Linus has already chimed in, and wants Zstd to just use the standard way of checking for CPU features instead of rolling their own. Apparently, that would be even faster, compiling to a single jump instruction.