| Author |
Message |
Maynard Handley
Guest
|
 Posted:
Mon Feb 21, 2005 10:56 pm Post subject:
Re: Apple and Sony Cell chip alliance predicted |
|
|
In article <cvck7d$897$1@io.uvsq.fr>,
TOUATI Sid <touati-nospamplease@nospam-prism.uvsq.fr> wrote:
| Quote: | Isn't it the well studied massive parallel machine paradigm (repainted
with the concept of cellular programming) ? I mean isn't it a massively
parallel architecture scaled to smaller sizes ?
S
|
Except that this is rather harder than standard massive parallelism.
Standard parallel programming does
* not require you to manage segment swapping so as to be able to fit
address+data into a 256KiB address space
* not require you to write in a mixture of C + assembly (so that you can
access the various vector goodness on the SPU). Sure trivial vector
problems can be auto-vectorized, but interesting problems require use of
permute which doesn't obviously translate into the primitives of any
language I know of.
* allow you to ignore messy issues like the cost of branches and loads
through helpers like branch prediction, L1 and OoOE.
You're taking an unsolved hard problem and just making it that much
harder. Imagining that Apple is going to jump into this is delusional.
Apple, after the 68K->PPC transition, while currently having to deal
with two different GPU "languages", and after having made various (never
very satisfactory) dalliances with DSPs in the past, is well aware of
just how important binary compatibility is, and of just how hard it is
to move code from one place to another (when it's the type of code that
Apple deals with --- OS and driver level stuff, video/audio stuff,
basically highly tuned machine-specific stuff). Maybe in three years,
when (IF!!!) it is clear that this concept works, that it's here for the
long term, and that IST have some sort of real answer for how to program
it, then Apple might start to look at it.
Maynard |
|
| Back to top |
|
 |
TOUATI Sid
Guest
|
| Posted:
Tue Feb 22, 2005 2:16 pm Post subject:
Re: Apple and Sony Cell chip alliance predicted |
|
|
| Quote: |
But (a) don't assume that well-studied means well-understood, and
(b) don't assume that all designs are similar. The devil is in
the detail.
|
Yes of course. Many fundamental problem in massively parallel machine
programming are still not solved, even if well studied.
S |
|
| Back to top |
|
|
Sander Vesik
Guest
|
| Posted:
Tue Feb 22, 2005 11:57 pm Post subject:
Re: Apple and Sony Cell chip alliance predicted |
|
|
TOUATI Sid <touati-nospamplease@nospam-prism.uvsq.fr> wrote:
| Quote: |
But (a) don't assume that well-studied means well-understood, and
(b) don't assume that all designs are similar. The devil is in
the detail.
Yes of course. Many fundamental problem in massively parallel machine
programming are still not solved, even if well studied.
|
The majority of the study of massively parallel machine programming
tends to rather be on the level of basics. Its only marginaly better
for the non-massive case.
<comment type="sarcastic">who wants to tackle the hard problems if
you can instead sit around all day rehashing old simple ones?</comment>
--
Sander
+++ Out of cheese error +++ |
|
| Back to top |
|
|
TOUATI Sid
Guest
|
| Posted:
Wed Feb 23, 2005 3:26 pm Post subject:
Re: Apple and Sony Cell chip alliance predicted |
|
|
All depends what do you define as "basics" and what do you call a
"solution".
A simplified problem doesn't mean that it is easy to "solve". Many basic
problems in massively parallel programming are still difficult, others
are till unsolved, even if simplfied. Such problems are inherited by
cellular programming paradigm.
Of course, for every complex problem in cellular programming, you can
find something called a "technical solution". Some people may be
satisfied, and some others not : they would require "real mathematical
solutions" for the basic problems instead of proposing an 999-th
heuristics that allows to produce good plots, as any of the 998-th
previous ones.
<comment type="smile">
All is marketting inside.
</comment>
S
Sander Vesik wrote:
| Quote: | TOUATI Sid <touati-nospamplease@nospam-prism.uvsq.fr> wrote:
But (a) don't assume that well-studied means well-understood, and
(b) don't assume that all designs are similar. The devil is in
the detail.
Yes of course. Many fundamental problem in massively parallel machine
programming are still not solved, even if well studied.
The majority of the study of massively parallel machine programming
tends to rather be on the level of basics. Its only marginaly better
for the non-massive case.
comment type="sarcastic">who wants to tackle the hard problems if
you can instead sit around all day rehashing old simple ones?</comment
S
|
|
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in