A method for video processing is provided. The method includes determining, for a conversion between a current video block of a video that is a chroma block and a coded representation of the video, parameters of a cross-component linear model that are completely determinable by two chroma samples and corresponding two luma samples; and performing the conversion based on the determining.