/*
 * Copyright (C) 2011 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#ifndef ART_COMPILER_DEX_QUICK_X86_CODEGEN_X86_H_
#define ART_COMPILER_DEX_QUICK_X86_CODEGEN_X86_H_

#include "base/logging.h"
#include "dex/compiler_ir.h"
#include "dex/mir_graph.h"
#include "dex/quick/mir_to_lir.h"
#include "x86_lir.h"

#include <map>
#include <vector>

namespace art {

class X86Mir2Lir FINAL : public Mir2Lir {
 protected:
  class InToRegStorageX86_64Mapper : public InToRegStorageMapper {
   public:
    explicit InToRegStorageX86_64Mapper(Mir2Lir* m2l)
        : m2l_(m2l), cur_core_reg_(0), cur_fp_reg_(0) {}
    virtual RegStorage GetNextReg(ShortyArg arg);
    virtual void Reset() OVERRIDE {
      cur_core_reg_ = 0;
      cur_fp_reg_ = 0;
    }
   protected:
    Mir2Lir* m2l_;
    size_t cur_core_reg_;
    size_t cur_fp_reg_;
  };

  class InToRegStorageX86Mapper : public InToRegStorageX86_64Mapper {
   public:
    explicit InToRegStorageX86Mapper(Mir2Lir* m2l)
        : InToRegStorageX86_64Mapper(m2l) { }
    virtual RegStorage GetNextReg(ShortyArg arg);
  };

  InToRegStorageX86_64Mapper in_to_reg_storage_x86_64_mapper_;
  InToRegStorageX86Mapper in_to_reg_storage_x86_mapper_;
  InToRegStorageMapper* GetResetedInToRegStorageMapper() OVERRIDE {
    InToRegStorageMapper* res;
    if (cu_->target64) {
      res = &in_to_reg_storage_x86_64_mapper_;
    } else {
      res = &in_to_reg_storage_x86_mapper_;
    }
    res->Reset();
    return res;
  }

  class ExplicitTempRegisterLock {
  public:
    ExplicitTempRegisterLock(X86Mir2Lir* mir_to_lir, int n_regs, ...);
    ~ExplicitTempRegisterLock();
  protected:
    std::vector<RegStorage> temp_regs_;
    X86Mir2Lir* const mir_to_lir_;
  };

  virtual int GenDalvikArgsBulkCopy(CallInfo* info, int first, int count) OVERRIDE;

 public:
  X86Mir2Lir(CompilationUnit* cu, MIRGraph* mir_graph, ArenaAllocator* arena);

  // Required for target - codegen helpers.
  bool SmallLiteralDivRem(Instruction::Code dalvik_opcode, bool is_div, RegLocation rl_src,
                          RegLocation rl_dest, int lit) OVERRIDE;
  bool EasyMultiply(RegLocation rl_src, RegLocation rl_dest, int lit) OVERRIDE;
  void GenMultiplyByConstantFloat(RegLocation rl_dest, RegLocation rl_src1,
                                  int32_t constant) OVERRIDE;
  void GenMultiplyByConstantDouble(RegLocation rl_dest, RegLocation rl_src1,
                                   int64_t constant) OVERRIDE;
  LIR* CheckSuspendUsingLoad() OVERRIDE;
  RegStorage LoadHelper(QuickEntrypointEnum trampoline) OVERRIDE;
  LIR* LoadBaseDisp(RegStorage r_base, int displacement, RegStorage r_dest,
                    OpSize size, VolatileKind is_volatile) OVERRIDE;
  LIR* LoadBaseIndexed(RegStorage r_base, RegStorage r_index, RegStorage r_dest, int scale,
                       OpSize size) OVERRIDE;
  LIR* LoadConstantNoClobber(RegStorage r_dest, int value);
  LIR* LoadConstantWide(RegStorage r_dest, int64_t value);
  void GenLongToInt(RegLocation rl_dest, RegLocation rl_src);
  LIR* StoreBaseDisp(RegStorage r_base, int displacement, RegStorage r_src,
                     OpSize size, VolatileKind is_volatile) OVERRIDE;
  LIR* StoreBaseIndexed(RegStorage r_base, RegStorage r_index, RegStorage r_src, int scale,
                        OpSize size) OVERRIDE;

  /// @copydoc Mir2Lir::UnconditionallyMarkGCCard(RegStorage)
  void UnconditionallyMarkGCCard(RegStorage tgt_addr_reg) OVERRIDE;

  bool CanUseOpPcRelDexCacheArrayLoad() const OVERRIDE;
  void OpPcRelDexCacheArrayLoad(const DexFile* dex_file, int offset, RegStorage r_dest, bool wide)
      OVERRIDE;

  void GenImplicitNullCheck(RegStorage reg, int opt_flags) OVERRIDE;

  // Required for target - register utilities.
  RegStorage TargetReg(SpecialTargetRegister reg) OVERRIDE;
  RegStorage TargetReg(SpecialTargetRegister symbolic_reg, WideKind wide_kind) OVERRIDE {
    if (wide_kind == kWide) {
      if (cu_->target64) {
        return As64BitReg(TargetReg32(symbolic_reg));
      } else {
        if (symbolic_reg >= kFArg0 && symbolic_reg <= kFArg3) {
          // We want an XMM, not a pair.
          return As64BitReg(TargetReg32(symbolic_reg));
        }
        // x86: construct a pair.
        DCHECK((kArg0 <= symbolic_reg && symbolic_reg < kArg3) ||
               (kRet0 == symbolic_reg));
        return RegStorage::MakeRegPair(TargetReg32(symbolic_reg),
                                 TargetReg32(static_cast<SpecialTargetRegister>(symbolic_reg + 1)));
      }
    } else if (wide_kind == kRef && cu_->target64) {
      return As64BitReg(TargetReg32(symbolic_reg));
    } else {
      return TargetReg32(symbolic_reg);
    }
  }
  RegStorage TargetPtrReg(SpecialTargetRegister symbolic_reg) OVERRIDE {
    return TargetReg(symbolic_reg, cu_->target64 ? kWide : kNotWide);
  }

  RegLocation GetReturnAlt() OVERRIDE;
  RegLocation GetReturnWideAlt() OVERRIDE;
  RegLocation LocCReturn() OVERRIDE;
  RegLocation LocCReturnRef() OVERRIDE;
  RegLocation LocCReturnDouble() OVERRIDE;
  RegLocation LocCReturnFloat() OVERRIDE;
  RegLocation LocCReturnWide() OVERRIDE;

  ResourceMask GetRegMaskCommon(const RegStorage& reg) const OVERRIDE;
  void AdjustSpillMask() OVERRIDE;
  void ClobberCallerSave() OVERRIDE;
  void FreeCallTemps() OVERRIDE;
  void LockCallTemps() OVERRIDE;

  void CompilerInitializeRegAlloc() OVERRIDE;
  int VectorRegisterSize() OVERRIDE;
  int NumReservableVectorRegisters(bool long_or_fp) OVERRIDE;

  // Required for target - miscellaneous.
  void AssembleLIR() OVERRIDE;
  void DumpResourceMask(LIR* lir, const ResourceMask& mask, const char* prefix) OVERRIDE;
  void SetupTargetResourceMasks(LIR* lir, uint64_t flags,
                                ResourceMask* use_mask, ResourceMask* def_mask) OVERRIDE;
  const char* GetTargetInstFmt(int opcode) OVERRIDE;
  const char* GetTargetInstName(int opcode) OVERRIDE;
  std::string BuildInsnString(const char* fmt, LIR* lir, unsigned char* base_addr) OVERRIDE;
  ResourceMask GetPCUseDefEncoding() const OVERRIDE;
  uint64_t GetTargetInstFlags(int opcode) OVERRIDE;
  size_t GetInsnSize(LIR* lir) OVERRIDE;
  bool IsUnconditionalBranch(LIR* lir) OVERRIDE;

  // Get the register class for load/store of a field.
  RegisterClass RegClassForFieldLoadStore(OpSize size, bool is_volatile) OVERRIDE;

  // Required for target - Dalvik-level generators.
  void GenArrayGet(int opt_flags, OpSize size, RegLocation rl_array, RegLocation rl_index,
                   RegLocation rl_dest, int scale) OVERRIDE;
  void GenArrayPut(int opt_flags, OpSize size, RegLocation rl_array,
                   RegLocation rl_index, RegLocation rl_src, int scale, bool card_mark) OVERRIDE;

  void GenArithOpDouble(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                        RegLocation rl_src2) OVERRIDE;
  void GenArithOpFloat(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                       RegLocation rl_src2) OVERRIDE;
  void GenCmpFP(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                RegLocation rl_src2) OVERRIDE;
  void GenConversion(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src) OVERRIDE;

  bool GenInlinedCas(CallInfo* info, bool is_long, bool is_object) OVERRIDE;
  bool GenInlinedMinMax(CallInfo* info, bool is_min, bool is_long) OVERRIDE;
  bool GenInlinedMinMaxFP(CallInfo* info, bool is_min, bool is_double) OVERRIDE;
  bool GenInlinedReverseBits(CallInfo* info, OpSize size) OVERRIDE;
  bool GenInlinedSqrt(CallInfo* info) OVERRIDE;
  bool GenInlinedAbsFloat(CallInfo* info) OVERRIDE;
  bool GenInlinedAbsDouble(CallInfo* info) OVERRIDE;
  bool GenInlinedPeek(CallInfo* info, OpSize size) OVERRIDE;
  bool GenInlinedPoke(CallInfo* info, OpSize size) OVERRIDE;
  bool GenInlinedCharAt(CallInfo* info) OVERRIDE;

  // Long instructions.
  void GenArithOpLong(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                      RegLocation rl_src2, int flags) OVERRIDE;
  void GenArithImmOpLong(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                         RegLocation rl_src2, int flags) OVERRIDE;
  void GenShiftImmOpLong(Instruction::Code opcode, RegLocation rl_dest,
                         RegLocation rl_src1, RegLocation rl_shift, int flags) OVERRIDE;
  void GenCmpLong(RegLocation rl_dest, RegLocation rl_src1, RegLocation rl_src2) OVERRIDE;
  void GenIntToLong(RegLocation rl_dest, RegLocation rl_src) OVERRIDE;
  void GenShiftOpLong(Instruction::Code opcode, RegLocation rl_dest,
                      RegLocation rl_src1, RegLocation rl_shift) OVERRIDE;

  /*
   * @brief Generate a two address long operation with a constant value
   * @param rl_dest location of result
   * @param rl_src constant source operand
   * @param op Opcode to be generated
   * @return success or not
   */
  bool GenLongImm(RegLocation rl_dest, RegLocation rl_src, Instruction::Code op);

  /*
   * @brief Generate a three address long operation with a constant value
   * @param rl_dest location of result
   * @param rl_src1 source operand
   * @param rl_src2 constant source operand
   * @param op Opcode to be generated
   * @return success or not
   */
  bool GenLongLongImm(RegLocation rl_dest, RegLocation rl_src1, RegLocation rl_src2,
                      Instruction::Code op);
  /**
   * @brief Generate a long arithmetic operation.
   * @param rl_dest The destination.
   * @param rl_src1 First operand.
   * @param rl_src2 Second operand.
   * @param op The DEX opcode for the operation.
   * @param is_commutative The sources can be swapped if needed.
   */
  virtual void GenLongArith(RegLocation rl_dest, RegLocation rl_src1, RegLocation rl_src2,
                            Instruction::Code op, bool is_commutative);

  /**
   * @brief Generate a two operand long arithmetic operation.
   * @param rl_dest The destination.
   * @param rl_src Second operand.
   * @param op The DEX opcode for the operation.
   */
  void GenLongArith(RegLocation rl_dest, RegLocation rl_src, Instruction::Code op);

  /**
   * @brief Generate a long operation.
   * @param rl_dest The destination.  Must be in a register
   * @param rl_src The other operand.  May be in a register or in memory.
   * @param op The DEX opcode for the operation.
   */
  virtual void GenLongRegOrMemOp(RegLocation rl_dest, RegLocation rl_src, Instruction::Code op);


  // TODO: collapse reg_lo, reg_hi
  RegLocation GenDivRem(RegLocation rl_dest, RegStorage reg_lo, RegStorage reg_hi, bool is_div)
      OVERRIDE;
  RegLocation GenDivRemLit(RegLocation rl_dest, RegStorage reg_lo, int lit, bool is_div) OVERRIDE;
  void GenDivZeroCheckWide(RegStorage reg) OVERRIDE;
  void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method) OVERRIDE;
  void GenExitSequence() OVERRIDE;
  void GenSpecialExitSequence() OVERRIDE;
  void GenSpecialEntryForSuspend() OVERRIDE;
  void GenSpecialExitForSuspend() OVERRIDE;
  void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double) OVERRIDE;
  void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir) OVERRIDE;
  void GenSelect(BasicBlock* bb, MIR* mir) OVERRIDE;
  void GenSelectConst32(RegStorage left_op, RegStorage right_op, ConditionCode code,
                        int32_t true_val, int32_t false_val, RegStorage rs_dest,
                        RegisterClass dest_reg_class) OVERRIDE;
  bool GenMemBarrier(MemBarrierKind barrier_kind) OVERRIDE;
  void GenMoveException(RegLocation rl_dest) OVERRIDE;
  void GenMultiplyByTwoBitMultiplier(RegLocation rl_src, RegLocation rl_result, int lit,
                                     int first_bit, int second_bit) OVERRIDE;
  void GenNegDouble(RegLocation rl_dest, RegLocation rl_src) OVERRIDE;
  void GenNegFloat(RegLocation rl_dest, RegLocation rl_src) OVERRIDE;
  void GenLargePackedSwitch(MIR* mir, DexOffset table_offset, RegLocation rl_src) OVERRIDE;
  void GenLargeSparseSwitch(MIR* mir, DexOffset table_offset, RegLocation rl_src) OVERRIDE;

  /**
   * @brief Implement instanceof a final class with x86 specific code.
   * @param use_declaring_class 'true' if we can use the class itself.
   * @param type_idx Type index to use if use_declaring_class is 'false'.
   * @param rl_dest Result to be set to 0 or 1.
   * @param rl_src Object to be tested.
   */
  void GenInstanceofFinal(bool use_declaring_class, uint32_t type_idx, RegLocation rl_dest,
                          RegLocation rl_src) OVERRIDE;

  // Single operation generators.
  LIR* OpUnconditionalBranch(LIR* target) OVERRIDE;
  LIR* OpCmpBranch(ConditionCode cond, RegStorage src1, RegStorage src2, LIR* target) OVERRIDE;
  LIR* OpCmpImmBranch(ConditionCode cond, RegStorage reg, int check_value, LIR* target) OVERRIDE;
  LIR* OpCondBranch(ConditionCode cc, LIR* target) OVERRIDE;
  LIR* OpDecAndBranch(ConditionCode c_code, RegStorage reg, LIR* target) OVERRIDE;
  LIR* OpFpRegCopy(RegStorage r_dest, RegStorage r_src) OVERRIDE;
  LIR* OpIT(ConditionCode cond, const char* guide) OVERRIDE;
  void OpEndIT(LIR* it) OVERRIDE;
  LIR* OpMem(OpKind op, RegStorage r_base, int disp) OVERRIDE;
  void OpPcRelLoad(RegStorage reg, LIR* target) OVERRIDE;
  LIR* OpReg(OpKind op, RegStorage r_dest_src) OVERRIDE;
  void OpRegCopy(RegStorage r_dest, RegStorage r_src) OVERRIDE;
  LIR* OpRegCopyNoInsert(RegStorage r_dest, RegStorage r_src) OVERRIDE;
  LIR* OpRegImm(OpKind op, RegStorage r_dest_src1, int value) OVERRIDE;
  LIR* OpRegReg(OpKind op, RegStorage r_dest_src1, RegStorage r_src2) OVERRIDE;
  LIR* OpMovRegMem(RegStorage r_dest, RegStorage r_base, int offset, MoveType move_type) OVERRIDE;
  LIR* OpMovMemReg(RegStorage r_base, int offset, RegStorage r_src, MoveType move_type) OVERRIDE;
  LIR* OpCondRegReg(OpKind op, ConditionCode cc, RegStorage r_dest, RegStorage r_src) OVERRIDE;
  LIR* OpRegRegImm(OpKind op, RegStorage r_dest, RegStorage r_src1, int value) OVERRIDE;
  LIR* OpRegRegReg(OpKind op, RegStorage r_dest, RegStorage r_src1, RegStorage r_src2) OVERRIDE;
  LIR* OpTestSuspend(LIR* target) OVERRIDE;
  LIR* OpVldm(RegStorage r_base, int count) OVERRIDE;
  LIR* OpVstm(RegStorage r_base, int count) OVERRIDE;
  void OpRegCopyWide(RegStorage dest, RegStorage src) OVERRIDE;
  bool GenInlinedCurrentThread(CallInfo* info) OVERRIDE;

  bool InexpensiveConstantInt(int32_t value) OVERRIDE;
  bool InexpensiveConstantFloat(int32_t value) OVERRIDE;
  bool InexpensiveConstantLong(int64_t value) OVERRIDE;
  bool InexpensiveConstantDouble(int64_t value) OVERRIDE;

  /*
   * @brief Should try to optimize for two address instructions?
   * @return true if we try to avoid generating three operand instructions.
   */
  virtual bool GenerateTwoOperandInstructions() const { return true; }

  /*
   * @brief x86 specific codegen for int operations.
   * @param opcode Operation to perform.
   * @param rl_dest Destination for the result.
   * @param rl_lhs Left hand operand.
   * @param rl_rhs Right hand operand.
   * @param flags The instruction optimization flags.
   */
  void GenArithOpInt(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_lhs,
                     RegLocation rl_rhs, int flags) OVERRIDE;

  /*
   * @brief Load the Method* of a dex method into the register.
   * @param target_method The MethodReference of the method to be invoked.
   * @param type How the method will be invoked.
   * @param register that will contain the code address.
   * @note register will be passed to TargetReg to get physical register.
   */
  void LoadMethodAddress(const MethodReference& target_method, InvokeType type,
                         SpecialTargetRegister symbolic_reg) OVERRIDE;

  /*
   * @brief Load the Class* of a Dex Class type into the register.
   * @param dex DexFile that contains the class type.
   * @param type How the method will be invoked.
   * @param register that will contain the code address.
   * @note register will be passed to TargetReg to get physical register.
   */
  void LoadClassType(const DexFile& dex_file, uint32_t type_idx,
                     SpecialTargetRegister symbolic_reg) OVERRIDE;

  NextCallInsn GetNextSDCallInsn() OVERRIDE;

  /*
   * @brief Generate a relative call to the method that will be patched at link time.
   * @param target_method The MethodReference of the method to be invoked.
   * @param type How the method will be invoked.
   * @returns Call instruction
   */
  LIR* CallWithLinkerFixup(const MethodReference& target_method, InvokeType type);

  /*
   * @brief Generate the actual call insn based on the method info.
   * @param method_info the lowering info for the method call.
   * @returns Call instruction
   */
  LIR* GenCallInsn(const MirMethodLoweringInfo& method_info) OVERRIDE;

  void AnalyzeMIR(RefCounts* core_counts, MIR* mir, uint32_t weight) OVERRIDE;
  void CountRefs(RefCounts* core_counts, RefCounts* fp_counts, size_t num_regs) OVERRIDE;
  void DoPromotion() OVERRIDE;

  /*
   * @brief Handle x86 specific literals
   */
  void InstallLiteralPools() OVERRIDE;

  LIR* InvokeTrampoline(OpKind op, RegStorage r_tgt, QuickEntrypointEnum trampoline) OVERRIDE;

 protected:
  RegStorage TargetReg32(SpecialTargetRegister reg) const;
  // Casting of RegStorage
  RegStorage As32BitReg(RegStorage reg) {
    DCHECK(!reg.IsPair());
    if ((kFailOnSizeError || kReportSizeError) && !reg.Is64Bit()) {
      if (kFailOnSizeError) {
        LOG(FATAL) << "Expected 64b register " << reg.GetReg();
      } else {
        LOG(WARNING) << "Expected 64b register " << reg.GetReg();
        return reg;
      }
    }
    RegStorage ret_val = RegStorage(RegStorage::k32BitSolo,
                                    reg.GetRawBits() & RegStorage::kRegTypeMask);
    DCHECK_EQ(GetRegInfo(reg)->FindMatchingView(RegisterInfo::k32SoloStorageMask)
                             ->GetReg().GetReg(),
              ret_val.GetReg());
    return ret_val;
  }

  RegStorage As64BitReg(RegStorage reg) {
    DCHECK(!reg.IsPair());
    if ((kFailOnSizeError || kReportSizeError) && !reg.Is32Bit()) {
      if (kFailOnSizeError) {
        LOG(FATAL) << "Expected 32b register " << reg.GetReg();
      } else {
        LOG(WARNING) << "Expected 32b register " << reg.GetReg();
        return reg;
      }
    }
    RegStorage ret_val = RegStorage(RegStorage::k64BitSolo,
                                    reg.GetRawBits() & RegStorage::kRegTypeMask);
    DCHECK_EQ(GetRegInfo(reg)->FindMatchingView(RegisterInfo::k64SoloStorageMask)
                             ->GetReg().GetReg(),
              ret_val.GetReg());
    return ret_val;
  }

  LIR* LoadBaseIndexedDisp(RegStorage r_base, RegStorage r_index, int scale, int displacement,
                           RegStorage r_dest, OpSize size);
  LIR* StoreBaseIndexedDisp(RegStorage r_base, RegStorage r_index, int scale, int displacement,
                            RegStorage r_src, OpSize size, int opt_flags = 0);

  int AssignInsnOffsets();
  void AssignOffsets();
  AssemblerStatus AssembleInstructions(LIR* first_lir_insn, CodeOffset start_addr);

  size_t ComputeSize(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_index,
                     int32_t raw_base, int32_t displacement);
  void CheckValidByteRegister(const X86EncodingMap* entry, int32_t raw_reg);
  void EmitPrefix(const X86EncodingMap* entry,
                  int32_t raw_reg_r, int32_t raw_reg_x, int32_t raw_reg_b);
  void EmitOpcode(const X86EncodingMap* entry);
  void EmitPrefixAndOpcode(const X86EncodingMap* entry,
                           int32_t reg_r, int32_t reg_x, int32_t reg_b);
  void EmitDisp(uint8_t base, int32_t disp);
  void EmitModrmThread(uint8_t reg_or_opcode);
  void EmitModrmDisp(uint8_t reg_or_opcode, uint8_t base, int32_t disp);
  void EmitModrmSibDisp(uint8_t reg_or_opcode, uint8_t base, uint8_t index, int scale,
                        int32_t disp);
  void EmitImm(const X86EncodingMap* entry, int64_t imm);
  void EmitNullary(const X86EncodingMap* entry);
  void EmitOpRegOpcode(const X86EncodingMap* entry, int32_t raw_reg);
  void EmitOpReg(const X86EncodingMap* entry, int32_t raw_reg);
  void EmitOpMem(const X86EncodingMap* entry, int32_t raw_base, int32_t disp);
  void EmitOpArray(const X86EncodingMap* entry, int32_t raw_base, int32_t raw_index, int scale,
                   int32_t disp);
  void EmitMemReg(const X86EncodingMap* entry, int32_t raw_base, int32_t disp, int32_t raw_reg);
  void EmitRegMem(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_base, int32_t disp);
  void EmitRegArray(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_base,
                    int32_t raw_index, int scale, int32_t disp);
  void EmitArrayReg(const X86EncodingMap* entry, int32_t raw_base, int32_t raw_index, int scale,
                    int32_t disp, int32_t raw_reg);
  void EmitMemImm(const X86EncodingMap* entry, int32_t raw_base, int32_t disp, int32_t imm);
  void EmitArrayImm(const X86EncodingMap* entry, int32_t raw_base, int32_t raw_index, int scale,
                    int32_t raw_disp, int32_t imm);
  void EmitRegThread(const X86EncodingMap* entry, int32_t raw_reg, int32_t disp);
  void EmitRegReg(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_reg2);
  void EmitRegRegImm(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_reg2, int32_t imm);
  void EmitRegMemImm(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_base, int32_t disp,
                     int32_t imm);
  void EmitMemRegImm(const X86EncodingMap* entry, int32_t base, int32_t disp, int32_t raw_reg1,
                     int32_t imm);
  void EmitRegImm(const X86EncodingMap* entry, int32_t raw_reg, int32_t imm);
  void EmitThreadImm(const X86EncodingMap* entry, int32_t disp, int32_t imm);
  void EmitMovRegImm(const X86EncodingMap* entry, int32_t raw_reg, int64_t imm);
  void EmitShiftRegImm(const X86EncodingMap* entry, int32_t raw_reg, int32_t imm);
  void EmitShiftRegCl(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_cl);
  void EmitShiftMemCl(const X86EncodingMap* entry, int32_t raw_base, int32_t disp, int32_t raw_cl);
  void EmitShiftRegRegCl(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_reg2,
                         int32_t raw_cl);
  void EmitShiftMemImm(const X86EncodingMap* entry, int32_t raw_base, int32_t disp, int32_t imm);
  void EmitRegCond(const X86EncodingMap* entry, int32_t raw_reg, int32_t cc);
  void EmitMemCond(const X86EncodingMap* entry, int32_t raw_base, int32_t disp, int32_t cc);
  void EmitRegRegCond(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_reg2, int32_t cc);
  void EmitRegMemCond(const X86EncodingMap* entry, int32_t raw_reg1, int32_t raw_base, int32_t disp,
                      int32_t cc);

  void EmitJmp(const X86EncodingMap* entry, int32_t rel);
  void EmitJcc(const X86EncodingMap* entry, int32_t rel, int32_t cc);
  void EmitCallMem(const X86EncodingMap* entry, int32_t raw_base, int32_t disp);
  void EmitCallImmediate(const X86EncodingMap* entry, int32_t disp);
  void EmitCallThread(const X86EncodingMap* entry, int32_t disp);
  void EmitPcRel(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_base_or_table,
                 int32_t raw_index, int scale, int32_t table_or_disp);
  void EmitUnimplemented(const X86EncodingMap* entry, LIR* lir);
  void GenFusedLongCmpImmBranch(BasicBlock* bb, RegLocation rl_src1,
                                int64_t val, ConditionCode ccode);
  void GenConstWide(RegLocation rl_dest, int64_t value);
  void GenMultiplyVectorSignedByte(RegStorage rs_dest_src1, RegStorage rs_src2);
  void GenMultiplyVectorLong(RegStorage rs_dest_src1, RegStorage rs_src2);
  void GenShiftByteVector(MIR* mir);
  void AndMaskVectorRegister(RegStorage rs_src1, uint32_t m1, uint32_t m2, uint32_t m3,
                             uint32_t m4);
  void MaskVectorRegister(X86OpCode opcode, RegStorage rs_src1, uint32_t m1, uint32_t m2,
                          uint32_t m3, uint32_t m4);
  void AppendOpcodeWithConst(X86OpCode opcode, int reg, MIR* mir);
  virtual void LoadVectorRegister(RegStorage rs_dest, RegStorage rs_src, OpSize opsize,
                                  int op_mov);

  static bool ProvidesFullMemoryBarrier(X86OpCode opcode);

  /*
   * @brief Ensure that a temporary register is byte addressable.
   * @returns a temporary guarenteed to be byte addressable.
   */
  virtual RegStorage AllocateByteRegister();

  /*
   * @brief Use a wide temporary as a 128-bit register
   * @returns a 128-bit temporary register.
   */
  virtual RegStorage Get128BitRegister(RegStorage reg);

  /*
   * @brief Check if a register is byte addressable.
   * @returns true if a register is byte addressable.
   */
  bool IsByteRegister(RegStorage reg) const;

  void GenDivRemLongLit(RegLocation rl_dest, RegLocation rl_src, int64_t imm, bool is_div);

  bool GenInlinedArrayCopyCharArray(CallInfo* info) OVERRIDE;

  /*
   * @brief generate inline code for fast case of Strng.indexOf.
   * @param info Call parameters
   * @param zero_based 'true' if the index into the string is 0.
   * @returns 'true' if the call was inlined, 'false' if a regular call needs to be
   * generated.
   */
  bool GenInlinedIndexOf(CallInfo* info, bool zero_based);

  /**
   * @brief Used to reserve a range of vector registers.
   * @see kMirOpReserveVectorRegisters
   * @param mir The extended MIR for reservation.
   */
  void ReserveVectorRegisters(MIR* mir);

  /**
   * @brief Used to return a range of vector registers.
   * @see kMirOpReturnVectorRegisters
   * @param mir The extended MIR for returning vector regs.
   */
  void ReturnVectorRegisters(MIR* mir);

  /*
   * @brief Load 128 bit constant into vector register.
   * @param mir The MIR whose opcode is kMirConstVector
   * @note vA is the TypeSize for the register.
   * @note vB is the destination XMM register. arg[0..3] are 32 bit constant values.
   */
  void GenConst128(MIR* mir);

  /*
   * @brief MIR to move a vectorized register to another.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination
   * @note vC: source
   */
  void GenMoveVector(MIR* mir);

  /*
   * @brief Packed multiply of units in two vector registers: vB = vB .* @note vC using vA to know
   * the type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenMultiplyVector(MIR* mir);

  /*
   * @brief Packed addition of units in two vector registers: vB = vB .+ vC using vA to know the
   * type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenAddVector(MIR* mir);

  /*
   * @brief Packed subtraction of units in two vector registers: vB = vB .- vC using vA to know the
   * type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenSubtractVector(MIR* mir);

  /*
   * @brief Packed shift left of units in two vector registers: vB = vB .<< vC using vA to know the
   * type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: immediate
   */
  void GenShiftLeftVector(MIR* mir);

  /*
   * @brief Packed signed shift right of units in two vector registers: vB = vB .>> vC using vA to
   * know the type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: immediate
   */
  void GenSignedShiftRightVector(MIR* mir);

  /*
   * @brief Packed unsigned shift right of units in two vector registers: vB = vB .>>> vC using vA
   * to know the type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: immediate
   */
  void GenUnsignedShiftRightVector(MIR* mir);

  /*
   * @brief Packed bitwise and of units in two vector registers: vB = vB .& vC using vA to know the
   * type of the vector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenAndVector(MIR* mir);

  /*
   * @brief Packed bitwise or of units in two vector registers: vB = vB .| vC using vA to know the
   * type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenOrVector(MIR* mir);

  /*
   * @brief Packed bitwise xor of units in two vector registers: vB = vB .^ vC using vA to know the
   * type of the vector.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination and source
   * @note vC: source
   */
  void GenXorVector(MIR* mir);

  /*
   * @brief Reduce a 128-bit packed element into a single VR by taking lower bits
   * @param mir The MIR whose opcode is kMirConstVector.
   * @details Instruction does a horizontal addition of the packed elements and then adds it to VR.
   * @note vA: TypeSize
   * @note vB: destination and source VR (not vector register)
   * @note vC: source (vector register)
   */
  void GenAddReduceVector(MIR* mir);

  /*
   * @brief Extract a packed element into a single VR.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize
   * @note vB: destination VR (not vector register)
   * @note vC: source (vector register)
   * @note arg[0]: The index to use for extraction from vector register (which packed element).
   */
  void GenReduceVector(MIR* mir);

  /*
   * @brief Create a vector value, with all TypeSize values equal to vC
   * @param bb The basic block in which the MIR is from.
   * @param mir The MIR whose opcode is kMirConstVector.
   * @note vA: TypeSize.
   * @note vB: destination vector register.
   * @note vC: source VR (not vector register).
   */
  void GenSetVector(MIR* mir);

  /**
   * @brief Used to generate code for kMirOpPackedArrayGet.
   * @param bb The basic block of MIR.
   * @param mir The mir whose opcode is kMirOpPackedArrayGet.
   */
  void GenPackedArrayGet(BasicBlock* bb, MIR* mir);

  /**
   * @brief Used to generate code for kMirOpPackedArrayPut.
   * @param bb The basic block of MIR.
   * @param mir The mir whose opcode is kMirOpPackedArrayPut.
   */
  void GenPackedArrayPut(BasicBlock* bb, MIR* mir);

  /*
   * @brief Generate code for a vector opcode.
   * @param bb The basic block in which the MIR is from.
   * @param mir The MIR whose opcode is a non-standard opcode.
   */
  void GenMachineSpecificExtendedMethodMIR(BasicBlock* bb, MIR* mir);

  /*
   * @brief Return the correct x86 opcode for the Dex operation
   * @param op Dex opcode for the operation
   * @param loc Register location of the operand
   * @param is_high_op 'true' if this is an operation on the high word
   * @param value Immediate value for the operation.  Used for byte variants
   * @returns the correct x86 opcode to perform the operation
   */
  X86OpCode GetOpcode(Instruction::Code op, RegLocation loc, bool is_high_op, int32_t value);

  /*
   * @brief Return the correct x86 opcode for the Dex operation
   * @param op Dex opcode for the operation
   * @param dest location of the destination.  May be register or memory.
   * @param rhs Location for the rhs of the operation.  May be in register or memory.
   * @param is_high_op 'true' if this is an operation on the high word
   * @returns the correct x86 opcode to perform the operation
   * @note at most one location may refer to memory
   */
  X86OpCode GetOpcode(Instruction::Code op, RegLocation dest, RegLocation rhs,
                      bool is_high_op);

  /*
   * @brief Is this operation a no-op for this opcode and value
   * @param op Dex opcode for the operation
   * @param value Immediate value for the operation.
   * @returns 'true' if the operation will have no effect
   */
  bool IsNoOp(Instruction::Code op, int32_t value);

  /**
   * @brief Calculate magic number and shift for a given divisor
   * @param divisor divisor number for calculation
   * @param magic hold calculated magic number
   * @param shift hold calculated shift
   * @param is_long 'true' if divisor is jlong, 'false' for jint.
   */
  void CalculateMagicAndShift(int64_t divisor, int64_t& magic, int& shift, bool is_long);

  /*
   * @brief Generate an integer div or rem operation.
   * @param rl_dest Destination Location.
   * @param rl_src1 Numerator Location.
   * @param rl_src2 Divisor Location.
   * @param is_div 'true' if this is a division, 'false' for a remainder.
   * @param flags The instruction optimization flags. It can include information
   * if exception check can be elided.
   */
  RegLocation GenDivRem(RegLocation rl_dest, RegLocation rl_src1, RegLocation rl_src2,
                        bool is_div, int flags);

  /*
   * @brief Generate an integer div or rem operation by a literal.
   * @param rl_dest Destination Location.
   * @param rl_src Numerator Location.
   * @param lit Divisor.
   * @param is_div 'true' if this is a division, 'false' for a remainder.
   */
  RegLocation GenDivRemLit(RegLocation rl_dest, RegLocation rl_src, int lit, bool is_div);

  /*
   * Generate code to implement long shift operations.
   * @param opcode The DEX opcode to specify the shift type.
   * @param rl_dest The destination.
   * @param rl_src The value to be shifted.
   * @param shift_amount How much to shift.
   * @param flags The instruction optimization flags.
   * @returns the RegLocation of the result.
   */
  RegLocation GenShiftImmOpLong(Instruction::Code opcode, RegLocation rl_dest,
                                RegLocation rl_src, int shift_amount, int flags);
  /*
   * Generate an imul of a register by a constant or a better sequence.
   * @param dest Destination Register.
   * @param src Source Register.
   * @param val Constant multiplier.
   */
  void GenImulRegImm(RegStorage dest, RegStorage src, int val);

  /*
   * Generate an imul of a memory location by a constant or a better sequence.
   * @param dest Destination Register.
   * @param sreg Symbolic register.
   * @param displacement Displacement on stack of Symbolic Register.
   * @param val Constant multiplier.
   */
  void GenImulMemImm(RegStorage dest, int sreg, int displacement, int val);

  /*
   * @brief Compare memory to immediate, and branch if condition true.
   * @param cond The condition code that when true will branch to the target.
   * @param temp_reg A temporary register that can be used if compare memory is not
   * supported by the architecture.
   * @param base_reg The register holding the base address.
   * @param offset The offset from the base.
   * @param check_value The immediate to compare to.
   * @param target branch target (or nullptr)
   * @param compare output for getting LIR for comparison (or nullptr)
   */
  LIR* OpCmpMemImmBranch(ConditionCode cond, RegStorage temp_reg, RegStorage base_reg,
                         int offset, int check_value, LIR* target, LIR** compare);

  void GenRemFP(RegLocation rl_dest, RegLocation rl_src1, RegLocation rl_src2, bool is_double);

  /*
   * Can this operation be using core registers without temporaries?
   * @param rl_lhs Left hand operand.
   * @param rl_rhs Right hand operand.
   * @returns 'true' if the operation can proceed without needing temporary regs.
   */
  bool IsOperationSafeWithoutTemps(RegLocation rl_lhs, RegLocation rl_rhs);

  /**
   * @brief Generates inline code for conversion of long to FP by using x87/
   * @param rl_dest The destination of the FP.
   * @param rl_src The source of the long.
   * @param is_double 'true' if dealing with double, 'false' for float.
   */
  virtual void GenLongToFP(RegLocation rl_dest, RegLocation rl_src, bool is_double);

  void GenArrayBoundsCheck(RegStorage index, RegStorage array_base, int32_t len_offset);
  void GenArrayBoundsCheck(int32_t index, RegStorage array_base, int32_t len_offset);

  LIR* OpRegMem(OpKind op, RegStorage r_dest, RegStorage r_base, int offset);
  LIR* OpRegMem(OpKind op, RegStorage r_dest, RegLocation value);
  LIR* OpMemReg(OpKind op, RegLocation rl_dest, int value);
  LIR* OpThreadMem(OpKind op, ThreadOffset<4> thread_offset);
  LIR* OpThreadMem(OpKind op, ThreadOffset<8> thread_offset);
  void OpRegThreadMem(OpKind op, RegStorage r_dest, ThreadOffset<4> thread_offset);
  void OpRegThreadMem(OpKind op, RegStorage r_dest, ThreadOffset<8> thread_offset);
  void OpTlsCmp(ThreadOffset<4> offset, int val);
  void OpTlsCmp(ThreadOffset<8> offset, int val);

  void OpLea(RegStorage r_base, RegStorage reg1, RegStorage reg2, int scale, int offset);

  // Try to do a long multiplication where rl_src2 is a constant. This simplified setup might fail,
  // in which case false will be returned.
  bool GenMulLongConst(RegLocation rl_dest, RegLocation rl_src1, int64_t val, int flags);
  void GenMulLong(Instruction::Code opcode, RegLocation rl_dest, RegLocation rl_src1,
                  RegLocation rl_src2, int flags);
  void GenNotLong(RegLocation rl_dest, RegLocation rl_src);
  void GenNegLong(RegLocation rl_dest, RegLocation rl_src);
  void GenDivRemLong(Instruction::Code, RegLocation rl_dest, RegLocation rl_src1,
                     RegLocation rl_src2, bool is_div, int flags);

  void SpillCoreRegs();
  void UnSpillCoreRegs();
  void UnSpillFPRegs();
  void SpillFPRegs();

  /*
   * Mir2Lir's UpdateLoc() looks to see if the Dalvik value is currently live in any temp register
   * without regard to data type.  In practice, this can result in UpdateLoc returning a
   * location record for a Dalvik float value in a core register, and vis-versa.  For targets
   * which can inexpensively move data between core and float registers, this can often be a win.
   * However, for x86 this is generally not a win.  These variants of UpdateLoc()
   * take a register class argument - and will return an in-register location record only if
   * the value is live in a temp register of the correct class.  Additionally, if the value is in
   * a temp register of the wrong register class, it will be clobbered.
   */
  RegLocation UpdateLocTyped(RegLocation loc);
  RegLocation UpdateLocWideTyped(RegLocation loc);

  /*
   * @brief Analyze one MIR float/double instruction
   * @param opcode MIR instruction opcode.
   * @param mir Instruction to analyze.
   * @return true iff the instruction needs to load a literal using PC-relative addressing.
   */
  bool AnalyzeFPInstruction(int opcode, MIR* mir);

  /*
   * @brief Analyze one use of a double operand.
   * @param rl_use Double RegLocation for the operand.
   * @return true iff the instruction needs to load a literal using PC-relative addressing.
   */
  bool AnalyzeDoubleUse(RegLocation rl_use);

  /*
   * @brief Analyze one invoke-static MIR instruction
   * @param mir Instruction to analyze.
   * @return true iff the instruction needs to load a literal using PC-relative addressing.
   */
  bool AnalyzeInvokeStaticIntrinsic(MIR* mir);

  // Information derived from analysis of MIR

  // The base register for PC-relative addressing if promoted (32-bit only).
  RegStorage pc_rel_base_reg_;

  // Have we actually used the pc_rel_base_reg_?
  bool pc_rel_base_reg_used_;

  // Pointer to the "call +0" insn that sets up the promoted register for PC-relative addressing.
  // The anchor "pop" insn is NEXT_LIR(setup_pc_rel_base_reg_). The whole "call +0; pop <reg>"
  // sequence will be removed in AssembleLIR() if we do not actually use PC-relative addressing.
  LIR* setup_pc_rel_base_reg_;  // There are 2 chained insns (no reordering allowed).

  // Instructions needing patching with Method* values.
  ArenaVector<LIR*> method_address_insns_;

  // Instructions needing patching with Class Type* values.
  ArenaVector<LIR*> class_type_address_insns_;

  // Instructions needing patching with PC relative code addresses.
  ArenaVector<LIR*> call_method_insns_;

  // Instructions needing patching with PC relative code addresses.
  ArenaVector<LIR*> dex_cache_access_insns_;

  // The list of const vector literals.
  LIR* const_vectors_;

  /*
   * @brief Search for a matching vector literal
   * @param constants An array of size 4 which contains all of 32-bit constants.
   * @returns pointer to matching LIR constant, or nullptr if not found.
   */
  LIR* ScanVectorLiteral(int32_t* constants);

  /*
   * @brief Add a constant vector literal
   * @param constants An array of size 4 which contains all of 32-bit constants.
   */
  LIR* AddVectorLiteral(int32_t* constants);

  bool WideGPRsAreAliases() const OVERRIDE {
    return cu_->target64;  // On 64b, we have 64b GPRs.
  }

  bool WideFPRsAreAliases() const OVERRIDE {
    return true;  // xmm registers have 64b views even on x86.
  }

  /*
   * @brief Dump a RegLocation using printf
   * @param loc Register location to dump
   */
  static void DumpRegLocation(RegLocation loc);

 private:
  void SwapBits(RegStorage result_reg, int shift, int32_t value);
  void SwapBits64(RegStorage result_reg, int shift, int64_t value);

  static int X86NextSDCallInsn(CompilationUnit* cu, CallInfo* info,
                               int state, const MethodReference& target_method,
                               uint32_t,
                               uintptr_t direct_code, uintptr_t direct_method,
                               InvokeType type);

  LIR* OpLoadPc(RegStorage r_dest);
  RegStorage GetPcAndAnchor(LIR** anchor, RegStorage r_tmp = RegStorage::InvalidReg());

  // When we don't know the proper offset for the value, pick one that will force
  // 4 byte offset.  We will fix this up in the assembler or linker later to have
  // the right value.
  static constexpr int kDummy32BitOffset = 256;

  static const X86EncodingMap EncodingMap[kX86Last];

  friend std::ostream& operator<<(std::ostream& os, const X86OpCode& rhs);
  friend class QuickAssembleX86Test;
  friend class QuickAssembleX86MacroTest;
  friend class QuickAssembleX86LowLevelTest;

  DISALLOW_COPY_AND_ASSIGN(X86Mir2Lir);
};

}  // namespace art

#endif  // ART_COMPILER_DEX_QUICK_X86_CODEGEN_X86_H_