summaryrefslogtreecommitdiffstats
path: root/fast_encoder.cc
diff options
context:
space:
mode:
Diffstat (limited to 'fast_encoder.cc')
-rw-r--r--fast_encoder.cc316
1 files changed, 84 insertions, 232 deletions
diff --git a/fast_encoder.cc b/fast_encoder.cc
index ba7c34c..319250a 100644
--- a/fast_encoder.cc
+++ b/fast_encoder.cc
@@ -1,5 +1,5 @@
/* Lzip - Data compressor based on the LZMA algorithm
- Copyright (C) 2008, 2009, 2010, 2011 Antonio Diaz Diaz.
+ Copyright (C) 2008, 2009, 2010, 2011, 2012 Antonio Diaz Diaz.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -30,97 +30,6 @@
#include "fast_encoder.h"
-bool Fmatchfinder::read_block()
- {
- if( !at_stream_end && stream_pos < buffer_size )
- {
- const int size = buffer_size - stream_pos;
- const int rd = readblock( infd, buffer + stream_pos, size );
- stream_pos += rd;
- if( rd != size && errno ) throw Error( "Read error" );
- at_stream_end = ( rd < size );
- }
- return pos < stream_pos;
- }
-
-
-Fmatchfinder::Fmatchfinder( const int ifd )
- :
- partial_data_pos( 0 ),
- prev_positions( new int32_t[num_prev_positions] ),
- pos( 0 ),
- cyclic_pos( 0 ),
- key4( 0 ),
- stream_pos( 0 ),
- match_len_limit_( 16 ),
- infd( ifd ),
- at_stream_end( false )
- {
- const int dict_size = 65536;
- const int buffer_size_limit = ( 16 * dict_size ) + before_size + after_size;
- buffer_size = dict_size;
- buffer = (uint8_t *)std::malloc( buffer_size );
- if( !buffer ) { delete[] prev_positions; throw std::bad_alloc(); }
- if( read_block() && !at_stream_end && buffer_size < buffer_size_limit )
- {
- buffer_size = buffer_size_limit;
- uint8_t * const tmp = (uint8_t *)std::realloc( buffer, buffer_size );
- if( !tmp )
- { std::free( buffer ); delete[] prev_positions; throw std::bad_alloc(); }
- buffer = tmp;
- read_block();
- }
- if( at_stream_end && stream_pos < dict_size )
- dictionary_size_ = std::max( (int)min_dictionary_size, stream_pos );
- else dictionary_size_ = dict_size;
- pos_limit = buffer_size;
- if( !at_stream_end ) pos_limit -= after_size;
- prev_pos_chain = new( std::nothrow ) int32_t[dictionary_size_];
- if( !prev_pos_chain )
- { std::free( buffer ); delete[] prev_positions; throw std::bad_alloc(); }
- for( int i = 0; i < num_prev_positions; ++i ) prev_positions[i] = -1;
- }
-
-
-void Fmatchfinder::reset()
- {
- const int size = stream_pos - pos;
- if( size > 0 ) std::memmove( buffer, buffer + pos, size );
- partial_data_pos = 0;
- stream_pos -= pos;
- pos = 0;
- cyclic_pos = 0;
- key4 = 0;
- for( int i = 0; i < num_prev_positions; ++i ) prev_positions[i] = -1;
- read_block();
- }
-
-
-void Fmatchfinder::move_pos()
- {
- if( ++cyclic_pos >= dictionary_size_ ) cyclic_pos = 0;
- if( ++pos >= pos_limit )
- {
- if( pos > stream_pos )
- internal_error( "pos > stream_pos in Fmatchfinder::move_pos" );
- if( !at_stream_end )
- {
- const int offset = pos - dictionary_size_ - before_size;
- const int size = stream_pos - offset;
- std::memmove( buffer, buffer + offset, size );
- partial_data_pos += offset;
- pos -= offset;
- stream_pos -= offset;
- for( int i = 0; i < num_prev_positions; ++i )
- if( prev_positions[i] >= 0 ) prev_positions[i] -= offset;
- for( int i = 0; i < dictionary_size_; ++i )
- if( prev_pos_chain[i] >= 0 ) prev_pos_chain[i] -= offset;
- read_block();
- }
- }
- }
-
-
int Fmatchfinder::longest_match_len( int * const distance ) throw()
{
int len_limit = match_len_limit_;
@@ -130,28 +39,26 @@ int Fmatchfinder::longest_match_len( int * const distance ) throw()
if( len_limit < 4 ) return 0;
}
- const uint8_t * const data = buffer + pos;
- key4 = ( ( key4 << 4 ) ^ data[3] ) & ( num_prev_positions - 1 );
+ key4 = ( ( key4 << 4 ) ^ buffer[pos+3] ) & ( num_prev_pos - 1 );
int newpos = prev_positions[key4];
prev_positions[key4] = pos;
- int32_t * ptr0 = prev_pos_chain + cyclic_pos;
+ int32_t * ptr0 = pos_array + cyclic_pos;
int maxlen = 0;
for( int count = 4; ; )
{
if( newpos < (pos - dictionary_size_ + 1) || newpos < 0 || --count < 0 )
{ *ptr0 = -1; break; }
- const uint8_t * const newdata = buffer + newpos;
int len = 0;
- if( newdata[maxlen] == data[maxlen] )
- while( len < len_limit && newdata[len] == data[len] ) ++len;
+ if( buffer[maxlen+newpos] == buffer[maxlen+pos] )
+ while( len < len_limit && buffer[len+newpos] == buffer[len+pos] ) ++len;
const int delta = pos - newpos;
if( maxlen < len ) { maxlen = len; *distance = delta - 1; }
- int32_t * const newptr = prev_pos_chain +
+ int32_t * const newptr = pos_array +
( cyclic_pos - delta +
( ( cyclic_pos >= delta ) ? 0 : dictionary_size_ ) );
@@ -167,6 +74,8 @@ int Fmatchfinder::longest_match_len( int * const distance ) throw()
break;
}
}
+ if( maxlen == match_len_limit_ && maxlen < max_match_len )
+ maxlen += true_match_len( maxlen, *distance + 1, max_match_len - maxlen );
return maxlen;
}
@@ -180,68 +89,86 @@ void Fmatchfinder::longest_match_len() throw()
if( len_limit < 4 ) return;
}
- const uint8_t * const data = buffer + pos;
- key4 = ( ( key4 << 4 ) ^ data[3] ) & ( num_prev_positions - 1 );
+ key4 = ( ( key4 << 4 ) ^ buffer[pos+3] ) & ( num_prev_pos - 1 );
const int newpos = prev_positions[key4];
prev_positions[key4] = pos;
- int32_t * const ptr0 = prev_pos_chain + cyclic_pos;
+ int32_t * const ptr0 = pos_array + cyclic_pos;
- if( newpos < (pos - dictionary_size_ + 1) || newpos < 0 ) *ptr0 = -1;
+ if( newpos < (pos - dictionary_size_ + 1) || newpos < 0 )
+ *ptr0 = -1;
+ else if( buffer[len_limit-1+newpos] != buffer[len_limit-1+pos] ||
+ std::memcmp( buffer + newpos, buffer + pos, len_limit - 1 ) )
+ *ptr0 = newpos;
else
{
- const uint8_t * const newdata = buffer + newpos;
- if( newdata[len_limit-1] != data[len_limit-1] ||
- std::memcmp( newdata, data, len_limit - 1 ) ) *ptr0 = newpos;
- else
- {
- int idx = cyclic_pos - pos + newpos;
- if( idx < 0 ) idx += dictionary_size_;
- *ptr0 = prev_pos_chain[idx];
- }
+ int idx = cyclic_pos - pos + newpos;
+ if( idx < 0 ) idx += dictionary_size_;
+ *ptr0 = pos_array[idx];
}
}
-// Return value == number of bytes advanced (len).
-// *disp returns the distance to encode.
-// ( *disp == -1 && len == 1 ) means literal.
-int FLZ_encoder::sequence_optimizer( const int reps[num_rep_distances],
- int * const disp, const State & state )
+void FLZ_encoder::sequence_optimizer( int reps[num_rep_distances],
+ State & state )
{
- const int main_len = read_match_distances();
+ int match_distance;
+ const int main_len = fmatchfinder.longest_match_len( &match_distance );
+ const int pos_state = fmatchfinder.data_position() & pos_state_mask;
+ int dis = 0;
+ int len = 0;
- int replen = 0;
- int rep_index = 0;
for( int i = 0; i < num_rep_distances; ++i )
{
- const int len = fmatchfinder.true_match_len( 0, reps[i] + 1, max_match_len );
- if( len > replen ) { replen = len; rep_index = i; }
+ const int tlen =
+ fmatchfinder.true_match_len( 0, reps[i] + 1, max_match_len );
+ if( tlen > len ) { len = tlen; dis = i; }
}
- if( replen > min_match_len && replen + 4 > main_len )
+ if( len > min_match_len && len + 4 > main_len )
{
- *disp = rep_index;
- move_pos( replen, true );
- return replen;
+ crc32.update( crc_, fmatchfinder.ptr_to_current_pos(), len );
+ mtf_reps( dis, reps );
+ range_encoder.encode_bit( bm_match[state()][pos_state], 1 );
+ range_encoder.encode_bit( bm_rep[state()], 1 );
+ const bool bit = ( dis == 0 );
+ range_encoder.encode_bit( bm_rep0[state()], !bit );
+ if( bit )
+ range_encoder.encode_bit( bm_len[state()][pos_state], 1 );
+ else
+ {
+ range_encoder.encode_bit( bm_rep1[state()], dis > 1 );
+ if( dis > 1 )
+ range_encoder.encode_bit( bm_rep2[state()], dis > 2 );
+ }
+ rep_match_len_encoder.encode( range_encoder, len, pos_state );
+ state.set_rep();
+ move_pos( len );
+ return;
}
if( main_len > min_match_len ||
( main_len == min_match_len && match_distance < modeled_distances ) )
{
- *disp = num_rep_distances + match_distance;
- move_pos( main_len, true );
- return main_len;
+ crc32.update( crc_, fmatchfinder.ptr_to_current_pos(), main_len );
+ dis = match_distance;
+ mtf_reps( dis + num_rep_distances, reps );
+ range_encoder.encode_bit( bm_match[state()][pos_state], 1 );
+ range_encoder.encode_bit( bm_rep[state()], 0 );
+ encode_pair( dis, main_len, pos_state );
+ state.set_match();
+ move_pos( main_len );
+ return;
}
+ const uint8_t prev_byte = fmatchfinder[-1];
const uint8_t cur_byte = fmatchfinder[0];
const uint8_t match_byte = fmatchfinder[-reps[0]-1];
+ crc32.update( crc_, cur_byte );
+ fmatchfinder.move_pos();
- *disp = -1;
if( match_byte == cur_byte )
{
- const uint8_t prev_byte = fmatchfinder[-1];
- const int pos_state = fmatchfinder.data_position() & pos_state_mask;
int price = price0( bm_match[state()][pos_state] );
if( state.is_char() )
price += literal_encoder.price_symbol( prev_byte, cur_byte );
@@ -251,44 +178,25 @@ int FLZ_encoder::sequence_optimizer( const int reps[num_rep_distances],
price1( bm_rep[state()] ) +
price0( bm_rep0[state()] ) +
price0( bm_len[state()][pos_state] );
- if( short_rep_price < price ) *disp = 0;
+ if( short_rep_price < price )
+ {
+ range_encoder.encode_bit( bm_match[state()][pos_state], 1 );
+ range_encoder.encode_bit( bm_rep[state()], 1 );
+ range_encoder.encode_bit( bm_rep0[state()], 0 );
+ range_encoder.encode_bit( bm_len[state()][pos_state], 0 );
+ state.set_short_rep();
+ return;
+ }
}
- fmatchfinder.move_pos();
- return 1;
- }
-
-
- // End Of Stream mark => (dis == 0xFFFFFFFFU, len == min_match_len)
-void FLZ_encoder::full_flush( const State & state )
- {
- const int pos_state = fmatchfinder.data_position() & pos_state_mask;
- range_encoder.encode_bit( bm_match[state()][pos_state], 1 );
- range_encoder.encode_bit( bm_rep[state()], 0 );
- encode_pair( 0xFFFFFFFFU, min_match_len, pos_state );
- range_encoder.flush();
- File_trailer trailer;
- trailer.data_crc( crc() );
- trailer.data_size( fmatchfinder.data_position() );
- trailer.member_size( range_encoder.member_position() + File_trailer::size() );
- for( int i = 0; i < File_trailer::size(); ++i )
- range_encoder.put_byte( trailer.data[i] );
- range_encoder.flush_data();
- }
-
-
-FLZ_encoder::FLZ_encoder( Fmatchfinder & mf, const File_header & header,
- const int outfd )
- :
- crc_( 0xFFFFFFFFU ),
- fmatchfinder( mf ),
- range_encoder( outfd ),
- len_encoder( fmatchfinder.match_len_limit() ),
- rep_match_len_encoder( fmatchfinder.match_len_limit() ),
- num_dis_slots( 2 * real_bits( fmatchfinder.dictionary_size() - 1 ) )
- {
- for( int i = 0; i < File_header::size; ++i )
- range_encoder.put_byte( header.data[i] );
+ // literal byte
+ range_encoder.encode_bit( bm_match[state()][pos_state], 0 );
+ if( state.is_char() )
+ literal_encoder.encode( range_encoder, prev_byte, cur_byte );
+ else
+ literal_encoder.encode_matched( range_encoder,
+ prev_byte, cur_byte, match_byte );
+ state.set_char();
}
@@ -304,77 +212,21 @@ bool FLZ_encoder::encode_member( const long long member_size )
range_encoder.member_position() != File_header::size )
return false; // can be called only once
- if( !fmatchfinder.finished() ) // encode first byte
+ if( !fmatchfinder.finished() ) // encode first byte
{
const uint8_t prev_byte = 0;
const uint8_t cur_byte = fmatchfinder[0];
range_encoder.encode_bit( bm_match[state()][0], 0 );
literal_encoder.encode( range_encoder, prev_byte, cur_byte );
crc32.update( crc_, cur_byte );
- move_pos( 1 );
+ fmatchfinder.longest_match_len();
+ fmatchfinder.move_pos();
}
- while( true )
- {
- if( fmatchfinder.finished() ) { full_flush( state ); return true; }
-
- const int pos_state = fmatchfinder.data_position() & pos_state_mask;
- int dis;
- const int len = sequence_optimizer( rep_distances, &dis, state );
- if( len <= 0 ) return false;
+ while( !fmatchfinder.finished() &&
+ range_encoder.member_position() < member_size_limit )
+ sequence_optimizer( rep_distances, state );
- bool bit = ( dis < 0 && len == 1 );
- range_encoder.encode_bit( bm_match[state()][pos_state], !bit );
- if( bit ) // literal byte
- {
- const uint8_t prev_byte = fmatchfinder[-len-1];
- const uint8_t cur_byte = fmatchfinder[-len];
- crc32.update( crc_, cur_byte );
- if( state.is_char() )
- literal_encoder.encode( range_encoder, prev_byte, cur_byte );
- else
- {
- const uint8_t match_byte = fmatchfinder[-len-rep_distances[0]-1];
- literal_encoder.encode_matched( range_encoder,
- prev_byte, cur_byte, match_byte );
- }
- state.set_char();
- }
- else // match or repeated match
- {
- crc32.update( crc_, fmatchfinder.ptr_to_current_pos() - len, len );
- mtf_reps( dis, rep_distances );
- bit = ( dis < num_rep_distances );
- range_encoder.encode_bit( bm_rep[state()], bit );
- if( bit )
- {
- bit = ( dis == 0 );
- range_encoder.encode_bit( bm_rep0[state()], !bit );
- if( bit )
- range_encoder.encode_bit( bm_len[state()][pos_state], len > 1 );
- else
- {
- range_encoder.encode_bit( bm_rep1[state()], dis > 1 );
- if( dis > 1 )
- range_encoder.encode_bit( bm_rep2[state()], dis > 2 );
- }
- if( len == 1 ) state.set_short_rep();
- else
- {
- rep_match_len_encoder.encode( range_encoder, len, pos_state );
- state.set_rep();
- }
- }
- else
- {
- encode_pair( dis - num_rep_distances, len, pos_state );
- state.set_match();
- }
- }
- if( range_encoder.member_position() >= member_size_limit )
- {
- full_flush( state );
- return true;
- }
- }
+ full_flush( fmatchfinder.data_position(), state );
+ return true;
}