Open Source 3D stacks and WINE

[nanonyme]

3. Wine relies on Windows-only extensions being present in the X implementation.

I've never heard of that one. Care to explain a bit better what you mean by it?

[bridgman]

I was thinking of pbuffers. As far as I know, the pbuffers extension was only officially added to WGL (Windows). I believe it was "experimental" in GLX (as GLX_SGIX_PBUFFER) but never became an official extension. I haven't been able to reconcile that with the fact that every third word in the GLX 1.4 spec is "pbuffer", but...

I think we support pbuffer anyways since a number of apps use it (including fgl_glxgears apparently ).

[dungeon]

We? Oh that is in fglrx, but that is not the case with DRI project which is build on top of GLX 1.2 specs and of course because of that it miss render_to_texture feature in hw which was also used in many Windows games. In games like SW:Jedi Academy, COD 1, etc. that effects will slow down games to 0 fps.

[bridgman]

We? Oh that is in fglrx....

D'oh !! This is an open source thread, isn't it. Sorry

...but that is not the case with DRI project which is build on top of GLX 1.2 specs and of course because of that it miss render_to_texture feature in hw which was also used in many Windows games.

I think that will all change fairly quickly once the open driver stack is running solidly on the memory manager. Without a common memory manager it didn't make a lot of sense to try extending the 3D driver much past its current level of support, but once everything is running nicely over GEM/TTM/DRI2 it's a totally different game.

[dungeon]

Do you know bridgman what is state of accumbuffer which is also not available in drivers, also auxiliary buffers... All those are also reason of slow down games in, as i know, pbuffer and other bo situations. Will those buffers might be imlemented in that futurity GEM/TTM? In intel's GEM they are not supported also, only framebuffer_object is supported there as GL extension.

P.S. Those two buffers are used in Max Payne 2 for Motion Blur effects, also "Mirrors" is render_to_texture i think which is there but 0 fps.

[bridgman]

Pretty much everything with "buffer" in the name is on the "do it after we have memory management" list, and most of those are also "do it after we have Gallium3D so with luck it doesn't have to be done once for each GPU".

I don't think there's any specific roadmap for any of these features, just that maybe 3-6 months from now it should get a lot easier to add them.

On the other hand, it appears that osiris got Occlusion Queries working today. I think it was OQ anyways

[dungeon]

Oh holly buffers. Good thing that occlusion queries for Sauerbraten and for Quake Wars to even start Where is that code?

[sandain]

I was thinking of pbuffers. As far as I know, the pbuffers extension was only officially added to WGL (Windows). I believe it was "experimental" in GLX (as GLX_SGIX_PBUFFER) but never became an official extension. I haven't been able to reconcile that with the fact that every third word in the GLX 1.4 spec is "pbuffer", but...

I think we support pbuffer anyways since a number of apps use it (including fgl_glxgears apparently ).

The FGLRX driver does export the GLX_SGIX_PBUFFER extension in Linux, and the Windows driver exports the WGL_ARB_PBUFFER extension. However, I am fairly certain that the implementation is bugged in both operating systems. As I posted on a previous thread:

Here is an example of pbuffer rendering in Windows, notice the transparent mini-map?


Here is the way it looks when rendered in Wine.


And finally, the way it should look, when rendered with d3d.


Note, that Nvidia drivers work just fine in Windows and Linux.


Update: Since this thread has been resurrected, I thought I should mention that AMD has fixed this bug as of FGLRX 9.8. Thanks AMD!

[bridgman]

Oh holly buffers. Good thing that occlusion queries for Sauerbraten and for Quake Wars to even start Where is that code?

I think osiris has been working in radeon-rewrite plus drm and radeon ddx from F11 / Rawhide, not 100% sure though.

Doesn't look like the latest OQ work has been checked in yet though... all I have to go on is :

http://people.freedesktop.org/~cbril...ate=2009-04-23

... around 08:17 log time. Osiris announced that it worked then disappeared, presumably for some serious celebration.

I'm not even 100% sure it's OQ we're talking about, but it seems likely.

[bridgman]

However, I am fairly certain that the implementation is bugged in both operating systems.

Note, that Nvidia drivers work just fine in Windows and Linux.

Has anyone found a pbuffers bug that shows up in a simple test app ?

Not behaving the same way as a competitor's driver is not always the same as having a bug (although of course sometimes it is).