BaseTools: Update Brotli Compress to the latest one 1.0.6

[mirror_edk2.git] / BaseTools / Source / C / BrotliCompress / enc / block_splitter_inc.h

/* NOLINT(build/header_guard) */\r
/* Copyright 2013 Google Inc. All Rights Reserved.\r
\r
   Distributed under MIT license.\r
   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT\r
*/\r
\r
/* template parameters: FN, DataType */\r
\r
#define HistogramType FN(Histogram)\r
\r
static void FN(InitialEntropyCodes)(const DataType* data, size_t length,\r
                                    size_t stride,\r
                                    size_t num_histograms,\r
                                    HistogramType* histograms) {\r
  uint32_t seed = 7;\r
  size_t block_length = length / num_histograms;\r
  size_t i;\r
  FN(ClearHistograms)(histograms, num_histograms);\r
  for (i = 0; i < num_histograms; ++i) {\r
    size_t pos = length * i / num_histograms;\r
    if (i != 0) {\r
      pos += MyRand(&seed) % block_length;\r
    }\r
    if (pos + stride >= length) {\r
      pos = length - stride - 1;\r
    }\r
    FN(HistogramAddVector)(&histograms[i], data + pos, stride);\r
  }\r
}\r
\r
static void FN(RandomSample)(uint32_t* seed,\r
                             const DataType* data,\r
                             size_t length,\r
                             size_t stride,\r
                             HistogramType* sample) {\r
  size_t pos = 0;\r
  if (stride >= length) {\r
    stride = length;\r
  } else {\r
    pos = MyRand(seed) % (length - stride + 1);\r
  }\r
  FN(HistogramAddVector)(sample, data + pos, stride);\r
}\r
\r
static void FN(RefineEntropyCodes)(const DataType* data, size_t length,\r
                                   size_t stride,\r
                                   size_t num_histograms,\r
                                   HistogramType* histograms) {\r
  size_t iters =\r
      kIterMulForRefining * length / stride + kMinItersForRefining;\r
  uint32_t seed = 7;\r
  size_t iter;\r
  iters = ((iters + num_histograms - 1) / num_histograms) * num_histograms;\r
  for (iter = 0; iter < iters; ++iter) {\r
    HistogramType sample;\r
    FN(HistogramClear)(&sample);\r
    FN(RandomSample)(&seed, data, length, stride, &sample);\r
    FN(HistogramAddHistogram)(&histograms[iter % num_histograms], &sample);\r
  }\r
}\r
\r
/* Assigns a block id from the range [0, num_histograms) to each data element\r
   in data[0..length) and fills in block_id[0..length) with the assigned values.\r
   Returns the number of blocks, i.e. one plus the number of block switches. */\r
static size_t FN(FindBlocks)(const DataType* data, const size_t length,\r
                             const double block_switch_bitcost,\r
                             const size_t num_histograms,\r
                             const HistogramType* histograms,\r
                             double* insert_cost,\r
                             double* cost,\r
                             uint8_t* switch_signal,\r
                             uint8_t* block_id) {\r
  const size_t data_size = FN(HistogramDataSize)();\r
  const size_t bitmaplen = (num_histograms + 7) >> 3;\r
  size_t num_blocks = 1;\r
  size_t i;\r
  size_t j;\r
  BROTLI_DCHECK(num_histograms <= 256);\r
  if (num_histograms <= 1) {\r
    for (i = 0; i < length; ++i) {\r
      block_id[i] = 0;\r
    }\r
    return 1;\r
  }\r
  memset(insert_cost, 0, sizeof(insert_cost[0]) * data_size * num_histograms);\r
  for (i = 0; i < num_histograms; ++i) {\r
    insert_cost[i] = FastLog2((uint32_t)histograms[i].total_count_);\r
  }\r
  for (i = data_size; i != 0;) {\r
    --i;\r
    for (j = 0; j < num_histograms; ++j) {\r
      insert_cost[i * num_histograms + j] =\r
          insert_cost[j] - BitCost(histograms[j].data_[i]);\r
    }\r
  }\r
  memset(cost, 0, sizeof(cost[0]) * num_histograms);\r
  memset(switch_signal, 0, sizeof(switch_signal[0]) * length * bitmaplen);\r
  /* After each iteration of this loop, cost[k] will contain the difference\r
     between the minimum cost of arriving at the current byte position using\r
     entropy code k, and the minimum cost of arriving at the current byte\r
     position. This difference is capped at the block switch cost, and if it\r
     reaches block switch cost, it means that when we trace back from the last\r
     position, we need to switch here. */\r
  for (i = 0; i < length; ++i) {\r
    const size_t byte_ix = i;\r
    size_t ix = byte_ix * bitmaplen;\r
    size_t insert_cost_ix = data[byte_ix] * num_histograms;\r
    double min_cost = 1e99;\r
    double block_switch_cost = block_switch_bitcost;\r
    size_t k;\r
    for (k = 0; k < num_histograms; ++k) {\r
      /* We are coding the symbol in data[byte_ix] with entropy code k. */\r
      cost[k] += insert_cost[insert_cost_ix + k];\r
      if (cost[k] < min_cost) {\r
        min_cost = cost[k];\r
        block_id[byte_ix] = (uint8_t)k;\r
      }\r
    }\r
    /* More blocks for the beginning. */\r
    if (byte_ix < 2000) {\r
      block_switch_cost *= 0.77 + 0.07 * (double)byte_ix / 2000;\r
    }\r
    for (k = 0; k < num_histograms; ++k) {\r
      cost[k] -= min_cost;\r
      if (cost[k] >= block_switch_cost) {\r
        const uint8_t mask = (uint8_t)(1u << (k & 7));\r
        cost[k] = block_switch_cost;\r
        BROTLI_DCHECK((k >> 3) < bitmaplen);\r
        switch_signal[ix + (k >> 3)] |= mask;\r
      }\r
    }\r
  }\r
  {  /* Trace back from the last position and switch at the marked places. */\r
    size_t byte_ix = length - 1;\r
    size_t ix = byte_ix * bitmaplen;\r
    uint8_t cur_id = block_id[byte_ix];\r
    while (byte_ix > 0) {\r
      const uint8_t mask = (uint8_t)(1u << (cur_id & 7));\r
      BROTLI_DCHECK(((size_t)cur_id >> 3) < bitmaplen);\r
      --byte_ix;\r
      ix -= bitmaplen;\r
      if (switch_signal[ix + (cur_id >> 3)] & mask) {\r
        if (cur_id != block_id[byte_ix]) {\r
          cur_id = block_id[byte_ix];\r
          ++num_blocks;\r
        }\r
      }\r
      block_id[byte_ix] = cur_id;\r
    }\r
  }\r
  return num_blocks;\r
}\r
\r
static size_t FN(RemapBlockIds)(uint8_t* block_ids, const size_t length,\r
                                uint16_t* new_id, const size_t num_histograms) {\r
  static const uint16_t kInvalidId = 256;\r
  uint16_t next_id = 0;\r
  size_t i;\r
  for (i = 0; i < num_histograms; ++i) {\r
    new_id[i] = kInvalidId;\r
  }\r
  for (i = 0; i < length; ++i) {\r
    BROTLI_DCHECK(block_ids[i] < num_histograms);\r
    if (new_id[block_ids[i]] == kInvalidId) {\r
      new_id[block_ids[i]] = next_id++;\r
    }\r
  }\r
  for (i = 0; i < length; ++i) {\r
    block_ids[i] = (uint8_t)new_id[block_ids[i]];\r
    BROTLI_DCHECK(block_ids[i] < num_histograms);\r
  }\r
  BROTLI_DCHECK(next_id <= num_histograms);\r
  return next_id;\r
}\r
\r
static void FN(BuildBlockHistograms)(const DataType* data, const size_t length,\r
                                     const uint8_t* block_ids,\r
                                     const size_t num_histograms,\r
                                     HistogramType* histograms) {\r
  size_t i;\r
  FN(ClearHistograms)(histograms, num_histograms);\r
  for (i = 0; i < length; ++i) {\r
    FN(HistogramAdd)(&histograms[block_ids[i]], data[i]);\r
  }\r
}\r
\r
static void FN(ClusterBlocks)(MemoryManager* m,\r
                              const DataType* data, const size_t length,\r
                              const size_t num_blocks,\r
                              uint8_t* block_ids,\r
                              BlockSplit* split) {\r
  uint32_t* histogram_symbols = BROTLI_ALLOC(m, uint32_t, num_blocks);\r
  uint32_t* block_lengths = BROTLI_ALLOC(m, uint32_t, num_blocks);\r
  const size_t expected_num_clusters = CLUSTERS_PER_BATCH *\r
      (num_blocks + HISTOGRAMS_PER_BATCH - 1) / HISTOGRAMS_PER_BATCH;\r
  size_t all_histograms_size = 0;\r
  size_t all_histograms_capacity = expected_num_clusters;\r
  HistogramType* all_histograms =\r
      BROTLI_ALLOC(m, HistogramType, all_histograms_capacity);\r
  size_t cluster_size_size = 0;\r
  size_t cluster_size_capacity = expected_num_clusters;\r
  uint32_t* cluster_size = BROTLI_ALLOC(m, uint32_t, cluster_size_capacity);\r
  size_t num_clusters = 0;\r
  HistogramType* histograms = BROTLI_ALLOC(m, HistogramType,\r
      BROTLI_MIN(size_t, num_blocks, HISTOGRAMS_PER_BATCH));\r
  size_t max_num_pairs =\r
      HISTOGRAMS_PER_BATCH * HISTOGRAMS_PER_BATCH / 2;\r
  size_t pairs_capacity = max_num_pairs + 1;\r
  HistogramPair* pairs = BROTLI_ALLOC(m, HistogramPair, pairs_capacity);\r
  size_t pos = 0;\r
  uint32_t* clusters;\r
  size_t num_final_clusters;\r
  static const uint32_t kInvalidIndex = BROTLI_UINT32_MAX;\r
  uint32_t* new_index;\r
  size_t i;\r
  uint32_t sizes[HISTOGRAMS_PER_BATCH] = { 0 };\r
  uint32_t new_clusters[HISTOGRAMS_PER_BATCH] = { 0 };\r
  uint32_t symbols[HISTOGRAMS_PER_BATCH] = { 0 };\r
  uint32_t remap[HISTOGRAMS_PER_BATCH] = { 0 };\r
\r
  if (BROTLI_IS_OOM(m)) return;\r
\r
  memset(block_lengths, 0, num_blocks * sizeof(uint32_t));\r
\r
  {\r
    size_t block_idx = 0;\r
    for (i = 0; i < length; ++i) {\r
      BROTLI_DCHECK(block_idx < num_blocks);\r
      ++block_lengths[block_idx];\r
      if (i + 1 == length || block_ids[i] != block_ids[i + 1]) {\r
        ++block_idx;\r
      }\r
    }\r
    BROTLI_DCHECK(block_idx == num_blocks);\r
  }\r
\r
  for (i = 0; i < num_blocks; i += HISTOGRAMS_PER_BATCH) {\r
    const size_t num_to_combine =\r
        BROTLI_MIN(size_t, num_blocks - i, HISTOGRAMS_PER_BATCH);\r
    size_t num_new_clusters;\r
    size_t j;\r
    for (j = 0; j < num_to_combine; ++j) {\r
      size_t k;\r
      FN(HistogramClear)(&histograms[j]);\r
      for (k = 0; k < block_lengths[i + j]; ++k) {\r
        FN(HistogramAdd)(&histograms[j], data[pos++]);\r
      }\r
      histograms[j].bit_cost_ = FN(BrotliPopulationCost)(&histograms[j]);\r
      new_clusters[j] = (uint32_t)j;\r
      symbols[j] = (uint32_t)j;\r
      sizes[j] = 1;\r
    }\r
    num_new_clusters = FN(BrotliHistogramCombine)(\r
        histograms, sizes, symbols, new_clusters, pairs, num_to_combine,\r
        num_to_combine, HISTOGRAMS_PER_BATCH, max_num_pairs);\r
    BROTLI_ENSURE_CAPACITY(m, HistogramType, all_histograms,\r
        all_histograms_capacity, all_histograms_size + num_new_clusters);\r
    BROTLI_ENSURE_CAPACITY(m, uint32_t, cluster_size,\r
        cluster_size_capacity, cluster_size_size + num_new_clusters);\r
    if (BROTLI_IS_OOM(m)) return;\r
    for (j = 0; j < num_new_clusters; ++j) {\r
      all_histograms[all_histograms_size++] = histograms[new_clusters[j]];\r
      cluster_size[cluster_size_size++] = sizes[new_clusters[j]];\r
      remap[new_clusters[j]] = (uint32_t)j;\r
    }\r
    for (j = 0; j < num_to_combine; ++j) {\r
      histogram_symbols[i + j] = (uint32_t)num_clusters + remap[symbols[j]];\r
    }\r
    num_clusters += num_new_clusters;\r
    BROTLI_DCHECK(num_clusters == cluster_size_size);\r
    BROTLI_DCHECK(num_clusters == all_histograms_size);\r
  }\r
  BROTLI_FREE(m, histograms);\r
\r
  max_num_pairs =\r
      BROTLI_MIN(size_t, 64 * num_clusters, (num_clusters / 2) * num_clusters);\r
  if (pairs_capacity < max_num_pairs + 1) {\r
    BROTLI_FREE(m, pairs);\r
    pairs = BROTLI_ALLOC(m, HistogramPair, max_num_pairs + 1);\r
    if (BROTLI_IS_OOM(m)) return;\r
  }\r
\r
  clusters = BROTLI_ALLOC(m, uint32_t, num_clusters);\r
  if (BROTLI_IS_OOM(m)) return;\r
  for (i = 0; i < num_clusters; ++i) {\r
    clusters[i] = (uint32_t)i;\r
  }\r
  num_final_clusters = FN(BrotliHistogramCombine)(\r
      all_histograms, cluster_size, histogram_symbols, clusters, pairs,\r
      num_clusters, num_blocks, BROTLI_MAX_NUMBER_OF_BLOCK_TYPES,\r
      max_num_pairs);\r
  BROTLI_FREE(m, pairs);\r
  BROTLI_FREE(m, cluster_size);\r
\r
  new_index = BROTLI_ALLOC(m, uint32_t, num_clusters);\r
  if (BROTLI_IS_OOM(m)) return;\r
  for (i = 0; i < num_clusters; ++i) new_index[i] = kInvalidIndex;\r
  pos = 0;\r
  {\r
    uint32_t next_index = 0;\r
    for (i = 0; i < num_blocks; ++i) {\r
      HistogramType histo;\r
      size_t j;\r
      uint32_t best_out;\r
      double best_bits;\r
      FN(HistogramClear)(&histo);\r
      for (j = 0; j < block_lengths[i]; ++j) {\r
        FN(HistogramAdd)(&histo, data[pos++]);\r
      }\r
      best_out = (i == 0) ? histogram_symbols[0] : histogram_symbols[i - 1];\r
      best_bits =\r
          FN(BrotliHistogramBitCostDistance)(&histo, &all_histograms[best_out]);\r
      for (j = 0; j < num_final_clusters; ++j) {\r
        const double cur_bits = FN(BrotliHistogramBitCostDistance)(\r
            &histo, &all_histograms[clusters[j]]);\r
        if (cur_bits < best_bits) {\r
          best_bits = cur_bits;\r
          best_out = clusters[j];\r
        }\r
      }\r
      histogram_symbols[i] = best_out;\r
      if (new_index[best_out] == kInvalidIndex) {\r
        new_index[best_out] = next_index++;\r
      }\r
    }\r
  }\r
  BROTLI_FREE(m, clusters);\r
  BROTLI_FREE(m, all_histograms);\r
  BROTLI_ENSURE_CAPACITY(\r
      m, uint8_t, split->types, split->types_alloc_size, num_blocks);\r
  BROTLI_ENSURE_CAPACITY(\r
      m, uint32_t, split->lengths, split->lengths_alloc_size, num_blocks);\r
  if (BROTLI_IS_OOM(m)) return;\r
  {\r
    uint32_t cur_length = 0;\r
    size_t block_idx = 0;\r
    uint8_t max_type = 0;\r
    for (i = 0; i < num_blocks; ++i) {\r
      cur_length += block_lengths[i];\r
      if (i + 1 == num_blocks ||\r
          histogram_symbols[i] != histogram_symbols[i + 1]) {\r
        const uint8_t id = (uint8_t)new_index[histogram_symbols[i]];\r
        split->types[block_idx] = id;\r
        split->lengths[block_idx] = cur_length;\r
        max_type = BROTLI_MAX(uint8_t, max_type, id);\r
        cur_length = 0;\r
        ++block_idx;\r
      }\r
    }\r
    split->num_blocks = block_idx;\r
    split->num_types = (size_t)max_type + 1;\r
  }\r
  BROTLI_FREE(m, new_index);\r
  BROTLI_FREE(m, block_lengths);\r
  BROTLI_FREE(m, histogram_symbols);\r
}\r
\r
static void FN(SplitByteVector)(MemoryManager* m,\r
                                const DataType* data, const size_t length,\r
                                const size_t literals_per_histogram,\r
                                const size_t max_histograms,\r
                                const size_t sampling_stride_length,\r
                                const double block_switch_cost,\r
                                const BrotliEncoderParams* params,\r
                                BlockSplit* split) {\r
  const size_t data_size = FN(HistogramDataSize)();\r
  size_t num_histograms = length / literals_per_histogram + 1;\r
  HistogramType* histograms;\r
  if (num_histograms > max_histograms) {\r
    num_histograms = max_histograms;\r
  }\r
  if (length == 0) {\r
    split->num_types = 1;\r
    return;\r
  } else if (length < kMinLengthForBlockSplitting) {\r
    BROTLI_ENSURE_CAPACITY(m, uint8_t,\r
        split->types, split->types_alloc_size, split->num_blocks + 1);\r
    BROTLI_ENSURE_CAPACITY(m, uint32_t,\r
        split->lengths, split->lengths_alloc_size, split->num_blocks + 1);\r
    if (BROTLI_IS_OOM(m)) return;\r
    split->num_types = 1;\r
    split->types[split->num_blocks] = 0;\r
    split->lengths[split->num_blocks] = (uint32_t)length;\r
    split->num_blocks++;\r
    return;\r
  }\r
  histograms = BROTLI_ALLOC(m, HistogramType, num_histograms);\r
  if (BROTLI_IS_OOM(m)) return;\r
  /* Find good entropy codes. */\r
  FN(InitialEntropyCodes)(data, length,\r
                          sampling_stride_length,\r
                          num_histograms, histograms);\r
  FN(RefineEntropyCodes)(data, length,\r
                         sampling_stride_length,\r
                         num_histograms, histograms);\r
  {\r
    /* Find a good path through literals with the good entropy codes. */\r
    uint8_t* block_ids = BROTLI_ALLOC(m, uint8_t, length);\r
    size_t num_blocks = 0;\r
    const size_t bitmaplen = (num_histograms + 7) >> 3;\r
    double* insert_cost = BROTLI_ALLOC(m, double, data_size * num_histograms);\r
    double* cost = BROTLI_ALLOC(m, double, num_histograms);\r
    uint8_t* switch_signal = BROTLI_ALLOC(m, uint8_t, length * bitmaplen);\r
    uint16_t* new_id = BROTLI_ALLOC(m, uint16_t, num_histograms);\r
    const size_t iters = params->quality < HQ_ZOPFLIFICATION_QUALITY ? 3 : 10;\r
    size_t i;\r
    if (BROTLI_IS_OOM(m)) return;\r
    for (i = 0; i < iters; ++i) {\r
      num_blocks = FN(FindBlocks)(data, length,\r
                                  block_switch_cost,\r
                                  num_histograms, histograms,\r
                                  insert_cost, cost, switch_signal,\r
                                  block_ids);\r
      num_histograms = FN(RemapBlockIds)(block_ids, length,\r
                                         new_id, num_histograms);\r
      FN(BuildBlockHistograms)(data, length, block_ids,\r
                               num_histograms, histograms);\r
    }\r
    BROTLI_FREE(m, insert_cost);\r
    BROTLI_FREE(m, cost);\r
    BROTLI_FREE(m, switch_signal);\r
    BROTLI_FREE(m, new_id);\r
    BROTLI_FREE(m, histograms);\r
    FN(ClusterBlocks)(m, data, length, num_blocks, block_ids, split);\r
    if (BROTLI_IS_OOM(m)) return;\r
    BROTLI_FREE(m, block_ids);\r
  }\r
}\r
\r
#undef HistogramType\r