Quelle  zstd_compress_superblock.c

// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
/*
 * Copyright (c) Meta Platforms, Inc. and affiliates.
 * 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.
 */

 /*-*************************************
 *  Dependencies
 ***************************************/
#include "zstd_compress_superblock.h"

#include "../common/zstd_internal.h"  /* ZSTD_getSequenceLength */
#include "hist.h"                     /* HIST_countFast_wksp */
#include "zstd_compress_internal.h"   /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
#include "zstd_compress_sequences.h"
#include "zstd_compress_literals.h"

/* ZSTD_compressSubBlock_literal() :
 *  Compresses literals section for a sub-block.
 *  When we have to write the Huffman table we will sometimes choose a header
 *  size larger than necessary. This is because we have to pick the header size
 *  before we know the table size + compressed size, so we have a bound on the
 *  table size. If we guessed incorrectly, we fall back to uncompressed literals.
 *
 *  We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
 *  in writing the header, otherwise it is set to 0.
 *
 *  hufMetadata->hType has literals block type info.
 *      If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
 *      If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
 *      If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
 *      If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
 *      and the following sub-blocks' literals sections will be Treeless_Literals_Block.
 *  @return : compressed size of literals section of a sub-block
 *            Or 0 if unable to compress.
 *            Or error code */
static size_t
ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
                              const ZSTD_hufCTablesMetadata_t* hufMetadata,
                              const BYTE* literals, size_t litSize,
                              void* dst, size_t dstSize,
                              const int bmi2, int writeEntropy, int* entropyWritten)
{
    size_t const header = writeEntropy ? 200 : 0;
    size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
    BYTE* const ostart = (BYTE*)dst;
    BYTE* const oend = ostart + dstSize;
    BYTE* op = ostart + lhSize;
    U32 const singleStream = lhSize == 3;
    SymbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
    size_t cLitSize = 0;

    DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);

    *entropyWritten = 0;
    if (litSize == 0 || hufMetadata->hType == set_basic) {
      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
      return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
    } else if (hufMetadata->hType == set_rle) {
      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
      return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
    }

    assert(litSize > 0);
    assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);

    if (writeEntropy && hufMetadata->hType == set_compressed) {
        ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
        op += hufMetadata->hufDesSize;
        cLitSize += hufMetadata->hufDesSize;
        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
    }

    {   int const flags = bmi2 ? HUF_flags_bmi2 : 0;
        const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags)
                                          : HUF_compress4X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags);
        op += cSize;
        cLitSize += cSize;
        if (cSize == 0 || ERR_isError(cSize)) {
            DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
            return 0;
        }
        /* If we expand and we aren't writing a header then emit uncompressed */
        if (!writeEntropy && cLitSize >= litSize) {
            DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
        }
        /* If we are writing headers then allow expansion that doesn't change our header size. */
        if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
            assert(cLitSize > litSize);
            DEBUGLOG(5, "Literals expanded beyond allowed header size");
            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
        }
        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
    }

    /* Build header */
    switch(lhSize)
    {
    case 3: /* 2 - 2 - 10 - 10 */
        {   U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
            MEM_writeLE24(ostart, lhc);
            break;
        }
    case 4: /* 2 - 2 - 14 - 14 */
        {   U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
            MEM_writeLE32(ostart, lhc);
            break;
        }
    case 5: /* 2 - 2 - 18 - 18 */
        {   U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
            MEM_writeLE32(ostart, lhc);
            ostart[4] = (BYTE)(cLitSize >> 10);
            break;
        }
    default:  /* not possible : lhSize is {3,4,5} */
        assert(0);
    }
    *entropyWritten = 1;
    DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
    return (size_t)(op-ostart);
}

static size_t
ZSTD_seqDecompressedSize(SeqStore_t const* seqStore,
                   const SeqDef* sequences, size_t nbSeqs,
                         size_t litSize, int lastSubBlock)
{
    size_t matchLengthSum = 0;
    size_t litLengthSum = 0;
    size_t n;
    for (n=0; n<nbSeqs; n++) {
        const ZSTD_SequenceLength seqLen = ZSTD_getSequenceLength(seqStore, sequences+n);
        litLengthSum += seqLen.litLength;
        matchLengthSum += seqLen.matchLength;
    }
    DEBUGLOG(5, "ZSTD_seqDecompressedSize: %u sequences from %p: %u literals + %u matchlength",
                (unsigned)nbSeqs, (const void*)sequences,
                (unsigned)litLengthSum, (unsigned)matchLengthSum);
    if (!lastSubBlock)
        assert(litLengthSum == litSize);
    else
        assert(litLengthSum <= litSize);
    (void)litLengthSum;
    return matchLengthSum + litSize;
}

/* ZSTD_compressSubBlock_sequences() :
 *  Compresses sequences section for a sub-block.
 *  fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
 *  symbol compression modes for the super-block.
 *  The first successfully compressed block will have these in its header.
 *  We set entropyWritten=1 when we succeed in compressing the sequences.
 *  The following sub-blocks will always have repeat mode.
 *  @return : compressed size of sequences section of a sub-block
 *            Or 0 if it is unable to compress
 *            Or error code. */
static size_t
ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
                                const ZSTD_fseCTablesMetadata_t* fseMetadata,
                                const SeqDef* sequences, size_t nbSeq,
                                const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
                                const ZSTD_CCtx_params* cctxParams,
                                void* dst, size_t dstCapacity,
                                const int bmi2, int writeEntropy, int* entropyWritten)
{
    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
    BYTE* const ostart = (BYTE*)dst;
    BYTE* const oend = ostart + dstCapacity;
    BYTE* op = ostart;
    BYTE* seqHead;

    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);

    *entropyWritten = 0;
    /* Sequences Header */
    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
                    dstSize_tooSmall, "");
    if (nbSeq < 128)
        *op++ = (BYTE)nbSeq;
    else if (nbSeq < LONGNBSEQ)
        op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
    else
        op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
    if (nbSeq==0) {
        return (size_t)(op - ostart);
    }

    /* seqHead : flags for FSE encoding type */
    seqHead = op++;

    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));

    if (writeEntropy) {
        const U32 LLtype = fseMetadata->llType;
        const U32 Offtype = fseMetadata->ofType;
        const U32 MLtype = fseMetadata->mlType;
        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
        *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
        ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
        op += fseMetadata->fseTablesSize;
    } else {
        const U32 repeat = set_repeat;
        *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
    }

    {   size_t const bitstreamSize = ZSTD_encodeSequences(
                                        op, (size_t)(oend - op),
                                        fseTables->matchlengthCTable, mlCode,
                                        fseTables->offcodeCTable, ofCode,
                                        fseTables->litlengthCTable, llCode,
                                        sequences, nbSeq,
                                        longOffsets, bmi2);
        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
        op += bitstreamSize;
        /* zstd versions <= 1.3.4 mistakenly report corruption when
         * FSE_readNCount() receives a buffer < 4 bytes.
         * Fixed by https://github.com/facebook/zstd/pull/1146.
         * This can happen when the last set_compressed table present is 2
         * bytes and the bitstream is only one byte.
         * In this exceedingly rare case, we will simply emit an uncompressed
         * block, since it isn't worth optimizing.
         */
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
        if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
            /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
            assert(fseMetadata->lastCountSize + bitstreamSize == 3);
            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
                        "emitting an uncompressed block.");
            return 0;
        }
#endif
        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
    }

    /* zstd versions <= 1.4.0 mistakenly report error when
     * sequences section body size is less than 3 bytes.
     * Fixed by https://github.com/facebook/zstd/pull/1664.
     * This can happen when the previous sequences section block is compressed
     * with rle mode and the current block's sequences section is compressed
     * with repeat mode where sequences section body size can be 1 byte.
     */
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
    if (op-seqHead < 4) {
        DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
                    "an uncompressed block when sequences are < 4 bytes");
        return 0;
    }
#endif

    *entropyWritten = 1;
    return (size_t)(op - ostart);
}

/* ZSTD_compressSubBlock() :
 *  Compresses a single sub-block.
 *  @return : compressed size of the sub-block
 *            Or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
                                    const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
                                    const SeqDef* sequences, size_t nbSeq,
                                    const BYTE* literals, size_t litSize,
                                    const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
                                    const ZSTD_CCtx_params* cctxParams,
                                    void* dst, size_t dstCapacity,
                                    const int bmi2,
                                    int writeLitEntropy, int writeSeqEntropy,
                                    int* litEntropyWritten, int* seqEntropyWritten,
                                    U32 lastBlock)
{
    BYTE* const ostart = (BYTE*)dst;
    BYTE* const oend = ostart + dstCapacity;
    BYTE* op = ostart + ZSTD_blockHeaderSize;
    DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
                litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
    {   size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
                                                        &entropyMetadata->hufMetadata, literals, litSize,
                                                        op, (size_t)(oend-op),
                                                        bmi2, writeLitEntropy, litEntropyWritten);
        FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
        if (cLitSize == 0) return 0;
        op += cLitSize;
    }
    {   size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
                                                  &entropyMetadata->fseMetadata,
                                                  sequences, nbSeq,
                                                  llCode, mlCode, ofCode,
                                                  cctxParams,
                                                  op, (size_t)(oend-op),
                                                  bmi2, writeSeqEntropy, seqEntropyWritten);
        FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
        if (cSeqSize == 0) return 0;
        op += cSeqSize;
    }
    /* Write block header */
    {   size_t cSize = (size_t)(op-ostart) - ZSTD_blockHeaderSize;
        U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
        MEM_writeLE24(ostart, cBlockHeader24);
    }
    return (size_t)(op-ostart);
}

static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
                                                const ZSTD_hufCTables_t* huf,
                                                const ZSTD_hufCTablesMetadata_t* hufMetadata,
                                                void* workspace, size_t wkspSize,
                                                int writeEntropy)
{
    unsigned* const countWksp = (unsigned*)workspace;
    unsigned maxSymbolValue = 255;
    size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */

    if (hufMetadata->hType == set_basic) return litSize;
    else if (hufMetadata->hType == set_rle) return 1;
    else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
        size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
        if (ZSTD_isError(largest)) return litSize;
        {   size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
            if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
            return cLitSizeEstimate + literalSectionHeaderSize;
    }   }
    assert(0); /* impossible */
    return 0;
}

static size_t ZSTD_estimateSubBlockSize_symbolType(SymbolEncodingType_e type,
                        const BYTE* codeTable, unsigned maxCode,
                        size_t nbSeq, const FSE_CTable* fseCTable,
                        const U8* additionalBits,
                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
                        void* workspace, size_t wkspSize)
{
    unsigned* const countWksp = (unsigned*)workspace;
    const BYTE* ctp = codeTable;
    const BYTE* const ctStart = ctp;
    const BYTE* const ctEnd = ctStart + nbSeq;
    size_t cSymbolTypeSizeEstimateInBits = 0;
    unsigned max = maxCode;

    HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
    if (type == set_basic) {
        /* We selected this encoding type, so it must be valid. */
        assert(max <= defaultMax);
        cSymbolTypeSizeEstimateInBits = max <= defaultMax
                ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
                : ERROR(GENERIC);
    } else if (type == set_rle) {
        cSymbolTypeSizeEstimateInBits = 0;
    } else if (type == set_compressed || type == set_repeat) {
        cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
    }
    if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
    while (ctp < ctEnd) {
        if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
        else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
        ctp++;
    }
    return cSymbolTypeSizeEstimateInBits / 8;
}

static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
                                                  const BYTE* llCodeTable,
                                                  const BYTE* mlCodeTable,
                                                  size_t nbSeq,
                                                  const ZSTD_fseCTables_t* fseTables,
                                                  const ZSTD_fseCTablesMetadata_t* fseMetadata,
                                                  void* workspace, size_t wkspSize,
                                                  int writeEntropy)
{
    size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
    size_t cSeqSizeEstimate = 0;
    if (nbSeq == 0) return sequencesSectionHeaderSize;
    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
                                         nbSeq, fseTables->offcodeCTable, NULL,
                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
                                         workspace, wkspSize);
    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
                                         nbSeq, fseTables->litlengthCTable, LL_bits,
                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
                                         workspace, wkspSize);
    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
                                         nbSeq, fseTables->matchlengthCTable, ML_bits,
                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
                                         workspace, wkspSize);
    if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
    return cSeqSizeEstimate + sequencesSectionHeaderSize;
}

typedef struct {
    size_t estLitSize;
    size_t estBlockSize;
} EstimatedBlockSize;
static EstimatedBlockSize ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
                                        const BYTE* ofCodeTable,
                                        const BYTE* llCodeTable,
                                        const BYTE* mlCodeTable,
                                        size_t nbSeq,
                                        const ZSTD_entropyCTables_t* entropy,
                                        const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
                                        void* workspace, size_t wkspSize,
                                        int writeLitEntropy, int writeSeqEntropy)
{
    EstimatedBlockSize ebs;
    ebs.estLitSize = ZSTD_estimateSubBlockSize_literal(literals, litSize,
                                                        &entropy->huf, &entropyMetadata->hufMetadata,
                                                        workspace, wkspSize, writeLitEntropy);
    ebs.estBlockSize = ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
                                                         nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
                                                         workspace, wkspSize, writeSeqEntropy);
    ebs.estBlockSize += ebs.estLitSize + ZSTD_blockHeaderSize;
    return ebs;
}

static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
{
    if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
        return 1;
    if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
        return 1;
    if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
        return 1;
    return 0;
}

static size_t countLiterals(SeqStore_t const* seqStore, const SeqDef* sp, size_t seqCount)
{
    size_t n, total = 0;
    assert(sp != NULL);
    for (n=0; n<seqCount; n++) {
        total += ZSTD_getSequenceLength(seqStore, sp+n).litLength;
    }
    DEBUGLOG(6, "countLiterals for %zu sequences from %p => %zu bytes", seqCount, (const void*)sp, total);
    return total;
}

#define BYTESCALE 256

static size_t sizeBlockSequences(const SeqDef* sp, size_t nbSeqs,
                size_t targetBudget, size_t avgLitCost, size_t avgSeqCost,
                int firstSubBlock)
{
    size_t n, budget = 0, inSize=0;
    /* entropy headers */
    size_t const headerSize = (size_t)firstSubBlock * 120 * BYTESCALE; /* generous estimate */
    assert(firstSubBlock==0 || firstSubBlock==1);
    budget += headerSize;

    /* first sequence => at least one sequence*/
    budget += sp[0].litLength * avgLitCost + avgSeqCost;
    if (budget > targetBudget) return 1;
    inSize = sp[0].litLength + (sp[0].mlBase+MINMATCH);

    /* loop over sequences */
    for (n=1; n<nbSeqs; n++) {
        size_t currentCost = sp[n].litLength * avgLitCost + avgSeqCost;
        budget += currentCost;
        inSize += sp[n].litLength + (sp[n].mlBase+MINMATCH);
        /* stop when sub-block budget is reached */
        if ( (budget > targetBudget)
            /* though continue to expand until the sub-block is deemed compressible */
          && (budget < inSize * BYTESCALE) )
            break;
    }

    return n;
}

/* ZSTD_compressSubBlock_multi() :
 *  Breaks super-block into multiple sub-blocks and compresses them.
 *  Entropy will be written into the first block.
 *  The following blocks use repeat_mode to compress.
 *  Sub-blocks are all compressed, except the last one when beneficial.
 *  @return : compressed size of the super block (which features multiple ZSTD blocks)
 *            or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock_multi(const SeqStore_t* seqStorePtr,
                            const ZSTD_compressedBlockState_t* prevCBlock,
                            ZSTD_compressedBlockState_t* nextCBlock,
                            const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
                            const ZSTD_CCtx_params* cctxParams,
                                  void* dst, size_t dstCapacity,
                            const void* src, size_t srcSize,
                            const int bmi2, U32 lastBlock,
                            void* workspace, size_t wkspSize)
{
    const SeqDef* const sstart = seqStorePtr->sequencesStart;
    const SeqDef* const send = seqStorePtr->sequences;
    const SeqDef* sp = sstart; /* tracks progresses within seqStorePtr->sequences */
    size_t const nbSeqs = (size_t)(send - sstart);
    const BYTE* const lstart = seqStorePtr->litStart;
    const BYTE* const lend = seqStorePtr->lit;
    const BYTE* lp = lstart;
    size_t const nbLiterals = (size_t)(lend - lstart);
    BYTE const* ip = (BYTE const*)src;
    BYTE const* const iend = ip + srcSize;
    BYTE* const ostart = (BYTE*)dst;
    BYTE* const oend = ostart + dstCapacity;
    BYTE* op = ostart;
    const BYTE* llCodePtr = seqStorePtr->llCode;
    const BYTE* mlCodePtr = seqStorePtr->mlCode;
    const BYTE* ofCodePtr = seqStorePtr->ofCode;
    size_t const minTarget = ZSTD_TARGETCBLOCKSIZE_MIN; /* enforce minimum size, to reduce undesirable side effects */
    size_t const targetCBlockSize = MAX(minTarget, cctxParams->targetCBlockSize);
    int writeLitEntropy = (entropyMetadata->hufMetadata.hType == set_compressed);
    int writeSeqEntropy = 1;

    DEBUGLOG(5, "ZSTD_compressSubBlock_multi (srcSize=%u, litSize=%u, nbSeq=%u)",
               (unsigned)srcSize, (unsigned)(lend-lstart), (unsigned)(send-sstart));

        /* let's start by a general estimation for the full block */
    if (nbSeqs > 0) {
        EstimatedBlockSize const ebs =
                ZSTD_estimateSubBlockSize(lp, nbLiterals,
                                        ofCodePtr, llCodePtr, mlCodePtr, nbSeqs,
                                        &nextCBlock->entropy, entropyMetadata,
                                        workspace, wkspSize,
                                        writeLitEntropy, writeSeqEntropy);
        /* quick estimation */
        size_t const avgLitCost = nbLiterals ? (ebs.estLitSize * BYTESCALE) / nbLiterals : BYTESCALE;
        size_t const avgSeqCost = ((ebs.estBlockSize - ebs.estLitSize) * BYTESCALE) / nbSeqs;
        const size_t nbSubBlocks = MAX((ebs.estBlockSize + (targetCBlockSize/2)) / targetCBlockSize, 1);
        size_t n, avgBlockBudget, blockBudgetSupp=0;
        avgBlockBudget = (ebs.estBlockSize * BYTESCALE) / nbSubBlocks;
        DEBUGLOG(5, "estimated fullblock size=%u bytes ; avgLitCost=%.2f ; avgSeqCost=%.2f ; targetCBlockSize=%u, nbSubBlocks=%u ; avgBlockBudget=%.0f bytes",
                    (unsigned)ebs.estBlockSize, (double)avgLitCost/BYTESCALE, (double)avgSeqCost/BYTESCALE,
                    (unsigned)targetCBlockSize, (unsigned)nbSubBlocks, (double)avgBlockBudget/BYTESCALE);
        /* simplification: if estimates states that the full superblock doesn't compress, just bail out immediately
         * this will result in the production of a single uncompressed block covering @srcSize.*/
        if (ebs.estBlockSize > srcSize) return 0;

        /* compress and write sub-blocks */
        assert(nbSubBlocks>0);
        for (n=0; n < nbSubBlocks-1; n++) {
            /* determine nb of sequences for current sub-block + nbLiterals from next sequence */
            size_t const seqCount = sizeBlockSequences(sp, (size_t)(send-sp),
                                        avgBlockBudget + blockBudgetSupp, avgLitCost, avgSeqCost, n==0);
            /* if reached last sequence : break to last sub-block (simplification) */
            assert(seqCount <= (size_t)(send-sp));
            if (sp + seqCount == send) break;
            assert(seqCount > 0);
            /* compress sub-block */
            {   int litEntropyWritten = 0;
                int seqEntropyWritten = 0;
                size_t litSize = countLiterals(seqStorePtr, sp, seqCount);
                const size_t decompressedSize =
                        ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 0);
                size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
                                                sp, seqCount,
                                                lp, litSize,
                                                llCodePtr, mlCodePtr, ofCodePtr,
                                                cctxParams,
                                                op, (size_t)(oend-op),
                                                bmi2, writeLitEntropy, writeSeqEntropy,
                                                &litEntropyWritten, &seqEntropyWritten,
                                                0);
                FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");

                /* check compressibility, update state components */
                if (cSize > 0 && cSize < decompressedSize) {
                    DEBUGLOG(5, "Committed sub-block compressing %u bytes => %u bytes",
                                (unsigned)decompressedSize, (unsigned)cSize);
                    assert(ip + decompressedSize <= iend);
                    ip += decompressedSize;
                    lp += litSize;
                    op += cSize;
                    llCodePtr += seqCount;
                    mlCodePtr += seqCount;
                    ofCodePtr += seqCount;
                    /* Entropy only needs to be written once */
                    if (litEntropyWritten) {
                        writeLitEntropy = 0;
                    }
                    if (seqEntropyWritten) {
                        writeSeqEntropy = 0;
                    }
                    sp += seqCount;
                    blockBudgetSupp = 0;
            }   }
            /* otherwise : do not compress yet, coalesce current sub-block with following one */
        }
    } /* if (nbSeqs > 0) */

    /* write last block */
    DEBUGLOG(5, "Generate last sub-block: %u sequences remaining", (unsigned)(send - sp));
    {   int litEntropyWritten = 0;
        int seqEntropyWritten = 0;
        size_t litSize = (size_t)(lend - lp);
        size_t seqCount = (size_t)(send - sp);
        const size_t decompressedSize =
                ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 1);
        size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
                                            sp, seqCount,
                                            lp, litSize,
                                            llCodePtr, mlCodePtr, ofCodePtr,
                                            cctxParams,
                                            op, (size_t)(oend-op),
                                            bmi2, writeLitEntropy, writeSeqEntropy,
                                            &litEntropyWritten, &seqEntropyWritten,
                                            lastBlock);
        FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");

        /* update pointers, the nb of literals borrowed from next sequence must be preserved */
        if (cSize > 0 && cSize < decompressedSize) {
            DEBUGLOG(5, "Last sub-block compressed %u bytes => %u bytes",
                        (unsigned)decompressedSize, (unsigned)cSize);
            assert(ip + decompressedSize <= iend);
            ip += decompressedSize;
            lp += litSize;
            op += cSize;
            llCodePtr += seqCount;
            mlCodePtr += seqCount;
            ofCodePtr += seqCount;
            /* Entropy only needs to be written once */
            if (litEntropyWritten) {
                writeLitEntropy = 0;
            }
            if (seqEntropyWritten) {
                writeSeqEntropy = 0;
            }
            sp += seqCount;
        }
    }


    if (writeLitEntropy) {
        DEBUGLOG(5, "Literal entropy tables were never written");
        ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
    }
    if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
        /* If we haven't written our entropy tables, then we've violated our contract and
         * must emit an uncompressed block.
         */
        DEBUGLOG(5, "Sequence entropy tables were never written => cancel, emit an uncompressed block");
        return 0;
    }

    if (ip < iend) {
        /* some data left : last part of the block sent uncompressed */
        size_t const rSize = (size_t)((iend - ip));
        size_t const cSize = ZSTD_noCompressBlock(op, (size_t)(oend - op), ip, rSize, lastBlock);
        DEBUGLOG(5, "Generate last uncompressed sub-block of %u bytes", (unsigned)(rSize));
        FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
        assert(cSize != 0);
        op += cSize;
        /* We have to regenerate the repcodes because we've skipped some sequences */
        if (sp < send) {
            const SeqDef* seq;
            Repcodes_t rep;
            ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
            for (seq = sstart; seq < sp; ++seq) {
                ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
            }
            ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
        }
    }

    DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed all subBlocks: total compressed size = %u",
                (unsigned)(op-ostart));
    return (size_t)(op-ostart);
}

size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
                               void* dst, size_t dstCapacity,
                               const void* src, size_t srcSize,
                               unsigned lastBlock)
{
    ZSTD_entropyCTablesMetadata_t entropyMetadata;

    FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
          &zc->blockState.prevCBlock->entropy,
          &zc->blockState.nextCBlock->entropy,
          &zc->appliedParams,
          &entropyMetadata,
          zc->tmpWorkspace, zc->tmpWkspSize /* statically allocated in resetCCtx */), "");

    return ZSTD_compressSubBlock_multi(&zc->seqStore,
            zc->blockState.prevCBlock,
            zc->blockState.nextCBlock,
            &entropyMetadata,
            &zc->appliedParams,
            dst, dstCapacity,
            src, srcSize,
            zc->bmi2, lastBlock,
            zc->tmpWorkspace, zc->tmpWkspSize /* statically allocated in resetCCtx */);
}