tor-browser

The Tor Browser
git clone https://git.dasho.dev/tor-browser.git
Log | Files | Refs | README | LICENSE

backward_references_inc.h (8181B)


      1 /* NOLINT(build/header_guard) */
      2 /* Copyright 2013 Google Inc. All Rights Reserved.
      3 
      4   Distributed under MIT license.
      5   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
      6 */
      7 
      8 /* template parameters: EXPORT_FN, FN */
      9 
     10 static BROTLI_NOINLINE void EXPORT_FN(CreateBackwardReferences)(
     11    size_t num_bytes, size_t position,
     12    const uint8_t* ringbuffer, size_t ringbuffer_mask,
     13    ContextLut literal_context_lut, const BrotliEncoderParams* params,
     14    Hasher* hasher, int* dist_cache, size_t* last_insert_len,
     15    Command* commands, size_t* num_commands, size_t* num_literals) {
     16  HASHER()* privat = &hasher->privat.FN(_);
     17  /* Set maximum distance, see section 9.1. of the spec. */
     18  const size_t max_backward_limit = BROTLI_MAX_BACKWARD_LIMIT(params->lgwin);
     19  const size_t position_offset = params->stream_offset;
     20 
     21  const Command* const orig_commands = commands;
     22  size_t insert_length = *last_insert_len;
     23  const size_t pos_end = position + num_bytes;
     24  const size_t store_end = num_bytes >= FN(StoreLookahead)() ?
     25      position + num_bytes - FN(StoreLookahead)() + 1 : position;
     26 
     27  /* For speed up heuristics for random data. */
     28  const size_t random_heuristics_window_size =
     29      LiteralSpreeLengthForSparseSearch(params);
     30  size_t apply_random_heuristics = position + random_heuristics_window_size;
     31  const size_t gap = params->dictionary.compound.total_size;
     32 
     33  /* Minimum score to accept a backward reference. */
     34  const score_t kMinScore = BROTLI_SCORE_BASE + 100;
     35 
     36  FN(PrepareDistanceCache)(privat, dist_cache);
     37 
     38  while (position + FN(HashTypeLength)() < pos_end) {
     39    size_t max_length = pos_end - position;
     40    size_t max_distance = BROTLI_MIN(size_t, position, max_backward_limit);
     41    size_t dictionary_start = BROTLI_MIN(size_t,
     42        position + position_offset, max_backward_limit);
     43    HasherSearchResult sr;
     44    int dict_id = 0;
     45    uint8_t p1 = 0;
     46    uint8_t p2 = 0;
     47    if (params->dictionary.contextual.context_based) {
     48      p1 = position >= 1 ?
     49          ringbuffer[(size_t)(position - 1) & ringbuffer_mask] : 0;
     50      p2 = position >= 2 ?
     51          ringbuffer[(size_t)(position - 2) & ringbuffer_mask] : 0;
     52      dict_id = params->dictionary.contextual.context_map[
     53          BROTLI_CONTEXT(p1, p2, literal_context_lut)];
     54    }
     55    sr.len = 0;
     56    sr.len_code_delta = 0;
     57    sr.distance = 0;
     58    sr.score = kMinScore;
     59    FN(FindLongestMatch)(privat, params->dictionary.contextual.dict[dict_id],
     60        ringbuffer, ringbuffer_mask, dist_cache, position, max_length,
     61        max_distance, dictionary_start + gap, params->dist.max_distance, &sr);
     62    if (ENABLE_COMPOUND_DICTIONARY) {
     63      LookupCompoundDictionaryMatch(&params->dictionary.compound, ringbuffer,
     64          ringbuffer_mask, dist_cache, position, max_length,
     65          dictionary_start, params->dist.max_distance, &sr);
     66    }
     67    if (sr.score > kMinScore) {
     68      /* Found a match. Let's look for something even better ahead. */
     69      int delayed_backward_references_in_row = 0;
     70      --max_length;
     71      for (;; --max_length) {
     72        const score_t cost_diff_lazy = 175;
     73        HasherSearchResult sr2;
     74        sr2.len = params->quality < MIN_QUALITY_FOR_EXTENSIVE_REFERENCE_SEARCH ?
     75            BROTLI_MIN(size_t, sr.len - 1, max_length) : 0;
     76        sr2.len_code_delta = 0;
     77        sr2.distance = 0;
     78        sr2.score = kMinScore;
     79        max_distance = BROTLI_MIN(size_t, position + 1, max_backward_limit);
     80        dictionary_start = BROTLI_MIN(size_t,
     81            position + 1 + position_offset, max_backward_limit);
     82        if (params->dictionary.contextual.context_based) {
     83          p2 = p1;
     84          p1 = ringbuffer[position & ringbuffer_mask];
     85          dict_id = params->dictionary.contextual.context_map[
     86              BROTLI_CONTEXT(p1, p2, literal_context_lut)];
     87        }
     88        FN(FindLongestMatch)(privat,
     89            params->dictionary.contextual.dict[dict_id],
     90            ringbuffer, ringbuffer_mask, dist_cache, position + 1, max_length,
     91            max_distance, dictionary_start + gap, params->dist.max_distance,
     92            &sr2);
     93        if (ENABLE_COMPOUND_DICTIONARY) {
     94          LookupCompoundDictionaryMatch(
     95              &params->dictionary.compound, ringbuffer,
     96              ringbuffer_mask, dist_cache, position + 1, max_length,
     97              dictionary_start, params->dist.max_distance, &sr2);
     98        }
     99        if (sr2.score >= sr.score + cost_diff_lazy) {
    100          /* Ok, let's just write one byte for now and start a match from the
    101             next byte. */
    102          ++position;
    103          ++insert_length;
    104          sr = sr2;
    105          if (++delayed_backward_references_in_row < 4 &&
    106              position + FN(HashTypeLength)() < pos_end) {
    107            continue;
    108          }
    109        }
    110        break;
    111      }
    112      apply_random_heuristics =
    113          position + 2 * sr.len + random_heuristics_window_size;
    114      dictionary_start = BROTLI_MIN(size_t,
    115          position + position_offset, max_backward_limit);
    116      {
    117        /* The first 16 codes are special short-codes,
    118           and the minimum offset is 1. */
    119        size_t distance_code = ComputeDistanceCode(
    120            sr.distance, dictionary_start + gap, dist_cache);
    121        if ((sr.distance <= (dictionary_start + gap)) && distance_code > 0) {
    122          dist_cache[3] = dist_cache[2];
    123          dist_cache[2] = dist_cache[1];
    124          dist_cache[1] = dist_cache[0];
    125          dist_cache[0] = (int)sr.distance;
    126          FN(PrepareDistanceCache)(privat, dist_cache);
    127        }
    128        InitCommand(commands++, &params->dist, insert_length,
    129            sr.len, sr.len_code_delta, distance_code);
    130      }
    131      *num_literals += insert_length;
    132      insert_length = 0;
    133      /* Put the hash keys into the table, if there are enough bytes left.
    134         Depending on the hasher implementation, it can push all positions
    135         in the given range or only a subset of them.
    136         Avoid hash poisoning with RLE data. */
    137      {
    138        size_t range_start = position + 2;
    139        size_t range_end = BROTLI_MIN(size_t, position + sr.len, store_end);
    140        if (sr.distance < (sr.len >> 2)) {
    141          range_start = BROTLI_MIN(size_t, range_end, BROTLI_MAX(size_t,
    142              range_start, position + sr.len - (sr.distance << 2)));
    143        }
    144        FN(StoreRange)(privat, ringbuffer, ringbuffer_mask, range_start,
    145                       range_end);
    146      }
    147      position += sr.len;
    148    } else {
    149      ++insert_length;
    150      ++position;
    151      /* If we have not seen matches for a long time, we can skip some
    152         match lookups. Unsuccessful match lookups are very very expensive
    153         and this kind of a heuristic speeds up compression quite
    154         a lot. */
    155      if (position > apply_random_heuristics) {
    156        /* Going through uncompressible data, jump. */
    157        if (position >
    158            apply_random_heuristics + 4 * random_heuristics_window_size) {
    159          /* It is quite a long time since we saw a copy, so we assume
    160             that this data is not compressible, and store hashes less
    161             often. Hashes of non compressible data are less likely to
    162             turn out to be useful in the future, too, so we store less of
    163             them to not to flood out the hash table of good compressible
    164             data. */
    165          const size_t kMargin =
    166              BROTLI_MAX(size_t, FN(StoreLookahead)() - 1, 4);
    167          size_t pos_jump =
    168              BROTLI_MIN(size_t, position + 16, pos_end - kMargin);
    169          for (; position < pos_jump; position += 4) {
    170            FN(Store)(privat, ringbuffer, ringbuffer_mask, position);
    171            insert_length += 4;
    172          }
    173        } else {
    174          const size_t kMargin =
    175              BROTLI_MAX(size_t, FN(StoreLookahead)() - 1, 2);
    176          size_t pos_jump =
    177              BROTLI_MIN(size_t, position + 8, pos_end - kMargin);
    178          for (; position < pos_jump; position += 2) {
    179            FN(Store)(privat, ringbuffer, ringbuffer_mask, position);
    180            insert_length += 2;
    181          }
    182        }
    183      }
    184    }
    185  }
    186  insert_length += pos_end - position;
    187  *last_insert_len = insert_length;
    188  *num_commands += (size_t)(commands - orig_commands);
    189 }