ChanServ changed the topic of #etnaviv to: #etnaviv - the home of the reverse-engineered Vivante GPU driver - Logs https://oftc.irclog.whitequark.org/etnaviv
pcercuei has quit [Quit: dodo]
dv_ has quit [Ping timeout: 480 seconds]
dv_ has joined #etnaviv
pcercuei has joined #etnaviv
DPA- has joined #etnaviv
DPA- has quit []
DPA- has joined #etnaviv
DPA has quit [Ping timeout: 480 seconds]
mntmn has quit [Ping timeout: 480 seconds]
mntmn has joined #etnaviv
dv_ has quit []
dv_ has joined #etnaviv
dv_ has quit []
dv_ has joined #etnaviv
JohnnyonFlame has joined #etnaviv
pcercuei has quit [Ping timeout: 480 seconds]
pcercuei has joined #etnaviv
JohnnyonFlame has quit [Ping timeout: 480 seconds]
otavio__ has joined #etnaviv
otavio_ has quit [Read error: Connection reset by peer]
<sravn> Is it correct understood that the ARB_shadow lowering patch enables GPU acceleration with chrome on GC880 (which is halti == 0)?
<austriancoder> sravn: not sure what the requirements for Chrome are.
<sravn> austriancoder: In another thread Stefan (@gizmo9) indicated that chrome would disable hwacc if OpenGL 2.1 was not supported
<sravn> And if I read the ARB_shadow patch correct gc880 will now have OpenGL 2.1 support
<sravn> So I hope we are good here.
<sravn> I need to go back and look af how we start chrome on out target, as we may override something.
<sravn> I hope it helps as performance on i.MX6 SoloCore (with GC880) is bad, and if this can help then hurray!
<cphealy> sravn: what are you using for compositing on the i.MX6 Solo? X? Wayland? Other? From a performance standpoint, are you able to hit GPU composition bypass and instead scanout the contents of Chrome buffer on one of the IPU HW planes?
<cphealy> What I've found in the past is that under many use cases, even if dealing with a single fullscreen app, the compositor (Wayland in my case) will still use the 3D GPU to composite the single fullscreen surface before scanout. If this is the case, there is an opportunity for increased performance by bypassing the unnecessary composition blit operations.
<cphealy> With the GPU and display controller in the i.MX6 Solo, the GPU renders to a tiled buffer format but the IPU does not support this format. As such, the GPU needs to be told to perform a detile operation at the end of rendering a frame so the frame can be directly scanned out.
<sravn> cphealy: chrome running ozone on top of weston
<cphealy> What version of Weston are you running?
<cphealy> Are you running Chrome fullscreen?
<sravn> We have pengutronix to help us and there was a compositor bypass patch that we dropped because colors was wrong.
<sravn> I am not really involved in these parts today so not really on top of things here.
<cphealy> Got it. Getting compositor bypass working is likely something that would provide significant performance improvement if you can get the colors fixed.
<sravn> cphealy: weston 8.0.0 and running chrome in fullscreen (kiosk) mode
<cphealy> Also, there's a new wayland protocol feature which may help provide some performance improvement. dmabuf_feedback. With this, it should be possible to have detiling performed on render instead of on scanout, thus reducing the amount of detiling the GPU performs when rendering at < VSYNC rate.
<cphealy> I think the team at Pengutronix is aware of this opportunity.
<sravn> I will points the right guys at the fine patch from austriancoder but as it superseedes a patch from Zable I guess they are aware.
<sravn> Sounds promising, will include the dmabuf_feedback info too just to make sure it is considered.
pcercuei has quit [Quit: dodo]