pureikyubu/Docs/RE/szp.txt

    Description of 'Yay0', by org <ogamespec@gmail.com>
    Last updated 5 Mar 2004

    All terms are mine, so dont kick me, if something is wrong..

    ---------------------------------------------------------------------------

    First 16 bytes are 'Yay0' header :

    offset size     description
      +0    4       'Yay0' signature
      +4    4       size of decoded data, in bytes
      +8    4       link table offset
      +12   4       non-linked chunks and count modifiers table offset

      +16           chunk description mask (words).
                    each word describes incoming 32 chunks.
                    if bit is set, then next chunk
                    is non-linked (1 byte), otherwise
                    next chunk is copied from previous data

    Example mask : 0b 1101 0010 0000 0010 0000 0000 1101 1001,
        bit 0 = 1, chunk 0 is non-linked, grab next byte from [+12] buffer
        bit 1 = 1, chunk 1 is non-linked, grab next byte from [+12] buffer
        bit 2 = 0, chunk 2 is linked, do block copy, using [+8] table
        and so on...

     [+8]           chunks link table (offsets)

    [+12]           non-linked chunks and count modifiers buffer

    Total size of decoded data buffer calculated as header size (16 bytes)
    and size of decoded data ([+4] in header). Typical values are :

        ANSI font = 65824 bytes
        SJIS font = 593636 bytes

    Average packing ratio is 20-50% (depends on data size)

    Description of 'decoding' :

    I dont know much about other data compression algorithms, so I will not
    be surprised, if Yay0 is just LZ modification :)

    Recipe : the main idea is repeat previously unpacked block of data.

    1. if next bit in mask is '1', copy one byte to destination buffer
       from [+12] table, otherwise do following steps.

    2. read 16-bit halfword from 'chunks link table' area :

        [cccc] [oooo oooo oooo]

       first 4-bits are 'count' modifier. it counts bytes to copy.
       next 12-bits are 'offset' modifier.

    3. adjust modifiers : 

        id('count' == 0)
        {
            'cmod' = grab next byte from 'count modifiers buffer'
            'count' = 'cmod' + 18;
        }
        else 'count' = 'count' + 2;

        'offset' = 'offset' - 1;

        (I assume, that current offset is in destination buffer)

    4. copy 'count' bytes from 'current offset' - 'offset'.

        See also <yay0.png>

    ---------------------------------------------------------------------------

    Reversing of decoding routine :

    void Decode(void *s, void *d)
    {
        i, j, k
        p, q
        cnt

        i = r21 = *(u32 *)(s + 4);      // size of decoded data
        j = r29 = *(u32 *)(s + 8);      // link table
        k = r23 = *(u32 *)(s + 12);     // byte chunks and count modifiers

        q = r31 = 0                     // current offset in dest buffer
        cnt = r28 = 0                   // mask bit counter
        p = r24 = 16                    // current offset in mask table

        do
        {
            // if all bits are done, get next mask
            if(cnt == 0)
            {
                // read word from mask data block
                r22 = *(u32 *)(s + p);
                p += 4;
                cnt = 32;   // bit counter
            }

            // if next bit is set, chunk is non-linked
            if(r22 & 0x80000000)
            {
                // get next byte
                *(u8 *)(d + q) = *(u8 *)(s + k);
                k++, q++;
            }
            // do copy, otherwise
            else
            {
                // read 16-bit from link table
                r26 = *(u16 *)(s + j);
                j += 2;

                // 'offset'
                r25 = q - (r26 & 0xfff);

                // 'count'
                r30 = r26 >> 12;

                if(r30 == 0)
                {
                    // get 'count' modifier
                    r5 = *(u8 *)(s + k);
                    k++;
                    r30 = r5 + 18
                }
                else r30 += 2;

                // do block copy
                r5 = d + r25
                for(i=0; i<r30; i++)
                {
                    *(u8 *)(d + q) = *(u8 *)(r5 - 1);
                    q++, r5++;
                }
            }

            // next bit in mask
            r22 <<= 1;
            cnt--;
        } while(q < i);
    }


    Notes : I found that some files on games DVDs which have "SZP" extension
    are also compressed as "Yay0", so probably Nintendo's term for this
    compression alg. is 'SZP'-compression.