| [ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
5.2 The coding contexts
These contexts (‘Bit_model’ in the source), are integers or arrays of integers representing the probability of the corresponding bit being 0.
The indices used in these arrays are:
- ‘state’
A state machine (‘State’ in the source) with 12 states (0 to 11), coding the latest 2 to 4 types of sequences processed. The initial state is 0.
- ‘pos_state’
Value of the 2 least significant bits of the current position in the decoded data.
- ‘literal_state’
Value of the 3 most significant bits of the latest byte decoded.
- ‘len_state’
Coded value of length (length - 2), with a maximum of 3. The resulting value is in the range 0 to 3.
In the following table, ‘!literal’ is any sequence except a literal byte. ‘rep’ is any one of ‘rep0’, ‘rep1’, ‘rep2’ or ‘rep3’. The types of previous sequences corresponding to each state are:
| State | Types of previous sequences |
|---|---|
| 0 | literal, literal, literal |
| 1 | match, literal, literal |
| 2 | (rep or shortrep), literal, literal |
| 3 | literal, shortrep, literal, literal |
| 4 | match, literal |
| 5 | (rep or shortrep), literal |
| 6 | literal, shortrep, literal |
| 7 | literal, match |
| 8 | literal, rep |
| 9 | literal, shortrep |
| 10 | !literal, match |
| 11 | !literal, (rep or shortrep) |
The contexts for decoding the type of coding sequence are:
| Name | Indices | Used when |
|---|---|---|
| bm_match | state, pos_state | sequence start |
| bm_rep | state | after sequence 1 |
| bm_rep0 | state | after sequence 11 |
| bm_rep1 | state | after sequence 111 |
| bm_rep2 | state | after sequence 1111 |
| bm_len | state, pos_state | after sequence 110 |
The contexts for decoding distances are:
| Name | Indices | Used when |
|---|---|---|
| bm_dis_slot | len_state, bit tree | distance start |
| bm_dis | reverse bit tree | after slots 4 to 13 |
| bm_align | reverse bit tree | for distances >= 128, after fixed probability bits |
There are two separate sets of contexts for lengths (‘Len_model’ in the source). One for normal matches, the other for repeated matches. The contexts in each Len_model are (see ‘decode_len’ in the source):
| Name | Indices | Used when |
|---|---|---|
| choice1 | none | length start |
| choice2 | none | after sequence 1 |
| bm_low | pos_state, bit tree | after sequence 0 |
| bm_mid | pos_state, bit tree | after sequence 10 |
| bm_high | bit tree | after sequence 11 |
The context array ‘bm_literal’ is special. In principle it acts as a normal bit tree context, the one selected by ‘literal_state’. But if the previous decoded byte was not a literal, two other bit tree contexts are used depending on the value of each bit in ‘match_byte’ (the byte at the latest used distance), until a bit is decoded that is different from its corresponding bit in ‘match_byte’. After the first difference is found, the rest of the byte is decoded using the normal bit tree context. (See ‘decode_matched’ in the source).
| [ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This document was generated on October 10, 2013 using texi2html 5.0.