mos6507's Blog

It might be possible to save two cycles on some queue accesses if you had a "delayed access" function. When enabled, hotspots at $40-$7F would cause the second following cartridge fetch to read from the queue after the previous one.

 

Thus, if it's necessary to fetch data from queues 4-6 and store them to PF0-PF2, one could use code:

 lda CH_QUEUE_FAST+4 ; fetch from queue 4
 sta PF0+$40 ; Triggers queue 5 access
 lda #0   ; Accesses queue 5
 sta PF1+$40 ; Triggers queue 6 access
 lda #0   ; Accesses queue 6
 sta PF2

A total of 4+3+2+3+2+3 = 17 cycles, compared with 21 cycles using queues the "normal" way. If timing allowed the Chimera chip to see the immediate operand fetched from "normal" memory before putting the replacement value on the data bus, that would allow for even more interesting options. Even if there wasn't time to act immediately upon the fetched value, if it could be used to select the queue for the next $40-hotspot access that could be useful.

 

Such a facility would probably make it possible to extend my Ruby Runner 10-character color kernel to 11 characters. There might even be enough time to extend it to 12, though that would likely require using a rather funky HMOVE trick and I don't know if the edges could be made to look 'clean'.

 

Still, I can see how something like Chimera might make something like Robotron almost practical.

Supercat, you should read my next post about queue timing.