RSXGL: OpenGL 3.1 Support For The PlayStation 3

[przemoli]

Dev in question do not pretend to implement whole OpenGL 3.1 core profile.

But to treat it as a goal IF hardware support it (and if mesa support GLSL 1.40 :P ).

On the github is STATUS file go and read it.



On the other hand OpenGL 2.1 =\= OpenGL ES, so there is no guarantee that there would be full OpenGL 2.1 on PS3 either.

[const]

The RSX chip today hasn't secrets : [url]http://en.wikipedia.org/wiki/RSX_%27Reality_Synthesizer%27[/url

i think we all are missing the point, because RSX doesn't even support OpenGL 2.1 - it supports "PSGL", which is "PlayStation GL" and it's subset of OpenGL for PlayStation. so, if Alexander Betts called OpenGL 2.x then it would be again wrong and probably cause people criticizing again. that's why i think what Alexander Betts did is support of "PSGL" as subset of OpenGL 3.1, i.e. it's not full OpenGL 3.1 and it's not even full OpenGL 2.x - it's just what's necessary for "PSGL" as subset of the highest possible OpenGL version. we all just miss too much knowledge to fully understand and i think Dave Airlie that commented earlier as respected Xord developer is not that familiar with "PlayStation3" platform and architecture too and judges from not entirely correct point of view.

[airlied]

i think we all are missing the point, because RSX doesn't even support OpenGL 2.1 - it supports "PSGL", which is "PlayStation GL" and it's subset of OpenGL for PlayStation. so, if Alexander Betts called OpenGL 2.x then it would be again wrong and probably cause people criticizing again. that's why i think what Alexander Betts did is support of "PSGL" as subset of OpenGL 3.1, i.e. it's not full OpenGL 3.1 and it's not even full OpenGL 2.x - it's just what's necessary for "PSGL" as subset of the highest possible OpenGL version. we all just miss too much knowledge to fully understand and i think Dave Airlie that commented earlier as respected Xord developer is not that familiar with "PlayStation3" platform and architecture too and judges from not entirely correct point of view.

PSGL is OpenGLES 1.0 + some GLES 2.0 bits in it, so yeah it might not be able to pull of desktop GL at all, so why bother tackling desktop GL on it, just use mesa to implement the GLES bits.

so eyah he should aim even lower than I suggest!
Dave.

[const]

PSGL is OpenGLES 1.0 + some GLES 2.0 bits in it, so yeah it might not be able to pull of desktop GL at all, so why bother tackling desktop GL on it, just use mesa to implement the GLES bits.

so eyah he should aim even lower than I suggest!
Dave.

Dave, thanks to clarifying things, but i've just found in Japanese (even before i read your reply):

http://pc.watch.impress.co.jp/docs/2.../kaigai104.htm

that it seems Sony claims "OpenGL ES 2.0" support for Playstation3. so, i guess they do the same, i.e. read "OpenGLES 1.0 + some GLES 2.0 " what is actually the case as you said, as "OpenGL ES 2.0 support", when technically there is no full ES 2.0 support. also, in another link that i lost, i read that graphic libraries that Sony developed for OpenGL support on Playstation3 use not only the RSX (nVidia graphic chip), but also the "Synergistic Processing Elements (SPE)" in the CellBE CPU - that i guess adds even more features compared to using just the graphic RSX processor, because SPE are 7-cores in CellBE CPU and very fast, i.e. like software emulation of graphic features.

so, could it be that Alexander Betts work really implements parts of OpenGL 3.1 and they are really supported by Playstation3 hardware, because of its strange architecture, i.e. "RSX (nVidia chip)+PowerPC 2-core CPU+7 cores of SPE inside the PowerPC CPU" and how Sony drivers and libraries tied together those parts to achieve what is called "PSGL for PS3"? i hope you follow me - i mean maybe what can be supported by Playstation3 hardware includes different features of different OpenGL versions including OpenGL 3.1 and that mix is actually "PSGL for PS3". as far as i understand Dave's opinion that entirely technically correct should be called as "OpenGL core version which features are fully covered + this and that extensions", but in any case in my opinion the work that Alexander Betts did is significant progress.

[Lynxeye]

The RSX is just a regular NV4x class chip. The only differences are in the number of execution engines and different memory access. The featureset is exactly the same. We already have a relatively good running driver for this chips MESA, it's called nv30 and implements OpenGL 2.1 + extensions.

I definitely don't want to downplay his achievements, all I'm saying is that he should really call it what it is and that he could have achieved even more if he would have taken the MESA infrastructure and driver we have today and implement missing/wrong bits.

[const]

The RSX is just a regular NV4x class chip. The only differences are in the number of execution engines and different memory access. The featureset is exactly the same. We already have a relatively good running driver for this chips MESA, it's called nv30 and implements OpenGL 2.1 + extensions.

I definitely don't want to downplay his achievements, all I'm saying is that he should really call it what it is and that he could have achieved even more if he would have taken the MESA infrastructure and driver we have today and implement missing/wrong bits.

ok, but if that's the case could you care to explain the tremendous performance difference:

* without RSXGL: http://dl.dropbox.com/u/22642664/expedite_psl1ght.log

* with RSXGL: http://dl.dropbox.com/u/22642664/expedite_rsxgl.log

that's the main reason i can't agree with you that until now there was anything for Playstation3 that do the same thing or at least do it with awful performance.

[curaga]

ok, but if that's the case could you care to explain the tremendous performance difference:

Perhaps it's simply: psl1ght sucks, his work is better?

Performance is unrelated to calling it "GL X".

[Squarepusher]

Wanted to correct some (mis)information here:

http://pc.watch.impress.co.jp/docs/2.../kaigai104.htm

that it seems Sony claims "OpenGL ES 2.0" support for Playstation3. so, i guess they do the same, i.e. read "OpenGLES 1.0 + some GLES 2.0 " what is actually the case as you said, as "OpenGL ES 2.0 support", when technically there is no full ES 2.0 support. also, in another link that i lost, i read that graphic libraries that Sony developed for OpenGL support on Playstation3 use not only the RSX (nVidia graphic chip), but also the "Synergistic Processing Elements (SPE)" in the CellBE CPU - that i guess adds even more features compared to using just the graphic RSX processor, because SPE are 7-cores in CellBE CPU and very fast, i.e. like software emulation of graphic features.

Let's get a couple of things straight.

Technically, the RSX doesn't support 'jack' regardless of what tech specs tell you. PSGL is just a wrapper library for libgcm (the lower-level graphics API available for the PS3). Developers have the option of using it or not and must do so by either linking in precompiled static libraries or compiling the (bundled) source themselves (advantages being they can customize it themselves to meet their performance requirements).

In terms of conformance, it's conformant with OpenGL ES 1.1 (plus some extensions). There are also some Sony-specific extensions such as texture references and attribute sets but this falls outside the scope of this discussion.

PSGL has been abandoned since 2007 and hasn't been updated since - originally the plan was to go for OpenGL ES 2.0 conformance but developers were so disappointed by the performance they were getting out of it that they simply cut their losses and create raw libgcm drivers instead of going through a GL wrapper.

About the CPU (it stuns me that there are still people out there that think this processor is 'powerful', but whatever - it might be they never coded for it):

so, could it be that Alexander Betts work really implements parts of OpenGL 3.1 and they are really supported by Playstation3 hardware, because of its strange architecture, i.e. "RSX (nVidia chip)+PowerPC 2-core CPU+7 cores of SPE inside the PowerPC CPU" and how Sony drivers and libraries tied together those parts to achieve what is called "PSGL for PS3"?

Wrong - it is a 1-core PowerPC design with 2 hardware threads. Second correction with regards to the CPU -

but also the "Synergistic Processing Elements (SPE)" in the CellBE CPU - that i guess adds even more features compared to using just the graphic RSX processor, because SPE are 7-cores in CellBE CPU and very fast, i.e. like software emulation of graphic features.

Those 7 SPU cores are not meant for scalar code at all - and while those SPUs are very fast, they only have 256K RAM each available to them. If you don't have enough workload to offload to those 7, you'll find you are wasting more CPU time on IPC overhead than the gains you will be able to achieve by going multithreaded in this way.

Overall, the Cell architecture was a nice design on paper but it's really quite the failure - the 1-core in-order PowerPC CPU performs like a Pentium 4 2.4GHz for scalar code (on a good sunny day) and those 7 SPUs aren't really as flexible as you seem to think.