And just a few shells could be added, to make this optimization right? If the sequence is 1, 4, 10, 23, .. one could start by just adding a pass with 4 and then 1.
https://stackoverflow.com/questions/2539545/fastest-gap-sequ...
I don't know how to select the fallback sort loop size (related to cache line size maybe?); in his github code he uses 16 elements as the threshold.