Does AOS4.0 use any Altivec specific routines?

In C? In assembler?

G3s don't have Altivec, so maybe there aren't any Altivec specific components?

Have any of the graphics.library, or P96 routines been able to take advantage of Altivec?

Considering it's vector based (I think that's what Altivec is all about), it could really help Flash work wonders (and maybe Java?).

G5s continue using Altivec? (I assume.)


AmigaOne! AOS4.0! More potential? We've got it in spades!!!!!!

The stated position has always been that OS4.0 components would not include Altivec instructions. There is no need since the OS would not get a significant advantage from Altivec.

Those of us ( ) who cajoled Eyetech into releasing G4 processors on the AOne did so in the hope that third parties would eventually release software that made good use of the additional instructions if there were sufficient G4 systems in use. Both Eyetech and Hyperion have told us that this would be unlikely in the short term.

IIRC the first hurdle is to find a version of the GCC (is that right?) compiler that can optimise code to use Altivec: hand coding in assembler will not bring any advantages unless carried out by an Altivec specialist.

From what I have read, the real advantages would come from using Altivec in video encoders and decoders and, possibly, for 2D and 3D rendering.

Regards
Roy Leith

If you look at the OS4 feature set at amiga.com, that mentions an "altivec.resource", that is supposed to be the OS-legal way of accessing the altivec unit on the CPU.

@anyone at Hyperion: is that feature set still valid, or are there things that need to be added/removed?

IIRC thats scheduled for 4.2 or thereabouts, so expect it around 2015 ;)

/me trundles off under his bridge.

_________________
[quote]Amiga were also offered Amithlon before anyone else. I was the first to run it. It ROCKED HARD. I begged them to use it, we had a WINNER and could sell a bajillion of them. We owned all the rights to it! But sadly, Bill and Fleecy didn't want peopl

In AOS4.0 there's no altivec specific instructions. So, while the G4 800Mhz is only slightly faster then the G3 800Mhz I'm doubtful you'll see any real improvements.

I believe altivec upgrades are planned in future version of 4 such .2 or .3.

Today with the G5 3rd party companies, such as Adobe, have added their own 64bit optimization schemes plug-ins. Hopefully, something similar could be available on the Amiga but with Altivec instructions instead. One thing I'd want is an Altivec plug-in for ImageFX. Then at some point in the future Amiga will pick up altivec usage into it's OS.

Can a regular instruction AND an Altivec instruction be executed simultaneously, like hyper-threading on P IVs?

Like, in a 68060, where the CPU and FPU were on the same chip, were they able to do a CPU AND a FPU computation at the same time?

I'm just trying to understand what the benefit of Altivec could be. I'm sure it wasn't easy to design and jam onto the crowded die, let's see what it can DO for us! Would you want yours to just be sitting there rusting????


AmigaOne! AOS4.0! Pull out all the stops!

Altivec can do a 4 way parallelism of 32bit integer and Floating Point calcuation. In practice this is seen as a 3.5-3.9x improvement of Megaflop performance.

Thus, if you have a G3 800Mhz against a G4 800Mhz and you run a ImageFX or Aladdin 4D or some other heavily floating point dependant application you're going to see roughly a 10% improvement. If the OS had altivec optimizations and ImageFX used them you'd see a 300% improvement. Now we're talking.

At Softwarehut in the US the G3 is priced about $100 lower then the G4. Which translates into roughly a 12% increase for the G4 over the G3. W/o Altivec you're paying 12% more for a 10% return not great. But with Altivech if I see a 300% improvement on some applications with a 12% more money outlay now you're talking a more attractive system.

As OSes are mainly integer computations and while the # of floating point calcuations are small in comparison to the # of interger computations you won't see much improvement in AOS4. (Of course TCP/IP does use floating point so conceivably you might be able to get some performace in communications.) But, where you'd see the performance is in those graphically intensive image and video manipulation software. You know one of the reasons the Amiga was so great from day 1.

Fingers crossed that those gfx apps will support Altivec too.

_________________

José

I suspect that what we'll see is applications having the "plain" version and the altivec version, a bit like we have apps with '020/'030/fpu optimized binaries in aminet.

I know there are special provisions in AmigaDE for MMX on the Intels so the Altivec problem will be resolved by the time we get to AOS 5. As for AOS 4 we'll just have to see what happens.

Quote:

@anyone at Hyperion: is that feature set still valid, or are there things that need to be added/removed?

There is only one modification required to use Altivec, and that is support in the scheduler. While I don't think that the upcoming dev release will feature this modification, chances are good that the final release will.

With that in mind, adding Altvec support in the system proper is going to be quite easy.

So the answer is, we're still planning to have Altivec support in the final OS 4.0 product.

_________________
Seriously, if you want to contact me do not bother sending me a PM here. Write me a mail

Quote:

Samurai_Crow wrote: I know there are special provisions in AmigaDE for MMX on the Intels so the Altivec problem will be resolved by the time we get to AOS 5. As for AOS 4 we'll just have to see what happens.

Is AmigaOS5 still supposed to be hosted on AmigaDE? I thought that was shelved, because of lack of interest on Tao's part to improve Intent to the point where it actually could host an OS. (ie: Mem protection, communication outside of box, etc)

Quote:

Is AmigaOS5 still supposed to be hosted on AmigaDE? I thought that was shelved, because of lack of interest on Tao's part to improve Intent to the point where it actually could host an OS. (ie: Mem protection, communication outside of box, etc)

This is news to me. Can anyone confirm?

Quote:

Rogue wrote: With that in mind, adding Altvec support in the system proper is going to be quite easy. So the answer is, we're still planning to have Altivec support in the final OS 4.0 product.

FAN - TAS - TIC !!!

We need every edge we can put our grubby little fingers on!


AOS4.0! The base is stable!

@Rogue:

Quote:

There is only one modification required to use Altivec, and that is support in the scheduler

I see, that would be saving and restoring of the current task's altivec context. I'm not that familiar with PPC CPUs, how much is that likely to add to the context switching overhead? On the x86 (for the floating point unit) there's a trick one can use to pretty much eliminate the overhead for tasks that don't use it (see eg the variable "last_task_used_math" on Linux), is there something similar on the PPC?