Project

General

Profile

Download (21.1 KB) Statistics
| Branch: | Revision:
#include "aricoder.h"

#include "bitops.h"

#include <algorithm>
#include <cstdlib>
#include <functional>
#include <limits>

template <std::uint8_t bit>
void ArithmeticBitWriter::write_bit() {
// add bit at last position
curr_byte_ = (curr_byte_ << 1) | bit;
// increment bit position
curr_bit_++;

// write bit if done
if (curr_bit_ == 8) {
data_.emplace_back(curr_byte_);
curr_bit_ = 0;
}
}

void ArithmeticBitWriter::write_n_zero_bits(std::size_t n) {
if (n + curr_bit_ >= 8) {
auto remainingBits = 8 - curr_bit_;
n -= remainingBits;
curr_byte_ <<= remainingBits;
data_.emplace_back(curr_byte_);
curr_bit_ = 0;
}

while (n >= 8) {
data_.emplace_back(0);
n -= 8;
}

curr_byte_ <<= n;
curr_bit_ += n;
}

void ArithmeticBitWriter::write_n_one_bits(std::size_t n) {
constexpr std::uint8_t all_ones = std::numeric_limits<std::uint8_t>::max();
if (n + curr_bit_ >= 8) {
auto remainingBits = 8 - curr_bit_;
n -= remainingBits;
curr_byte_ <<= remainingBits;
curr_byte_ |= all_ones >> (8 - remainingBits);
data_.emplace_back(curr_byte_);
curr_bit_ = 0;
}

while (n >= 8) {
data_.emplace_back(all_ones);
n -= 8;
}

curr_byte_ = (curr_byte_ << n) | (all_ones >> (8 - n));
curr_bit_ += n;
}

void ArithmeticBitWriter::pad() {
while (curr_bit_ > 0) {
write_bit<0>();
}
}

std::vector<std::uint8_t> ArithmeticBitWriter::get_data() const {
return data_;
}

ArithmeticDecoder::ArithmeticDecoder(Reader& reader) : reader_(reader) {
// code buffer has to be filled before starting decoding
for (std::uint32_t i = 0; i < CODER_USE_BITS; i++ ) {
ccode = ( ccode << 1 ) | read_bit();
}
}

ArithmeticEncoder::ArithmeticEncoder(Writer& writer) : writer_(writer) {}

ArithmeticEncoder::~ArithmeticEncoder() {
if (!finalized) {
this->finalize();
}
}

void ArithmeticEncoder::finalize() {
if (finalized) {
return;
}
// due to clow < CODER_LIMIT050, and chigh >= CODER_LIMIT050
// there are only two possible cases
if (clow < CODER_LIMIT025) {
bitwriter_->write_bit<0>();
bitwriter_->write_bit<1>();
bitwriter_->write_n_one_bits(nrbits);
nrbits = 0;
} else {
// case b.), clow >= CODER_LIMIT025
bitwriter_->write_bit<1>();
}
// done, zeroes are auto-read by the decoder

bitwriter_->pad(); // Pad code with zeroes.
writer_.write(bitwriter_->get_data());
finalized = true;
}

/* -----------------------------------------------
arithmetic encoder function
----------------------------------------------- */
void ArithmeticEncoder::encode( symbol* s )
{
// Make local copies of clow_ and chigh_ for cache performance:
uint32_t clow_local = clow;
uint32_t chigh_local = chigh;
// update steps, low count, high count
cstep = (chigh_local - clow_local + 1) / s->scale;
chigh_local = clow_local + (cstep * s->high_count) - 1;
clow_local = clow_local + (cstep * s->low_count);
// e3 scaling is performed for speed and to avoid underflows
// if both, low and high are either in the lower half or in the higher half
// one bit can be safely shifted out
while ( clow_local >= CODER_LIMIT050 || chigh_local < CODER_LIMIT050 ) {
if (chigh_local < CODER_LIMIT050 ) { // this means both, high and low are below, and 0 can be safely shifted out
// write 0 bit
bitwriter_->write_bit<0>();
// shift out remaing e3 bits
bitwriter_->write_n_one_bits(nrbits);
nrbits = 0;
}
else { // if the first wasn't the case, it's clow >= CODER_LIMIT050
// write 1 bit
bitwriter_->write_bit<1>();
clow_local &= CODER_LIMIT050 - 1;
chigh_local &= CODER_LIMIT050 - 1;
// shift out remaing e3 bits
bitwriter_->write_n_zero_bits(nrbits);
nrbits = 0;
}
clow_local <<= 1;
chigh_local <<= 1;
chigh_local++;
}
// e3 scaling, to make sure that theres enough space between low and high
while ( (clow_local >= CODER_LIMIT025 ) && (chigh_local < CODER_LIMIT075 ) ) {
nrbits++;
clow_local &= CODER_LIMIT025 - 1;
chigh_local ^= CODER_LIMIT025 + CODER_LIMIT050;
// clow -= CODER_LIMIT025;
// chigh -= CODER_LIMIT025;
clow_local <<= 1;
chigh_local <<= 1;
chigh_local++;
}

clow = clow_local;
chigh = chigh_local;
}


/* -----------------------------------------------
arithmetic decoder get count function
----------------------------------------------- */
unsigned int ArithmeticDecoder::decode_count( symbol* s )
{
// update cstep, which is needed to remove the symbol from the stream later
cstep = ( ( chigh - clow ) + 1 ) / s->scale;
// return counts, needed to decode the symbol from the statistical model
return ( ccode - clow ) / cstep;
}

/* -----------------------------------------------
arithmetic decoder function
----------------------------------------------- */
void ArithmeticDecoder::decode( symbol* s )
{
// no actual decoding takes place, as this has to happen in the statistical model
// the symbol has to be removed from the stream, though
// alread have steps updated from decoder_count
// update low count and high count
uint32_t ccode_local = ccode;
uint32_t clow_local = clow;
uint32_t chigh_local = clow_local + (cstep * s->high_count) - 1;
clow_local = clow_local + (cstep * s->low_count);
// e3 scaling is performed for speed and to avoid underflows
// if both, low and high are either in the lower half or in the higher half
// one bit can be safely shifted out
while ( (clow_local >= CODER_LIMIT050 ) || (chigh_local < CODER_LIMIT050 ) ) {
if (clow_local >= CODER_LIMIT050 ) {
clow_local &= CODER_LIMIT050 - 1;
chigh_local &= CODER_LIMIT050 - 1;
ccode_local &= CODER_LIMIT050 - 1;
} // if the first wasn't the case, it's chigh < CODER_LIMIT050
clow_local <<= 1;
chigh_local <<= 1;
chigh_local++;
ccode_local <<= 1;
ccode_local |= read_bit();
}
// e3 scaling, to make sure that theres enough space between low and high
while ( (clow_local >= CODER_LIMIT025 ) && (chigh_local < CODER_LIMIT075 ) ) {
clow_local &= CODER_LIMIT025 - 1;
chigh_local ^= CODER_LIMIT025 + CODER_LIMIT050;
// clow -= CODER_LIMIT025;
// chigh -= CODER_LIMIT025;
ccode_local -= CODER_LIMIT025;
clow_local <<= 1;
chigh_local <<= 1;
chigh_local++;
ccode_local <<= 1;
ccode_local |= read_bit();
}
chigh = chigh_local;
clow = clow_local;
ccode = ccode_local;
}

/* -----------------------------------------------
bit reader function
----------------------------------------------- */
unsigned char ArithmeticDecoder::read_bit()
{
// read in new byte if needed
if ( cbit == 0 ) {
if ( !reader_.read_byte(&bbyte)) // read next byte if available
bbyte = 0; // if no more data is left in the stream
cbit = 8;
}
// decrement current bit position
cbit--;
// return bit at cbit position
return BITN( bbyte, cbit );
}


/* -----------------------------------------------
universal statistical model for arithmetic coding
boundaries of this model:
max_s (maximum symbol) -> 1 <= max_s <= 1024 (???)
max_c (maximum context) -> 1 <= max_c <= 1024 (???)
max_o (maximum order) -> -1 <= max_o <= 4
c_lim (maximum count) -> 2 <= c_lim <= 4096 (???)
WARNING: this can be memory intensive, so don't overdo it
max_s == 256; max_c == 256; max_o == 4 would be way too much
----------------------------------------------- */
model_s::model_s( int max_s, int max_c, int max_o, int c_lim ) :
// Copy settings into the model:
max_symbol(max_s),
max_context(max_c),
max_order(max_o + 1),
max_count(c_lim),

current_order(max_o + 1),
sb0_count(max_s),

totals(max_s + 2),
scoreboard(new bool[max_s]),
contexts(max_o + 3)
{
std::fill(scoreboard, scoreboard + max_symbol, false);
// set up null table
table_s* null_table = new table_s;
null_table->counts = std::vector<uint16_t>(max_symbol, uint16_t(1)); // Set all probabilities to 1.

// set up internal counts
null_table->max_count = 1;
null_table->max_symbol = max_symbol;
// set up start table
table_s* start_table = new table_s;
start_table->links = std::vector<table_s*>(max_context);
// integrate tables into contexts
contexts[ 0 ] = null_table;
contexts[ 1 ] = start_table;
// build initial 'normal' tables
for (int i = 2; i <= max_order; i++ ) {
// set up current order table
contexts[i] = new table_s;
// build forward links
if ( i < max_order ) {
contexts[i]->links = std::vector<table_s*>(max_context);
}
contexts[ i - 1 ]->links[ 0 ] = contexts[ i ];
}
}


/* -----------------------------------------------
model class destructor - recursive cleanup of memory is done here
----------------------------------------------- */

model_s::~model_s()
{
// clean up each 'normal' table
delete contexts[1];
// clean up null table
delete contexts[0];
// free everything else
delete[] scoreboard;
}


/* -----------------------------------------------
Updates statistics for a specific symbol / resets to highest order.
Use -1 if you just want to reset without updating statistics.
----------------------------------------------- */
void model_s::update_model( int symbol )
{
// only contexts, that were actually used to encode
// the symbol get its count updated
if ( symbol >= 0 ) {
for (int local_order = ( current_order < 1 ) ? 1 : current_order;
local_order <= max_order; local_order++ ) {
table_s* context = contexts[ local_order ];
auto& count = context->counts[symbol];
// update count for specific symbol & scale
count++;
// store side information for totalize_table
context->max_count = std::max(count, context->max_count);
context->max_symbol = std::max(uint16_t(symbol + 1), context->max_symbol);
// if count for that symbol have gone above the maximum count
// the table has to be resized (scale factor 2)
if (count == max_count) {
context->rescale_table();
}
}
}
// reset scoreboard and current order
current_order = max_order;
std::fill(scoreboard, scoreboard + max_symbol, false);
sb0_count = max_symbol;
}


/* -----------------------------------------------
shift in one context (max no of contexts is max_c)
----------------------------------------------- */
void model_s::shift_context( int c )
{
// shifting is not possible if max_order is below 1
// or context index is negative
if ( ( max_order < 2 ) || ( c < 0 ) ) return;
// shift each orders' context
for (int i = max_order; i > 1; i-- ) {
// this is the new current order context
table_s* context = contexts[ i - 1 ]->links[ c ];
// check if context exists, build if needed
if ( context == nullptr ) {
// reserve memory for next table_s
context = new table_s;
// finished here if this is a max order context
if ( i < max_order ) {
// build links to higher order tables otherwise
context->links.resize(max_context);
}
// put context to its right place
contexts[ i - 1 ]->links[ c ] = context;
}
// switch context
contexts[ i ] = context;
}
}


/* -----------------------------------------------
Flushes the entire model by calling rescale_table on all contexts.
----------------------------------------------- */
void model_s::flush_model()
{
contexts[1]->recursive_flush();
}


/* -----------------------------------------------
Excludes every symbol above c.
----------------------------------------------- */
void model_s::exclude_symbols(int c)
{
// exclusions are back to normal after update_model is used

for ( c = c + 1; c < max_symbol; c++ ) {
if ( !scoreboard[ c ] ) {
scoreboard[ c ] = true;
sb0_count--;
}
}
}


/* -----------------------------------------------
converts an int to a symbol, needed only when encoding
----------------------------------------------- */
int model_s::convert_int_to_symbol( int c, symbol *s )
{
// search the symbol c in the current context table_s,
// return scale, low- and high counts

// totalize table for the current context
table_s* context = contexts[ current_order ];
totalize_table( context );
// finding the scale is easy
s->scale = totals[ 0 ];
// check if that symbol exists in the current table. send escape otherwise
if ( context->counts[ c ] > 0 ) {
// return high and low count for the current symbol
s->low_count = totals[ c + 2 ];
s->high_count = totals[ c + 1 ];
return 0;
}
// return high and low count for the escape symbol
s->low_count = totals[ 1 ];
s->high_count = totals[ 0 ];
current_order--;
return 1;
}


/* -----------------------------------------------
returns the current context scale needed only when decoding
----------------------------------------------- */
void model_s::get_symbol_scale( symbol *s )
{
// getting the scale is easy: totalize the table_s, use accumulated count -> done
totalize_table( contexts[ current_order ] );
s->scale = totals[ 0 ];
}


/* -----------------------------------------------
converts a count to an int, called after get_symbol_scale
----------------------------------------------- */
int model_s::convert_symbol_to_int(uint32_t count, symbol *s)
{
// seek the symbol that matches the count,
// also, set low- and high count for the symbol - it has to be removed from the stream


// go through the totals table, search the symbol that matches the count
int c;
for (c = 1; c < int(totals.size()); c++) {
if (count >= totals[c]) {
break;
}
}
// set up the current symbol
s->low_count = totals[c]; // It is guaranteed that there exists such a symbol.
s->high_count = totals[c - 1]; // This is guaranteed to not go out of bounds since the search started at index 1 of totals.
// send escape if escape symbol encountered
if (c == 1) {
current_order--;
return ESCAPE_SYMBOL;
}
// return symbol value
return c - 2 ; // Since c is not one and is a positive number, this will be nonnegative.
}


/* -----------------------------------------------
totals are calculated by accumulating counts in the current table_s
----------------------------------------------- */

void model_s::totalize_table( table_s* context )
{
// update exclusion is used, so this has to be done each time
// escape probability calculation also takes place here
// accumulated counts must never exceed CODER_MAXSCALE
// as CODER_MAXSCALE is big enough, though, (2^29), this shouldn't happen and is not checked
const auto& counts = context->counts;
// check counts
if (!counts.empty()) { // if counts are already set
// locally store current fill/symbol count
int local_symb = sb0_count;

// set the last symbol of the totals to zero
int i = context->max_symbol - 1;
totals[i + 2] = 0;

// (re)set current total
uint32_t curr_total = 0;

// go reverse though the whole counts table and accumulate counts
// leave space at the beginning of the table for the escape symbol
for (; i >= 0; i--) {
// only count probability if the current symbol is not 'scoreboard - excluded'
if (!scoreboard[i]) {
uint16_t curr_count = counts[i];
if (curr_count > 0) {
// add counts for the current symbol
curr_total += curr_count;
// exclude symbol from scoreboard
scoreboard[i] = true;
sb0_count--;
}
}
totals[i + 1] = curr_total;
}
// here the escape calculation needs to take place
uint32_t esc_prob;
if (local_symb == sb0_count) {
esc_prob = 1;
} else if (sb0_count == 0) {
esc_prob = 0;
} else {
// esc_prob = 1;
esc_prob = sb0_count * ( local_symb - sb0_count );
esc_prob /= ( local_symb * context->max_count );
esc_prob++;
}
// include escape probability in totals table
totals[ 0 ] = totals[ 1 ] + esc_prob;
} else { // if counts are not already set
// setup counts for current table
context->counts.resize(max_symbol);
// set totals table -> only escape probability included
totals[ 0 ] = 1;
totals[ 1 ] = 0;
}
}

/* -----------------------------------------------
special version of model_s for binary coding
boundaries of this model:
... (maximum symbol) -> 2 (0 or 1 )
max_c (maximum context) -> 1 <= max_c <= 1024 (???)
max_o (maximum order) -> -1 <= max_o <= 4
----------------------------------------------- */

model_b::model_b( int max_c, int max_o, int c_lim ) :
// Copy settings into the model:
max_context(max_c),
max_order(max_o + 1),
max_count(c_lim),

contexts(max_o + 3)
{
// set up null table
table* null_table = new table;
null_table->counts = std::vector<uint16_t>(2, uint16_t(1));
null_table->scale = uint32_t(2);
// set up start table
table* start_table = new table;
start_table->links = std::vector<table*>(max_context);
// integrate tables into contexts
contexts[ 0 ] = null_table;
contexts[ 1 ] = start_table;
// build initial 'normal' tables
for (int i = 2; i <= max_order; i++ ) {
// set up current order table
contexts[i] = new table;
// build forward links
if ( i < max_order ) {
contexts[i]->links = std::vector<table*>(max_context);
}
contexts[ i - 1 ]->links[ 0 ] = contexts[ i ];
}
}


/* -----------------------------------------------
model class destructor - recursive cleanup of memory is done here
----------------------------------------------- */
model_b::~model_b()
{
// clean up each 'normal' table
delete contexts[1];
// clean up null table
delete contexts[0];
}


/* -----------------------------------------------
updates statistics for a specific symbol / resets to highest order
----------------------------------------------- */
void model_b::update_model( int symbol )
{
// use -1 if you just want to reset without updating statistics
table* context = contexts[ max_order ];
// only contexts, that were actually used to encode
// the symbol get their counts updated
if ( ( symbol >= 0 ) && ( max_order >= 0 ) ) {
// update count for specific symbol & scale
context->counts[ symbol ]++;
context->scale++;
// if counts for that symbol have gone above the maximum count
// the table has to be resized (scale factor 2)
if ( context->counts[ symbol ] >= max_count )
context->rescale_table();
}
}


/* -----------------------------------------------
shift in one context (max no of contexts is max_c)
----------------------------------------------- */
void model_b::shift_context( int c )
{
// shifting is not possible if max_order is below 1
// or context index is negative
if ( (max_order < 2 ) || ( c < 0 ) ) return;
// shift each orders' context
for (int i = max_order; i > 1; i-- ) {
// this is the new current order context
table* context = contexts[ i - 1 ]->links[ c ];
// check if context exists, build if needed
if ( context == nullptr ) {
// reserve memory for next table
context = new table;
// finished here if this is a max order context
if ( i < max_order) {
// build links to higher order tables otherwise
context->links.resize(max_context);
}
// put context to its right place
contexts[ i - 1 ]->links[ c ] = context;
}
// switch context
contexts[ i ] = context;
}
}


/* -----------------------------------------------
Flushes the entire model by calling rescale_table on all contexts.
----------------------------------------------- */
void model_b::flush_model()
{
contexts[1]->recursive_flush();
}


/* -----------------------------------------------
converts an int to a symbol, needed only when encoding
----------------------------------------------- */
int model_b::convert_int_to_symbol( int c, symbol *s )
{
table* context = contexts[ max_order ];
// check if counts are available
context->check_counts();
// finding the scale is easy
s->scale = context->scale;
// return high and low count for current symbol
if ( c == 0 ) { // if 0 is to be encoded
s->low_count = uint32_t(0);
s->high_count = context->counts[ 0 ];
}
else { // if 1 is to be encoded
s->low_count = context->counts[ 0 ];
s->high_count = context->scale;
}
return 1;
}


/* -----------------------------------------------
returns the current context scale needed only when decoding
----------------------------------------------- */
void model_b::get_symbol_scale( symbol *s )
{
table* context = contexts[ max_order ];
// check if counts are available
context->check_counts();
// getting the scale is easy
s->scale = context->scale;
}


/* -----------------------------------------------
converts a count to an int, called after get_symbol_scale
----------------------------------------------- */
int model_b::convert_symbol_to_int(uint32_t count, symbol *s)
{
table* context = contexts[ max_order ];
auto counts0 = context->counts[ 0 ];
// set up the current symbol
if ( count < counts0 ) {
s->low_count = uint32_t(0);
s->high_count = counts0;
return 0;
}
else {
s->low_count = counts0;
s->high_count = s->scale;
return 1;
}
}
(4-4/16)