ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard + Bifrost + Valhall - Logs https://oftc.irclog.whitequark.org/panfrost
<mntirc> anarsoul: as far as i can see it's gnome-shell's built in compositor.
<mntirc> but i noticed now that the issue is most pronounced here because one of my displays is rotated and at scale 2, and when a window at the border between the two displays is repainted, it slows down everything. (not yet overlapping, but there is probably some invisible overlapping area)
<mntirc> ok, the problem is gtk3. it's reproducible with the gtk3-demo's "fishbowl" demo. resizing that demo to a big size will drag down the whole desktop's performance to <10fps. but the same demo in gtk4-demo stays at 120 fps (it's a 120hz screen)
Daanct12 has joined #panfrost
rcf has quit [Quit: WeeChat 4.4.4]
rcf has joined #panfrost
Daanct12 has quit [Ping timeout: 480 seconds]
Daanct12 has joined #panfrost
Daanct12 has quit [Quit: WeeChat 4.5.1]
Daanct12 has joined #panfrost
ity has quit [Remote host closed the connection]
ity has joined #panfrost
lucas_ has joined #panfrost
lcagustini has quit [Ping timeout: 480 seconds]
urja has quit [Read error: Connection reset by peer]
urja has joined #panfrost
rasterman has joined #panfrost
ity has quit [Quit: WeeChat 4.5.1]
ity has joined #panfrost
<mntirc> after some discussion with rmader he pointed out that mutter somehow picks desktop GL on panfrost when it should use GLES. using COGL_DRIVER=gles2 fixes the perf issues
rasterman has quit [Quit: Gettin' stinky!]
Daanct12 has quit [Ping timeout: 480 seconds]
<mntirc> btw apparently there is a panfrost regression for a311d (mali g-57) that makes the full screen flicker on certain workloads, contributors bisected and reverted this commit to fix it https://source.mnt.re/reform/reform-debian-packages/-/merge_requests/85/diffs
<mntirc> now i have some funky bugs with firefox 135 on wayland mutter + latest git mesa + rk3588, where it often fails to finish drawing tiles, and those remain transparent, leaking parts of desktop through and messing with damage tracking (cursor trails)
Lucretia has joined #panfrost
kusma has joined #panfrost
warpme has joined #panfrost
warpme has quit []
warpme has joined #panfrost
sally has quit []
rasterman has joined #panfrost
warpme has quit []
paulk-bis has joined #panfrost
paulk has quit [Ping timeout: 480 seconds]
CME_ has joined #panfrost
CME has quit [Ping timeout: 480 seconds]
sally has joined #panfrost
rasterman has quit [Quit: Gettin' stinky!]
cphealy_ has quit []
cphealy has joined #panfrost
<cphealy> mntirc: Unless I'm mistaken, the Amlogic A311D has a Mali G52 (Bifrost) and not the G57 (Valhall).
<mntirc> cphealy: you're right, my memory failed me there.
francis[m] has joined #panfrost
<cphealy> mntirc: regarding the patch that fix you pointed out above which reverts the addition of 32x8 AFBC, thinking out loud, I wonder if this is resulting in a path which forces GPU composition or bouncing between GPU and display plane composition. Historically, Panfrost has only supported 16x16 tiling when writing out AFBC for scanout. The GPU HW supports both 16x16 and 32x8 (and possibly others depending on generation). The
<cphealy> addition of 32x8 is to add support for HW platforms which only support 32x8 AFBC tile layout in the scanout plane HW.
Lucretia has quit [Remote host closed the connection]