Parameterization of Cross-token Relations with Relative Positional Encoding for Vision MLP