Superfast compression library
DENSITY is a free C99, open-source, BSD licensed compression library.
It is focused on high-speed compression, at the best ratio possible. DENSITY features a buffer and stream API to enable quick integration in any project.
Quick bench
File used : enwik8 (100 MB)
Platform : MacBook Pro, OSX 10.10.2, 2.3 GHz Intel Core i7, 8Gb 1600 MHz DDR, SSD
Timing : using the time function, and taking the best user output after multiple runs
Note : sharc -c1 uses the chameleon algorithm, sharc -c2 the cheetah algorithm, and sharc -c3 the lion algorithm (see About the algorithms further down).
| Program | Library | Compress | Decompress | Size | Ratio | Round trip |
|---|---|---|---|---|---|---|
| sharc -c1 | density 0.12.1 | 0.109s (914 MB/s) | 0.077s (1299 MB/s) | 61 524 502 | 61,52% | 0.186s |
| lz4 -1 | lz4 r126 | 0.461s (217 MB/s) | 0.091s (1099 MB/s) | 56 995 497 | 57,00% | 0.552s |
| lzop -1 | lzo 2.08 | 0.367s (272 MB/s) | 0.309s (324 MB/s) | 56 709 096 | 56,71% | 0.676s |
| sharc -c2 | density 0.12.1 | 0.211s (473 MB/s) | 0.177s (565 MB/s) | 53 156 782 | 53,16% | 0.388s |
| sharc -c3 | density 0.12.1 | 0.349s (286 MB/s) | 0.358s (280 MB/s) | 48 101 942 | 48,10% | 0.707s |
| lz4 -3 | lz4 r126 | 1.520s (66 MB/s) | 0.087s (1149 MB/s) | 47 082 421 | 47,08% | 1.607s |
| lzop -7 | lzo 2.08 | 9.562s (10 MB/s) | 0.319s (313 MB/s) | 41 720 721 | 41,72% | 9.881s |
Squash
Squash is an abstraction layer for compression algorithms, and has an extremely exhaustive set of benchmark results, including density's, available here. You can choose between system architecture and compressed file type. There are even ARM boards tested ! A great tool for selecting a compression library.
FsBench
FsBench is a command line utility that enables real-time testing of compression algorithms, but also hashes and much more. A fork with the latest density releases is available here for easy access. The original author's repository can be found here. Very informative tool as well.
Here are the results of a couple of test runs on a MacBook Pro, OSX 10.10.2, 2.3 GHz Intel Core i7, 8Gb 1600 MHz DDR, SSD :
enwik8 (100,000,000 bytes)
Codec version args
C.Size (C.Ratio) E.Speed D.Speed E.Eff. D.Eff.
density::chameleon 2015-03-30
61524474 (x 1.625) 921 MB/s 1370 MB/s 354e6 527e6
density::cheetah 2015-03-30
53156746 (x 1.881) 473 MB/s 564 MB/s 221e6 264e6
density::lion 2015-03-30
48101906 (x 2.079) 286 MB/s 288 MB/s 148e6 149e6
LZ4 r127
56973103 (x 1.755) 256 MB/s 1647 MB/s 110e6 708e6
LZF 3.6 very
53945381 (x 1.854) 191 MB/s 368 MB/s 87e6 169e6
LZO 2.08 1x1
55792795 (x 1.792) 289 MB/s 373 MB/s 127e6 165e6
QuickLZ 1.5.1b6 1
52334371 (x 1.911) 279 MB/s 353 MB/s 133e6 168e6
Snappy 1.1.0
56539845 (x 1.769) 245 MB/s 797 MB/s 106e6 346e6
wfLZ r10
63521804 (x 1.574) 151 MB/s 515 MB/s 55e6 188e6
silesia (211,960,320 bytes)
Codec version args
C.Size (C.Ratio) E.Speed D.Speed E.Eff. D.Eff.
density::chameleon 2015-03-30
133118910 (x 1.592) 1044 MB/s 1382 MB/s 388e6 514e6
density::cheetah 2015-03-30
101751474 (x 2.083) 530 MB/s 662 MB/s 275e6 344e6
density::lion 2015-03-30
89309921 (x 2.373) 305 MB/s 292 MB/s 176e6 169e6
LZ4 r127
101634462 (x 2.086) 365 MB/s 1831 MB/s 189e6 953e6
LZF 3.6 very
102043866 (x 2.077) 252 MB/s 495 MB/s 130e6 256e6
LZO 2.08 1x1
100592662 (x 2.107) 429 MB/s 583 MB/s 225e6 306e6
QuickLZ 1.5.1b6 1
94727961 (x 2.238) 374 MB/s 432 MB/s 206e6 239e6
Snappy 1.1.0
101385885 (x 2.091) 356 MB/s 1085 MB/s 185e6 565e6
wfLZ r10
109610020 (x 1.934) 197 MB/s 698 MB/s 95e6 337e6
DENSITY is fully C99 compliant and can therefore be built on a number of platforms. You need a C compiler (gcc, clang ...), and a make utility.
Just cd into the density directory, then run the following command :
make
And that's it !
DENSITY outputs compressed data in a simple format, which enables file storage and optional parallelization for both compression and decompression. Inside the main header and footer, a number of blocks can be found, each having its own header and footer. Inside each block, compressed data has a structure determined by the compression algorithm used.
It is possible to add an integrity checksum to the compressed output by using the DENSITY_BLOCK_TYPE_WITH_HASHSUM_INTEGRITY_CHECK block type. The 128-bit checksum is calculated using the excellent SpookyHash algorithm, which is extremely fast and offers a near-zero performance penalty. An additional integrity check will then be automatically performed during decompression.
DENSITY features a buffer API and a stream API which are very simple to use, yet powerful enough to keep users' creativity unleashed. Please see the quick start at the bottom of this page.
Copy ( DENSITY_COMPRESSION_MODE_COPY )
This is not a so-to-speak algorithm as the name implies. It embeds data inside the density block structures. It can be used to quickly add integrity checks to input data, but it has another important purpose inside each block : if data is marked as incompressible using the target algorithm, a mode reversion occurs and copy mode is instead used for the remainder of the block. On the next block the target algorithm is tried again.
Chameleon ( DENSITY_COMPRESSION_MODE_CHAMELEON_ALGORITHM )
Chameleon is a dictionary lookup based compression algorithm. It is designed for absolute speed and usually reaches a 60% compression ratio on compressible data. Decompression is just as fast. This algorithm is a great choice when main concern is speed.
Cheetah ( DENSITY_COMPRESSION_MODE_CHEETAH_ALGORITHM )
Cheetah was developed in conjunction with Piotr Tarsa. It is derived from chameleon and uses swapped double dictionary lookups and predictions. It can be extremely good with highly compressible data (ratio reaching 10% or less). On typical compressible data compression ratio is about 50% or less. It is still extremely fast for both compression and decompression and is a great, efficient all-rounder algorithm.
Lion ( DENSITY_COMPRESSION_MODE_LION_ALGORITHM )
Lion is a multiform compression algorithm derived from cheetah. It goes further in the areas of dynamic adaptation and fine-grained analysis. It uses swapped double dictionary lookups, multiple predictions, shifting sub-word dictionary lookups and forms rank entropy coding. Lion provides the best compression ratio of all three algorithms under any circumstance, and is still very fast.
Using DENSITY in your application couldn't be any simpler.
First you need to include the 2 following files in your project :
- density_api.h
- density_api_data_structures.h
When this is done you can start using the DENSITY APIs :
#include "density_api.h"
#define MY_TEXT "This is a simple example on how to use Density API bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla bla"
#define BUFFER_SIZE DENSITY_MINIMUM_OUT_BUFFER_SIZE
...
uint8_t *outCompressed = (uint8_t *) malloc(BUFFER_SIZE * sizeof(uint8_t));
uint8_t *outDecompressed = (uint8_t *) malloc(BUFFER_SIZE * sizeof(uint8_t));
/**************
* Buffer API *
**************/
density_buffer_processing_result result;
result = density_buffer_compress((uint8_t *) TEXT, strlen(TEXT), outCompressed, BUFFER_SIZE, DENSITY_COMPRESSION_MODE_CHAMELEON_ALGORITHM, DENSITY_BLOCK_TYPE_DEFAULT, NULL, NULL);
if(!result.state)
printf("%llu bytes >> %llu bytes\n", result.bytesRead, result.bytesWritten);
result = density_buffer_decompress(outCompressed, result.bytesWritten, outDecompressed, BUFFER_SIZE, NULL, NULL);
if(!result.state)
printf("%llu bytes >> %llu bytes\n", result.bytesRead, result.bytesWritten);
/**************
* Stream API *
**************/
// We create the stream using the standard malloc and free functions
density_stream* stream = density_stream_create(NULL, NULL);
DENSITY_STREAM_STATE streamState;
// Let's compress our text, using the Chameleon algorithm (extremely fast compression and decompression)
if ((streamState = density_stream_prepare(stream, (uint8_t *) TEXT, strlen(TEXT), outCompressed, BUFFER_SIZE)))
fprintf(stderr, "Error %i when preparing compression\n", streamState);
if ((streamState = density_stream_compress_init(stream, DENSITY_COMPRESSION_MODE_CHAMELEON_ALGORITHM, DENSITY_BLOCK_TYPE_DEFAULT)))
fprintf(stderr, "Error %i when initializing compression\n", streamState);
if ((streamState = density_stream_compress_continue(stream))) if (streamState != DENSITY_STREAM_STATE_STALL_ON_INPUT)
fprintf(stderr, "Error %i occured during compression\n", streamState);
if ((streamState = density_stream_compress_finish(stream)))
fprintf(stderr, "Error %i occured while finishing compression\n", streamState);
printf("%llu bytes >> %llu bytes\n", *stream->totalBytesRead, *stream->totalBytesWritten);
// Now let's decompress it, using the density_stream_output_available_for_use() method to know how many bytes were made available
if ((streamState = density_stream_prepare(stream, outCompressed, density_stream_output_available_for_use(stream), outDecompressed, BUFFER_SIZE)))
fprintf(stderr, "Error %i when preparing decompression\n", streamState);
if ((streamState = density_stream_decompress_init(stream, NULL)))
fprintf(stderr, "Error %i when initializing decompression\n", streamState);
if ((streamState = density_stream_decompress_continue(stream))) if (streamState != DENSITY_STREAM_STATE_STALL_ON_INPUT)
fprintf(stderr, "Error %i occured during decompression\n", streamState);
if ((streamState = density_stream_decompress_finish(stream)))
fprintf(stderr, "Error %i occured while finishing compression\n", streamState);
printf("%llu bytes >> %llu bytes\n", *stream->totalBytesRead, *stream->totalBytesWritten);
// Free memory
density_stream_destroy(stream);
free(outCompressed);
free(outDecompressed);
And that's it ! We've done two compression/decompression round trips with a few lines !
If you want a more elaborate example you can checkout the SHARC project.