Am I the only one weirded out that the author didn't even consider writing/benchmarking the obvious byte-at-a-time version (in a loop, and/or unrolled) before r

Am I the only one weirded out that the author didn't even consider writing/benchmarking the obvious byte-at-a-time version (in a loop, and/or unrolled) before resorting to a nonportable/incorrect 'optimized' version? Versions of GCC old enough to drive are pretty good at generating fast code from byte operations and summing up bytes seems like the kind of transparently analyzable case where modern compilers really are sufficiently smart