Bug 56534

Summary: [r600g] cayman msaa/fxaa buggy
Product: Mesa Reporter: Thomas Rohloff <v10lator>
Component: Drivers/Gallium/r600Assignee: Default DRI bug account <dri-devel>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: medium CC: alexandre.f.demers, mcclurem, vmerlet
Version: unspecified   
Hardware: x86-64 (AMD64)   
OS: Linux (All)   
See Also: https://bugs.freedesktop.org/show_bug.cgi?id=64695
Whiteboard:
i915 platform: i915 features:
Attachments: A in-game screenshot without any anti-aliasing method active.
A in-game screenshot with 8x MSAA active.
A in-game screenshot with 32x MLAA active.
A in-game screenshot with FXAA active.
sRGB fix

Description Thomas Rohloff 2012-10-29 17:50:20 UTC
Created attachment 69244 [details]
A in-game screenshot without any anti-aliasing method active.

MSAA looks just ugly (not like anti-aliasing should look like) and Oil Rush spams "OpenGL error: invalid operation", MLAA doesn't seem to do anything and FXAA just gives a white screen.

All anti-aliasing methods where tested with Oil Rush.

Exact settings:
MSAA: 8x, setted from the game launcher.
FXAA: Setted from the game launcher.
MLAA: 32x (for both, 2D and 3D) forced by drirc.
Comment 1 Thomas Rohloff 2012-10-29 17:51:01 UTC
Created attachment 69245 [details]
A in-game screenshot with 8x MSAA active.
Comment 2 Thomas Rohloff 2012-10-29 17:51:41 UTC
Created attachment 69246 [details]
A in-game screenshot with 32x MLAA active.
Comment 3 Thomas Rohloff 2012-10-29 17:52:09 UTC
Created attachment 69247 [details]
A in-game screenshot with FXAA active.
Comment 4 Thomas Rohloff 2012-10-29 17:57:34 UTC
Sorry for setting the wrong mime type for the images, they are png images. If you need me to re-attach just tell. :)
Comment 5 Alexandre Demers 2013-01-21 02:20:53 UTC
Forcing MLAA through driconf on Cayman 6950 crashes Gnome Shell if both mlaa and mlaa color (2D) are set (tested with both set to 8). If one of them is not set, Gnome Shell doesn't crashes.

MLAA color (2D) seems to have an effect on my Gnome Shell display.
Comment 6 Alexandre Demers 2013-01-21 20:10:37 UTC
Should differents anti-aliasing methods be tracked in differents bugs?
Comment 7 Alex Deucher 2013-04-24 18:43:05 UTC
(In reply to comment #6)
> Should differents anti-aliasing methods be tracked in differents bugs?

Yes.  MSAA is the only one that's part of the driver pipeline. MLAA is post processing effect.
Comment 8 Alex Deucher 2013-05-17 12:25:06 UTC
Is this still an issue with mesa from git?  There were some recent fixes for MSAA on cayman.
Comment 9 Thomas Rohloff 2013-12-04 22:40:06 UTC
(In reply to comment #8)
> Is this still an issue with mesa from git?  There were some recent fixes for
> MSAA on cayman.

I used fglrx for some time now cause of other bugs, but now I'm on mesa again (10.0). Right now MSAA doesn't seem to do anything at all. I didn't test other anti-aliasing methods for now, but you told to track them in different bugs anyway.

Should I still try latest mesa from git?
Comment 10 Marek Olšák 2013-12-05 17:46:03 UTC
It could be a bug in the game. Does MSAA work in other games?
Comment 11 Thomas Rohloff 2013-12-05 17:52:40 UTC
(In reply to comment #10)
> Does MSAA work in other games?

I tested:
- Oil Rush
- Left 4 Dead 2
- Garry's mod
- Worms Reloaded
- Portal
- Team Fortress 2
- Serious Sam 3
- Kerbal Space Program
- Killing Floor
and maybe other games I don't remember right now.
Comment 12 Marek Olšák 2013-12-13 21:35:49 UTC
Is it any better with Catalyst?
Comment 13 Marek Olšák 2013-12-14 17:32:34 UTC
First of all, I'm only interested in MSAA.

In addition to testing Catalyst, is it also any better with current Mesa master?
Comment 14 Marek Olšák 2013-12-14 17:34:39 UTC
Created attachment 90778 [details] [review]
sRGB fix

Also, does this patch for Mesa improve MSAA?
Comment 15 Alexandre Demers 2013-12-14 22:51:47 UTC
Removed MLAA from the topic since it was covered, tested and fixed in bug 64695.
Comment 16 Thomas Rohloff 2013-12-16 13:58:30 UTC
(In reply to comment #12)
> Is it any better with Catalyst?

Yes. With catalyst the FPS goes down while the scene looks anti-aliased. With r600g the FPS and the scene stay the same.

(In reply to comment #13)
> is it also any better with current Mesa
> master?
I would love to answer you that question but master is currently a crash party for me (3 of 4 tested games crashed at launch, with the one that worked it's really hard to tell if anti-aliasing is working. So my only answer right now can be a "maybe"). I might have to investigate more into this later and open a new bug report.

(In reply to comment #14)
> Created attachment 90778 [details] [review] [review]
> sRGB fix
> 
> Also, does this patch for Mesa improve MSAA?

Yes! But there are new bugs with it (tested with gary's mod: Colors are generally darker, sometimes when looking at some things they get even more dark. But screenshots look normal. Should I open a new bug report for it or is that already known/fixed?).
Comment 17 Marek Olšák 2014-01-14 11:37:59 UTC
Is MSAA any better with current mesa git? If you enabled the LLVM backend, please turn it off for this bug.
Comment 18 Alexandre Demers 2014-02-01 00:53:03 UTC
(In reply to comment #17)
> Is MSAA any better with current mesa git? If you enabled the LLVM backend,
> please turn it off for this bug.

Tested with TF2 with latest git code: the FPS does go down now and the display is definitively different.
Comment 19 Marek Olšák 2014-02-04 00:24:28 UTC
(In reply to comment #18)
> (In reply to comment #17)
> > Is MSAA any better with current mesa git? If you enabled the LLVM backend,
> > please turn it off for this bug.
> 
> Tested with TF2 with latest git code: the FPS does go down now and the
> display is definitively different.

What does it mean?
Comment 20 Alexandre Demers 2014-02-04 04:59:50 UTC
(In reply to comment #19)
> (In reply to comment #18)
> > (In reply to comment #17)
> > > Is MSAA any better with current mesa git? If you enabled the LLVM backend,
> > > please turn it off for this bug.
> > 
> > Tested with TF2 with latest git code: the FPS does go down now and the
> > display is definitively different.
> 
> What does it mean?

It's working, as expected.
Comment 21 Marek Olšák 2014-02-04 12:00:28 UTC
Okay. Closing.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.