Table Test 0

VLC coding for runs will follow a pattern:

 Run Length B(1) (Prefix, Range, Extra):
 257-264,   3- 10, 0
 265-268,  11- 18, 1
 269-272,  19- 34, 2
 273-276,  35- 66, 3
 277-280,  67-130, 4
 281-284, 131-258, 5
 285,     258|3-65538, 0|16

The type of length table used will depend on the block type.

• Fixed Huffman 2 and Dynamic Huffman 2 will use A.
• Fixed Huffman and Dynamic Huffman will used B.
  • The encoding of symbol 285 will depend on the mode:
    • Deflate Mode: 258, 0
    • Deflate64 Mode: 3-65538, 16
    • BTLZA Mode: 3-65538, 16
  • Symbols beyond 285 are not used.

 Extended Runs
 318-321,     1-    4, 0
 322/323,     5-    8, 1
 324/325,     9-   16, 2
 326/327,    17-   32, 3
 328/329,    33-   64, 4
 330/331,    65-  128, 5
 332/333,   129-  256, 6

More Generalized Form of this scheme:

 VLCB (Prefix, Range, Extra):
  0- 7,   0-  7, 0
  8-11,   8- 15, 1
 12-15,  16- 31, 2
 16-19,  32- 63, 3
 20-23,  64-127, 4
 24-27, 128-255, 5
 28-31, 256-511, 6
 -
 32-33,  512-1023, 7
 34-35, 1024-2047, 8
 ...
 
 VLC-VLN (Prefix, Range, Extra):
  0/1,      0-    1, 0 (Zero Run)
  2,        2-    3, 1 (Zero Run)
  3,        4-    7, 2 (Zero Run)
  4,        8-   15, 3 (Zero Run)
  5,       16-   31, 4 (Zero Run)
  6,       32-   63, 5 (Zero Run)
  7,       64-  127, 6 (Value)
  8,      128-  255, 7 (Value)
  9,      256-  511, 8
 10,      512- 1023, 9
 11,     1024- 2047, 10
 12,     2048- 4095, 11
 13,     4096- 8191, 12
 14,     8192-16383, 13
 15,    16384-32767, 14
 -
 16,    32768- 65535, 15
 17,    65536-131071, 16
 18,   131072-262143, 17
 ...

Some values and arguments may have the sign folded into the LSB:

• 0, -1, 1, -2, 2, -3, 3, -4, ...

These cases will be called SVLCA and SVLCB.

Arithmetic Coder

To be filled in more once it stabilizes.

Basically, it uses a bit-at-a-time encoder, with a variable context size.

The context basically consists of the last N preceding bits, which is used as an index into a probabilities table. After encoding each bit, the context is shifted 1 bit and the encoded bit is added to the LSB (or MSB).

This context is masked by the context-size, namely ((1<<n)-1), weights. ="" ="&#10;&#10;&#10;Logically," range="max-min:" bound="(range"></n)-1),>>8)*weight.

• Values above this point (rval > dval) will correspond to a logical 1 bit
  • min=bound+1;
  • weight=weight-(w>>5);
• Values less than or equal to this value will be interpreted as 0
  • max=bound;
  • weight=weight+((256-w)>>5);

Normalization happens when the high 8 bits of min and max become equal.

For encoding:

• Write the high 8 bits to the output;
  • WriteByte(min>>24);
• min=(min<<8); max=(max<<8)|255;

For decoding:

• min=(min<<8); max=(max<<8)|255;
• val=(val<<8)|ReadByte();