Hi Emilian,

This has all the hallmarks of driver bug.  Driver bugs come and go
from driver version to driver version, we can account for this a
little by use the OSG's extension disable feature but adding hardwired
hacks into the code as you have done introduces the hack for all OSG
uses, regardless of OS, hardware, driver version for evermore.

If we were to start adding in random hacks like this for driver bugs
the OSG would very rapidly end up a complete quagmire of hacks and be
slow and completely unmaintainable.

As for a viable fix for the problem you are seeing?  Updating the
driver might resolve the issue.  Running the viewer single threaded
would also be lightweight way to avoid the driver bug.

Robert.

On 17 May 2015 at 20:45, Emilian Mihalache <[email protected]> wrote:
> Hello,
>
> While going through the "OpenSceneGraph for Beginners" book and testing out
> the "rendering more scenes at one time" example, I came across a SIGSEGV in
> fglrx. Attached zip file contains stack trace (as reported by gdb) + source
> code of program that can be used to reproduce the crash 100% of the time on
> my machine (main.cpp) + proposed solution (tested to work, at least on my
> machine) (State.cpp).
>
> I'm not 100% sure why this works, adding the Mutex was just something that
> occurred to me after adding some debug prints in State::frameCompleted and
> observing the behavior: 1. the first two Viewers block inside the
> 'glGetInteger64v' call; 2. the third Viewer comes along; 3.
> 'glGetInteger64v' un-blocks one of the callers; 4. SIGSEGV.
>
> Assumption is that fglrx does not properly synchronise access.
>
> I don't know if this solution is in any way acceptable (since I guess a
> Mutex would not be cheap, especially in that part of the framework), or if
> anyone else is seeing this problem, but I figured I might share my fix
> anyway.
>
> Additional info: Ubuntu 14.04, fglrx 13.35.5 [Mar 12 2014] on a Radeon
> Mobility HD 5650.
>
> Please let me know if any additional information is required.
>
> Best Regards,
> Emilian
>
>
> _______________________________________________
> osg-submissions mailing list
> [email protected]
> http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org
>
_______________________________________________
osg-submissions mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org

Reply via email to