/* LzmaDec.c -- LZMA Decoder
   2009-09-20 : Igor Pavlov : Public domain */
#include <stdint.h>
#include <string.h>

#include "LzmaDec.h"


#define kNumTopBits 24
#define kTopValue ((uint32_t)1 << kNumTopBits)

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

#define RC_INIT_SIZE 5

#define LZMADEC_NORMALIZE if (range < kTopValue) { range <<= 8; code = (code << 8) | (*buf++); }

#define LZMADEC_IF_BIT_0(prob) ttt = *(prob); LZMADEC_NORMALIZE; bound = (range >> kNumBitModelTotalBits) * ttt; if (code < bound)
#define UPDATE_0(p) range = bound; *(p) = (uint16_t)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits));
#define LZMADEC_UPDATE_1(p) range -= bound; code -= bound; *(p) = (uint16_t)(ttt - (ttt >> kNumMoveBits));
#define GET_BIT2(p, i, A0, A1) LZMADEC_IF_BIT_0(p) \
{ UPDATE_0(p); i = (i + i); A0; } else \
{ LZMADEC_UPDATE_1(p); i = (i + i) + 1; A1; }
#define GET_BIT(p, i) GET_BIT2(p, i, ; , ;)

#define TREE_GET_BIT(probs, i) { GET_BIT((probs + i), i); }
#define TREE_DECODE(probs, limit, i) \
{ i = 1; do { TREE_GET_BIT(probs, i); } while (i < limit); i -= limit; }

/* #define _LZMA_SIZE_OPT */

#ifdef _LZMA_SIZE_OPT
#define TREE_6_DECODE(probs, i) TREE_DECODE(probs, (1 << 6), i)
#else
#define TREE_6_DECODE(probs, i) \
{ i = 1; \
   TREE_GET_BIT(probs, i); \
   TREE_GET_BIT(probs, i); \
   TREE_GET_BIT(probs, i); \
   TREE_GET_BIT(probs, i); \
   TREE_GET_BIT(probs, i); \
   TREE_GET_BIT(probs, i); \
   i -= 0x40; }
#endif

#define NORMALIZE_CHECK if (range < kTopValue) { if (buf >= bufLimit) return DUMMY_ERROR; range <<= 8; code = (code << 8) | (*buf++); }

#define IF_BIT_0_CHECK(p) ttt = *(p); NORMALIZE_CHECK; bound = (range >> kNumBitModelTotalBits) * ttt; if (code < bound)
#define UPDATE_0_CHECK range = bound;
#define UPDATE_1_CHECK range -= bound; code -= bound;
#define GET_BIT2_CHECK(p, i, A0, A1) IF_BIT_0_CHECK(p) \
{ UPDATE_0_CHECK; i = (i + i); A0; } else \
{ UPDATE_1_CHECK; i = (i + i) + 1; A1; }
#define GET_BIT_CHECK(p, i) GET_BIT2_CHECK(p, i, ; , ;)
#define TREE_DECODE_CHECK(probs, limit, i) \
{ i = 1; do { GET_BIT_CHECK(probs + i, i) } while (i < limit); i -= limit; }


#define kNumPosBitsMax 4
#define kNumPosStatesMax (1 << kNumPosBitsMax)

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

#define LenChoice 0
#define LenChoice2 (LenChoice + 1)
#define LenLow (LenChoice2 + 1)
#define LenMid (LenLow + (kNumPosStatesMax << kLenNumLowBits))
#define LenHigh (LenMid + (kNumPosStatesMax << kLenNumMidBits))
#define kNumLenProbs (LenHigh + kLenNumHighSymbols)


#define kNumStates 12
#define kNumLitStates 7

#define kStartPosModelIndex 4
#define kEndPosModelIndex 14
#define kNumFullDistances (1 << (kEndPosModelIndex >> 1))

#define kNumPosSlotBits 6
#define kNumLenToPosStates 4

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

#define kMatchMinLen 2
#define kMatchSpecLenStart (kMatchMinLen + kLenNumLowSymbols + kLenNumMidSymbols + kLenNumHighSymbols)

#define IsMatch 0
#define IsRep (IsMatch + (kNumStates << kNumPosBitsMax))
#define IsRepG0 (IsRep + kNumStates)
#define IsRepG1 (IsRepG0 + kNumStates)
#define IsRepG2 (IsRepG1 + kNumStates)
#define IsRep0Long (IsRepG2 + kNumStates)
#define PosSlot (IsRep0Long + (kNumStates << kNumPosBitsMax))
#define SpecPos (PosSlot + (kNumLenToPosStates << kNumPosSlotBits))
#define Align (SpecPos + kNumFullDistances - kEndPosModelIndex)
#define LenCoder (Align + kAlignTableSize)
#define RepLenCoder (LenCoder + kNumLenProbs)
#define Literal (RepLenCoder + kNumLenProbs)

#define LZMA_BASE_SIZE 1846
#define LZMA_LIT_SIZE 768

#define LzmaProps_GetNumProbs(p) ((uint32_t)LZMA_BASE_SIZE + (LZMA_LIT_SIZE << ((p)->lc + (p)->lp)))

#if Literal != LZMA_BASE_SIZE
StopCompilingDueBUG
#endif

#define LZMA_DIC_MIN (1 << 12)

/* First LZMA-symbol is always decoded.
   And it decodes new LZMA-symbols while (buf < bufLimit), but "buf" is without last normalization
   Out:
   Result:
   SZ_OK - OK
   SZ_ERROR_DATA - Error
   p->remainLen:
   < kMatchSpecLenStart : normal remain
   = kMatchSpecLenStart : finished
   = kMatchSpecLenStart + 1 : Flush marker
   = kMatchSpecLenStart + 2 : State Init Marker
   */

static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, size_t limit, const uint8_t *bufLimit)
{
   uint16_t *probs = p->probs;

   unsigned state = p->state;
   uint32_t rep0 = p->reps[0], rep1 = p->reps[1], rep2 = p->reps[2], rep3 = p->reps[3];
   unsigned pbMask = ((unsigned)1 << (p->prop.pb)) - 1;
   unsigned lpMask = ((unsigned)1 << (p->prop.lp)) - 1;
   unsigned lc = p->prop.lc;

   uint8_t *dic = p->dic;
   size_t dicBufSize = p->dicBufSize;
   size_t dicPos = p->dicPos;

   uint32_t processedPos = p->processedPos;
   uint32_t checkDicSize = p->checkDicSize;
   unsigned len = 0;

   const uint8_t *buf = p->buf;
   uint32_t range = p->range;
   uint32_t code = p->code;

   do
   {
      uint16_t *prob;
      uint32_t bound;
      unsigned ttt;
      unsigned posState = processedPos & pbMask;

      prob = probs + IsMatch + (state << kNumPosBitsMax) + posState;
      ttt = *(prob);
      LZMADEC_NORMALIZE;
      bound = (range >> kNumBitModelTotalBits) * ttt;
      
      if (code < bound)
      {
         unsigned symbol;
         UPDATE_0(prob);
         prob = probs + Literal;
         if (checkDicSize != 0 || processedPos != 0)
            prob += (LZMA_LIT_SIZE * (((processedPos & lpMask) << lc) +
                     (dic[(dicPos == 0 ? dicBufSize : dicPos) - 1] >> (8 - lc))));

         if (state < kNumLitStates)
         {
            state -= (state < 4) ? state : 3;
            symbol = 1;
            do { GET_BIT(prob + symbol, symbol) } while (symbol < 0x100);
         }
         else
         {
            unsigned matchuint8_t = p->dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
            unsigned offs = 0x100;
            state -= (state < 10) ? 3 : 6;
            symbol = 1;
            do
            {
               unsigned bit;
               uint16_t *probLit;
               matchuint8_t <<= 1;
               bit = (matchuint8_t & offs);
               probLit = prob + offs + bit + symbol;
               GET_BIT2(probLit, symbol, offs &= ~bit, offs &= bit)
            }
            while (symbol < 0x100);
         }
         dic[dicPos++] = (uint8_t)symbol;
         processedPos++;
         continue;
      }
      else
      {
         LZMADEC_UPDATE_1(prob);
         prob = probs + IsRep + state;

         ttt = *(prob);
         LZMADEC_NORMALIZE;
         bound = (range >> kNumBitModelTotalBits) * ttt;

         if (code < bound)
         {
            UPDATE_0(prob);
            state += kNumStates;
            prob = probs + LenCoder;
         }
         else
         {
            LZMADEC_UPDATE_1(prob);
            if (checkDicSize == 0 && processedPos == 0)
               return SZ_ERROR_DATA;
            prob = probs + IsRepG0 + state;

            ttt = *(prob);
            LZMADEC_NORMALIZE;
            bound = (range >> kNumBitModelTotalBits) * ttt;

            if (code < bound)
            {
               UPDATE_0(prob);
               prob = probs + IsRep0Long + (state << kNumPosBitsMax) + posState;

               ttt = *(prob);
               LZMADEC_NORMALIZE;
               bound = (range >> kNumBitModelTotalBits) * ttt;

               if (code < bound)
               {
                  UPDATE_0(prob);
                  dic[dicPos] = dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
                  dicPos++;
                  processedPos++;
                  state = state < kNumLitStates ? 9 : 11;
                  continue;
               }
               LZMADEC_UPDATE_1(prob);
            }
            else
            {
               uint32_t distance;
               LZMADEC_UPDATE_1(prob);
               prob = probs + IsRepG1 + state;

               ttt = *(prob);
               LZMADEC_NORMALIZE;
               bound = (range >> kNumBitModelTotalBits) * ttt;

               if (code < bound)
               {
                  UPDATE_0(prob);
                  distance = rep1;
               }
               else
               {
                  LZMADEC_UPDATE_1(prob);
                  prob = probs + IsRepG2 + state;

                  ttt = *(prob);
                  LZMADEC_NORMALIZE;
                  bound = (range >> kNumBitModelTotalBits) * ttt;

                  if (code < bound)
                  {
                     UPDATE_0(prob);
                     distance = rep2;
                  }
                  else
                  {
                     LZMADEC_UPDATE_1(prob);
                     distance = rep3;
                     rep3 = rep2;
                  }
                  rep2 = rep1;
               }
               rep1 = rep0;
               rep0 = distance;
            }
            state = state < kNumLitStates ? 8 : 11;
            prob = probs + RepLenCoder;
         }
         {
            unsigned limit, offset;
            uint16_t *probLen = prob + LenChoice;
               
            ttt = *(probLen);
            LZMADEC_NORMALIZE;
            bound = (range >> kNumBitModelTotalBits) * ttt;
            if (code < bound)
            {
               UPDATE_0(probLen);
               probLen = prob + LenLow + (posState << kLenNumLowBits);
               offset = 0;
               limit = (1 << kLenNumLowBits);
            }
            else
            {
               LZMADEC_UPDATE_1(probLen);
               probLen = prob + LenChoice2;

               ttt = *(probLen);
               LZMADEC_NORMALIZE;
               bound = (range >> kNumBitModelTotalBits) * ttt;
               if (code < bound)
               {
                  UPDATE_0(probLen);
                  probLen = prob + LenMid + (posState << kLenNumMidBits);
                  offset = kLenNumLowSymbols;
                  limit = (1 << kLenNumMidBits);
               }
               else
               {
                  LZMADEC_UPDATE_1(probLen);
                  probLen = prob + LenHigh;
                  offset = kLenNumLowSymbols + kLenNumMidSymbols;
                  limit = (1 << kLenNumHighBits);
               }
            }
            TREE_DECODE(probLen, limit, len);
            len += offset;
         }

         if (state >= kNumStates)
         {
            uint32_t distance;
            prob = probs + PosSlot +
               ((len < kNumLenToPosStates ? len : kNumLenToPosStates - 1) << kNumPosSlotBits);
            TREE_6_DECODE(prob, distance);
            if (distance >= kStartPosModelIndex)
            {
               unsigned posSlot = (unsigned)distance;
               int numDirectBits = (int)(((distance >> 1) - 1));
               distance = (2 | (distance & 1));
               if (posSlot < kEndPosModelIndex)
               {
                  distance <<= numDirectBits;
                  prob = probs + SpecPos + distance - posSlot - 1;
                  {
                     uint32_t mask = 1;
                     unsigned i = 1;
                     do
                     {
                        GET_BIT2(prob + i, i, ; , distance |= mask);
                        mask <<= 1;
                     }
                     while (--numDirectBits != 0);
                  }
               }
               else
               {
                  numDirectBits -= kNumAlignBits;
                  do
                  {
                     LZMADEC_NORMALIZE
                        range >>= 1;

                     {
                        uint32_t t;
                        code -= range;
                        t = (0 - ((uint32_t)code >> 31)); /* (uint32_t)((Int32)code >> 31) */
                        distance = (distance << 1) + (t + 1);
                        code += range & t;
                     }
                     /*
                        distance <<= 1;
                        if (code >= range)
                        {
                        code -= range;
                        distance |= 1;
                        }
                        */
                  }
                  while (--numDirectBits != 0);
                  prob = probs + Align;
                  distance <<= kNumAlignBits;
                  {
                     unsigned i = 1;
                     GET_BIT2(prob + i, i, ; , distance |= 1);
                     GET_BIT2(prob + i, i, ; , distance |= 2);
                     GET_BIT2(prob + i, i, ; , distance |= 4);
                     GET_BIT2(prob + i, i, ; , distance |= 8);
                  }
                  if (distance == (uint32_t)0xFFFFFFFF)
                  {
                     len += kMatchSpecLenStart;
                     state -= kNumStates;
                     break;
                  }
               }
            }
            rep3 = rep2;
            rep2 = rep1;
            rep1 = rep0;
            rep0 = distance + 1;
            if (checkDicSize == 0)
            {
               if (distance >= processedPos)
                  return SZ_ERROR_DATA;
            }
            else if (distance >= checkDicSize)
               return SZ_ERROR_DATA;
            state = (state < kNumStates + kNumLitStates) ? kNumLitStates : kNumLitStates + 3;
         }

         len += kMatchMinLen;

         if (limit == dicPos)
            return SZ_ERROR_DATA;
         {
            size_t rem = limit - dicPos;
            unsigned curLen = ((rem < len) ? (unsigned)rem : len);
            size_t pos = (dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0);

            processedPos += curLen;

            len -= curLen;
            if (pos + curLen <= dicBufSize)
            {
               uint8_t *dest = dic + dicPos;
               ptrdiff_t src = (ptrdiff_t)pos - (ptrdiff_t)dicPos;
               const uint8_t *lim = dest + curLen;
               dicPos += curLen;
               do
                  *(dest) = (uint8_t)*(dest + src);
               while (++dest != lim);
            }
            else
            {
               do
               {
                  dic[dicPos++] = dic[pos];
                  if (++pos == dicBufSize)
                     pos = 0;
               }
               while (--curLen != 0);
            }
         }
      }
   }while (dicPos < limit && buf < bufLimit);

   LZMADEC_NORMALIZE;

   p->buf = buf;
   p->range = range;
   p->code = code;
   p->remainLen = len;
   p->dicPos = dicPos;
   p->processedPos = processedPos;
   p->reps[0] = rep0;
   p->reps[1] = rep1;
   p->reps[2] = rep2;
   p->reps[3] = rep3;
   p->state = state;

   return SZ_OK;
}

static void MY_FAST_CALL LzmaDec_WriteRem(CLzmaDec *p, size_t limit)
{
   if (p->remainLen != 0 && p->remainLen < kMatchSpecLenStart)
   {
      uint8_t *dic = p->dic;
      size_t dicPos = p->dicPos;
      size_t dicBufSize = p->dicBufSize;
      unsigned len = p->remainLen;
      uint32_t rep0 = p->reps[0];
      if (limit - dicPos < len)
         len = (unsigned)(limit - dicPos);

      if (p->checkDicSize == 0 && p->prop.dicSize - p->processedPos <= len)
         p->checkDicSize = p->prop.dicSize;

      p->processedPos += len;
      p->remainLen -= len;
      while (len-- != 0)
      {
         dic[dicPos] = dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
         dicPos++;
      }
      p->dicPos = dicPos;
   }
}

static int MY_FAST_CALL LzmaDec_DecodeReal2(CLzmaDec *p, size_t limit, const uint8_t *bufLimit)
{
   do
   {
      size_t limit2 = limit;
      if (p->checkDicSize == 0)
      {
         uint32_t rem = p->prop.dicSize - p->processedPos;
         if (limit - p->dicPos > rem)
            limit2 = p->dicPos + rem;
      }
      RINOK(LzmaDec_DecodeReal(p, limit2, bufLimit));
      if (p->processedPos >= p->prop.dicSize)
         p->checkDicSize = p->prop.dicSize;
      LzmaDec_WriteRem(p, limit);
   }
   while (p->dicPos < limit && p->buf < bufLimit && p->remainLen < kMatchSpecLenStart);

   if (p->remainLen > kMatchSpecLenStart)
   {
      p->remainLen = kMatchSpecLenStart;
   }
   return 0;
}

typedef enum
{
   DUMMY_ERROR, /* unexpected end of input stream */
   DUMMY_LIT,
   DUMMY_MATCH,
   DUMMY_REP
} ELzmaDummy;

static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const uint8_t *buf, size_t inSize)
{
   uint32_t range = p->range;
   uint32_t code = p->code;
   const uint8_t *bufLimit = buf + inSize;
   uint16_t *probs = p->probs;
   unsigned state = p->state;
   ELzmaDummy res;

   {
      uint16_t *prob;
      uint32_t bound;
      unsigned ttt;
      unsigned posState = (p->processedPos) & ((1 << p->prop.pb) - 1);

      prob = probs + IsMatch + (state << kNumPosBitsMax) + posState;
      IF_BIT_0_CHECK(prob)
      {
         UPDATE_0_CHECK

            /* if (bufLimit - buf >= 7) return DUMMY_LIT; */

            prob = probs + Literal;
         if (p->checkDicSize != 0 || p->processedPos != 0)
            prob += (LZMA_LIT_SIZE *
                  ((((p->processedPos) & ((1 << (p->prop.lp)) - 1)) << p->prop.lc) +
                   (p->dic[(p->dicPos == 0 ? p->dicBufSize : p->dicPos) - 1] >> (8 - p->prop.lc))));

         if (state < kNumLitStates)
         {
            unsigned symbol = 1;
            do { GET_BIT_CHECK(prob + symbol, symbol) } while (symbol < 0x100);
         }
         else
         {
            unsigned matchuint8_t = p->dic[p->dicPos - p->reps[0] +
               ((p->dicPos < p->reps[0]) ? p->dicBufSize : 0)];
            unsigned offs = 0x100;
            unsigned symbol = 1;
            do
            {
               unsigned bit;
               uint16_t *probLit;
               matchuint8_t <<= 1;
               bit = (matchuint8_t & offs);
               probLit = prob + offs + bit + symbol;
               GET_BIT2_CHECK(probLit, symbol, offs &= ~bit, offs &= bit)
            }
            while (symbol < 0x100);
         }
         res = DUMMY_LIT;
      }
      else
      {
         unsigned len;
         UPDATE_1_CHECK;

         prob = probs + IsRep + state;
         IF_BIT_0_CHECK(prob)
         {
            UPDATE_0_CHECK;
            state = 0;
            prob = probs + LenCoder;
            res = DUMMY_MATCH;
         }
         else
         {
            UPDATE_1_CHECK;
            res = DUMMY_REP;
            prob = probs + IsRepG0 + state;
            IF_BIT_0_CHECK(prob)
            {
               UPDATE_0_CHECK;
               prob = probs + IsRep0Long + (state << kNumPosBitsMax) + posState;
               IF_BIT_0_CHECK(prob)
               {
                  UPDATE_0_CHECK;
                  NORMALIZE_CHECK;
                  return DUMMY_REP;
               }
               else
               {
                  UPDATE_1_CHECK;
               }
            }
            else
            {
               UPDATE_1_CHECK;
               prob = probs + IsRepG1 + state;
               IF_BIT_0_CHECK(prob)
               {
                  UPDATE_0_CHECK;
               }
               else
               {
                  UPDATE_1_CHECK;
                  prob = probs + IsRepG2 + state;
                  IF_BIT_0_CHECK(prob)
                  {
                     UPDATE_0_CHECK;
                  }
                  else
                  {
                     UPDATE_1_CHECK;
                  }
               }
            }
            state = kNumStates;
            prob = probs + RepLenCoder;
         }
         {
            unsigned limit, offset;
            uint16_t *probLen = prob + LenChoice;
            IF_BIT_0_CHECK(probLen)
            {
               UPDATE_0_CHECK;
               probLen = prob + LenLow + (posState << kLenNumLowBits);
               offset = 0;
               limit = 1 << kLenNumLowBits;
            }
            else
            {
               UPDATE_1_CHECK;
               probLen = prob + LenChoice2;
               IF_BIT_0_CHECK(probLen)
               {
                  UPDATE_0_CHECK;
                  probLen = prob + LenMid + (posState << kLenNumMidBits);
                  offset = kLenNumLowSymbols;
                  limit = 1 << kLenNumMidBits;
               }
               else
               {
                  UPDATE_1_CHECK;
                  probLen = prob + LenHigh;
                  offset = kLenNumLowSymbols + kLenNumMidSymbols;
                  limit = 1 << kLenNumHighBits;
               }
            }
            TREE_DECODE_CHECK(probLen, limit, len);
            len += offset;
         }

         if (state < 4)
         {
            unsigned posSlot;
            prob = probs + PosSlot +
               ((len < kNumLenToPosStates ? len : kNumLenToPosStates - 1) <<
                kNumPosSlotBits);
            TREE_DECODE_CHECK(prob, 1 << kNumPosSlotBits, posSlot);
            if (posSlot >= kStartPosModelIndex)
            {
               int numDirectBits = ((posSlot >> 1) - 1);

               /* if (bufLimit - buf >= 8) return DUMMY_MATCH; */

               if (posSlot < kEndPosModelIndex)
               {
                  prob = probs + SpecPos + ((2 | (posSlot & 1)) << numDirectBits) - posSlot - 1;
               }
               else
               {
                  numDirectBits -= kNumAlignBits;
                  do
                  {
                     NORMALIZE_CHECK
                        range >>= 1;
                     code -= range & (((code - range) >> 31) - 1);
                     /* if (code >= range) code -= range; */
                  }
                  while (--numDirectBits != 0);
                  prob = probs + Align;
                  numDirectBits = kNumAlignBits;
               }
               {
                  unsigned i = 1;
                  do
                  {
                     GET_BIT_CHECK(prob + i, i);
                  }
                  while (--numDirectBits != 0);
               }
            }
         }
      }
   }
   NORMALIZE_CHECK;
   return res;
}


static void LzmaDec_InitRc(CLzmaDec *p, const uint8_t *data)
{
   p->code = ((uint32_t)data[1] << 24) | ((uint32_t)data[2] << 16) | ((uint32_t)data[3] << 8) | ((uint32_t)data[4]);
   p->range = 0xFFFFFFFF;
   p->needFlush = 0;
}

void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState);

void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState)
{
   p->needFlush = 1;
   p->remainLen = 0;
   p->tempBufSize = 0;

   if (initDic)
   {
      p->processedPos = 0;
      p->checkDicSize = 0;
      p->needInitState = 1;
   }
   if (initState)
      p->needInitState = 1;
}

void LzmaDec_Init(CLzmaDec *p)
{
   p->dicPos = 0;
   LzmaDec_InitDicAndState(p, True, True);
}

static void LzmaDec_InitStateReal(CLzmaDec *p)
{
   uint32_t numProbs = Literal + ((uint32_t)LZMA_LIT_SIZE << (p->prop.lc + p->prop.lp));
   uint32_t i;
   uint16_t *probs = p->probs;
   for (i = 0; i < numProbs; i++)
      probs[i] = kBitModelTotal >> 1;
   p->reps[0] = p->reps[1] = p->reps[2] = p->reps[3] = 1;
   p->state = 0;
   p->needInitState = 0;
}

SRes LzmaDec_DecodeToDic(CLzmaDec *p, size_t dicLimit, const uint8_t *src, size_t *srcLen,
      ELzmaFinishMode finishMode, ELzmaStatus *status)
{
   size_t inSize = *srcLen;
   (*srcLen) = 0;
   LzmaDec_WriteRem(p, dicLimit);

   *status = LZMA_STATUS_NOT_SPECIFIED;

   while (p->remainLen != kMatchSpecLenStart)
   {
      int checkEndMarkNow;

      if (p->needFlush != 0)
      {
         for (; inSize > 0 && p->tempBufSize < RC_INIT_SIZE; (*srcLen)++, inSize--)
            p->tempBuf[p->tempBufSize++] = *src++;
         if (p->tempBufSize < RC_INIT_SIZE)
         {
            *status = LZMA_STATUS_NEEDS_MORE_INPUT;
            return SZ_OK;
         }
         if (p->tempBuf[0] != 0)
            return SZ_ERROR_DATA;

         LzmaDec_InitRc(p, p->tempBuf);
         p->tempBufSize = 0;
      }

      checkEndMarkNow = 0;
      if (p->dicPos >= dicLimit)
      {
         if (p->remainLen == 0 && p->code == 0)
         {
            *status = LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK;
            return SZ_OK;
         }
         if (finishMode == LZMA_FINISH_ANY)
         {
            *status = LZMA_STATUS_NOT_FINISHED;
            return SZ_OK;
         }
         if (p->remainLen != 0)
         {
            *status = LZMA_STATUS_NOT_FINISHED;
            return SZ_ERROR_DATA;
         }
         checkEndMarkNow = 1;
      }

      if (p->needInitState)
         LzmaDec_InitStateReal(p);

      if (p->tempBufSize == 0)
      {
         size_t processed;
         const uint8_t *bufLimit;
         if (inSize < LZMA_REQUIRED_INPUT_MAX || checkEndMarkNow)
         {
            int dummyRes = LzmaDec_TryDummy(p, src, inSize);
            if (dummyRes == DUMMY_ERROR)
            {
               memcpy(p->tempBuf, src, inSize);
               p->tempBufSize = (unsigned)inSize;
               (*srcLen) += inSize;
               *status = LZMA_STATUS_NEEDS_MORE_INPUT;
               return SZ_OK;
            }
            if (checkEndMarkNow && dummyRes != DUMMY_MATCH)
            {
               *status = LZMA_STATUS_NOT_FINISHED;
               return SZ_ERROR_DATA;
            }
            bufLimit = src;
         }
         else
            bufLimit = src + inSize - LZMA_REQUIRED_INPUT_MAX;
         p->buf = src;
         if (LzmaDec_DecodeReal2(p, dicLimit, bufLimit) != 0)
            return SZ_ERROR_DATA;
         processed = (size_t)(p->buf - src);
         (*srcLen) += processed;
         src += processed;
         inSize -= processed;
      }
      else
      {
         unsigned rem = p->tempBufSize, lookAhead = 0;
         while (rem < LZMA_REQUIRED_INPUT_MAX && lookAhead < inSize)
            p->tempBuf[rem++] = src[lookAhead++];
         p->tempBufSize = rem;
         if (rem < LZMA_REQUIRED_INPUT_MAX || checkEndMarkNow)
         {
            int dummyRes = LzmaDec_TryDummy(p, p->tempBuf, rem);
            if (dummyRes == DUMMY_ERROR)
            {
               (*srcLen) += lookAhead;
               *status = LZMA_STATUS_NEEDS_MORE_INPUT;
               return SZ_OK;
            }
            if (checkEndMarkNow && dummyRes != DUMMY_MATCH)
            {
               *status = LZMA_STATUS_NOT_FINISHED;
               return SZ_ERROR_DATA;
            }
         }
         p->buf = p->tempBuf;
         if (LzmaDec_DecodeReal2(p, dicLimit, p->buf) != 0)
            return SZ_ERROR_DATA;
         lookAhead -= (rem - (unsigned)(p->buf - p->tempBuf));
         (*srcLen) += lookAhead;
         src += lookAhead;
         inSize -= lookAhead;
         p->tempBufSize = 0;
      }
   }
   if (p->code == 0)
      *status = LZMA_STATUS_FINISHED_WITH_MARK;
   return (p->code == 0) ? SZ_OK : SZ_ERROR_DATA;
}

SRes LzmaDec_DecodeToBuf(CLzmaDec *p, uint8_t *dest, size_t *destLen, const uint8_t *src, size_t *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
{
   size_t outSize = *destLen;
   size_t inSize = *srcLen;
   *srcLen = *destLen = 0;
   for (;;)
   {
      size_t inSizeCur = inSize, outSizeCur, dicPos;
      ELzmaFinishMode curFinishMode;
      SRes res;
      if (p->dicPos == p->dicBufSize)
         p->dicPos = 0;
      dicPos = p->dicPos;
      if (outSize > p->dicBufSize - dicPos)
      {
         outSizeCur = p->dicBufSize;
         curFinishMode = LZMA_FINISH_ANY;
      }
      else
      {
         outSizeCur = dicPos + outSize;
         curFinishMode = finishMode;
      }

      res = LzmaDec_DecodeToDic(p, outSizeCur, src, &inSizeCur, curFinishMode, status);
      src += inSizeCur;
      inSize -= inSizeCur;
      *srcLen += inSizeCur;
      outSizeCur = p->dicPos - dicPos;
      memcpy(dest, p->dic + dicPos, outSizeCur);
      dest += outSizeCur;
      outSize -= outSizeCur;
      *destLen += outSizeCur;
      if (res != 0)
         return res;
      if (outSizeCur == 0 || outSize == 0)
         return SZ_OK;
   }
}

void LzmaDec_FreeProbs(CLzmaDec *p, ISzAlloc *alloc)
{
   alloc->Free(alloc, p->probs);
   p->probs = 0;
}

static void LzmaDec_FreeDict(CLzmaDec *p, ISzAlloc *alloc)
{
   alloc->Free(alloc, p->dic);
   p->dic = 0;
}

void LzmaDec_Free(CLzmaDec *p, ISzAlloc *alloc)
{
   LzmaDec_FreeProbs(p, alloc);
   LzmaDec_FreeDict(p, alloc);
}

SRes LzmaProps_Decode(CLzmaProps *p, const uint8_t *data, unsigned size)
{
   uint32_t dicSize;
   uint8_t d;

   if (size < LZMA_PROPS_SIZE)
      return SZ_ERROR_UNSUPPORTED;
   else
      dicSize = data[1] | ((uint32_t)data[2] << 8) | ((uint32_t)data[3] << 16) | ((uint32_t)data[4] << 24);

   if (dicSize < LZMA_DIC_MIN)
      dicSize = LZMA_DIC_MIN;
   p->dicSize = dicSize;

   d = data[0];
   if (d >= (9 * 5 * 5))
      return SZ_ERROR_UNSUPPORTED;

   p->lc = d % 9;
   d /= 9;
   p->pb = d / 5;
   p->lp = d % 5;

   return SZ_OK;
}

static SRes LzmaDec_AllocateProbs2(CLzmaDec *p, const CLzmaProps *propNew, ISzAlloc *alloc)
{
   uint32_t numProbs = LzmaProps_GetNumProbs(propNew);
   if (p->probs == 0 || numProbs != p->numProbs)
   {
      LzmaDec_FreeProbs(p, alloc);
      p->probs = (uint16_t *)alloc->Alloc(alloc, numProbs * sizeof(uint16_t));
      p->numProbs = numProbs;
      if (p->probs == 0)
         return SZ_ERROR_MEM;
   }
   return SZ_OK;
}

SRes LzmaDec_AllocateProbs(CLzmaDec *p, const uint8_t *props, unsigned propsSize, ISzAlloc *alloc)
{
   CLzmaProps propNew;
   RINOK(LzmaProps_Decode(&propNew, props, propsSize));
   RINOK(LzmaDec_AllocateProbs2(p, &propNew, alloc));
   p->prop = propNew;
   return SZ_OK;
}

SRes LzmaDec_Allocate(CLzmaDec *p, const uint8_t *props, unsigned propsSize, ISzAlloc *alloc)
{
   CLzmaProps propNew;
   size_t dicBufSize;
   RINOK(LzmaProps_Decode(&propNew, props, propsSize));
   RINOK(LzmaDec_AllocateProbs2(p, &propNew, alloc));
   dicBufSize = propNew.dicSize;
   if (p->dic == 0 || dicBufSize != p->dicBufSize)
   {
      LzmaDec_FreeDict(p, alloc);
      p->dic = (uint8_t *)alloc->Alloc(alloc, dicBufSize);
      if (p->dic == 0)
      {
         LzmaDec_FreeProbs(p, alloc);
         return SZ_ERROR_MEM;
      }
   }
   p->dicBufSize = dicBufSize;
   p->prop = propNew;
   return SZ_OK;
}

SRes LzmaDecode(uint8_t *dest, size_t *destLen, const uint8_t *src, size_t *srcLen,
      const uint8_t *propData, unsigned propSize, ELzmaFinishMode finishMode,
      ELzmaStatus *status, ISzAlloc *alloc)
{
   CLzmaDec p;
   SRes res;
   size_t inSize = *srcLen;
   size_t outSize = *destLen;
   *srcLen = *destLen = 0;
   if (inSize < RC_INIT_SIZE)
      return SZ_ERROR_INPUT_EOF;

   LzmaDec_Construct(&p);
   res = LzmaDec_AllocateProbs(&p, propData, propSize, alloc);
   if (res != 0)
      return res;
   p.dic = dest;
   p.dicBufSize = outSize;

   LzmaDec_Init(&p);

   *srcLen = inSize;
   res = LzmaDec_DecodeToDic(&p, outSize, src, srcLen, finishMode, status);

   if (res == SZ_OK && *status == LZMA_STATUS_NEEDS_MORE_INPUT)
      res = SZ_ERROR_INPUT_EOF;

   (*destLen) = p.dicPos;
   LzmaDec_FreeProbs(&p, alloc);
   return res;
}