I know it might be a contrived case, but it's rather disappointing to see that even with the -O2 optimisation level, the final generated code contains 43% (3/7) useless instructions to mess around with RBP. I mean, it should be bleeding obvious to the compiler that this function doesn't require any stack allocation, much less an RBP-based one (the tradeoff between using RBP/RSP-based allocation is far more subtle, and I can forgive a compiler for choosing nonoptimally in that case), so why did it emit those useless instructions?

There's plenty of articles about how intelligent compiler optimisers can be, and I don't doubt that they can do some pretty advanced transformations, having read a lot of compiler output; and then, I see things like this, which just make you go why!?!? It's puzzling how compilers behave, applying very sophisticated optimisation techniques but then missing the easy cases. The effect might not be as great as more advanced optimisations, but so is the effort required to do this type of optimisation. This is not even low-hanging fruit, it's sitting on the ground.

However, the compiler will use as few registers as possible.

Compilers usually try to use all the registers they can (although sometimes they do exhibit odd non-optimal behaviour, as described above); eschewing register use and using memory instead will increase code size and time.