An example method includes coding a value of a motion vector difference (MVD) for a current block of video data; obtaining, from a motion vector (MV) buffer, a value of a motion vector predictor (MVP) for the current block of video data; determining a resolution of the value of the MVD for the current block; obtaining a storage resolution shift amount for MVs; rounding, based on the storage resolution shift amount, the value of the MVD obtained from the MV buffer directly to the resolution of the value of the MVD; adding the rounded value of the MVP to the value of the MVD to obtain a value of a MV for the current block; obtaining, based on the MV for the current block, samples of a predictor block for the current block; and reconstructing samples of the current block based on the samples of the predictor block.