ -29,8 +29,8 @@
 #include "a64/decoder-a64.h"
 
 namespace vixl {
-// Top-level instruction decode function.
-void Decoder::Decode(Instruction *instr) {
+
+void Decoder::DecodeInstruction(const Instruction *instr) {
   if (instr->Bits(28, 27) == 0) {
     VisitUnallocated(instr);
   } else {
@@ -109,20 +109,17 @@ void Decoder::Decode(Instruction *instr) {
 }
 
 void Decoder::AppendVisitor(DecoderVisitor* new_visitor) {
-  visitors_.remove(new_visitor);
-  visitors_.push_front(new_visitor);
+  visitors_.push_back(new_visitor);
 }
 
 
 void Decoder::PrependVisitor(DecoderVisitor* new_visitor) {
-  visitors_.remove(new_visitor);
-  visitors_.push_back(new_visitor);
+  visitors_.push_front(new_visitor);
 }
 
 
 void Decoder::InsertVisitorBefore(DecoderVisitor* new_visitor,
                                   DecoderVisitor* registered_visitor) {
-  visitors_.remove(new_visitor);
   std::list<DecoderVisitor*>::iterator it;
   for (it = visitors_.begin(); it != visitors_.end(); it++) {
     if (*it == registered_visitor) {
@@ -139,7 +136,6 @@ void Decoder::InsertVisitorBefore(DecoderVisitor* new_visitor,
 
 void Decoder::InsertVisitorAfter(DecoderVisitor* new_visitor,
                                  DecoderVisitor* registered_visitor) {
-  visitors_.remove(new_visitor);
   std::list<DecoderVisitor*>::iterator it;
   for (it = visitors_.begin(); it != visitors_.end(); it++) {
     if (*it == registered_visitor) {
@@ -160,7 +156,7 @@ void Decoder::RemoveVisitor(DecoderVisitor* visitor) {
 }
 
 
-void Decoder::DecodePCRelAddressing(Instruction* instr) {
+void Decoder::DecodePCRelAddressing(const Instruction* instr) {
   VIXL_ASSERT(instr->Bits(27, 24) == 0x0);
   // We know bit 28 is set, as <b28:b27> = 0 is filtered out at the top level
   // decode.
@@ -169,7 +165,7 @@ void Decoder::DecodePCRelAddressing(Instruction* instr) {
 }
 
 
-void Decoder::DecodeBranchSystemException(Instruction* instr) {
+void Decoder::DecodeBranchSystemException(const Instruction* instr) {
   VIXL_ASSERT((instr->Bits(27, 24) == 0x4) ||
               (instr->Bits(27, 24) == 0x5) ||
               (instr->Bits(27, 24) == 0x6) ||
@@ -270,7 +266,7 @@ void Decoder::DecodeBranchSystemException(Instruction* instr) {
 }
 
 
-void Decoder::DecodeLoadStore(Instruction* instr) {
+void Decoder::DecodeLoadStore(const Instruction* instr) {
   VIXL_ASSERT((instr->Bits(27, 24) == 0x8) ||
               (instr->Bits(27, 24) == 0x9) ||
               (instr->Bits(27, 24) == 0xC) ||
@@ -388,7 +384,7 @@ void Decoder::DecodeLoadStore(Instruction* instr) {
 }
 
 
-void Decoder::DecodeLogical(Instruction* instr) {
+void Decoder::DecodeLogical(const Instruction* instr) {
   VIXL_ASSERT(instr->Bits(27, 24) == 0x2);
 
   if (instr->Mask(0x80400000) == 0x00400000) {
@@ -407,7 +403,7 @@ void Decoder::DecodeLogical(Instruction* instr) {
 }
 
 
-void Decoder::DecodeBitfieldExtract(Instruction* instr) {
+void Decoder::DecodeBitfieldExtract(const Instruction* instr) {
   VIXL_ASSERT(instr->Bits(27, 24) == 0x3);
 
   if ((instr->Mask(0x80400000) == 0x80000000) ||
@@ -432,7 +428,7 @@ void Decoder::DecodeBitfieldExtract(Instruction* instr) {
 }
 
 
-void Decoder::DecodeAddSubImmediate(Instruction* instr) {
+void Decoder::DecodeAddSubImmediate(const Instruction* instr) {
   VIXL_ASSERT(instr->Bits(27, 24) == 0x1);
   if (instr->Bit(23) == 1) {
     VisitUnallocated(instr);
@@ -442,7 +438,7 @@ void Decoder::DecodeAddSubImmediate(Instruction* instr) {
 }
 
 
-void Decoder::DecodeDataProcessing(Instruction* instr) {
+void Decoder::DecodeDataProcessing(const Instruction* instr) {
   VIXL_ASSERT((instr->Bits(27, 24) == 0xA) ||
               (instr->Bits(27, 24) == 0xB));
 
@@ -557,7 +553,7 @@ void Decoder::DecodeDataProcessing(Instruction* instr) {
 }
 
 
-void Decoder::DecodeFP(Instruction* instr) {
+void Decoder::DecodeFP(const Instruction* instr) {
   VIXL_ASSERT((instr->Bits(27, 24) == 0xE) ||
               (instr->Bits(27, 24) == 0xF));
 
@@ -684,14 +680,14 @@ void Decoder::DecodeFP(Instruction* instr) {
 }
 
 
-void Decoder::DecodeAdvSIMDLoadStore(Instruction* instr) {
+void Decoder::DecodeAdvSIMDLoadStore(const Instruction* instr) {
   // TODO: Implement Advanced SIMD load/store instruction decode.
   VIXL_ASSERT(instr->Bits(29, 25) == 0x6);
   VisitUnimplemented(instr);
 }
 
 
-void Decoder::DecodeAdvSIMDDataProcessing(Instruction* instr) {
+void Decoder::DecodeAdvSIMDDataProcessing(const Instruction* instr) {
   // TODO: Implement Advanced SIMD data processing instruction decode.
   VIXL_ASSERT(instr->Bits(27, 25) == 0x7);
   VisitUnimplemented(instr);
@@ -699,7 +695,7 @@ void Decoder::DecodeAdvSIMDDataProcessing(Instruction* instr) {
 
 
 #define DEFINE_VISITOR_CALLERS(A)                                              \
-  void Decoder::Visit##A(Instruction *instr) {                                 \
+  void Decoder::Visit##A(const Instruction *instr) {                           \
     VIXL_ASSERT(instr->Mask(A##FMask) == A##Fixed);                            \
     std::list<DecoderVisitor*>::iterator it;                                   \
     for (it = visitors_.begin(); it != visitors_.end(); it++) {                \
diff --git a/disas/libvixl/a64/decoder-a64.h b/disas/libvixl/a64/decoder-a64.h
index 72c15196ce..172594c89b 100644
--- a/disas/libvixl/a64/decoder-a64.h
+++ b/disas/libvixl/a64/decoder-a64.h
@@ -88,112 +88,152 @@ namespace vixl {
 // must provide implementations for all of these functions.
 class DecoderVisitor {
  public:
-  #define DECLARE(A) virtual void Visit##A(Instruction* instr) = 0;
+  enum VisitorConstness {
+    kConstVisitor,
+    kNonConstVisitor
+  };
+  explicit DecoderVisitor(VisitorConstness constness = kConstVisitor)
+      : constness_(constness) {}
+
+  virtual ~DecoderVisitor() {}
+
+  #define DECLARE(A) virtual void Visit##A(const Instruction* instr) = 0;
   VISITOR_LIST(DECLARE)
   #undef DECLARE
 
-  virtual ~DecoderVisitor() {}
+  bool IsConstVisitor() const { return constness_ == kConstVisitor; }
+  Instruction* MutableInstruction(const Instruction* instr) {
+    VIXL_ASSERT(!IsConstVisitor());
+    return const_cast<Instruction*>(instr);
+  }
 
  private:
-  // Visitors are registered in a list.
-  std::list<DecoderVisitor*> visitors_;
-
-  friend class Decoder;
+  VisitorConstness constness_;
 };
 
 
-class Decoder: public DecoderVisitor {
+class Decoder {
  public:
   Decoder() {}
 
-  // Top-level instruction decoder function. Decodes an instruction and calls
-  // the visitor functions registered with the Decoder class.
-  void Decode(Instruction *instr);
+  // Top-level wrappers around the actual decoding function.
+  void Decode(const Instruction* instr) {
+    std::list<DecoderVisitor*>::iterator it;
+    for (it = visitors_.begin(); it != visitors_.end(); it++) {
+      VIXL_ASSERT((*it)->IsConstVisitor());
+    }
+    DecodeInstruction(instr);
+  }
+  void Decode(Instruction* instr) {
+    DecodeInstruction(const_cast<const Instruction*>(instr));
+  }
 
   // Register a new visitor class with the decoder.
   // Decode() will call the corresponding visitor method from all registered
   // visitor classes when decoding reaches the leaf node of the instruction
   // decode tree.
-  // Visitors are called in the order.
-  // A visitor can only be registered once.
-  // Registering an already registered visitor will update its position.
+  // Visitors are called in order.
+  // A visitor can be registered multiple times.
   //
   //   d.AppendVisitor(V1);
   //   d.AppendVisitor(V2);
-  //   d.PrependVisitor(V2);            // Move V2 at the start of the list.
-  //   d.InsertVisitorBefore(V3, V2);
-  //   d.AppendVisitor(V4);
-  //   d.AppendVisitor(V4);             // No effect.
+  //   d.PrependVisitor(V2);
+  //   d.AppendVisitor(V3);
   //
   //   d.Decode(i);
   //
-  // will call in order visitor methods in V3, V2, V1, V4.
+  // will call in order visitor methods in V2, V1, V2, V3.
   void AppendVisitor(DecoderVisitor* visitor);
   void PrependVisitor(DecoderVisitor* visitor);
+  // These helpers register `new_visitor` before or after the first instance of
+  // `registered_visiter` in the list.
+  // So if
+  //   V1, V2, V1, V2
+  // are registered in this order in the decoder, calls to
+  //   d.InsertVisitorAfter(V3, V1);
+  //   d.InsertVisitorBefore(V4, V2);
+  // will yield the order
+  //   V1, V3, V4, V2, V1, V2
+  //
+  // For more complex modifications of the order of registered visitors, one can
+  // directly access and modify the list of visitors via the `visitors()'
+  // accessor.
   void InsertVisitorBefore(DecoderVisitor* new_visitor,
                            DecoderVisitor* registered_visitor);
   void InsertVisitorAfter(DecoderVisitor* new_visitor,
                           DecoderVisitor* registered_visitor);
 
-  // Remove a previously registered visitor class from the list of visitors
-  // stored by the decoder.
+  // Remove all instances of a previously registered visitor class from the list
+  // of visitors stored by the decoder.
   void RemoveVisitor(DecoderVisitor* visitor);
 
-  #define DECLARE(A) void Visit##A(Instruction* instr);
+  #define DECLARE(A) void Visit##A(const Instruction* instr);
   VISITOR_LIST(DECLARE)
   #undef DECLARE
 
+
+  std::list<DecoderVisitor*>* visitors() { return &visitors_; }
+
  private:
+  // Decodes an instruction and calls the visitor functions registered with the
+  // Decoder class.
+  void DecodeInstruction(const Instruction* instr);
+
   // Decode the PC relative addressing instruction, and call the corresponding
   // visitors.
   // On entry, instruction bits 27:24 = 0x0.
-  void DecodePCRelAddressing(Instruction* instr);
+  void DecodePCRelAddressing(const Instruction* instr);
 
   // Decode the add/subtract immediate instruction, and call the correspoding
   // visitors.
   // On entry, instruction bits 27:24 = 0x1.
-  void DecodeAddSubImmediate(Instruction* instr);
+  void DecodeAddSubImmediate(const Instruction* instr);
 
   // Decode the branch, system command, and exception generation parts of
   // the instruction tree, and call the corresponding visitors.
   // On entry, instruction bits 27:24 = {0x4, 0x5, 0x6, 0x7}.
-  void DecodeBranchSystemException(Instruction* instr);
+  void DecodeBranchSystemException(const Instruction* instr);
 
   // Decode the load and store parts of the instruction tree, and call
   // the corresponding visitors.
   // On entry, instruction bits 27:24 = {0x8, 0x9, 0xC, 0xD}.
-  void DecodeLoadStore(Instruction* instr);
+  void DecodeLoadStore(const Instruction* instr);
 
   // Decode the logical immediate and move wide immediate parts of the
   // instruction tree, and call the corresponding visitors.
   // On entry, instruction bits 27:24 = 0x2.
-  void DecodeLogical(Instruction* instr);
+  void DecodeLogical(const Instruction* instr);
 
   // Decode the bitfield and extraction parts of the instruction tree,
   // and call the corresponding visitors.
   // On entry, instruction bits 27:24 = 0x3.
-  void DecodeBitfieldExtract(Instruction* instr);
+  void DecodeBitfieldExtract(const Instruction* instr);
 
   // Decode the data processing parts of the instruction tree, and call the
   // corresponding visitors.
   // On entry, instruction bits 27:24 = {0x1, 0xA, 0xB}.
-  void DecodeDataProcessing(Instruction* instr);
+  void DecodeDataProcessing(const Instruction* instr);
 
   // Decode the floating point parts of the instruction tree, and call the
   // corresponding visitors.
   // On entry, instruction bits 27:24 = {0xE, 0xF}.
-  void DecodeFP(Instruction* instr);
+  void DecodeFP(const Instruction* instr);
 
   // 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(Instruction* instr);
+  void DecodeAdvSIMDLoadStore(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(Instruction* instr);
+  void DecodeAdvSIMDDataProcessing(const Instruction* instr);
+
+ private:
+  // Visitors are registered in a list.
+  std::list<DecoderVisitor*> visitors_;
 };
+
 }  // namespace vixl
 
 #endif  // VIXL_A64_DECODER_A64_H_
diff --git a/disas/libvixl/a64/disasm-a64.cc b/disas/libvixl/a64/disasm-a64.cc
index 248ebfd436..e4a74aa57c 100644
--- a/disas/libvixl/a64/disasm-a64.cc
+++ b/disas/libvixl/a64/disasm-a64.cc
@@ -57,7 +57,7 @@ char* Disassembler::GetOutput() {
 }
 
 
-void Disassembler::VisitAddSubImmediate(Instruction* instr) {
+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;
@@ -102,7 +102,7 @@ void Disassembler::VisitAddSubImmediate(Instruction* instr) {
 }
 
 
-void Disassembler::VisitAddSubShifted(Instruction* instr) {
+void Disassembler::VisitAddSubShifted(const Instruction* instr) {
   bool rd_is_zr = RdIsZROrSP(instr);
   bool rn_is_zr = RnIsZROrSP(instr);
   const char *mnemonic = "";
@@ -149,7 +149,7 @@ void Disassembler::VisitAddSubShifted(Instruction* instr) {
 }
 
 
-void Disassembler::VisitAddSubExtended(Instruction* instr) {
+void Disassembler::VisitAddSubExtended(const Instruction* instr) {
   bool rd_is_zr = RdIsZROrSP(instr);
   const char *mnemonic = "";
   Extend mode = static_cast<Extend>(instr->ExtendMode());
@@ -187,7 +187,7 @@ void Disassembler::VisitAddSubExtended(Instruction* instr) {
 }
 
 
-void Disassembler::VisitAddSubWithCarry(Instruction* instr) {
+void Disassembler::VisitAddSubWithCarry(const Instruction* instr) {
   bool rn_is_zr = RnIsZROrSP(instr);
   const char *mnemonic = "";
   const char *form = "'Rd, 'Rn, 'Rm";
@@ -222,7 +222,7 @@ void Disassembler::VisitAddSubWithCarry(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLogicalImmediate(Instruction* instr) {
+void Disassembler::VisitLogicalImmediate(const Instruction* instr) {
   bool rd_is_zr = RdIsZROrSP(instr);
   bool rn_is_zr = RnIsZROrSP(instr);
   const char *mnemonic = "";
@@ -294,7 +294,7 @@ bool Disassembler::IsMovzMovnImm(unsigned reg_size, uint64_t value) {
 }
 
 
-void Disassembler::VisitLogicalShifted(Instruction* instr) {
+void Disassembler::VisitLogicalShifted(const Instruction* instr) {
   bool rd_is_zr = RdIsZROrSP(instr);
   bool rn_is_zr = RnIsZROrSP(instr);
   const char *mnemonic = "";
@@ -345,7 +345,7 @@ void Disassembler::VisitLogicalShifted(Instruction* instr) {
 }
 
 
-void Disassembler::VisitConditionalCompareRegister(Instruction* instr) {
+void Disassembler::VisitConditionalCompareRegister(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rn, 'Rm, 'INzcv, 'Cond";
 
@@ -360,7 +360,7 @@ void Disassembler::VisitConditionalCompareRegister(Instruction* instr) {
 }
 
 
-void Disassembler::VisitConditionalCompareImmediate(Instruction* instr) {
+void Disassembler::VisitConditionalCompareImmediate(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rn, 'IP, 'INzcv, 'Cond";
 
@@ -375,7 +375,7 @@ void Disassembler::VisitConditionalCompareImmediate(Instruction* instr) {
 }
 
 
-void Disassembler::VisitConditionalSelect(Instruction* instr) {
+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 = "";
@@ -428,7 +428,7 @@ void Disassembler::VisitConditionalSelect(Instruction* instr) {
 }
 
 
-void Disassembler::VisitBitfield(Instruction* instr) {
+void Disassembler::VisitBitfield(const Instruction* instr) {
   unsigned s = instr->ImmS();
   unsigned r = instr->ImmR();
   unsigned rd_size_minus_1 =
@@ -506,7 +506,7 @@ void Disassembler::VisitBitfield(Instruction* instr) {
 }
 
 
-void Disassembler::VisitExtract(Instruction* instr) {
+void Disassembler::VisitExtract(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rd, 'Rn, 'Rm, 'IExtract";
 
@@ -527,7 +527,7 @@ void Disassembler::VisitExtract(Instruction* instr) {
 }
 
 
-void Disassembler::VisitPCRelAddressing(Instruction* instr) {
+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;
@@ -536,7 +536,7 @@ void Disassembler::VisitPCRelAddressing(Instruction* instr) {
 }
 
 
-void Disassembler::VisitConditionalBranch(Instruction* instr) {
+void Disassembler::VisitConditionalBranch(const Instruction* instr) {
   switch (instr->Mask(ConditionalBranchMask)) {
     case B_cond: Format(instr, "b.'CBrn", "'BImmCond"); break;
     default: VIXL_UNREACHABLE();
@@ -544,7 +544,8 @@ void Disassembler::VisitConditionalBranch(Instruction* instr) {
 }
 
 
-void Disassembler::VisitUnconditionalBranchToRegister(Instruction* instr) {
+void Disassembler::VisitUnconditionalBranchToRegister(
+    const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'Xn";
 
@@ -564,7 +565,7 @@ void Disassembler::VisitUnconditionalBranchToRegister(Instruction* instr) {
 }
 
 
-void Disassembler::VisitUnconditionalBranch(Instruction* instr) {
+void Disassembler::VisitUnconditionalBranch(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'BImmUncn";
 
@@ -577,7 +578,7 @@ void Disassembler::VisitUnconditionalBranch(Instruction* instr) {
 }
 
 
-void Disassembler::VisitDataProcessing1Source(Instruction* instr) {
+void Disassembler::VisitDataProcessing1Source(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rd, 'Rn";
 
@@ -598,7 +599,7 @@ void Disassembler::VisitDataProcessing1Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitDataProcessing2Source(Instruction* instr) {
+void Disassembler::VisitDataProcessing2Source(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'Rd, 'Rn, 'Rm";
 
@@ -619,7 +620,7 @@ void Disassembler::VisitDataProcessing2Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitDataProcessing3Source(Instruction* instr) {
+void Disassembler::VisitDataProcessing3Source(const Instruction* instr) {
   bool ra_is_zr = RaIsZROrSP(instr);
   const char *mnemonic = "";
   const char *form = "'Xd, 'Wn, 'Wm, 'Xa";
@@ -697,7 +698,7 @@ void Disassembler::VisitDataProcessing3Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitCompareBranch(Instruction* instr) {
+void Disassembler::VisitCompareBranch(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rt, 'BImmCmpa";
 
@@ -712,7 +713,7 @@ void Disassembler::VisitCompareBranch(Instruction* instr) {
 }
 
 
-void Disassembler::VisitTestBranch(Instruction* instr) {
+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
@@ -729,7 +730,7 @@ void Disassembler::VisitTestBranch(Instruction* instr) {
 }
 
 
-void Disassembler::VisitMoveWideImmediate(Instruction* instr) {
+void Disassembler::VisitMoveWideImmediate(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rd, 'IMoveImm";
 
@@ -768,7 +769,7 @@ void Disassembler::VisitMoveWideImmediate(Instruction* instr) {
   V(LDR_s, "ldr", "'St")      \
   V(LDR_d, "ldr", "'Dt")
 
-void Disassembler::VisitLoadStorePreIndex(Instruction* instr) {
+void Disassembler::VisitLoadStorePreIndex(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStorePreIndex)";
 
@@ -782,7 +783,7 @@ void Disassembler::VisitLoadStorePreIndex(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStorePostIndex(Instruction* instr) {
+void Disassembler::VisitLoadStorePostIndex(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStorePostIndex)";
 
@@ -796,7 +797,7 @@ void Disassembler::VisitLoadStorePostIndex(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStoreUnsignedOffset(Instruction* instr) {
+void Disassembler::VisitLoadStoreUnsignedOffset(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStoreUnsignedOffset)";
 
@@ -811,7 +812,7 @@ void Disassembler::VisitLoadStoreUnsignedOffset(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStoreRegisterOffset(Instruction* instr) {
+void Disassembler::VisitLoadStoreRegisterOffset(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStoreRegisterOffset)";
 
@@ -826,7 +827,7 @@ void Disassembler::VisitLoadStoreRegisterOffset(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStoreUnscaledOffset(Instruction* instr) {
+void Disassembler::VisitLoadStoreUnscaledOffset(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'Wt, ['Xns'ILS]";
   const char *form_x = "'Xt, ['Xns'ILS]";
@@ -857,7 +858,7 @@ void Disassembler::VisitLoadStoreUnscaledOffset(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadLiteral(Instruction* instr) {
+void Disassembler::VisitLoadLiteral(const Instruction* instr) {
   const char *mnemonic = "ldr";
   const char *form = "(LoadLiteral)";
 
@@ -866,6 +867,11 @@ void Disassembler::VisitLoadLiteral(Instruction* instr) {
     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;
+    }
     default: mnemonic = "unimplemented";
   }
   Format(instr, mnemonic, form);
@@ -883,7 +889,7 @@ void Disassembler::VisitLoadLiteral(Instruction* instr) {
   V(STP_d, "stp", "'Dt, 'Dt2", "8")     \
   V(LDP_d, "ldp", "'Dt, 'Dt2", "8")
 
-void Disassembler::VisitLoadStorePairPostIndex(Instruction* instr) {
+void Disassembler::VisitLoadStorePairPostIndex(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStorePairPostIndex)";
 
@@ -897,7 +903,7 @@ void Disassembler::VisitLoadStorePairPostIndex(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStorePairPreIndex(Instruction* instr) {
+void Disassembler::VisitLoadStorePairPreIndex(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStorePairPreIndex)";
 
@@ -911,7 +917,7 @@ void Disassembler::VisitLoadStorePairPreIndex(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStorePairOffset(Instruction* instr) {
+void Disassembler::VisitLoadStorePairOffset(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(LoadStorePairOffset)";
 
@@ -925,7 +931,7 @@ void Disassembler::VisitLoadStorePairOffset(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStorePairNonTemporal(Instruction* instr) {
+void Disassembler::VisitLoadStorePairNonTemporal(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form;
 
@@ -944,7 +950,7 @@ void Disassembler::VisitLoadStorePairNonTemporal(Instruction* instr) {
 }
 
 
-void Disassembler::VisitLoadStoreExclusive(Instruction* instr) {
+void Disassembler::VisitLoadStoreExclusive(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form;
 
@@ -987,7 +993,7 @@ void Disassembler::VisitLoadStoreExclusive(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPCompare(Instruction* instr) {
+void Disassembler::VisitFPCompare(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'Fn, 'Fm";
   const char *form_zero = "'Fn, #0.0";
@@ -1003,7 +1009,7 @@ void Disassembler::VisitFPCompare(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPConditionalCompare(Instruction* instr) {
+void Disassembler::VisitFPConditionalCompare(const Instruction* instr) {
   const char *mnemonic = "unmplemented";
   const char *form = "'Fn, 'Fm, 'INzcv, 'Cond";
 
@@ -1018,7 +1024,7 @@ void Disassembler::VisitFPConditionalCompare(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPConditionalSelect(Instruction* instr) {
+void Disassembler::VisitFPConditionalSelect(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Fd, 'Fn, 'Fm, 'Cond";
 
@@ -1031,7 +1037,7 @@ void Disassembler::VisitFPConditionalSelect(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPDataProcessing1Source(Instruction* instr) {
+void Disassembler::VisitFPDataProcessing1Source(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'Fd, 'Fn";
 
@@ -1059,7 +1065,7 @@ void Disassembler::VisitFPDataProcessing1Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPDataProcessing2Source(Instruction* instr) {
+void Disassembler::VisitFPDataProcessing2Source(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Fd, 'Fn, 'Fm";
 
@@ -1083,7 +1089,7 @@ void Disassembler::VisitFPDataProcessing2Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPDataProcessing3Source(Instruction* instr) {
+void Disassembler::VisitFPDataProcessing3Source(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Fd, 'Fn, 'Fm, 'Fa";
 
@@ -1102,7 +1108,7 @@ void Disassembler::VisitFPDataProcessing3Source(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPImmediate(Instruction* instr) {
+void Disassembler::VisitFPImmediate(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "(FPImmediate)";
 
@@ -1115,7 +1121,7 @@ void Disassembler::VisitFPImmediate(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPIntegerConvert(Instruction* instr) {
+void Disassembler::VisitFPIntegerConvert(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "(FPIntegerConvert)";
   const char *form_rf = "'Rd, 'Fn";
@@ -1171,7 +1177,7 @@ void Disassembler::VisitFPIntegerConvert(Instruction* instr) {
 }
 
 
-void Disassembler::VisitFPFixedPointConvert(Instruction* instr) {
+void Disassembler::VisitFPFixedPointConvert(const Instruction* instr) {
   const char *mnemonic = "";
   const char *form = "'Rd, 'Fn, 'IFPFBits";
   const char *form_fr = "'Fd, 'Rn, 'IFPFBits";
@@ -1199,7 +1205,7 @@ void Disassembler::VisitFPFixedPointConvert(Instruction* instr) {
 }
 
 
-void Disassembler::VisitSystem(Instruction* instr) {
+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.
@@ -1267,7 +1273,7 @@ void Disassembler::VisitSystem(Instruction* instr) {
 }
 
 
-void Disassembler::VisitException(Instruction* instr) {
+void Disassembler::VisitException(const Instruction* instr) {
   const char *mnemonic = "unimplemented";
   const char *form = "'IDebug";
 
@@ -1286,22 +1292,75 @@ void Disassembler::VisitException(Instruction* instr) {
 }
 
 
-void Disassembler::VisitUnimplemented(Instruction* instr) {
+void Disassembler::VisitUnimplemented(const Instruction* instr) {
   Format(instr, "unimplemented", "(Unimplemented)");
 }
 
 
-void Disassembler::VisitUnallocated(Instruction* instr) {
+void Disassembler::VisitUnallocated(const Instruction* instr) {
   Format(instr, "unallocated", "(Unallocated)");
 }
 
 
-void Disassembler::ProcessOutput(Instruction* /*instr*/) {
+void Disassembler::ProcessOutput(const Instruction* /*instr*/) {
   // The base disasm does nothing more than disassembling into a buffer.
 }
 
 
-void Disassembler::Format(Instruction* instr, const char* mnemonic,
+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 %p)", 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::Format(const Instruction* instr, const char* mnemonic,
                           const char* format) {
   VIXL_ASSERT(mnemonic != NULL);
   ResetOutput();
@@ -1315,7 +1374,7 @@ void Disassembler::Format(Instruction* instr, const char* mnemonic,
 }
 
 
-void Disassembler::Substitute(Instruction* instr, const char* string) {
+void Disassembler::Substitute(const Instruction* instr, const char* string) {
   char chr = *string++;
   while (chr != '\0') {
     if (chr == '\'') {
@@ -1328,7 +1387,8 @@ void Disassembler::Substitute(Instruction* instr, const char* string) {
 }
 
 
-int Disassembler::SubstituteField(Instruction* instr, const char* format) {
+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.
@@ -1354,7 +1414,7 @@ int Disassembler::SubstituteField(Instruction* instr, const char* format) {
 }
 
 
-int Disassembler::SubstituteRegisterField(Instruction* instr,
+int Disassembler::SubstituteRegisterField(const Instruction* instr,
                                           const char* format) {
   unsigned reg_num = 0;
   unsigned field_len = 2;
@@ -1381,34 +1441,47 @@ int Disassembler::SubstituteRegisterField(Instruction* instr,
     field_len = 3;
   }
 
-  char reg_type;
+  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 = instr->SixtyFourBits() ? 'x' : '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 = ((instr->FPType() & 1) == 0) ? 's' : 'd';
+    reg_type = CPURegister::kFPRegister;
+    reg_size = ((instr->FPType() & 1) == 0) ? kSRegSize : kDRegSize;
   } else {
-    // Register type is specified. Make it lower case.
-    reg_type = format[0] + 0x20;
+    // 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_num != kZeroRegCode) || (reg_type == 's') || (reg_type == 'd')) {
-    // A normal register: w0 - w30, x0 - x30, s0 - s31, d0 - d31.
-    AppendToOutput("%c%d", reg_type, reg_num);
-  } else if (format[2] == 's') {
-    // Disassemble w31/x31 as stack pointer wsp/sp.
-    AppendToOutput("%s", (reg_type == 'w') ? "wsp" : "sp");
-  } else {
-    // Disassemble w31/x31 as zero register wzr/xzr.
-    AppendToOutput("%czr", reg_type);
+  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(Instruction* instr,
+int Disassembler::SubstituteImmediateField(const Instruction* instr,
                                            const char* format) {
   VIXL_ASSERT(format[0] == 'I');
 
@@ -1458,8 +1531,7 @@ int Disassembler::SubstituteImmediateField(Instruction* instr,
     }
     case 'C': {  // ICondB - Immediate Conditional Branch.
       int64_t offset = instr->ImmCondBranch() << 2;
-      char sign = (offset >= 0) ? '+' : '-';
-      AppendToOutput("#%c0x%" PRIx64, sign, offset);
+      AppendPCRelativeOffsetToOutput(instr, offset);
       return 6;
     }
     case 'A': {  // IAddSub.
@@ -1522,7 +1594,7 @@ int Disassembler::SubstituteImmediateField(Instruction* instr,
 }
 
 
-int Disassembler::SubstituteBitfieldImmediateField(Instruction* instr,
+int Disassembler::SubstituteBitfieldImmediateField(const Instruction* instr,
                                                    const char* format) {
   VIXL_ASSERT((format[0] == 'I') && (format[1] == 'B'));
   unsigned r = instr->ImmR();
@@ -1557,7 +1629,7 @@ int Disassembler::SubstituteBitfieldImmediateField(Instruction* instr,
 }
 
 
-int Disassembler::SubstituteLiteralField(Instruction* instr,
+int Disassembler::SubstituteLiteralField(const Instruction* instr,
                                          const char* format) {
   VIXL_ASSERT(strncmp(format, "LValue", 6) == 0);
   USE(format);
@@ -1565,16 +1637,21 @@ int Disassembler::SubstituteLiteralField(Instruction* instr,
   switch (instr->Mask(LoadLiteralMask)) {
     case LDR_w_lit:
     case LDR_x_lit:
+    case LDRSW_x_lit:
     case LDR_s_lit:
-    case LDR_d_lit: AppendToOutput("(addr %p)", instr->LiteralAddress()); break;
-    default: VIXL_UNREACHABLE();
+    case LDR_d_lit:
+      AppendDataAddressToOutput(instr, instr->LiteralAddress());
+      break;
+    default:
+      VIXL_UNREACHABLE();
   }
 
   return 6;
 }
 
 
-int Disassembler::SubstituteShiftField(Instruction* instr, const char* format) {
+int Disassembler::SubstituteShiftField(const Instruction* instr,
+                                       const char* format) {
   VIXL_ASSERT(format[0] == 'H');
   VIXL_ASSERT(instr->ShiftDP() <= 0x3);
 
@@ -1597,7 +1674,7 @@ int Disassembler::SubstituteShiftField(Instruction* instr, const char* format) {
 }
 
 
-int Disassembler::SubstituteConditionField(Instruction* instr,
+int Disassembler::SubstituteConditionField(const Instruction* instr,
                                            const char* format) {
   VIXL_ASSERT(format[0] == 'C');
   const char* condition_code[] = { "eq", "ne", "hs", "lo",
@@ -1618,27 +1695,28 @@ int Disassembler::SubstituteConditionField(Instruction* instr,
 }
 
 
-int Disassembler::SubstitutePCRelAddressField(Instruction* instr,
+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();
-  Instruction * base = instr;
+  const Instruction * base = instr;
 
   if (format[9] == 'P') {
     offset *= kPageSize;
     base = AlignDown(base, kPageSize);
   }
 
-  char sign = (offset < 0) ? '-' : '+';
-  void * target = reinterpret_cast<void *>(base + offset);
-  AppendToOutput("#%c0x%" PRIx64 " (addr %p)", sign, std::abs(offset), target);
+  const void* target = reinterpret_cast<const void*>(base + offset);
+  AppendPCRelativeOffsetToOutput(instr, offset);
+  AppendToOutput(" ");
+  AppendAddressToOutput(instr, target);
   return 13;
 }
 
 
-int Disassembler::SubstituteBranchTargetField(Instruction* instr,
+int Disassembler::SubstituteBranchTargetField(const Instruction* instr,
                                               const char* format) {
   VIXL_ASSERT(strncmp(format, "BImm", 4) == 0);
 
@@ -1655,19 +1733,18 @@ int Disassembler::SubstituteBranchTargetField(Instruction* instr,
     default: VIXL_UNIMPLEMENTED();
   }
   offset <<= kInstructionSizeLog2;
-  char sign = '+';
-  if (offset < 0) {
-    offset = -offset;
-    sign = '-';
-  }
+  const void* target_address = reinterpret_cast<const void*>(instr + offset);
   VIXL_STATIC_ASSERT(sizeof(*instr) == 1);
-  void * address = reinterpret_cast<void *>(instr + offset);
-  AppendToOutput("#%c0x%" PRIx64 " (addr %p)", sign, offset, address);
+
+  AppendPCRelativeOffsetToOutput(instr, offset);
+  AppendToOutput(" ");
+  AppendCodeAddressToOutput(instr, target_address);
+
   return 8;
 }
 
 
-int Disassembler::SubstituteExtendField(Instruction* instr,
+int Disassembler::SubstituteExtendField(const Instruction* instr,
                                         const char* format) {
   VIXL_ASSERT(strncmp(format, "Ext", 3) == 0);
   VIXL_ASSERT(instr->ExtendMode() <= 7);
@@ -1694,7 +1771,7 @@ int Disassembler::SubstituteExtendField(Instruction* instr,
 }
 
 
-int Disassembler::SubstituteLSRegOffsetField(Instruction* instr,
+int Disassembler::SubstituteLSRegOffsetField(const Instruction* instr,
                                              const char* format) {
   VIXL_ASSERT(strncmp(format, "Offsetreg", 9) == 0);
   const char* extend_mode[] = { "undefined", "undefined", "uxtw", "lsl",
@@ -1723,7 +1800,7 @@ int Disassembler::SubstituteLSRegOffsetField(Instruction* instr,
 }
 
 
-int Disassembler::SubstitutePrefetchField(Instruction* instr,
+int Disassembler::SubstitutePrefetchField(const Instruction* instr,
                                           const char* format) {
   VIXL_ASSERT(format[0] == 'P');
   USE(format);
@@ -1738,7 +1815,7 @@ int Disassembler::SubstitutePrefetchField(Instruction* instr,
   return 6;
 }
 
-int Disassembler::SubstituteBarrierField(Instruction* instr,
+int Disassembler::SubstituteBarrierField(const Instruction* instr,
                                          const char* format) {
   VIXL_ASSERT(format[0] == 'M');
   USE(format);
@@ -1770,7 +1847,7 @@ void Disassembler::AppendToOutput(const char* format, ...) {
 }
 
 
-void PrintDisassembler::ProcessOutput(Instruction* instr) {
+void PrintDisassembler::ProcessOutput(const Instruction* instr) {
   fprintf(stream_, "0x%016" PRIx64 "  %08" PRIx32 "\t\t%s\n",
           reinterpret_cast<uint64_t>(instr),
           instr->InstructionBits(),
diff --git a/disas/libvixl/a64/disasm-a64.h b/disas/libvixl/a64/disasm-a64.h
index 06ee43fdeb..db043375c5 100644
--- a/disas/libvixl/a64/disasm-a64.h
+++ b/disas/libvixl/a64/disasm-a64.h
@@ -31,6 +31,7 @@
 #include "utils.h"
 #include "instructions-a64.h"
 #include "decoder-a64.h"
+#include "assembler-a64.h"
 
 namespace vixl {
 
@@ -42,48 +43,83 @@ class Disassembler: public DecoderVisitor {
   char* GetOutput();
 
   // Declare all Visitor functions.
-  #define DECLARE(A)  void Visit##A(Instruction* instr);
+  #define DECLARE(A)  void Visit##A(const Instruction* instr);
   VISITOR_LIST(DECLARE)
   #undef DECLARE
 
  protected:
-  virtual void ProcessOutput(Instruction* instr);
+  virtual void ProcessOutput(const Instruction* instr);
+
+  // Default output functions.  The functions below implement a default way of
+  // printing elements in the disassembly. A sub-class can override these to
+  // customize the disassembly output.
+
+  // Prints the name of a register.
+  virtual void AppendRegisterNameToOutput(const Instruction* instr,
+                                          const CPURegister& reg);
+
+  // Prints a PC-relative offset. This is used for example when disassembling
+  // branches to immediate offsets.
+  virtual void AppendPCRelativeOffsetToOutput(const Instruction* instr,
+                                              int64_t offset);
+
+  // Prints an address, in the general case. It can be code or data. This is
+  // used for example to print the target address of an ADR instruction.
+  virtual void AppendAddressToOutput(const Instruction* instr,
+                                     const void* addr);
+
+  // Prints the address of some code.
+  // This is used for example to print the target address of a branch to an
+  // immediate offset.
+  // A sub-class can for example override this method to lookup the address and
+  // print an appropriate name.
+  virtual void AppendCodeAddressToOutput(const Instruction* instr,
+                                         const void* addr);
+
+  // Prints the address of some data.
+  // This is used for example to print the source address of a load literal
+  // instruction.
+  virtual void AppendDataAddressToOutput(const Instruction* instr,
+                                         const void* addr);
 
  private:
-  void Format(Instruction* instr, const char* mnemonic, const char* format);
-  void Substitute(Instruction* instr, const char* string);
-  int SubstituteField(Instruction* instr, const char* format);
-  int SubstituteRegisterField(Instruction* instr, const char* format);
-  int SubstituteImmediateField(Instruction* instr, const char* format);
-  int SubstituteLiteralField(Instruction* instr, const char* format);
-  int SubstituteBitfieldImmediateField(Instruction* instr, const char* format);
-  int SubstituteShiftField(Instruction* instr, const char* format);
-  int SubstituteExtendField(Instruction* instr, const char* format);
-  int SubstituteConditionField(Instruction* instr, const char* format);
-  int SubstitutePCRelAddressField(Instruction* instr, const char* format);
-  int SubstituteBranchTargetField(Instruction* instr, const char* format);
-  int SubstituteLSRegOffsetField(Instruction* instr, const char* format);
-  int SubstitutePrefetchField(Instruction* instr, const char* format);
-  int SubstituteBarrierField(Instruction* instr, const char* format);
-
-  inline bool RdIsZROrSP(Instruction* instr) const {
+  void Format(
+      const Instruction* instr, const char* mnemonic, const char* format);
+  void Substitute(const Instruction* instr, const char* string);
+  int SubstituteField(const Instruction* instr, const char* format);
+  int SubstituteRegisterField(const Instruction* instr, const char* format);
+  int SubstituteImmediateField(const Instruction* instr, const char* format);
+  int SubstituteLiteralField(const Instruction* instr, const char* format);
+  int SubstituteBitfieldImmediateField(
+      const Instruction* instr, const char* format);
+  int SubstituteShiftField(const Instruction* instr, const char* format);
+  int SubstituteExtendField(const Instruction* instr, const char* format);
+  int SubstituteConditionField(const Instruction* instr, const char* format);
+  int SubstitutePCRelAddressField(const Instruction* instr, const char* format);
+  int SubstituteBranchTargetField(const Instruction* instr, const char* format);
+  int SubstituteLSRegOffsetField(const Instruction* instr, const char* format);
+  int SubstitutePrefetchField(const Instruction* instr, const char* format);
+  int SubstituteBarrierField(const Instruction* instr, const char* format);
+
+  inline bool RdIsZROrSP(const Instruction* instr) const {
     return (instr->Rd() == kZeroRegCode);
   }
 
-  inline bool RnIsZROrSP(Instruction* instr) const {
+  inline bool RnIsZROrSP(const Instruction* instr) const {
     return (instr->Rn() == kZeroRegCode);
   }
 
-  inline bool RmIsZROrSP(Instruction* instr) const {
+  inline bool RmIsZROrSP(const Instruction* instr) const {
     return (instr->Rm() == kZeroRegCode);
   }
 
-  inline bool RaIsZROrSP(Instruction* instr) const {
+  inline bool RaIsZROrSP(const Instruction* instr) const {
     return (instr->Ra() == kZeroRegCode);
   }
 
   bool IsMovzMovnImm(unsigned reg_size, uint64_t value);
 
+ protected:
   void ResetOutput();
   void AppendToOutput(const char* string, ...) PRINTF_CHECK(2, 3);
 
@@ -97,10 +133,10 @@ class Disassembler: public DecoderVisitor {
 class PrintDisassembler: public Disassembler {
  public:
   explicit PrintDisassembler(FILE* stream) : stream_(stream) { }
-  ~PrintDisassembler() { }
+  virtual ~PrintDisassembler() { }
 
  protected:
-  virtual void ProcessOutput(Instruction* instr);
+  virtual void ProcessOutput(const Instruction* instr);
 
  private:
   FILE *stream_;
diff --git a/disas/libvixl/a64/instructions-a64.cc b/disas/libvixl/a64/instructions-a64.cc
index e9caceb37b..1f08c781eb 100644
--- a/disas/libvixl/a64/instructions-a64.cc
+++ b/disas/libvixl/a64/instructions-a64.cc
@@ -57,7 +57,7 @@ static uint64_t RepeatBitsAcrossReg(unsigned reg_size,
 // 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() {
+uint64_t Instruction::ImmLogical() const {
   unsigned reg_size = SixtyFourBits() ? kXRegSize : kWRegSize;
   int64_t n = BitN();
   int64_t imm_s = ImmSetBits();
@@ -108,7 +108,7 @@ uint64_t Instruction::ImmLogical() {
 }
 
 
-float Instruction::ImmFP32() {
+float Instruction::ImmFP32() const {
   //  ImmFP: abcdefgh (8 bits)
   // Single: aBbb.bbbc.defg.h000.0000.0000.0000.0000 (32 bits)
   // where B is b ^ 1
@@ -122,7 +122,7 @@ float Instruction::ImmFP32() {
 }
 
 
-double Instruction::ImmFP64() {
+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)
@@ -148,8 +148,8 @@ LSDataSize CalcLSPairDataSize(LoadStorePairOp op) {
 }
 
 
-Instruction* Instruction::ImmPCOffsetTarget() {
-  Instruction * base = this;
+const Instruction* Instruction::ImmPCOffsetTarget() const {
+  const Instruction * base = this;
   ptrdiff_t offset;
   if (IsPCRelAddressing()) {
     // ADR and ADRP.
@@ -182,7 +182,7 @@ inline int Instruction::ImmBranch() const {
 }
 
 
-void Instruction::SetImmPCOffsetTarget(Instruction* target) {
+void Instruction::SetImmPCOffsetTarget(const Instruction* target) {
   if (IsPCRelAddressing()) {
     SetPCRelImmTarget(target);
   } else {
@@ -191,7 +191,7 @@ void Instruction::SetImmPCOffsetTarget(Instruction* target) {
 }
 
 
-void Instruction::SetPCRelImmTarget(Instruction* target) {
+void Instruction::SetPCRelImmTarget(const Instruction* target) {
   int32_t imm21;
   if ((Mask(PCRelAddressingMask) == ADR)) {
     imm21 = target - this;
@@ -207,7 +207,7 @@ void Instruction::SetPCRelImmTarget(Instruction* target) {
 }
 
 
-void Instruction::SetBranchImmTarget(Instruction* target) {
+void Instruction::SetBranchImmTarget(const Instruction* target) {
   VIXL_ASSERT(((target - this) & 3) == 0);
   Instr branch_imm = 0;
   uint32_t imm_mask = 0;
@@ -239,9 +239,9 @@ void Instruction::SetBranchImmTarget(Instruction* target) {
 }
 
 
-void Instruction::SetImmLLiteral(Instruction* source) {
-  VIXL_ASSERT(((source - this) & 3) == 0);
-  int offset = (source - this) >> kLiteralEntrySizeLog2;
+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;
 
diff --git a/disas/libvixl/a64/instructions-a64.h b/disas/libvixl/a64/instructions-a64.h
index d5b90c52e1..38f079f967 100644
--- a/disas/libvixl/a64/instructions-a64.h
+++ b/disas/libvixl/a64/instructions-a64.h
@@ -44,6 +44,7 @@ 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;
@@ -201,9 +202,9 @@ class Instruction {
     return signed_bitextract_32(width-1, 0, offset);
   }
 
-  uint64_t ImmLogical();
-  float ImmFP32();
-  double ImmFP64();
+  uint64_t ImmLogical() const;
+  float ImmFP32() const;
+  double ImmFP64() const;
 
   inline LSDataSize SizeLSPair() const {
     return CalcLSPairDataSize(
@@ -311,46 +312,49 @@ class Instruction {
 
   // Find the target of this instruction. 'this' may be a branch or a
   // PC-relative addressing instruction.
-  Instruction* ImmPCOffsetTarget();
+  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(Instruction* target);
+  void SetImmPCOffsetTarget(const Instruction* target);
   // Patch a literal load instruction to load from 'source'.
-  void SetImmLLiteral(Instruction* source);
+  void SetImmLLiteral(const Instruction* source);
 
-  inline uint8_t* LiteralAddress() {
+  inline uint8_t* LiteralAddress() const {
     int offset = ImmLLiteral() << kLiteralEntrySizeLog2;
-    return reinterpret_cast<uint8_t*>(this) + offset;
+    const uint8_t* address = reinterpret_cast<const uint8_t*>(this) + offset;
+    // Note that the result is safely mutable only if the backing buffer is
+    // safely mutable.
+    return const_cast<uint8_t*>(address);
   }
 
-  inline uint32_t Literal32() {
+  inline uint32_t Literal32() const {
     uint32_t literal;
     memcpy(&literal, LiteralAddress(), sizeof(literal));
 
     return literal;
   }
 
-  inline uint64_t Literal64() {
+  inline uint64_t Literal64() const {
     uint64_t literal;
     memcpy(&literal, LiteralAddress(), sizeof(literal));
 
     return literal;
   }
 
-  inline float LiteralFP32() {
+  inline float LiteralFP32() const {
     return rawbits_to_float(Literal32());
   }
 
-  inline double LiteralFP64() {
+  inline double LiteralFP64() const {
     return rawbits_to_double(Literal64());
   }
 
-  inline Instruction* NextInstruction() {
+  inline const Instruction* NextInstruction() const {
     return this + kInstructionSize;
   }
 
-  inline Instruction* InstructionAtOffset(int64_t offset) {
+  inline const Instruction* InstructionAtOffset(int64_t offset) const {
     VIXL_ASSERT(IsWordAligned(this + offset));
     return this + offset;
   }
@@ -359,11 +363,15 @@ class Instruction {
     return reinterpret_cast<Instruction*>(src);
   }
 
+  template<typename T> static inline const Instruction* CastConst(T src) {
+    return reinterpret_cast<const Instruction*>(src);
+  }
+
  private:
   inline int ImmBranch() const;
 
-  void SetPCRelImmTarget(Instruction* target);
-  void SetBranchImmTarget(Instruction* target);
+  void SetPCRelImmTarget(const Instruction* target);
+  void SetBranchImmTarget(const Instruction* target);
 };
 }  // namespace vixl
 
diff --git a/disas/libvixl/code-buffer.h b/disas/libvixl/code-buffer.h
new file mode 100644
index 0000000000..da6233dd80
--- /dev/null
+++ b/disas/libvixl/code-buffer.h
@@ -0,0 +1,113 @@
+// Copyright 2014, 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_CODE_BUFFER_H
+#define VIXL_CODE_BUFFER_H
+
+#include <string.h>
+#include "globals.h"
+
+namespace vixl {
+
+class CodeBuffer {
+ public:
+  explicit CodeBuffer(size_t capacity = 4 * KBytes);
+  CodeBuffer(void* buffer, size_t capacity);
+  ~CodeBuffer();
+
+  void Reset();
+
+  ptrdiff_t OffsetFrom(ptrdiff_t offset) const {
+    ptrdiff_t cursor_offset = cursor_ - buffer_;
+    VIXL_ASSERT((offset >= 0) && (offset <= cursor_offset));
+    return cursor_offset - offset;
+  }
+
+  ptrdiff_t CursorOffset() const {
+    return OffsetFrom(0);
+  }
+
+  template <typename T>
+  T GetOffsetAddress(ptrdiff_t offset) const {
+    VIXL_ASSERT((offset >= 0) && (offset <= (cursor_ - buffer_)));
+    return reinterpret_cast<T>(buffer_ + offset);
+  }
+
+  size_t RemainingBytes() const {
+    VIXL_ASSERT((cursor_ >= buffer_) && (cursor_ <= (buffer_ + capacity_)));
+    return (buffer_ + capacity_) - cursor_;
+  }
+
+  // A code buffer can emit:
+  //  * 32-bit data: instruction and constant.
+  //  * 64-bit data: constant.
+  //  * string: debug info.
+  void Emit32(uint32_t data) { Emit(data); }
+
+  void Emit64(uint64_t data) { Emit(data); }
+
+  void EmitString(const char* string);
+
+  // Align to kInstructionSize.
+  void Align();
+
+  size_t capacity() const { return capacity_; }
+
+  bool IsManaged() const { return managed_; }
+
+  void Grow(size_t new_capacity);
+
+  bool IsDirty() const { return dirty_; }
+
+  void SetClean() { dirty_ = false; }
+
+ private:
+  template <typename T>
+  void Emit(T value) {
+    VIXL_ASSERT(RemainingBytes() >= sizeof(value));
+    dirty_ = true;
+    memcpy(cursor_, &value, sizeof(value));
+    cursor_ += sizeof(value);
+  }
+
+  // Backing store of the buffer.
+  byte* buffer_;
+  // If true the backing store is allocated and deallocated by the buffer. The
+  // backing store can then grow on demand. If false the backing store is
+  // provided by the user and cannot be resized internally.
+  bool managed_;
+  // Pointer to the next location to be written.
+  byte* cursor_;
+  // True if there has been any write since the buffer was created or cleaned.
+  bool dirty_;
+  // Capacity in bytes of the backing store.
+  size_t capacity_;
+};
+
+}  // namespace vixl
+
+#endif  // VIXL_CODE_BUFFER_H
+
diff --git a/disas/libvixl/utils.cc b/disas/libvixl/utils.cc
index 4d4fcbdad4..21965d7a1f 100644
--- a/disas/libvixl/utils.cc
+++ b/disas/libvixl/utils.cc
@@ -134,4 +134,5 @@ uint64_t LowestSetBit(uint64_t value) {
 bool IsPowerOf2(int64_t value) {
   return (value != 0) && ((value & (value - 1)) == 0);
 }
+
 }  // namespace vixl
diff --git a/disas/libvixl/utils.h b/disas/libvixl/utils.h
index b472f0e6cd..1540c3060b 100644
--- a/disas/libvixl/utils.h
+++ b/disas/libvixl/utils.h
@@ -171,7 +171,7 @@ bool IsPowerOf2(int64_t value);
 template<typename T>
 bool IsWordAligned(T pointer) {
   VIXL_ASSERT(sizeof(pointer) == sizeof(intptr_t));   // NOLINT(runtime/sizeof)
-  return (reinterpret_cast<intptr_t>(pointer) & 3) == 0;
+  return ((intptr_t)(pointer) & 3) == 0;
 }
 
 // Increment a pointer until it has the specified alignment.
@@ -204,7 +204,6 @@ T AlignDown(T pointer, size_t alignment) {
   return (T)(pointer_raw - align_step);
 }
 
-
 }  // namespace vixl
 
 #endif  // VIXL_UTILS_H