After you entered a segment of speed code at a certain spot by making use of some method discussed in the Article “Dispatch on a byte” we now want to exit that code at a certain spot.
Some code can be written that way, that we can also determine the runlength by the point where we enter and let it run until its end. But if we have explicit target addresses in our speedcode this just won't work.
;<- enter here sta $0400 sta $0401 sta $0402 sta $0403 sta $0404 sta $0405 sta $0406 sta $0407 sta $0408 ;<- leave here sta $0409 sta $040a sta $040b sta $040c sta $040d sta $040e sta $040f
So given the above example, we might want to enter at sta $0400 and leave after sta $0408, to do so we can modify the code and transform the upcoming sta to an rts command. After exiting the speedcode we then modify the rts back to a sta. Pretty nuch overhead and we would even need to call our speedcode as a subroutine. If we speedocde with predictable runlengths (so no random penalty cycles apply) we could set up a timer irq that will trigger just at the right moment. Thus, no more code modification is needed and the setup is pretty simple.
All you need is setting up a IRQ/NMI handler once beforehand as the exit point of your routine, setup the timer with 2 writes to e.g. $dd04/$dd05 and start a single shot timer run by setting $dd0e to $09. After that, jump to your speedcode segment and wait for the time interrupt to happen at the right spot.
An example could look like this:
;setup lda #$08 sta $dd0e lda #$00 sta $dd04 sta $dd05 lda $dd0d lda #$81 sta $dd0d lda #<exit sta $fffa lda #>exit sta $fffb ... lda runlength_lo,x sta $dd04 ;this can even be ommitted if we do not run more than 255 cycles ;lda runlength_hi,x ;sta $dd05 tsx lda #$09 sta $dd0e ptr jmp speedcode speedcode sta $0400 sta $0401 sta $0402 sta $0403 sta $0404 sta $0405 sta $0406 sta $0407 sta $0408 sta $0409 sta $040a sta $040b sta $040c sta $040d sta $040e sta $040f runlength_lo !byte $04,$08,$0c,$10,$14 ... runlength_hi !byte $00,$00,$00,$00,$00 ... exit txs cli lda $dd0d ...
Regarding the cli instruction: This is important, as else normal irqs might be blocked until we restore the I-flag on the terminating rti, so find out for yourself if it is needed in your case. If so, you can just pull the original PC and flags from stack via pla/plp, instead of manipulating the stack pointer, or simply forgo on the cli if no further irq needs to happen.