Probably best to stick to assembly for that, since the C version has lots of undefined behavior (division by zero is UB, as is bitwise shift of a negative numbe

Probably best to stick to assembly for that, since the C version has lots of undefined behavior (division by zero is UB, as is bitwise shift of a negative number) an optimizing compiler can do what it want with that code. :) I'm not trying to be annoying, but I think there's value in not publishing broken code without at least noting that it is broken. I realize that the code still communicates to humans how to compute signum(), but it's meaningless if compiled which is kind of important sometimes too.