Updates to Brotli compression format, decoder and encoder

This commit contains a batch of changes that were made to the Brotli
compression algorithm in the last three weeks. Most important changes:

  * Added UTF8 context model for good text compression.
  * Simplified context modeling by having only 4 context modes.
  * Per-block context mode selection.
  * Faster backward copying and bit reading functions.
  * More efficient histogram coding.
  * Streaming support for the decoder and encoder.
This commit is contained in:
Zoltan Szabadka
2013-11-15 19:02:17 +01:00
parent 79e99afe46
commit 1571db36a9
23 changed files with 1647 additions and 870 deletions
+16 -3
View File
@@ -79,19 +79,32 @@ typedef Histogram<kNumCommandPrefixes> HistogramCommand;
typedef Histogram<kNumDistancePrefixes> HistogramDistance;
typedef Histogram<kNumBlockLenPrefixes> HistogramBlockLength;
static const int kLiteralContextBits = 6;
static const int kDistanceContextBits = 2;
void BuildHistograms(
const std::vector<Command>& cmds,
const BlockSplit& literal_split,
const BlockSplit& insert_and_copy_split,
const BlockSplit& dist_split,
const uint8_t* input_buffer,
const uint8_t* ringbuffer,
size_t pos,
int context_mode,
int distance_context_mode,
size_t mask,
const std::vector<int>& context_modes,
std::vector<HistogramLiteral>* literal_histograms,
std::vector<HistogramCommand>* insert_and_copy_histograms,
std::vector<HistogramDistance>* copy_dist_histograms);
void BuildLiteralHistogramsForBlockType(
const std::vector<Command>& cmds,
const BlockSplit& literal_split,
const uint8_t* ringbuffer,
size_t pos,
size_t mask,
int block_type,
int context_mode,
std::vector<HistogramLiteral>* histograms);
} // namespace brotli
#endif // BROTLI_ENC_HISTOGRAM_H_