Comment by Jacobi

I think that it can be implemented in constant time for floating point numbers. For integers I know that the exponentiation by squaring takes O(log(n))