That's the surprisingly aesthetic output from the debug version of the latest program:

This is halfway between the look of a cellular automaton and a procedurally-generated maze on an 8-bit console... Every "_" is a datum enqueued by the "slow backwards" thread and every "|" is enqueued by the "fast forward" thread.

And it runs fast !

$ gcc -g -Wall -lpthread -D_REENTRANT -DWIDTH=16 -Os pt_orbit_07.c -o pt_orbit && /usr/bin/time ./pt_orbit
Measuring the first orbit of the Pisano-Carry state space
 for WIDTH = 16 bits, or about 2147516416 iterations.
Starting to scan backwards from Step 0 : X=1, Y=0, C=0 
Starting to scan forward from Step 0 : X=1, Y=0, C=0 

  EUREKA !!!!  Backwards scan reached forward at X=45918 Y=0 C=0
after 822785867 + 1324730547 = 2147516414 steps
 and 12706 + 20329 = 33035 crossings

C'est fini.
3.09user 0.00system 0:01.55elapsed 199%CPU

The "fast-forward" loop alone runs at about 2.4s and running the slower version in parallel saved 0.9s ! Amdahl's law applies of course and it's not possible to get to 1.2s because the slower part will always hold everything else back. But my flexible synchronisation method with 2 FIFOs works well, at least to keep the speed maxed.

However it doesn't work as expected all the time and there are many misses. I'll have to solve that but you can already play at home with pt_orbit_07.c.

Enjoy !