I have had some thoughts abou ASMP for the G5.
Is the memory seperate per processor, 2gb+2gb or 4gb+4gb, or can it be made so?
Can Quark be set up to set aside a certain percentage of timeslices to other process' besides Abox?
A could process be set up in Abox to overlay information to a created MUI/Ambient window?

[ Edited by Jim 19.05.2013 - 14:21 ]

Yasu,
Quote:
Only explicitely AMP aware applications can take advantage of the additional cores. All other software can use teh main core only. On SMP the OS schedules all tasks/threads more or less well balanced to all available cores.

Zylesea,
Quote:

And that would be more than fine for us I think. Old legacy programs run happily with single core and more cpu hungry programs are still in development so they could be adapted. I don't think full SMP would give that much more in comparison to AMP in our case. Getting AMP would be big enough step here and probably possible before breaking everything :) Getting mplayer, reggae and similar stuff to use second cpu would be a good start...

jPV,

Quote:

I hereby agree...

Jim,
Quote:
I guess it might be doable, BUT at the time of task creation, there's no way to know which other tasks it needs to share memory with. Most often, that includes system tasks btw. which would restrict most tasks to "system running core".

jPV,
Quote:

I do not think this will happen. With the limited resources, there is no reason to waste any.

Better go for the right target instead of doing it wrong on purpose. Thats the same reason why no one will go the Cygnix way, too. It is a solution that in the end no one wants.

Geit

geit,
Quote:

Doesn't that depend on how hard it is to implement? Without knowing too much about programming, AMP does sound like a relatively easy way to get multicores until MOS changes architecture. I think you are right to say that doing it the right way from the start is the way to go, but we should not forget the existing hardware as well. If we would get multicore G5 working, our hardware would suddenly not be so far behind anymore. As a modern OS, MOS would gain a couple of more years before becoming too old for everyday usage.

+1 for ASMP support.
Even if it is unlikely, it would make a nice feature.

@Geit

Guess that depends on what is seen as "the right way".

The final goal should be a MorphOS running on none-PPC and PPC HW useing as many cores as availbale to run SW compiled against an Amiga-inspired legacy-free API (+ maybe some box to run old 68k stuff in one way or another).

On this route the 1st step could be to transform Quark into something with an SMP-capable API.

Make this API accessable from ABox (sind Quark won't be ready to support full apps including GUI, network etc from the start) and it would indead be seen as AMP from an ABox-perspective.

Still a lot of man-hours required, but assuming there actually is a plan to achieve "the right way" it shouldn't be much of a detour.

Kronos,
Quote:

Thanks Kronos,
As I intend to keep a PPC based system after the ISA change, I would like as many features as possible.

> the 1st step could be to transform Quark into something with an SMP-capable API.

I believe that Quark provides that already.

http://en.wikipedia.org/wiki/Quark_%28kernel%29#Design_goals

> Quark won't be ready to support full apps including GUI, network etc from the start

Will it ever? I mean it's just a kernel.

>> Quark won't be ready to support full apps including GUI, network etc from the start

>Will it ever? I mean it's just a kernel.

Micro kernel at that, so even more services than normal occupy user space.