Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(669)

Unified Diff: third_party/brotli/enc/block_splitter_inc.h

Issue 2537133002: Update brotli to v1.0.0-snapshot. (Closed)
Patch Set: Fixed typo Created 4 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « third_party/brotli/enc/block_splitter.cc ('k') | third_party/brotli/enc/brotli_bit_stream.h » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: third_party/brotli/enc/block_splitter_inc.h
diff --git a/third_party/brotli/enc/block_splitter.cc b/third_party/brotli/enc/block_splitter_inc.h
similarity index 15%
rename from third_party/brotli/enc/block_splitter.cc
rename to third_party/brotli/enc/block_splitter_inc.h
index db8d9c606d6fab28641f815e0cffdf2aa67612b2..8574fb9a63300d4601c4ad0a34008d06aaa6031e 100644
--- a/third_party/brotli/enc/block_splitter.cc
+++ b/third_party/brotli/enc/block_splitter_inc.h
@@ -1,98 +1,23 @@
+/* NOLINT(build/header_guard) */
/* Copyright 2013 Google Inc. All Rights Reserved.
Distributed under MIT license.
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
*/
-// Block split point selection utilities.
+/* template parameters: FN, DataType */
-#include "./block_splitter.h"
+#define HistogramType FN(Histogram)
-#include <assert.h>
-#include <math.h>
-
-#include <algorithm>
-#include <cstring>
-#include <vector>
-
-#include "./cluster.h"
-#include "./command.h"
-#include "./fast_log.h"
-#include "./histogram.h"
-
-namespace brotli {
-
-static const size_t kMaxLiteralHistograms = 100;
-static const size_t kMaxCommandHistograms = 50;
-static const double kLiteralBlockSwitchCost = 28.1;
-static const double kCommandBlockSwitchCost = 13.5;
-static const double kDistanceBlockSwitchCost = 14.6;
-static const size_t kLiteralStrideLength = 70;
-static const size_t kCommandStrideLength = 40;
-static const size_t kSymbolsPerLiteralHistogram = 544;
-static const size_t kSymbolsPerCommandHistogram = 530;
-static const size_t kSymbolsPerDistanceHistogram = 544;
-static const size_t kMinLengthForBlockSplitting = 128;
-static const size_t kIterMulForRefining = 2;
-static const size_t kMinItersForRefining = 100;
-
-void CopyLiteralsToByteArray(const Command* cmds,
- const size_t num_commands,
- const uint8_t* data,
- const size_t offset,
- const size_t mask,
- std::vector<uint8_t>* literals) {
- // Count how many we have.
- size_t total_length = 0;
- for (size_t i = 0; i < num_commands; ++i) {
- total_length += cmds[i].insert_len_;
- }
- if (total_length == 0) {
- return;
- }
-
- // Allocate.
- literals->resize(total_length);
-
- // Loop again, and copy this time.
- size_t pos = 0;
- size_t from_pos = offset & mask;
- for (size_t i = 0; i < num_commands && pos < total_length; ++i) {
- size_t insert_len = cmds[i].insert_len_;
- if (from_pos + insert_len > mask) {
- size_t head_size = mask + 1 - from_pos;
- memcpy(&(*literals)[pos], data + from_pos, head_size);
- from_pos = 0;
- pos += head_size;
- insert_len -= head_size;
- }
- if (insert_len > 0) {
- memcpy(&(*literals)[pos], data + from_pos, insert_len);
- pos += insert_len;
- }
- from_pos = (from_pos + insert_len + cmds[i].copy_len()) & mask;
- }
-}
-
-inline static unsigned int MyRand(unsigned int* seed) {
- *seed *= 16807U;
- if (*seed == 0) {
- *seed = 1;
- }
- return *seed;
-}
-
-template<typename HistogramType, typename DataType>
-void InitialEntropyCodes(const DataType* data, size_t length,
- size_t stride,
- size_t num_histograms,
- HistogramType* histograms) {
- for (size_t i = 0; i < num_histograms; ++i) {
- histograms[i].Clear();
- }
+static void FN(InitialEntropyCodes)(const DataType* data, size_t length,
+ size_t stride,
+ size_t num_histograms,
+ HistogramType* histograms) {
unsigned int seed = 7;
size_t block_length = length / num_histograms;
- for (size_t i = 0; i < num_histograms; ++i) {
+ size_t i;
+ FN(ClearHistograms)(histograms, num_histograms);
+ for (i = 0; i < num_histograms; ++i) {
size_t pos = length * i / num_histograms;
if (i != 0) {
pos += MyRand(&seed) % block_length;
@@ -100,16 +25,15 @@ void InitialEntropyCodes(const DataType* data, size_t length,
if (pos + stride >= length) {
pos = length - stride - 1;
}
- histograms[i].Add(data + pos, stride);
+ FN(HistogramAddVector)(&histograms[i], data + pos, stride);
}
}
-template<typename HistogramType, typename DataType>
-void RandomSample(unsigned int* seed,
- const DataType* data,
- size_t length,
- size_t stride,
- HistogramType* sample) {
+static void FN(RandomSample)(unsigned int* seed,
+ const DataType* data,
+ size_t length,
+ size_t stride,
+ HistogramType* sample) {
size_t pos = 0;
if (stride >= length) {
pos = 0;
@@ -117,389 +41,392 @@ void RandomSample(unsigned int* seed,
} else {
pos = MyRand(seed) % (length - stride + 1);
}
- sample->Add(data + pos, stride);
+ FN(HistogramAddVector)(sample, data + pos, stride);
}
-template<typename HistogramType, typename DataType>
-void RefineEntropyCodes(const DataType* data, size_t length,
- size_t stride,
- size_t num_histograms,
- HistogramType* histograms) {
+static void FN(RefineEntropyCodes)(const DataType* data, size_t length,
+ size_t stride,
+ size_t num_histograms,
+ HistogramType* histograms) {
size_t iters =
kIterMulForRefining * length / stride + kMinItersForRefining;
unsigned int seed = 7;
+ size_t iter;
iters = ((iters + num_histograms - 1) / num_histograms) * num_histograms;
- for (size_t iter = 0; iter < iters; ++iter) {
+ for (iter = 0; iter < iters; ++iter) {
HistogramType sample;
- RandomSample(&seed, data, length, stride, &sample);
- size_t ix = iter % num_histograms;
- histograms[ix].AddHistogram(sample);
+ FN(HistogramClear)(&sample);
+ FN(RandomSample)(&seed, data, length, stride, &sample);
+ FN(HistogramAddHistogram)(&histograms[iter % num_histograms], &sample);
}
}
-inline static double BitCost(size_t count) {
- return count == 0 ? -2.0 : FastLog2(count);
-}
-
-// Assigns a block id from the range [0, vec.size()) to each data element
-// in data[0..length) and fills in block_id[0..length) with the assigned values.
-// Returns the number of blocks, i.e. one plus the number of block switches.
-template<typename DataType, int kSize>
-size_t FindBlocks(const DataType* data, const size_t length,
- const double block_switch_bitcost,
- const size_t num_histograms,
- const Histogram<kSize>* histograms,
- double* insert_cost,
- double* cost,
- uint8_t* switch_signal,
- uint8_t *block_id) {
+/* Assigns a block id from the range [0, num_histograms) to each data element
+ in data[0..length) and fills in block_id[0..length) with the assigned values.
+ Returns the number of blocks, i.e. one plus the number of block switches. */
+static size_t FN(FindBlocks)(const DataType* data, const size_t length,
+ const double block_switch_bitcost,
+ const size_t num_histograms,
+ const HistogramType* histograms,
+ double* insert_cost,
+ double* cost,
+ uint8_t* switch_signal,
+ uint8_t *block_id) {
+ const size_t data_size = FN(HistogramDataSize)();
+ const size_t bitmaplen = (num_histograms + 7) >> 3;
+ size_t num_blocks = 1;
+ size_t i;
+ size_t j;
+ assert(num_histograms <= 256);
if (num_histograms <= 1) {
- for (size_t i = 0; i < length; ++i) {
+ for (i = 0; i < length; ++i) {
block_id[i] = 0;
}
return 1;
}
- const size_t bitmaplen = (num_histograms + 7) >> 3;
- assert(num_histograms <= 256);
- memset(insert_cost, 0, sizeof(insert_cost[0]) * kSize * num_histograms);
- for (size_t j = 0; j < num_histograms; ++j) {
- insert_cost[j] = FastLog2(static_cast<uint32_t>(
- histograms[j].total_count_));
+ memset(insert_cost, 0, sizeof(insert_cost[0]) * data_size * num_histograms);
+ for (i = 0; i < num_histograms; ++i) {
+ insert_cost[i] = FastLog2((uint32_t)histograms[i].total_count_);
}
- for (size_t i = kSize; i != 0;) {
+ for (i = data_size; i != 0;) {
--i;
- for (size_t j = 0; j < num_histograms; ++j) {
+ for (j = 0; j < num_histograms; ++j) {
insert_cost[i * num_histograms + j] =
insert_cost[j] - BitCost(histograms[j].data_[i]);
}
}
memset(cost, 0, sizeof(cost[0]) * num_histograms);
memset(switch_signal, 0, sizeof(switch_signal[0]) * length * bitmaplen);
- // After each iteration of this loop, cost[k] will contain the difference
- // between the minimum cost of arriving at the current byte position using
- // entropy code k, and the minimum cost of arriving at the current byte
- // position. This difference is capped at the block switch cost, and if it
- // reaches block switch cost, it means that when we trace back from the last
- // position, we need to switch here.
- for (size_t byte_ix = 0; byte_ix < length; ++byte_ix) {
+ /* After each iteration of this loop, cost[k] will contain the difference
+ between the minimum cost of arriving at the current byte position using
+ entropy code k, and the minimum cost of arriving at the current byte
+ position. This difference is capped at the block switch cost, and if it
+ reaches block switch cost, it means that when we trace back from the last
+ position, we need to switch here. */
+ for (i = 0; i < length; ++i) {
+ const size_t byte_ix = i;
size_t ix = byte_ix * bitmaplen;
size_t insert_cost_ix = data[byte_ix] * num_histograms;
double min_cost = 1e99;
- for (size_t k = 0; k < num_histograms; ++k) {
- // We are coding the symbol in data[byte_ix] with entropy code k.
+ double block_switch_cost = block_switch_bitcost;
+ size_t k;
+ for (k = 0; k < num_histograms; ++k) {
+ /* We are coding the symbol in data[byte_ix] with entropy code k. */
cost[k] += insert_cost[insert_cost_ix + k];
if (cost[k] < min_cost) {
min_cost = cost[k];
- block_id[byte_ix] = static_cast<uint8_t>(k);
+ block_id[byte_ix] = (uint8_t)k;
}
}
- double block_switch_cost = block_switch_bitcost;
- // More blocks for the beginning.
+ /* More blocks for the beginning. */
if (byte_ix < 2000) {
- block_switch_cost *= 0.77 + 0.07 * static_cast<double>(byte_ix) / 2000;
+ block_switch_cost *= 0.77 + 0.07 * (double)byte_ix / 2000;
}
- for (size_t k = 0; k < num_histograms; ++k) {
+ for (k = 0; k < num_histograms; ++k) {
cost[k] -= min_cost;
if (cost[k] >= block_switch_cost) {
+ const uint8_t mask = (uint8_t)(1u << (k & 7));
cost[k] = block_switch_cost;
- const uint8_t mask = static_cast<uint8_t>(1u << (k & 7));
assert((k >> 3) < bitmaplen);
switch_signal[ix + (k >> 3)] |= mask;
}
}
}
- // Now trace back from the last position and switch at the marked places.
- size_t byte_ix = length - 1;
- size_t ix = byte_ix * bitmaplen;
- uint8_t cur_id = block_id[byte_ix];
- size_t num_blocks = 1;
- while (byte_ix > 0) {
- --byte_ix;
- ix -= bitmaplen;
- const uint8_t mask = static_cast<uint8_t>(1u << (cur_id & 7));
- assert((static_cast<size_t>(cur_id) >> 3) < bitmaplen);
- if (switch_signal[ix + (cur_id >> 3)] & mask) {
- if (cur_id != block_id[byte_ix]) {
- cur_id = block_id[byte_ix];
- ++num_blocks;
+ { /* Trace back from the last position and switch at the marked places. */
+ size_t byte_ix = length - 1;
+ size_t ix = byte_ix * bitmaplen;
+ uint8_t cur_id = block_id[byte_ix];
+ while (byte_ix > 0) {
+ const uint8_t mask = (uint8_t)(1u << (cur_id & 7));
+ assert(((size_t)cur_id >> 3) < bitmaplen);
+ --byte_ix;
+ ix -= bitmaplen;
+ if (switch_signal[ix + (cur_id >> 3)] & mask) {
+ if (cur_id != block_id[byte_ix]) {
+ cur_id = block_id[byte_ix];
+ ++num_blocks;
+ }
}
+ block_id[byte_ix] = cur_id;
}
- block_id[byte_ix] = cur_id;
}
return num_blocks;
}
-static size_t RemapBlockIds(uint8_t* block_ids, const size_t length,
- uint16_t* new_id, const size_t num_histograms) {
+static size_t FN(RemapBlockIds)(uint8_t* block_ids, const size_t length,
+ uint16_t* new_id, const size_t num_histograms) {
static const uint16_t kInvalidId = 256;
- for (size_t i = 0; i < num_histograms; ++i) {
+ uint16_t next_id = 0;
+ size_t i;
+ for (i = 0; i < num_histograms; ++i) {
new_id[i] = kInvalidId;
}
- uint16_t next_id = 0;
- for (size_t i = 0; i < length; ++i) {
+ for (i = 0; i < length; ++i) {
assert(block_ids[i] < num_histograms);
if (new_id[block_ids[i]] == kInvalidId) {
new_id[block_ids[i]] = next_id++;
}
}
- for (size_t i = 0; i < length; ++i) {
- block_ids[i] = static_cast<uint8_t>(new_id[block_ids[i]]);
+ for (i = 0; i < length; ++i) {
+ block_ids[i] = (uint8_t)new_id[block_ids[i]];
assert(block_ids[i] < num_histograms);
}
assert(next_id <= num_histograms);
return next_id;
}
-template<typename HistogramType, typename DataType>
-void BuildBlockHistograms(const DataType* data, const size_t length,
- const uint8_t* block_ids,
- const size_t num_histograms,
- HistogramType* histograms) {
- for (size_t i = 0; i < num_histograms; ++i) {
- histograms[i].Clear();
- }
- for (size_t i = 0; i < length; ++i) {
- histograms[block_ids[i]].Add(data[i]);
+static void FN(BuildBlockHistograms)(const DataType* data, const size_t length,
+ const uint8_t* block_ids,
+ const size_t num_histograms,
+ HistogramType* histograms) {
+ size_t i;
+ FN(ClearHistograms)(histograms, num_histograms);
+ for (i = 0; i < length; ++i) {
+ FN(HistogramAdd)(&histograms[block_ids[i]], data[i]);
}
}
-template<typename HistogramType, typename DataType>
-void ClusterBlocks(const DataType* data, const size_t length,
- const size_t num_blocks,
- uint8_t* block_ids,
- BlockSplit* split) {
- static const size_t kMaxNumberOfBlockTypes = 256;
- static const size_t kHistogramsPerBatch = 64;
- static const size_t kClustersPerBatch = 16;
- std::vector<uint32_t> histogram_symbols(num_blocks);
- std::vector<uint32_t> block_lengths(num_blocks);
+static void FN(ClusterBlocks)(MemoryManager* m,
+ const DataType* data, const size_t length,
+ const size_t num_blocks,
+ uint8_t* block_ids,
+ BlockSplit* split) {
+ uint32_t* histogram_symbols = BROTLI_ALLOC(m, uint32_t, num_blocks);
+ uint32_t* block_lengths = BROTLI_ALLOC(m, uint32_t, num_blocks);
+ const size_t expected_num_clusters = CLUSTERS_PER_BATCH *
+ (num_blocks + HISTOGRAMS_PER_BATCH - 1) / HISTOGRAMS_PER_BATCH;
+ size_t all_histograms_size = 0;
+ size_t all_histograms_capacity = expected_num_clusters;
+ HistogramType* all_histograms =
+ BROTLI_ALLOC(m, HistogramType, all_histograms_capacity);
+ size_t cluster_size_size = 0;
+ size_t cluster_size_capacity = expected_num_clusters;
+ uint32_t* cluster_size = BROTLI_ALLOC(m, uint32_t, cluster_size_capacity);
+ size_t num_clusters = 0;
+ HistogramType* histograms = BROTLI_ALLOC(m, HistogramType,
+ BROTLI_MIN(size_t, num_blocks, HISTOGRAMS_PER_BATCH));
+ size_t max_num_pairs =
+ HISTOGRAMS_PER_BATCH * HISTOGRAMS_PER_BATCH / 2;
+ size_t pairs_capacity = max_num_pairs + 1;
+ HistogramPair* pairs = BROTLI_ALLOC(m, HistogramPair, pairs_capacity);
+ size_t pos = 0;
+ uint32_t* clusters;
+ size_t num_final_clusters;
+ static const uint32_t kInvalidIndex = BROTLI_UINT32_MAX;
+ uint32_t* new_index;
+ uint8_t max_type = 0;
+ size_t i;
+ uint32_t sizes[HISTOGRAMS_PER_BATCH] = { 0 };
+ uint32_t new_clusters[HISTOGRAMS_PER_BATCH] = { 0 };
+ uint32_t symbols[HISTOGRAMS_PER_BATCH] = { 0 };
+ uint32_t remap[HISTOGRAMS_PER_BATCH] = { 0 };
+
+ if (BROTLI_IS_OOM(m)) return;
- size_t block_idx = 0;
- for (size_t i = 0; i < length; ++i) {
- assert(block_idx < num_blocks);
- ++block_lengths[block_idx];
- if (i + 1 == length || block_ids[i] != block_ids[i + 1]) {
- ++block_idx;
+ memset(block_lengths, 0, num_blocks * sizeof(uint32_t));
+
+ {
+ size_t block_idx = 0;
+ for (i = 0; i < length; ++i) {
+ assert(block_idx < num_blocks);
+ ++block_lengths[block_idx];
+ if (i + 1 == length || block_ids[i] != block_ids[i + 1]) {
+ ++block_idx;
+ }
}
+ assert(block_idx == num_blocks);
}
- assert(block_idx == num_blocks);
- const size_t expected_num_clusters =
- kClustersPerBatch *
- (num_blocks + kHistogramsPerBatch - 1) / kHistogramsPerBatch;
- std::vector<HistogramType> all_histograms;
- std::vector<uint32_t> cluster_size;
- all_histograms.reserve(expected_num_clusters);
- cluster_size.reserve(expected_num_clusters);
- size_t num_clusters = 0;
- std::vector<HistogramType> histograms(
- std::min(num_blocks, kHistogramsPerBatch));
- size_t max_num_pairs = kHistogramsPerBatch * kHistogramsPerBatch / 2;
- std::vector<HistogramPair> pairs(max_num_pairs + 1);
- size_t pos = 0;
- for (size_t i = 0; i < num_blocks; i += kHistogramsPerBatch) {
- const size_t num_to_combine = std::min(num_blocks - i, kHistogramsPerBatch);
- uint32_t sizes[kHistogramsPerBatch];
- uint32_t clusters[kHistogramsPerBatch];
- uint32_t symbols[kHistogramsPerBatch];
- uint32_t remap[kHistogramsPerBatch];
- for (size_t j = 0; j < num_to_combine; ++j) {
- histograms[j].Clear();
- for (size_t k = 0; k < block_lengths[i + j]; ++k) {
- histograms[j].Add(data[pos++]);
+ for (i = 0; i < num_blocks; i += HISTOGRAMS_PER_BATCH) {
+ const size_t num_to_combine =
+ BROTLI_MIN(size_t, num_blocks - i, HISTOGRAMS_PER_BATCH);
+ size_t num_new_clusters;
+ size_t j;
+ for (j = 0; j < num_to_combine; ++j) {
+ size_t k;
+ FN(HistogramClear)(&histograms[j]);
+ for (k = 0; k < block_lengths[i + j]; ++k) {
+ FN(HistogramAdd)(&histograms[j], data[pos++]);
}
- histograms[j].bit_cost_ = PopulationCost(histograms[j]);
- symbols[j] = clusters[j] = static_cast<uint32_t>(j);
+ histograms[j].bit_cost_ = FN(BrotliPopulationCost)(&histograms[j]);
+ new_clusters[j] = (uint32_t)j;
+ symbols[j] = (uint32_t)j;
sizes[j] = 1;
}
- size_t num_new_clusters = HistogramCombine(
- &histograms[0], sizes, symbols, clusters, &pairs[0], num_to_combine,
- num_to_combine, kHistogramsPerBatch, max_num_pairs);
- for (size_t j = 0; j < num_new_clusters; ++j) {
- all_histograms.push_back(histograms[clusters[j]]);
- cluster_size.push_back(sizes[clusters[j]]);
- remap[clusters[j]] = static_cast<uint32_t>(j);
+ num_new_clusters = FN(BrotliHistogramCombine)(
+ histograms, sizes, symbols, new_clusters, pairs, num_to_combine,
+ num_to_combine, HISTOGRAMS_PER_BATCH, max_num_pairs);
+ BROTLI_ENSURE_CAPACITY(m, HistogramType, all_histograms,
+ all_histograms_capacity, all_histograms_size + num_new_clusters);
+ BROTLI_ENSURE_CAPACITY(m, uint32_t, cluster_size,
+ cluster_size_capacity, cluster_size_size + num_new_clusters);
+ if (BROTLI_IS_OOM(m)) return;
+ for (j = 0; j < num_new_clusters; ++j) {
+ all_histograms[all_histograms_size++] = histograms[new_clusters[j]];
+ cluster_size[cluster_size_size++] = sizes[new_clusters[j]];
+ remap[new_clusters[j]] = (uint32_t)j;
}
- for (size_t j = 0; j < num_to_combine; ++j) {
- histogram_symbols[i + j] =
- static_cast<uint32_t>(num_clusters) + remap[symbols[j]];
+ for (j = 0; j < num_to_combine; ++j) {
+ histogram_symbols[i + j] = (uint32_t)num_clusters + remap[symbols[j]];
}
num_clusters += num_new_clusters;
- assert(num_clusters == cluster_size.size());
- assert(num_clusters == all_histograms.size());
+ assert(num_clusters == cluster_size_size);
+ assert(num_clusters == all_histograms_size);
}
+ BROTLI_FREE(m, histograms);
max_num_pairs =
- std::min(64 * num_clusters, (num_clusters / 2) * num_clusters);
- pairs.resize(max_num_pairs + 1);
+ BROTLI_MIN(size_t, 64 * num_clusters, (num_clusters / 2) * num_clusters);
+ if (pairs_capacity < max_num_pairs + 1) {
+ BROTLI_FREE(m, pairs);
+ pairs = BROTLI_ALLOC(m, HistogramPair, max_num_pairs + 1);
+ if (BROTLI_IS_OOM(m)) return;
+ }
- std::vector<uint32_t> clusters(num_clusters);
- for (size_t i = 0; i < num_clusters; ++i) {
- clusters[i] = static_cast<uint32_t>(i);
+ clusters = BROTLI_ALLOC(m, uint32_t, num_clusters);
+ if (BROTLI_IS_OOM(m)) return;
+ for (i = 0; i < num_clusters; ++i) {
+ clusters[i] = (uint32_t)i;
}
- size_t num_final_clusters =
- HistogramCombine(&all_histograms[0], &cluster_size[0],
- &histogram_symbols[0],
- &clusters[0], &pairs[0], num_clusters,
- num_blocks, kMaxNumberOfBlockTypes, max_num_pairs);
+ num_final_clusters = FN(BrotliHistogramCombine)(
+ all_histograms, cluster_size, histogram_symbols, clusters, pairs,
+ num_clusters, num_blocks, BROTLI_MAX_NUMBER_OF_BLOCK_TYPES,
+ max_num_pairs);
+ BROTLI_FREE(m, pairs);
+ BROTLI_FREE(m, cluster_size);
- static const uint32_t kInvalidIndex = std::numeric_limits<uint32_t>::max();
- std::vector<uint32_t> new_index(num_clusters, kInvalidIndex);
- uint32_t next_index = 0;
+ new_index = BROTLI_ALLOC(m, uint32_t, num_clusters);
+ if (BROTLI_IS_OOM(m)) return;
+ for (i = 0; i < num_clusters; ++i) new_index[i] = kInvalidIndex;
pos = 0;
- for (size_t i = 0; i < num_blocks; ++i) {
- HistogramType histo;
- for (size_t j = 0; j < block_lengths[i]; ++j) {
- histo.Add(data[pos++]);
- }
- uint32_t best_out =
- i == 0 ? histogram_symbols[0] : histogram_symbols[i - 1];
- double best_bits = HistogramBitCostDistance(
- histo, all_histograms[best_out]);
- for (size_t j = 0; j < num_final_clusters; ++j) {
- const double cur_bits = HistogramBitCostDistance(
- histo, all_histograms[clusters[j]]);
- if (cur_bits < best_bits) {
- best_bits = cur_bits;
- best_out = clusters[j];
+ {
+ uint32_t next_index = 0;
+ for (i = 0; i < num_blocks; ++i) {
+ HistogramType histo;
+ size_t j;
+ uint32_t best_out;
+ double best_bits;
+ FN(HistogramClear)(&histo);
+ for (j = 0; j < block_lengths[i]; ++j) {
+ FN(HistogramAdd)(&histo, data[pos++]);
+ }
+ best_out = (i == 0) ? histogram_symbols[0] : histogram_symbols[i - 1];
+ best_bits =
+ FN(BrotliHistogramBitCostDistance)(&histo, &all_histograms[best_out]);
+ for (j = 0; j < num_final_clusters; ++j) {
+ const double cur_bits = FN(BrotliHistogramBitCostDistance)(
+ &histo, &all_histograms[clusters[j]]);
+ if (cur_bits < best_bits) {
+ best_bits = cur_bits;
+ best_out = clusters[j];
+ }
+ }
+ histogram_symbols[i] = best_out;
+ if (new_index[best_out] == kInvalidIndex) {
+ new_index[best_out] = next_index++;
}
- }
- histogram_symbols[i] = best_out;
- if (new_index[best_out] == kInvalidIndex) {
- new_index[best_out] = next_index++;
}
}
- uint8_t max_type = 0;
- uint32_t cur_length = 0;
- block_idx = 0;
- split->types.resize(num_blocks);
- split->lengths.resize(num_blocks);
- for (size_t i = 0; i < num_blocks; ++i) {
- cur_length += block_lengths[i];
- if (i + 1 == num_blocks ||
- histogram_symbols[i] != histogram_symbols[i + 1]) {
- const uint8_t id = static_cast<uint8_t>(new_index[histogram_symbols[i]]);
- split->types[block_idx] = id;
- split->lengths[block_idx] = cur_length;
- max_type = std::max(max_type, id);
- cur_length = 0;
- ++block_idx;
+ BROTLI_FREE(m, clusters);
+ BROTLI_FREE(m, all_histograms);
+ BROTLI_ENSURE_CAPACITY(
+ m, uint8_t, split->types, split->types_alloc_size, num_blocks);
+ BROTLI_ENSURE_CAPACITY(
+ m, uint32_t, split->lengths, split->lengths_alloc_size, num_blocks);
+ if (BROTLI_IS_OOM(m)) return;
+ {
+ uint32_t cur_length = 0;
+ size_t block_idx = 0;
+ for (i = 0; i < num_blocks; ++i) {
+ cur_length += block_lengths[i];
+ if (i + 1 == num_blocks ||
+ histogram_symbols[i] != histogram_symbols[i + 1]) {
+ const uint8_t id = (uint8_t)new_index[histogram_symbols[i]];
+ split->types[block_idx] = id;
+ split->lengths[block_idx] = cur_length;
+ max_type = BROTLI_MAX(uint8_t, max_type, id);
+ cur_length = 0;
+ ++block_idx;
+ }
}
+ split->num_blocks = block_idx;
+ split->num_types = (size_t)max_type + 1;
}
- split->types.resize(block_idx);
- split->lengths.resize(block_idx);
- split->num_types = static_cast<size_t>(max_type) + 1;
+ BROTLI_FREE(m, new_index);
+ BROTLI_FREE(m, block_lengths);
+ BROTLI_FREE(m, histogram_symbols);
}
-template<int kSize, typename DataType>
-void SplitByteVector(const std::vector<DataType>& data,
- const size_t literals_per_histogram,
- const size_t max_histograms,
- const size_t sampling_stride_length,
- const double block_switch_cost,
- BlockSplit* split) {
- if (data.empty()) {
+static void FN(SplitByteVector)(MemoryManager* m,
+ const DataType* data, const size_t length,
+ const size_t literals_per_histogram,
+ const size_t max_histograms,
+ const size_t sampling_stride_length,
+ const double block_switch_cost,
+ const BrotliEncoderParams* params,
+ BlockSplit* split) {
+ const size_t data_size = FN(HistogramDataSize)();
+ size_t num_histograms = length / literals_per_histogram + 1;
+ HistogramType* histograms;
+ if (num_histograms > max_histograms) {
+ num_histograms = max_histograms;
+ }
+ if (length == 0) {
split->num_types = 1;
return;
- } else if (data.size() < kMinLengthForBlockSplitting) {
+ } else if (length < kMinLengthForBlockSplitting) {
+ BROTLI_ENSURE_CAPACITY(m, uint8_t,
+ split->types, split->types_alloc_size, split->num_blocks + 1);
+ BROTLI_ENSURE_CAPACITY(m, uint32_t,
+ split->lengths, split->lengths_alloc_size, split->num_blocks + 1);
+ if (BROTLI_IS_OOM(m)) return;
split->num_types = 1;
- split->types.push_back(0);
- split->lengths.push_back(static_cast<uint32_t>(data.size()));
+ split->types[split->num_blocks] = 0;
+ split->lengths[split->num_blocks] = (uint32_t)length;
+ split->num_blocks++;
return;
}
- size_t num_histograms = data.size() / literals_per_histogram + 1;
- if (num_histograms > max_histograms) {
- num_histograms = max_histograms;
- }
- Histogram<kSize>* histograms = new Histogram<kSize>[num_histograms];
- // Find good entropy codes.
- InitialEntropyCodes(&data[0], data.size(),
- sampling_stride_length,
- num_histograms, histograms);
- RefineEntropyCodes(&data[0], data.size(),
- sampling_stride_length,
- num_histograms, histograms);
- // Find a good path through literals with the good entropy codes.
- std::vector<uint8_t> block_ids(data.size());
- size_t num_blocks;
- const size_t bitmaplen = (num_histograms + 7) >> 3;
- double* insert_cost = new double[kSize * num_histograms];
- double *cost = new double[num_histograms];
- uint8_t* switch_signal = new uint8_t[data.size() * bitmaplen];
- uint16_t* new_id = new uint16_t[num_histograms];
- for (size_t i = 0; i < 10; ++i) {
- num_blocks = FindBlocks(&data[0], data.size(),
- block_switch_cost,
- num_histograms, histograms,
- insert_cost, cost, switch_signal,
- &block_ids[0]);
- num_histograms = RemapBlockIds(&block_ids[0], data.size(),
- new_id, num_histograms);
- BuildBlockHistograms(&data[0], data.size(), &block_ids[0],
+ histograms = BROTLI_ALLOC(m, HistogramType, num_histograms);
+ if (BROTLI_IS_OOM(m)) return;
+ /* Find good entropy codes. */
+ FN(InitialEntropyCodes)(data, length,
+ sampling_stride_length,
+ num_histograms, histograms);
+ FN(RefineEntropyCodes)(data, length,
+ sampling_stride_length,
num_histograms, histograms);
- }
- delete[] insert_cost;
- delete[] cost;
- delete[] switch_signal;
- delete[] new_id;
- delete[] histograms;
- ClusterBlocks<Histogram<kSize> >(&data[0], data.size(), num_blocks,
- &block_ids[0], split);
-}
-
-void SplitBlock(const Command* cmds,
- const size_t num_commands,
- const uint8_t* data,
- const size_t pos,
- const size_t mask,
- BlockSplit* literal_split,
- BlockSplit* insert_and_copy_split,
- BlockSplit* dist_split) {
{
- // Create a continuous array of literals.
- std::vector<uint8_t> literals;
- CopyLiteralsToByteArray(cmds, num_commands, data, pos, mask, &literals);
- // Create the block split on the array of literals.
- // Literal histograms have alphabet size 256.
- SplitByteVector<256>(
- literals,
- kSymbolsPerLiteralHistogram, kMaxLiteralHistograms,
- kLiteralStrideLength, kLiteralBlockSwitchCost,
- literal_split);
- }
-
- {
- // Compute prefix codes for commands.
- std::vector<uint16_t> insert_and_copy_codes(num_commands);
- for (size_t i = 0; i < num_commands; ++i) {
- insert_and_copy_codes[i] = cmds[i].cmd_prefix_;
- }
- // Create the block split on the array of command prefixes.
- SplitByteVector<kNumCommandPrefixes>(
- insert_and_copy_codes,
- kSymbolsPerCommandHistogram, kMaxCommandHistograms,
- kCommandStrideLength, kCommandBlockSwitchCost,
- insert_and_copy_split);
- }
-
- {
- // Create a continuous array of distance prefixes.
- std::vector<uint16_t> distance_prefixes(num_commands);
- size_t pos = 0;
- for (size_t i = 0; i < num_commands; ++i) {
- const Command& cmd = cmds[i];
- if (cmd.copy_len() && cmd.cmd_prefix_ >= 128) {
- distance_prefixes[pos++] = cmd.dist_prefix_;
- }
+ /* Find a good path through literals with the good entropy codes. */
+ uint8_t* block_ids = BROTLI_ALLOC(m, uint8_t, length);
+ size_t num_blocks = 0;
+ const size_t bitmaplen = (num_histograms + 7) >> 3;
+ double* insert_cost = BROTLI_ALLOC(m, double, data_size * num_histograms);
+ double* cost = BROTLI_ALLOC(m, double, num_histograms);
+ uint8_t* switch_signal = BROTLI_ALLOC(m, uint8_t, length * bitmaplen);
+ uint16_t* new_id = BROTLI_ALLOC(m, uint16_t, num_histograms);
+ const size_t iters = params->quality < HQ_ZOPFLIFICATION_QUALITY ? 3 : 10;
+ size_t i;
+ if (BROTLI_IS_OOM(m)) return;
+ for (i = 0; i < iters; ++i) {
+ num_blocks = FN(FindBlocks)(data, length,
+ block_switch_cost,
+ num_histograms, histograms,
+ insert_cost, cost, switch_signal,
+ block_ids);
+ num_histograms = FN(RemapBlockIds)(block_ids, length,
+ new_id, num_histograms);
+ FN(BuildBlockHistograms)(data, length, block_ids,
+ num_histograms, histograms);
}
- distance_prefixes.resize(pos);
- // Create the block split on the array of distance prefixes.
- SplitByteVector<kNumDistancePrefixes>(
- distance_prefixes,
- kSymbolsPerDistanceHistogram, kMaxCommandHistograms,
- kCommandStrideLength, kDistanceBlockSwitchCost,
- dist_split);
+ BROTLI_FREE(m, insert_cost);
+ BROTLI_FREE(m, cost);
+ BROTLI_FREE(m, switch_signal);
+ BROTLI_FREE(m, new_id);
+ BROTLI_FREE(m, histograms);
+ FN(ClusterBlocks)(m, data, length, num_blocks, block_ids, split);
+ if (BROTLI_IS_OOM(m)) return;
+ BROTLI_FREE(m, block_ids);
}
}
-} // namespace brotli
+#undef HistogramType
« no previous file with comments | « third_party/brotli/enc/block_splitter.cc ('k') | third_party/brotli/enc/brotli_bit_stream.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698