A device for decoding video data determines that a current block of video data is coded using a linear model prediction mode; for the luma component of the current block, determines reconstructed luma samples; based on luma samples in a luma component of one or more already decoded neighboring blocks and chroma samples in a chroma component of the one or more already decoded neighboring blocks, determines values for linear parameters, wherein the luma samples in the luma component of the one or more already decoded neighboring blocks comprise luma samples from a starting line in the luma component of the one or more already decoded neighboring blocks, wherein the starting line in the luma component of the one or more already decoded neighboring blocks is at least one line removed from a border line of the luma component of the current block.