SpiceWare's Blog

cd-w · March 11, 2012

That is some very tight code!

You could save a few cycles by arranging all the repositioning kernels to be in the same bank. This would free up 5 cycles, although you would need to set this before the kernel:

lda #<ReposBank
sta EarlyHMjmp+1

For the ones that won't fit in the bank, you can put a placeholder in the repositioning bank to jump to the real kernel:

P1EarlyHM25:
 jmp RealP1EarlyHM25

That would still save you 2 cycles and might just be enough for those missing AUDV0 writes?

Chris

SpiceWare · March 11, 2012

Interesting idea.

Just checked and the P1EarlyHMxx kernels go from $f2dd to $f5f so most of them would end up with a jmp RealP1EarlyHMxx.

The LDA # part must still occur as that's stored in the ColorPlayer1 data stream and it needs to stay in sync with the other datastreams (ie: the 10th byte in the datastream is for the 10th scanline). As such any kernel with the jmp Realxxx would end up taking exactly the same amount of time as it does now. However, only EarlyHM65 and 70 need the extra time and if they're in the "default" page it might buy enough time to add back the audio update...

SpiceWare · March 12, 2012

Thanks, it panned out! New source in blog entry.

Turned out my EarlyHM60 routine was wrong, it had the exact same code as for 55. When I fixed it I ran into the same timing conflict as 65 & 70 had where I had to drop the AUDV0 update before. Luckily enough 3 of the EarlyHM routines ended up on the same page as all the jmp RealEarlyxxx entries, so it worked out.