| constants.c |
|
619 |
| constants.h |
@file
Common constants used in decoder and encoder API.
|
7977 |
| context.c |
Common context lookup table for all context modes. |
7983 |
| context.h |
Lookup table to map the previous two bytes to a context id.
There are four different context modeling modes defined here:
CONTEXT_LSB6: context id is the least significant 6 bits of the last byte,
CONTEXT_MSB6: context id is the most significant 6 bits of the last byte,
CONTEXT_UTF8: second-order context model tuned for UTF8-encoded text,
CONTEXT_SIGNED: second-order context model tuned for signed integers.
If |p1| and |p2| are the previous two bytes, and |mode| is current context
mode, we calculate the context as:
context = ContextLut(mode)[p1] | ContextLut(mode)[p2 + 256].
For CONTEXT_UTF8 mode, if the previous two bytes are ASCII characters
(i.e. < 128), this will be equivalent to
context = 4 * context1(p1) + context2(p2),
where context1 is based on the previous byte in the following way:
0 : non-ASCII control
1 : \t, \n, \r
2 : space
3 : other punctuation
4 : " '
5 : %
6 : ( < [ {
7 : ) > ] }
8 : , ; :
9 : .
10 : =
11 : number
12 : upper-case vowel
13 : upper-case consonant
14 : lower-case vowel
15 : lower-case consonant
and context2 is based on the second last byte:
0 : control, space
1 : punctuation
2 : upper-case letter, number
3 : lower-case letter
If the last byte is ASCII, and the second last byte is not (in a valid UTF8
stream it will be a continuation byte, value between 128 and 191), the
context is the same as if the second last byte was an ASCII control or space.
If the last byte is a UTF8 lead byte (value >= 192), then the next byte will
be a continuation byte and the context id is 2 or 3 depending on the LSB of
the last byte and to a lesser extent on the second last byte if it is ASCII.
If the last byte is a UTF8 continuation byte, the second last byte can be:
- continuation byte: the next byte is probably ASCII or lead byte (assuming
4-byte UTF8 characters are rare) and the context id is 0 or 1.
- lead byte (192 - 207): next byte is ASCII or lead byte, context is 0 or 1
- lead byte (208 - 255): next byte is continuation byte, context is 2 or 3
The possible value combinations of the previous two bytes, the range of
context ids and the type of the next byte is summarized in the table below:
|--------\-----------------------------------------------------------------|
| \ Last byte |
| Second \---------------------------------------------------------------|
| last byte \ ASCII | cont. byte | lead byte |
| \ (0-127) | (128-191) | (192-) |
|=============|===================|=====================|==================|
| ASCII | next: ASCII/lead | not valid | next: cont. |
| (0-127) | context: 4 - 63 | | context: 2 - 3 |
|-------------|-------------------|---------------------|------------------|
| cont. byte | next: ASCII/lead | next: ASCII/lead | next: cont. |
| (128-191) | context: 4 - 63 | context: 0 - 1 | context: 2 - 3 |
|-------------|-------------------|---------------------|------------------|
| lead byte | not valid | next: ASCII/lead | not valid |
| (192-207) | | context: 0 - 1 | |
|-------------|-------------------|---------------------|------------------|
| lead byte | not valid | next: cont. | not valid |
| (208-) | | context: 2 - 3 | |
|-------------|-------------------|---------------------|------------------|
|
4637 |
| dictionary.c |
Embed kBrotliDictionaryData |
1558 |
| dictionary.h |
Collection of static dictionary words. |
1922 |
| dictionary_inc.h |
|
470432 |
| platform.c |
Default brotli_alloc_func |
465 |
| platform.h |
Macros for compiler / platform specific features and build options.
Build options are:
BROTLI_BUILD_32_BIT disables 64-bit optimizations
BROTLI_BUILD_64_BIT forces to use 64-bit optimizations
BROTLI_BUILD_BIG_ENDIAN forces to use big-endian optimizations
BROTLI_BUILD_ENDIAN_NEUTRAL disables endian-aware optimizations
BROTLI_BUILD_LITTLE_ENDIAN forces to use little-endian optimizations
BROTLI_BUILD_NO_RBIT disables "rbit" optimization for ARM CPUs
BROTLI_BUILD_NO_UNALIGNED_READ_FAST forces off the fast-unaligned-read
optimizations (mainly for testing purposes)
BROTLI_DEBUG dumps file name and line number when decoder detects stream
or memory error
BROTLI_ENABLE_LOG enables asserts and dumps various state information
BROTLI_ENABLE_DUMP overrides default "dump" behaviour
|
23753 |
| shared_dictionary.c |
Shared Dictionary definition and utilities. |
17347 |
| shared_dictionary_internal.h |
(Transparent) Shared Dictionary definition. |
2409 |
| static_init.h |
Central point for static initialization.
In case of "lazy" mode `BrotliXxxLazyStaticInit` is not provided by the
library. Embedder is responsible for providing it. This function should call
`BrotliXxxLazyStaticInitInner` on the first invocation. This function should
not return until execution of `BrotliXxxLazyStaticInitInner` is finished.
In C or before C++11 it is possible to call `BrotliXxxLazyStaticInitInner`
on start-up path and then `BrotliEncoderLazyStaticInit` is could be no-op;
another option is to use available thread execution controls to meet the
requirements. For possible C++11 implementation see static_init_lazy.cc.
|
1996 |
| transform.c |
RFC 7932 transforms string data |
10787 |
| transform.h |
transforms is a part of ABI, but not API.
It means that there are some functions that are supposed to be in "common"
library, but header itself is not placed into include/brotli. This way,
aforementioned functions will be available only to brotli internals.
|
3180 |
| version.h |
Version definition. |
1723 |