2018-07-26_TKEX0.txt

Thoughts:
* Sort of similar to PE/COFF and WEX.
* Will use LZ compressed sections.

LZ Scheme:
* Try to be fast while also compressing acceptably.
** Will use byte-oriented data, but some DWORD structures.
** Designed around BJX2 having unaligned access, but high per-access cost
* Will use 32-bit Match Records


Match records will be a 16 or 32 bits, with the low 2 bits as a tag.
* Tag=00: Normal Match
** Bits 31..16: Match Distance (Byte, 1..65536)
** Bits 15..8: Match Length (Byte, 4..259)
** Bits 7..2: Raw Length (Byte, 0..63)

* Tag=10: Short Match (16-bit)
** 15..8: Distance (1..256)
** 7..4: Length (4..17)
** 3..2: Raw (Byte, 0..3)

* Tag=11: PP Match (16-bit)
** Bits 3..2: TagB

** TagB=00: Raw Data (No Match)
*** Bits 15..4: Raw Length (1..4096)
** TagB=01: Predicted Length and Distance
*** Bits 15..4: Raw Length

** TagB=10: Predicted Distance
*** Bits 15..8: Match Length
*** Bits 7..4: Raw Length

** TagB=11: Predicted Length
*** Bits 15..8: Match Distance
*** Bits 7..4: Raw Length