Entered by the JR in SPRITE_ROW, as many 8-byte units from the end as the sprite is wide. Each unit takes the next mask and image pair off the stack and does one buffer byte: cleared under the mask (CPL, OR E, CPL), the image ORed in, stored. The last unit has no INC BC, so a row leaves BC on its last byte, which is what the step patched into SPRITE_SHIFTED_RUN allows for. Knight Lore's sprite_aligned and sprite_aligned_tail.