This paper presents a software devised for the purpose of optimal motions computation for robots. Generated motions satisfy a stability constraint and motors limitations while minimizing energy consumption. Motion optimization is solved with the IPOPT optimization package. In order to achieve fast and reliable optimization convergence, calculation of gradients is carried out. Almost rarely considered in the literature, joint friction, which has a non-negligible effect on the optimized motion, is taken into account. Efficiency of the proposed software is demonstrated through the optimization for the 30 degrees of freedom humanoid robot HRP-2 of three dynamic motions: a kicking motion, a throwing motion and a lifting motion. We have applied successfully the obtained optimal motions to the real robot. We show that the motion optimization approach is a powerful tool to generate various types of motions, and to take into account nonlinear limitations.