/*

 * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.

 * All rights reserved.

 *

 * This source code is licensed under both the BSD-style license (found in the

 * LICENSE file in the root directory of this source tree) and the GPLv2 (found

 * in the COPYING file in the root directory of this source tree).

 * You may select, at your option, one of the above-listed licenses.

 */

#include "zstd_compress_internal.h"

#include "hist.h"

#include "zstd_opt.h"

#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */

#define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */

#define ZSTD_MAX_PRICE     (1<<30)

#define ZSTD_PREDEF_THRESHOLD 1024   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */

/*-*************************************

*  Price functions for optimal parser

***************************************/

#if 0    /* approximation at bit level */

#  define BITCOST_ACCURACY 0

#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)

#  define WEIGHT(stat)  ((void)opt, ZSTD_bitWeight(stat))

#elif 0  /* fractional bit accuracy */

#  define BITCOST_ACCURACY 8

#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)

#  define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))

#else    /* opt==approx, ultra==accurate */

#  define BITCOST_ACCURACY 8

#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)

#  define WEIGHT(stat,opt) (opt ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))

#endif

MEM_STATIC U32 ZSTD_bitWeight(U32 stat)

{

    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);

}

MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)

{

    U32 const stat = rawStat + 1;

    U32 const hb = ZSTD_highbit32(stat);

    U32 const BWeight = hb * BITCOST_MULTIPLIER;

    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;

    U32 const weight = BWeight + FWeight;

    assert(hb + BITCOST_ACCURACY < 31);

    return weight;

}

#if (DEBUGLEVEL>=2)

/* debugging function,

 * @return price in bytes as fractional value

 * for debug messages only */

MEM_STATIC double ZSTD_fCost(U32 price)

{

    return (double)price / (BITCOST_MULTIPLIER*8);

}

#endif

static int ZSTD_compressedLiterals(optState_t const* const optPtr)

{

    return optPtr->literalCompressionMode != ZSTD_lcm_uncompressed;

}

static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)

{

    if (ZSTD_compressedLiterals(optPtr))

        optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);

    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);

    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);

    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);

}

/* ZSTD_downscaleStat() :

 * reduce all elements in table by a factor 2^(ZSTD_FREQ_DIV+malus)

 * return the resulting sum of elements */

static U32 ZSTD_downscaleStat(unsigned* table, U32 lastEltIndex, int malus)

{

    U32 s, sum=0;

    DEBUGLOG(5, "ZSTD_downscaleStat (nbElts=%u)", (unsigned)lastEltIndex+1);

    assert(ZSTD_FREQ_DIV+malus > 0 && ZSTD_FREQ_DIV+malus < 31);

    for (s=0; s<lastEltIndex+1; s++) {

        table[s] = 1 + (table[s] >> (ZSTD_FREQ_DIV+malus));

        sum += table[s];

    }

    return sum;

}

/* ZSTD_rescaleFreqs() :

 * if first block (detected by optPtr->litLengthSum == 0) : init statistics

 *    take hints from dictionary if there is one

 *    or init from zero, using src for literals stats, or flat 1 for match symbols

 * otherwise downscale existing stats, to be used as seed for next block.

 */

static void

ZSTD_rescaleFreqs(optState_t* const optPtr,

            const BYTE* const src, size_t const srcSize,

                  int const optLevel)

{

    int const compressedLiterals = ZSTD_compressedLiterals(optPtr);

    DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);

    optPtr->priceType = zop_dynamic;

    if (optPtr->litLengthSum == 0) {  /* first block : init */

        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {  /* heuristic */

            DEBUGLOG(5, "(srcSize <= ZSTD_PREDEF_THRESHOLD) => zop_predef");

            optPtr->priceType = zop_predef;

        }

        assert(optPtr->symbolCosts != NULL);

        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {

            /* huffman table presumed generated by dictionary */

            optPtr->priceType = zop_dynamic;

            if (compressedLiterals) {

                unsigned lit;

                assert(optPtr->litFreq != NULL);

                optPtr->litSum = 0;

                for (lit=0; lit<=MaxLit; lit++) {

                    U32 const scaleLog = 11;   /* scale to 2K */

                    U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);

                    assert(bitCost <= scaleLog);

                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;

                    optPtr->litSum += optPtr->litFreq[lit];

            }   }

            {   unsigned ll;

                FSE_CState_t llstate;

                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);

                optPtr->litLengthSum = 0;

                for (ll=0; ll<=MaxLL; ll++) {

                    U32 const scaleLog = 10;   /* scale to 1K */

                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);

                    assert(bitCost < scaleLog);

                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;

                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];

            }   }

            {   unsigned ml;

                FSE_CState_t mlstate;

                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);

                optPtr->matchLengthSum = 0;

                for (ml=0; ml<=MaxML; ml++) {

                    U32 const scaleLog = 10;

                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);

                    assert(bitCost < scaleLog);

                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;

                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];

            }   }

            {   unsigned of;

                FSE_CState_t ofstate;

                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);

                optPtr->offCodeSum = 0;

                for (of=0; of<=MaxOff; of++) {

                    U32 const scaleLog = 10;

                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);

                    assert(bitCost < scaleLog);

                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;

                    optPtr->offCodeSum += optPtr->offCodeFreq[of];

            }   }

        } else {  /* not a dictionary */

            assert(optPtr->litFreq != NULL);

            if (compressedLiterals) {

                unsigned lit = MaxLit;

                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */

                optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);

            }

            {   unsigned ll;

                for (ll=0; ll<=MaxLL; ll++)

                    optPtr->litLengthFreq[ll] = 1;

            }

            optPtr->litLengthSum = MaxLL+1;

            {   unsigned ml;

                for (ml=0; ml<=MaxML; ml++)

                    optPtr->matchLengthFreq[ml] = 1;

            }

            optPtr->matchLengthSum = MaxML+1;

            {   unsigned of;

                for (of=0; of<=MaxOff; of++)

                    optPtr->offCodeFreq[of] = 1;

            }

            optPtr->offCodeSum = MaxOff+1;

        }

    } else {   /* new block : re-use previous statistics, scaled down */

        if (compressedLiterals)

            optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);

        optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);

        optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);

        optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);

    }

    ZSTD_setBasePrices(optPtr, optLevel);

}

/* ZSTD_rawLiteralsCost() :

 * price of literals (only) in specified segment (which length can be 0).

 * does not include price of literalLength symbol */

static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,

                                const optState_t* const optPtr,

                                int optLevel)

{

    if (litLength == 0) return 0;

    if (!ZSTD_compressedLiterals(optPtr))

        return (litLength << 3) * BITCOST_MULTIPLIER;  /* Uncompressed - 8 bytes per literal. */

    if (optPtr->priceType == zop_predef)

        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */

    /* dynamic statistics */

    {   U32 price = litLength * optPtr->litSumBasePrice;

        U32 u;

        for (u=0; u < litLength; u++) {

            assert(WEIGHT(optPtr->litFreq[literals[u]], optLevel) <= optPtr->litSumBasePrice);   /* literal cost should never be negative */

            price -= WEIGHT(optPtr->litFreq[literals[u]], optLevel);

        }

        return price;

    }

}

/* ZSTD_litLengthPrice() :

 * cost of literalLength symbol */

static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)

{

    if (optPtr->priceType == zop_predef) return WEIGHT(litLength, optLevel);

    /* dynamic statistics */

    {   U32 const llCode = ZSTD_LLcode(litLength);

        return (LL_bits[llCode] * BITCOST_MULTIPLIER)

             + optPtr->litLengthSumBasePrice

             - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);

    }

}

/* ZSTD_litLengthContribution() :

 * @return ( cost(litlength) - cost(0) )

 * this value can then be added to rawLiteralsCost()

 * to provide a cost which is directly comparable to a match ending at same position */

static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr, int optLevel)

{

    if (optPtr->priceType >= zop_predef) return (int)WEIGHT(litLength, optLevel);

    /* dynamic statistics */

    {   U32 const llCode = ZSTD_LLcode(litLength);

        int const contribution = (int)(LL_bits[llCode] * BITCOST_MULTIPLIER)

                               + (int)WEIGHT(optPtr->litLengthFreq[0], optLevel)   /* note: log2litLengthSum cancel out */

                               - (int)WEIGHT(optPtr->litLengthFreq[llCode], optLevel);

#if 1

        return contribution;

#else

        return MAX(0, contribution); /* sometimes better, sometimes not ... */

#endif

    }

}

/* ZSTD_literalsContribution() :

 * creates a fake cost for the literals part of a sequence

 * which can be compared to the ending cost of a match

 * should a new match start at this position */

static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLength,

                                     const optState_t* const optPtr,

                                     int optLevel)

{

    int const contribution = (int)ZSTD_rawLiteralsCost(literals, litLength, optPtr, optLevel)

                           + ZSTD_litLengthContribution(litLength, optPtr, optLevel);

    return contribution;

}

/* ZSTD_getMatchPrice() :

 * Provides the cost of the match part (offset + matchLength) of a sequence

 * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.

 * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */

FORCE_INLINE_TEMPLATE U32

ZSTD_getMatchPrice(U32 const offset,

                   U32 const matchLength,

             const optState_t* const optPtr,

                   int const optLevel)

{

    U32 price;

    U32 const offCode = ZSTD_highbit32(offset+1);

    U32 const mlBase = matchLength - MINMATCH;

    assert(matchLength >= MINMATCH);

    if (optPtr->priceType == zop_predef)  /* fixed scheme, do not use statistics */

        return WEIGHT(mlBase, optLevel) + ((16 + offCode) * BITCOST_MULTIPLIER);

    /* dynamic statistics */

    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));

    if ((optLevel<2) /*static*/ && offCode >= 20)

        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */

    /* match Length */

    {   U32 const mlCode = ZSTD_MLcode(mlBase);

        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));

    }

    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */

    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);

    return price;

}

/* ZSTD_updateStats() :

 * assumption : literals + litLengtn <= iend */

static void ZSTD_updateStats(optState_t* const optPtr,

                             U32 litLength, const BYTE* literals,

                             U32 offsetCode, U32 matchLength)

{

    /* literals */

    if (ZSTD_compressedLiterals(optPtr)) {

        U32 u;

        for (u=0; u < litLength; u++)

            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;

        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;

    }

    /* literal Length */

    {   U32 const llCode = ZSTD_LLcode(litLength);

        optPtr->litLengthFreq[llCode]++;

        optPtr->litLengthSum++;

    }

    /* match offset code (0-2=>repCode; 3+=>offset+2) */

    {   U32 const offCode = ZSTD_highbit32(offsetCode+1);

        assert(offCode <= MaxOff);

        optPtr->offCodeFreq[offCode]++;

        optPtr->offCodeSum++;

    }

    /* match Length */

    {   U32 const mlBase = matchLength - MINMATCH;

        U32 const mlCode = ZSTD_MLcode(mlBase);

        optPtr->matchLengthFreq[mlCode]++;

        optPtr->matchLengthSum++;

    }

}

/* ZSTD_readMINMATCH() :

 * function safe only for comparisons

 * assumption : memPtr must be at least 4 bytes before end of buffer */

MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)

{

    switch (length)

    {

    default :

    case 4 : return MEM_read32(memPtr);

    case 3 : if (MEM_isLittleEndian())

                return MEM_read32(memPtr)<<8;

             else

                return MEM_read32(memPtr)>>8;

    }

}

/* Update hashTable3 up to ip (excluded)

   Assumption : always within prefix (i.e. not within extDict) */

static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,

                                              U32* nextToUpdate3,

                                              const BYTE* const ip)

{

    U32* const hashTable3 = ms->hashTable3;

    U32 const hashLog3 = ms->hashLog3;

    const BYTE* const base = ms->window.base;

    U32 idx = *nextToUpdate3;

    U32 const target = (U32)(ip - base);

    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);

    assert(hashLog3 > 0);

    while(idx < target) {

        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;

        idx++;

    }

    *nextToUpdate3 = target;

    return hashTable3[hash3];

}

/*-*************************************

*  Binary Tree search

***************************************/

/** ZSTD_insertBt1() : add one or multiple positions to tree.

 *  ip : assumed <= iend-8 .

 * @return : nb of positions added */

static U32 ZSTD_insertBt1(

                ZSTD_matchState_t* ms,

                const BYTE* const ip, const BYTE* const iend,

                U32 const mls, const int extDict)

{

    const ZSTD_compressionParameters* const cParams = &ms->cParams;

    U32*   const hashTable = ms->hashTable;

    U32    const hashLog = cParams->hashLog;

    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);

    U32*   const bt = ms->chainTable;

    U32    const btLog  = cParams->chainLog - 1;

    U32    const btMask = (1 << btLog) - 1;

    U32 matchIndex = hashTable[h];

    size_t commonLengthSmaller=0, commonLengthLarger=0;

    const BYTE* const base = ms->window.base;

    const BYTE* const dictBase = ms->window.dictBase;

    const U32 dictLimit = ms->window.dictLimit;

    const BYTE* const dictEnd = dictBase + dictLimit;

    const BYTE* const prefixStart = base + dictLimit;

    const BYTE* match;

    const U32 current = (U32)(ip-base);

    const U32 btLow = btMask >= current ? 0 : current - btMask;

    U32* smallerPtr = bt + 2*(current&btMask);

    U32* largerPtr  = smallerPtr + 1;

    U32 dummy32;   /* to be nullified at the end */

    U32 const windowLow = ms->window.lowLimit;

    U32 matchEndIdx = current+8+1;

    size_t bestLength = 8;

    U32 nbCompares = 1U << cParams->searchLog;

#ifdef ZSTD_C_PREDICT

    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);

    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);

    predictedSmall += (predictedSmall>0);

    predictedLarge += (predictedLarge>0);

#endif /* ZSTD_C_PREDICT */

    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);

    assert(ip <= iend-8);   /* required for h calculation */

    hashTable[h] = current;   /* Update Hash Table */

    assert(windowLow > 0);

    while (nbCompares-- && (matchIndex >= windowLow)) {

        U32* const nextPtr = bt + 2*(matchIndex & btMask);

        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */

        assert(matchIndex < current);

#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */

        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */

        if (matchIndex == predictedSmall) {

            /* no need to check length, result known */

            *smallerPtr = matchIndex;

            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */

            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */

            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */

            predictedSmall = predictPtr[1] + (predictPtr[1]>0);

            continue;

        }

        if (matchIndex == predictedLarge) {

            *largerPtr = matchIndex;

            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */

            largerPtr = nextPtr;

            matchIndex = nextPtr[0];

            predictedLarge = predictPtr[0] + (predictPtr[0]>0);

            continue;

        }

#endif

        if (!extDict || (matchIndex+matchLength >= dictLimit)) {

            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */

            match = base + matchIndex;

            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);

        } else {

            match = dictBase + matchIndex;

            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);

            if (matchIndex+matchLength >= dictLimit)

                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */

        }

        if (matchLength > bestLength) {

            bestLength = matchLength;

            if (matchLength > matchEndIdx - matchIndex)

                matchEndIdx = matchIndex + (U32)matchLength;

        }

        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */

            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */

        }

        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */

            /* match is smaller than current */

            *smallerPtr = matchIndex;             /* update smaller idx */

            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */

            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */

            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */

            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */

        } else {

            /* match is larger than current */

            *largerPtr = matchIndex;

            commonLengthLarger = matchLength;

            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */

            largerPtr = nextPtr;

            matchIndex = nextPtr[0];