View Single Post
Old 04-17-12, 02:11 AM   #13
Registered User
Join Date: Jul 2011
Location: Moscow, Russian Federation
Posts: 58
Send a message via ICQ to lexa2 Send a message via Skype™ to lexa2
Default Re: Regression in "KB:The Legend" when run under Wine with drivers v.285.05.09

Thanks for infos. In anycase, it turns out that reproducing this bug would be a tough thing: tonight I've got glretrace playing back the problematic trace in loop during all over the night, and checking on the morning if the render gone bogus come to a no-go - everything had been rendered correctly, as it should. So I kept glretrace executing loop to does its job in background and went on with doing my usual morning business (mail checking, news reading, e.t.c.) and after a bit while suddenly noticed that the trace rendering is corrupt, featuring the bug. To confuse the things even more, renderings were corrupt for several subsequent runs, and then magically auto-healed.

The behavior I observe currently made me believe that the problem comes not from the trace itself, but rather somewhere from general system environment I've got here, like the processes running in background at the same time this bug shows up. For now what does look most suspect for me are web browsers with notorious flash plugin installed. As an indirect confirmation to this guess is a fact that I've got Mozilla Firefox and/or Google Chrome almost always running "at background" since the very startup of the desktop environment session, thus if that is flash who somehow corrupts the driver state by using vdpau and/or opengl - it's not a surprising that I had been hitting this bug on a regular basis before. On the other hand if that's the real case - it might be hard to reproduce it reliably now due to the version update I had installed for Chrome (switched into using 18.x series), and that version had a broken flash player version bundled with it which experience "blue people" and "parts of flash content bleed out into other apps" bugs in case I have the hardware acceleration turned on in flash settings. Using a VDPAU_TRACE hack (Plagman, thanks once again for implementing and publishing it) it seems to be possible to fix this issues, but that does certainly change the flash behavior with regards to GPU usage and might be a relevant change that would prevent flash from triggering the "render corruption" bug.

All in all, I would have to do more observations and testing prior to uploading the problematic trace to nVIDIA, because currently reproducing the bug is not like "proceed with the following steps", but rather something like "dance with tambourine around the PC for about two hours on a new moon night". I'm still wondering how does it come to happen that I've been hitting this bug almost always when been doing my previous tests but now it seems to be quite a big problem triggering this one.
lexa2 is offline   Reply With Quote