01:44 alyssa: Kayden: congratulations :)
03:00 janesma: anarsoul: https://gitlab.freedesktop.org/mesa/mesa/issues/2410
03:00 gitbot: Mesa issue 2410 in mesa "[regression][bdw,hsw,bsw,gen9,icl] vulkancts and crucible tests crashing on mesa=3c754900b5f689b821cbb1162b821ce12bcaf23e" [Bisected, Regression, Vulkan, Opened]
03:00 janesma: unless you have an immediate idea of what is going on, we will revert the patch soon. It prevents intel platforms from completing vulkancts
03:29 janesma: anarsoul: https://paste.centos.org/view/27511476
03:37 anarsoul: janesma: do you have a backtrace?
03:37 janesma: https://paste.centos.org/view/7741f9b1
03:37 janesma: anarsoul: jekstrand already has a fix that I am sending through intel ci
03:38 anarsoul: OK, thanks
03:39 Kayden: (anv not initializing nir_builder quite early enough...doh)
10:29 doras: agd5f_: any chance to get this reviewed https://patchwork.kernel.org/patch/11318791/
10:29 doras: ?
10:31 doras: Resuming from suspend doesn't work for me without it
13:58 kherbst: ehh, we don't implement all spir-v 1.1 features :(
14:12 dj-death: is there a tool take look at different in failures of piglit runs?
14:12 kherbst: dj-death: piglit summaries?
14:13 kherbst: piglit-summary-html.py is the command
14:18 dj-death: kherbst: it doesn't really say that it supports 2 reports
14:19 kherbst: you can add multiple reports
14:19 dj-death: oh yeah
14:19 dj-death:reads README
14:19 dj-death: thanks!
14:42 imirkin: dj-death: it produces something like this: https://people.freedesktop.org/~imirkin/all-comparison/problems.html
14:42 imirkin: there's also a text summary one, which produces something simpler, i think it's used by CI
14:42 imirkin: which is just like testname: pass/fail
14:47 dj-death: imirkin: thanks, seen that before, but didn't know it was piglit
15:25 tomeu: seanpaul: have you seen that drmhwc can call drmWaitVBlank before a mode has been set? and then it hits WARN_ON(!vop->is_enabled) in vop_crtc_enable_vblank
15:25 seanpaul: tomeu: can't say that i have, however i don't think i've ever tried drmhwc on rk :-)
15:25 seanpaul: jstultz: ^^
15:26 tomeu: ok, that's just some verbosity on the serial console, but thought it could be worth mentioning
15:27 tomeu: more important is that the monitor isn't liking the mode that is being set
15:31 vsyrjala: sounds like a driver bug if vblank enable gets called on a disabled crtcs. the drm_vblank_on/off stuff is supposed to prevent that
15:41 tomeu: ah, interesting
16:29 Venemo: jekstrand: what's up with MR 2339? you mention in there that it causes some regressions but you didn't go into too much detail about that
16:38 jekstrand: Venemo: It causes some perf regressions. I don't remember what all regressed.
16:38 jekstrand: Venemo: At least on Intel, it can end up squashing too much stuff towards the top so our scheduler doesn't get the chance to space things out to avoid latency.
16:39 Venemo: that's interesting
16:39 Venemo: I wonder how it would affect us
16:40 jekstrand: No idea
16:40 jekstrand: It should be functionally correct though bnieuwenhuizen had a suggestion IIRC
16:40 Venemo: could it be toned down to be less aggressive so that it doesn't hurt perf on intel?
16:40 jekstrand: oh, no, it was dschuermann
16:40 jekstrand: Venemo: possibly
16:41 jekstrand: Venemo: But it's kind-of tricky to know exactly how to do that because it moves control-flow and typically we don't schedule stuff past control-flow.
16:41 Venemo: mhm
16:42 Venemo: maybe worth trying to see what it does to aco
16:42 Venemo: what caught my eye is the mention that it could help dxvk
19:16 jstultz: tomeu: so yes! that's been a recent issue. On the boards I see, there seems to be some trouble w/ hotplug, or getting spurious hotplug events that causes surfaceflinger to get confused. its been happening since android10 merged, and there was bunch of new vsync logic that landed then..
19:17 tomeu: jstultz: ok, I'm going to have to look at why the kms driver is setting a wrong clock after bootanim
19:17 tomeu: after a long weekend, though :)
19:18 jstultz: tomeu: i've not been able to focus enough to really run down where the problem is, but it could well be something where drmhwc isn't doing quite what surfaceflinger now expects, or just a chance in behavior that is bringing out bugs in the hdmi bridge driver I'm using.
19:19 jstultz: (adv7511 has some spurious hotplug notification issues, but the driver did work ok prior to the android10 merge)
19:28 dj-death: does the same rules apply to crucible as for piglit regarding MRs?
19:28 dj-death: aka, nobody looks at it for 2weeks and you can merge it
19:33 bnieuwenhuizen: dj-death: what do you need reviewed?
20:01 dj-death: bnieuwenhuizen: oh you reviewed it :)
20:01 dj-death: thanks!
22:10 kisak: what's going on with https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3376 ? Marge-bot waiting on a CI run on a MR with CI disabled?
22:10 gitbot: Mesa issue (Merge request) 3376 in mesa "intel/compiler: Clear accumulator register before EOT (Gen12 WA)" [Intel-Fs, Intel-Vec4, Opened]
22:18 anholt_: did someone disable ci in their repo or something?
22:22 anholt_: fixed their repo
22:27 anholt_: and unwedged marge
22:33 anholt_: and filed a bug on marge
22:56 anholt_: krh, robclark: did fd-farm get unplugged or something?
22:56 robclark: probably not intentionally.. fritz has been working on the new farm
22:57 anholt_: yeah, seems to be down. if someone could take a look, pipelines are blocked rightn ow
23:01 anholt_: ok, looks like it's back and unblocked
23:32 anholt_: so, apparently fd-farm was only not suspending due to being logged in at the desktop, and once it logged out it started autosuspending. that should hopefully be fixed now, apologies for the backup on the CI pipelines.
23:39 robclark: heh, lol.. ok
23:40 robclark: I'll disconnect the monitor.. let me know if anything goes badly