[ydl-gen] So, PS3

[Ted Goranson]

David Seikel wrote on 11/18/06:
>  > Since we'd be depending on TerraSoft to build and maintain rpms for
>>  the yum archive, we'd have to go with what they decide is in their
>>  interest to support. And we have no way yet of knowing that, do we?
>
>As with any linux distro, you are free to compile anything from source
>that you want.
>
>>  Programming the cell seems limited in certain respects. A C compiler
>>  (octopiler) is available. And it appears that some Eclipse-based Java
>>  tools are as well. But if you wanted to explore some languages not of
>>  interest to Sony/IBM/terrasoft , like lisp for instance, you'd be on
>>  your own.
>
>A C compiler is all you need to compile interpreters and compilers for
>other languages, even if you have to compile some other language first
>(C++ maybe).  If you are interested in exploring other languages, then
>you would be more than capable of performing the usual three step that
>is required to install things from scratch.  Since Cell technology is
>also aimed at the scientific community for super computer use, many
>obscure languages are likely to be of interest to Sony/IBM/Terrasoft.
>
>I understand that the library used to run code on the Cell co processors
>was designed to be language agnostic.  So I don't see any problems with
>using it from whatever language you choose to use.
>
>When creating a linux distro based on some other distro that has been
>around for a very long time, you would expect all sorts of strange
>languages to creep into the mix.  Emacs is likely to be in there, so
>lisp will likely to be in there.  B-)
>
>Since YDL 5 is based on Fedora Core 5, PPC Fedora Core 5 packages can
>probably install quite easily.  Fedora Core is the community version
>of Red Hat, so PPC Red Hat packages will probably install quite easily.
>
>On the other hand, it would not surprise me if you could install PPC
>RPMS from other sources, and even .debs.  There are rather large repos
>out there of such things.  Lots of them for Fedora Core / Red Hat type
>systems.
>
>Judging by what I have seen on some web sites, it should be possible to
>install other PPC based distros on the PS3, but that is off topic for a
>YDL mailing list.  B-)
>
>People that want to explore obscure programming languages should have a
>minimal amount of technical knowledge, they can head out on their own
>without any hand holding from Terrasoft.  Heading out on your own is not
>any harder than it would be on any other linux distro.  Linux is not a
>closed source, proprietary, locked down to one vendor kind of OS, there
>are very many choices, some of them very easy to implement.

Well. Let's see. As I suggested, you might want a YDL PS3 for three reasons:

1) for the BlueRay

2) as an "ordinary" linux box, meaning that the types of things you 
may choose to do would be the same you'd do on any linux box.

3) as a platform to explore what's unique about the processor.

I think you may be wrong about a few things in what you've said above.

Concerning item 2) I'm under the impression that the YDL PS3 distro 
is a different beast that the MacPPC distro. I'm not sure what the 
differences may be but I expect them to be significant. We have the 
PS3 distro now, or soon. And we are a few months or so out on the 
PPC. So I wouldn't assume that rpms for one would work on the other. 
Maybe you are right. I don't know, but it just doesn't feel right. 
They really are different chips and the Cell wasn't obliged to worry 
about PPC compatibility.

We are dealing with several things here I think. We have a new chip 
architecture. We have a new state management paradigm. We have a new 
sort of multiprocessing notion, quite different than Intel's multiple 
cores. From looking at the literature and speaking to some game PS3 
developers, you have to accommodate that all through your stack, 
starting with how programs are designed, rippling through extended 
language syntax and compiler awareness. And then you'd need special 
dooking done to your compiler or interpreter).

So if you wanted to do 3). And you wanted to do it with, say Ruby, or 
Python, you might just need a tiny bit more than having " a minimal 
amount of technical knowledge," so "they can head out on their own 
without any hand holding from Terrasoft." But suppose you wanted to 
REALLY explore the possibilities. You might want to exploit the 
cell's lazy scheduling paradigm with something like Erlang, which is 
designed for such a thing.

My original post wasn't asking for help in this, just trying to 
discover the basic facts of what we might expect from Terrasoft, 
Sony, IBM and Barcelona. I plan a trip to Barcelona in a couple weeks 
and will report back. I think you are profoundly wrong on what you 
said so far as 3) and possibly wrong on 2).

For instance, even though TS doesn't seem interested in KDE, the way 
KDE is designed, it should be possible to pull the QT burden away 
from the GPU and slough it onto the slave processors using the same 
oct-tet tricks some of the gamers did, and add features like 
anticipated focus. Its not my interest, but there are a world of 
possibilities that can be imagined here.

Even more interesting would be Darwin on Cell. Yeah?

Best, Ted




-- 
__________
Ted Goranson
Sirius-Beta