C#: Added compression algorithms for the LZ-Overlay format. At the moment, the file will either be fully compressed or not compressed at all. The 'optimal' method also has a higher chance of being sub-optimal due to the nature of overlay files (usually double-compressed files).

This commit is contained in:
barubary 2011-05-14 18:52:46 +00:00
parent b8d11b3e9d
commit 48f189c991

View File

@ -2,6 +2,7 @@
using System.Collections.Generic;
using System.Text;
using System.IO;
using DSDecmp.Utils;
namespace DSDecmp.Formats
{
@ -13,6 +14,17 @@ namespace DSDecmp.Formats
/// </summary>
public class LZOvl : CompressionFormat
{
private static bool lookAhead = false;
/// <summary>
/// Sets the flag that determines if 'look-ahead'/DP should be used when compressing
/// with the LZ-Ovl format. The default is false, which is what is used in the original
/// implementation.
/// </summary>
public static bool LookAhead
{
set { lookAhead = value; }
}
#region Method: Supports(string file)
public override bool Supports(string file)
{
@ -90,6 +102,7 @@ namespace DSDecmp.Formats
}
#endregion
#region Decompression method
public override long Decompress(System.IO.Stream instream, long inLength, System.IO.Stream outstream)
{
#region Format description
@ -274,10 +287,316 @@ namespace DSDecmp.Formats
return decompressedLength + (inLength - headerSize - compressedSize);
}
}
#endregion
public override int Compress(System.IO.Stream instream, long inLength, System.IO.Stream outstream)
{
throw new NotImplementedException();
// don't bother trying to get the optimal not-compressed - compressed ratio for now.
// Either compress fully or don't compress (as the format cannot handle decompressed
// sizes that are smaller than the compressed file).
if (inLength > 0xFFFFFF)
throw new InputTooLargeException();
// read the input and reverse it
byte[] indata = new byte[inLength];
instream.Read(indata, 0, (int)inLength);
Array.Reverse(indata);
MemoryStream inMemStream = new MemoryStream(indata);
MemoryStream outMemStream = new MemoryStream();
int compressedLength = this.CompressNormal(inMemStream, inLength, outMemStream);
int totalCompFileLength = compressedLength + 8;
// make the file 4-byte aligned with padding in the header
if (totalCompFileLength % 4 != 0)
totalCompFileLength += 4 - totalCompFileLength % 4;
if (totalCompFileLength < inLength)
{
byte[] compData = outMemStream.ToArray();
Array.Reverse(compData);
outstream.Write(compData, 0, compressedLength);
int writtenBytes = compressedLength;
// there always seem to be some padding FFs. Let's pad to make the file 4-byte aligned
while ((writtenBytes + 8) % 4 != 0)
{
outstream.WriteByte(0xFF);
writtenBytes++;
}
outstream.WriteByte((byte)((compressedLength) & 0xFF));
outstream.WriteByte((byte)((compressedLength >> 8) & 0xFF));
outstream.WriteByte((byte)((compressedLength >> 16) & 0xFF));
int headerLength = totalCompFileLength - compressedLength;
outstream.WriteByte((byte)headerLength);
int extraSize = (int)inLength - totalCompFileLength;
outstream.WriteByte((byte)((extraSize) & 0xFF));
outstream.WriteByte((byte)((extraSize >> 8) & 0xFF));
outstream.WriteByte((byte)((extraSize >> 16) & 0xFF));
outstream.WriteByte((byte)((extraSize >> 24) & 0xFF));
return totalCompFileLength;
}
else
{
Array.Reverse(indata);
outstream.Write(indata, 0, (int)inLength);
outstream.WriteByte(0); outstream.WriteByte(0); outstream.WriteByte(0); outstream.WriteByte(0);
return (int)inLength + 4;
}
}
#region 'Normal' compression method. Delegates to CompressWithLA when LookAhead is set
/// <summary>
/// Compresses the given input stream with the LZ-Ovl compression, but compresses _forward_
/// instad of backwards.
/// </summary>
/// <param name="instream">The input stream to compress.</param>
/// <param name="inLength">The length of the input stream.</param>
/// <param name="outstream">The stream to write to.</param>
private unsafe int CompressNormal(Stream instream, long inLength, Stream outstream)
{
// make sure the decompressed size fits in 3 bytes.
// There should be room for four bytes, however I'm not 100% sure if that can be used
// in every game, as it may not be a built-in function.
if (inLength > 0xFFFFFF)
throw new InputTooLargeException();
// use the other method if lookahead is enabled
if (lookAhead)
{
return CompressWithLA(instream, inLength, outstream);
}
// save the input data in an array to prevent having to go back and forth in a file
byte[] indata = new byte[inLength];
int numReadBytes = instream.Read(indata, 0, (int)inLength);
if (numReadBytes != inLength)
throw new StreamTooShortException();
int compressedLength = 0;
fixed (byte* instart = &indata[0])
{
// we do need to buffer the output, as the first byte indicates which blocks are compressed.
// this version does not use a look-ahead, so we do not need to buffer more than 8 blocks at a time.
byte[] outbuffer = new byte[8 * 2 + 1];
outbuffer[0] = 0;
int bufferlength = 1, bufferedBlocks = 0;
int readBytes = 0;
while (readBytes < inLength)
{
#region If 8 blocks are bufferd, write them and reset the buffer
// we can only buffer 8 blocks at a time.
if (bufferedBlocks == 8)
{
outstream.Write(outbuffer, 0, bufferlength);
compressedLength += bufferlength;
// reset the buffer
outbuffer[0] = 0;
bufferlength = 1;
bufferedBlocks = 0;
}
#endregion
// determine if we're dealing with a compressed or raw block.
// it is a compressed block when the next 3 or more bytes can be copied from
// somewhere in the set of already compressed bytes.
int disp;
int oldLength = Math.Min(readBytes, 0x1001);
int length = LZUtil.GetOccurrenceLength(instart + readBytes, (int)Math.Min(inLength - readBytes, 0x12),
instart + readBytes - oldLength, oldLength, out disp);
// disp = 1 cannot be stored.
if (disp == 1)
{
length = 1;
}
// disp = 2 cannot be saved properly. use a too large disp instead.
// however since I'm not sure if that's actually how that's handled, don't compress instead.
else if (disp == 2)
{
length = 1;
/*if (readBytes < 0x1001)
disp = readBytes + 1;
else
length = 1;/**/
}
// length not 3 or more? next byte is raw data
if (length < 3)
{
outbuffer[bufferlength++] = *(instart + (readBytes++));
}
else
{
// 3 or more bytes can be copied? next (length) bytes will be compressed into 2 bytes
readBytes += length;
// mark the next block as compressed
outbuffer[0] |= (byte)(1 << (7 - bufferedBlocks));
outbuffer[bufferlength] = (byte)(((length - 3) << 4) & 0xF0);
outbuffer[bufferlength] |= (byte)(((disp - 3) >> 8) & 0x0F);
bufferlength++;
outbuffer[bufferlength] = (byte)((disp - 3) & 0xFF);
bufferlength++;
}
bufferedBlocks++;
}
// copy the remaining blocks to the output
if (bufferedBlocks > 0)
{
outstream.Write(outbuffer, 0, bufferlength);
compressedLength += bufferlength;
/*/ make the compressed file 4-byte aligned.
while ((compressedLength % 4) != 0)
{
outstream.WriteByte(0);
compressedLength++;
}/**/
}
}
return compressedLength;
}
#endregion
#region Dynamic Programming compression method
/// <summary>
/// Variation of the original compression method, making use of Dynamic Programming to 'look ahead'
/// and determine the optimal 'length' values for the compressed blocks. Is not 100% optimal,
/// as the flag-bytes are not taken into account.
/// </summary>
private unsafe int CompressWithLA(Stream instream, long inLength, Stream outstream)
{
// save the input data in an array to prevent having to go back and forth in a file
byte[] indata = new byte[inLength];
int numReadBytes = instream.Read(indata, 0, (int)inLength);
if (numReadBytes != inLength)
throw new StreamTooShortException();
int compressedLength = 0;
fixed (byte* instart = &indata[0])
{
// we do need to buffer the output, as the first byte indicates which blocks are compressed.
// this version does not use a look-ahead, so we do not need to buffer more than 8 blocks at a time.
byte[] outbuffer = new byte[8 * 2 + 1];
outbuffer[0] = 0;
int bufferlength = 1, bufferedBlocks = 0;
int readBytes = 0;
// get the optimal choices for len and disp
int[] lengths, disps;
this.GetOptimalCompressionLengths(instart, indata.Length, out lengths, out disps);
while (readBytes < inLength)
{
// we can only buffer 8 blocks at a time.
if (bufferedBlocks == 8)
{
outstream.Write(outbuffer, 0, bufferlength);
compressedLength += bufferlength;
// reset the buffer
outbuffer[0] = 0;
bufferlength = 1;
bufferedBlocks = 0;
}
if (lengths[readBytes] == 1)
{
outbuffer[bufferlength++] = *(instart + (readBytes++));
}
else
{
// mark the next block as compressed
outbuffer[0] |= (byte)(1 << (7 - bufferedBlocks));
outbuffer[bufferlength] = (byte)(((lengths[readBytes] - 3) << 4) & 0xF0);
outbuffer[bufferlength] |= (byte)(((disps[readBytes] - 3) >> 8) & 0x0F);
bufferlength++;
outbuffer[bufferlength] = (byte)((disps[readBytes] - 3) & 0xFF);
bufferlength++;
readBytes += lengths[readBytes];
}
bufferedBlocks++;
}
// copy the remaining blocks to the output
if (bufferedBlocks > 0)
{
outstream.Write(outbuffer, 0, bufferlength);
compressedLength += bufferlength;
}
}
return compressedLength;
}
#endregion
#region DP compression helper method; GetOptimalCompressionLengths
/// <summary>
/// Gets the optimal compression lengths for each start of a compressed block using Dynamic Programming.
/// This takes O(n^2) time.
/// </summary>
/// <param name="indata">The data to compress.</param>
/// <param name="inLength">The length of the data to compress.</param>
/// <param name="lengths">The optimal 'length' of the compressed blocks. For each byte in the input data,
/// this value is the optimal 'length' value. If it is 1, the block should not be compressed.</param>
/// <param name="disps">The 'disp' values of the compressed blocks. May be less than 3, in which case the
/// corresponding length will never be anything other than 1.</param>
private unsafe void GetOptimalCompressionLengths(byte* indata, int inLength, out int[] lengths, out int[] disps)
{
lengths = new int[inLength];
disps = new int[inLength];
int[] minLengths = new int[inLength];
for (int i = inLength - 1; i >= 0; i--)
{
// first get the compression length when the next byte is not compressed
minLengths[i] = int.MaxValue;
lengths[i] = 1;
if (i + 1 >= inLength)
minLengths[i] = 1;
else
minLengths[i] = 1 + minLengths[i + 1];
// then the optimal compressed length
int oldLength = Math.Min(0x1001, i);
// get the appropriate disp while at it. Takes at most O(n) time if oldLength is considered O(n)
// be sure to bound the input length with 0x12, as that's the maximum length for LZ-Ovl compressed blocks.
int maxLen = LZUtil.GetOccurrenceLength(indata + i, Math.Min(inLength - i, 0x12),
indata + i - oldLength, oldLength, out disps[i]);
if (disps[i] > i)
throw new Exception("disp is too large");
// disp < 3 cannot be stored explicitly.
if (disps[i] < 3)
maxLen = 1;
for (int j = 3; j <= maxLen; j++)
{
int newCompLen;
if (i + j >= inLength)
newCompLen = 2;
else
newCompLen = 2 + minLengths[i + j];
if (newCompLen < minLengths[i])
{
lengths[i] = j;
minLengths[i] = newCompLen;
}
}
}
// we could optimize this further to also optimize it with regard to the flag-bytes, but that would require 8 times
// more space and time (one for each position in the block) for only a potentially tiny increase in compression ratio.
}
#endregion
}
}