Remove hardwired alignment, align to sizeof(long) which should be more
portable (though the hardwired alignment of 16 would also have worked just
fine and perhaps even resulted in better L1 cache operation).