Computing correctly rounded integer powers in floating-point arithmetic