float ** 2
It would be a relatively frequent case. It is still slower than `float * float` because `*` has a dedicated VM instruction (opt_mult), though.