On Wed, Nov 24, 2010 at 6:06 PM, Robert Osfield <[email protected]>wrote:
> Hi Tim,
>
> I have just checked in the following workaround:
>
> const GLubyte* renderer = glGetString(GL_RENDERER);
> std::string rendererString = renderer ? (const char*)renderer : "";
> if (rendererString.find("Radeon")!=std::string::npos ||
> rendererString.find("RADEON")!=std::string::npos)
> {
> // AMD/ATI drivers are producing an invalid enumerate error on
> the
> // glGetQueryiv(GL_TIMESTAMP, GL_QUERY_COUNTER_BITS_ARB, &bits);
> // call so work around it by assuming 64 bits for counter.
> setTimestampBits(64);
> //setTimestampBits(0);
> }
> else
> {
> GLint bits = 0;
> extensions->glGetQueryiv(GL_TIMESTAMP,
> GL_QUERY_COUNTER_BITS_ARB, &bits);
> setTimestampBits(bits);
> }
>
> I've tested this out on my ATI 4670 + Kubnutu 10.10 system and it
> kinda works, but not too well. Sometimes I get the orange bars
> appearing but typically quite a few frames in, and the bars often
> appear and then disappear, sometimes never to appear again. I've
> experimented with using adding a glFinish() to
> osg::State::frameCompleted() and this does stabilize things but it's
> still not 100% reliable.
>
> Any thoughts on why the timing doesn't seem reliable? It might simply
> be the the GL_TIMESTAMP code in the driver is flaky and just needs to
> be disabled completely for ATI cards. I guess there is chance that
> you see the instability under NVidia as well.
>
> I believe that the GPU activity for a frame is happening well after the
call to swapBuffers() for that frame. This is one of those modern
deeply-pipelined GPU phenomena. You have vsync on, correct? The fact that
inserting a glFinish() seems to improve the display would seem to be
evidence for this theory. I have a plan to change the colors in the stats
display to be per frame, not per traversal, which would visibly confirm
this.
> Could you do an svn update and test the new code out to make sure
> there hasn't been a regression under NVidia.
>
> I will do this later and report if there are any problems.
Thanks,
Tim
> Cheers,
> Robert.
>
> Robert.
> _______________________________________________
> 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
