arstechnica commentary about the Octopiler

Well I doubt it has anything at all to do with PS3 running late. Since Sony isn't currently using IBM compilers.

And franky I'm highly skeptical of any auto paralellising system. The compiler simply doesn't have enough knowledge to do it except in trivial cases. It's not like no one has tried to do this before. We'll see I guess.
 
I think Hannibal's just kind of doing his own thing here - he's never really 'gotten' the Cell architecture, so I think he's always looking to find reasons for it's existence wherever they may lie. Here he's seemingly fixated on Octopiler, but the truth is Octopiler is a 'shoot the moon' long-term strategy that no one is banking on whatsoever.

Honestly I think he tends to give programmers less credit than they're due, and puts too much emphasis on the 'tools' that make the job easy; in this case Octopiler, in past articles OOE. I mean these are important - and mind I'm not even a programmer - but I think he feels the odds of programmers ever making use of the SPEs power otherwise are like 1000:1, and that's just not the case.

Certainly any PS3 launch depending on Octopiler would be a launch that would occur in time for the Xbox 720 (maybe).
 
As I pointed out in the other thread, it just relies on OpenMP hints at the moment for parallelization (apart from vectorization of course, which is easier).
 
ERP said:
Well I doubt it has anything at all to do with PS3 running late. Since Sony isn't currently using IBM compilers.

And franky I'm highly skeptical of any auto paralellising system. The compiler simply doesn't have enough knowledge to do it except in trivial cases. It's not like no one has tried to do this before. We'll see I guess.

ERP if i understand you correct the OP is in the future more better suited to those "bad" newly examened programmers that dont know shit? :)
 
overclocked said:
ERP if i understand you correct the OP is in the future more better suited to those "bad" newly examened programmers that dont know shit? :)

It'll be in the hands of programmers at some level, languages will evolve, there is a good thread on this in the Sweeney thread in the tech forums.

Frankly I'd be more interested in something that did automatic virtualization of the local store, so we could run more general code on the SPU's, even if it runs slower. I think that early software will just use the SPU's for trivial eyecandy and some of the more obvious parallel tasks, but I really hope we can get beyond that.
 
ERP said:
Frankly I'd be more interested in something that did automatic virtualization of the local store, so we could run more general code on the SPU's, even if it runs slower. I think that early software will just use the SPU's for trivial eyecandy and some of the more obvious parallel tasks, but I really hope we can get beyond that.
Such a solution was briefly discussed towards the end of a small thread in the Console Tech forum (you may or may not have read it): http://www.beyond3d.com/forum/showthread.php?t=28057.
 
Last edited by a moderator:
ERP said:
Well I doubt it has anything at all to do with PS3 running late. Since Sony isn't currently using IBM compilers.

Yeah, I couldn't help but smile ever time he mentioned "Meanwhile, the PS3 is due out in 2006.". As if the timeliness of PS3's release was in any way tied to the emergence of a compiler fulfilling the ambition of Octopiler..me thinks he was fishing for traffic hoping for more on the PS3 delay brew-ha-ha. The headline is plain laughable.
 
Last edited by a moderator:
Titanio said:
Yeah, I couldn't help but smile ever time he mentioned "Meanwhile, the PS3 is due out in 2006.". As if the timeliness of PS3's release was in any way tied to the emergence of a compiler fulfilling the ambition of Octopiler..me thinks he was fishing for traffic hoping for more on the PS3 delay brew-ha-ha. The headline is plain laughable.
That just depends on what the reason for the delay is. If it's hardware, then certainly not. But if it's because Sony (or the studios) don't feel like the launch titles are ready, then it could certainly have an impact. The point Hannibal is trying to make is that managing all the overhead of getting data and code smoothly flowing to and from the SPEs is hard. If they don't have libraries or a compiler that manages that for you, then the extra time that developers have to spend doing that is time that they aren't working on the rest of the engine. I wouldn't call it laughable. And I would never accuse Hannibal of fishing for traffic; he is definitely above that.
 
Hannibal is generally knows what he's talking about if you ask me. I've been reading his whitepapers for years, since I was an undergrad. While he's often not the first to spot trends, he has a clear view of microprocessor design over the last decade, and he's good at connecting the dots. Once in a while I'll be working on something, and I'll randomly remember something he wrote, and realize how on-the-money he was.

But just remember, he's coming from the PC space, so everything he writes more or less reflects that. His skepticism of Cell because of its asymmetric multiprocessing design is well founded, at least in PC/mainframe space - the problem has been well-known and unsuccessfully tackled for at least two decades. He just doesn't realize that console programmers have been doing just that for years.

So cut him some slack - and don't forget the guy wrote some excellent stuff about the PS2, way back when most of us knew nothing about it.
 
Cut him some slack for what? He's a grown man that too smart to keep repeating PS3. Titiano was right when he said the man is fishing for traffic. It's cool or should I say hip to talk about the PS3 delay these days.

They know hundreds of thousands of people talk and search for this information.
 
Sethamin said:
That just depends on what the reason for the delay is. If it's hardware, then certainly not. But if it's because Sony (or the studios) don't feel like the launch titles are ready, then it could certainly have an impact. The point Hannibal is trying to make is that managing all the overhead of getting data and code smoothly flowing to and from the SPEs is hard. If they don't have libraries or a compiler that manages that for you, then the extra time that developers have to spend doing that is time that they aren't working on the rest of the engine. I wouldn't call it laughable. And I would never accuse Hannibal of fishing for traffic; he is definitely above that.

I think specifically relating the availability of a mature version of Octopiler to availability of PS3 is silly, and yes, laughable. If he had more generally discussed software issues and development issues, it might have made more sense. But even then, I'm not sure if that's what would delay PS3.

I don't know, it seems coincidental that there's been a lot of fuss about PS3 delays lately, and Hannibal seems to have taken the most indirect of routes to fold that into a story about Octopiler - it could have easily done without it, PS3 game development has nothing to do with Octopiler, and likely won't for a long time to come (if ever). Call me cynical, but it's just my two cents.
 
Last edited by a moderator:
ERP said:
It'll be in the hands of programmers at some level, languages will evolve, there is a good thread on this in the Sweeney thread in the tech forums.

Frankly I'd be more interested in something that did automatic virtualization of the local store, so we could run more general code on the SPU's, even if it runs slower. I think that early software will just use the SPU's for trivial eyecandy and some of the more obvious parallel tasks, but I really hope we can get beyond that.

Scummed through Sweeneys presentation but aswell as its interesting i think he is overestimating some things, frankly the hype of UE3 has gotten trough his mind imo.

I dont get if i follow you as its not the easiest thing( :) ), but with virtualization of LS do you mean storing in the XDR aswell and/or a combination of both with smart streaming?

Just for satisfaction of my own interest what do you mean by "trivial eyecandy", i mean its maybe trivial to you but the rest of us go "wow Holy Sh#t!" how is that possible. Could you mention some of these trivial cases for ex?
 
Last edited by a moderator:
Titanio said:
I think specifically relating the availability of a mature version of Octopiler to availability of PS3 is silly, and yes, laughable. If he had more generally discussed software issues and development issues, it might have made more sense. But even then, I'm not sure if that's what would delay PS3.
Sure, I agree that the absence of a fully mature version of the Octopiler is certainly not having a direct impact on the PS3 launch. But that's a straw man argument if ever I've heard one. The clear but unspoken implication (to me, leastways) is that the state of this particular compiler IS directly related to the state of general software development tools on the platform. After all, if IBM (or Sony or Toshiba) had already tackled and solved some of those problems alluded to, then they certainly would have been implemented already. The fact that he states that there is a generation's worth of academic work that still needs to be explored here pretty much implies that those problems have not been solved yet in general, much less in existing tools on CELL.

Titanio said:
I don't know, it seems coincidental that there's been a lot of fuss about PS3 delays lately, and Hannibal seems to have taken the most indirect of routes to fold that into a story about Octopiler. Call me cynical, but it's just my two cents.
Sorry, I think Hannibal's reputation is beyond reproach in this regard. He writes the absolute best articles on the Net when it comes to modern microprocessors, and his analysis is always fair, astute, and on point. Even mentioning Ars in the same company as any game rumor mongering sites is grossly unfair, IMO.

I mean, you might as well accuse Dave of fishing for hits on B3D with any of his Xenos or RSX coverage, because that's roughly what it's equivalent to.
 
Sethamin said:
The clear but unspoken implication (to me, leastways) is that the state of this particular compiler IS directly related to the state of general software development tools on the platform. After all, if IBM (or Sony or Toshiba) had already tackled and solved some of those problems alluded to, then they certainly would have been implemented already. The fact that he states that there is a generation's worth of academic work that still needs to be explored here pretty much implies that those problems have not been solved yet in general, much less in existing tools on CELL.

This is all true, but it has nothing to do with when PS3 launches. Or whether PS3 Cell will be well utilised going forward. I don't expect Octopiler will ever be available for PS3 development, but I don't expect that will negatively impact the learning curve that's expected..you won't be able to take x86 code and run it well with little modification on Cell, but then that isn't strictly necessary (however desireable that might be). Even if Octopiler was available, and available for launch title development, while it'd make things friendlier, I'm not sure how well it could make lazy code use the chip anyway, for launch titles.

Sethamin said:
Sorry, I think Hannibal's reputation is beyond reproach in this regard. He writes the absolute best articles on the Net when it comes to modern microprocessors, and his analysis is always fair, astute, and on point. Even mentioning Ars in the same company as any game rumor mongering sites is grossly unfair, IMO.

I mean, you might as well accuse Dave of fishing for hits on B3D with any of his Xenos or RSX coverage, because that's roughly what it's equivalent to.

Nah, this is a bit different. This isn't a technical analysis as such, as in his EE articles or Dave's Xenos articles. It's an article about a subject that probably wouldn't attract as much attention (or traffic) if he hadn't mixed in some "PS3 delay" themes. I don't blame him, I just think it makes the article a shade on the side of funny.
 
Last edited by a moderator:
Sethamin said:
Sure, I agree that the absence of a fully mature version of the Octopiler is certainly not having a direct impact on the PS3 launch. But that's a straw man argument if ever I've heard one. The clear but unspoken implication (to me, leastways) is that the state of this particular compiler IS directly related to the state of general software development tools on the platform. After all, if IBM (or Sony or Toshiba) had already tackled and solved some of those problems alluded to, then they certainly would have been implemented already. The fact that he states that there is a generation's worth of academic work that still needs to be explored here pretty much implies that those problems have not been solved yet in general, much less in existing tools on CELL.

So I guess you missed the threads here at B3D that have PS2 developers given their positive opinions of the PS3 Cell processor right?
 
mckmas8808 said:
So I guess you missed the threads here at B3D that have PS2 developers given their positive opinions of the PS3 Cell processor right?

Actually, I've missed that.

I have read that PS2 developers are very happy with the graphics sub-system in PS3 vs. PS2 (which should be a given, considering the straighforwardness of nVidia graphics.)

I haven't heard many positive opinions on Cell processor itself...and if anything a bit negative about multithreadedness in general.
 
Joe DeFuria said:
I haven't heard many positive opinions on Cell processor itself...and if anything a bit negative about multithreadedness in general.

Ah well they do exist. Devs actually do think positive stuff about the Cell processor. A search wouldn't hurt cha.:p
 
mckmas8808 said:
Ah well they do exist. Devs actually do think positive stuff about the Cell processor. A search wouldn't hurt cha.:p

You made the vague claim, not me. Why and how should I search for something I don't recall happening on this board? I assume you have some reccolection of a specific conversation or two.

Again, what I recall when it comes to "PS2 developers", is that they are much happier with the PS3 graphics sub-sytem relative to PS2. I don't recall hearing much at all about Cell programming relative to PS2 programming, but I do hear lots of grumbling in general about multithreaded programming in general.
 
Back
Top