/* LzmaEnc.c -- LZMA Encoder
2009-02-02 : Igor Pavlov : Public domain */

#include <string.h>

/* #define SHOW_STAT */
/* #define SHOW_STAT2 */

#if defined(SHOW_STAT) || defined(SHOW_STAT2)
#include <stdio.h>
#endif

#include "LzmaEnc.h"

#include "LzFind.h"
#ifdef COMPRESS_MF_MT
#include "LzFindMt.h"
#endif

#ifdef SHOW_STAT
static int ttt = 0;
#endif

#define kBlockSizeMax ((1 << LZMA_NUM_BLOCK_SIZE_BITS) - 1)

#define kBlockSize (9 << 10)
#define kUnpackBlockSize (1 << 18)
#define kMatchArraySize (1 << 21)
#define kMatchRecordMaxSize ((LZMA_MATCH_LEN_MAX * 2 + 3) * LZMA_MATCH_LEN_MAX)

#define kNumMaxDirectBits (31)

#define kNumTopBits 24
#define kTopValue ((UInt32)1 << kNumTopBits)

#define kNumBitModelTotalBits 11
#define kBitModelTotal (1 << kNumBitModelTotalBits)
#define kNumMoveBits 5
#define kProbInitValue (kBitModelTotal >> 1)

#define kNumMoveReducingBits 4
#define kNumBitPriceShiftBits 4
#define kBitPrice (1 << kNumBitPriceShiftBits)

void LzmaEncProps_Init(CLzmaEncProps *p)
{
  p->level = 5;
  p->dictSize = p->mc = 0;
  p->lc = p->lp = p->pb = p->algo = p->fb = p->btMode = p->numHashBytes = p->numThreads = -1;
  p->writeEndMark = 0;
}

void LzmaEncProps_Normalize(CLzmaEncProps *p)
{
  int level = p->level;
  if (level < 0) level = 5;
  p->level = level;
  if (p->dictSize == 0) p->dictSize = (level <= 5 ? (1 << (level * 2 + 14)) : (level == 6 ? (1 << 25) : (1 << 26)));
  if (p->lc < 0) p->lc = 3;
  if (p->lp < 0) p->lp = 0;
  if (p->pb < 0) p->pb = 2;
  if (p->algo < 0) p->algo = (level < 5 ? 0 : 1);
  if (p->fb < 0) p->fb = (level < 7 ? 32 : 64);
  if (p->btMode < 0) p->btMode = (p->algo == 0 ? 0 : 1);
  if (p->numHashBytes < 0) p->numHashBytes = 4;
  if (p->mc == 0)  p->mc = (16 + (p->fb >> 1)) >> (p->btMode ? 0 : 1);
  if (p->numThreads < 0)
    p->numThreads =
      #ifdef COMPRESS_MF_MT
      ((p->btMode && p->algo) ? 2 : 1);
      #else
      1;
      #endif
}

UInt32 LzmaEncProps_GetDictSize(const CLzmaEncProps *props2)
{
  CLzmaEncProps props = *props2;
  LzmaEncProps_Normalize(&props);
  return props.dictSize;
}

/* #define LZMA_LOG_BSR */
/* Define it for Intel's CPU */


#ifdef LZMA_LOG_BSR

#define kDicLogSizeMaxCompress 30

#define BSR2_RET(pos, res) { unsigned long i; _BitScanReverse(&i, (pos)); res = (i + i) + ((pos >> (i - 1)) & 1); }

UInt32 GetPosSlot1(UInt32 pos)
{
  UInt32 res;
  BSR2_RET(pos, res);
  return res;
}
#define GetPosSlot2(pos, res) { BSR2_RET(pos, res); }
#define GetPosSlot(pos, res) { if (pos < 2) res = pos; else BSR2_RET(pos, res); }

#else

#define kNumLogBits (9 + (int)sizeof(size_t) / 2)
#define kDicLogSizeMaxCompress ((kNumLogBits - 1) * 2 + 7)

static void LzmaEnc_FastPosInit(Byte *g_FastPos)
{
  int c = 2, slotFast;
  g_FastPos[0] = 0;
  g_FastPos[1] = 1;
  
  for (slotFast = 2; slotFast < kNumLogBits * 2; slotFast++)
  {
    UInt32 k = (1 << ((slotFast >> 1) - 1));
    UInt32 j;
    for (j = 0; j < k; j++, c++)
      g_FastPos[c] = (Byte)slotFast;
  }
}

#define BSR2_RET(pos, res) { UInt32 i = 6 + ((kNumLogBits - 1) & \
  (0 - (((((UInt32)1 << (kNumLogBits + 6)) - 1) - pos) >> 31))); \
  res = p->g_FastPos[pos >> i] + (i * 2); }
/*
#define BSR2_RET(pos, res) { res = (pos < (1 << (kNumLogBits + 6))) ? \
  p->g_FastPos[pos >> 6] + 12 : \
  p->g_FastPos[pos >> (6 + kNumLogBits - 1)] + (6 + (kNumLogBits - 1)) * 2; }
*/

#define GetPosSlot1(pos) p->g_FastPos[pos]
#define GetPosSlot2(pos, res) { BSR2_RET(pos, res); }
#define GetPosSlot(pos, res) { if (pos < kNumFullDistances) res = p->g_FastPos[pos]; else BSR2_RET(pos, res); }

#endif


#define LZMA_NUM_REPS 4

typedef unsigned CState;

typedef struct _COptimal
{
  UInt32 price;

  CState state;
  int prev1IsChar;
  int prev2;

  UInt32 posPrev2;
  UInt32 backPrev2;

  UInt32 posPrev;
  UInt32 backPrev;
  UInt32 backs[LZMA_NUM_REPS];
} COptimal;

#define kNumOpts (1 << 12)

#define kNumLenToPosStates 4
#define kNumPosSlotBits 6
#define kDicLogSizeMin 0
#define kDicLogSizeMax 32
#define kDistTableSizeMax (kDicLogSizeMax * 2)


#define kNumAlignBits 4
#define kAlignTableSize (1 << kNumAlignBits)
#define kAlignMask (kAlignTableSize - 1)

#define kStartPosModelIndex 4
#define kEndPosModelIndex 14
#define kNumPosModels (kEndPosModelIndex - kStartPosModelIndex)

#define kNumFullDistances (1 << (kEndPosModelIndex / 2))

#ifdef _LZMA_PROB32
#define CLzmaProb UInt32
#else
#define CLzmaProb UInt16
#endif

#define LZMA_PB_MAX 4
#define LZMA_LC_MAX 8
#define LZMA_LP_MAX 4

#define LZMA_NUM_PB_STATES_MAX (1 << LZMA_PB_MAX)


#define kLenNumLowBits 3
#define kLenNumLowSymbols (1 << kLenNumLowBits)
#define kLenNumMidBits 3
#define kLenNumMidSymbols (1 << kLenNumMidBits)
#define kLenNumHighBits 8
#define kLenNumHighSymbols (1 << kLenNumHighBits)

#define kLenNumSymbolsTotal (kLenNumLowSymbols + kLenNumMidSymbols + kLenNumHighSymbols)

#define LZMA_MATCH_LEN_MIN 2
#define LZMA_MATCH_LEN_MAX (LZMA_MATCH_LEN_MIN + kLenNumSymbolsTotal - 1)

#define kNumStates 12

typedef struct
{
  CLzmaProb choice;
  CLzmaProb choice2;
  CLzmaProb low[LZMA_NUM_PB_STATES_MAX << kLenNumLowBits];
  CLzmaProb mid[LZMA_NUM_PB_STATES_MAX << kLenNumMidBits];
  CLzmaProb high[kLenNumHighSymbols];
} CLenEnc;

typedef struct
{
  CLenEnc p;
  UInt32 prices[LZMA_NUM_PB_STATES_MAX][kLenNumSymbolsTotal];
  UInt32 tableSize;
  UInt32 counters[LZMA_NUM_PB_STATES_MAX];
} CLenPriceEnc;

typedef struct _CRangeEnc
{
  UInt32 range;
  Byte cache;
  UInt64 low;
  UInt64 cacheSize;
  Byte *buf;
  Byte *bufLim;
  Byte *bufBase;
  ISeqOutStream *outStream;
  UInt64 processed;
  SRes res;
} CRangeEnc;

typedef struct _CSeqInStreamBuf
{
  ISeqInStream funcTable;
  const Byte *data;
  SizeT rem;
} CSeqInStreamBuf;

static SRes MyRead(void *pp, void *data, size_t *size)
{
  size_t curSize = *size;
  CSeqInStreamBuf *p = (CSeqInStreamBuf *)pp;
  if (p->rem < curSize)
    curSize = p->rem;
  memcpy(data, p->data, curSize);
  p->rem -= curSize;
  p->data += curSize;
  *size = curSize;
  return SZ_OK;
}

typedef struct
{
  CLzmaProb *litProbs;

  CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX];
  CLzmaProb isRep[kNumStates];
  CLzmaProb isRepG0[kNumStates];
  CLzmaProb isRepG1[kNumStates];
  CLzmaProb isRepG2[kNumStates];
  CLzmaProb isRep0Long[kNumStates][LZMA_NUM_PB_STATES_MAX];

  CLzmaProb posSlotEncoder[kNumLenToPosStates][1 << kNumPosSlotBits];
  CLzmaProb posEncoders[kNumFullDistances - kEndPosModelIndex];
  CLzmaProb posAlignEncoder[1 << kNumAlignBits];
  
  CLenPriceEnc lenEnc;
  CLenPriceEnc repLenEnc;

  UInt32 reps[LZMA_NUM_REPS];
  UInt32 state;
} CSaveState;

typedef struct _CLzmaEnc
{
  IMatchFinder matchFinder;
  void *matchFinderObj;

  #ifdef COMPRESS_MF_MT
  Bool mtMode;
  CMatchFinderMt matchFinderMt;
  #endif

  CMatchFinder matchFinderBase;

  #ifdef COMPRESS_MF_MT
  Byte pad[128];
  #endif
  
  UInt32 optimumEndIndex;
  UInt32 optimumCurrentIndex;

  UInt32 longestMatchLength;
  UInt32 numPairs;
  UInt32 numAvail;
  COptimal opt[kNumOpts];
  
  #ifndef LZMA_LOG_BSR
  Byte g_FastPos[1 << kNumLogBits];
  #endif

  UInt32 ProbPrices[kBitModelTotal >> kNumMoveReducingBits];
  UInt32 matches[LZMA_MATCH_LEN_MAX * 2 + 2 + 1];
  UInt32 numFastBytes;
  UInt32 additionalOffset;
  UInt32 reps[LZMA_NUM_REPS];
  UInt32 state;

  UInt32 posSlotPrices[kNumLenToPosStates][kDistTableSizeMax];
  UInt32 distancesPrices[kNumLenToPosStates][kNumFullDistances];
  UInt32 alignPrices[kAlignTableSize];
  UInt32 alignPriceCount;

  UInt32 distTableSize;

  unsigned lc, lp, pb;
  unsigned lpMask, pbMask;

  CLzmaProb *litProbs;

  CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX];
  CLzmaProb isRep[kNumStates];
  CLzmaProb isRepG0[kNumStates];
  CLzmaProb isRepG1[kNumStates];
  CLzmaProb isRepG2[kNumStates];
  CLzmaProb isRep0Long[kNumStates][LZMA_NUM_PB_STATES_MAX];

  CLzmaProb posSlotEncoder[kNumLenToPosStates][1 << kNumPosSlotBits];
  CLzmaProb posEncoders[kNumFullDistances - kEndPosModelIndex];
  CLzmaProb posAlignEncoder[1 << kNumAlignBits];
  
  CLenPriceEnc lenEnc;
  CLenPriceEnc repLenEnc;

  unsigned lclp;

  Bool fastMode;
  
  CRangeEnc rc;

  Bool writeEndMark;
  UInt64 nowPos64;
  UInt32 matchPriceCount;
  Bool finished;
  Bool multiThread;

  SRes result;
  UInt32 dictSize;
  UInt32 matchFinderCycles;

  ISeqInStream *inStream;
  CSeqInStreamBuf seqBufInStream;

  CSaveState saveState;
} CLzmaEnc;

/*static void LzmaEnc_SaveState(CLzmaEncHandle pp)
{
  CLzmaEnc *p = (CLzmaEnc *)pp;
  CSaveState *dest = &p->saveState;
  int i;
  dest->lenEnc = p->lenEnc;
  dest->repLenEnc = p->repLenEnc;
  dest->state = p->state;

  for (i = 0; i < kNumStates; i++)
  {
    memcpy(dest->isMatch[i], p->isMatch[i], sizeof(p->isMatch[i]));
    memcpy(dest->isRep0Long[i], p->isRep0Long[i], sizeof(p->isRep0Long[i]));
  }
  for (i = 0; i < kNumLenToPosStates; i++)
    memcpy(dest->posSlotEncoder[i], p->posSlotEncoder[i], sizeof(p->posSlotEncoder[i]));
  memcpy(dest->isRep, p->isRep, sizeof(p->isRep));
  memcpy(dest->isRepG0, p->isRepG0, sizeof(p->isRepG0));
  memcpy(dest->isRepG1, p->isRepG1, sizeof(p->isRepG1));
  memcpy(dest->isRepG2, p->isRepG2, sizeof(p->isRepG2));
  memcpy(dest->posEncoders, p->posEncoders, sizeof(p->posEncoders));
  memcpy(dest->posAlignEncoder, p->posAlignEncoder, sizeof(p->posAlignEncoder));
  memcpy(dest->reps, p->reps, sizeof(p->reps));
  memcpy(dest->litProbs, p->litProbs, (0x300 << p->lclp) * sizeof(CLzmaProb));
}*/

/*static void LzmaEnc_RestoreState(CLzmaEncHandle pp)
{
  CLzmaEnc *dest = (CLzmaEnc *)pp;
  const CSaveState *p = &dest->saveState;
  int i;
  dest->lenEnc = p->lenEnc;
  dest->repLenEnc = p->repLenEnc;
  dest->state = p->state;

  for (i = 0; i < kNumStates; i++)
  {
    memcpy(dest->isMatch[i], p->isMatch[i], sizeof(p->isMatch[i]));
    memcpy(dest->isRep0Long[i], p->isRep0Long[i], sizeof(p->isRep0Long[i]));
  }
  for (i = 0; i < kNumLenToPosStates; i++)
    memcpy(dest->posSlotEncoder[i], p->posSlotEncoder[i], sizeof(p->posSlotEncoder[i]));
  memcpy(dest->isRep, p->isRep, sizeof(p->isRep));
  memcpy(dest->isRepG0, p->isRepG0, sizeof(p->isRepG0));
  memcpy(dest->isRepG1, p->isRepG1, sizeof(p->isRepG1));
  memcpy(dest->isRepG2, p->isRepG2, sizeof(p->isRepG2));
  memcpy(dest->posEncoders, p->posEncoders, sizeof(p->posEncoders));
  memcpy(dest->posAlignEncoder, p->posAlignEncoder, sizeof(p->posAlignEncoder));
  memcpy(dest->reps, p->reps, sizeof(p->reps));
  memcpy(dest->litProbs, p->litProbs, (0x300 << dest->lclp) * sizeof(CLzmaProb));
}*/

SRes LzmaEnc_SetProps(CLzmaEncHandle pp, const CLzmaEncProps *props2)
{
  CLzmaEnc *p = (CLzmaEnc *)pp;
  CLzmaEncProps props = *props2;
  LzmaEncProps_Normalize(&props);

  if (props.lc > LZMA_LC_MAX || props.lp > LZMA_LP_MAX || props.pb > LZMA_PB_MAX ||
      props.dictSize > (1 << kDicLogSizeMaxCompress) || props.dictSize > (1 << 30))
    return SZ_ERROR_PARAM;
  p->dictSize = props.dictSize;
  p->matchFinderCycles = props.mc;
  {
    unsigned fb = props.fb;
    if (fb < 5)
      fb = 5;
    if (fb > LZMA_MATCH_LEN_MAX)
      fb = LZMA_MATCH_LEN_MAX;
    p->numFastBytes = fb;
  }
  p->lc = props.lc;
  p->lp = props.lp;
  p->pb = props.pb;
  p->fastMode = (props.algo == 0);
  p->matchFinderBase.btMode = props.btMode;
  {
    UInt32 numHashBytes = 4;
    if (props.btMode)
    {
      if (props.numHashBytes < 2)
        numHashBytes = 2;
      else if (props.numHashBytes < 4)
        numHashBytes = props.numHashBytes;
    }
    p->matchFinderBase.numHashBytes = numHashBytes;
  }

  p->matchFinderBase.cutValue = props.mc;

  p->writeEndMark = props.writeEndMark;

  #ifdef COMPRESS_MF_MT
  /*
  if (newMultiThread != _multiThread)
  {
    ReleaseMatchFinder();
    _multiThread = newMultiThread;