Home | History | Annotate | Download | only in lz4 
      1 LZ4 Format Description
      2 Last revised: 2012-02-27
      3 Author : Y. Collet
      4 
      5 
      6 
      7 This small specification intents to provide enough information
      8 to anyone willing to produce LZ4-compatible compressed data blocks
      9 using any programming language.
     10 
     11 LZ4 is an LZ77-type compressor with a fixed, byte-oriented encoding.
     12 The most important design principle behind LZ4 is simplicity.
     13 It helps to create an easy to read and maintain source code.
     14 It also helps later on for optimisations, compactness, and speed.
     15 There is no entropy encoder backend nor framing layer.
     16 The latter is assumed to be handled by other parts of the system.
     17 
     18 This document only describes the format,
     19 not how the LZ4 compressor nor decompressor actually work.
     20 The correctness of the decompressor should not depend
     21 on implementation details of the compressor, and vice versa.
     22 
     23 
     24 
     25 -- Compressed block format --
     26 
     27 An LZ4 compressed block is composed of sequences.
     28 Schematically, a sequence is a suite of literals, followed by a match copy.
     29 
     30 Each sequence starts with a token.
     31 The token is a one byte value, separated into two 4-bits fields.
     32 Therefore each field ranges from 0 to 15.
     33 
     34 
     35 The first field uses the 4 high-bits of the token.
     36 It provides the length of literals to follow.
     37 (Note : a literal is a not-compressed byte).
     38 If the field value is 0, then there is no literal.
     39 If it is 15, then we need to add some more bytes to indicate the full length.
     40 Each additionnal byte then represent a value from 0 to 255,
     41 which is added to the previous value to produce a total length.
     42 When the byte value is 255, another byte is output.
     43 There can be any number of bytes following the token. There is no "size limit".
     44 (Sidenote this is why a not-compressible input block is expanded by 0.4%).
     45 
     46 Example 1 : A length of 48 will be represented as :
     47 - 15 : value for the 4-bits High field
     48 - 33 : (=48-15) remaining length to reach 48
     49 
     50 Example 2 : A length of 280 will be represented as :
     51 - 15  : value for the 4-bits High field
     52 - 255 : following byte is maxed, since 280-15 >= 255
     53 - 10  : (=280 - 15 - 255) ) remaining length to reach 280
     54 
     55 Example 3 : A length of 15 will be represented as :
     56 - 15 : value for the 4-bits High field
     57 - 0  : (=15-15) yes, the zero must be output
     58 
     59 Following the token and optional length bytes, are the literals themselves.
     60 They are exactly as numerous as previously decoded (length of literals).
     61 It's possible that there are zero literal.
     62 
     63 
     64 Following the literals is the match copy operation.
     65 
     66 It starts by the offset.
     67 This is a 2 bytes value, in little endian format.
     68 
     69 The offset represents the position of the match to be copied from.
     70 1 means "current position - 1 byte".
     71 The maximum offset value is 65535, 65536 cannot be coded.
     72 Note that 0 is an invalid value, not used. 
     73 
     74 Then we need to extract the match length.
     75 For this, we use the second token field, the low 4-bits.
     76 Value, obviously, ranges from 0 to 15.
     77 However here, 0 means that the copy operation will be minimal.
     78 The minimum length of a match, called minmatch, is 4. 
     79 As a consequence, a 0 value means 4 bytes, and a value of 15 means 19+ bytes.
     80 Similar to literal length, on reaching the highest possible value (15), 
     81 we output additional bytes, one at a time, with values ranging from 0 to 255.
     82 They are added to total to provide the final match length.
     83 A 255 value means there is another byte to read and add.
     84 There is no limit to the number of optional bytes that can be output this way.
     85 (This points towards a maximum achievable compression ratio of ~250).
     86 
     87 With the offset and the matchlength,
     88 the decoder can now proceed to copy the data from the already decoded buffer.
     89 On decoding the matchlength, we reach the end of the compressed sequence,
     90 and therefore start another one.
     91 
     92 
     93 -- Parsing restrictions --
     94 
     95 There are specific parsing rules to respect in order to remain compatible
     96 with assumptions made by the decoder :
     97 1) The last 5 bytes are always literals
     98 2) The last match must start at least 12 bytes before end of block
     99 Consequently, a block with less than 13 bytes cannot be compressed.
    100 These rules are in place to ensure that the decoder
    101 will never read beyond the input buffer, nor write beyond the output buffer.
    102 
    103 Note that the last sequence is also incomplete,
    104 and stops right after literals.
    105 
    106 
    107 -- Additional notes --
    108 
    109 There is no assumption nor limits to the way the compressor
    110 searches and selects matches within the source data block.
    111 It could be a fast scan, a multi-probe, a full search using BST,
    112 standard hash chains or MMC, well whatever.
    113 
    114 Advanced parsing strategies can also be implemented, such as lazy match,
    115 or full optimal parsing.
    116 
    117 All these trade-off offer distinctive speed/memory/compression advantages.
    118 Whatever the method used by the compressor, its result will be decodable
    119 by any LZ4 decoder if it follows the format specification described above.
    120 
    121