wc

>none of the ones I’ve read have posted the test sets up I think the earliest(?) entry[1] in this "series" (the one done in Haskell) had their test input shared alongside the source code, and it has been referenced in some others (often multiplied several times over, as the original isn't very big). Beyond that it's ASCII only, the contents matter little. >nobody seems to want to go to that much effort to get coherent comparisons I think nobody really wants to get any coherent comparisons, because this thing isn't really a competition between the entries themselves. [1]: https://github.com/ChrisPenner/wc (see data/big.txt)

Some timings to show just how quickly Dyalog is improving, taken with the original function from OP and big.txt (https://github.com/ChrisPenner/wc/blob/master/data/big.txt) on a Kaby Lake i7. Dyalog releases about once a year, and 17.1 had almost no performance changes. 15.0: 93ms 16.0: 82ms 17.0: 7.4ms 17.1: 7.5ms 18.0: 5.3ms There's still more room for improvement: although arithmetic functions use AVX2 when available, functions based on SSSE3 shuffle don't yet. That includes Membership (∊) here, which takes the majority of the runtime in my improved version of wc and should be twice as fast with AVX2.