The rounding was wrong due to incorrect ue of floats,
changed to rationals and av_rescale.

The results were not properly passed to SDL.