| 1 | /* Copyright 2013 Google Inc. All Rights Reserved. |
| 2 | |
| 3 | Distributed under MIT license. |
| 4 | See file LICENSE for detail or copy at https://opensource.org/licenses/MIT |
| 5 | */ |
| 6 | |
| 7 | /* Lookup table to map the previous two bytes to a context id. |
| 8 | |
| 9 | There are four different context modeling modes defined here: |
| 10 | CONTEXT_LSB6: context id is the least significant 6 bits of the last byte, |
| 11 | CONTEXT_MSB6: context id is the most significant 6 bits of the last byte, |
| 12 | CONTEXT_UTF8: second-order context model tuned for UTF8-encoded text, |
| 13 | CONTEXT_SIGNED: second-order context model tuned for signed integers. |
| 14 | |
| 15 | If |p1| and |p2| are the previous two bytes, and |mode| is current context |
| 16 | mode, we calculate the context as: |
| 17 | |
| 18 | context = ContextLut(mode)[p1] | ContextLut(mode)[p2 + 256]. |
| 19 | |
| 20 | For CONTEXT_UTF8 mode, if the previous two bytes are ASCII characters |
| 21 | (i.e. < 128), this will be equivalent to |
| 22 | |
| 23 | context = 4 * context1(p1) + context2(p2), |
| 24 | |
| 25 | where context1 is based on the previous byte in the following way: |
| 26 | |
| 27 | 0 : non-ASCII control |
| 28 | 1 : \t, \n, \r |
| 29 | 2 : space |
| 30 | 3 : other punctuation |
| 31 | 4 : " ' |
| 32 | 5 : % |
| 33 | 6 : ( < [ { |
| 34 | 7 : ) > ] } |
| 35 | 8 : , ; : |
| 36 | 9 : . |
| 37 | 10 : = |
| 38 | 11 : number |
| 39 | 12 : upper-case vowel |
| 40 | 13 : upper-case consonant |
| 41 | 14 : lower-case vowel |
| 42 | 15 : lower-case consonant |
| 43 | |
| 44 | and context2 is based on the second last byte: |
| 45 | |
| 46 | 0 : control, space |
| 47 | 1 : punctuation |
| 48 | 2 : upper-case letter, number |
| 49 | 3 : lower-case letter |
| 50 | |
| 51 | If the last byte is ASCII, and the second last byte is not (in a valid UTF8 |
| 52 | stream it will be a continuation byte, value between 128 and 191), the |
| 53 | context is the same as if the second last byte was an ASCII control or space. |
| 54 | |
| 55 | If the last byte is a UTF8 lead byte (value >= 192), then the next byte will |
| 56 | be a continuation byte and the context id is 2 or 3 depending on the LSB of |
| 57 | the last byte and to a lesser extent on the second last byte if it is ASCII. |
| 58 | |
| 59 | If the last byte is a UTF8 continuation byte, the second last byte can be: |
| 60 | - continuation byte: the next byte is probably ASCII or lead byte (assuming |
| 61 | 4-byte UTF8 characters are rare) and the context id is 0 or 1. |
| 62 | - lead byte (192 - 207): next byte is ASCII or lead byte, context is 0 or 1 |
| 63 | - lead byte (208 - 255): next byte is continuation byte, context is 2 or 3 |
| 64 | |
| 65 | The possible value combinations of the previous two bytes, the range of |
| 66 | context ids and the type of the next byte is summarized in the table below: |
| 67 | |
| 68 | |--------\-----------------------------------------------------------------| |
| 69 | | \ Last byte | |
| 70 | | Second \---------------------------------------------------------------| |
| 71 | | last byte \ ASCII | cont. byte | lead byte | |
| 72 | | \ (0-127) | (128-191) | (192-) | |
| 73 | |=============|===================|=====================|==================| |
| 74 | | ASCII | next: ASCII/lead | not valid | next: cont. | |
| 75 | | (0-127) | context: 4 - 63 | | context: 2 - 3 | |
| 76 | |-------------|-------------------|---------------------|------------------| |
| 77 | | cont. byte | next: ASCII/lead | next: ASCII/lead | next: cont. | |
| 78 | | (128-191) | context: 4 - 63 | context: 0 - 1 | context: 2 - 3 | |
| 79 | |-------------|-------------------|---------------------|------------------| |
| 80 | | lead byte | not valid | next: ASCII/lead | not valid | |
| 81 | | (192-207) | | context: 0 - 1 | | |
| 82 | |-------------|-------------------|---------------------|------------------| |
| 83 | | lead byte | not valid | next: cont. | not valid | |
| 84 | | (208-) | | context: 2 - 3 | | |
| 85 | |-------------|-------------------|---------------------|------------------| |
| 86 | */ |
| 87 | |
| 88 | #ifndef BROTLI_COMMON_CONTEXT_H_ |
| 89 | #define BROTLI_COMMON_CONTEXT_H_ |
| 90 | |
| 91 | #include <brotli/port.h> |
| 92 | #include <brotli/types.h> |
| 93 | |
| 94 | typedef enum ContextType { |
| 95 | CONTEXT_LSB6 = 0, |
| 96 | CONTEXT_MSB6 = 1, |
| 97 | CONTEXT_UTF8 = 2, |
| 98 | CONTEXT_SIGNED = 3 |
| 99 | } ContextType; |
| 100 | |
| 101 | /* "Soft-private", it is exported, but not "advertised" as API. */ |
| 102 | /* Common context lookup table for all context modes. */ |
| 103 | BROTLI_COMMON_API extern const uint8_t _kBrotliContextLookupTable[2048]; |
| 104 | |
| 105 | typedef const uint8_t* ContextLut; |
| 106 | |
| 107 | /* typeof(MODE) == ContextType; returns ContextLut */ |
| 108 | #define BROTLI_CONTEXT_LUT(MODE) (&_kBrotliContextLookupTable[(MODE) << 9]) |
| 109 | |
| 110 | /* typeof(LUT) == ContextLut */ |
| 111 | #define BROTLI_CONTEXT(P1, P2, LUT) ((LUT)[P1] | ((LUT) + 256)[P2]) |
| 112 | |
| 113 | #endif /* BROTLI_COMMON_CONTEXT_H_ */ |
| 114 | |