A video coder determines, based at least in part on a distance between a current sub-block of a transform coefficient block and a sub-block that contains a last significant coefficient (LSC) of the transform coefficient block, coding contexts for significance flags for transform coefficients of the current sub-block. Furthermore, the video coder entropy codes, based on the coding contexts for the significance flags for the transform coefficients of the current sub-block, the significance flags for the transform coefficients of the current sub-block.