Difference between revisions of "Compression"
(described exact compression of game data) |
m (added missing detail to the game's algorithm) |
||
Line 80: | Line 80: | ||
##* Call the position at which the match is found '''best_offset''' and the length of the match '''best_length'''. | ##* Call the position at which the match is found '''best_offset''' and the length of the match '''best_length'''. | ||
##* If there are multiple matches of the same length, use the one with the highest offset. | ##* If there are multiple matches of the same length, use the one with the highest offset. | ||
## if '''2 <= best_length <= 5''' and '''best_offset''' is no less than '''inpus_pos-0x100''', then encode the match using mode '''00'''. | ## if '''input_pos + best_length > input_size''', then '''best_length''' = '''input_size - input_pos''' | ||
## Encode the match: | |||
##* if '''2 <= best_length <= 5''' and '''best_offset''' is no less than '''inpus_pos-0x100''', then encode the match using mode '''00'''. | |||
##* else if '''best_length <= 2''', then encode the current input byte as literal and set '''best_length = 1'''. | |||
##* else if '''best_length <= 9''', then encode the match using mode '''01'''. | |||
##* else encode the match using mode '''01L'''. | |||
## '''inpus_pos''' += '''best_length''' | ## '''inpus_pos''' += '''best_length''' | ||
Revision as of 17:35, 6 March 2016
Terranigma uses a compression for most of the data.
Format
Compressed data always starts with a 4 byte header:
Offset | Length | Description |
---|---|---|
0 | 1 | Chunk Counter ??, always 0 (not used in the game) |
1 | 2 | Uncompressed data size (LE) |
3 | 1 | First byte of uncompressed data |
After that follows the compressed data. It consists of data bytes intersparsed with a control code bitstream, both use MSB first (BE) encoding. Data bytes are always aligned at byte boundary. The control code stream is fit in between the data bytes so that a new cc byte is only placed when a bit needs to be used and the previous cc byte is full.
cc | cc | cc | cc | 8 bit | 5 bit | 3 bit | 8 bit | Description |
---|---|---|---|---|---|---|---|---|
1 | - | Literal | - | Literal Byte | ||||
0 | 1 | - | Offset | Length != 0 | - | Copy Length+2 bytes from next_output_pos-0x2000+offset to the output | ||
0 | Length != 0 | Copy Length+1 bytes from next_output_pos-0x2000+offset to the output | ||||||
0 | End of Data (usually in this case Offset is also 0) | |||||||
0 | Length | Offset | - | Copy Length+2 bytes from next_output_pos-0x100+offset to the output |
In the table, each cc stands for a control code bit. The second data byte is divided into 5+3 bits for mode 01.
Compression used for the Game
In general there is more than one way to compress a chunk of data with the given format. But the game's developers used a very specific greedy algorithm to compress all the data in the game. It can be reproduced with the following algorithm:
- Append 0x100 zero bytes to the (uncompressed) input data. These don't count into input_size and are only there for padding!
- Output the header, using 00, input_size, input[0]
- input_pos = 1
- While there is still input to process, repeat the following:
- Find the longest match of data for the current input_pos:
- Look in the range max(0,input_pos-0x2000)..(input_pos-1) for a match of up to 0x100 bytes.
- Call the position at which the match is found best_offset and the length of the match best_length.
- If there are multiple matches of the same length, use the one with the highest offset.
- if input_pos + best_length > input_size, then best_length = input_size - input_pos
- Encode the match:
- if 2 <= best_length <= 5 and best_offset is no less than inpus_pos-0x100, then encode the match using mode 00.
- else if best_length <= 2, then encode the current input byte as literal and set best_length = 1.
- else if best_length <= 9, then encode the match using mode 01.
- else encode the match using mode 01L.
- inpus_pos += best_length
- Find the longest match of data for the current input_pos:
Tool
Crediar re'ed the decompression from the code and translated it to C code.
You can download a tool with the source to decompress data from any given offset here.