A great old-timey game-programming hack (2013)

A programmer has nearly completed a side-scrolling game for an 8-bit PC, but its software-rendered graphics run too slowly. The game must copy background tiles pixel data into a 320-by-240 screen buffer, and the 6809 processor has no graphics hardware to do the work. A straightforward routine copies each tile row with a loop; unrolling that loop removes its overhead, and replacing auto-increment addressing with fixed offsets cuts the estimated time per row from 157 to 98 cycles. The resulting tile-copy routine is about 60 percent faster, yet still cannot make the game playable.

After initially finding no further improvement, the narrator consults a 6809 manual and notices that its stack instructions can move multiple registers at once. The proposed hack repurposes the stack pointers: point one at the tile and the other at the screen buffer, then use PULS and PSHU to transfer the D, X, and Y registers as six-byte chunks. Since a tile row contains 14 bytes, the plan uses two six-byte transfers and a final two-byte transfer, with the loops unrolled. This approach exploits an unusual processor feature to treat stack operations as fast bulk-memory copying, rather than merely as a way to save and restore registers. The excerpt ends as the narrator sketches the pointer adjustments for the routine.