wasm-binary.h

/*
 * Copyright 2015 WebAssembly Community Group participants
 *
 * 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.
 */

//
// Parses and emits WebAssembly binary code
//

#ifndef wasm_wasm_binary_h
#define wasm_wasm_binary_h

#include <cassert>
#include <ostream>
#include <type_traits>

#include "ir/import-utils.h"
#include "ir/module-utils.h"
#include "parsing.h"
#include "source-map.h"
#include "wasm-builder.h"
#include "wasm-features.h"
#include "wasm-ir-builder.h"
#include "wasm-traversal.h"
#include "wasm-validator.h"
#include "wasm.h"

namespace wasm {

enum {
  // the maximum amount of bytes we emit per LEB
  MaxLEB32Bytes = 5,
};

template<typename T, typename MiniT> struct LEB {
  static_assert(sizeof(MiniT) == 1, "MiniT must be a byte");

  T value;

  LEB() = default;
  LEB(T value) : value(value) {}

  bool hasMore(T temp, MiniT byte) {
    // for signed, we must ensure the last bit has the right sign, as it will
    // zero extend
    return std::is_signed<T>::value
             ? (temp != 0 && temp != T(-1)) || (value >= 0 && (byte & 64)) ||
                 (value < 0 && !(byte & 64))
             : (temp != 0);
  }

  void write(std::vector<uint8_t>* out) {
    T temp = value;
    bool more;
    do {
      uint8_t byte = temp & 127;
      temp >>= 7;
      more = hasMore(temp, byte);
      if (more) {
        byte = byte | 128;
      }
      out->push_back(byte);
    } while (more);
  }

  // @minimum: a minimum number of bytes to write, padding as necessary
  // returns the number of bytes written
  size_t writeAt(std::vector<uint8_t>* out, size_t at, size_t minimum = 0) {
    T temp = value;
    size_t offset = 0;
    bool more;
    do {
      uint8_t byte = temp & 127;
      temp >>= 7;
      more = hasMore(temp, byte) || offset + 1 < minimum;
      if (more) {
        byte = byte | 128;
      }
      (*out)[at + offset] = byte;
      offset++;
    } while (more);
    return offset;
  }

  LEB<T, MiniT>& read(std::function<MiniT()> get) {
    value = 0;
    T shift = 0;
    MiniT byte;
    while (1) {
      byte = get();
      bool last = !(byte & 128);
      T payload = byte & 127;
      using mask_type = typename std::make_unsigned<T>::type;
      auto payload_mask = 0 == shift
                            ? ~mask_type(0)
                            : ((mask_type(1) << (sizeof(T) * 8 - shift)) - 1u);
      T significant_payload = payload_mask & payload;
      value |= significant_payload << shift;
      T unused_bits_mask = ~payload_mask & 127;
      T unused_bits = payload & unused_bits_mask;
      if (std::is_signed_v<T> && value < 0) {
        if (unused_bits != unused_bits_mask) {
          throw ParseException("Unused negative LEB bits must be 1s");
        }
      } else {
        if (unused_bits != 0) {
          throw ParseException("Unused non-negative LEB bits must be 0s");
        }
      }
      if (last) {
        break;
      }
      shift += 7;
      if (size_t(shift) >= sizeof(T) * 8) {
        throw ParseException("LEB overflow");
      }
    }
    // If signed LEB, then we might need to sign-extend.
    if constexpr (std::is_signed_v<T>) {
      shift += 7;
      if ((byte & 64) && size_t(shift) < 8 * sizeof(T)) {
        size_t sext_bits = 8 * sizeof(T) - size_t(shift);
        value <<= sext_bits;
        value >>= sext_bits;
        if (value >= 0) {
          throw ParseException(
            " LEBsign-extend should produce a negative value");
        }
      }
    }
    return *this;
  }
};

using U32LEB = LEB<uint32_t, uint8_t>;
using U64LEB = LEB<uint64_t, uint8_t>;
using S32LEB = LEB<int32_t, int8_t>;
using S64LEB = LEB<int64_t, int8_t>;

//
// We mostly stream into a buffer as we create the binary format, however,
// sometimes we need to backtrack and write to a location behind us - wasm
// is optimized for reading, not writing.
//
class BufferWithRandomAccess : public std::vector<uint8_t> {
public:
  BufferWithRandomAccess() = default;

  BufferWithRandomAccess& operator<<(int8_t x) {
    push_back(x);
    return *this;
  }
  BufferWithRandomAccess& operator<<(int16_t x) {
    push_back(x & 0xff);
    push_back(x >> 8);
    return *this;
  }
  BufferWithRandomAccess& operator<<(int32_t x) {
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    return *this;
  }
  BufferWithRandomAccess& operator<<(int64_t x) {
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    x >>= 8;
    push_back(x & 0xff);
    return *this;
  }
  BufferWithRandomAccess& operator<<(U32LEB x) {
    x.write(this);
    return *this;
  }
  BufferWithRandomAccess& operator<<(U64LEB x) {
    x.write(this);
    return *this;
  }
  BufferWithRandomAccess& operator<<(S32LEB x) {
    x.write(this);
    return *this;
  }
  BufferWithRandomAccess& operator<<(S64LEB x) {
    x.write(this);
    return *this;
  }

  BufferWithRandomAccess& operator<<(uint8_t x) { return *this << (int8_t)x; }
  BufferWithRandomAccess& operator<<(uint16_t x) { return *this << (int16_t)x; }
  BufferWithRandomAccess& operator<<(uint32_t x) { return *this << (int32_t)x; }
  BufferWithRandomAccess& operator<<(uint64_t x) { return *this << (int64_t)x; }

  BufferWithRandomAccess& operator<<(float x) {
    return *this << Literal(x).reinterpreti32();
  }
  BufferWithRandomAccess& operator<<(double x) {
    return *this << Literal(x).reinterpreti64();
  }

  void writeAt(size_t i, uint16_t x) {
    (*this)[i] = x & 0xff;
    (*this)[i + 1] = x >> 8;
  }
  void writeAt(size_t i, uint32_t x) {
    (*this)[i] = x & 0xff;
    x >>= 8;
    (*this)[i + 1] = x & 0xff;
    x >>= 8;
    (*this)[i + 2] = x & 0xff;
    x >>= 8;
    (*this)[i + 3] = x & 0xff;
  }

  // writes out an LEB to an arbitrary location. this writes the LEB as a full
  // 5 bytes, the fixed amount that can easily be set aside ahead of time
  void writeAtFullFixedSize(size_t i, U32LEB x) {
    // fill all 5 bytes, we have to do this when backpatching
    x.writeAt(this, i, MaxLEB32Bytes);
  }
  // writes out an LEB of normal size
  // returns how many bytes were written
  size_t writeAt(size_t i, U32LEB x) { return x.writeAt(this, i); }

  template<typename T> void writeTo(T& o) {
    for (auto c : *this) {
      o << c;
    }
  }

  std::vector<char> getAsChars() {
    std::vector<char> ret;
    ret.resize(size());
    std::copy(begin(), end(), ret.begin());
    return ret;
  }

  // Writes bytes in the maximum amount for a U32 LEB placeholder. Return the
  // offset we wrote it at. The LEB can then be patched with the proper value
  // later, when the size is known.
  BinaryLocation writeU32LEBPlaceholder() {
    BinaryLocation ret = size();
    *this << int32_t(0);
    *this << int8_t(0);
    return ret;
  }

  // Given the location of a maximum-size LEB placeholder, as returned from
  // writeU32LEBPlaceholder, use the current buffer size to figure out the size
  // that should be written there, and emit an optimal-size LEB. Move contents
  // backwards if we used fewer bytes, and return the number of bytes we moved.
  // (Thus, if we return >0, we moved code backwards, and the caller may need to
  // adjust things.)
  BinaryLocation emitRetroactiveSectionSizeLEB(BinaryLocation start) {
    // Do not include the LEB itself in the section size.
    auto sectionSize = size() - start - MaxLEB32Bytes;
    auto sizeFieldSize = writeAt(start, U32LEB(sectionSize));

    // We can move things back if the actual LEB for the size doesn't use the
    // maximum 5 bytes. In that case we need to adjust offsets after we move
    // things backwards.
    auto adjustmentForLEBShrinking = MaxLEB32Bytes - sizeFieldSize;
    if (adjustmentForLEBShrinking) {
      // We can save some room.
      assert(sizeFieldSize < MaxLEB32Bytes);
      std::move(&(*this)[start] + MaxLEB32Bytes,
                &(*this)[start] + MaxLEB32Bytes + sectionSize,
                &(*this)[start] + sizeFieldSize);
      resize(size() - adjustmentForLEBShrinking);
    }

    return adjustmentForLEBShrinking;
  }

  void writeInlineString(std::string_view name) {
    auto size = name.size();
    auto data = name.data();
    *this << U32LEB(size);
    for (size_t i = 0; i < size; i++) {
      *this << int8_t(data[i]);
    }
  }
};

namespace BinaryConsts {

enum Meta { Magic = 0x6d736100, Version = 0x01 };

enum Section {
  Custom = 0,
  Type = 1,
  Import = 2,
  Function = 3,
  Table = 4,
  Memory = 5,
  Global = 6,
  Export = 7,
  Start = 8,
  Element = 9,
  Code = 10,
  Data = 11,
  DataCount = 12,
  Tag = 13,
  Strings = 14,
};

// A passive segment is a segment that will not be automatically copied into a
//   memory or table on instantiation, and must instead be applied manually
//   using the instructions memory.init or table.init.
// An active segment is equivalent to a passive segment, but with an implicit
//   memory.init followed by a data.drop (or table.init followed by a elem.drop)
//   that is prepended to the module's start function.
// A declarative element segment is not available at runtime but merely serves
//   to forward-declare references that are formed in code with instructions
//   like ref.func.
enum SegmentFlag {
  // Bit 0: 0 = active, 1 = passive
  IsPassive = 1 << 0,
  // Bit 1 if passive: 0 = passive, 1 = declarative
  IsDeclarative = 1 << 1,
  // Bit 1 if active: 0 = index 0, 1 = index given
  HasIndex = 1 << 1,
  // Table element segments only:
  // Bit 2: 0 = elemType is funcref and a vector of func indexes given
  //        1 = elemType is given and a vector of ref expressions is given
  UsesExpressions = 1 << 2
};

enum BrOnCastFlag {
  InputNullable = 1 << 0,
  OutputNullable = 1 << 1,
};

constexpr uint32_t ExactImport = 1 << 5;

constexpr uint32_t HasMemoryOrderMask = 1 << 5;
constexpr uint32_t HasMemoryIndexMask = 1 << 6;

enum EncodedType {
  // value types
  i32 = -0x1,  // 0x7f
  i64 = -0x2,  // 0x7e
  f32 = -0x3,  // 0x7d
  f64 = -0x4,  // 0x7c
  v128 = -0x5, // 0x7b
  // packed types
  i8 = -0x8,         // 0x78
  i16 = -0x9,        // 0x77
  waitQueue = -0x24, // 0x5c
  // reference types
  nullfuncref = -0xd,   // 0x73
  nullexternref = -0xe, // 0x72
  nullref = -0xf,       // 0x71
  i31ref = -0x14,       // 0x6c
  structref = -0x15,    // 0x6b
  arrayref = -0x16,     // 0x6a
  funcref = -0x10,      // 0x70
  externref = -0x11,    // 0x6f
  anyref = -0x12,       // 0x6e
  eqref = -0x13,        // 0x6d
  nonnullable = -0x1c,  // 0x64
  nullable = -0x1d,     // 0x63
  contref = -0x18,      // 0x68
  nullcontref = -0x0b,  // 0x75
  // exception handling
  exnref = -0x17,    // 0x69
  nullexnref = -0xc, // 0x74
  // string reference types
  stringref = -0x19, // 0x67
  // type forms
  Func = 0x60,
  Cont = 0x5d,
  Struct = 0x5f,
  Array = 0x5e,
  Sub = 0x50,
  SubFinal = 0x4f,
  Shared = 0x65,
  SharedLEB = -0x1b, // Also 0x65 as an SLEB128
  Exact = 0x62,
  ExactLEB = -0x1e, // Also 0x62 as an SLEB128
  Rec = 0x4e,
  Descriptor = 0x4d,
  Describes = 0x4c,
  // block_type
  Empty = -0x40, // 0x40
};

enum EncodedHeapType {
  nofunc = -0xd,   // 0x73
  noext = -0xe,    // 0x72
  none = -0xf,     // 0x71
  func = -0x10,    // 0x70
  ext = -0x11,     // 0x6f
  any = -0x12,     // 0x6e
  eq = -0x13,      // 0x6d
  exn = -0x17,     // 0x69
  noexn = -0xc,    // 0x74
  cont = -0x18,    // 0x68
  nocont = -0x0b,  // 0x75
  i31 = -0x14,     // 0x6c
  struct_ = -0x15, // 0x6b
  array = -0x16,   // 0x6a
  string = -0x19,  // 0x67
};

namespace CustomSections {
extern const char* Name;
extern const char* SourceMapUrl;
extern const char* Dylink;
extern const char* Dylink0;
extern const char* Linking;
extern const char* Producers;
extern const char* BuildId;
extern const char* TargetFeatures;

extern const char* AtomicsFeature;
extern const char* BulkMemoryFeature;
extern const char* ExceptionHandlingFeature;
extern const char* MutableGlobalsFeature;
extern const char* TruncSatFeature;
extern const char* SignExtFeature;
extern const char* SIMD128Feature;
extern const char* ExceptionHandlingFeature;
extern const char* TailCallFeature;
extern const char* ReferenceTypesFeature;
extern const char* MultivalueFeature;
extern const char* GCFeature;
extern const char* Memory64Feature;
extern const char* RelaxedSIMDFeature;
extern const char* ExtendedConstFeature;
extern const char* StringsFeature;
extern const char* MultiMemoryFeature;
extern const char* StackSwitchingFeature;
extern const char* SharedEverythingFeature;
extern const char* FP16Feature;
extern const char* BulkMemoryOptFeature;
extern const char* CallIndirectOverlongFeature;
extern const char* CustomDescriptorsFeature;
extern const char* RelaxedAtomicsFeature;

enum Subsection {
  NameModule = 0,
  NameFunction = 1,
  NameLocal = 2,
  // see: https://github.com/WebAssembly/extended-name-section
  NameLabel = 3,
  NameType = 4,
  NameTable = 5,
  NameMemory = 6,
  NameGlobal = 7,
  NameElem = 8,
  NameData = 9,
  // see: https://github.com/WebAssembly/gc/issues/193
  NameField = 10,
  NameTag = 11,

  DylinkMemInfo = 1,
  DylinkNeeded = 2,
};

} // namespace CustomSections

enum ASTNodes {
  Unreachable = 0x00,
  Nop = 0x01,
  Block = 0x02,
  Loop = 0x03,
  If = 0x04,
  Else = 0x05,

  End = 0x0b,
  Br = 0x0c,
  BrIf = 0x0d,
  BrTable = 0x0e,
  Return = 0x0f,

  CallFunction = 0x10,
  CallIndirect = 0x11,
  RetCallFunction = 0x12,
  RetCallIndirect = 0x13,

  Drop = 0x1a,
  Select = 0x1b,
  SelectWithType = 0x1c, // added in reference types proposal

  LocalGet = 0x20,
  LocalSet = 0x21,
  LocalTee = 0x22,
  GlobalGet = 0x23,
  GlobalSet = 0x24,

  TableGet = 0x25,
  TableSet = 0x26,

  I32LoadMem = 0x28,
  I64LoadMem = 0x29,
  F32LoadMem = 0x2a,
  F64LoadMem = 0x2b,

  I32LoadMem8S = 0x2c,
  I32LoadMem8U = 0x2d,
  I32LoadMem16S = 0x2e,
  I32LoadMem16U = 0x2f,
  I64LoadMem8S = 0x30,
  I64LoadMem8U = 0x31,
  I64LoadMem16S = 0x32,
  I64LoadMem16U = 0x33,
  I64LoadMem32S = 0x34,
  I64LoadMem32U = 0x35,

  I32StoreMem = 0x36,
  I64StoreMem = 0x37,
  F32StoreMem = 0x38,
  F64StoreMem = 0x39,

  I32StoreMem8 = 0x3a,
  I32StoreMem16 = 0x3b,
  I64StoreMem8 = 0x3c,
  I64StoreMem16 = 0x3d,
  I64StoreMem32 = 0x3e,

  MemorySize = 0x3f,
  MemoryGrow = 0x40,

  I32Const = 0x41,
  I64Const = 0x42,
  F32Const = 0x43,
  F64Const = 0x44,

  I32EqZ = 0x45,
  I32Eq = 0x46,
  I32Ne = 0x47,
  I32LtS = 0x48,
  I32LtU = 0x49,
  I32GtS = 0x4a,
  I32GtU = 0x4b,
  I32LeS = 0x4c,
  I32LeU = 0x4d,
  I32GeS = 0x4e,
  I32GeU = 0x4f,
  I64EqZ = 0x50,
  I64Eq = 0x51,
  I64Ne = 0x52,
  I64LtS = 0x53,
  I64LtU = 0x54,
  I64GtS = 0x55,
  I64GtU = 0x56,
  I64LeS = 0x57,
  I64LeU = 0x58,
  I64GeS = 0x59,
  I64GeU = 0x5a,
  F32Eq = 0x5b,
  F32Ne = 0x5c,
  F32Lt = 0x5d,
  F32Gt = 0x5e,
  F32Le = 0x5f,
  F32Ge = 0x60,
  F64Eq = 0x61,
  F64Ne = 0x62,
  F64Lt = 0x63,
  F64Gt = 0x64,
  F64Le = 0x65,
  F64Ge = 0x66,

  I32Clz = 0x67,
  I32Ctz = 0x68,
  I32Popcnt = 0x69,
  I32Add = 0x6a,
  I32Sub = 0x6b,
  I32Mul = 0x6c,
  I32DivS = 0x6d,
  I32DivU = 0x6e,
  I32RemS = 0x6f,
  I32RemU = 0x70,
  I32And = 0x71,
  I32Or = 0x72,
  I32Xor = 0x73,
  I32Shl = 0x74,
  I32ShrS = 0x75,
  I32ShrU = 0x76,
  I32RotL = 0x77,
  I32RotR = 0x78,

  I64Clz = 0x79,
  I64Ctz = 0x7a,
  I64Popcnt = 0x7b,
  I64Add = 0x7c,
  I64Sub = 0x7d,
  I64Mul = 0x7e,
  I64DivS = 0x7f,
  I64DivU = 0x80,
  I64RemS = 0x81,
  I64RemU = 0x82,
  I64And = 0x83,
  I64Or = 0x84,
  I64Xor = 0x85,
  I64Shl = 0x86,
  I64ShrS = 0x87,
  I64ShrU = 0x88,
  I64RotL = 0x89,
  I64RotR = 0x8a,

  F32Abs = 0x8b,
  F32Neg = 0x8c,
  F32Ceil = 0x8d,
  F32Floor = 0x8e,
  F32Trunc = 0x8f,
  F32Nearest = 0x90,
  F32Sqrt = 0x91,
  F32Add = 0x92,
  F32Sub = 0x93,
  F32Mul = 0x94,
  F32Div = 0x95,
  F32Min = 0x96,
  F32Max = 0x97,
  F32CopySign = 0x98,

  F64Abs = 0x99,
  F64Neg = 0x9a,
  F64Ceil = 0x9b,
  F64Floor = 0x9c,
  F64Trunc = 0x9d,
  F64Nearest = 0x9e,
  F64Sqrt = 0x9f,
  F64Add = 0xa0,
  F64Sub = 0xa1,
  F64Mul = 0xa2,
  F64Div = 0xa3,
  F64Min = 0xa4,
  F64Max = 0xa5,
  F64CopySign = 0xa6,

  I32WrapI64 = 0xa7,
  I32STruncF32 = 0xa8,
  I32UTruncF32 = 0xa9,
  I32STruncF64 = 0xaa,
  I32UTruncF64 = 0xab,
  I64SExtendI32 = 0xac,
  I64UExtendI32 = 0xad,
  I64STruncF32 = 0xae,
  I64UTruncF32 = 0xaf,
  I64STruncF64 = 0xb0,
  I64UTruncF64 = 0xb1,
  F32SConvertI32 = 0xb2,
  F32UConvertI32 = 0xb3,
  F32SConvertI64 = 0xb4,
  F32UConvertI64 = 0xb5,
  F32DemoteI64 = 0xb6,
  F64SConvertI32 = 0xb7,
  F64UConvertI32 = 0xb8,
  F64SConvertI64 = 0xb9,
  F64UConvertI64 = 0xba,
  F64PromoteF32 = 0xbb,

  I32ReinterpretF32 = 0xbc,
  I64ReinterpretF64 = 0xbd,
  F32ReinterpretI32 = 0xbe,
  F64ReinterpretI64 = 0xbf,

  I32ExtendS8 = 0xc0,
  I32ExtendS16 = 0xc1,
  I64ExtendS8 = 0xc2,
  I64ExtendS16 = 0xc3,
  I64ExtendS32 = 0xc4,

  // prefixes

  GCPrefix = 0xfb,
  MiscPrefix = 0xfc,
  SIMDPrefix = 0xfd,
  AtomicPrefix = 0xfe,

  // atomic opcodes

  AtomicNotify = 0x00,
  I32AtomicWait = 0x01,
  I64AtomicWait = 0x02,
  AtomicFence = 0x03,
  Pause = 0x04,
  StructWait = 0x05,
  StructNotify = 0x06,

  I32AtomicLoad = 0x10,
  I64AtomicLoad = 0x11,
  I32AtomicLoad8U = 0x12,
  I32AtomicLoad16U = 0x13,
  I64AtomicLoad8U = 0x14,
  I64AtomicLoad16U = 0x15,
  I64AtomicLoad32U = 0x16,
  I32AtomicStore = 0x17,
  I64AtomicStore = 0x18,
  I32AtomicStore8 = 0x19,
  I32AtomicStore16 = 0x1a,
  I64AtomicStore8 = 0x1b,
  I64AtomicStore16 = 0x1c,
  I64AtomicStore32 = 0x1d,

  AtomicRMWOps_Begin = 0x1e,
  I32AtomicRMWAdd = 0x1e,
  I64AtomicRMWAdd = 0x1f,
  I32AtomicRMWAdd8U = 0x20,
  I32AtomicRMWAdd16U = 0x21,
  I64AtomicRMWAdd8U = 0x22,
  I64AtomicRMWAdd16U = 0x23,
  I64AtomicRMWAdd32U = 0x24,
  I32AtomicRMWSub = 0x25,
  I64AtomicRMWSub = 0x26,
  I32AtomicRMWSub8U = 0x27,
  I32AtomicRMWSub16U = 0x28,
  I64AtomicRMWSub8U = 0x29,
  I64AtomicRMWSub16U = 0x2a,
  I64AtomicRMWSub32U = 0x2b,
  I32AtomicRMWAnd = 0x2c,
  I64AtomicRMWAnd = 0x2d,
  I32AtomicRMWAnd8U = 0x2e,
  I32AtomicRMWAnd16U = 0x2f,
  I64AtomicRMWAnd8U = 0x30,
  I64AtomicRMWAnd16U = 0x31,
  I64AtomicRMWAnd32U = 0x32,
  I32AtomicRMWOr = 0x33,
  I64AtomicRMWOr = 0x34,
  I32AtomicRMWOr8U = 0x35,
  I32AtomicRMWOr16U = 0x36,
  I64AtomicRMWOr8U = 0x37,
  I64AtomicRMWOr16U = 0x38,
  I64AtomicRMWOr32U = 0x39,
  I32AtomicRMWXor = 0x3a,
  I64AtomicRMWXor = 0x3b,
  I32AtomicRMWXor8U = 0x3c,
  I32AtomicRMWXor16U = 0x3d,
  I64AtomicRMWXor8U = 0x3e,
  I64AtomicRMWXor16U = 0x3f,
  I64AtomicRMWXor32U = 0x40,
  I32AtomicRMWXchg = 0x41,
  I64AtomicRMWXchg = 0x42,
  I32AtomicRMWXchg8U = 0x43,
  I32AtomicRMWXchg16U = 0x44,
  I64AtomicRMWXchg8U = 0x45,
  I64AtomicRMWXchg16U = 0x46,
  I64AtomicRMWXchg32U = 0x47,
  AtomicRMWOps_End = 0x47,

  AtomicCmpxchgOps_Begin = 0x48,
  I32AtomicCmpxchg = 0x48,
  I64AtomicCmpxchg = 0x49,
  I32AtomicCmpxchg8U = 0x4a,
  I32AtomicCmpxchg16U = 0x4b,
  I64AtomicCmpxchg8U = 0x4c,
  I64AtomicCmpxchg16U = 0x4d,
  I64AtomicCmpxchg32U = 0x4e,
  AtomicCmpxchgOps_End = 0x4e,

  // truncsat opcodes

  I32STruncSatF32 = 0x00,
  I32UTruncSatF32 = 0x01,
  I32STruncSatF64 = 0x02,
  I32UTruncSatF64 = 0x03,
  I64STruncSatF32 = 0x04,
  I64UTruncSatF32 = 0x05,
  I64STruncSatF64 = 0x06,
  I64UTruncSatF64 = 0x07,

  // SIMD opcodes

  V128Load = 0x00,
  V128Load8x8S = 0x01,
  V128Load8x8U = 0x02,
  V128Load16x4S = 0x03,
  V128Load16x4U = 0x04,
  V128Load32x2S = 0x05,
  V128Load32x2U = 0x06,
  V128Load8Splat = 0x07,
  V128Load16Splat = 0x08,
  V128Load32Splat = 0x09,
  V128Load64Splat = 0x0a,
  V128Store = 0x0b,

  V128Const = 0x0c,
  I8x16Shuffle = 0x0d,
  I8x16Swizzle = 0x0e,

  I8x16Splat = 0x0f,
  I16x8Splat = 0x10,
  I32x4Splat = 0x11,
  I64x2Splat = 0x12,
  F32x4Splat = 0x13,
  F64x2Splat = 0x14,

  I8x16ExtractLaneS = 0x15,
  I8x16ExtractLaneU = 0x16,
  I8x16ReplaceLane = 0x17,
  I16x8ExtractLaneS = 0x18,
  I16x8ExtractLaneU = 0x19,
  I16x8ReplaceLane = 0x1a,
  I32x4ExtractLane = 0x1b,
  I32x4ReplaceLane = 0x1c,
  I64x2ExtractLane = 0x1d,
  I64x2ReplaceLane = 0x1e,
  F32x4ExtractLane = 0x1f,
  F32x4ReplaceLane = 0x20,
  F64x2ExtractLane = 0x21,
  F64x2ReplaceLane = 0x22,

  I8x16Eq = 0x23,
  I8x16Ne = 0x24,
  I8x16LtS = 0x25,
  I8x16LtU = 0x26,
  I8x16GtS = 0x27,
  I8x16GtU = 0x28,
  I8x16LeS = 0x29,
  I8x16LeU = 0x2a,
  I8x16GeS = 0x2b,
  I8x16GeU = 0x2c,
  I16x8Eq = 0x2d,
  I16x8Ne = 0x2e,
  I16x8LtS = 0x2f,
  I16x8LtU = 0x30,
  I16x8GtS = 0x31,
  I16x8GtU = 0x32,
  I16x8LeS = 0x33,
  I16x8LeU = 0x34,
  I16x8GeS = 0x35,
  I16x8GeU = 0x36,
  I32x4Eq = 0x37,
  I32x4Ne = 0x38,
  I32x4LtS = 0x39,
  I32x4LtU = 0x3a,
  I32x4GtS = 0x3b,
  I32x4GtU = 0x3c,
  I32x4LeS = 0x3d,
  I32x4LeU = 0x3e,
  I32x4GeS = 0x3f,
  I32x4GeU = 0x40,
  F32x4Eq = 0x41,
  F32x4Ne = 0x42,
  F32x4Lt = 0x43,
  F32x4Gt = 0x44,
  F32x4Le = 0x45,
  F32x4Ge = 0x46,
  F64x2Eq = 0x47,
  F64x2Ne = 0x48,
  F64x2Lt = 0x49,
  F64x2Gt = 0x4a,
  F64x2Le = 0x4b,
  F64x2Ge = 0x4c,

  V128Not = 0x4d,
  V128And = 0x4e,
  V128Andnot = 0x4f,
  V128Or = 0x50,
  V128Xor = 0x51,
  V128Bitselect = 0x52,
  V128AnyTrue = 0x53,

  V128Load8Lane = 0x54,
  V128Load16Lane = 0x55,
  V128Load32Lane = 0x56,
  V128Load64Lane = 0x57,
  V128Store8Lane = 0x58,
  V128Store16Lane = 0x59,
  V128Store32Lane = 0x5a,
  V128Store64Lane = 0x5b,
  V128Load32Zero = 0x5c,
  V128Load64Zero = 0x5d,

  F32x4DemoteF64x2Zero = 0x5e,
  F64x2PromoteLowF32x4 = 0x5f,

  I8x16Abs = 0x60,
  I8x16Neg = 0x61,
  I8x16Popcnt = 0x62,
  I8x16AllTrue = 0x63,
  I8x16Bitmask = 0x64,
  I8x16NarrowI16x8S = 0x65,
  I8x16NarrowI16x8U = 0x66,
  F32x4Ceil = 0x67,
  F32x4Floor = 0x68,
  F32x4Trunc = 0x69,
  F32x4Nearest = 0x6a,
  I8x16Shl = 0x6b,
  I8x16ShrS = 0x6c,
  I8x16ShrU = 0x6d,
  I8x16Add = 0x6e,
  I8x16AddSatS = 0x6f,
  I8x16AddSatU = 0x70,
  I8x16Sub = 0x71,
  I8x16SubSatS = 0x72,
  I8x16SubSatU = 0x73,
  F64x2Ceil = 0x74,
  F64x2Floor = 0x75,
  I8x16MinS = 0x76,
  I8x16MinU = 0x77,
  I8x16MaxS = 0x78,
  I8x16MaxU = 0x79,
  F64x2Trunc = 0x7a,
  I8x16AvgrU = 0x7b,
  I16x8ExtaddPairwiseI8x16S = 0x7c,
  I16x8ExtaddPairwiseI8x16U = 0x7d,
  I32x4ExtaddPairwiseI16x8S = 0x7e,
  I32x4ExtaddPairwiseI16x8U = 0x7f,

  I16x8Abs = 0x80,
  I16x8Neg = 0x81,
  I16x8Q15MulrSatS = 0x82,
  I16x8AllTrue = 0x83,
  I16x8Bitmask = 0x84,
  I16x8NarrowI32x4S = 0x85,
  I16x8NarrowI32x4U = 0x86,
  I16x8ExtendLowI8x16S = 0x87,
  I16x8ExtendHighI8x16S = 0x88,
  I16x8ExtendLowI8x16U = 0x89,
  I16x8ExtendHighI8x16U = 0x8a,
  I16x8Shl = 0x8b,
  I16x8ShrS = 0x8c,
  I16x8ShrU = 0x8d,
  I16x8Add = 0x8e,
  I16x8AddSatS = 0x8f,
  I16x8AddSatU = 0x90,
  I16x8Sub = 0x91,
  I16x8SubSatS = 0x92,
  I16x8SubSatU = 0x93,
  F64x2Nearest = 0x94,
  I16x8Mul = 0x95,
  I16x8MinS = 0x96,
  I16x8MinU = 0x97,
  I16x8MaxS = 0x98,
  I16x8MaxU = 0x99,
  // 0x9a unused
  I16x8AvgrU = 0x9b,
  I16x8ExtmulLowI8x16S = 0x9c,
  I16x8ExtmulHighI8x16S = 0x9d,
  I16x8ExtmulLowI8x16U = 0x9e,
  I16x8ExtmulHighI8x16U = 0x9f,

  I32x4Abs = 0xa0,
  I32x4Neg = 0xa1,
  // 0xa2 unused
  I32x4AllTrue = 0xa3,
  I32x4Bitmask = 0xa4,
  // 0xa5 unused
  // 0xa6 unused
  I32x4ExtendLowI16x8S = 0xa7,
  I32x4ExtendHighI16x8S = 0xa8,
  I32x4ExtendLowI16x8U = 0xa9,
  I32x4ExtendHighI16x8U = 0xaa,
  I32x4Shl = 0xab,
  I32x4ShrS = 0xac,
  I32x4ShrU = 0xad,
  I32x4Add = 0xae,
  // 0xaf unused
  // 0xb0 unused
  I32x4Sub = 0xb1,
  // 0xb2 unused
  // 0xb3 unused
  // 0xb4 unused
  I32x4Mul = 0xb5,
  I32x4MinS = 0xb6,
  I32x4MinU = 0xb7,
  I32x4MaxS = 0xb8,
  I32x4MaxU = 0xb9,
  I32x4DotI16x8S = 0xba,
  // 0xbb unused
  I32x4ExtmulLowI16x8S = 0xbc,
  I32x4ExtmulHighI16x8S = 0xbd,
  I32x4ExtmulLowI16x8U = 0xbe,
  I32x4ExtmulHighI16x8U = 0xbf,

  I64x2Abs = 0xc0,
  I64x2Neg = 0xc1,
  // 0xc2 unused
  I64x2AllTrue = 0xc3,
  I64x2Bitmask = 0xc4,
  // 0xc5 unused
  // 0xc6 unused
  I64x2ExtendLowI32x4S = 0xc7,
  I64x2ExtendHighI32x4S = 0xc8,
  I64x2ExtendLowI32x4U = 0xc9,
  I64x2ExtendHighI32x4U = 0xca,
  I64x2Shl = 0xcb,
  I64x2ShrS = 0xcc,
  I64x2ShrU = 0xcd,
  I64x2Add = 0xce,
  // 0xcf unused
  // 0xd0 unused
  I64x2Sub = 0xd1,
  // 0xd2 unused
  // 0xd3 unused
  // 0xd4 unused
  I64x2Mul = 0xd5,
  I64x2Eq = 0xd6,
  I64x2Ne = 0xd7,
  I64x2LtS = 0xd8,
  I64x2GtS = 0xd9,
  I64x2LeS = 0xda,
  I64x2GeS = 0xdb,
  I64x2ExtmulLowI32x4S = 0xdc,
  I64x2ExtmulHighI32x4S = 0xdd,
  I64x2ExtmulLowI32x4U = 0xde,
  I64x2ExtmulHighI32x4U = 0xdf,

  F32x4Abs = 0xe0,
  F32x4Neg = 0xe1,
  // 0xe2 unused
  F32x4Sqrt = 0xe3,
  F32x4Add = 0xe4,
  F32x4Sub = 0xe5,
  F32x4Mul = 0xe6,
  F32x4Div = 0xe7,
  F32x4Min = 0xe8,
  F32x4Max = 0xe9,
  F32x4Pmin = 0xea,
  F32x4Pmax = 0xeb,

  F64x2Abs = 0xec,
  F64x2Neg = 0xed,
  // 0xee unused
  F64x2Sqrt = 0xef,
  F64x2Add = 0xf0,
  F64x2Sub = 0xf1,
  F64x2Mul = 0xf2,
  F64x2Div = 0xf3,
  F64x2Min = 0xf4,
  F64x2Max = 0xf5,
  F64x2Pmin = 0xf6,
  F64x2Pmax = 0xf7,

  I32x4TruncSatF32x4S = 0xf8,
  I32x4TruncSatF32x4U = 0xf9,
  F32x4ConvertI32x4S = 0xfa,
  F32x4ConvertI32x4U = 0xfb,
  I32x4TruncSatF64x2SZero = 0xfc,
  I32x4TruncSatF64x2UZero = 0xfd,
  F64x2ConvertLowI32x4S = 0xfe,
  F64x2ConvertLowI32x4U = 0xff,

  // relaxed SIMD opcodes
  I8x16RelaxedSwizzle = 0x100,
  I32x4RelaxedTruncF32x4S = 0x101,
  I32x4RelaxedTruncF32x4U = 0x102,
  I32x4RelaxedTruncF64x2SZero = 0x103,
  I32x4RelaxedTruncF64x2UZero = 0x104,
  F16x8Madd = 0x14e,
  F16x8Nmadd = 0x14f,
  F32x4RelaxedMadd = 0x105,
  F32x4RelaxedNmadd = 0x106,
  F64x2RelaxedMadd = 0x107,
  F64x2RelaxedNmadd = 0x108,
  I8x16Laneselect = 0x109,
  I16x8Laneselect = 0x10a,
  I32x4Laneselect = 0x10b,
  I64x2Laneselect = 0x10c,
  F32x4RelaxedMin = 0x10d,
  F32x4RelaxedMax = 0x10e,
  F64x2RelaxedMin = 0x10f,
  F64x2RelaxedMax = 0x110,
  I16x8RelaxedQ15MulrS = 0x111,
  I16x8DotI8x16I7x16S = 0x112,
  I32x4DotI8x16I7x16AddS = 0x113,

  // half precision opcodes
  F32_F16LoadMem = 0x30,
  F32_F16StoreMem = 0x31,
  F16x8Splat = 0x120,
  F16x8ExtractLane = 0x121,
  F16x8ReplaceLane = 0x122,
  F16x8Abs = 0x130,
  F16x8Neg = 0x131,
  F16x8Sqrt = 0x132,
  F16x8Ceil = 0x133,
  F16x8Floor = 0x134,
  F16x8Trunc = 0x135,
  F16x8Nearest = 0x136,
  F16x8Eq = 0x137,
  F16x8Ne = 0x138,
  F16x8Lt = 0x139,
  F16x8Gt = 0x13a,
  F16x8Le = 0x13b,
  F16x8Ge = 0x13c,
  F16x8Add = 0x13d,
  F16x8Sub = 0x13e,
  F16x8Mul = 0x13f,
  F16x8Div = 0x140,
  F16x8Min = 0x141,
  F16x8Max = 0x142,
  F16x8Pmin = 0x143,
  F16x8Pmax = 0x144,
  I16x8TruncSatF16x8S = 0x145,
  I16x8TruncSatF16x8U = 0x146,
  F16x8ConvertI16x8S = 0x147,
  F16x8ConvertI16x8U = 0x148,

  // bulk memory opcodes

  MemoryInit = 0x08,
  DataDrop = 0x09,
  MemoryCopy = 0x0a,
  MemoryFill = 0x0b,

  // reference types opcodes

  TableGrow = 0x0f,
  TableSize = 0x10,
  TableFill = 0x11,
  TableCopy = 0x0e,
  TableInit = 0x0c,
  ElemDrop = 0x0d,
  RefNull = 0xd0,
  RefIsNull = 0xd1,
  RefFunc = 0xd2,
  RefEq = 0xd3,
  RefAsNonNull = 0xd4,
  BrOnNull = 0xd5,
  BrOnNonNull = 0xd6,

  // exception handling opcodes

  Try = 0x06,
  Catch_Legacy = 0x07,    // Legacy 'catch'
  CatchAll_Legacy = 0x19, // Legacy 'catch_all'
  Delegate = 0x18,
  Throw = 0x08,
  Rethrow = 0x09,
  TryTable = 0x1f,
  Catch = 0x00,
  CatchRef = 0x01,
  CatchAll = 0x02,
  CatchAllRef = 0x03,
  ThrowRef = 0x0a,

  // typed function references opcodes

  CallRef = 0x14,
  RetCallRef = 0x15,

  // gc opcodes

  StructNew = 0x00,
  StructNewDefault = 0x01,
  StructNewDesc = 0x20,
  StructNewDefaultDesc = 0x21,
  StructGet = 0x02,
  StructGetS = 0x03,
  StructGetU = 0x04,
  StructSet = 0x05,
  ArrayNew = 0x06,
  ArrayNewDefault = 0x07,
  ArrayNewFixed = 0x08,
  ArrayNewData = 0x09,
  ArrayNewElem = 0x0a,
  ArrayGet = 0x0b,
  ArrayGetS = 0x0c,
  ArrayGetU = 0x0d,
  ArraySet = 0x0e,
  ArrayLen = 0x0f,
  ArrayFill = 0x10,
  ArrayCopy = 0x11,
  ArrayInitData = 0x12,
  ArrayInitElem = 0x13,
  RefTest = 0x14,
  RefTestNull = 0x15,
  RefCast = 0x16,
  RefCastNull = 0x17,
  RefCastDescEq = 0x23,
  RefCastDescEqNull = 0x24,
  BrOnCast = 0x18,
  BrOnCastFail = 0x19,
  BrOnCastDescEq = 0x25,
  BrOnCastDescEqFail = 0x26,
  AnyConvertExtern = 0x1a,
  ExternConvertAny = 0x1b,
  RefI31 = 0x1c,
  I31GetS = 0x1d,
  I31GetU = 0x1e,
  RefI31Shared = 0x1f,
  RefGetDesc = 0x22,

  // Shared GC Opcodes

  OrderSeqCst = 0x0,
  OrderAcqRel = 0x1,
  StructAtomicGet = 0x5c,
  StructAtomicGetS = 0x5d,
  StructAtomicGetU = 0x5e,
  StructAtomicSet = 0x5f,
  StructAtomicRMWAdd = 0x60,
  StructAtomicRMWSub = 0x61,
  StructAtomicRMWAnd = 0x62,
  StructAtomicRMWOr = 0x63,
  StructAtomicRMWXor = 0x64,
  StructAtomicRMWXchg = 0x65,
  StructAtomicRMWCmpxchg = 0x66,
  ArrayAtomicGet = 0x67,
  ArrayAtomicGetS = 0x68,
  ArrayAtomicGetU = 0x69,
  ArrayAtomicSet = 0x6a,
  ArrayAtomicRMWAdd = 0x6b,
  ArrayAtomicRMWSub = 0x6c,
  ArrayAtomicRMWAnd = 0x6d,
  ArrayAtomicRMWOr = 0x6e,
  ArrayAtomicRMWXor = 0x6f,
  ArrayAtomicRMWXchg = 0x70,
  ArrayAtomicRMWCmpxchg = 0x71,

  // stringref opcodes

  StringConst = 0x82,
  StringMeasureUTF8 = 0x83,
  StringMeasureWTF16 = 0x85,
  StringConcat = 0x88,
  StringEq = 0x89,
  StringIsUSV = 0x8a,
  StringTest = 0x8b,
  StringAsWTF16 = 0x98,
  StringViewWTF16GetCodePoint = 0x9a,
  StringViewWTF16Slice = 0x9c,
  StringCompare = 0xa8,
  StringFromCodePoint = 0xa9,
  StringHash = 0xaa,
  StringNewWTF16Array = 0xb1,
  StringEncodeWTF16Array = 0xb3,
  StringNewLossyUTF8Array = 0xb4,
  StringEncodeLossyUTF8Array = 0xb6,

  // stack switching opcodes
  ContNew = 0xe0,
  ContBind = 0xe1,
  Suspend = 0xe2,
  Resume = 0xe3,
  ResumeThrow = 0xe4,
  ResumeThrowRef = 0xe5,
  Switch = 0xe6,  // NOTE(dhil): the internal class is known as
                  // StackSwitch to avoid conflict with the existing
                  // 'switch table'.
  OnLabel = 0x00, // (on $tag $label)
  OnSwitch = 0x01 // (on $tag switch)
};

enum MemoryAccess {
  Offset = 0x10,    // bit 4
  Alignment = 0x80, // bit 7
  NaturalAlignment = 0
};

enum MemoryFlags { HasMaximum = 1 << 0, IsShared = 1 << 1, Is64 = 1 << 2 };

enum FeaturePrefix { FeatureUsed = '+', FeatureDisallowed = '-' };

} // namespace BinaryConsts

// (local index in IR, tuple index) => binary local index
using MappedLocals = std::unordered_map<std::pair<Index, Index>, size_t>;

// Writes out wasm to the binary format

class WasmBinaryWriter {
  // Computes the indexes in a wasm binary, i.e., with function imports
  // and function implementations sharing a single index space, etc.,
  // and with the imports first (the Module's functions and globals
  // arrays are not assumed to be in a particular order, so we can't
  // just use them directly).
  struct BinaryIndexes {
    std::unordered_map<Name, Index> functionIndexes;
    std::unordered_map<Name, Index> tagIndexes;
    std::unordered_map<Name, Index> globalIndexes;
    std::unordered_map<Name, Index> tableIndexes;
    std::unordered_map<Name, Index> elemIndexes;
    std::unordered_map<Name, Index> memoryIndexes;
    std::unordered_map<Name, Index> dataIndexes;

    BinaryIndexes(Module& wasm) {
      auto addIndexes = [&](auto& source, auto& indexes) {
        auto addIndex = [&](auto* curr) {
          auto index = indexes.size();
          indexes[curr->name] = index;
        };
        for (auto& curr : source) {
          if (curr->imported()) {
            addIndex(curr.get());
          }
        }
        for (auto& curr : source) {
          if (!curr->imported()) {
            addIndex(curr.get());
          }
        }
      };
      addIndexes(wasm.functions, functionIndexes);
      addIndexes(wasm.tags, tagIndexes);
      addIndexes(wasm.tables, tableIndexes);
      addIndexes(wasm.memories, memoryIndexes);

      for (auto& curr : wasm.elementSegments) {
        auto index = elemIndexes.size();
        elemIndexes[curr->name] = index;
      }

      for (auto& curr : wasm.dataSegments) {
        auto index = dataIndexes.size();
        dataIndexes[curr->name] = index;
      }

      // Globals may have tuple types in the IR, in which case they lower to
      // multiple globals, one for each tuple element, in the binary. Tuple
      // globals therefore occupy multiple binary indices, and we have to take
      // that into account when calculating indices.
      Index globalCount = 0;
      auto addGlobal = [&](auto* curr) {
        globalIndexes[curr->name] = globalCount;
        globalCount += curr->type.size();
      };
      for (auto& curr : wasm.globals) {
        if (curr->imported()) {
          addGlobal(curr.get());
        }
      }
      for (auto& curr : wasm.globals) {
        if (!curr->imported()) {
          addGlobal(curr.get());
        }
      }
    }
  };

public:
  WasmBinaryWriter(Module* input,
                   BufferWithRandomAccess& o,
                   const PassOptions& options)
    : wasm(input), o(o), options(options), indexes(*input) {
    prepare();
  }

  // locations in the output binary for the various parts of the module
  struct TableOfContents {
    struct Entry {
      Name name;
      size_t offset; // where the entry starts
      size_t size;   // the size of the entry
      Entry(Name name, size_t offset, size_t size)
        : name(name), offset(offset), size(size) {}
    };
    std::vector<Entry> functionBodies;
  } tableOfContents;

  void setNamesSection(bool set) {
    debugInfo = set;
    emitModuleName = set;
  }
  void setEmitModuleName(bool set) { emitModuleName = set; }
  void setSourceMap(std::ostream* set, std::string url) {
    sourceMap = set;
    sourceMapUrl = url;
  }
  void setSymbolMap(std::string set) { symbolMap = set; }

  void write();
  void writeHeader();
  int32_t writeU32LEBPlaceholder();
  void writeResizableLimits(
    Address initial, Address maximum, bool hasMaximum, bool shared, bool is64);
  template<typename T> int32_t startSection(T code);
  void finishSection(int32_t start);
  int32_t startSubsection(BinaryConsts::CustomSections::Subsection code);
  void finishSubsection(int32_t start);
  void writeStart();
  void writeMemories();
  void writeTypes();
  void writeImports();

  void writeFunctionSignatures();
  void writeExpression(Expression* curr);
  void writeFunctions();
  void writeStrings();
  void writeGlobals();
  void writeExports();
  void writeDataCount();
  void writeDataSegments();
  void writeTags();

  uint32_t getFunctionIndex(Name name) const;
  uint32_t getTableIndex(Name name) const;
  uint32_t getMemoryIndex(Name name) const;
  uint32_t getGlobalIndex(Name name) const;
  uint32_t getTagIndex(Name name) const;
  uint32_t getDataSegmentIndex(Name name) const;
  uint32_t getElementSegmentIndex(Name name) const;
  uint32_t getTypeIndex(HeapType type) const;
  uint32_t getSignatureIndex(Signature sig) const;
  uint32_t getStringIndex(Name string) const;

  void writeTableDeclarations();
  void writeElementSegments();
  void writeNames();
  void writeSourceMapUrl();
  void writeSymbolMap();
  void writeLateCustomSections();
  void writeCustomSection(const CustomSection& section);
  void writeFeaturesSection();
  void writeDylinkSection();
  void writeLegacyDylinkSection();

  void initializeDebugInfo();
  void writeSourceMapProlog();
  void writeSourceMapEpilog();
  void writeDebugLocation(const Function::DebugLocation& loc);
  void writeNoDebugLocation();
  void writeSourceMapLocation(Expression* curr, Function* func);

  // Track where expressions go in the binary format.
  void trackExpressionStart(Expression* curr, Function* func);
  void trackExpressionEnd(Expression* curr, Function* func);
  void trackExpressionDelimiter(Expression* curr, Function* func, size_t id);

  // Writes code annotations into a buffer and returns it. We cannot write them
  // directly into the output since we write function code first (to get the
  // offsets for the annotations), and only then can write annotations, which we
  // must then insert before the code (as the spec requires that).
  std::optional<BufferWithRandomAccess> writeCodeAnnotations();

  std::optional<BufferWithRandomAccess> getBranchHintsBuffer();
  std::optional<BufferWithRandomAccess> getInlineHintsBuffer();
  std::optional<BufferWithRandomAccess> getRemovableIfUnusedHintsBuffer();
  std::optional<BufferWithRandomAccess> getJSCalledHintsBuffer();
  std::optional<BufferWithRandomAccess> getIdempotentHintsBuffer();

  // helpers
  void writeInlineString(std::string_view name);
  void writeEscapedName(std::string_view name);
  void writeInlineBuffer(const char* data, size_t size);
  void writeData(const char* data, size_t size);

  struct Buffer {
    const char* data;
    size_t size;
    size_t pointerLocation;
    Buffer(const char* data, size_t size, size_t pointerLocation)
      : data(data), size(size), pointerLocation(pointerLocation) {}
  };

  Module* getModule() { return wasm; }

  void writeType(Type type);

  // Writes an arbitrary heap type, which may be indexed or one of the
  // basic types like funcref.
  void writeHeapType(HeapType type, Exactness exact);
  // Writes an indexed heap type. Note that this is encoded differently than a
  // general heap type because it does not allow negative values for basic heap
  // types.
  void writeIndexedHeapType(HeapType type);

  void writeField(const Field& field);

  void writeMemoryOrder(MemoryOrder order, bool isRMW = false);

private:
  Module* wasm;
  BufferWithRandomAccess& o;
  const PassOptions& options;

  BinaryIndexes indexes;
  ModuleUtils::IndexedHeapTypes indexedTypes;
  std::unordered_map<Signature, uint32_t> signatureIndexes;

  bool debugInfo = true;

  // TODO: Remove `emitModuleName` in the future once there are better ways to
  // ensure modules have meaningful names in stack traces.For example, using
  // ObjectURLs works in FireFox, but not Chrome. See
  // https://bugs.chromium.org/p/v8/issues/detail?id=11808.
  bool emitModuleName = true;

  std::ostream* sourceMap = nullptr;
  std::string sourceMapUrl;
  std::string symbolMap;

  MixedArena allocator;

  // Storage of source map locations until the section is placed at its final
  // location (shrinking LEBs may cause changes there).
  //
  // A null DebugLocation* indicates we have no debug information for that
  // location.
  std::vector<std::pair<size_t, const Function::DebugLocation*>>
    sourceMapLocations;
  size_t sourceMapLocationsSizeAtSectionStart;
  Function::DebugLocation lastDebugLocation;

  std::unique_ptr<ImportInfo> importInfo;

  // General debugging info: track locations as we write.
  BinaryLocations binaryLocations;
  size_t binaryLocationsSizeAtSectionStart;
  // Track the expressions that we added for the current function being
  // written, so that we can update those specific binary locations when
  // the function is written out.
  std::vector<Expression*> binaryLocationTrackedExpressionsForFunc;

  // Maps function names to their mapped locals. This is used when we emit the
  // local names section: we map the locals when writing the function, save that
  // info here, and then use it when writing the names.
  std::unordered_map<Name, MappedLocals> funcMappedLocals;

  // Indexes in the string literal section of each StringConst in the wasm.
  std::unordered_map<Name, Index> stringIndexes;

  void prepare();

  // Internal helper for emitting a code annotation section for a hint that is
  // expression offset based. Receives the name of the section and two
  // functions, one to check if the annotation we care about exists (receiving
  // the annotation), and another to emit it (receiving the annotation and the
  // buffer to write in).
  template<typename HasFunc, typename EmitFunc>
  std::optional<BufferWithRandomAccess>
  writeExpressionHints(Name sectionName, HasFunc has, EmitFunc emit);
};

extern std::vector<char> defaultEmptySourceMap;

class WasmBinaryReader {
  Module& wasm;
  MixedArena& allocator;
  const std::vector<char>& input;

  // Settings.

  bool debugInfo = true;
  bool DWARF = false;
  bool skipFunctionBodies = false;

  // Internal state.

  size_t pos = 0;
  Index startIndex = -1;
  size_t codeSectionLocation;
  std::unordered_set<uint8_t> seenSections;

  IRBuilder builder;
  SourceMapReader sourceMapReader;

  // All types defined in the type section
  std::vector<HeapType> types;

public:
  WasmBinaryReader(Module& wasm,
                   FeatureSet features,
                   const std::vector<char>& input,
                   std::vector<char>& sourceMap = defaultEmptySourceMap);

  void setDebugInfo(bool value) { debugInfo = value; }
  void setDWARF(bool value) { DWARF = value; }
  void setSkipFunctionBodies(bool skipFunctionBodies_) {
    skipFunctionBodies = skipFunctionBodies_;
  }
  void read();
  void readCustomSection(size_t payloadLen);

  bool more() { return pos < input.size(); }

  std::string_view getByteView(size_t size);
  uint8_t getInt8();
  uint16_t getInt16();
  uint32_t getInt32();
  uint64_t getInt64();
  uint8_t getLaneIndex(size_t lanes);
  // it is unsafe to return a float directly, due to ABI issues with the
  // signalling bit
  Literal getFloat32Literal();
  Literal getFloat64Literal();
  Literal getVec128Literal();
  uint32_t getU32LEB();
  uint64_t getU64LEB();
  int32_t getS32LEB();
  int64_t getS64LEB();
  uint64_t getUPtrLEB();

  bool getBasicType(int32_t code, Type& out);
  bool getBasicHeapType(int64_t code, HeapType& out);
  // Get the signature of a control flow structure.
  Signature getBlockType();
  // Read a value and get a type for it.
  Type getType();
  // Get a type given the initial S32LEB has already been read, and is provided.
  Type getType(int code);
  std::pair<HeapType, Exactness> getHeapType();
  HeapType getIndexedHeapType();

  Type getConcreteType();
  Name getInlineString(bool requireValid = true);
  void verifyInt8(int8_t x);
  void verifyInt16(int16_t x);
  void verifyInt32(int32_t x);
  void verifyInt64(int64_t x);
  void readHeader();
  void readStart();
  void readMemories();
  void readTypes();

  // gets a name in the combined import+defined space
  Name getFunctionName(Index index);
  Name getTableName(Index index);
  Name getMemoryName(Index index);
  Name getGlobalName(Index index);
  Name getTagName(Index index);
  Name getDataName(Index index);
  Name getElemName(Index index);

  // gets a memory in the combined import+defined space
  Memory* getMemory(Index index);
  // gets a table in the combined import+defined space
  Table* getTable(Index index);

  void getResizableLimits(Address& initial,
                          Address& max,
                          bool& shared,
                          Type& addressType,
                          Address defaultIfNoMax);
  void readImports();

  // The signatures of each function, including imported functions, given in the
  // import and function sections. Store HeapTypes instead of Signatures because
  // reconstructing the HeapTypes from the Signatures is expensive.
  std::vector<HeapType> functionTypes;

  // Used to make sure the number of imported functions, signatures, and
  // declared functions all match up.
  Index numFuncImports = 0;
  Index numFuncBodies = 0;

  void readFunctionSignatures();
  HeapType getTypeByIndex(Index index);
  HeapType getTypeByFunctionIndex(Index index);
  Signature getSignatureByTypeIndex(Index index);
  Signature getSignatureByFunctionIndex(Index index);

  Name getNextLabel();

  // We read the names section first so we know in advance what names various
  // elements should have. Store the information for use when building
  // expressions.
  std::unordered_map<Index, Name> functionNames;
  std::unordered_map<Index, std::unordered_map<Index, Name>> localNames;
  std::unordered_map<Index, Name> typeNames;
  std::unordered_map<Index, std::unordered_map<Index, Name>> fieldNames;
  std::unordered_map<Index, Name> tableNames;
  std::unordered_map<Index, Name> memoryNames;
  std::unordered_map<Index, Name> globalNames;
  std::unordered_map<Index, Name> tagNames;
  std::unordered_map<Index, Name> dataNames;
  std::unordered_map<Index, Name> elemNames;

  // The names that are already used (either from the names section, or that we
  // generate as internal names for un-named things).
  std::unordered_set<Name> usedFunctionNames, usedTableNames, usedMemoryNames,
    usedGlobalNames, usedTagNames;

  Function* currFunction = nullptr;
  // before we see a function (like global init expressions), there is no end of
  // function to check
  Index endOfFunction = -1;

  // Throws a parsing error if we are not in a function context
  void requireFunctionContext(const char* error);

  void readFunctions();
  void readVars();
  void setLocalNames(Function& func, Index i);

  Result<> readInst();

  void readExports();

  // The strings in the strings section (which are referred to by StringConst).
  std::vector<Name> strings;
  void readStrings();
  Name getIndexedString();

  Expression* readExpression();
  void readGlobals();

  // validations that cannot be performed on the Module
  void validateBinary();

  size_t dataCount = 0;
  bool hasDataCount = false;

  void createDataSegments(Index count);
  void readDataSegments();
  void readDataSegmentCount();

  void readTableDeclarations();
  void readElementSegments();

  void readTags();

  static Name escape(Name name);
  void readNames(size_t sectionPos, size_t payloadLen);
  void readFeatures(size_t sectionPos, size_t payloadLen);
  void readDylink(size_t payloadLen);
  void readDylink0(size_t payloadLen);

  // We read code annotations *after* the code section, even though they appear
  // earlier. That is simpler for us as we note expression locations as we scan
  // code, and then just need to match them up. To do this, we note the
  // positions of annotation sections in the first pass, and handle them later.
  struct AnnotationSectionInfo {
    // The start position of the section. We will rewind to there to read it.
    size_t pos;
    // A lambda that will read the section, from that position.
    std::function<void()> read;
  };
  std::vector<AnnotationSectionInfo> deferredAnnotationSections;

  void readBranchHints(size_t payloadLen);
  void readInlineHints(size_t payloadLen);
  void readRemovableIfUnusedHints(size_t payloadLen);
  void readJSCalledHints(size_t payloadLen);
  void readIdempotentHints(size_t payloadLen);

  std::tuple<Address, Address, Index, MemoryOrder>
  readMemoryAccess(bool isAtomic, bool isRMW);
  std::tuple<Name, Address, Address, MemoryOrder> getAtomicMemarg();
  std::tuple<Name, Address, Address, MemoryOrder> getRMWMemarg();
  std::tuple<Name, Address, Address> getMemarg();
  MemoryOrder getMemoryOrder(bool isRMW = false);

  [[noreturn]] void throwError(std::string text) {
    throw ParseException(text, 0, pos);
  }

  // Allow users to query the target features section features after parsing.
  FeatureSet getFeaturesSectionFeatures() { return featuresSectionFeatures; }

private:
  // In certain modes we need to note the locations of expressions, to match
  // them against sections like DWARF or custom annotations. As this incurs
  // overhead, we only note locations when we actually need to.
  bool needCodeLocations = false;

  // The features enabled by the target features section, which may be a subset
  // of the features enabled for the module.
  FeatureSet featuresSectionFeatures;

  // Scans ahead in the binary to check certain conditions like
  // needCodeLocations.
  void preScan();

  // Internal helper for reading a code annotation section for a hint that is
  // expression offset based. Receives the section name, payload length of the
  // section and a function to read a single hint (receiving the annotation to
  // update).
  template<typename ReadFunc>
  void readExpressionHints(Name sectionName, size_t payloadLen, ReadFunc read);
};

} // namespace wasm

#undef DEBUG_TYPE

#endif // wasm_wasm_binary_h