lib/Target/ARM/Disassembler/ARMDisassembler.cpp

//===- ARMDisassembler.cpp - Disassembler for ARM/Thumb ISA -----*- C++ -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "arm-disassembler"

#include "ARMDisassembler.h"
#include "ARM.h"
#include "ARMRegisterInfo.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "MCTargetDesc/ARMBaseInfo.h"
#include "llvm/MC/EDInstInfo.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCContext.h"
#include "llvm/Target/TargetRegistry.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MemoryObject.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"

// Pull DecodeStatus and its enum values into the global namespace.
typedef llvm::MCDisassembler::DecodeStatus DecodeStatus;
#define Success llvm::MCDisassembler::Success
#define Unpredictable llvm::MCDisassembler::SoftFail
#define Fail llvm::MCDisassembler::Fail

// Helper macro to perform setwise reduction of the current running status
// and another status, and return if the new status is Fail.
#define CHECK(S,X) do {                           \
    S = (DecodeStatus) ((int)S & (X));            \
    if (S == Fail) return Fail;                   \
  } while(0)

// Forward declare these because the autogenerated code will reference them.
// Definitions are further down.
static DecodeStatus DecodeGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeGPRnopcRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodetGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodetcGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecoderGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDPR_8RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDPR_VFP2RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);
static DecodeStatus DecodeQPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder);

static DecodeStatus DecodePredicateOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeCCOutOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSOImmOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeBLTargetOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeRegListOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSPRRegListOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDPRRegListOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);

static DecodeStatus DecodeBitfieldMaskOperand(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeCopMemInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrMode2IdxInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSORegMemOperand(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrMode3Instruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSORegImmOperand(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSORegRegOperand(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);

static DecodeStatus DecodeMemMultipleWritebackInstruction(llvm::MCInst & Inst,
                                                  unsigned Insn,
                                                  uint64_t Adddress,
                                                  const void *Decoder);
static DecodeStatus DecodeSMLAInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeCPSInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrModeImm12Operand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrMode5Operand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrMode7Operand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeBranchImmInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVCVTImmOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeAddrMode6Operand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLDInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVSTInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD1DupInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD2DupInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD3DupInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD4DupInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeNEONModImmInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVSHLMaxInstruction(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeShiftRight8Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeShiftRight16Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeShiftRight32Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeShiftRight64Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeTBLInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVFPfpImm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodePostIdxReg(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeCoprocessor(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeMemBarrierOption(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeMSRMask(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDoubleRegLoad(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeDoubleRegStore(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSTRPreImm(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeSTRPreReg(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD1LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD2LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD3LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVLD4LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVST1LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVST2LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVST3LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeVST4LN(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder);


static DecodeStatus DecodeThumbAddSpecialReg(llvm::MCInst &Inst, uint16_t Insn,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbBROperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2BROperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbCmpBROperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddrModeRR(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddrModeIS(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddrModePC(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddrModeSP(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2AddrModeSOReg(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2LoadShift(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2Imm8S4(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2AddrModeImm8s4(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2Imm8(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2AddrModeImm8(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddSPImm(llvm::MCInst &Inst, uint16_t Val,
                               uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbAddSPReg(llvm::MCInst &Inst, uint16_t Insn,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbCPS(llvm::MCInst &Inst, uint16_t Insn,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbBLXOffset(llvm::MCInst &Inst, unsigned Insn,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2AddrModeImm12(llvm::MCInst &Inst, unsigned Val,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumb2BCCInstruction(llvm::MCInst &Inst, unsigned Val,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeT2SOImm(llvm::MCInst &Inst, unsigned Val,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbBCCTargetOperand(llvm::MCInst &Inst, unsigned Val,
                                uint64_t Address, const void *Decoder);
static DecodeStatus DecodeThumbBLTargetOperand(llvm::MCInst &Inst, unsigned Val,
                                uint64_t Address, const void *Decoder);

#include "ARMGenDisassemblerTables.inc"
#include "ARMGenInstrInfo.inc"
#include "ARMGenEDInfo.inc"

using namespace llvm;

static MCDisassembler *createARMDisassembler(const Target &T) {
  return new ARMDisassembler;
}

static MCDisassembler *createThumbDisassembler(const Target &T) {
  return new ThumbDisassembler;
}

EDInstInfo *ARMDisassembler::getEDInfo() const {
  return instInfoARM;
}

EDInstInfo *ThumbDisassembler::getEDInfo() const {
  return instInfoARM;
}

DecodeStatus ARMDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
                                             const MemoryObject &Region,
                                             uint64_t Address,raw_ostream &os) const {
  uint8_t bytes[4];

  // We want to read exactly 4 bytes of data.
  if (Region.readBytes(Address, 4, (uint8_t*)bytes, NULL) == -1)
    return Fail;

  // Encoded as a small-endian 32-bit word in the stream.
  uint32_t insn = (bytes[3] << 24) |
                  (bytes[2] << 16) |
                  (bytes[1] <<  8) |
                  (bytes[0] <<  0);

  // Calling the auto-generated decoder function.
  DecodeStatus result = decodeARMInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    return result;
  }

  // Instructions that are shared between ARM and Thumb modes.
  // FIXME: This shouldn't really exist.  It's an artifact of the
  // fact that we fail to encode a few instructions properly for Thumb.
  MI.clear();
  result = decodeCommonInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    return result;
  }

  // VFP and NEON instructions, similarly, are shared between ARM
  // and Thumb modes.
  MI.clear();
  result = decodeVFPInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    return result;
  }

  MI.clear();
  result = decodeNEONDataInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    // Add a fake predicate operand, because we share these instruction
    // definitions with Thumb2 where these instructions are predicable.
    if (!DecodePredicateOperand(MI, 0xE, Address, this)) return Fail;
    return result;
  }

  MI.clear();
  result = decodeNEONLoadStoreInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    // Add a fake predicate operand, because we share these instruction
    // definitions with Thumb2 where these instructions are predicable.
    if (!DecodePredicateOperand(MI, 0xE, Address, this)) return Fail;
    return result;
  }

  MI.clear();
  result = decodeNEONDupInstruction32(MI, insn, Address, this);
  if (result != Fail) {
    Size = 4;
    // Add a fake predicate operand, because we share these instruction
    // definitions with Thumb2 where these instructions are predicable.
    if (!DecodePredicateOperand(MI, 0xE, Address, this)) return Fail;
    return result;
  }

  MI.clear();

  return Fail;
}

namespace llvm {
extern MCInstrDesc ARMInsts[];
}

// Thumb1 instructions don't have explicit S bits.  Rather, they
// implicitly set CPSR.  Since it's not represented in the encoding, the
// auto-generated decoder won't inject the CPSR operand.  We need to fix
// that as a post-pass.
static void AddThumb1SBit(MCInst &MI, bool InITBlock) {
  const MCOperandInfo *OpInfo = ARMInsts[MI.getOpcode()].OpInfo;
  unsigned short NumOps = ARMInsts[MI.getOpcode()].NumOperands;
  MCInst::iterator I = MI.begin();
  for (unsigned i = 0; i < NumOps; ++i, ++I) {
    if (I == MI.end()) break;
    if (OpInfo[i].isOptionalDef() && OpInfo[i].RegClass == ARM::CCRRegClassID) {
      if (i > 0 && OpInfo[i-1].isPredicate()) continue;
      MI.insert(I, MCOperand::CreateReg(InITBlock ? 0 : ARM::CPSR));
      return;
    }
  }

  MI.insert(I, MCOperand::CreateReg(InITBlock ? 0 : ARM::CPSR));
}

// Most Thumb instructions don't have explicit predicates in the
// encoding, but rather get their predicates from IT context.  We need
// to fix up the predicate operands using this context information as a
// post-pass.
void ThumbDisassembler::AddThumbPredicate(MCInst &MI) const {
  // A few instructions actually have predicates encoded in them.  Don't
  // try to overwrite it if we're seeing one of those.
  switch (MI.getOpcode()) {
    case ARM::tBcc:
    case ARM::t2Bcc:
      return;
    default:
      break;
  }

  // If we're in an IT block, base the predicate on that.  Otherwise,
  // assume a predicate of AL.
  unsigned CC;
  if (!ITBlock.empty()) {
    CC = ITBlock.back();
    ITBlock.pop_back();
  } else
    CC = ARMCC::AL;

  const MCOperandInfo *OpInfo = ARMInsts[MI.getOpcode()].OpInfo;
  unsigned short NumOps = ARMInsts[MI.getOpcode()].NumOperands;
  MCInst::iterator I = MI.begin();
  for (unsigned i = 0; i < NumOps; ++i, ++I) {
    if (I == MI.end()) break;
    if (OpInfo[i].isPredicate()) {
      I = MI.insert(I, MCOperand::CreateImm(CC));
      ++I;
      if (CC == ARMCC::AL)
        MI.insert(I, MCOperand::CreateReg(0));
      else
        MI.insert(I, MCOperand::CreateReg(ARM::CPSR));
      return;
    }
  }

  I = MI.insert(I, MCOperand::CreateImm(CC));
  ++I;
  if (CC == ARMCC::AL)
    MI.insert(I, MCOperand::CreateReg(0));
  else
    MI.insert(I, MCOperand::CreateReg(ARM::CPSR));
}

// Thumb VFP instructions are a special case.  Because we share their
// encodings between ARM and Thumb modes, and they are predicable in ARM
// mode, the auto-generated decoder will give them an (incorrect)
// predicate operand.  We need to rewrite these operands based on the IT
// context as a post-pass.
void ThumbDisassembler::UpdateThumbVFPPredicate(MCInst &MI) const {
  unsigned CC;
  if (!ITBlock.empty()) {
    CC = ITBlock.back();
    ITBlock.pop_back();
  } else
    CC = ARMCC::AL;

  const MCOperandInfo *OpInfo = ARMInsts[MI.getOpcode()].OpInfo;
  MCInst::iterator I = MI.begin();
  for (unsigned i = 0, e = MI.size(); i < e; ++i, ++I) {
    if (OpInfo[i].isPredicate() ) {
      I->setImm(CC);
      ++I;
      if (CC == ARMCC::AL)
        I->setReg(0);
      else
        I->setReg(ARM::CPSR);
      return;
    }
  }
}

DecodeStatus ThumbDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
                                               const MemoryObject &Region,
                                               uint64_t Address,raw_ostream &os) const {
  uint8_t bytes[4];

  // We want to read exactly 2 bytes of data.
  if (Region.readBytes(Address, 2, (uint8_t*)bytes, NULL) == -1)
    return Fail;

  uint16_t insn16 = (bytes[1] << 8) | bytes[0];
  DecodeStatus result = decodeThumbInstruction16(MI, insn16, Address, this);
  if (result != Fail) {
    Size = 2;
    AddThumbPredicate(MI);
    return result;
  }

  MI.clear();
  result = decodeThumbSBitInstruction16(MI, insn16, Address, this);
  if (result) {
    Size = 2;
    bool InITBlock = !ITBlock.empty();
    AddThumbPredicate(MI);
    AddThumb1SBit(MI, InITBlock);
    return result;
  }

  MI.clear();
  result = decodeThumb2Instruction16(MI, insn16, Address, this);
  if (result != Fail) {
    Size = 2;
    AddThumbPredicate(MI);

    // If we find an IT instruction, we need to parse its condition
    // code and mask operands so that we can apply them correctly
    // to the subsequent instructions.
    if (MI.getOpcode() == ARM::t2IT) {
      unsigned firstcond = MI.getOperand(0).getImm();
      uint32_t mask = MI.getOperand(1).getImm();
      unsigned zeros = CountTrailingZeros_32(mask);
      mask >>= zeros+1;

      for (unsigned i = 0; i < 4 - (zeros+1); ++i) {
        if (firstcond ^ (mask & 1))
          ITBlock.push_back(firstcond ^ 1);
        else
          ITBlock.push_back(firstcond);
        mask >>= 1;
      }
      ITBlock.push_back(firstcond);
    }

    return result;
  }

  // We want to read exactly 4 bytes of data.
  if (Region.readBytes(Address, 4, (uint8_t*)bytes, NULL) == -1)
    return Fail;

  uint32_t insn32 = (bytes[3] <<  8) |
                    (bytes[2] <<  0) |
                    (bytes[1] << 24) |
                    (bytes[0] << 16);
  MI.clear();
  result = decodeThumbInstruction32(MI, insn32, Address, this);
  if (result != Fail) {
    Size = 4;
    bool InITBlock = ITBlock.size();
    AddThumbPredicate(MI);
    AddThumb1SBit(MI, InITBlock);
    return result;
  }

  MI.clear();
  result = decodeThumb2Instruction32(MI, insn32, Address, this);
  if (result != Fail) {
    Size = 4;
    AddThumbPredicate(MI);
    return result;
  }

  MI.clear();
  result = decodeCommonInstruction32(MI, insn32, Address, this);
  if (result != Fail) {
    Size = 4;
    AddThumbPredicate(MI);
    return result;
  }

  MI.clear();
  result = decodeVFPInstruction32(MI, insn32, Address, this);
  if (result != Fail) {
    Size = 4;
    UpdateThumbVFPPredicate(MI);
    return result;
  }

  MI.clear();
  result = decodeNEONDupInstruction32(MI, insn32, Address, this);
  if (result != Fail) {
    Size = 4;
    AddThumbPredicate(MI);
    return result;
  }

  if (fieldFromInstruction32(insn32, 24, 8) == 0xF9) {
    MI.clear();
    uint32_t NEONLdStInsn = insn32;
    NEONLdStInsn &= 0xF0FFFFFF;
    NEONLdStInsn |= 0x04000000;
    result = decodeNEONLoadStoreInstruction32(MI, NEONLdStInsn, Address, this);
    if (result != Fail) {
      Size = 4;
      AddThumbPredicate(MI);
      return result;
    }
  }

  if (fieldFromInstruction32(insn32, 24, 4) == 0xF) {
    MI.clear();
    uint32_t NEONDataInsn = insn32;
    NEONDataInsn &= 0xF0FFFFFF; // Clear bits 27-24
    NEONDataInsn |= (NEONDataInsn & 0x10000000) >> 4; // Move bit 28 to bit 24
    NEONDataInsn |= 0x12000000; // Set bits 28 and 25
    result = decodeNEONDataInstruction32(MI, NEONDataInsn, Address, this);
    if (result != Fail) {
      Size = 4;
      AddThumbPredicate(MI);
      return result;
    }
  }

  return Fail;
}


extern "C" void LLVMInitializeARMDisassembler() {
  TargetRegistry::RegisterMCDisassembler(TheARMTarget,
                                         createARMDisassembler);
  TargetRegistry::RegisterMCDisassembler(TheThumbTarget,
                                         createThumbDisassembler);
}

static const unsigned GPRDecoderTable[] = {
  ARM::R0, ARM::R1, ARM::R2, ARM::R3,
  ARM::R4, ARM::R5, ARM::R6, ARM::R7,
  ARM::R8, ARM::R9, ARM::R10, ARM::R11,
  ARM::R12, ARM::SP, ARM::LR, ARM::PC
};

static DecodeStatus DecodeGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 15)
    return Fail;

  unsigned Register = GPRDecoderTable[RegNo];
  Inst.addOperand(MCOperand::CreateReg(Register));
  return Success;
}

static DecodeStatus DecodeGPRnopcRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                       uint64_t Address, const void *Decoder) {
  if (RegNo == 15) return Fail;
  return DecodeGPRRegisterClass(Inst, RegNo, Address, Decoder);
}

static DecodeStatus DecodetGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 7)
    return Fail;
  return DecodeGPRRegisterClass(Inst, RegNo, Address, Decoder);
}

static DecodeStatus DecodetcGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  unsigned Register = 0;
  switch (RegNo) {
    case 0:
      Register = ARM::R0;
      break;
    case 1:
      Register = ARM::R1;
      break;
    case 2:
      Register = ARM::R2;
      break;
    case 3:
      Register = ARM::R3;
      break;
    case 9:
      Register = ARM::R9;
      break;
    case 12:
      Register = ARM::R12;
      break;
    default:
      return Fail;
    }

  Inst.addOperand(MCOperand::CreateReg(Register));
  return Success;
}

static DecodeStatus DecoderGPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo == 13 || RegNo == 15) return Fail;
  return DecodeGPRRegisterClass(Inst, RegNo, Address, Decoder);
}

static const unsigned SPRDecoderTable[] = { 
     ARM::S0,  ARM::S1,  ARM::S2,  ARM::S3,
     ARM::S4,  ARM::S5,  ARM::S6,  ARM::S7,
     ARM::S8,  ARM::S9, ARM::S10, ARM::S11,
    ARM::S12, ARM::S13, ARM::S14, ARM::S15,
    ARM::S16, ARM::S17, ARM::S18, ARM::S19,
    ARM::S20, ARM::S21, ARM::S22, ARM::S23,
    ARM::S24, ARM::S25, ARM::S26, ARM::S27,
    ARM::S28, ARM::S29, ARM::S30, ARM::S31
};

static DecodeStatus DecodeSPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 31)
    return Fail;

  unsigned Register = SPRDecoderTable[RegNo];
  Inst.addOperand(MCOperand::CreateReg(Register));
  return Success;
}

static const unsigned DPRDecoderTable[] = { 
     ARM::D0,  ARM::D1,  ARM::D2,  ARM::D3,
     ARM::D4,  ARM::D5,  ARM::D6,  ARM::D7,
     ARM::D8,  ARM::D9, ARM::D10, ARM::D11,
    ARM::D12, ARM::D13, ARM::D14, ARM::D15,
    ARM::D16, ARM::D17, ARM::D18, ARM::D19,
    ARM::D20, ARM::D21, ARM::D22, ARM::D23,
    ARM::D24, ARM::D25, ARM::D26, ARM::D27,
    ARM::D28, ARM::D29, ARM::D30, ARM::D31
};

static DecodeStatus DecodeDPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 31)
    return Fail;

  unsigned Register = DPRDecoderTable[RegNo];
  Inst.addOperand(MCOperand::CreateReg(Register));
  return Success;
}

static DecodeStatus DecodeDPR_8RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 7)
    return Fail;
  return DecodeDPRRegisterClass(Inst, RegNo, Address, Decoder);
}

static DecodeStatus DecodeDPR_VFP2RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 15)
    return Fail;
  return DecodeDPRRegisterClass(Inst, RegNo, Address, Decoder);
}

static const unsigned QPRDecoderTable[] = { 
     ARM::Q0,  ARM::Q1,  ARM::Q2,  ARM::Q3,
     ARM::Q4,  ARM::Q5,  ARM::Q6,  ARM::Q7,
     ARM::Q8,  ARM::Q9, ARM::Q10, ARM::Q11,
    ARM::Q12, ARM::Q13, ARM::Q14, ARM::Q15
};


static DecodeStatus DecodeQPRRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
                                   uint64_t Address, const void *Decoder) {
  if (RegNo > 31)
    return Fail;
  RegNo >>= 1;

  unsigned Register = QPRDecoderTable[RegNo];
  Inst.addOperand(MCOperand::CreateReg(Register));
  return Success;
}

static DecodeStatus DecodePredicateOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  if (Val == 0xF) return Fail;
  // AL predicate is not allowed on Thumb1 branches.
  if (Inst.getOpcode() == ARM::tBcc && Val == 0xE)
    return Fail;
  Inst.addOperand(MCOperand::CreateImm(Val));
  if (Val == ARMCC::AL) {
    Inst.addOperand(MCOperand::CreateReg(0));
  } else
    Inst.addOperand(MCOperand::CreateReg(ARM::CPSR));
  return Success;
}

static DecodeStatus DecodeCCOutOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  if (Val)
    Inst.addOperand(MCOperand::CreateReg(ARM::CPSR));
  else
    Inst.addOperand(MCOperand::CreateReg(0));
  return Success;
}

static DecodeStatus DecodeSOImmOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  uint32_t imm = Val & 0xFF;
  uint32_t rot = (Val & 0xF00) >> 7;
  uint32_t rot_imm = (imm >> rot) | (imm << (32-rot));
  Inst.addOperand(MCOperand::CreateImm(rot_imm));
  return Success;
}

static DecodeStatus DecodeBLTargetOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  Val <<= 2;
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<26>(Val)));
  return Success;
}

static DecodeStatus DecodeSORegImmOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rm = fieldFromInstruction32(Val, 0, 4);
  unsigned type = fieldFromInstruction32(Val, 5, 2);
  unsigned imm = fieldFromInstruction32(Val, 7, 5);

  // Register-immediate
  CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));

  ARM_AM::ShiftOpc Shift = ARM_AM::lsl;
  switch (type) {
    case 0:
      Shift = ARM_AM::lsl;
      break;
    case 1:
      Shift = ARM_AM::lsr;
      break;
    case 2:
      Shift = ARM_AM::asr;
      break;
    case 3:
      Shift = ARM_AM::ror;
      break;
  }

  if (Shift == ARM_AM::ror && imm == 0)
    Shift = ARM_AM::rrx;

  unsigned Op = Shift | (imm << 3);
  Inst.addOperand(MCOperand::CreateImm(Op));

  return S;
}

static DecodeStatus DecodeSORegRegOperand(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rm = fieldFromInstruction32(Val, 0, 4);
  unsigned type = fieldFromInstruction32(Val, 5, 2);
  unsigned Rs = fieldFromInstruction32(Val, 8, 4);

  // Register-register
  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rm, Address, Decoder));
  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rs, Address, Decoder));

  ARM_AM::ShiftOpc Shift = ARM_AM::lsl;
  switch (type) {
    case 0:
      Shift = ARM_AM::lsl;
      break;
    case 1:
      Shift = ARM_AM::lsr;
      break;
    case 2:
      Shift = ARM_AM::asr;
      break;
    case 3:
      Shift = ARM_AM::ror;
      break;
  }

  Inst.addOperand(MCOperand::CreateImm(Shift));

  return S;
}

static DecodeStatus DecodeRegListOperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  // Empty register lists are not allowed.
  if (CountPopulation_32(Val) == 0) return Fail;
  for (unsigned i = 0; i < 16; ++i) {
    if (Val & (1 << i)) {
      CHECK(S, DecodeGPRRegisterClass(Inst, i, Address, Decoder));
    }
  }

  return S;
}

static DecodeStatus DecodeSPRRegListOperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Vd = fieldFromInstruction32(Val, 8, 4);
  unsigned regs = Val & 0xFF;

  CHECK(S, DecodeSPRRegisterClass(Inst, Vd, Address, Decoder));
  for (unsigned i = 0; i < (regs - 1); ++i) {
    CHECK(S, DecodeSPRRegisterClass(Inst, ++Vd, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeDPRRegListOperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Vd = fieldFromInstruction32(Val, 8, 4);
  unsigned regs = (Val & 0xFF) / 2;

  CHECK(S, DecodeDPRRegisterClass(Inst, Vd, Address, Decoder));
  for (unsigned i = 0; i < (regs - 1); ++i) {
    CHECK(S, DecodeDPRRegisterClass(Inst, ++Vd, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeBitfieldMaskOperand(llvm::MCInst &Inst, unsigned Val,
                                      uint64_t Address, const void *Decoder) {
  // This operand encodes a mask of contiguous zeros between a specified MSB
  // and LSB.  To decode it, we create the mask of all bits MSB-and-lower,
  // the mask of all bits LSB-and-lower, and then xor them to create
  // the mask of that's all ones on [msb, lsb].  Finally we not it to 
  // create the final mask.
  unsigned msb = fieldFromInstruction32(Val, 5, 5);
  unsigned lsb = fieldFromInstruction32(Val, 0, 5);
  uint32_t msb_mask = (1 << (msb+1)) - 1;
  uint32_t lsb_mask = (1 << lsb) - 1;
  Inst.addOperand(MCOperand::CreateImm(~(msb_mask ^ lsb_mask)));
  return Success;
}

static DecodeStatus DecodeCopMemInstruction(llvm::MCInst &Inst, unsigned Insn,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned pred = fieldFromInstruction32(Insn, 28, 4);
  unsigned CRd = fieldFromInstruction32(Insn, 12, 4);
  unsigned coproc = fieldFromInstruction32(Insn, 8, 4);
  unsigned imm = fieldFromInstruction32(Insn, 0, 8);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned U = fieldFromInstruction32(Insn, 23, 1);

  switch (Inst.getOpcode()) {
    case ARM::LDC_OFFSET:
    case ARM::LDC_PRE:
    case ARM::LDC_POST:
    case ARM::LDC_OPTION:
    case ARM::LDCL_OFFSET:
    case ARM::LDCL_PRE:
    case ARM::LDCL_POST:
    case ARM::LDCL_OPTION:
    case ARM::STC_OFFSET:
    case ARM::STC_PRE:
    case ARM::STC_POST:
    case ARM::STC_OPTION:
    case ARM::STCL_OFFSET:
    case ARM::STCL_PRE:
    case ARM::STCL_POST:
    case ARM::STCL_OPTION:
      if (coproc == 0xA || coproc == 0xB)
        return Fail;
      break;
    default:
      break;
  }

  Inst.addOperand(MCOperand::CreateImm(coproc));
  Inst.addOperand(MCOperand::CreateImm(CRd));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  switch (Inst.getOpcode()) {
    case ARM::LDC_OPTION:
    case ARM::LDCL_OPTION:
    case ARM::LDC2_OPTION:
    case ARM::LDC2L_OPTION:
    case ARM::STC_OPTION:
    case ARM::STCL_OPTION:
    case ARM::STC2_OPTION:
    case ARM::STC2L_OPTION:
    case ARM::LDCL_POST:
    case ARM::STCL_POST:
      break;
    default:
      Inst.addOperand(MCOperand::CreateReg(0));
      break;
  }

  unsigned P = fieldFromInstruction32(Insn, 24, 1);
  unsigned W = fieldFromInstruction32(Insn, 21, 1);

  bool writeback = (P == 0) || (W == 1);
  unsigned idx_mode = 0;
  if (P && writeback)
    idx_mode = ARMII::IndexModePre;
  else if (!P && writeback)
    idx_mode = ARMII::IndexModePost;

  switch (Inst.getOpcode()) {
    case ARM::LDCL_POST:
    case ARM::STCL_POST:
      imm |= U << 8;
    case ARM::LDC_OPTION:
    case ARM::LDCL_OPTION:
    case ARM::LDC2_OPTION:
    case ARM::LDC2L_OPTION:
    case ARM::STC_OPTION:
    case ARM::STCL_OPTION:
    case ARM::STC2_OPTION:
    case ARM::STC2L_OPTION:
      Inst.addOperand(MCOperand::CreateImm(imm));
      break;
    default:
      if (U)
        Inst.addOperand(MCOperand::CreateImm(
            ARM_AM::getAM2Opc(ARM_AM::add, imm, ARM_AM::lsl, idx_mode)));
      else
        Inst.addOperand(MCOperand::CreateImm(
            ARM_AM::getAM2Opc(ARM_AM::sub, imm, ARM_AM::lsl, idx_mode)));
      break;
  }

  switch (Inst.getOpcode()) {
    case ARM::LDC_OFFSET:
    case ARM::LDC_PRE:
    case ARM::LDC_POST:
    case ARM::LDC_OPTION:
    case ARM::LDCL_OFFSET:
    case ARM::LDCL_PRE:
    case ARM::LDCL_POST:
    case ARM::LDCL_OPTION:
    case ARM::STC_OFFSET:
    case ARM::STC_PRE:
    case ARM::STC_POST:
    case ARM::STC_OPTION:
    case ARM::STCL_OFFSET:
    case ARM::STCL_PRE:
    case ARM::STCL_POST:
    case ARM::STCL_OPTION:
      CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));
      break;
    default:
      break;
  }

  return S;
}

static DecodeStatus DecodeAddrMode2IdxInstruction(llvm::MCInst &Inst, unsigned Insn,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned imm = fieldFromInstruction32(Insn, 0, 12);
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);
  unsigned reg = fieldFromInstruction32(Insn, 25, 1);
  unsigned P = fieldFromInstruction32(Insn, 24, 1);
  unsigned W = fieldFromInstruction32(Insn, 21, 1);

  // On stores, the writeback operand precedes Rt.
  switch (Inst.getOpcode()) {
    case ARM::STR_POST_IMM:
    case ARM::STR_POST_REG:
    case ARM::STRB_POST_IMM:
    case ARM::STRB_POST_REG:
    case ARM::STRT_POST_REG:
    case ARM::STRT_POST_IMM:
    case ARM::STRBT_POST_REG:
    case ARM::STRBT_POST_IMM:
      CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
      break;
    default:
      break;
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));

  // On loads, the writeback operand comes after Rt.
  switch (Inst.getOpcode()) {
    case ARM::LDR_POST_IMM:
    case ARM::LDR_POST_REG:
    case ARM::LDRB_POST_IMM:
    case ARM::LDRB_POST_REG:
    case ARM::LDR_PRE:
    case ARM::LDRB_PRE:
    case ARM::LDRBT_POST_REG:
    case ARM::LDRBT_POST_IMM:
    case ARM::LDRT_POST_REG:
    case ARM::LDRT_POST_IMM:
      CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
      break;
    default:
      break;
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));

  ARM_AM::AddrOpc Op = ARM_AM::add;
  if (!fieldFromInstruction32(Insn, 23, 1))
    Op = ARM_AM::sub;

  bool writeback = (P == 0) || (W == 1);
  unsigned idx_mode = 0;
  if (P && writeback)
    idx_mode = ARMII::IndexModePre;
  else if (!P && writeback)
    idx_mode = ARMII::IndexModePost;

  if (writeback && (Rn == 15 || Rn == Rt)) S = Unpredictable; // UNPREDICTABLE

  if (reg) {
    CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rm, Address, Decoder));
    ARM_AM::ShiftOpc Opc = ARM_AM::lsl;
    switch( fieldFromInstruction32(Insn, 5, 2)) {
      case 0:
        Opc = ARM_AM::lsl;
        break;
      case 1:
        Opc = ARM_AM::lsr;
        break;
      case 2:
        Opc = ARM_AM::asr;
        break;
      case 3:
        Opc = ARM_AM::ror;
        break;
      default:
        return Fail;
    }
    unsigned amt = fieldFromInstruction32(Insn, 7, 5);
    unsigned imm = ARM_AM::getAM2Opc(Op, amt, Opc, idx_mode);

    Inst.addOperand(MCOperand::CreateImm(imm));
  } else {
    Inst.addOperand(MCOperand::CreateReg(0));
    unsigned tmp = ARM_AM::getAM2Opc(Op, imm, ARM_AM::lsl, idx_mode);
    Inst.addOperand(MCOperand::CreateImm(tmp));
  }

  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeSORegMemOperand(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 13, 4);
  unsigned Rm = fieldFromInstruction32(Val,  0, 4);
  unsigned type = fieldFromInstruction32(Val, 5, 2);
  unsigned imm = fieldFromInstruction32(Val, 7, 5);
  unsigned U = fieldFromInstruction32(Val, 12, 1);

  ARM_AM::ShiftOpc ShOp = ARM_AM::lsl;
  switch (type) {
    case 0:
      ShOp = ARM_AM::lsl;
      break;
    case 1:
      ShOp = ARM_AM::lsr;
      break;
    case 2:
      ShOp = ARM_AM::asr;
      break;
    case 3:
      ShOp = ARM_AM::ror;
      break;
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  unsigned shift;
  if (U)
    shift = ARM_AM::getAM2Opc(ARM_AM::add, imm, ShOp);
  else
    shift = ARM_AM::getAM2Opc(ARM_AM::sub, imm, ShOp);
  Inst.addOperand(MCOperand::CreateImm(shift));

  return S;
}

static DecodeStatus DecodeAddrMode3Instruction(llvm::MCInst &Inst, unsigned Insn,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned type = fieldFromInstruction32(Insn, 22, 1);
  unsigned imm = fieldFromInstruction32(Insn, 8, 4);
  unsigned U = ((~fieldFromInstruction32(Insn, 23, 1)) & 1) << 8;
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);
  unsigned W = fieldFromInstruction32(Insn, 21, 1);
  unsigned P = fieldFromInstruction32(Insn, 24, 1);

  bool writeback = (W == 1) | (P == 0);

  // For {LD,ST}RD, Rt must be even, else undefined.
  switch (Inst.getOpcode()) {
    case ARM::STRD:
    case ARM::STRD_PRE:
    case ARM::STRD_POST:
    case ARM::LDRD:
    case ARM::LDRD_PRE:
    case ARM::LDRD_POST:
      if (Rt & 0x1) return Fail;
      break;
  default:
    break;
  }

  if (writeback) { // Writeback
    if (P)
      U |= ARMII::IndexModePre << 9;
    else
      U |= ARMII::IndexModePost << 9;

    // On stores, the writeback operand precedes Rt.
    switch (Inst.getOpcode()) {
    case ARM::STRD:
    case ARM::STRD_PRE:
    case ARM::STRD_POST:
    case ARM::STRH:
    case ARM::STRH_PRE:
    case ARM::STRH_POST:
      CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
      break;
    default:
      break;
    }
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  switch (Inst.getOpcode()) {
    case ARM::STRD:
    case ARM::STRD_PRE:
    case ARM::STRD_POST:
    case ARM::LDRD:
    case ARM::LDRD_PRE:
    case ARM::LDRD_POST:
      CHECK(S, DecodeGPRRegisterClass(Inst, Rt+1, Address, Decoder));
      break;
    default:
      break;
  }

  if (writeback) {
    // On loads, the writeback operand comes after Rt.
    switch (Inst.getOpcode()) {
    case ARM::LDRD:
    case ARM::LDRD_PRE:
    case ARM::LDRD_POST:
    case ARM::LDRH:
    case ARM::LDRH_PRE:
    case ARM::LDRH_POST:
    case ARM::LDRSH:
    case ARM::LDRSH_PRE:
    case ARM::LDRSH_POST:
    case ARM::LDRSB:
    case ARM::LDRSB_PRE:
    case ARM::LDRSB_POST:
    case ARM::LDRHTr:
    case ARM::LDRSBTr:
      CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
      break;
    default:
      break;
    }
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));

  if (type) {
    Inst.addOperand(MCOperand::CreateReg(0));
    Inst.addOperand(MCOperand::CreateImm(U | (imm << 4) | Rm));
  } else {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
    Inst.addOperand(MCOperand::CreateImm(U));
  }

  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeRFEInstruction(llvm::MCInst &Inst, unsigned Insn,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned mode = fieldFromInstruction32(Insn, 23, 2);

  switch (mode) {
    case 0:
      mode = ARM_AM::da;
      break;
    case 1:
      mode = ARM_AM::ia;
      break;
    case 2:
      mode = ARM_AM::db;
      break;
    case 3:
      mode = ARM_AM::ib;
      break;
  }

  Inst.addOperand(MCOperand::CreateImm(mode));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));

  return S;
}

static DecodeStatus DecodeMemMultipleWritebackInstruction(llvm::MCInst &Inst,
                                  unsigned Insn,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);
  unsigned reglist = fieldFromInstruction32(Insn, 0, 16);

  if (pred == 0xF) {
    switch (Inst.getOpcode()) {
      case ARM::STMDA:
        Inst.setOpcode(ARM::RFEDA);
        break;
      case ARM::STMDA_UPD:
        Inst.setOpcode(ARM::RFEDA_UPD);
        break;
      case ARM::STMDB:
        Inst.setOpcode(ARM::RFEDB);
        break;
      case ARM::STMDB_UPD:
        Inst.setOpcode(ARM::RFEDB_UPD);
        break;
      case ARM::STMIA:
        Inst.setOpcode(ARM::RFEIA);
        break;
      case ARM::STMIA_UPD:
        Inst.setOpcode(ARM::RFEIA_UPD);
        break;
      case ARM::STMIB:
        Inst.setOpcode(ARM::RFEIB);
        break;
      case ARM::STMIB_UPD:
        Inst.setOpcode(ARM::RFEIB_UPD);
        break;
    }
    return DecodeRFEInstruction(Inst, Insn, Address, Decoder);
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder)); // Tied
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));
  CHECK(S, DecodeRegListOperand(Inst, reglist, Address, Decoder));

  return S;
}

static DecodeStatus DecodeCPSInstruction(llvm::MCInst &Inst, unsigned Insn,
                                 uint64_t Address, const void *Decoder) {
  unsigned imod = fieldFromInstruction32(Insn, 18, 2);
  unsigned M = fieldFromInstruction32(Insn, 17, 1);
  unsigned iflags = fieldFromInstruction32(Insn, 6, 3);
  unsigned mode = fieldFromInstruction32(Insn, 0, 5);

  // imod == '01' --> UNPREDICTABLE
  if (imod == 1) return Fail;

  if (M && mode && imod && iflags) {
    Inst.setOpcode(ARM::CPS3p);
    Inst.addOperand(MCOperand::CreateImm(imod));
    Inst.addOperand(MCOperand::CreateImm(iflags));
    Inst.addOperand(MCOperand::CreateImm(mode));
    return Success;
  } else if (!mode && !M) {
    Inst.setOpcode(ARM::CPS2p);
    Inst.addOperand(MCOperand::CreateImm(imod));
    Inst.addOperand(MCOperand::CreateImm(iflags));
    return Success;
  } else if (!imod && !iflags && M) {
    Inst.setOpcode(ARM::CPS1p);
    Inst.addOperand(MCOperand::CreateImm(mode));
    return Success;
  }

  return Fail;
}

static DecodeStatus DecodeSMLAInstruction(llvm::MCInst &Inst, unsigned Insn,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 8, 4);
  unsigned Ra = fieldFromInstruction32(Insn, 12, 4);
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);

  if (pred == 0xF)
    return DecodeCPSInstruction(Inst, Insn, Address, Decoder);

  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Rm, Address, Decoder));
  CHECK(S, DecodeGPRnopcRegisterClass(Inst, Ra, Address, Decoder));

  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeAddrModeImm12Operand(llvm::MCInst &Inst, unsigned Val,
                           uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned add = fieldFromInstruction32(Val, 12, 1);
  unsigned imm = fieldFromInstruction32(Val, 0, 12);
  unsigned Rn = fieldFromInstruction32(Val, 13, 4);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));

  if (!add) imm *= -1;
  if (imm == 0 && !add) imm = INT32_MIN;
  Inst.addOperand(MCOperand::CreateImm(imm));

  return S;
}

static DecodeStatus DecodeAddrMode5Operand(llvm::MCInst &Inst, unsigned Val,
                                   uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 9, 4);
  unsigned U = fieldFromInstruction32(Val, 8, 1);
  unsigned imm = fieldFromInstruction32(Val, 0, 8);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));

  if (U)
    Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::add, imm)));
  else
    Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::sub, imm)));

  return S;
}

static DecodeStatus DecodeAddrMode7Operand(llvm::MCInst &Inst, unsigned Val,
                                   uint64_t Address, const void *Decoder) {
  return DecodeGPRRegisterClass(Inst, Val, Address, Decoder);
}

static DecodeStatus DecodeBranchImmInstruction(llvm::MCInst &Inst, unsigned Insn,
                                   uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned pred = fieldFromInstruction32(Insn, 28, 4);
  unsigned imm = fieldFromInstruction32(Insn, 0, 24) << 2;

  if (pred == 0xF) {
    Inst.setOpcode(ARM::BLXi);
    imm |= fieldFromInstruction32(Insn, 24, 1) << 1;
    Inst.addOperand(MCOperand::CreateImm(SignExtend32<26>(imm)));
    return S;
  }

  Inst.addOperand(MCOperand::CreateImm(SignExtend32<26>(imm)));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}


static DecodeStatus DecodeVCVTImmOperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(64 - Val));
  return Success;
}

static DecodeStatus DecodeAddrMode6Operand(llvm::MCInst &Inst, unsigned Val,
                                   uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rm = fieldFromInstruction32(Val, 0, 4);
  unsigned align = fieldFromInstruction32(Val, 4, 2);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  if (!align)
    Inst.addOperand(MCOperand::CreateImm(0));
  else
    Inst.addOperand(MCOperand::CreateImm(4 << align));

  return S;
}

static DecodeStatus DecodeVLDInstruction(llvm::MCInst &Inst, unsigned Insn,
                                   uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned wb = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  Rn |= fieldFromInstruction32(Insn, 4, 2) << 4;
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);

  // First output register
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));

  // Second output register
  switch (Inst.getOpcode()) {
    case ARM::VLD1q8:
    case ARM::VLD1q16:
    case ARM::VLD1q32:
    case ARM::VLD1q64:
    case ARM::VLD1q8_UPD:
    case ARM::VLD1q16_UPD:
    case ARM::VLD1q32_UPD:
    case ARM::VLD1q64_UPD:
    case ARM::VLD1d8T:
    case ARM::VLD1d16T:
    case ARM::VLD1d32T:
    case ARM::VLD1d64T:
    case ARM::VLD1d8T_UPD:
    case ARM::VLD1d16T_UPD:
    case ARM::VLD1d32T_UPD:
    case ARM::VLD1d64T_UPD:
    case ARM::VLD1d8Q:
    case ARM::VLD1d16Q:
    case ARM::VLD1d32Q:
    case ARM::VLD1d64Q:
    case ARM::VLD1d8Q_UPD:
    case ARM::VLD1d16Q_UPD:
    case ARM::VLD1d32Q_UPD:
    case ARM::VLD1d64Q_UPD:
    case ARM::VLD2d8:
    case ARM::VLD2d16:
    case ARM::VLD2d32:
    case ARM::VLD2d8_UPD:
    case ARM::VLD2d16_UPD:
    case ARM::VLD2d32_UPD:
    case ARM::VLD2q8:
    case ARM::VLD2q16:
    case ARM::VLD2q32:
    case ARM::VLD2q8_UPD:
    case ARM::VLD2q16_UPD:
    case ARM::VLD2q32_UPD:
    case ARM::VLD3d8:
    case ARM::VLD3d16:
    case ARM::VLD3d32:
    case ARM::VLD3d8_UPD:
    case ARM::VLD3d16_UPD:
    case ARM::VLD3d32_UPD:
    case ARM::VLD4d8:
    case ARM::VLD4d16:
    case ARM::VLD4d32:
    case ARM::VLD4d8_UPD:
    case ARM::VLD4d16_UPD:
    case ARM::VLD4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+1)%32, Address, Decoder));
      break;
    case ARM::VLD2b8:
    case ARM::VLD2b16:
    case ARM::VLD2b32:
    case ARM::VLD2b8_UPD:
    case ARM::VLD2b16_UPD:
    case ARM::VLD2b32_UPD:
    case ARM::VLD3q8:
    case ARM::VLD3q16:
    case ARM::VLD3q32:
    case ARM::VLD3q8_UPD:
    case ARM::VLD3q16_UPD:
    case ARM::VLD3q32_UPD:
    case ARM::VLD4q8:
    case ARM::VLD4q16:
    case ARM::VLD4q32:
    case ARM::VLD4q8_UPD:
    case ARM::VLD4q16_UPD:
    case ARM::VLD4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2)%32, Address, Decoder));
    default:
      break;
  }

  // Third output register
  switch(Inst.getOpcode()) {
    case ARM::VLD1d8T:
    case ARM::VLD1d16T:
    case ARM::VLD1d32T:
    case ARM::VLD1d64T:
    case ARM::VLD1d8T_UPD:
    case ARM::VLD1d16T_UPD:
    case ARM::VLD1d32T_UPD:
    case ARM::VLD1d64T_UPD:
    case ARM::VLD1d8Q:
    case ARM::VLD1d16Q:
    case ARM::VLD1d32Q:
    case ARM::VLD1d64Q:
    case ARM::VLD1d8Q_UPD:
    case ARM::VLD1d16Q_UPD:
    case ARM::VLD1d32Q_UPD:
    case ARM::VLD1d64Q_UPD:
    case ARM::VLD2q8:
    case ARM::VLD2q16:
    case ARM::VLD2q32:
    case ARM::VLD2q8_UPD:
    case ARM::VLD2q16_UPD:
    case ARM::VLD2q32_UPD:
    case ARM::VLD3d8:
    case ARM::VLD3d16:
    case ARM::VLD3d32:
    case ARM::VLD3d8_UPD:
    case ARM::VLD3d16_UPD:
    case ARM::VLD3d32_UPD:
    case ARM::VLD4d8:
    case ARM::VLD4d16:
    case ARM::VLD4d32:
    case ARM::VLD4d8_UPD:
    case ARM::VLD4d16_UPD:
    case ARM::VLD4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2)%32, Address, Decoder));
      break;
    case ARM::VLD3q8:
    case ARM::VLD3q16:
    case ARM::VLD3q32:
    case ARM::VLD3q8_UPD:
    case ARM::VLD3q16_UPD:
    case ARM::VLD3q32_UPD:
    case ARM::VLD4q8:
    case ARM::VLD4q16:
    case ARM::VLD4q32:
    case ARM::VLD4q8_UPD:
    case ARM::VLD4q16_UPD:
    case ARM::VLD4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+4)%32, Address, Decoder));
      break;
    default:
      break;
  }

  // Fourth output register
  switch (Inst.getOpcode()) {
    case ARM::VLD1d8Q:
    case ARM::VLD1d16Q:
    case ARM::VLD1d32Q:
    case ARM::VLD1d64Q:
    case ARM::VLD1d8Q_UPD:
    case ARM::VLD1d16Q_UPD:
    case ARM::VLD1d32Q_UPD:
    case ARM::VLD1d64Q_UPD:
    case ARM::VLD2q8:
    case ARM::VLD2q16:
    case ARM::VLD2q32:
    case ARM::VLD2q8_UPD:
    case ARM::VLD2q16_UPD:
    case ARM::VLD2q32_UPD:
    case ARM::VLD4d8:
    case ARM::VLD4d16:
    case ARM::VLD4d32:
    case ARM::VLD4d8_UPD:
    case ARM::VLD4d16_UPD:
    case ARM::VLD4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+3)%32, Address, Decoder));
      break;
    case ARM::VLD4q8:
    case ARM::VLD4q16:
    case ARM::VLD4q32:
    case ARM::VLD4q8_UPD:
    case ARM::VLD4q16_UPD:
    case ARM::VLD4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+6)%32, Address, Decoder));
      break;
    default:
      break;
  }

  // Writeback operand
  switch (Inst.getOpcode()) {
    case ARM::VLD1d8_UPD:
    case ARM::VLD1d16_UPD:
    case ARM::VLD1d32_UPD:
    case ARM::VLD1d64_UPD:
    case ARM::VLD1q8_UPD:
    case ARM::VLD1q16_UPD:
    case ARM::VLD1q32_UPD:
    case ARM::VLD1q64_UPD:
    case ARM::VLD1d8T_UPD:
    case ARM::VLD1d16T_UPD:
    case ARM::VLD1d32T_UPD:
    case ARM::VLD1d64T_UPD:
    case ARM::VLD1d8Q_UPD:
    case ARM::VLD1d16Q_UPD:
    case ARM::VLD1d32Q_UPD:
    case ARM::VLD1d64Q_UPD:
    case ARM::VLD2d8_UPD:
    case ARM::VLD2d16_UPD:
    case ARM::VLD2d32_UPD:
    case ARM::VLD2q8_UPD:
    case ARM::VLD2q16_UPD:
    case ARM::VLD2q32_UPD:
    case ARM::VLD2b8_UPD:
    case ARM::VLD2b16_UPD:
    case ARM::VLD2b32_UPD:
    case ARM::VLD3d8_UPD:
    case ARM::VLD3d16_UPD:
    case ARM::VLD3d32_UPD:
    case ARM::VLD3q8_UPD:
    case ARM::VLD3q16_UPD:
    case ARM::VLD3q32_UPD:
    case ARM::VLD4d8_UPD:
    case ARM::VLD4d16_UPD:
    case ARM::VLD4d32_UPD:
    case ARM::VLD4q8_UPD:
    case ARM::VLD4q16_UPD:
    case ARM::VLD4q32_UPD:
      CHECK(S, DecodeGPRRegisterClass(Inst, wb, Address, Decoder));
      break;
    default:
      break;
  }

  // AddrMode6 Base (register+alignment)
  CHECK(S, DecodeAddrMode6Operand(Inst, Rn, Address, Decoder));

  // AddrMode6 Offset (register)
  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeVSTInstruction(llvm::MCInst &Inst, unsigned Insn,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned wb = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  Rn |= fieldFromInstruction32(Insn, 4, 2) << 4;
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);

  // Writeback Operand
  switch (Inst.getOpcode()) {
    case ARM::VST1d8_UPD:
    case ARM::VST1d16_UPD:
    case ARM::VST1d32_UPD:
    case ARM::VST1d64_UPD:
    case ARM::VST1q8_UPD:
    case ARM::VST1q16_UPD:
    case ARM::VST1q32_UPD:
    case ARM::VST1q64_UPD:
    case ARM::VST1d8T_UPD:
    case ARM::VST1d16T_UPD:
    case ARM::VST1d32T_UPD:
    case ARM::VST1d64T_UPD:
    case ARM::VST1d8Q_UPD:
    case ARM::VST1d16Q_UPD:
    case ARM::VST1d32Q_UPD:
    case ARM::VST1d64Q_UPD:
    case ARM::VST2d8_UPD:
    case ARM::VST2d16_UPD:
    case ARM::VST2d32_UPD:
    case ARM::VST2q8_UPD:
    case ARM::VST2q16_UPD:
    case ARM::VST2q32_UPD:
    case ARM::VST2b8_UPD:
    case ARM::VST2b16_UPD:
    case ARM::VST2b32_UPD:
    case ARM::VST3d8_UPD:
    case ARM::VST3d16_UPD:
    case ARM::VST3d32_UPD:
    case ARM::VST3q8_UPD:
    case ARM::VST3q16_UPD:
    case ARM::VST3q32_UPD:
    case ARM::VST4d8_UPD:
    case ARM::VST4d16_UPD:
    case ARM::VST4d32_UPD:
    case ARM::VST4q8_UPD:
    case ARM::VST4q16_UPD:
    case ARM::VST4q32_UPD:
      CHECK(S, DecodeGPRRegisterClass(Inst, wb, Address, Decoder));
      break;
    default:
      break;
  }

  // AddrMode6 Base (register+alignment)
  CHECK(S, DecodeAddrMode6Operand(Inst, Rn, Address, Decoder));

  // AddrMode6 Offset (register)
  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  // First input register
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));

  // Second input register
  switch (Inst.getOpcode()) {
    case ARM::VST1q8:
    case ARM::VST1q16:
    case ARM::VST1q32:
    case ARM::VST1q64:
    case ARM::VST1q8_UPD:
    case ARM::VST1q16_UPD:
    case ARM::VST1q32_UPD:
    case ARM::VST1q64_UPD:
    case ARM::VST1d8T:
    case ARM::VST1d16T:
    case ARM::VST1d32T:
    case ARM::VST1d64T:
    case ARM::VST1d8T_UPD:
    case ARM::VST1d16T_UPD:
    case ARM::VST1d32T_UPD:
    case ARM::VST1d64T_UPD:
    case ARM::VST1d8Q:
    case ARM::VST1d16Q:
    case ARM::VST1d32Q:
    case ARM::VST1d64Q:
    case ARM::VST1d8Q_UPD:
    case ARM::VST1d16Q_UPD:
    case ARM::VST1d32Q_UPD:
    case ARM::VST1d64Q_UPD:
    case ARM::VST2d8:
    case ARM::VST2d16:
    case ARM::VST2d32:
    case ARM::VST2d8_UPD:
    case ARM::VST2d16_UPD:
    case ARM::VST2d32_UPD:
    case ARM::VST2q8:
    case ARM::VST2q16:
    case ARM::VST2q32:
    case ARM::VST2q8_UPD:
    case ARM::VST2q16_UPD:
    case ARM::VST2q32_UPD:
    case ARM::VST3d8:
    case ARM::VST3d16:
    case ARM::VST3d32:
    case ARM::VST3d8_UPD:
    case ARM::VST3d16_UPD:
    case ARM::VST3d32_UPD:
    case ARM::VST4d8:
    case ARM::VST4d16:
    case ARM::VST4d32:
    case ARM::VST4d8_UPD:
    case ARM::VST4d16_UPD:
    case ARM::VST4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+1)%32, Address, Decoder));
      break;
    case ARM::VST2b8:
    case ARM::VST2b16:
    case ARM::VST2b32:
    case ARM::VST2b8_UPD:
    case ARM::VST2b16_UPD:
    case ARM::VST2b32_UPD:
    case ARM::VST3q8:
    case ARM::VST3q16:
    case ARM::VST3q32:
    case ARM::VST3q8_UPD:
    case ARM::VST3q16_UPD:
    case ARM::VST3q32_UPD:
    case ARM::VST4q8:
    case ARM::VST4q16:
    case ARM::VST4q32:
    case ARM::VST4q8_UPD:
    case ARM::VST4q16_UPD:
    case ARM::VST4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2)%32, Address, Decoder));
      break;
    default:
      break;
  }

  // Third input register
  switch (Inst.getOpcode()) {
    case ARM::VST1d8T:
    case ARM::VST1d16T:
    case ARM::VST1d32T:
    case ARM::VST1d64T:
    case ARM::VST1d8T_UPD:
    case ARM::VST1d16T_UPD:
    case ARM::VST1d32T_UPD:
    case ARM::VST1d64T_UPD:
    case ARM::VST1d8Q:
    case ARM::VST1d16Q:
    case ARM::VST1d32Q:
    case ARM::VST1d64Q:
    case ARM::VST1d8Q_UPD:
    case ARM::VST1d16Q_UPD:
    case ARM::VST1d32Q_UPD:
    case ARM::VST1d64Q_UPD:
    case ARM::VST2q8:
    case ARM::VST2q16:
    case ARM::VST2q32:
    case ARM::VST2q8_UPD:
    case ARM::VST2q16_UPD:
    case ARM::VST2q32_UPD:
    case ARM::VST3d8:
    case ARM::VST3d16:
    case ARM::VST3d32:
    case ARM::VST3d8_UPD:
    case ARM::VST3d16_UPD:
    case ARM::VST3d32_UPD:
    case ARM::VST4d8:
    case ARM::VST4d16:
    case ARM::VST4d32:
    case ARM::VST4d8_UPD:
    case ARM::VST4d16_UPD:
    case ARM::VST4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2)%32, Address, Decoder));
      break;
    case ARM::VST3q8:
    case ARM::VST3q16:
    case ARM::VST3q32:
    case ARM::VST3q8_UPD:
    case ARM::VST3q16_UPD:
    case ARM::VST3q32_UPD:
    case ARM::VST4q8:
    case ARM::VST4q16:
    case ARM::VST4q32:
    case ARM::VST4q8_UPD:
    case ARM::VST4q16_UPD:
    case ARM::VST4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+4)%32, Address, Decoder));
      break;
    default:
      break;
  }

  // Fourth input register
  switch (Inst.getOpcode()) {
    case ARM::VST1d8Q:
    case ARM::VST1d16Q:
    case ARM::VST1d32Q:
    case ARM::VST1d64Q:
    case ARM::VST1d8Q_UPD:
    case ARM::VST1d16Q_UPD:
    case ARM::VST1d32Q_UPD:
    case ARM::VST1d64Q_UPD:
    case ARM::VST2q8:
    case ARM::VST2q16:
    case ARM::VST2q32:
    case ARM::VST2q8_UPD:
    case ARM::VST2q16_UPD:
    case ARM::VST2q32_UPD:
    case ARM::VST4d8:
    case ARM::VST4d16:
    case ARM::VST4d32:
    case ARM::VST4d8_UPD:
    case ARM::VST4d16_UPD:
    case ARM::VST4d32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+3)%32, Address, Decoder));
      break;
    case ARM::VST4q8:
    case ARM::VST4q16:
    case ARM::VST4q32:
    case ARM::VST4q8_UPD:
    case ARM::VST4q16_UPD:
    case ARM::VST4q32_UPD:
      CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+6)%32, Address, Decoder));
      break;
    default:
      break;
  }

  return S;
}

static DecodeStatus DecodeVLD1DupInstruction(llvm::MCInst &Inst, unsigned Insn,
                                    uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned align = fieldFromInstruction32(Insn, 4, 1);
  unsigned size = fieldFromInstruction32(Insn, 6, 2);
  unsigned regs = fieldFromInstruction32(Insn, 5, 1) + 1;

  align *= (1 << size);

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  if (regs == 2) {
    CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+1)%32, Address, Decoder));
  }
  if (Rm == 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));

  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeVLD2DupInstruction(llvm::MCInst &Inst, unsigned Insn,
                                    uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned align = fieldFromInstruction32(Insn, 4, 1);
  unsigned size = 1 << fieldFromInstruction32(Insn, 6, 2);
  unsigned inc = fieldFromInstruction32(Insn, 5, 1) + 1;
  align *= 2*size;

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+inc)%32, Address, Decoder));
  if (Rm == 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));

  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeVLD3DupInstruction(llvm::MCInst &Inst, unsigned Insn,
                                    uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned inc = fieldFromInstruction32(Insn, 5, 1) + 1;

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+inc)%32, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2*inc)%32, Address, Decoder));
  if (Rm == 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(0));

  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeVLD4DupInstruction(llvm::MCInst &Inst, unsigned Insn,
                                    uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned size = fieldFromInstruction32(Insn, 6, 2);
  unsigned inc = fieldFromInstruction32(Insn, 5, 1) + 1;
  unsigned align = fieldFromInstruction32(Insn, 4, 1);

  if (size == 0x3) {
    size = 4;
    align = 16;
  } else {
    if (size == 2) {
      size = 1 << size;
      align *= 8;
    } else {
      size = 1 << size;
      align *= 4*size;
    }
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+inc)%32, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+2*inc)%32, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, (Rd+3*inc)%32, Address, Decoder));
  if (Rm == 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));

  if (Rm == 0xD)
    Inst.addOperand(MCOperand::CreateReg(0));
  else if (Rm != 0xF) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeNEONModImmInstruction(llvm::MCInst &Inst, unsigned Insn,
                                        uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned imm = fieldFromInstruction32(Insn, 0, 4);
  imm |= fieldFromInstruction32(Insn, 16, 3) << 4;
  imm |= fieldFromInstruction32(Insn, 24, 1) << 7;
  imm |= fieldFromInstruction32(Insn, 8, 4) << 8;
  imm |= fieldFromInstruction32(Insn, 5, 1) << 12;
  unsigned Q = fieldFromInstruction32(Insn, 6, 1);

  if (Q) {
    CHECK(S, DecodeQPRRegisterClass(Inst, Rd, Address, Decoder));
  } else {
    CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  }

  Inst.addOperand(MCOperand::CreateImm(imm));

  switch (Inst.getOpcode()) {
    case ARM::VORRiv4i16:
    case ARM::VORRiv2i32:
    case ARM::VBICiv4i16:
    case ARM::VBICiv2i32:
      CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
      break;
    case ARM::VORRiv8i16:
    case ARM::VORRiv4i32:
    case ARM::VBICiv8i16:
    case ARM::VBICiv4i32:
      CHECK(S, DecodeQPRRegisterClass(Inst, Rd, Address, Decoder));
      break;
    default:
      break;
  }

  return S;
}

static DecodeStatus DecodeVSHLMaxInstruction(llvm::MCInst &Inst, unsigned Insn,
                                        uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  Rm |= fieldFromInstruction32(Insn, 5, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 18, 2);

  CHECK(S, DecodeQPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rm, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(8 << size));

  return S;
}

static DecodeStatus DecodeShiftRight8Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(8 - Val));
  return Success;
}

static DecodeStatus DecodeShiftRight16Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(16 - Val));
  return Success;
}

static DecodeStatus DecodeShiftRight32Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(32 - Val));
  return Success;
}

static DecodeStatus DecodeShiftRight64Imm(llvm::MCInst &Inst, unsigned Val,
                               uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(64 - Val));
  return Success;
}

static DecodeStatus DecodeTBLInstruction(llvm::MCInst &Inst, unsigned Insn,
                               uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  Rn |= fieldFromInstruction32(Insn, 7, 1) << 4;
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  Rm |= fieldFromInstruction32(Insn, 5, 1) << 4;
  unsigned op = fieldFromInstruction32(Insn, 6, 1);
  unsigned length = fieldFromInstruction32(Insn, 8, 2) + 1;

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  if (op) {
    CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder)); // Writeback
  }

  for (unsigned i = 0; i < length; ++i) {
    CHECK(S, DecodeDPRRegisterClass(Inst, (Rn+i)%32, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rm, Address, Decoder));

  return S;
}

static DecodeStatus DecodeVFPfpImm(llvm::MCInst &Inst, unsigned Val,
                            uint64_t Address, const void *Decoder) {
  // The immediate needs to be a fully instantiated float.  However, the
  // auto-generated decoder is only able to fill in some of the bits
  // necessary.  For instance, the 'b' bit is replicated multiple times,
  // and is even present in inverted form in one bit.  We do a little
  // binary parsing here to fill in those missing bits, and then
  // reinterpret it all as a float.
  union {
    uint32_t integer;
    float fp;
  } fp_conv;

  fp_conv.integer = Val;
  uint32_t b = fieldFromInstruction32(Val, 25, 1);
  fp_conv.integer |= b << 26;
  fp_conv.integer |= b << 27;
  fp_conv.integer |= b << 28;
  fp_conv.integer |= b << 29;
  fp_conv.integer |= (~b & 0x1) << 30;

  Inst.addOperand(MCOperand::CreateFPImm(fp_conv.fp));
  return Success;
}

static DecodeStatus DecodeThumbAddSpecialReg(llvm::MCInst &Inst, uint16_t Insn,
                                     uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned dst = fieldFromInstruction16(Insn, 8, 3);
  unsigned imm = fieldFromInstruction16(Insn, 0, 8);

  CHECK(S, DecodetGPRRegisterClass(Inst, dst, Address, Decoder));

  if (Inst.getOpcode() == ARM::tADR)
    Inst.addOperand(MCOperand::CreateReg(ARM::PC));
  else if (Inst.getOpcode() == ARM::tADDrSPi)
    Inst.addOperand(MCOperand::CreateReg(ARM::SP));
  else
    return Fail;

  Inst.addOperand(MCOperand::CreateImm(imm));
  return S;
}

static DecodeStatus DecodeThumbBROperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<12>(Val << 1)));
  return Success;
}

static DecodeStatus DecodeT2BROperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<21>(Val)));
  return Success;
}

static DecodeStatus DecodeThumbCmpBROperand(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<7>(Val << 1)));
  return Success;
}

static DecodeStatus DecodeThumbAddrModeRR(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 0, 3);
  unsigned Rm = fieldFromInstruction32(Val, 3, 3);

  CHECK(S, DecodetGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodetGPRRegisterClass(Inst, Rm, Address, Decoder));

  return S;
}

static DecodeStatus DecodeThumbAddrModeIS(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 0, 3);
  unsigned imm = fieldFromInstruction32(Val, 3, 5);

  CHECK(S, DecodetGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(imm));

  return S;
}

static DecodeStatus DecodeThumbAddrModePC(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(Val << 2));

  return Success;
}

static DecodeStatus DecodeThumbAddrModeSP(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateReg(ARM::SP));
  Inst.addOperand(MCOperand::CreateImm(Val << 2));

  return Success;
}

static DecodeStatus DecodeT2AddrModeSOReg(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 6, 4);
  unsigned Rm = fieldFromInstruction32(Val, 2, 4);
  unsigned imm = fieldFromInstruction32(Val, 0, 2);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecoderGPRRegisterClass(Inst, Rm, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(imm));

  return S;
}

static DecodeStatus DecodeT2LoadShift(llvm::MCInst &Inst, unsigned Insn,
                              uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  if (Inst.getOpcode() != ARM::t2PLDs) {
    unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
    CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  }

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  if (Rn == 0xF) {
    switch (Inst.getOpcode()) {
      case ARM::t2LDRBs:
        Inst.setOpcode(ARM::t2LDRBpci);
        break;
      case ARM::t2LDRHs:
        Inst.setOpcode(ARM::t2LDRHpci);
        break;
      case ARM::t2LDRSHs:
        Inst.setOpcode(ARM::t2LDRSHpci);
        break;
      case ARM::t2LDRSBs:
        Inst.setOpcode(ARM::t2LDRSBpci);
        break;
      case ARM::t2PLDs:
        Inst.setOpcode(ARM::t2PLDi12);
        Inst.addOperand(MCOperand::CreateReg(ARM::PC));
        break;
      default:
        return Fail;
    }

    int imm = fieldFromInstruction32(Insn, 0, 12);
    if (!fieldFromInstruction32(Insn, 23, 1)) imm *= -1;
    Inst.addOperand(MCOperand::CreateImm(imm));

    return S;
  }

  unsigned addrmode = fieldFromInstruction32(Insn, 4, 2);
  addrmode |= fieldFromInstruction32(Insn, 0, 4) << 2;
  addrmode |= fieldFromInstruction32(Insn, 16, 4) << 6;
  CHECK(S, DecodeT2AddrModeSOReg(Inst, addrmode, Address, Decoder));

  return S;
}

static DecodeStatus DecodeT2Imm8S4(llvm::MCInst &Inst, unsigned Val,
                           uint64_t Address, const void *Decoder) {
  int imm = Val & 0xFF;
  if (!(Val & 0x100)) imm *= -1;
  Inst.addOperand(MCOperand::CreateImm(imm << 2));

  return Success;
}

static DecodeStatus DecodeT2AddrModeImm8s4(llvm::MCInst &Inst, unsigned Val,
                                   uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 9, 4);
  unsigned imm = fieldFromInstruction32(Val, 0, 9);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeT2Imm8S4(Inst, imm, Address, Decoder));

  return S;
}

static DecodeStatus DecodeT2Imm8(llvm::MCInst &Inst, unsigned Val,
                         uint64_t Address, const void *Decoder) {
  int imm = Val & 0xFF;
  if (!(Val & 0x100)) imm *= -1;
  Inst.addOperand(MCOperand::CreateImm(imm));

  return Success;
}


static DecodeStatus DecodeT2AddrModeImm8(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 9, 4);
  unsigned imm = fieldFromInstruction32(Val, 0, 9);

  // Some instructions always use an additive offset.
  switch (Inst.getOpcode()) {
    case ARM::t2LDRT:
    case ARM::t2LDRBT:
    case ARM::t2LDRHT:
    case ARM::t2LDRSBT:
    case ARM::t2LDRSHT:
      imm |= 0x100;
      break;
    default:
      break;
  }

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeT2Imm8(Inst, imm, Address, Decoder));

  return S;
}


static DecodeStatus DecodeT2AddrModeImm12(llvm::MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Val, 13, 4);
  unsigned imm = fieldFromInstruction32(Val, 0, 12);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(imm));

  return S;
}


static DecodeStatus DecodeThumbAddSPImm(llvm::MCInst &Inst, uint16_t Insn,
                                uint64_t Address, const void *Decoder) {
  unsigned imm = fieldFromInstruction16(Insn, 0, 7);

  Inst.addOperand(MCOperand::CreateReg(ARM::SP));
  Inst.addOperand(MCOperand::CreateReg(ARM::SP));
  Inst.addOperand(MCOperand::CreateImm(imm));

  return Success;
}

static DecodeStatus DecodeThumbAddSPReg(llvm::MCInst &Inst, uint16_t Insn,
                                uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  if (Inst.getOpcode() == ARM::tADDrSP) {
    unsigned Rdm = fieldFromInstruction16(Insn, 0, 3);
    Rdm |= fieldFromInstruction16(Insn, 7, 1) << 3;

    CHECK(S, DecodeGPRRegisterClass(Inst, Rdm, Address, Decoder));
    Inst.addOperand(MCOperand::CreateReg(ARM::SP));
    CHECK(S, DecodeGPRRegisterClass(Inst, Rdm, Address, Decoder));
  } else if (Inst.getOpcode() == ARM::tADDspr) {
    unsigned Rm = fieldFromInstruction16(Insn, 3, 4);

    Inst.addOperand(MCOperand::CreateReg(ARM::SP));
    Inst.addOperand(MCOperand::CreateReg(ARM::SP));
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  return S;
}

static DecodeStatus DecodeThumbCPS(llvm::MCInst &Inst, uint16_t Insn,
                           uint64_t Address, const void *Decoder) {
  unsigned imod = fieldFromInstruction16(Insn, 4, 1) | 0x2;
  unsigned flags = fieldFromInstruction16(Insn, 0, 3);

  Inst.addOperand(MCOperand::CreateImm(imod));
  Inst.addOperand(MCOperand::CreateImm(flags));

  return Success;
}

static DecodeStatus DecodePostIdxReg(llvm::MCInst &Inst, unsigned Insn,
                             uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned add = fieldFromInstruction32(Insn, 4, 1);

  CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder)) ;
  Inst.addOperand(MCOperand::CreateImm(add));

  return S;
}

static DecodeStatus DecodeThumbBLXOffset(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder) {
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<22>(Val << 1)));
  return Success;
}

static DecodeStatus DecodeCoprocessor(llvm::MCInst &Inst, unsigned Val,
                              uint64_t Address, const void *Decoder) {
  if (Val == 0xA || Val == 0xB)
    return Fail;

  Inst.addOperand(MCOperand::CreateImm(Val));
  return Success;
}

static DecodeStatus DecodeThumb2BCCInstruction(llvm::MCInst &Inst, unsigned Insn,
                                       uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned pred = fieldFromInstruction32(Insn, 22, 4);
  if (pred == 0xE || pred == 0xF) {
    unsigned opc = fieldFromInstruction32(Insn, 4, 2);
    switch (opc) {
      default:
        return Fail;
      case 0:
        Inst.setOpcode(ARM::t2DSB);
        break;
      case 1:
        Inst.setOpcode(ARM::t2DMB);
        break;
      case 2:
        Inst.setOpcode(ARM::t2ISB);
        return Success;
    }

    unsigned imm = fieldFromInstruction32(Insn, 0, 4);
    return DecodeMemBarrierOption(Inst, imm, Address, Decoder);
  }

  unsigned brtarget = fieldFromInstruction32(Insn, 0, 11) << 1;
  brtarget |= fieldFromInstruction32(Insn, 11, 1) << 19;
  brtarget |= fieldFromInstruction32(Insn, 13, 1) << 18;
  brtarget |= fieldFromInstruction32(Insn, 16, 6) << 12;
  brtarget |= fieldFromInstruction32(Insn, 26, 1) << 20;

  CHECK(S, DecodeT2BROperand(Inst, brtarget, Address, Decoder));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

// Decode a shifted immediate operand.  These basically consist
// of an 8-bit value, and a 4-bit directive that specifies either
// a splat operation or a rotation.
static DecodeStatus DecodeT2SOImm(llvm::MCInst &Inst, unsigned Val,
                          uint64_t Address, const void *Decoder) {
  unsigned ctrl = fieldFromInstruction32(Val, 10, 2);
  if (ctrl == 0) {
    unsigned byte = fieldFromInstruction32(Val, 8, 2);
    unsigned imm = fieldFromInstruction32(Val, 0, 8);
    switch (byte) {
      case 0:
        Inst.addOperand(MCOperand::CreateImm(imm));
        break;
      case 1:
        Inst.addOperand(MCOperand::CreateImm((imm << 16) | imm));
        break;
      case 2:
        Inst.addOperand(MCOperand::CreateImm((imm << 24) | (imm << 8)));
        break;
      case 3:
        Inst.addOperand(MCOperand::CreateImm((imm << 24) | (imm << 16) |
                                             (imm << 8)  |  imm));
        break;
    }
  } else {
    unsigned unrot = fieldFromInstruction32(Val, 0, 7) | 0x80;
    unsigned rot = fieldFromInstruction32(Val, 7, 5);
    unsigned imm = (unrot >> rot) | (unrot << ((32-rot)&31));
    Inst.addOperand(MCOperand::CreateImm(imm));
  }

  return Success;
}

static DecodeStatus DecodeThumbBCCTargetOperand(llvm::MCInst &Inst, unsigned Val,
                                        uint64_t Address, const void *Decoder){
  Inst.addOperand(MCOperand::CreateImm(Val << 1));
  return Success;
}

static DecodeStatus DecodeThumbBLTargetOperand(llvm::MCInst &Inst, unsigned Val,
                                       uint64_t Address, const void *Decoder){
  Inst.addOperand(MCOperand::CreateImm(SignExtend32<22>(Val << 1)));
  return Success;
}

static DecodeStatus DecodeMemBarrierOption(llvm::MCInst &Inst, unsigned Val,
                                   uint64_t Address, const void *Decoder) {
  switch (Val) {
  default:
    return Fail;
  case 0xF: // SY
  case 0xE: // ST
  case 0xB: // ISH
  case 0xA: // ISHST
  case 0x7: // NSH
  case 0x6: // NSHST
  case 0x3: // OSH
  case 0x2: // OSHST
    break;
  }

  Inst.addOperand(MCOperand::CreateImm(Val));
  return Success;
}

static DecodeStatus DecodeMSRMask(llvm::MCInst &Inst, unsigned Val,
                          uint64_t Address, const void *Decoder) {
  if (!Val) return Fail;
  Inst.addOperand(MCOperand::CreateImm(Val));
  return Success;
}

static DecodeStatus DecodeDoubleRegLoad(llvm::MCInst &Inst, unsigned Insn,
                                          uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);

  if ((Rt & 1) || Rt == 0xE || Rn == 0xF) return Fail;

  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rt+1, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}


static DecodeStatus DecodeDoubleRegStore(llvm::MCInst &Inst, unsigned Insn,
                                          uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  unsigned Rt = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);

  CHECK(S, DecoderGPRRegisterClass(Inst, Rd, Address, Decoder));

  if ((Rt & 1) || Rt == 0xE || Rn == 0xF) return Fail;
  if (Rd == Rn || Rd == Rt || Rd == Rt+1) return Fail;

  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rt+1, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeSTRPreImm(llvm::MCInst &Inst, unsigned Insn,
                            uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
  unsigned imm = fieldFromInstruction32(Insn, 0, 12);
  imm |= fieldFromInstruction32(Insn, 16, 4) << 13;
  imm |= fieldFromInstruction32(Insn, 23, 1) << 12;
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);

  if (Rn == 0xF || Rn == Rt) return Unpredictable; // UNPREDICTABLE

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  CHECK(S, DecodeAddrModeImm12Operand(Inst, imm, Address, Decoder));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeSTRPreReg(llvm::MCInst &Inst, unsigned Insn,
                            uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rt = fieldFromInstruction32(Insn, 12, 4);
  unsigned imm = fieldFromInstruction32(Insn, 0, 12);
  imm |= fieldFromInstruction32(Insn, 16, 4) << 13;
  imm |= fieldFromInstruction32(Insn, 23, 1) << 12;
  unsigned pred = fieldFromInstruction32(Insn, 28, 4);

  if (Rn == 0xF || Rn == Rt) return Unpredictable; // UNPREDICTABLE

  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  CHECK(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder));
  CHECK(S, DecodeSORegMemOperand(Inst, imm, Address, Decoder));
  CHECK(S, DecodePredicateOperand(Inst, pred, Address, Decoder));

  return S;
}

static DecodeStatus DecodeVLD1LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 5, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 6, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 4, 2) != 0)
        align = 4;
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}

static DecodeStatus DecodeVST1LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 5, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 6, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 4, 2) != 0)
        align = 4;
  }

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}


static DecodeStatus DecodeVLD2LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      index = fieldFromInstruction32(Insn, 5, 3);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 2;
      break;
    case 1:
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 4;
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 5, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 4, 1) != 0)
        align = 8;
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}

static DecodeStatus DecodeVST2LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      index = fieldFromInstruction32(Insn, 5, 3);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 2;
      break;
    case 1:
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 4;
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 5, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 4, 1) != 0)
        align = 8;
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}


static DecodeStatus DecodeVLD3LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 4, 2))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}

static DecodeStatus DecodeVST3LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 4, 1))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 4, 2))
        return Fail; // UNDEFINED
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}


static DecodeStatus DecodeVLD4LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 4;
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 8;
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 4, 2))
        align = 4 << fieldFromInstruction32(Insn, 4, 2);
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+3*inc, Address, Decoder));

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+3*inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}

static DecodeStatus DecodeVST4LN(llvm::MCInst &Inst, unsigned Insn,
                         uint64_t Address, const void *Decoder) {
  DecodeStatus S = Success;

  unsigned Rn = fieldFromInstruction32(Insn, 16, 4);
  unsigned Rm = fieldFromInstruction32(Insn, 0, 4);
  unsigned Rd = fieldFromInstruction32(Insn, 12, 4);
  Rd |= fieldFromInstruction32(Insn, 22, 1) << 4;
  unsigned size = fieldFromInstruction32(Insn, 10, 2);

  unsigned align = 0;
  unsigned index = 0;
  unsigned inc = 1;
  switch (size) {
    default:
      return Fail;
    case 0:
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 4;
      index = fieldFromInstruction32(Insn, 5, 3);
      break;
    case 1:
      if (fieldFromInstruction32(Insn, 4, 1))
        align = 8;
      index = fieldFromInstruction32(Insn, 6, 2);
      if (fieldFromInstruction32(Insn, 5, 1))
        inc = 2;
      break;
    case 2:
      if (fieldFromInstruction32(Insn, 4, 2))
        align = 4 << fieldFromInstruction32(Insn, 4, 2);
      index = fieldFromInstruction32(Insn, 7, 1);
      if (fieldFromInstruction32(Insn, 6, 1))
        inc = 2;
      break;
  }

  if (Rm != 0xF) { // Writeback
    CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  }
  CHECK(S, DecodeGPRRegisterClass(Inst, Rn, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(align));
  if (Rm != 0xF && Rm != 0xD) {
    CHECK(S, DecodeGPRRegisterClass(Inst, Rm, Address, Decoder));
  }

  CHECK(S, DecodeDPRRegisterClass(Inst, Rd, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+2*inc, Address, Decoder));
  CHECK(S, DecodeDPRRegisterClass(Inst, Rd+3*inc, Address, Decoder));
  Inst.addOperand(MCOperand::CreateImm(index));

  return S;
}