Context: ruby/ruby#17353 ([Feature #22118], String#bit_count)
adds str_count_bits() in string.c, which counts set bits 32 bytes at a time
using a manually unrolled loop of four 8-byte rb_popcount64() calls.
In review, @kou asked about that loop:
Is this faster than 8 bytes loop * 4?
This document benchmarks the current unrolled loop against the plain 8-byte loop to answer that question.