As I've mentioned a couple times, I'm toying with and planning to develop a game or three that are targeted at unexpanded OCS Amigas (specifically, my goal is to have them run playably on an Amiga 500 with 512KB chip RAM and 512KB slow RAM.) Most of what I have planned for them isn't particularily intensive, so this shouldn't be much of a problem.

But...four channels of sound just isn't enough for my compositional style. I could use chord samples, but that increases memory footprint (and loading times) and still carries limitations on the music I write. So I'm wondering about incorporating software mixing to give myself a boost.

Now, I do know that this is CPU-intensive, and I'm told that a 7-channel mixer only just barely runs on a stock Amiga, though I haven't actually tried it myself. I don't think I quite need that much, though - if I were able to mix four channels in software, double them up so that there's a copy on each speaker, and then use the remaining two channels as plain hardware channels (for drums and the like,) that would suit my needs fairly well. I also could do with fixed-length loop samples for the software channels (i.e. simple waveforms rather than full-length recorded audio,) which might make things faster.

My questions are as follows:

• How practical is this? I'm not looking at making a high-speed action game, so I don't need vast amounts of free CPU time, but if it's really as intensive as some folks make it out to be, it might conceivably impact the performance of even a less intensive game...
• Is there a noticeable drop in sound quality? Would oversampling the software channels during mixing help to alleviate it?
• Is there an example mixer for the Amiga whose source I could use as a guide?

Perhaps the source to the OSS replay routine might be useful as a reference?

http://aminet.net/package/dev/src/OcSS_src

It's barely practical if you're planning on doing anything more than a few sprite animations in the background.

Yes. The samples are reduced to 7 bits to speed up the mixing process on OctaMED. Oversampling down to 4 voices will make a huge improvement in sound quality since they will be mixes of 8 bit samples rather than 7-bit.

Is that with seven channels? The game I have planned isn't terribly intensive (it's a console-style RPG, so the most I need is a slowly-scrolling playfield for walkaround bits,) so I'm kind of hoping that cutting it down to four software channels would free up enough CPU time for my purposes...

Scrolling is doable if you use the autoscroll/superbitmap scrolling of the hardware. If you use a hybrid blitter scroller, you'll run out of DMA time. These figures are based on using the 8-voice routines of OctaMED.

-edit-
I just read your first post in entirety and see that you only want 6 voices. It won't buy much time to eliminate the mixing on the first 2 but if you can swing it, try it and see if it works.

I'll play around with it and see what I can find, at least. But it kind of sounds from what you're saying like DMA time is the limiting factor? I kind of assumed it'd be the CPU time.

On an A500 with Chip and slow RAM, all access is DMA time: including the CPU. It's not until you have real Fast RAM that the CPU gets its own memory accesses.

Really? I thought from the diagrams in the hardware reference manual that the CPU got the even memory accesses and the DMA worked on the odd; guess I'll have to go over that again...

I'm not really well versed in this but I think that when you've got enough bitplanes active that some cycles get stolen from the CPU by the chipset.

Yeah, I believe that's in 5-bit or 6-bit modes. IIRC it's one memory cycle per bitplane, out of a total of four, so at worst case that takes half the CPU's memory cycles. I really need to go over the documentation again...