#asahi-gpu · 2023-03-10 04:55
linaMesa sends me emails ^^#asahi-gpu · 2023-03-10 04:55
alyssa:-D#asahi-gpu · 2023-03-10 04:55
linaThat too!#asahi-gpu · 2023-03-10 04:55
alyssano mesa review?#asahi-gpu · 2023-03-10 04:55
alyssaah#asahi-gpu · 2023-03-10 04:55
linaM2 Pro/Max support and keep the review emails going ^^#asahi-gpu · 2023-03-10 04:52
alyssabtw what's your plan for the next [short term]?#asahi-gpu · 2023-03-10 04:52
alyssalina: awesome :)#asahi-gpu · 2023-03-10 04:10
lina*black#asahi-gpu · 2023-03-10 04:10
linaYou'd have to get some stderr output from that to debug it#asahi-gpu · 2023-03-10 04:10
linaPlasma going back usually means kwin_wayland crashed and restarted, the recovery from that is... not great#asahi-gpu · 2023-03-10 04:09
linajannau: I haven't stress tested 13.2 so it's totally possible that is broken in new and exciting ways ^^#asahi-gpu · 2023-03-10 04:07
linaAnd yeah, I was thinking of just adding it to the device params so mesa can optimize based on it#asahi-gpu · 2023-03-10 04:07
linaalyssa: Yes, it ignores writes#asahi-gpu · 2023-03-09 19:22
jannauso far no crashes on 12.3 and no pink boxes on G13C#asahi-gpu · 2023-03-09 19:13
jannaureproduce, insofar an application bug as the black screen is the result of a missing plasmashell, not sure why. nothing suspicious in dmesg#asahi-gpu · 2023-03-09 19:05
jannaumoving to top left corner displayed still the overview effect and pressing on the power button showed the logout screen#asahi-gpu · 2023-03-09 19:04
jannauno crashes with 12.3 so far, but I ended 2 times with a black desktop. it was flickering 2 or 3 times to black and stayed black then. Could easily be a application bug. desktop and panel were black but mouse pointer was still present and moving#asahi-gpu · 2023-03-09 18:53
jannauI was trying to reproduce the pink boxes during plasma startup so it didn't crash with 12.3 FW on j493 and j375#asahi-gpu · 2023-03-09 18:44
jannauthis is on the 13.2 firmware so it might not be a regression#asahi-gpu · 2023-03-09 18:43
jannaulina: rtkit crash dump: https://www.jannau.net/asahi/gpu_crash_agx_release_2023-03-9.log#asahi-gpu · 2023-03-09 18:41
jannauj314c (m1 max), firmware 13.2, kernel https://github.com/jannau/linux/tree/asahi-6.2-1_star_dcp_v13_2 (roughly gpu/explicit-sync + dcp 13.2) and alyssa's agx/release#asahi-gpu · 2023-03-09 18:40
alyssajannau: dmesg?#asahi-gpu · 2023-03-09 18:40
alyssawomp womp#asahi-gpu · 2023-03-09 18:38
jannaulina: I managed to crash the GPU (rtkit crash) with quick sddm - plasma/wayland login/logoff cycles#asahi-gpu · 2023-03-09 16:01
alyssaOtherwise you don't#asahi-gpu · 2023-03-09 16:01
alyssaI think that's a reasonable tradeoff. If you boot with the magic kernel modparam, you get a better debug experience at the cost of some performance and no sparse texture support#asahi-gpu · 2023-03-09 15:59
alyssa2. insert bounds checking code in a bunch of places#asahi-gpu · 2023-03-09 15:58
alyssa1. advertise sparse support#asahi-gpu · 2023-03-09 15:58
alyssaand then based on that mesa can decide whether to#asahi-gpu · 2023-03-09 15:58
alyssaI think that's ok, just means we need a param added to the UAPI `bool page_faults_supported/suppressed/robust/whatever`#asahi-gpu · 2023-03-09 15:57
alyssabe possible"#asahi-gpu · 2023-03-09 15:57
alyssafound a way to change it after initial load unfortunately, but it might#asahi-gpu · 2023-03-09 15:57
alyssa"it's a kernel module param, GPU global and I haven't#asahi-gpu · 2023-03-09 15:57
alyssaDoes it also ignore writes in this case?#asahi-gpu · 2023-03-09 15:57
alyssaand silently return all-00 on reads for shader accesses"#asahi-gpu · 2023-03-09 15:57
alyssalina: "On the other hand, macOS configures things so page faults are ignored#asahi-gpu · 2023-03-09 15:00
linaeric_engestrom: It was actually panfrost, but I just removed that and replaced it with a mention of xe_drm.h when I did the DRM submission because I don't think there's any of the original header left (I pretty much rewrote the entire UAPI with the explicit sync changes)#asahi-gpu · 2023-03-08 10:21
eric_engestromlina: just noticed asahi_drm.h has the comment "Based on asahi_drm.h" 🔁#asahi-gpu · 2023-03-08 10:21
eric_engestromlina: given the copyright on the next line (and its years) I'm guessing this should say "Based on v3d_drm.h" ?#asahi-gpu · 2023-03-08 08:37
jannaulina: it's the only thing I spotted but I've done only light testing. beside the repeaated login/logout I did only a glmark2 run, with rendering but I didn't looked constantly at it#asahi-gpu · 2023-03-08 02:42
linaIf it's just that though... I think at this point we should just release and see if anyone else finds something more serious...#asahi-gpu · 2023-03-08 02:41
linajannau: I thought I saw that once during a stream but I couldn't reproduce it and I kept doubting myself!#asahi-gpu · 2023-03-07 20:51
jannaunot reproducible on the m1 ultra in over 30 tries with the exact same m1n1/kernel/mesa#asahi-gpu · 2023-03-07 20:41
jannauno, still reproducible later. so maybe reproducible in 10 to 15%#asahi-gpu · 2023-03-07 20:33
jannaureproduces with ~50% in the first 3 login attempts after boot and seems to be very unlikely to reproduce after that (3 repeats)#asahi-gpu · 2023-03-07 20:30
jannauhttps://www.jannau.net/asahi/2023-03-07_gpu-explicit-sync-agx-release_m2.mov#asahi-gpu · 2023-03-07 20:16
jannaureproducible, boxes, maybe 16x8, 1 or 2 frames#asahi-gpu · 2023-03-07 20:13
alyssaNoooo#asahi-gpu · 2023-03-07 20:06
jannaupink/violet boxes/triangles during plasma/wayland login on m2, very briefly