/// halves method in loop
void prb0 (unsigned *begin, unsigned count)
{
  // usual Duff's device works worse than the decoupled code
  // unrolled loop works better than normal loop
  if (count)
  {
    switch (count & 3)
    {
    case 3: *begin = rbH5 (*begin); ++begin;
    case 2: *begin = rbH5 (*begin); ++begin;
    case 1: *begin = rbH5 (*begin); ++begin;
    default: ;// continue
    }
  
    for (unsigned n = count >> 2, i = 0; n != 0; i += 4, --n)
    {
      begin[i]     = rbH5 (begin[i]);
      begin[i + 1] = rbH5 (begin[i + 1]);
      begin[i + 2] = rbH5 (begin[i + 2]);
      begin[i + 3] = rbH5 (begin[i + 3]);
    }
  }
}