kinogaki-bytes

Fast search for one byte value in a byte slice, in safe code: find, rfind and count.

The slice is read eight bytes at a time as a u64 ("SWAR"). For a word w and the needle repeated in every byte, x = w ^ needle has a zero byte exactly where w holds the needle, and !(((x & 0x7f..7f) + 0x7f..7f) | x) & 0x80..80 sets the top bit of exactly those bytes (no carry crosses a byte, unlike the shorter (x - 0x01..01) & !x test). The lowest or highest such bit gives the first or last match; the bit count gives the number of matches. Blocks of four words are tested together before one is looked at.