| From: | Matt Sealey |
| Date: | 2 Sep 2000 at 15:53:44 |
| Subject: | Re: Amiga PPC UAE |
Hello Alastair
On 02-Sep-00, you wrote:
> Hmmm. Good point - I hadn't considered this. I suppose under a flexible
> enough operating system this could be done with MMU traps?
Possibly..
> In fact, I seem to remember a few years back someone talking on a
> newsgroup about the possibility of using an MMU to emulate AGA on
> gfx-card systems. Nothing ever came of it though.
Probably because there was no real way to program the MMU without
going extremely low level and learning how to work an MMU. Also this
brings all kinds of conflicts with 680x0.library, VMM, Shapeshifter..
.. but now we have mmu.library - so we can do it now and stay
semi-compatible.
But why would you want to emulate AGA on a gfx card system? I can only
see it of being any benefit for ECS users to get AGA stuff to work. But
nearly all AGA stuff is either a game - and therefore isn't going to sit
for emulation - or 'knows' cybergraphics.library anyway.
>> It's the checking that slows things down. The clever part of the
>> Transmeta technology is that they found a way to minimise these delays,
>> as explained in the Transmeta patents. This is not done in Amiga
>> emulators.
>
> Hmmm. I must read those patents sometime.
It's just clever pipelining and having a cache on-chip for holding
pre-translated code for certain addresses. When you consider that
the x86 'emulation' is getting it's code from a high-speed cache
which loads into another higher-speed-cache to be translated and
then STORED, you can surpass x86 speeds at the same clock (in tight
loops) just by having a better microcode kernel... and that isn't
hard to do ;)
>> Whether it could be, I don't know.
>
> Without infringing the patents? ;^p
I don't think hardware patents apply to software methodology that much.
Otherwise basically Transmeta have patented buffering :)
>> Of course if the 68000 had separate addresses for hardware ports, like
>> the Z80, there would be no problem. Puttting hardware in the same address
>> range as RAM was a short-sighted decision. The Z80 has a specific set of
>> commands for hardware access, so they can be easily converted in an
>> emulation.
>
> I somehow doubt that ease of emulation was the primary motivation in the
> design ;-) Seriously though, x86 has seperate IO space and memory space
> too, doesn't it?
Yes.
Using the standard main memory map reduces the complexity of the chips
though.
>> Why has nobody ever written an Atari ST emulation for the Amiga? Same
>> problem, I think.
>
> There are at least two - but the one I've got only works on A500 and needs
> two disk-drives. I think the other one is on Aminet, but needs the ROM to
> be grabbed from a real ST.
They work quite well, too :)
Thanks
Quote carefully and read all ADMIN:README mails