16:39 <marex> lynxeye: so you're now looking into the etnaviv threaded GL slowness ?

16:45 <lynxeye> marex: Not right now, as I'm still busy with other things, but it's very much on top of my "do very soon" list.

16:45 <marex> lynxeye: the "do very soon" list is the one which gets postponed for many months ? :)

16:46 <marex> lynxeye: or is it the one below that, priority-wise

16:47 <lynxeye> marex: Nope that one isn't the postpone list, as it's hurting things I care about enough that I can justify spending time on it. ;)

16:48 <marex> ha

16:48 <marex> lynxeye: things which use threaded GL or not ?

16:49 <mntmn> marex: which apps are using threaded gl?

16:49 <marex> mntmn: chromium,qtwebengine

16:51 <lynxeye> marex: Both. We are oversyncing for threaded GL, but the resource tracking overhead is also very visible in singlethreaded apps.

16:51 <marex> lynxeye: I know

16:52 <marex> lynxeye: maybe what we really need is the batching

16:54 <lynxeye> marex: I don't think so. The context/cmdstream stuff should work pretty well if we avoid shooting ourselves in the foot as much as we do by oversyncing and all the locks required to make that work.

16:55 <lynxeye> I already have a battle plan on how to change things, just need some time to type it out and properly test it.

16:57 <marex> lynxeye: well, let me know once you have it, I have a feeling I already have tests which will break it

16:58 <mntmn> ah, interesting