The dominator graph algorith completely makes this optimization obsolete.

Instead of the original laf-intel style instrumentation with nested 1
byte compare, we instead instrument each byte separately to provide afl
with hints. We then execute the original cmp/jcc sequence.

Advantages of this method are that we do not have to handle
signed/unsigned comparisons of bytes, nor do we have to deal with having
separate code to handle >=, <=.

Furthermore, we can easily guide AFL towards inducing division by zero,
by doing a compare against 0, i.e., logically add "cmp X, 0" before a "div X" instruction.