YAZ0 (File Format)

Yaz0 is a run length encoding (RLE compression) method. In Mario Kart Wii most of the SZS files are Yaz0 compressed U8 files.

Data structure

Header

The header of a Yaz0 file is always 16 bytes long. All numeric values stored as big endian values.

Offset	Type	Description
0x00	char[4]	always "Yaz0"
0x04	u32	size of the uncompressed data
0x08	char[8]	always zero (padding)

GNU C example

typedef struct yaz0_header_t
{
    char	magic[4];		// always "Yaz0"
    be32_t	uncompressed_size;	// total size of uncompressed data
    char	padding[8];		// always 0?
}
__attribute__ ((packed)) yaz0_header_t;

Data Groups

The complete compressed data is organized in data groups. Each data group consists of 1 group header byte an 8 chunks:

N	Size	Description
1	1	the group header byte
8	1-3	8 chunks

Each bit of the group header corespondents to one chunk:

• The MSB (most significant bit, 0x80) corespondents to chunk 1
• The LSB (lowest significant bit, 0x01) corespondents to chunk 8

A set bit (=1) in the group header means, that the chunk is 1 exact 1 byte long. This byte must be copied to the output stream 1:1. A cleared bit (=0) defines, that the chunk is 2 or 3 bytes long interpreted as a back reference to already decompressed data that must be copied.

Size	1.B	2.B	3.B	Comment
2	NR	RR	—	N=1..f, SIZE=N+2
3	0R	RR	NN	N=00..ff, SIZE=N+0x12

• RRR is a value between 0x000 and 0xfff. Go back RRR+1 bytes in the output stream to find the start of the data to be copied.
• SIZE is the number of bytes to be copied.
• It is important to know the a chunk may reference to itself. For example if RRR=0 and SIZE=10 the previos byte is copied 10 times.

Decoding data groups and chunks are done until the end of the destination data is reached.

Examples

Decompression

GNU C example

const u8 * src      = // pointer to start of source
const u8 * src_end  = // pointer to end of source (last byte +1)
u8 * dest           = // pointer to start of destination
u8 * dest_end       = // pointer to end of destination (last byte +1)

u8  code            = 0; // code ...
int code_len        = 0; // ... and code_len used to manage groups

while ( src < src_end && dest < dest_end ) 
{ 
    if (!code_len--)
    { 
        code = *src++;
        code_len = 7; 
    } 

    if ( code & 0x80 )
    {
        // copy 1 byte direct
        *dest++ = *src++;
    }
    else 
    { 
        // rle part 

        const u8  b1 = *src++;
        const u8  b2 = *src++;
        const u8 * copy_src = dest - (( b1 & 0x0f ) << 8 | b2 ) - 1;

        int n = b1 >> 4; 
        if (!n) 
            n = *src++ + 0x12; 
        else
            n += 2;
        ASSERT( n >= 3 && n <= 0x111 );

        if ( copy_src < szs->data && dest + n > dest_end )
            return ERROR("Corrupted data!\n");

        // don't use memcpy() or memmove() here because
        // they don't work with self referencing chunks.
        while ( n-- > 0 )
            *dest++ = *copy_src++;
    }
    code <<= 1;
}
ASSERT( src <= src_end );
ASSERT( dest <= dest_end );

This code example is taken from Wiimms SZS Tools: SVN repository lib-szs.c line 159