I've used popcnt in prod in last 10 years. We were looking for similar images in large sets, using a perceptual hash (64 bits, at least 60 bits being same consi

I've used popcnt in prod in last 10 years. We were looking for similar images in large sets, using a perceptual hash (64 bits, at least 60 bits being same considered a match). Tried a few different solutions. Fancy vector databases. Clustering algos to reduce search space. A really neat custom O(n) algo that searched the set of possible solutions instead of the intersection of n^2 possible matches. The old fashioned N^2 solution ended up being fastest, simplest, and could handle quite large Ns, when the core of the loop was optimized with popcnt (well xor -> popcnt). Computers are fast.