nV News Forums

 
 

nV News Forums (http://www.nvnews.net/vbulletin/index.php)
-   NVIDIA Linux (http://www.nvnews.net/vbulletin/forumdisplay.php?f=14)
-   -   Xcomposite Bug (http://www.nvnews.net/vbulletin/showthread.php?t=73508)

felix-bellaby 07-16-06 10:24 AM

Xcomposite Bug
 
1 Attachment(s)
I posted a simple GL compositor developed for the accelerated nvidia platform around a month ago. Since then, I have rewritten much of the code, ported it to a wider range of platforms, tested it under a range of different systems and discussed it with members of the xorg mailing list on freedesktop.org. I attach the latest version.

When I first posted the compositor here, I pointed out a pair of problems that I was having, which James Jones was kind enough to discuss with me. He was able to reproduce one of the problems on his platform and posted a bug report onto the nvidia tracker. However, he was unable to reproduce the other bug that I was finding.

Unfortunately, I have yet to identify a single platform using the nvidia drivers on which this bug does not appear. On the other hand, it does not seem to appear on any other platforms (including Mesa / Xorg nv). Therefore, I am driven to the conclusion that it must be a problem in the nvidia drivers.

When the attached compositor is built with COPY_BUFFER set to 1 on line 64 then GL apps can be composited successfully. When COPY_BUFFER is set to 0 then they freeze. I now know that this bug is reproducible with the 1.0.87.62 drivers over the following platforms:
6150 GPU:
Fedora 5 Xorg 7.0 (x86_64), Fedora Rawhide Xorg 7.1 (x86 & x86_64),
Vanilla Xorg 7.1 (x86_64)
7800 GTX GPU:
Xorg 7.0 (x86_64)

netllama 07-16-06 10:33 AM

Re: Xcomposite Bug
 
Please generate and post an nvidia-bug-report.log.

Thanks,
Lonni

felix-bellaby 07-17-06 01:27 PM

Re: Xcomposite Bug
 
1 Attachment(s)
Okay, here is an nvidia-bur-report.log file for the Fedora Rawhide (x86_64) GPU 6150 system.

I am afraid that I installed the driver on this system from the livna repositories, but given the wide range of platforms on which I have seen this error, I doubt that this point is significant.

Felix

netllama 07-17-06 03:13 PM

Re: Xcomposite Bug
 
Thanks. I just tested our your compositing mgr, and even without running any OpenGL apps, I'm seeing severe corruption. When I ran glxgears, it didn't hang, but the rendering, as with everything else was badly corrupted.

Is this expected behavior?

-Lonni

felix-bellaby 07-17-06 04:39 PM

Re: Xcomposite Bug
 
1 Attachment(s)
No, severe corruption is certainly not the expected behaviour on nvidia platforms, nor have I had such reports from other testers. The only problems that I have encountered have arisen from broken GLX setups on non-nvidia platforms and they resulted in failures to initialise GL correctly.

I have just rebuilt and tested the version that I posted in this thread, and it seems to be work fine on my machine with COPY_BUFFER set to 1. However, it does look like their might be a problem with COPY_BUFFER set to 0 on this version, so I have attached the previous version, which seems fine in both cases.

Note that when running with COPY_BUFFER set to 0, it is very slow and windows will take time to "come good". This is due to the restrictive steps needed to avoid another known bug.

What are you testing it on, and what kind of corruption are you seeing?

Felix

netllama 07-17-06 05:07 PM

Re: Xcomposite Bug
 
6150 GPU, FC5-x86_64 with 1.0-8762.

Thanks,
-Lonni

felix-bellaby 07-17-06 05:57 PM

Re: Xcomposite Bug
 
Shouldn't be a problem with a platform difference then. :)

Some of the very recent changes on Rawhide look like they have created some new problems for the COPY_BUFFER 0 case even in some of the versions before glcompmgr-0.4. I find that things seem to work much better in this case on xorg-x11-server-Xorg-1.1.0-13 and its predecessors than they do on the latest xorg-x11-server-Xorg-1.1.0-25.fc6. You might be better sticking to Xorg 7.0.

There is an issue with software cursors under this compositor, but it only causes minor corruption around the cursor itself. Turning to a hardware cursor eliminates this cursor problem.

Felix

felix-bellaby 07-20-06 10:43 AM

Re: Xcomposite Bug
 
1 Attachment(s)
Hi Lonni,

Here is a version of the compositor that works with or without COPY_BUFFER 1 on the latest RawHide Xorg.

It might help to isolate the problem, because it makes more focused use of the server grabs.

Felix

netllama 07-20-06 11:21 AM

Re: Xcomposite Bug
 
The previous problem is resolved. What I'm now seeing is the entire X screen flickering whenever an OpenGL app (glxgears, etc) is running with COPY_BUFFER=0. Is that the bug you're reporting?

Thanks,
Lonni

Dragoran 07-21-06 07:31 AM

Re: Xcomposite Bug
 
Quote:

Originally Posted by netllama
The previous problem is resolved. What I'm now seeing is the entire X screen flickering whenever an OpenGL app (glxgears, etc) is running with COPY_BUFFER=0. Is that the bug you're reporting?

Thanks,
Lonni

does glxgears draw anything or does it shows a black window? (like it happens for me)

netllama 07-21-06 09:35 AM

Re: Xcomposite Bug
 
glxgears draws, albeit slowly, in between the flickering.

-Lonni


All times are GMT -5. The time now is 11:28 AM.

Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.