Because readable and maintainable code is key. The Brain Compressor is an easily verifiable and portable lossless data compressor. Source codes count 635 lines in total. A core library is less than 400 lines in pure C.
Benchmark evaluates the compression of the reference enwik8 file. All compressors have been compiled with GCC 9.2 on 64-bit Linux. The reference system uses an AMD Ryzen Threadripper 2990WX. All measurements use default settings (no extra arguments). The elapsed Compression and Decompression times (wall clock) are given in seconds. The compression Ratio is given as uncompressed/compressed (more is better). SLOC means Source Lines Of Code. Bold font indicates the best result.
| Compressor | Ratio | Compression time | Decompression time | SLOC |
|---|---|---|---|---|
| lz4 1.9.2 | 1.75 | 0.29 | 0.11 | 20 619 |
| lzop 1.04 | 1.78 | 0.36 | 0.33 | 17 123 |
| Brain Compressor | 1.88 | 1.03 | 0.91 | 635 |
| gzip 1.9 | 2.74 | 4.69 | 0.63 | 48 552 |
| zstd 1.3.7 | 2.80 | 0.55 | 0.18 | 111 948 |
| bzip2 1.0.6 | 3.45 | 7.39 | 3.36 | 8 117 |
| xz 5.2.4 | 3.79 | 53.70 | 1.40 | 43 534 |
| brotli 1.0.7 | 3.88 | 3:05.59 | 0.34 | 35 372 |
The Brain Compressor uses an adaptive Golomb-Rice coding based on context modeling. The context model uses a single previous byte in the uncompressed stream to predict the next byte.
make BUILD=release
or
make build-pgo
Compress:
./encode INPUT-FILE [OUTPUT-FILE]
Decompress:
./decode INPUT-FILE [OUTPUT-FILE]
- David Barina, [email protected]
This project is licensed under the MIT License. See the LICENSE.md file for details.