13:41apteryx: hi! looking at perf data throuh the lense of 'hotspot', I see that 3D rendering on my 'nouveau' system appears ot use 'crocus_dri.so', which appears to be an intel-related driver. Is nouveau supposed to make use of that?
13:42MrCooper: nope
13:42MrCooper: either it's actually using the Intel GPU, or crocus_dri.so might be a red herring
13:43tnt: is it by anychance some laptop with intel+nvidia ?
13:43MrCooper: (all *_dri.so are hardlinks to a single file)
13:43apteryx: tnt: nope, a desktop with no integrated video and an old nvidia 8800 GTS discrete card
13:47apteryx: I thought it only happened with Qt 6 but any application making use of OpenGL seems to go through it
13:48apteryx: the hotspot flamegraph looks like this: https://pasteboard.co/1XncWlg6W9e3.jpg
13:48apteryx: this is for the mupen64plus emulator
13:51apteryx: it's spending 12.5% of processing time in crocus_dri.so
13:59zmike: I would strace the app or check LD_DEBUG info to see which driver is actually loaded
14:00MrCooper: there's a nouveau symbol on top of crocus_dri.so, so it's most certainly actually using nouveau
14:00MrCooper: probably something picks up the red-herring crocus_dri.so hardlink for some reason
14:00alice: the perils of hardlinks
14:01ccr: lordi - hard link hallelujah
14:02MrCooper: I'd try running the app with LIBGL_DEBUG=verbose
14:04MrCooper: that should say which Mesa driver is used
14:30apteryx: MrCooper: thanks, I'll try in a minute. here's another flamegraph from jami, which made it all the most alarming :-) https://pasteboard.co/q9ycN8BMeNrK.jpg
14:33apteryx: the LIBGL_DEBUG=verbose trick prints this: https://paste.debian.net/1316620/
14:33apteryx: looks like it's dlopen'ing nouveau
14:42apteryx: are video artifacts when using opengl backend with Qt 6 expected for the nouveau driver?
15:00apteryx: here's what it looks like in jami, the only application that exhibits the problem (the only one rendering via opengl that I use, I think): https://pasteboard.co/K5yVQlk3lccR.png
15:00apteryx: the contacts list is blurred by myself for privacy, but the rest is qt/nouveau's doing
15:06apteryx: better one, with more details: https://pasteboard.co/KRnGN58EedC8.png
15:06apteryx: you can see some of the text labels are corrupted, and there's a big white colored triangle covering half of the chat area
21:44mareko: robclark: a630 seems to behave randomly: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/58615905
22:05DavidHeidelberg: mareko: the failure was kinda OK, but the previous one was broken machine
22:06DavidHeidelberg: robclark: if someone can in the morning disconnect the cheza-3 from the CI that would be great, another failure on 3: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/58615732
22:06DavidHeidelberg: *Monday morning