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
+36
View File
@@ -20,9 +20,45 @@
#include <stddef.h>
#include <stdint.h>
#include <string>
#include <vector>
#include "./hash.h"
#include "./ringbuffer.h"
namespace brotli {
class BrotliCompressor {
public:
BrotliCompressor();
~BrotliCompressor();
// Writes the stream header into the internal output buffer.
void WriteStreamHeader();
// Encodes the data in input_buffer as a meta-block and writes it to
// encoded_buffer and sets *encoded_size to the number of bytes that was
// written.
void WriteMetaBlock(const size_t input_size,
const uint8_t* input_buffer,
size_t* encoded_size,
uint8_t* encoded_buffer);
// Writes a zero-length meta-block with end-of-input bit set to the
// internal output buffer and copies the output buffer to encoded_buffer and
// sets *encoded_size to the number of bytes written.
void FinishStream(size_t* encoded_size, uint8_t* encoded_buffer);
private:
Hasher* hasher_;
int dist_ringbuffer_[4];
size_t dist_ringbuffer_idx_;
size_t input_pos_;
RingBuffer ringbuffer_;
std::vector<float> literal_cost_;
int storage_ix_;
uint8_t* storage_;
};
// Compresses the data in input_buffer into encoded_buffer, and sets
// *encoded_size to the compressed length.
// Returns 0 if there was an error and 1 otherwise.