IBM releases Cell SDK and simulation tools

S

Silkworm

Newcomer
Clicky the linky.
Runs on/needs a Fedora Core 4 environment though, and fairly beefy system requirements: >2Ghz, 1GB ram, 5GB disk, etc..

Highlights include:
- a timing-accurate simulator
- a whole-system emulator, named "Mambo"
- GCC tool chain
- XL C (Alpha) compiler with auto-SIMD-ification
- example code, including the FFT demo presented earlier this year at Barcelona
 
Last edited by a moderator:
Where do you see the 1 GB RAM requirement? The page I am looking at states that you need double the memory of the system you wish to emulate. I suppose this is 1 GB minimum for all intents and purposes, but you could probably run it on less.

link

EDIT: Doh! Was looking at the emulator only. My eyes also see the 1GB requirement for the full SDK. :smile:
 
Last edited by a moderator:
Great news! Gonna try and set this up over the next couple of days, although I'm waiting on some new memory for my computer..

Does anyone know if there are any specific sites/discussion boards now for Cell programming, or still just IBM's own? An independent community really ought to grow around this..

edit - to add to the requirements saga, the FAQ says you may get away with less than 1GB if you're not using XL C..

By the way, do PS3 kits come with XL C? I vaguely recall someone saying it does not. Kind of weird to have that, or an alpha of it even, in a public release like this and not available for game devs, if that's the case.
 
Last edited by a moderator:
Fark... these PC's here only have 512 MB of RAM... does it refuse to run with less than that ? Bah, it probably would run crappy. Come on SCE give us PLAYSTATION 3 Linux soon :D.
 
Titanio said:
Great news! Gonna try and set this up over the next couple of days, although I'm waiting on some new memory for my computer..

Does anyone know if there are any specific sites/discussion boards now for Cell programming, or still just IBM's own? An independent community really ought to grow around this..

edit - to add to the requirements saga, the FAQ says you may get away with less than 1GB if you're not using XL C..

By the way, do PS3 kits come with XL C? I vaguely recall someone saying it does not. Kind of weird to have that, or an alpha of it even, in a public release like this and not available for game devs, if that's the case.
Click to expand...

SCE did not license XLC so no :). Maybe XLC will be available as middleware.

I have to install Fedora now :D.
 
Panajev2001a said:
I have to install Fedora now :D.
Click to expand...

Same. I'm on 512MB for the short term aswell, but I think it's worth a shot.

I wonder why SCE didn't license XLC, given that they were license-happy with things like Havok and Ageia. Maybe they think the GCC toolchain is so good ;) Or maybe XLC as it currently is doesn't offer much additional benefit (though things like auto-SIMDisation do sound nice!).
 
Last edited by a moderator:
Titanio said:
Same. I'm on 512MB for the short term aswell, but I think it's worth a shot.

I wonder why SCE didn't license XLC, given that they were license-happy with things like Havok and Ageia. Maybe they think the GCC toolchain is so good ;) Or maybe XLC as it currently is doesn't offer much additional benefit (though things like auto-SIMDisation do sound nice!).
Click to expand...

First, SCE knew that Faf and nAo, while bitching to have it, would have never touched it... what, using something high-level that does the dirty ASM/vectorization job for them ? What'cha talking about IBM ;) ?

Since SCE does not want them to buy toys they will not use, well it was one step ahead of them... look, they already listened, with RSX, to this nagging of theirs about some sort of "hey SCE, did you fall asleep when the teacher was talking about clipping in class?" problem with PSTwo and PSP, so come on :p.

Hopefully SCE and IBM (the team at the STI labs in Texas started the GCC and Linux basic work, but one of IBM Germany's Linux Technology R&D labs is finishing the job porting Linux to CELL and a full GCC and GDB solution for the PPE and the SPE's as well as the SPU Virtual File-System IIRC) can get a good GCC going that is at least not too many miles behind MS's own compiler and XLC.

XLC plans to do MUCH more than simple auto-vectorization: such things as software caching for the SPE, etc... are part of their plans.

http://www.research.ibm.com/cellcompiler/compiler.htm (where do you think that Octopiler's efforts will be delivered to... first ;) ? XLC :)).
 
Last edited by a moderator:
They do not even need to disclose to homebrew programmers the I/O CPU or too many inner details of RSX (hiding it behind an OpneGL ES 2.0 interface and/or Cg, but with these two we would be getting access to Vertex and Pixel Shaders and Cg based tools), so they can avoid worrying about people using the kit to learn how to pirate PLAYSTATION 3 games. They can wrap around Libraries/System Calls all that is related to the I/O processor and similar low level things.
 
One of the questions regarding development of software for PS3 Linux is 'where would Sony get their money from?' Do we think that the availibility of a cheap Cell platform will help push Cell as a viable mainsteam CPU? In essence taking a potential hit in one area (games sales, with people buy Linux games instead of PS3 titles. Yeah I know, no-one's going to be knocking out MGS4's and GT5's from their bedroom, but...) in order to increase software development into the wider field? Provide a ready base of experienced Cell developers for helping Cell make the transition into other devices?
 
Shifty Geezer said:
One of the questions regarding development of software for PS3 Linux is 'where would Sony get their money from?' Do we think that the availibility of a cheap Cell platform will help push Cell as a viable mainsteam CPU? In essence taking a potential hit in one area (games sales, with people buy Linux games instead of PS3 titles. Yeah I know, no-one's going to be knocking out MGS4's and GT5's from their bedroom, but...) in order to increase software development into the wider field? Provide a ready base of experienced Cell developers for helping Cell make the transition into other devices?
Click to expand...
The question is: Would it hurt them?
As long as the PS3 sells at a loss I would say yes, as soon as they break even not.
If a couple of people buy PS3s as Cheap Computer (FOH or similar stuff) and not game at all, theres only the margin/loss of the hardware.
I dont see Developers releasing their games for a Cell-Linux Plafform anytime soon, even with the probability of modchips PS3-Games (native-PS3 Games if you wish) will have lower piracy.
 
Maybe some relevancy to the topic...

http://money.cnn.com/2005/11/10/technology/linux.reut/index.htm
IBM, Sony, Philips form Linux alliance

The Open Invention Network (OIN), as the new firm unveiled Thursday is known, could mark a breakthrough in resolving how to protect vendors and customers from patent royalty disputes resulting from freely shared Linux code.

If OIN's approach to managing intellectual property wins acceptance, it could overcome a big stumbling block to wider corporate adoption of Linux and pose challenges for major opponent Microsoft Corp., which has argued that relying on "open source" software poses legal risks.

OIN is funded by the world's largest computer company, IBM, two consumer electronics giants -- Sony Corp. of Japan and Philips of the Netherlands -- and Linux distributors Red Hat Inc. and Novell Inc.
Click to expand...
 
Titanio said:
Same. I'm on 512MB for the short term aswell, but I think it's worth a shot.

I wonder why SCE didn't license XLC, given that they were license-happy with things like Havok and Ageia. Maybe they think the GCC toolchain is so good ;) Or maybe XLC as it currently is doesn't offer much additional benefit (though things like auto-SIMDisation do sound nice!).
Click to expand...

XLC has always been a mythical compiler for PPC development. However, every real world, though informal, test of GCC vs XLC indicated that the vast majority of code was roughly equal and each compiler having small areas where they beat the other significantly.

The problem for SIMD support in compilers, in my experience, has been that the areas you get for free from the compiler are either non-hotspots that make no noticeable performance difference or hotspots that you are going to be focusing heavily on anyways and will be tweaking by hand.

The best PPC compiler has been Metrowerk's, but GCC has rapidly narrowed the gap over the past few years. I don't know anyone who doesn't use Metrowerks, including myself, for their Mac game versions.
 
SubD said:
XLC has always been a mythical compiler for PPC development. However, every real world, though informal, test of GCC vs XLC indicated that the vast majority of code was roughly equal and each compiler having small areas where they beat the other significantly.

The problem for SIMD support in compilers, in my experience, has been that the areas you get for free from the compiler are either non-hotspots that make no noticeable performance difference or hotspots that you are going to be focusing heavily on anyways and will be tweaking by hand.

The best PPC compiler has been Metrowerk's, but GCC has rapidly narrowed the gap over the past few years. I don't know anyone who doesn't use Metrowerks, including myself, for their Mac game versions.
Click to expand...

SPE wise though, XLC might have definately the upper hand. Not only performance-wise (proper scheduling of instructions to handle misaligned scalar loads and stores), but ease of use wise: as that link I posted mentioned, managing Software Caches for the SPE's Local Storage, code overlays, etc... are the tip of the iceberg for what IBM wants to realize.

Ultimately IBM points to a single source compiler.

Sony/SCE, IBM and Toshiba will have pushed GCC well and improoved it a lot, but this is more important concerning the quality of the compiler compared to GCC 2.95.2 and VCL (which was not bad [all the contrary, I cannot imagine not having it :D], but still restricted you to a ASM-like coding style) than concernign how it stacks against XLC (which IMHO, if it is THAT important for game development, will be available as middleware for PLAYSTATION 3).
 
Has anyone tried this devkit yet? Got any insights? C'mon, there must be lots of info people can say like 'God, how do we work with these damned SPEs?' or 'run's okay. Written a BSP tree traversal in 64 kb' or even 'Hello World program still don't run.'
 
To install it... download from the BSC center you download the install.sh script from all the packages they provide on that page and put them together with the ones you downloaded from IBM's page (the script won't download ALL the others as it says in the installation guide).

Also remember to have freeglut and freeglut-devel installed in the system as without glut.h one of the samples will not compile.
 
Shifty Geezer said:
Has anyone tried this devkit yet? Got any insights? C'mon, there must be lots of info people can say like 'God, how do we work with these damned SPEs?' or 'run's okay. Written a BSP tree traversal in 64 kb' or even 'Hello World program still don't run.'
Click to expand...

I just got done getting the install.sh script to correctly (just to make sure I started it again from the very first step ;)) check, download and build the SDK and its components. Give people some time or at least enough to understand how to use this SDK :p.

It is funny though, basically documentation for the CPU, tools, compilers are being released by SCE and IBM before the console is even in the stores... PlayStation 2 Linux did not get out at system launch... and they never got Cygwin to release for free the EE simulator. I hope this is good news for those of us who are waiting for a PLAYSTATION 3 Linux kit to be released by SCE once the system launches :).
 
You must log in or register to reply here.
Back
Top