arstechnica commentary about the Octopiler

[ban25]

Based on what I've read, at least some (most?) developers are using GCC instead of IBM's XLC.

Tangentially, how many different compilers for the CELL architecture do you think there are? I would imagine that IBM wrote the compiler that most devs are using for the PS3, considering that IBM has a lot of expertise with this (certainly a lot more than Sony or Toshiba). The Octopiler is probably built off the same codebase as the PS3 compiler, or at least is a branch off that project that shares a lot of code with it. I would be very surprised if this were not the case.

 

[Carl B]

@Bobbler: Yeah, the situation is definitely 'shades of grey.' The intra-developer generations are a good point, but then we have to also consider that if what we're really getting at with the generation monikers is a sense of technical achievement relative to prior 'generations,' sometimes a dev with a first-time effort can come in and do quite well.

But it's a good point that the generational breakdowns for these games seem a little arbitrary in the end no matter how you slice it.

 

[Joe DeFuria]

OK, well we'll need to define our generations then. When I said "second generation", I was talking about late 2006 releases. If we define it as 2007, then obviously that's something we can't really comment on yet.

Leaving aside ambiguous "generation" classifications, I guess I mean that I expect that of late 2006 PS3 titles compared to late 2006 X360 titles.

I think late 2006 PS3 titles will be comparable to late 2006 X360 titles. About the same more or less.

Here's my generational differences:

I consider 1st gen titles to be all titles that had significant development with non-final dev kits.

I consider 2nd gen titles to have been developed entirely with final dev kits / hardware....and so on.

Since the most technically advanced games likely take 18-24 months to complete, we won't be seeing (by my definition) 2nd gen games on X360 until we get into 2007.

As an approximation, 18-24 months per generation. Of course, there is overlap. The best technical and art teams can likely have a 1st gen game look and play better than others' 2nd gen games, etc.

 

[Sethamin]

Based on what I've read, at least some (most?) developers are using GCC instead of IBM's XLC.

Hmm, after a bit of digging, it appears you're right. A CELL backend was written for the GNU Toolchain and Sony maintains a port of it. So the two compilers have completely different front ends, although IBM still uses that same backend to do the assembling and linking. Color me surprised.

 

[Titanio]

Based on what I've read, at least some (most?) developers are using GCC instead of IBM's XLC.

Further to that and Sethamin's query, SCE had a large hand in the GCC port, if they weren't entirely responsible. I think they did handle it themselves.

edit - beaten to it by Sethamin as it turns out I had this reply open for a while before posting it..

 

[Griffith]

I didn't say you were a troll, I said you were trolling. As they say: "attack the behavior, not the person". Anyway, I though this in particular was kind of a troll-like comment:

It just seemed to me like a ridiculous statement deliberately trying to inflame opinion. But, on second read, I suppose it's fine if you actually believe that. It seems pretty out there, but if that's really your view, then I apologize.

when I wrote that Good games will not launch in the first year of ps3, I was talking of the technical side, I'm sure that games with good gameplay, with good name and appeal will be there
I'm not shooting at the console, I was talking of my estimate use of the architecture in the first year of life

Cell is a different way to think, it's not simply "7 cores thing", there're a lot of issues and the devs have to study the way to workaround this issues in order to gain advantage from the spe's

with the right tools, I estimate 2-3 years to furfill the goal of "very good tech&graph" games

I've to explain my concept of software generation, usually first year = first gen, second year = second gen, but this is not an absolute rule, in xbox1 I can see 3 big steps, so I see only 3 generations in 4 years

for 360 I truly believe that second gen will start in 2007 with titles as HALO3, more or less the time when ps3 will reach europe with his first gen titles

so the key factors to great looking games, are two:
1) delay to provide final kit to developers
2) great tools to help devs to take advantage from hardware

in this perspective, octopiler can help a lot or make damage (if too late) the PS3 devs
it's usefull to remember that XNA is not here, for what I know, It's near 1-2 months to launch, this will be the fuel of second gen 2007 titles

really, I don't know how my thought, agreeable or not, can be classified as trolling

 

[Tap In]

I think that's an important and valid point raised; exactly what would one consider the generational breakdowns to be as far as games are concerned? I've always kind of though of Gears of War and the like as representing the second gen of Xox 360 titles, but maybe this isn't the best way to look at it. Perhaps the first crop of games to feature engines built around the tiling would be a better yardstick?

So that instead we have 'launch titles' and 'first-generation' games, which I see as kind of a half-step away from one another. So it'll be 360 'first-generation' games vs Sony 'launch titles,' for whatever that's worth.

I could get on board with this

I certainly disagree that PS3 launch willl be comparing to X360 2nd gen.

Hell IMO the X360 launch games were a placeholder (with a couple of exceptions) for the 1st gen. I'm assuming PS3 will have had more time to work on launch titles with close to final HW compared to X360's launch and allow for them to skip right over the initial X360 "launch" quality.

 

[one]

I'm not a huge fan of the 1st gen, 2nd gen, 3rd gen, 4th gen, etc, monikers.
(snip)
I dunno though, the term gets overused a bit I think.

I couldn't agree more.

Besides, Microsoft as a software company is supposed to have an advantage in this matter, so why their failure in bridging the gap between the alpha kit and the final kit is given strange excuses is beyond me.

 

[Acert93]

I think late 2006 PS3 titles will be comparable to late 2006 X360 titles. About the same more or less.

Here's my generational differences:

I consider 1st gen titles to be all titles that had significant development with non-final dev kits.

I consider 2nd gen titles to have been developed entirely with final dev kits / hardware....and so on.

Since the most technically advanced games likely take 18-24 months to complete, we won't be seeing (by my definition) 2nd gen games on X360 until we get into 2007.

As an approximation, 18-24 months per generation. Of course, there is overlap. The best technical and art teams can likely have a 1st gen game look and play better than others' 2nd gen games, etc.

Great summary Joe. I think it is common sense that the contrast between software developed on un-finalized hardware and software developed with the final hardware in hand from day one (and thus building an engine that plays to the strengths of the hardware and the art assets/quality are designed within the engine specs) is night and day. As I believe it was ERP who said a very long time ago the big difference between 1st and 2nd gen titles has to do with the development work flow as much as anything.

It takes time to idenitify bottlenecks and gauge what effects work best with a given architecture. Kind of hard to test out design schemes that surround features you cannot test.

In this sense Sony did a good job of contracting NV and providing developers with CELL and SM3.0 hardware very similar to the finished products spec sheet from day 1. MS should have bit the bullet and had 6800 series GPUs in the dev kits (heck, when they started MS had 9800 cards!)

Besides, Microsoft as a software company is supposed to have an advantage in this matter, so why their failure in bridging the gap between the alpha kit and the final kit is given strange excuses is beyond me.

How so? This is not a new phenomina by any means, it can be observed every generation. MS's was a big aid in being able to transition quickly from alpha hardware to beta hardware (and cross platform, one of the keys of their approach if anything, and so far we have seen a number of big 3rd parties co-release titles)--but I don't see the connection with how a more flexible and/or effecient software platform has anything to do with the challenges of building a game without having real hardware in hand.

For simplicities sake, if you build around a feature and find out it is 1/3rd as fast as you expected you are screwed. If you don't have certain techniques available to test you cannot use them.

Even PS3 devs here expect a gradual transition from early titles making light use of SPEs for "hevy lifting" with the PPE dominating, and yet at the end of the lifecycle the SPEs being heavily relied on. It takes time, good software or not, to learn and leverage new designs. I am sure we would be seeing similar issues from Sony if they had decided to adhear to the typically 5 year cycle.

 

[one]

How so? This is not a new phenomina by any means, it can be observed every generation.

Every generation? I've not known Microsoft had a previous generation before the Xbox 1.

MS's was a big aid in being able to transition quickly from alpha hardware to beta hardware (and cross platform, one of the keys of their approach if anything, and so far we have seen a number of big 3rd parties co-release titles)--but I don't see the connection with how a more flexible and/or effecient software platform has anything to do with the challenges of building a game without having real hardware in hand.

Hmm, what kind of things do you mean by a "more flexible and/or effecient software platform"? First, it is said Microsoft has a superior and robust API such as DirectX which is most likely not to have changed so much between the Xbox 1 (and Windows PC) and the XBox 360 unlike changes between PS1 -> PS2 and PS2 -> PS3. Second, they could write all sorts of emulators to test the latest features not found in alpha kits hardware by software. For example, PSP had a software emulator in its early development.

 

[AlNom]

First, it is said Microsoft has a superior and robust API such as DirectX which is most likely not to have changed so much between the Xbox 1 (and Windows PC) and the XBox 360 unlike changes between PS1 -> PS2 and PS2 -> PS3.

But DirectX does not make up all of the game engine. There are other hardware bits that are just inherently difficult to emulate wrt finding speed bottlenecks. From posted comments, it seems PC devs are just having a harder time than PS2 devs with the multicore approach, or at least they are more vocal about it . And then there's the IOE versus OOOE cpu issue.

*shrug*

 

[one]

But DirectX does not make up all of the game engine. There are other hardware bits that are just inherently difficult to emulate wrt finding speed bottlenecks. From posted comments, it seems PC devs are just having a harder time than PS2 devs with the multicore approach, or at least they are more vocal about it . And then there's the IOE versus OOOE cpu issue.

*shrug*

Yeah that's sure. Nonetheless you can save time on certain things, right?
An API set is kind of a virtual machine. You can get a working game early, then can concentrate on optmizations related to in-order CPU and other trivia. In addition, the "general purpose" perf in the Xbox 360 CPU could help.

 

[karlotta]

I see Mr John Stokes is full of shit as usual

And that is all you can say about it? It seems most of the the Sony guys here didnt understand what he was talking about and took the one thing they could attack....But then this is just a console forum with mostly flames and fights about pretty pictures.... You would do well to read the discusion in there forum. Heres a direct quote Mr Stokes:
"g65536,
Sorry if the post made it sound to you like the Octopiler is somehow hanging up PS3 development. I think that's pretty obviously not the case, considering how early on the project is--PS3 launch titles are already development and aren't waiting around for this. Current PS3 development is going ahead with whatever IBM has out right now, which seems largely like Tier II type tools.

When I said "or why the PS3 is running late", I meant that state of the project is indicative of the overall problems with programming the Cell. Everyone is still figuring out how to get their minds around this, and the deadlines are looming. If IBM invented this thing, and this is where they're at with it in 2/06...".

 

[liverkick]

When I said "or why the PS3 is running late", I meant that state of the project is indicative of the overall problems with programming the Cell. Everyone is still figuring out how to get their minds around this, and the deadlines are looming. If IBM invented this thing, and this is where they're at with it in 2/06...".

Even with this qualification I think the comparison still comes across as a bit specious since (as has already been noted in this thread) in the long run a particular compiler may not have anything to do with PS3 development whatsoever. There isnt exactly a correlation between the natural learning curve for developers with Cell and the status of this particular project of IBM's. Hardly an "indicative" one at least.

 

[Bobbler]

I think his problem is that he seems to think Cell is IBM's baby, when in reality it's likely much more Sony's baby than anything else. It doesn't really matter what state IBM's compilers/tools are in, they aren't what Sony is using anyways.

I'm not even sure IBM had much, if at all, to do with the PS3 dev tools.

 

[liverkick]

I think his problem is that he seems to think Cell is IBM's baby, when in reality it's likely much more Sony's baby than anything else. It doesn't really matter what state IBM's compilers/tools are in, they aren't what Sony is using anyways.

I'm not even sure IBM had much, if at all, to do with the PS3 dev tools.

Agreed, that seems to be where the static is coming from. Perhaps "specious" was a harsh adjective on my part. Still, even I was aware of IBM's limited role in PS3 dev tools and Im about as layman as this forum gets.

 

[one]

And that is all you can say about it? It seems most of the the Sony guys here didnt understand what he was talking about and took the one thing they could attack....But then this is just a console forum with mostly flames and fights about pretty pictures.... You would do well to read the discusion in there forum. Heres a direct quote Mr Stokes:
"g65536,
Sorry if the post made it sound to you like the Octopiler is somehow hanging up PS3 development. I think that's pretty obviously not the case, considering how early on the project is--PS3 launch titles are already development and aren't waiting around for this. Current PS3 development is going ahead with whatever IBM has out right now, which seems largely like Tier II type tools.

When I said "or why the PS3 is running late", I meant that state of the project is indicative of the overall problems with programming the Cell. Everyone is still figuring out how to get their minds around this, and the deadlines are looming. If IBM invented this thing, and this is where they're at with it in 2/06...".

Sorry, that post of Stokes just proves the point of Robert.L.