As the article says, overlapping halves turns out faster than using powers of two in the general case. But for small enough searches the bitwise method can be r

As the article says, overlapping halves turns out faster than using powers of two in the general case. But for small enough searches the bitwise method can be run in vector registers! I talked about this at [0], slides as zipped html at [1] (and I forgot the last click on the multi-byte data slide that shifts the indices for the final result!). The overlap method is probably also possible, but the simpler bitwise operations work out better in SIMD. [0] https://youtu.be/paxIkKBzqBU?t=610 https://youtu.be/paxIkKBzqBU?t=610 [1] https://www.dyalog.com/uploads/conference/dyalog19/presentations/D04_Tacit_Techniques.zip https://www.dyalog.com/uploads/conference/dyalog19/presentat...