Devices, systems and methods for unified rounding in sub-block based prediction are described. In a representative aspect, a method of video processing includes generating, for a processing of a current block of video, a pairwise merge candidate based on a pair of motion candidates, and performing, based on the pairwise merge candidate, a conversion between the current block and a bitstream representation of the video. In another representative aspect, a method of video processing includes generating, for a current block of video coded using a geometry partition mode, a uni-prediction motion candidate based on a scaled motion vector and a List0 motion vector, and performing, based on the uni-prediction motion candidate, a conversion between the current block and a bitstream representation of the video.