Did Akiko ever get used?

[JLF65]

Jupp3 wrote:
I think both Shapeshifter and Fusion had akiko enabled video drivers. Never tried, probably they didn't help too much.

Yes, they did, and on a CD32 they were a HUGE help. Remember what kind of system you're talking about there.  ;-)

For what it's worth, back before Doom was open sourced, I had been working on some code to play Doom on the Amiga. I had a number of c2p routines that covered everything from using nothing but the CPU, to using AKIKO, to CyberGfx. One of the routines did one pass with the CPU to reorder the data, then used the blitter to do the c2p. If the app could render in columns (like Doom), the CPU pass could be skipped leaving the blitter to do everything. I could break 20 FPS (at 320x200) on a stock A1200 with that.

[JLF65]

stefcep2 wrote:

JLF65 wrote:

Jupp3 wrote:
I think both Shapeshifter and Fusion had akiko enabled video drivers. Never tried, probably they didn't help too much.

Yes, they did, and on a CD32 they were a HUGE help. Remember what kind of system you're talking about there.  ;-)

For what it's worth, back before Doom was open sourced, I had been working on some code to play Doom on the Amiga. I had a number of c2p routines that covered everything from using nothing but the CPU, to using AKIKO, to CyberGfx. One of the routines did one pass with the CPU to reorder the data, then used the blitter to do the c2p. If the app could render in columns (like Doom), the CPU pass could be skipped leaving the blitter to do everything. I could break 20 FPS (at 320x200) on a stock A1200 with that.

And with Akiko?

If you interleaved the writes to chipmem just right, you could go faster. It's like the c2p done with the 030/040 and fastmem - you want enough time between writes to chipmem that you didn't stall waiting on a time slot. The moment you had the first long ready, you wrote it and then continued doing c2p to make time before you wrote the next long.

With AKIKO, too many programmers just looped over the entire screen sending eight longs to AKIKO, then writing back the eight long from it. While that works, they're wasting time that could be spent doing something else. Most of the time in such a loop is spent waiting for free chipmem slots.