it is not the optimial, for fast native mode, we rely on tcc for fast interpretion