deflateencoder.cpp

7-Zip 是一款号称有着现今最高压缩比的压缩软件

      minCount = 4;
    }
  }
}

NO_INLINE void CCoder::WriteBits(UInt32 value, int numBits)
{
  m_OutStream.WriteBits(value, numBits);
}

#define WRITE_HF2(codes, lens, i) m_OutStream.WriteBits(codes[i], lens[i])
#define WRITE_HF(i) WriteBits(codes[i], lens[i])

NO_INLINE void CCoder::LevelTableCode(const Byte *levels, int numLevels, const Byte *lens, const UInt32 *codes)
{
  int prevLen = 0xFF;
  int nextLen = levels[0];
  int count = 0;
  int maxCount = 7;
  int minCount = 4;
  if (nextLen == 0)
  {
    maxCount = 138;
    minCount = 3;
  }
  for (int n = 0; n < numLevels; n++)
  {
    int curLen = nextLen;
    nextLen = (n < numLevels - 1) ? levels[n + 1] : 0xFF;
    count++;
    if (count < maxCount && curLen == nextLen)
      continue;
    
    if (count < minCount)
      for(int i = 0; i < count; i++)
        WRITE_HF(curLen);
    else if (curLen != 0)
    {
      if (curLen != prevLen)
      {
        WRITE_HF(curLen);
        count--;
      }
      WRITE_HF(kTableLevelRepNumber);
      WriteBits(count - 3, 2);
    }
    else if (count <= 10)
    {
      WRITE_HF(kTableLevel0Number);
      WriteBits(count - 3, 3);
    }
    else
    {
      WRITE_HF(kTableLevel0Number2);
      WriteBits(count - 11, 7);
    }

    count = 0;
    prevLen = curLen;
    
    if (nextLen == 0)
    {
      maxCount = 138;
      minCount = 3;
    }
    else if (curLen == nextLen)
    {
      maxCount = 6;
      minCount = 3;
    }
    else
    {
      maxCount = 7;
      minCount = 4;
    }
  }
}

NO_INLINE void CCoder::MakeTables(unsigned maxHuffLen)
{
  Huffman_Generate(mainFreqs, mainCodes, m_NewLevels.litLenLevels, kFixedMainTableSize, maxHuffLen);
  Huffman_Generate(distFreqs, distCodes, m_NewLevels.distLevels, kDistTableSize64, maxHuffLen);
}

NO_INLINE UInt32 Huffman_GetPrice(const UInt32 *freqs, const Byte *lens, UInt32 num)
{
  UInt32 price = 0;
  UInt32 i;
  for (i = 0; i < num; i++)
    price += lens[i] * freqs[i];
  return price;
};

NO_INLINE UInt32 Huffman_GetPrice_Spec(const UInt32 *freqs, const Byte *lens, UInt32 num, const Byte *extraBits, UInt32 extraBase)
{
  return Huffman_GetPrice(freqs, lens, num) +
    Huffman_GetPrice(freqs + extraBase, extraBits, num - extraBase);
}

NO_INLINE UInt32 CCoder::GetLzBlockPrice() const
{
  return
    Huffman_GetPrice_Spec(mainFreqs, m_NewLevels.litLenLevels, kFixedMainTableSize, m_LenDirectBits, kSymbolMatch) +
    Huffman_GetPrice_Spec(distFreqs, m_NewLevels.distLevels, kDistTableSize64, kDistDirectBits, 0);
}

NO_INLINE void CCoder::TryBlock()
{
  memset(mainFreqs, 0, sizeof(mainFreqs));
  memset(distFreqs, 0, sizeof(distFreqs));

  m_ValueIndex = 0;
  UInt32 blockSize = BlockSizeRes;
  BlockSizeRes = 0;
  for (;;)
  {
    if (m_OptimumCurrentIndex == m_OptimumEndIndex)
    {
      if (m_Pos >= kMatchArrayLimit || BlockSizeRes >= blockSize || !m_SecondPass &&
          ((Inline_MatchFinder_GetNumAvailableBytes(&_lzInWindow) == 0) || m_ValueIndex >= m_ValueBlockSize))
        break;
    }
    UInt32 pos;
    UInt32 len;
    if (_fastMode)
      len = GetOptimalFast(pos);
    else
      len = GetOptimal(pos);
    CCodeValue &codeValue = m_Values[m_ValueIndex++];
    if (len >= kMatchMinLen)
    {
      UInt32 newLen = len - kMatchMinLen;
      codeValue.Len = (UInt16)newLen;
      mainFreqs[kSymbolMatch + g_LenSlots[newLen]]++;
      codeValue.Pos = (UInt16)pos;
      distFreqs[GetPosSlot(pos)]++;
    }
    else
    {
      Byte b = Inline_MatchFinder_GetIndexByte(&_lzInWindow, 0 - m_AdditionalOffset);
      mainFreqs[b]++;
      codeValue.SetAsLiteral();
      codeValue.Pos = b;
    }
    m_AdditionalOffset -= len;
    BlockSizeRes += len;
  }
  mainFreqs[kSymbolEndOfBlock]++;
  m_AdditionalOffset += BlockSizeRes;
  m_SecondPass = true;
}

NO_INLINE void CCoder::SetPrices(const CLevels &levels)
{
  if (_fastMode)
    return;
  UInt32 i;
  for(i = 0; i < 256; i++)
  {
    Byte price = levels.litLenLevels[i];
    m_LiteralPrices[i] = ((price != 0) ? price : kNoLiteralStatPrice);
  }
  
  for(i = 0; i < m_NumLenCombinations; i++)
  {
    UInt32 slot = g_LenSlots[i];
    Byte price = levels.litLenLevels[kSymbolMatch + slot];
    m_LenPrices[i] = (Byte)(((price != 0) ? price : kNoLenStatPrice) + m_LenDirectBits[slot]);
  }
  
  for(i = 0; i < kDistTableSize64; i++)
  {
    Byte price = levels.distLevels[i];
    m_PosPrices[i] = (Byte)(((price != 0) ? price: kNoPosStatPrice) + kDistDirectBits[i]);
  }
}

NO_INLINE void Huffman_ReverseBits(UInt32 *codes, const Byte *lens, UInt32 num)
{
  for (UInt32 i = 0; i < num; i++)
  {
    UInt32 x = codes[i];
    x = ((x & 0x5555) << 1) | ((x & 0xAAAA) >> 1);
    x = ((x & 0x3333) << 2) | ((x & 0xCCCC) >> 2);
    x = ((x & 0x0F0F) << 4) | ((x & 0xF0F0) >> 4);
    codes[i] = (((x & 0x00FF) << 8) | ((x & 0xFF00) >> 8)) >> (16 - lens[i]);
  }
}

NO_INLINE void CCoder::WriteBlock()
{
  Huffman_ReverseBits(mainCodes, m_NewLevels.litLenLevels, kFixedMainTableSize);
  Huffman_ReverseBits(distCodes, m_NewLevels.distLevels, kDistTableSize64);

  for (UInt32 i = 0; i < m_ValueIndex; i++)
  {
    const CCodeValue &codeValue = m_Values[i];
    if (codeValue.IsLiteral())
      WRITE_HF2(mainCodes, m_NewLevels.litLenLevels, codeValue.Pos);
    else
    {
      UInt32 len = codeValue.Len;
      UInt32 lenSlot = g_LenSlots[len];
      WRITE_HF2(mainCodes, m_NewLevels.litLenLevels, kSymbolMatch + lenSlot);
      m_OutStream.WriteBits(len - m_LenStart[lenSlot], m_LenDirectBits[lenSlot]);
      UInt32 dist = codeValue.Pos;
      UInt32 posSlot = GetPosSlot(dist);
      WRITE_HF2(distCodes, m_NewLevels.distLevels, posSlot);
      m_OutStream.WriteBits(dist - kDistStart[posSlot], kDistDirectBits[posSlot]);
    }
  }
  WRITE_HF2(mainCodes, m_NewLevels.litLenLevels, kSymbolEndOfBlock);
}

static UInt32 GetStorePrice(UInt32 blockSize, int bitPosition)
{
  UInt32 price = 0;
  do
  {
    UInt32 nextBitPosition = (bitPosition + kFinalBlockFieldSize + kBlockTypeFieldSize) & 7;
    int numBitsForAlign = nextBitPosition > 0 ? (8 - nextBitPosition): 0;
    UInt32 curBlockSize = (blockSize < (1 << 16)) ? blockSize : (1 << 16) - 1;
    price += kFinalBlockFieldSize + kBlockTypeFieldSize + numBitsForAlign + (2 + 2) * 8 + curBlockSize * 8;
    bitPosition = 0;
    blockSize -= curBlockSize;
  }
  while(blockSize != 0);
  return price;
}

void CCoder::WriteStoreBlock(UInt32 blockSize, UInt32 additionalOffset, bool finalBlock)
{
  do
  {
    UInt32 curBlockSize = (blockSize < (1 << 16)) ? blockSize : (1 << 16) - 1;
    blockSize -= curBlockSize;
    WriteBits((finalBlock && (blockSize == 0) ? NFinalBlockField::kFinalBlock: NFinalBlockField::kNotFinalBlock), kFinalBlockFieldSize);
    WriteBits(NBlockType::kStored, kBlockTypeFieldSize);
    m_OutStream.FlushByte();
    WriteBits((UInt16)curBlockSize, kStoredBlockLengthFieldSize);
    WriteBits((UInt16)~curBlockSize, kStoredBlockLengthFieldSize);
    const Byte *data = Inline_MatchFinder_GetPointerToCurrentPos(&_lzInWindow)- additionalOffset;
    for(UInt32 i = 0; i < curBlockSize; i++)
      m_OutStream.WriteByte(data[i]);
    additionalOffset -= curBlockSize;
  }
  while(blockSize != 0);
}

NO_INLINE UInt32 CCoder::TryDynBlock(int tableIndex, UInt32 numPasses)
{
  CTables &t = m_Tables[tableIndex];
  BlockSizeRes = t.BlockSizeRes;
  UInt32 posTemp = t.m_Pos;
  SetPrices(t);

  for (UInt32 p = 0; p < numPasses; p++)
  {
    m_Pos = posTemp;
    TryBlock();
    unsigned numHuffBits =
        (m_ValueIndex > 18000 ? 12 :
        (m_ValueIndex >  7000 ? 11 :
        (m_ValueIndex >  2000 ? 10 : 9)));
    MakeTables(numHuffBits);
    SetPrices(m_NewLevels);
  }

  (CLevels &)t = m_NewLevels;

  m_NumLitLenLevels = kMainTableSize;
  while(m_NumLitLenLevels > kNumLitLenCodesMin && m_NewLevels.litLenLevels[m_NumLitLenLevels - 1] == 0)
    m_NumLitLenLevels--;
  
  m_NumDistLevels = kDistTableSize64;
  while(m_NumDistLevels > kNumDistCodesMin && m_NewLevels.distLevels[m_NumDistLevels - 1] == 0)
    m_NumDistLevels--;
  
  UInt32 levelFreqs[kLevelTableSize];
  memset(levelFreqs, 0, sizeof(levelFreqs));

  LevelTableDummy(m_NewLevels.litLenLevels, m_NumLitLenLevels, levelFreqs);
  LevelTableDummy(m_NewLevels.distLevels, m_NumDistLevels, levelFreqs);
  
  Huffman_Generate(levelFreqs, levelCodes, levelLens, kLevelTableSize, kMaxLevelBitLength);
  
  m_NumLevelCodes = kNumLevelCodesMin;
  for (UInt32 i = 0; i < kLevelTableSize; i++)
  {
    Byte level = levelLens[kCodeLengthAlphabetOrder[i]];
    if (level > 0 && i >= m_NumLevelCodes)
      m_NumLevelCodes = i + 1;
    m_LevelLevels[i] = level;
  }
  
  return GetLzBlockPrice() +
      Huffman_GetPrice_Spec(levelFreqs, levelLens, kLevelTableSize, kLevelDirectBits, kTableDirectLevels) +
      kNumLenCodesFieldSize + kNumDistCodesFieldSize + kNumLevelCodesFieldSize +
      m_NumLevelCodes * kLevelFieldSize + kFinalBlockFieldSize + kBlockTypeFieldSize;
}

NO_INLINE UInt32 CCoder::TryFixedBlock(int tableIndex)
{
  CTables &t = m_Tables[tableIndex];
  BlockSizeRes = t.BlockSizeRes;
  m_Pos = t.m_Pos;
  m_NewLevels.SetFixedLevels();
  SetPrices(m_NewLevels);
  TryBlock();
  return kFinalBlockFieldSize + kBlockTypeFieldSize + GetLzBlockPrice();
}

NO_INLINE UInt32 CCoder::GetBlockPrice(int tableIndex, int numDivPasses)
{
  CTables &t = m_Tables[tableIndex];
  t.StaticMode = false;
  UInt32 price = TryDynBlock(tableIndex, m_NumPasses);
  t.BlockSizeRes = BlockSizeRes;
  UInt32 numValues = m_ValueIndex;
  UInt32 posTemp = m_Pos;
  UInt32 additionalOffsetEnd = m_AdditionalOffset;
  
  if (m_CheckStatic && m_ValueIndex <= kFixedHuffmanCodeBlockSizeMax)
  {
    const UInt32 fixedPrice = TryFixedBlock(tableIndex);
    t.StaticMode = (fixedPrice < price);
    if (t.StaticMode)
      price = fixedPrice;
  }

  const UInt32 storePrice = GetStorePrice(BlockSizeRes, 0); // bitPosition
  t.StoreMode = (storePrice <= price);
  if (t.StoreMode)
    price = storePrice;

  t.UseSubBlocks = false;

  if (numDivPasses > 1 && numValues >= kDivideCodeBlockSizeMin)
  {
    CTables &t0 = m_Tables[(tableIndex << 1)];
    (CLevels &)t0 = t;
    t0.BlockSizeRes = t.BlockSizeRes >> 1;
    t0.m_Pos = t.m_Pos;
    UInt32 subPrice = GetBlockPrice((tableIndex << 1), numDivPasses - 1);

    UInt32 blockSize2 = t.BlockSizeRes - t0.BlockSizeRes;
    if (t0.BlockSizeRes >= kDivideBlockSizeMin && blockSize2 >= kDivideBlockSizeMin)
    {
      CTables &t1 = m_Tables[(tableIndex << 1) + 1];
      (CLevels &)t1 = t;
      t1.BlockSizeRes = blockSize2;
      t1.m_Pos = m_Pos;
      m_AdditionalOffset -= t0.BlockSizeRes;
      subPrice += GetBlockPrice((tableIndex << 1) + 1, numDivPasses - 1);
      t.UseSubBlocks = (subPrice < price);
      if (t.UseSubBlocks)
        price = subPrice;
    }
  }
  m_AdditionalOffset = additionalOffsetEnd;
  m_Pos = posTemp;
  return price;
}

void CCoder::CodeBlock(int tableIndex, bool finalBlock)
{
  CTables &t = m_Tables[tableIndex];
  if (t.UseSubBlocks)
  {
    CodeBlock((tableIndex << 1), false);
    CodeBlock((tableIndex << 1) + 1, finalBlock);
  }
  else
  {
    if (t.StoreMode)
      WriteStoreBlock(t.BlockSizeRes, m_AdditionalOffset, finalBlock);
    else
    {
      WriteBits((finalBlock ? NFinalBlockField::kFinalBlock: NFinalBlockField::kNotFinalBlock), kFinalBlockFieldSize);
      if (t.StaticMode)
      {
        WriteBits(NBlockType::kFixedHuffman, kBlockTypeFieldSize);
        TryFixedBlock(tableIndex);
        int i;
        const int kMaxStaticHuffLen = 9;
        for (i = 0; i < kFixedMainTableSize; i++)
          mainFreqs[i] = (UInt32)1 << (kMaxStaticHuffLen - m_NewLevels.litLenLevels[i]);
        for (i = 0; i < kFixedDistTableSize; i++)
          distFreqs[i] = (UInt32)1 << (kMaxStaticHuffLen - m_NewLevels.distLevels[i]);
        MakeTables(kMaxStaticHuffLen);
      }
      else
      {
        if (m_NumDivPasses > 1 || m_CheckStatic)
          TryDynBlock(tableIndex, 1);
        WriteBits(NBlockType::kDynamicHuffman, kBlockTypeFieldSize);
        WriteBits(m_NumLitLenLevels - kNumLitLenCodesMin, kNumLenCodesFieldSize);
        WriteBits(m_NumDistLevels - kNumDistCodesMin, kNumDistCodesFieldSize);
        WriteBits(m_NumLevelCodes - kNumLevelCodesMin, kNumLevelCodesFieldSize);
        
        for (UInt32 i = 0; i < m_NumLevelCodes; i++)
          WriteBits(m_LevelLevels[i], kLevelFieldSize);
        
        Huffman_ReverseBits(levelCodes, levelLens, kLevelTableSize);
        LevelTableCode(m_NewLevels.litLenLevels, m_NumLitLenLevels, levelLens, levelCodes);
        LevelTableCode(m_NewLevels.distLevels, m_NumDistLevels, levelLens, levelCodes);
      }
      WriteBlock();
    }
    m_AdditionalOffset -= t.BlockSizeRes;
  }
}

SRes Read(void *object, void *data, size_t *size)
{
  const UInt32 kStepSize = (UInt32)1 << 31;
  UInt32 curSize = ((*size < kStepSize) ? (UInt32)*size : kStepSize);
  HRESULT res = ((CSeqInStream *)object)->RealStream->Read(data, curSize, &curSize);
  *size = curSize;
  return (SRes)res;
}

HRESULT CCoder::CodeReal(ISequentialInStream *inStream,
    ISequentialOutStream *outStream, const UInt64 * /* inSize */ , const UInt64 * /* outSize */ ,
    ICompressProgressInfo *progress)
{
  m_CheckStatic = (m_NumPasses != 1 || m_NumDivPasses != 1);
  m_IsMultiPass = (m_CheckStatic || (m_NumPasses != 1 || m_NumDivPasses != 1));

  RINOK(Create());

  m_ValueBlockSize = (7 << 10) + (1 << 12) * m_NumDivPasses;

  UInt64 nowPos = 0;

  _seqInStream.RealStream = inStream;
  _seqInStream.SeqInStream.Read = Read;
  _lzInWindow.stream = &_seqInStream.SeqInStream;

  MatchFinder_Init(&_lzInWindow);
  m_OutStream.SetStream(outStream);
  m_OutStream.Init();

  CCoderReleaser coderReleaser(this);

  m_OptimumEndIndex = m_OptimumCurrentIndex = 0;

  CTables &t = m_Tables[1];
  t.m_Pos = 0;
  t.InitStructures();

  m_AdditionalOffset = 0;
  do
  {
    t.BlockSizeRes = kBlockUncompressedSizeThreshold;
    m_SecondPass = false;
    GetBlockPrice(1, m_NumDivPasses);
    CodeBlock(1, Inline_MatchFinder_GetNumAvailableBytes(&_lzInWindow) == 0);
    nowPos += m_Tables[1].BlockSizeRes;
    if (progress != NULL)
    {
      UInt64 packSize = m_OutStream.GetProcessedSize();
      RINOK(progress->SetRatioInfo(&nowPos, &packSize));
    }
  }
  while (Inline_MatchFinder_GetNumAvailableBytes(&_lzInWindow) != 0);
  if (_lzInWindow.result != SZ_OK)
    return _lzInWindow.result;
  return m_OutStream.Flush();
}

HRESULT CCoder::BaseCode(ISequentialInStream *inStream,
    ISequentialOutStream *outStream, const UInt64 *inSize, const UInt64 *outSize,
    ICompressProgressInfo *progress)
{
  try { return CodeReal(inStream, outStream, inSize, outSize, progress); }
  catch(const COutBufferException &e) { return e.ErrorCode; }
  catch(...) { return E_FAIL; }
}

STDMETHODIMP CCOMCoder::Code(ISequentialInStream *inStream,
    ISequentialOutStream *outStream, const UInt64 *inSize, const UInt64 *outSize,
    ICompressProgressInfo *progress)
  { return BaseCode(inStream, outStream, inSize, outSize, progress); }

STDMETHODIMP CCOMCoder::SetCoderProperties(const PROPID *propIDs,
    const PROPVARIANT *properties, UInt32 numProperties)
  { return BaseSetEncoderProperties2(propIDs, properties, numProperties); }

STDMETHODIMP CCOMCoder64::Code(ISequentialInStream *inStream,
    ISequentialOutStream *outStream, const UInt64 *inSize, const UInt64 *outSize,
    ICompressProgressInfo *progress)
  { return BaseCode(inStream, outStream, inSize, outSize, progress); }

STDMETHODIMP CCOMCoder64::SetCoderProperties(const PROPID *propIDs,
    const PROPVARIANT *properties, UInt32 numProperties)
  { return BaseSetEncoderProperties2(propIDs, properties, numProperties); }

}}}