blob: addf4303f1657c810ce4085918dd16f7a873c6c4 [file] [log] [blame]
============================================================================
LZO -- a real-time data compression library
============================================================================
Author : Markus Franz Xaver Johannes Oberhumer
<markus@oberhumer.com>
http://www.oberhumer.com/opensource/lzo/
Version : 2.03
Date : 30 Apr 2008
Abstract
--------
LZO is a portable lossless data compression library written in ANSI C.
It offers pretty fast compression and very fast decompression.
Decompression requires no memory.
In addition there are slower compression levels achieving a quite
competitive compression ratio while still decompressing at
this very high speed.
The LZO algorithms and implementations are copyrighted OpenSource
distributed under the GNU General Public License.
Introduction
------------
LZO is a data compression library which is suitable for data
de-/compression in real-time. This means it favours speed
over compression ratio.
The acronym LZO is standing for Lempel-Ziv-Oberhumer.
LZO is written in ANSI C. Both the source code and the compressed
data format are designed to be portable across platforms.
LZO implements a number of algorithms with the following features:
- Decompression is simple and *very* fast.
- Requires no memory for decompression.
- Compression is pretty fast.
- Requires 64 kB of memory for compression.
- Allows you to dial up extra compression at a speed cost in the
compressor. The speed of the decompressor is not reduced.
- Includes compression levels for generating pre-compressed
data which achieve a quite competitive compression ratio.
- There is also a compression level which needs only 8 kB for compression.
- Algorithm is thread safe.
- Algorithm is lossless.
LZO supports overlapping compression and in-place decompression.
Design criteria
---------------
LZO was designed with speed in mind. Decompressor speed has been
favoured over compressor speed. Real-time decompression should be
possible for virtually any application. The implementation of the
LZO1X decompressor in optimized i386 assembler code runs about at
the third of the speed of a memcpy() - and even faster for many files.
In fact I first wrote the decompressor of each algorithm thereby
defining the compressed data format, verified it with manually
created test data and at last added the compressor.
Performance
-----------
To keep you interested, here is an overview of the average results
when compressing the Calgary Corpus test suite with a blocksize
of 256 kB, originally done on an ancient Intel Pentium 133.
The naming convention of the various algorithms goes LZOxx-N, where N is
the compression level. Range 1-9 indicates the fast standard levels using
64 kB memory for compression. Level 99 offers better compression at the
cost of more memory (256 kB), and is still reasonably fast.
Level 999 achieves nearly optimal compression - but it is slow
and uses much memory, and is mainly intended for generating
pre-compressed data.
The C version of LZO1X-1 is about 4-5 times faster than the fastest
zlib compression level, and it also outperforms other algorithms
like LZRW1-A and LZV in both compression ratio and compression speed
and decompression speed.
+------------------------------------------------------------------------+
| Algorithm Length CxB ComLen %Remn Bits Com K/s Dec K/s |
| --------- ------ --- ------ ----- ---- ------- ------- |
| |
| memcpy() 224401 1 224401 100.0 8.00 60956.83 59124.58 |
| |
| LZO1-1 224401 1 117362 53.1 4.25 4665.24 13341.98 |
| LZO1-99 224401 1 101560 46.7 3.73 1373.29 13823.40 |
| |
| LZO1A-1 224401 1 115174 51.7 4.14 4937.83 14410.35 |
| LZO1A-99 224401 1 99958 45.5 3.64 1362.72 14734.17 |
| |
| LZO1B-1 224401 1 109590 49.6 3.97 4565.53 15438.34 |
| LZO1B-2 224401 1 106235 48.4 3.88 4297.33 15492.79 |
| LZO1B-3 224401 1 104395 47.8 3.83 4018.21 15373.52 |
| LZO1B-4 224401 1 104828 47.4 3.79 3024.48 15100.11 |
| LZO1B-5 224401 1 102724 46.7 3.73 2827.82 15427.62 |
| LZO1B-6 224401 1 101210 46.0 3.68 2615.96 15325.68 |
| LZO1B-7 224401 1 101388 46.0 3.68 2430.89 15361.47 |
| LZO1B-8 224401 1 99453 45.2 3.62 2183.87 15402.77 |
| LZO1B-9 224401 1 99118 45.0 3.60 1677.06 15069.60 |
| LZO1B-99 224401 1 95399 43.6 3.48 1286.87 15656.11 |
| LZO1B-999 224401 1 83934 39.1 3.13 232.40 16445.05 |
| |
| LZO1C-1 224401 1 111735 50.4 4.03 4883.08 15570.91 |
| LZO1C-2 224401 1 108652 49.3 3.94 4424.24 15733.14 |
| LZO1C-3 224401 1 106810 48.7 3.89 4127.65 15645.69 |
| LZO1C-4 224401 1 105717 47.7 3.82 3007.92 15346.44 |
| LZO1C-5 224401 1 103605 47.0 3.76 2829.15 15153.88 |
| LZO1C-6 224401 1 102585 46.5 3.72 2631.37 15257.58 |
| LZO1C-7 224401 1 101937 46.2 3.70 2378.57 15492.49 |
| LZO1C-8 224401 1 100779 45.6 3.65 2171.93 15386.07 |
| LZO1C-9 224401 1 100255 45.4 3.63 1691.44 15194.68 |
| LZO1C-99 224401 1 97252 44.1 3.53 1462.88 15341.37 |
| LZO1C-999 224401 1 87740 40.2 3.21 306.44 16411.94 |
| |
| LZO1F-1 224401 1 113412 50.8 4.07 4755.97 16074.12 |
| LZO1F-999 224401 1 89599 40.3 3.23 280.68 16553.90 |
| |
| LZO1X-1(11) 224401 1 118810 52.6 4.21 4544.42 15879.04 |
| LZO1X-1(12) 224401 1 113675 50.6 4.05 4411.15 15721.59 |
| LZO1X-1 224401 1 109323 49.4 3.95 4991.76 15584.89 |
| LZO1X-1(15) 224401 1 108500 49.1 3.93 5077.50 15744.56 |
| LZO1X-999 224401 1 82854 38.0 3.04 135.77 16548.48 |
| |
| LZO1Y-1 224401 1 110820 49.8 3.98 4952.52 15638.82 |
| LZO1Y-999 224401 1 83614 38.2 3.05 135.07 16385.40 |
| |
| LZO1Z-999 224401 1 83034 38.0 3.04 133.31 10553.74 |
| |
| LZO2A-999 224401 1 87880 40.0 3.20 301.21 8115.75 |
+------------------------------------------------------------------------+
Notes:
- CxB is the number of blocks
- K/s is the speed measured in 1000 uncompressed bytes per second
- the assembler decompressors are even faster
Short documentation
-------------------
LZO is a block compression algorithm - it compresses and decompresses
a block of data. Block size must be the same for compression
and decompression.
LZO compresses a block of data into matches (a sliding dictionary)
and runs of non-matching literals. LZO takes care about long matches
and long literal runs so that it produces good results on highly
redundant data and deals acceptably with non-compressible data.
When dealing with uncompressible data, LZO expands the input
block by a maximum of 16 bytes per 1024 bytes input.
I have verified LZO using such tools as valgrind and other memory checkers.
And in addition to compressing gigabytes of files when tuning some parameters
I have also consulted various `lint' programs to spot potential portability
problems. LZO is free of any known bugs.
The algorithms
--------------
There are too many algorithms implemented. But I want to support
unlimited backward compatibility, so I will not reduce the LZO
distribution in the future.
As the many object files are mostly independent of each other, the
size overhead for an executable statically linked with the LZO library
is usually pretty low (just a few kB) because the linker will only add
the modules that you are actually using.
I first published LZO1 and LZO1A in the Internet newsgroups
comp.compression and comp.compression.research in March 1996.
They are mainly included for compatibility reasons. The LZO2A
decompressor is too slow, and there is no fast compressor anyway.
My experiments have shown that LZO1B is good with a large blocksize
or with very redundant data, LZO1F is good with a small blocksize or
with binary data and that LZO1X is often the best choice of all.
LZO1Y and LZO1Z are almost identical to LZO1X - they can achieve a
better compression ratio on some files.
Beware, your mileage may vary.
Usage of the library
--------------------
Despite of its size, the basic usage of LZO is really very simple.
Let's assume you want to compress some data with LZO1X-1:
A) compression
* include <lzo/lzo1x.h>
call lzo_init()
compress your data with lzo1x_1_compress()
* link your application with the LZO library
B) decompression
* include <lzo/lzo1x.h>
call lzo_init()
decompress your data with lzo1x_decompress()
* link your application with the LZO library
The program examples/simple.c shows a fully working example.
See also LZO.FAQ for more information.
Building LZO
------------
As LZO uses Autoconf+Automake+Libtool the building process under
UNIX systems should be very unproblematic. Shared libraries are
supported on many architectures as well.
For detailed instructions see the file INSTALL.
Please note that due to the design of the ELF executable format
the performance of a shared library on i386 systems (e.g. Linux)
is a little bit slower, so you may want to link your applications
with the static version (liblzo2.a) anyway.
For building under DOS, Win16, Win32, OS/2 and other systems
take a look at the file B/00readme.txt.
In case of troubles (like decompression data errors) try recompiling
everything without optimizations - LZO may break the optimizer
of your compiler. See the file BUGS.
LZO is written in ANSI C. In particular this means:
- your compiler must understand prototypes
- your compiler must understand prototypes in function pointers
- your compiler must correctly promote integrals ("value-preserving")
- your preprocessor must implement #elif, #error and stringizing
- you must have a conforming and correct <limits.h> header
- you must have <stddef.h>, <string.h> and other ANSI C headers
- you should have size_t and ptrdiff_t
Portability
-----------
I have built and tested LZO successfully on a variety of platforms
including DOS (16 + 32 bit), Windows 3.x (16-bit), Win32, Win64,
Linux, *BSD, HP-UX and many more.
LZO is also reported to work under AIX, ConvexOS, IRIX, MacOS, PalmOS (Pilot),
PSX (Sony Playstation), Solaris, SunOS, TOS (Atari ST) and VxWorks.
Furthermore it is said that its performance on a Cray is superior
to all other machines...
And I think it would be much fun to translate the decompressors
to Z-80 or 6502 assembly.
The future
----------
Here is what I'm planning for the next months. No promises, though...
- interfaces to .NET and Mono
- interfaces to Perl, Java, Python, Delphi, Visual Basic, ...
- improve documentation and API reference
Some comments about the source code
-----------------------------------
Be warned: the main source code in the `src' directory is a
real pain to understand as I've experimented with hundreds of slightly
different versions. It contains many #if and some gotos, and
is *completely optimized for speed* and not for readability.
Code sharing of the different algorithms is implemented by stressing
the preprocessor - this can be really confusing. Lots of marcos and
assertions don't make things better.
Nevertheless LZO compiles very quietly on a variety of
compilers with the highest warning levels turned on, even
in C++ mode.
Copyright
---------
LZO is Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
2005, 2006, 2007, 2008 Markus Franz Xaver Johannes Oberhumer
LZO is distributed under the terms of the GNU General Public License (GPL).
See the file COPYING.
Special licenses for commercial and other applications which
are not willing to accept the GNU General Public License
are available by contacting the author.