[Mesa-dev] Initial version of threaded GL dispatch V2

Timothy Arceri tarceri at itsqueeze.com
Thu Mar 9 12:05:48 UTC 2017



On 09/03/17 21:46, Mike Lothian wrote:
> Hi
>
> I've been testing this patch series along with patchwork
> series https://patchwork.freedesktop.org/series/20872/
> & https://patchwork.freedesktop.org/series/20873/
>
> It's been working well with the games I've tested on so far - I had one
> crash in the Shadow of Mordor benchmark, but I've reproduced that in
> master without these applied - just need to track the issue down

I suspect that is a very hard to reproduce bug that seems to happen with 
the shader cache enabled. I did manage to get a stack trace of it but it 
showed a crash in one of the games threads. However the feral dev's seem 
to think it could still be a Mesa stomping on the thread. We need a 
reliable way to reproduce to be able to say for certain.

>
> I'm currently running a full piglit test on both Skylake & Tonga - I'll
> let you know the results

Cool, thanks.

>
> Mike
>
>
>
> On Tue, 7 Mar 2017 at 06:22 Timothy Arceri <tarceri at itsqueeze.com
> <mailto:tarceri at itsqueeze.com>> wrote:
>
>     Since Marek is busy with other things at the moment I thought I'd
>     try keep the momentum going on getting the initial implementation
>     committed.
>
>     Gregory Hainaut has fixed the piglit crashes in the previous
>     version, and I've addressed the review feedback. I've also pushed
>     some standalone patches.
>
>     The only thing I didn't address was adding EGL/GBM support as that
>     code is unfamiliar to me and it looks like we can pass a -plaform=
>     arg to piglit to get waffle to use a different platform for testing
>     anyway (also Gregory did some basic EGL testing with PCSX2).
>
>     One thing that still concerns me is the use of _mesa_glthread_finish()
>     This seems to assume that the context itself is only accessed by a
>     single thread. See patch 14, 22 & 24 as an example of what I mean. I'm
>     not sure how this should be solved (assuming its a real problem).
>
>     Please review.
>
>     _______________________________________________
>     mesa-dev mailing list
>     mesa-dev at lists.freedesktop.org <mailto:mesa-dev at lists.freedesktop.org>
>     https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>


More information about the mesa-dev mailing list