The GPU part of the code is way too complex to be efficient.
Consider simplifying it (a lot!)
that is the point, the binary transformations involved create unique values(no repeatence, no collisions) and hit areas of the range that random or sequential may never reach,
yes, its not that fast, there is trade-off, but its unique in its own way
so threoretically there are multiple targets (horizontally or vertically or inversion or reversed) - hitting which automatically reveals the main private key, and i believe it could be effective
and because all transformations are in binary format they are quite accurate, hitting those tiny places, that random or sequential may never reach
for a better understanding i made a pc version of the same algorithm on python
https://github.com/puzzleman22/Bitcoin-puzzle-transformations-CPU
yes of course I realize that the range is still huge
I believe that with enough computational power this method can be very effective