view tools/oldlzma/lzma.txt @ 421:c685b3c5f40b

Update MPD's Makefile.
author Aleksandr Rybalko <>
date Fri, 14 Sep 2012 02:20:50 +0300
parents 62ac8289dac7
line wrap: on
line source
LZMA SDK 4.17 

LZMA SDK 4.17  Copyright (C) 1999-2005 Igor Pavlov

LZMA SDK provides developers with documentation, source code,
and sample code necessary to write software that uses LZMA compression. 

LZMA is default and general compression method of 7z format
in 7-Zip compression program ( LZMA provides high 
compression ratio and very fast decompression.

LZMA is an improved version of famous LZ77 compression algorithm. 
It was improved in way of maximum increasing of compression ratio,
keeping high decompression speed and low memory requirements for 


LZMA SDK is licensed under two licenses:

1) GNU Lesser General Public License (GNU LGPL)
2) Common Public License (CPL)

It means that you can select one of these two licenses and 
follow rules of that license.

Igor Pavlov, as the author of this code, expressly permits you 
to statically or dynamically link your code (or bind by name) 
to the files from LZMA SDK without subjecting your linked 
code to the terms of the CPL or GNU LGPL. 
Any modifications or additions to files from LZMA SDK, however, 
are subject to the GNU LGPL or CPL terms.

GNU LGPL and CPL licenses are pretty similar and both these
licenses are classified as 

1) "Free software licenses" at 
2) "OSI-approved" at

LZMA SDK also can be available under a proprietary license for 
those who cannot use the GNU LGPL or CPL in their code. To request
such proprietary license or any additional consultations,
send email message from that page:

You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

You should have received a copy of the Common Public License
along with this library.

LZMA SDK Contents

LZMA SDK includes:

  - C++ source code of LZMA Encoder and Decoder
  - C++ source code for file->file LZMA compressing and decompressing
  - ANSI-C compatible source code for LZMA decompressing
  - Compiled file->file LZMA compressing/decompressing program for Windows system

ANSI-C LZMA decompression code was ported from original C++ sources to C.
Also it was simplified and optimized for code size. 
But it is fully compatible with LZMA from 7-Zip.

UNIX/Linux version 
To compile C++ version of file->file LZMA, go to directory
and type "make" or "make clean all" to recompile all.

In some UNIX/Linux versions you must compile LZMA with static libraries.
To compile with static libraries, change string in makefile
LIB = -lm
to string  
LIB = -lm -static

SRC      - directory with source code
lzma.txt - LZMA SDK description (this file)
7zFormat.txt - 7z Format description
7zC.txt  - 7z ANSI-C Decoder description (this file)
methods.txt  - Compression method IDs for .7z
LGPL.txt - GNU Lesser General Public License
CPL.html - Common Public License
lzma.exe - Compiled file->file LZMA encoder/decoder for Windows
history.txt - history of the LZMA SDK

Source code structure

  Common  - common files for C++ projects
  Windows - common files for Windows related code
  7zip    - files related to 7-Zip Project
    Common   - common files for 7-Zip
    Compress - files related to compression/decompression
      LZ     - files related to LZ (Lempel-Ziv) compression algorithm
        BinTree    - Binary Tree Match Finder for LZ algorithm
        HashChain  - Hash Chain Match Finder for LZ algorithm
        Patricia   - Patricia Match Finder for LZ algorithm
      RangeCoder   - Range Coder (special code of compression/decompression)
      LZMA         - LZMA compression/decompression on C++
      LZMA_Alone   - file->file LZMA compression/decompression
      LZMA_C       - ANSI-C compatible LZMA decompressor
        LzmaDecode.h  - interface for LZMA decoding on ANSI-C
        LzmaDecode.c      - LZMA decoding on ANSI-C (new fastest version)
        LzmaDecodeSize.c  - LZMA decoding on ANSI-C (old size-optimized version)
        LzmaTest.c    - test application that decodes LZMA encoded file
      Branch       - Filters for x86, IA-64, ARM, ARM-Thumb, PowerPC and SPARC code
    Archive - files related to archiving
      7z_C     - 7z ANSI-C Decoder

Source code of LZMA SDK is only part of big 7-Zip project. That is 
why LZMA SDK uses such complex source code structure. 

You can find ANSI-C LZMA decompressing code at folder 
7-Zip doesn't use that ANSI-C LZMA code and that code was developed 
specially for this SDK. And files from LZMA_C do not need files from 
other directories of SDK for compiling.

7-Zip source code can be downloaded from 7-Zip's SourceForge page:

LZMA Decompression features
  - Variable dictionary size (up to 256 MB)
  - Estimated compressing speed: about 500 KB/s on 1 GHz CPU
  - Estimated decompressing speed: 
      - 8-12 MB/s on 1 GHz Intel Pentium 3 or AMD Athlon
      - 500-1000 KB/s on 100 MHz ARM, MIPS, PowerPC or other simple RISC
  - Small memory requirements for decompressing (8-32 KB + DictionarySize)
  - Small code size for decompressing: 2-8 KB (depending from 
    speed optimizations) 

LZMA decoder uses only integer operations and can be 
implemented in any modern 32-bit CPU (or on 16-bit CPU with some conditions).

Some critical operations that affect to speed of LZMA decompression:
  1) 32*16 bit integer multiply
  2) Misspredicted branches (penalty mostly depends from pipeline length)
  3) 32-bit shift and arithmetic operations

Speed of LZMA decompression mostly depends from CPU speed.
Memory speed has no big meaning. But if your CPU has small data cache, 
overall weight of memory speed will slightly increase.

How To Use

Using LZMA encoder/decoder executable

Usage:  LZMA <e|d> inputFile outputFile [<switches>...]

  e: encode file

  d: decode file

  b: Benchmark. There are two tests: compressing and decompressing 
     with LZMA method. Benchmark shows rating in MIPS (million 
     instructions per second). Rating value is calculated from 
     measured speed and it is normalized with AMD Athlon XP CPU
     results. Also Benchmark checks possible hardware errors (RAM 
     errors in most cases). Benchmark uses these settings:
     (-a1, -d21, -fb32, -mfbt4). You can change only -d. Also you 
     can change number of iterations. Example for 30 iterations:
	LZMA b 30
     Default number of iterations is 10.


  -a{N}:  set compression mode 0 = fast, 1 = normal, 2 = max
          default: 2 (max)

  d{N}:   Sets Dictionary size - [0, 28], default: 23 (8MB)
          The maximum value for dictionary size is 256 MB = 2^28 bytes.
          Dictionary size is calculated as DictionarySize = 2^N bytes. 
          For decompressing file compressed by LZMA method with dictionary 
          size D = 2^N you need about D bytes of memory (RAM).

  -fb{N}: set number of fast bytes - [5, 255], default: 128
          Usually big number gives a little bit better compression ratio 
          and slower compression process.

  -lc{N}: set number of literal context bits - [0, 8], default: 3
          Sometimes lc=4 gives gain for big files.

  -lp{N}: set number of literal pos bits - [0, 4], default: 0
          lp switch is intended for periodical data when period is 
          equal 2^N. For example, for 32-bit (4 bytes) 
          periodical data you can use lp=2. Often it's better to set lc0, 
          if you change lp switch.

  -pb{N}: set number of pos bits - [0, 4], default: 2
          pb switch is intended for periodical data 
          when period is equal 2^N.

  -mf{MF_ID}: set Match Finder. Default: bt4. 
              Compression ratio for all bt* and pat* almost the same.
              Algorithms from hc* group doesn't provide good compression 
              ratio, but they often works pretty fast in combination with 
              fast mode (-a0). Methods from bt* group require less memory 
              than methods from pat* group. Usually bt4 works faster than 
              any pat*, but for some types of files pat* can work faster. 

              Memory requirements depend from dictionary size 
              (parameter "d" in table below). 

               MF_ID     Memory                   Description

                bt2    d*9.5 +  1MB  Binary Tree with 2 bytes hashing.
                bt3    d*9.5 + 65MB  Binary Tree with 2-3(full) bytes hashing.
                bt4    d*9.5 +  6MB  Binary Tree with 2-3-4 bytes hashing.
                bt4b   d*9.5 + 34MB  Binary Tree with 2-3-4(big) bytes hashing.
                pat2r  d*26  +  1MB  Patricia Tree with 2-bits nodes, removing.
                pat2   d*38  +  1MB  Patricia Tree with 2-bits nodes.
                pat2h  d*38  + 77MB  Patricia Tree with 2-bits nodes, 2-3 bytes hashing.
                pat3h  d*62  + 85MB  Patricia Tree with 3-bits nodes, 2-3 bytes hashing.
                pat4h  d*110 +101MB  Patricia Tree with 4-bits nodes, 2-3 bytes hashing.
                hc3    d*5.5 +  1MB  Hash Chain with 2-3 bytes hashing.
                hc4    d*5.5 +  6MB  Hash Chain with 2-3-4 bytes hashing.

  -eos:   write End Of Stream marker. By default LZMA doesn't write 
          eos marker, since LZMA decoder knows uncompressed size 
          stored in .lzma file header.

  -si:    Read data from stdin (it will write End Of Stream marker).
  -so:    Write data to stdout


1) LZMA e file.bin file.lzma -d16 -lc0 

compresses file.bin to file.lzma with 64 KB dictionary (2^16=64K)  
and 0 literal context bits. -lc0 allows to reduce memory requirements 
for decompression.

2) LZMA e file.bin file.lzma -lc0 -lp2

compresses file.bin to file.lzma with settings suitable 
for 32-bit periodical data (for example, ARM or MIPS code).

3) LZMA d file.lzma file.bin

decompresses file.lzma to file.bin.

Compression ratio hints


To increase compression ratio for LZMA compressing it's desirable 
to have aligned data (if it's possible) and also it's desirable to locate
data in such order, where code is grouped in one place and data is 
grouped in other place (it's better than such mixing: code, data, code,
data, ...).

Using Filters
You can increase compression ratio for some data types, using
special filters before compressing. For example, it's possible to 
increase compression ratio on 5-10% for code for those CPU ISAs: 
x86, IA-64, ARM, ARM-Thumb, PowerPC, SPARC.

You can find C/C++ source code of such filters in folder "7zip/Compress/Branch"

You can check compression ratio gain of these filters with such 
7-Zip commands (example for ARM code):
No filter:
  7z a a1.7z a.bin -m0=lzma

With filter for little-endian ARM code:
  7z a a2.7z a.bin -m0=bc_arm -m1=lzma        

With filter for big-endian ARM code (using additional Swap4 filter):
  7z a a3.7z a.bin -m0=swap4 -m1=bc_arm -m2=lzma

It works in such manner:
Compressing    = Filter_encoding + LZMA_encoding
Decompressing  = LZMA_decoding + Filter_decoding

Compressing and decompressing speed of such filters is very high,
so it will not increase decompressing time too much.
Moreover, it reduces decompression time for LZMA_decoding, 
since compression ratio with filtering is higher.

These filters convert CALL (calling procedure) instructions 
from relative offsets to absolute addresses, so such data becomes more 
compressible. Source code of these CALL filters is pretty simple
(about 20 lines of C++), so you can convert it from C++ version yourself.

For some ISAs (for example, for MIPS) it's impossible to get gain from such filter.

LZMA compressed file format
Offset Size Description
  0     1   Special LZMA properties for compressed data
  1     4   Dictionary size (little endian)
  5     8   Uncompressed size (little endian). -1 means unknown size
 13         Compressed data


To use ANSI-C LZMA Decoder you need to files:
LzmaDecode.h and one of the following two files:
1) LzmaDecode.c      - LZMA decoding on ANSI-C (new fastest version)
2) LzmaDecodeSize.c  - LZMA decoding on ANSI-C (old size-optimized version)
use LzmaDecode.c, if you need fastest code.

Memory requirements for LZMA decoding

LZMA decoder doesn't allocate memory itself, so you must 
calculate required memory, allocate it and send it to LZMA.

Stack usage of LZMA function for local variables is not 
larger than 200 bytes.

Memory requirements for decompression depend 
from interface that you want to use:

  a) Memory to memory decompression:
    M1 = (inputSize + outputSize + lzmaInternalSize).

  b) Decompression with buffering:

    M2 = (inputBufferSize + outputBufferSize + dictionarySize + lzmaInternalSize)

How To decompress data

1) Read first byte of properties for LZMA compressed stream, 
   check that it has correct value and calculate three 
   LZMA property variables:

  int lc, lp, pb;
  unsigned char prop0 = properties[0];
  if (prop0 >= (9*5*5))
    sprintf(rs + strlen(rs), "\n properties error");
    return 1;
  for (pb = 0; prop0 >= (9 * 5); 
    pb++, prop0 -= (9 * 5));
  for (lp = 0; prop0 >= 9; 
    lp++, prop0 -= 9);
  lc = prop0;

2) Calculate required amount for LZMA lzmaInternalSize:

  lzmaInternalSize = (LZMA_BASE_SIZE + (LZMA_LIT_SIZE << (lc + lp))) * 


  LZMA decoder uses array of CProb variables as internal structure.
  By default, CProb is (unsigned short)
  But you can define _LZMA_PROB32 to make it (unsigned int)
  It can increase speed on some 32-bit CPUs, but memory usage will 
  be doubled in that case.

  2b) If you use Decompression with buffering, add 100 bytes to 
      #ifdef _LZMA_OUT_READ
      lzmaInternalSize += 100;

3) Allocate that memory with malloc or some other function:

  lzmaInternalData = malloc(lzmaInternalSize);

4) Decompress data:

  4a) If you use simple memory to memory decompression:

    int result = LzmaDecode(lzmaInternalData, lzmaInternalSize,
        lc, lp, pb,
        unsigned char *inStream, unsigned int inSize,
        unsigned char *outStream, unsigned int outSize, 

  4b) If you use Decompression with buffering

    4.1) Read dictionary size from properties

      unsigned int dictionarySize = 0;
      int i;
      for (i = 0; i < 4; i++)
        dictionarySize += (unsigned int)(b) << (i * 8);

    4.2) Allocate memory for dictionary

      unsigned char *dictionary = malloc(dictionarySize);

    4.3) Initialize LZMA decoder:

    LzmaDecoderInit((unsigned char *)lzmaInternalData, lzmaInternalSize,
        lc, lp, pb,
        dictionary, dictionarySize,

    4.4) In loop call LzmaDecoderCode function:

    for (nowPos = 0; nowPos < outSize;)
      unsigned int blockSize = outSize - nowPos;
      unsigned int kBlockSize = 0x10000;
      if (blockSize > kBlockSize)
        blockSize = kBlockSize;
      res = LzmaDecode((unsigned char *)lzmaInternalData, 
      ((unsigned char *)outStream) + nowPos, blockSize, &outSizeProcessed);
      if (res != 0)
        printf("\nerror = %d\n", res);
      nowPos += outSizeProcessed;
      if (outSizeProcessed == 0)
        outSize = nowPos;

EXIT codes

LZMA decoder can return one of the following codes:

#define LZMA_RESULT_OK 0

If you use callback function for input data and you return some 
error code, LZMA Decoder also returns that code.

LZMA Defines

_LZMA_IN_CB    - Use callback for input data

_LZMA_OUT_READ - Use read function for output data

_LZMA_LOC_OPT  - Enable local speed optimizations inside code.
                 _LZMA_LOC_OPT is only for LzmaDecodeSize.c (size-optimized version).
                 _LZMA_LOC_OPT doesn't affect LzmaDecode.c (speed-optimized version)

_LZMA_PROB32   - It can increase speed on some 32-bit CPUs, 
                 but memory usage will be doubled in that case

_LZMA_UINT32_IS_ULONG  - Define it if int is 16-bit on your compiler
                         and long is 32-bit.

1) please note that LzmaTest.c doesn't free allocated memory in some cases. 
But in your real applicaions you must free memory after decompression.

2) All numbers above were calculated for case when int is not more than 
  32-bit in your compiler. If in your compiler int is 64-bit or larger 
  probably LZMA can require more memory for some structures.

C++ LZMA Encoder/Decoder 
C++ LZMA code use COM-like interfaces. So if you want to use it, 
you can study basics of COM/OLE.

By default, LZMA Encoder contains all Match Finders.
But for compressing it's enough to have just one of them.
So for reducing size of compressing code you can define:
  #define COMPRESS_MF_BT
  #define COMPRESS_MF_BT4
and it will use only bt4 match finder.