A device for decoding 360-degree video data is configured to store a decoded picture of 360-degree video as a reference frame; derive an extended reference frame from the stored reference frame based on a padding amount by extending a first cube face in the reference frame; inter-predict a block of a current picture from a block of the extended reference frame by determining a motion vector for the block of the current picture; in response to a determination that the motion vector points to a cube face in the extended reference frame other than the first cube face, clipping the motion vector such that the motion vector points to a location in the first cube face; and locating a prediction block for a current block using the clipped motion vector.