In generally, techniques are described for efficiently selecting a prediction mode by which to predict predictive video data from reference video data. In particular, an apparatus may include a memory that stores at least a first and second reference coded unit that each includes a first and second reference video data unit, respectively. The apparatus may further comprise a motion compensation unit that performs default weighted prediction to predict a first version of a predictive video data unit from the first and second reference video data units and calculates an offset value for the first version of the predictive video data unit. The motion compensation unit may then perform, based on the calculated offset value, either implicit weighted prediction or explicit weighted prediction to predict a second version of the predictive video data unit and encode the predictive video data unit as either the first or second version.