Ok, I stand corrected. However, at lower optimisation levels they do not inline memcpy calls, which might matter for you. It's never fun if the debug build is 10 times slower in a hot function...