mirror of
https://github.com/martravi/wiiqt.git
synced 2024-11-15 05:45:11 +01:00
67064cb8bb
* make everything build again
820 lines
18 KiB
C++
820 lines
18 KiB
C++
#include "lz77.h"
|
|
|
|
//most functions here are taken from wii.cs by leathl
|
|
//they have just been tweeked slightly to work with Qt C++
|
|
|
|
LZ77::LZ77()
|
|
{
|
|
match_position = 0;
|
|
match_length = 0;
|
|
textsize = 0;
|
|
codesize = 0;
|
|
}
|
|
|
|
int LZ77::GetLz77Offset( const QByteArray &data )
|
|
{
|
|
QByteArray start = data.left( 5000 );
|
|
|
|
return start.indexOf( "LZ77" );
|
|
}
|
|
|
|
LZ77::CompressionType LZ77::GetCompressedType( const QByteArray &data, int *outOffset )
|
|
{
|
|
if( data.startsWith( 0x10 ) )
|
|
{
|
|
if( outOffset )
|
|
{
|
|
*outOffset = 0;
|
|
}
|
|
return v10;
|
|
}
|
|
if( data.startsWith( 0x11 ) )
|
|
{
|
|
if( outOffset )
|
|
{
|
|
*outOffset = 0;
|
|
}
|
|
return v11;
|
|
}
|
|
int t = GetLz77Offset( data );
|
|
if( t >= 0 )
|
|
{
|
|
if( outOffset )
|
|
{
|
|
*outOffset = t;
|
|
}
|
|
return v10_w_magic;
|
|
}
|
|
return None;
|
|
}
|
|
|
|
QByteArray LZ77::Decompress( const QByteArray &stuff, LZ77::CompressionType *outType )
|
|
{
|
|
int off;
|
|
CompressionType t = GetCompressedType( stuff, &off );
|
|
if( outType )
|
|
{
|
|
*outType = t;
|
|
}
|
|
if( t == v10 )
|
|
{
|
|
return Decompress_v10( stuff, 0 );
|
|
}
|
|
if( t == v10_w_magic )
|
|
{
|
|
return Decompress_v10( stuff, off + 4 );
|
|
}
|
|
if( t == v11 )
|
|
{
|
|
return LZ77_11::Decompress( stuff );
|
|
}
|
|
return stuff;
|
|
}
|
|
|
|
void LZ77::InitTree()
|
|
{
|
|
int i;
|
|
int N = 4096;
|
|
for( i = N + 1; i <= N + 256; i++ )
|
|
rson[ i ] = N;
|
|
|
|
for( i = 0; i < N; i++ )
|
|
dad[ i ] = N;
|
|
}
|
|
|
|
void LZ77::DeleteNode( int p )
|
|
{
|
|
int N = 4096;
|
|
int q;
|
|
|
|
if( dad[ p ] == N )
|
|
return; /* not in tree */
|
|
|
|
if( rson[ p ] == N )
|
|
q = lson[ p ];
|
|
|
|
else if( lson[ p ] == N )
|
|
q = rson[ p ];
|
|
|
|
else
|
|
{
|
|
q = lson[ p ];
|
|
|
|
if( rson[ q ] != N )
|
|
{
|
|
do
|
|
{
|
|
q = rson[ q ];
|
|
}
|
|
while( rson[ q ] != N );
|
|
|
|
rson[ dad[ q ] ] = lson[ q ];
|
|
dad[ lson[ q ] ] = dad[ q ];
|
|
|
|
lson[ q ] = lson[ p ];
|
|
dad[ lson[ p ] ] = q;
|
|
}
|
|
rson[ q ] = rson[ p ];
|
|
dad[ rson[ p ] ] = q;
|
|
}
|
|
dad[ q ] = dad[ p ];
|
|
|
|
if( rson[ dad[ p ] ] == p )
|
|
rson[ dad[ p ] ] = q;
|
|
else
|
|
lson[ dad[ p ] ] = q;
|
|
|
|
dad[ p ] = N;
|
|
}
|
|
|
|
void LZ77::InsertNode( int r )
|
|
{
|
|
int i, p, cmp;
|
|
int N = 4096;
|
|
int F = 18;
|
|
cmp = 1;
|
|
p = N + 1 + ( text_buf[ r ] == ( quint16 )0xffff ? ( quint16 )0 : text_buf[ r ] );
|
|
rson[ r ] = lson[ r ] = N;
|
|
match_length = 0;
|
|
for( ; ; )
|
|
{
|
|
if( cmp >= 0 )
|
|
{
|
|
if( rson[ p ] != N )
|
|
p = rson[ p ];
|
|
else
|
|
{
|
|
rson[ p ] = r;
|
|
dad[ r ] = p;
|
|
return;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if ( lson[ p ] != N )
|
|
p = lson[ p ];
|
|
|
|
else
|
|
{
|
|
lson[ p ] = r;
|
|
dad[ r ] = p;
|
|
return;
|
|
}
|
|
}
|
|
for( i = 1; i < F; i++ )
|
|
if( ( cmp = text_buf[ r + i ] - text_buf[ p + i ] ) != 0 )
|
|
break;
|
|
if( i > match_length )
|
|
{
|
|
match_position = p;
|
|
if( ( match_length = i ) >= F )
|
|
break;
|
|
}
|
|
}
|
|
|
|
dad[ r ] = dad[ p ];
|
|
lson[ r ] = lson[ p ];
|
|
rson[ r ] = rson[ p ];
|
|
dad[ lson[ p ] ] = r;
|
|
dad[ rson[ p ] ] = r;
|
|
|
|
if( rson[ dad[ p ] ] == p )
|
|
rson[ dad[ p ] ] = r;
|
|
|
|
else
|
|
lson[ dad[ p ] ] = r;
|
|
|
|
dad[ p ] = N;
|
|
}
|
|
|
|
quint32 LZ77::GetDecompressedSize( const QByteArray &data )
|
|
{
|
|
int off;
|
|
LZ77::CompressionType ct = LZ77::GetCompressedType( data, &off );
|
|
if( ct == None )
|
|
{
|
|
return data.size();
|
|
}
|
|
if( ct == v10_w_magic )
|
|
{
|
|
off += 4;
|
|
}
|
|
QByteArray ba = data;
|
|
QBuffer buf( &ba );
|
|
if( !buf.open( QBuffer::ReadOnly ) )
|
|
{
|
|
qWarning() << "LZ77::GetDecompressedSize -> Can't create buffer";
|
|
return 0;
|
|
}
|
|
buf.seek( off );
|
|
|
|
quint32 gbaheader;
|
|
buf.seek( off );
|
|
buf.read( (char*)&gbaheader, 4 );
|
|
quint32 ret = ( gbaheader >> 8 );
|
|
if( !ret )
|
|
{
|
|
buf.read( (char*)&ret, 4 );
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
QByteArray LZ77::Decompress_v10( const QByteArray &compressed, int offset )
|
|
{
|
|
int N = 4096;
|
|
int F = 18;
|
|
int threshold = 2;
|
|
int k, r, z, flags;
|
|
quint16 i, j;
|
|
char ch;
|
|
quint32 decomp_size;
|
|
quint32 cur_size = 0;
|
|
|
|
QByteArray crap = compressed;
|
|
QBuffer infile( &crap );
|
|
if( !infile.open( QBuffer::ReadOnly ) )
|
|
{
|
|
qWarning() << "Can't create buffer 1";
|
|
return QByteArray();
|
|
}
|
|
QByteArray ret;
|
|
QBuffer outfile( &ret );
|
|
if( !outfile.open( QBuffer::ReadWrite ) )
|
|
{
|
|
qWarning() << "Can't create buffer 2";
|
|
return QByteArray();
|
|
}
|
|
|
|
quint32 gbaheader;
|
|
infile.seek( offset );
|
|
infile.read( (char*)&gbaheader, 4 );
|
|
|
|
decomp_size = gbaheader >> 8;
|
|
if( !decomp_size )
|
|
{
|
|
infile.read( (char*)&decomp_size, 4 );
|
|
}
|
|
quint8 text_buf[ N + 17 ];
|
|
//qDebug() << "decomp_size:" << decomp_size;
|
|
|
|
for( i = 0; i < N - F; i++ )
|
|
text_buf[ i ] = 0xdf;
|
|
|
|
r = N - F;
|
|
flags = 7;
|
|
z = 7;
|
|
|
|
while( true )
|
|
{
|
|
flags <<= 1;
|
|
z++;
|
|
if( z == 8 )
|
|
{
|
|
if( !infile.getChar( &ch ) )
|
|
break;
|
|
flags = (quint8) ch;
|
|
z = 0;
|
|
}
|
|
if( ( flags & 0x80 ) == 0 )
|
|
{
|
|
if( !infile.getChar( &ch ) )
|
|
break;
|
|
|
|
if( cur_size < decomp_size )
|
|
outfile.putChar( ch );
|
|
|
|
text_buf[ r++ ] = ch;
|
|
r &= ( N - 1 );
|
|
cur_size++;
|
|
}
|
|
else
|
|
{
|
|
if( !infile.getChar( &ch ) )
|
|
break;
|
|
i = (quint8)ch;
|
|
|
|
if( !infile.getChar( &ch ) )
|
|
break;
|
|
j = (quint8)ch;
|
|
|
|
j = j | ( (i << 8) & 0xf00 );
|
|
i = ( ( i >> 4 ) & 0x0f ) + threshold;
|
|
|
|
for( k = 0; k <= i; k++ )
|
|
{
|
|
ch = text_buf[ ( r - j - 1 ) & ( N - 1 ) ];
|
|
if( cur_size < decomp_size )
|
|
outfile.putChar( ch );
|
|
|
|
text_buf[ r++ ] = ch;
|
|
r &= ( N - 1 );
|
|
cur_size++;
|
|
}
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
/*QByteArray LZ77::Decompress( const QByteArray &compressed )
|
|
{
|
|
int off = GetLz77Offset( compressed );
|
|
if( off < 0 )
|
|
{
|
|
qWarning() << "LZ77::Decompress -> data is not compressed";
|
|
return compressed;
|
|
}
|
|
return Decompress( compressed, off );
|
|
}*/
|
|
|
|
QByteArray LZ77::Compress( const QByteArray &ba, LZ77::CompressionType type )
|
|
{
|
|
if( type == v10 )
|
|
{
|
|
return Compress_v10( ba, false );
|
|
}
|
|
if( type == v10_w_magic )
|
|
{
|
|
return Compress_v10( ba, true );
|
|
}
|
|
if( type == v11 )
|
|
{
|
|
return LZ77_11::Compress( ba );
|
|
}
|
|
return ba;
|
|
}
|
|
|
|
QByteArray LZ77::Compress_v10( const QByteArray &ba, bool addMagic )
|
|
{
|
|
LZ77 lz;
|
|
QByteArray ret = lz.Compr_v10( ba, addMagic );
|
|
return ret;
|
|
}
|
|
|
|
QByteArray LZ77::Compr_v10( const QByteArray &ba, bool addMagic )
|
|
{
|
|
int i, len, r, s, last_match_length, code_buf_ptr;
|
|
char ch;
|
|
int code_buf[ 17 ];
|
|
int mask;
|
|
int N = 4096;
|
|
int F = 18;
|
|
int threshold = 2;
|
|
quint32 filesize;
|
|
|
|
QByteArray crap = ba;
|
|
QBuffer infile( &crap );
|
|
if( !infile.open( QBuffer::ReadOnly ) )
|
|
{
|
|
qWarning() << "Can't create buffer 1";
|
|
return QByteArray();
|
|
}
|
|
QByteArray ret;
|
|
QBuffer output( &ret );
|
|
if( !output.open( QBuffer::ReadWrite ) )
|
|
{
|
|
qWarning() << "Can't create buffer 2";
|
|
return QByteArray();
|
|
}
|
|
|
|
if( addMagic )
|
|
{
|
|
output.putChar( 'L' );
|
|
output.putChar( 'Z' );
|
|
output.putChar( '7' );
|
|
output.putChar( '7' );
|
|
}
|
|
|
|
if( ba.size() <= 0xffffff )
|
|
{
|
|
filesize = ( ba.size() << 8 ) | 0x10;
|
|
output.write( (const char*)&filesize, 4 );
|
|
//output.putChar( '\0' );
|
|
//output.putChar( '\0' );
|
|
//output.putChar( '\0' );
|
|
//output.putChar( '\0' );
|
|
}
|
|
else
|
|
{
|
|
filesize = 0x10;
|
|
output.write( (const char*)&filesize, 4 );
|
|
filesize = ba.size();
|
|
output.write( (const char*)&filesize, 4 );
|
|
}
|
|
|
|
InitTree();
|
|
code_buf[ 0 ] = 0;
|
|
code_buf_ptr = 1;
|
|
mask = 0x80;
|
|
s = 0;
|
|
r = N - F;
|
|
|
|
for( i = s; i < r; i++ )
|
|
text_buf[ i ] = 0xffff;
|
|
|
|
for( len = 0; len < F && infile.getChar( &ch ); len++ )
|
|
text_buf[ r + len ] = (quint8)ch;
|
|
|
|
if( ( textsize = len ) == 0)
|
|
return ba;
|
|
|
|
for( i = 1; i <= F; i++ )
|
|
InsertNode( r - i );
|
|
|
|
InsertNode( r );
|
|
|
|
do
|
|
{
|
|
if( match_length > len )
|
|
match_length = len;
|
|
|
|
if( match_length <= threshold )
|
|
{
|
|
match_length = 1;
|
|
code_buf[ code_buf_ptr++ ] = text_buf[ r ];
|
|
}
|
|
else
|
|
{
|
|
code_buf[ 0 ] |= mask;
|
|
|
|
code_buf[ code_buf_ptr++ ] = (quint8)
|
|
( ( ( r - match_position - 1 ) >> 8) & 0x0f ) |
|
|
( ( match_length - ( threshold + 1 ) ) << 4 );
|
|
|
|
code_buf[ code_buf_ptr++ ] = (quint8)( ( r - match_position - 1 ) & 0xff );
|
|
}
|
|
if( ( mask >>= 1 ) == 0 )
|
|
{
|
|
for( i = 0; i < code_buf_ptr; i++ )
|
|
output.putChar( (quint8)code_buf[ i ] );
|
|
|
|
codesize += code_buf_ptr;
|
|
code_buf[ 0 ] = 0; code_buf_ptr = 1;
|
|
mask = 0x80;
|
|
}
|
|
|
|
last_match_length = match_length;
|
|
for( i = 0; i < last_match_length && infile.getChar( &ch ); i++ )
|
|
{
|
|
DeleteNode( s );
|
|
text_buf[ s ] = (quint8)ch;
|
|
if( s < F - 1 )
|
|
text_buf[ s + N ] = (quint8)ch;
|
|
|
|
s = ( s + 1 ) & ( N - 1 );
|
|
r = ( r + 1 ) & ( N - 1 );
|
|
InsertNode( r );
|
|
}
|
|
|
|
while( i++ < last_match_length )
|
|
{
|
|
DeleteNode( s );
|
|
s = ( s + 1 ) & ( N - 1 );
|
|
r = ( r + 1 ) & ( N - 1 );
|
|
if( --len != 0 )
|
|
InsertNode( r );
|
|
}
|
|
}
|
|
while( len > 0 );
|
|
|
|
if( code_buf_ptr > 1 )
|
|
{
|
|
for( i = 0; i < code_buf_ptr; i++ )
|
|
output.putChar( (quint8)code_buf[ i ] );
|
|
|
|
codesize += code_buf_ptr;
|
|
}
|
|
int padding = codesize % 4;
|
|
if( padding != 0 )
|
|
output.write( QByteArray( 4 - padding, '\0' ) );
|
|
|
|
infile.close();
|
|
return ret;
|
|
}
|
|
|
|
LZ77_11::LZ77_11()
|
|
{
|
|
}
|
|
|
|
QByteArray LZ77_11::Compress( const QByteArray &stuff )
|
|
{
|
|
// Test if the file is too large to be compressed
|
|
if( (quint64)stuff.size() > 0xFFFFFFFF )
|
|
{
|
|
qDebug() << "LZ77_11::Compress -> Input file is too large to compress.";
|
|
return QByteArray();
|
|
}
|
|
|
|
quint32 decompressedSize = stuff.size();
|
|
|
|
quint32 SourcePointer = 0x0;
|
|
quint32 DestPointer = 0x4;
|
|
|
|
quint32 tmp;
|
|
QByteArray ret( decompressedSize, '\0' );//will reduce the size later
|
|
QBuffer buf( &ret );
|
|
buf.open( QIODevice::WriteOnly );
|
|
|
|
|
|
// Set up the Lz Compression Dictionary
|
|
LzWindowDictionary LzDictionary;
|
|
LzDictionary.SetWindowSize( 0x1000 );
|
|
LzDictionary.SetMaxMatchAmount( 0xFFFF + 273 );
|
|
|
|
// Figure out where we are going to write the decompressed file size
|
|
if( stuff.size() <= 0xFFFFFF )
|
|
{
|
|
tmp = ( decompressedSize << 8 ) | 0x11; //dont switch endian?
|
|
buf.write( (const char*)&tmp, 4 );
|
|
}
|
|
else
|
|
{
|
|
tmp = 0x11;
|
|
buf.write( (const char*)&tmp, 4 ); //dont switch endian?
|
|
tmp = decompressedSize;
|
|
buf.write( (const char*)&tmp, 4 ); //dont switch endian?
|
|
DestPointer += 0x4;
|
|
}
|
|
|
|
// Start compression
|
|
while( SourcePointer < decompressedSize )
|
|
{
|
|
quint8 Flag = 0x0;
|
|
quint32 FlagPosition = DestPointer;
|
|
// It will be filled in later
|
|
buf.putChar( Flag );
|
|
DestPointer++;
|
|
|
|
for( int i = 7; i >= 0; i-- )
|
|
{
|
|
QList<int>LzSearchMatch = LzDictionary.Search( stuff, SourcePointer, decompressedSize );
|
|
if( LzSearchMatch[ 1 ] > 0 ) // There is a compression match
|
|
{
|
|
Flag |= (quint8)( 1 << i );
|
|
|
|
// Write the distance/length pair
|
|
if( LzSearchMatch[ 1 ] <= 0xF + 1 ) // 2 bytes
|
|
{
|
|
buf.putChar( (quint8)( ( ( ( LzSearchMatch[ 1 ] - 1) & 0xF ) << 4 ) | ( ( ( LzSearchMatch[ 0 ] - 1 ) & 0xFFF ) >> 8 ) ) );
|
|
buf.putChar( (quint8)( ( LzSearchMatch[ 0 ] - 1 ) & 0xFF ) );
|
|
DestPointer += 2;
|
|
}
|
|
else if (LzSearchMatch[1] <= 0xFF + 17) // 3 bytes
|
|
{
|
|
buf.putChar( (quint8)(((LzSearchMatch[1] - 17) & 0xFF) >> 4) );
|
|
buf.putChar( (quint8)((((LzSearchMatch[1] - 17) & 0xF) << 4) | (((LzSearchMatch[0] - 1) & 0xFFF) >> 8)) );
|
|
buf.putChar( (quint8)((LzSearchMatch[0] - 1) & 0xFF) );
|
|
DestPointer += 3;
|
|
}
|
|
else // 4 bytes
|
|
{
|
|
buf.putChar( (quint8)((1 << 4) | (((LzSearchMatch[1] - 273) & 0xFFFF) >> 12)) );
|
|
buf.putChar( (quint8)(((LzSearchMatch[1] - 273) & 0xFFF) >> 4) );
|
|
buf.putChar( (quint8)((((LzSearchMatch[1] - 273) & 0xF) << 4) | (((LzSearchMatch[0] - 1) & 0xFFF) >> 8)) );
|
|
buf.putChar( (quint8)((LzSearchMatch[0] - 1) & 0xFF) );
|
|
DestPointer += 4;
|
|
}
|
|
|
|
LzDictionary.AddEntryRange( stuff, (int)SourcePointer, LzSearchMatch[ 1 ] );
|
|
LzDictionary.SlideWindow( LzSearchMatch[ 1 ] );
|
|
|
|
SourcePointer += (quint32)LzSearchMatch[ 1 ];
|
|
}
|
|
else // There wasn't a match
|
|
{
|
|
Flag |= (quint8)(0 << i);
|
|
|
|
buf.putChar( stuff.at( SourcePointer ) );
|
|
|
|
LzDictionary.AddEntry( stuff, (int)SourcePointer );
|
|
LzDictionary.SlideWindow( 1 );
|
|
|
|
SourcePointer++;
|
|
DestPointer++;
|
|
}
|
|
|
|
// Check for out of bounds
|
|
if( SourcePointer >= decompressedSize )
|
|
break;
|
|
}
|
|
|
|
// Write the flag.
|
|
// Note that the original position gets reset after writing.
|
|
buf.seek( FlagPosition );
|
|
buf.putChar( Flag );
|
|
buf.seek( DestPointer );
|
|
}
|
|
|
|
buf.close();
|
|
ret.resize( DestPointer );
|
|
return ret;
|
|
}
|
|
|
|
QByteArray LZ77_11::Decompress( QByteArray stuff )
|
|
{
|
|
if( !stuff.startsWith( 0x11 ) )
|
|
{
|
|
qWarning() << "LZ77_11::Decompress -> data doesnt start with 0x11";
|
|
return QByteArray();
|
|
}
|
|
|
|
// Compressed & Decompressed Data Information
|
|
QBuffer b( &stuff );
|
|
b.open( QIODevice::ReadOnly );
|
|
|
|
quint32 compressedSize = (quint32)stuff.size();
|
|
quint32 decompressedSize;
|
|
b.read( (char*)&decompressedSize, 4 ); //is this really little endian?
|
|
decompressedSize >>= 8;
|
|
|
|
quint32 sourcePointer = 0x4;
|
|
quint32 destPointer = 0x0;
|
|
unsigned char tempbuffer[ 10 ];
|
|
quint32 backwards_offset;
|
|
|
|
if( !decompressedSize ) // Next 4 bytes are the decompressed size
|
|
{
|
|
b.read( (char*)&decompressedSize, 4 );
|
|
sourcePointer += 0x4;
|
|
}
|
|
|
|
b.close();
|
|
|
|
QByteArray decompressedData( decompressedSize, '\0' );
|
|
if( (quint32)decompressedData.size() != decompressedSize )
|
|
{
|
|
qWarning() << "LZ77_11::Decompress -> failed to allocate" << hex << decompressedSize << "bytes";
|
|
return QByteArray();
|
|
}
|
|
|
|
// Start Decompression
|
|
quint32 num_bytes_to_copy;
|
|
quint32 copy_start_index;
|
|
while( sourcePointer < compressedSize && destPointer < decompressedSize )
|
|
{
|
|
quint8 flag = stuff[ sourcePointer++ ];
|
|
|
|
|
|
for( quint32 i = 0; i < 8; i++ )
|
|
{
|
|
if( flag & ( 0x80 >> i ) )
|
|
{
|
|
// Take encoded data
|
|
tempbuffer[ 0 ] = stuff[ sourcePointer++ ];
|
|
tempbuffer[ 1 ] = stuff[ sourcePointer++ ];
|
|
|
|
switch( tempbuffer[ 0 ] & 0xF0 )
|
|
{
|
|
case 0:
|
|
tempbuffer[ 2 ] = stuff[ sourcePointer++ ];
|
|
num_bytes_to_copy = ( ( tempbuffer[ 0 ] << 4 ) + ( tempbuffer[ 1 ] >> 4 ) + 0x11 );
|
|
backwards_offset = ( ( ( tempbuffer[ 1 ] & 0x0F ) << 8 ) + tempbuffer[ 2 ] + 1 );
|
|
break;
|
|
case 0x10:
|
|
tempbuffer[ 2 ] = stuff[ sourcePointer++ ];
|
|
tempbuffer[ 3 ] = stuff[ sourcePointer++ ];
|
|
num_bytes_to_copy = ( ( tempbuffer[ 0 ] & 0x0F ) << 12 ) + ( tempbuffer[ 1 ] << 4 ) + ( tempbuffer[ 2 ] >> 4 ) + 0x111;
|
|
backwards_offset = ( ( tempbuffer[ 2 ] & 0x0F ) << 8 ) + tempbuffer[ 3 ] + 1;
|
|
break;
|
|
default:
|
|
num_bytes_to_copy = ( tempbuffer[ 0 ] >> 4 ) + 0x01;
|
|
backwards_offset = ( ( tempbuffer[ 0 ] & 0x0F ) << 8 ) + tempbuffer[ 1 ] + 1;
|
|
break;
|
|
}
|
|
copy_start_index = destPointer - backwards_offset;
|
|
for( quint32 copy_counter = 0; copy_counter < num_bytes_to_copy; copy_counter++ )
|
|
{
|
|
if( ( copy_start_index + copy_counter ) >= destPointer )
|
|
{
|
|
qWarning() << "LZ77_11::Decompress -> Error occured while decompressing: The input seems to be telling us to copy uninitialized data.";
|
|
return QByteArray();
|
|
}
|
|
else
|
|
{
|
|
decompressedData[ destPointer++ ] = decompressedData[ copy_start_index + copy_counter ];
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
decompressedData[ destPointer++ ] = stuff[ sourcePointer++ ];
|
|
}
|
|
}
|
|
}
|
|
return decompressedData;
|
|
}
|
|
|
|
LzWindowDictionary::LzWindowDictionary()
|
|
{
|
|
WindowSize = 0x1000;
|
|
WindowStart = 0;
|
|
WindowLength = 0;
|
|
MinMatchAmount = 3;
|
|
MaxMatchAmount = 18;
|
|
BlockSize = 0;
|
|
}
|
|
|
|
QList<int> LzWindowDictionary::Search( const QByteArray &DecompressedData, quint32 offset, quint32 length )
|
|
{
|
|
RemoveOldEntries( DecompressedData[ offset ] ); // Remove old entries for this index
|
|
|
|
if( offset < (quint32)MinMatchAmount || length - offset < (quint32)MinMatchAmount ) // Can't find matches if there isn't enough data
|
|
return QList<int>() << 0 << 0;
|
|
|
|
QList<int>Match = QList<int>() << 0 << 0;
|
|
int MatchStart;
|
|
int MatchSize;
|
|
|
|
for( int i = OffsetList[ (quint8)( DecompressedData[ offset ] ) ].size() - 1; i >= 0; i-- )
|
|
{
|
|
MatchStart = OffsetList[ (quint8)( DecompressedData[ offset ] ) ][ i ];
|
|
MatchSize = 1;
|
|
|
|
while( MatchSize < MaxMatchAmount
|
|
&& MatchSize < WindowLength
|
|
&& (quint32)(MatchStart + MatchSize) < offset
|
|
&& offset + MatchSize < length
|
|
&& DecompressedData[ offset + MatchSize ] == DecompressedData[ MatchStart + MatchSize ] )
|
|
MatchSize++;
|
|
|
|
if( MatchSize >= MinMatchAmount && MatchSize > Match[ 1 ] ) // This is a good match
|
|
{
|
|
Match = QList<int>() << (int)(offset - MatchStart) << MatchSize;
|
|
|
|
if( MatchSize == MaxMatchAmount ) // Don't look for more matches
|
|
break;
|
|
}
|
|
}
|
|
|
|
// Return the match.
|
|
// If no match was made, the distance & length pair will be zero
|
|
return Match;
|
|
}
|
|
|
|
// Slide the window
|
|
void LzWindowDictionary::SlideWindow( int Amount )
|
|
{
|
|
if( WindowLength == WindowSize )
|
|
WindowStart += Amount;
|
|
else
|
|
{
|
|
if( WindowLength + Amount <= WindowSize )
|
|
WindowLength += Amount;
|
|
else
|
|
{
|
|
Amount -= ( WindowSize - WindowLength );
|
|
WindowLength = WindowSize;
|
|
WindowStart += Amount;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Slide the window to the next block
|
|
void LzWindowDictionary::SlideBlock()
|
|
{
|
|
WindowStart += BlockSize;
|
|
}
|
|
|
|
// Remove old entries
|
|
void LzWindowDictionary::RemoveOldEntries( quint8 index )
|
|
{
|
|
for( int i = 0; i < OffsetList[ index ].size(); ) // Don't increment i
|
|
{
|
|
if( OffsetList[ index ][ i ] >= WindowStart )
|
|
break;
|
|
else
|
|
OffsetList[ index ].removeAt( 0 );
|
|
}
|
|
}
|
|
|
|
// Set variables
|
|
void LzWindowDictionary::SetWindowSize( int size )
|
|
{
|
|
WindowSize = size;
|
|
}
|
|
void LzWindowDictionary::SetMinMatchAmount( int amount )
|
|
{
|
|
MinMatchAmount = amount;
|
|
}
|
|
void LzWindowDictionary::SetMaxMatchAmount( int amount )
|
|
{
|
|
MaxMatchAmount = amount;
|
|
}
|
|
void LzWindowDictionary::SetBlockSize( int size )
|
|
{
|
|
BlockSize = size;
|
|
WindowLength = size; // The window will work in blocks now
|
|
}
|
|
|
|
// Add entries
|
|
void LzWindowDictionary::AddEntry( const QByteArray &DecompressedData, int offset )
|
|
{
|
|
OffsetList[ (quint8)( DecompressedData[ offset ] ) ] << offset;
|
|
}
|
|
void LzWindowDictionary::AddEntryRange( const QByteArray &DecompressedData, int offset, int length )
|
|
{
|
|
for( int i = 0; i < length; i++ )
|
|
AddEntry( DecompressedData, offset + i );
|
|
}
|