00:50imirkin: vpelletier: well, you do get one interrupt... maybe the rearm logic isn't working properly. it's a 8-bit write to 88068 ... perhaps that works out poorly with the endian swap thing?
00:50imirkin: vpelletier: grab envytools and nvapeek it to see if it's working
00:50waltercool: guys, question ,what would that mean? disp: 0x6341: INIT_GENERIC_CONDITON: unknown 0x07 , since 4.9 or 4.10 I can't properly use external monitors (Maxwell2)
00:57skeggsb_: waltercool: i don't expect the two are related (though, it's possible).. i'm aware of that condition, but have nfi what is needed to implement it
00:58skeggsb_: waltercool: if you show me a kernel log of failing multiple monitors, that could be useful
00:58skeggsb_: "no fucking idea"
00:58waltercool: sure, let me fetch you that for you
01:00waltercool: some useful flag for nouveau to debug? drm.debug?
01:01skeggsb_: let's just see a normal log first, nothing else may be required
01:04waltercool: At least using KDE, using xrandr --auto, I didn't get any problem, but the external monitor just froze, but using ksystemsettings to manage the monitor, caused that error
01:05skeggsb_: ah, that's not at all what i was expecting to see
01:06skeggsb_: i'll try and have a play around this week and see if i can trigger something similar
01:07skeggsb_: imirkin: how goes bindless?
01:07waltercool: at least, under maxwell2, using systemsettings of KDE5, happens everytime
01:07skeggsb_: waltercool: it's a laptop, i presume?
01:07skeggsb_: with an intel gpu as primary?
01:08waltercool: yeah, I'm using modesetting for Intel GPU
01:08skeggsb_: ok. i can replicate a similar setup here, so, i'll see what i can come up with when i get a chance
01:08waltercool: Thanks :)
01:09waltercool: If you need any test, just ask me
01:09skeggsb_: you might want to try 4.11, or even better drm-next (one airlied has merged my pull request)
01:09waltercool: Do you need some freedesktop ticket?
01:09skeggsb_: yeah, that'd help to keep track of it
01:09waltercool: I'm using 4.11.4
01:09skeggsb_: ah, right
01:09skeggsb_: nvm then.
01:10waltercool: There is a current ticket for the monitor issue you mentioned?
01:11skeggsb_: your issue isn't related to that INIT_GENERIC_CONDITION thing
01:11skeggsb_: so, don't worry about that
01:11waltercool: oh, ok, so I just need to make a ticket about the stacktrace?
01:24waltercool: oh my bad, seems like I'm not using xf86-video-nouveau, let me try using it
01:29waltercool: well, it didn't made any difference
02:25imirkin: skeggsb_: didn't get time
02:30imirkin: skeggsb_: that tends to be the way things go with me nowadays, sadly =/
02:31skeggsb_: hm, maybe i'll take a look, it'll likely effect some of the things i'm working on too anyways
02:32skeggsb_: unless you think you'll get time sometime soon?
02:32airlied: skeggsb_: how is mt?
02:32skeggsb_: airlied: i'm going back and filling in the nv30/nv50 bits atm, and a couple of issues i spoke with imirkin about recently to solve still
02:32skeggsb_: still thinking about those while doing the mindless nv30/nv50 bits
02:33skeggsb_: airlied: i still need to find me some access to more "real" games to regression (and mt, if they support it) test against
02:34skeggsb_: i pinged daniels ages back as suggested, but no response
02:36airlied: skeggsb_: ah keep pinging him, he gets there eventually :)
02:57imirkin: skeggsb_: ping the feral guy too
02:58skeggsb_: imirkin: who would that be?
02:58imirkin: skeggsb_: https://www.feralinteractive.com/en/news/752/
02:58airlied: skeggsb_: I hope you've got a good internet connection :-P
02:59skeggsb_: airlied: managed to get 100Mbit, despite being on FTTN :P
02:59airlied: skeggsb_: quota?
02:59skeggsb_: the node is right outside my house, which helps
02:59airlied: ah cool, I nearly hit my 200GB limit last month
03:00skeggsb_: imirkin: thanks!
03:00airlied:is going to end up with nbn over hfc, so most likely absolutely no change, but I'll pay someone more
03:01skeggsb_: go australia! *sigh*
03:39imirkin:has gbit, theoretically
03:40imirkin: finally managed to get an apt in a fios building :)
03:42skeggsb_: imirkin: our government still believes 640KiB is enough for anyone
03:42imirkin: mmm... that sounds nice
03:42imirkin: i believe the definition of broadband in USA is >= 128kbit
03:45imirkin: oh wow, that's no longer the case
03:45imirkin: "... voted to change the definition of broadband by raising the minimum download speeds needed from 4Mbps to 25Mbps, and the minimum upload speed from 1Mbps to 3Mbps"
03:46skeggsb_: i'm not even sure what our minimum is... i believe NBN (our "brand new, shiny, modern, fast" national broadband network) is 12Mbps
03:46skeggsb_: that's not guaranteed though.. just the lowest tier
03:47imirkin: allegedly the fiber line in my apt runs directly to the CO, no aggregator in the building/etc. i was surprised.
03:47skeggsb_: the original idea was fiber everywhere... a change of government later, and it's a hodge-podge mix of things, with the intention of lowering costs.. that has actually lead to blowing out the original full-fiber budget anyway
03:57Namidairo: found the australian
03:58Horizon_Brave: evening eveyone
03:59Horizon_Brave: did I miss the deb 9 launch party? xD
04:10Horizon_Brave: :( doesn't paint the nouveau project in the best light....
04:11airlied: also recommends installing llvm for no reason
04:11skeggsb_: i don't think it says anything horribly unfair or inaccurate
04:14imirkin: i'd move the "(if compatible)" to the nvidia driver bit actually... heh
04:15airlied:is afraid to ask what kernel stretch is
04:15imirkin: 0.2.4.36 ;)
04:16imirkin:remembers when 2.4 came out...
04:16imirkin: those were the days... new USB stack, all kinds of goodies...
04:21Horizon_Brave: lol...stretch is totally cutting edge..
04:22Horizon_Brave: going to have to get used to thinking of stretch as stable...
04:22Tom^: Horizon_Brave: https://www.debian.org/News/2017/20170617.en.html , you atleast only got 1 year old kernel.
04:27Horizon_Brave: Tom^: lol yea like I said, we're riding the wave of new innovation
04:29Horizon_Brave: this bread tastes like I'm biting into pure salt...
07:51mangix: gnurou_: ping
09:17gnurou_: mangix: sorry, I wasn't sure what the latest status was - haven't you managed to get it to work?
09:27karolherbst: gnurou_: no, current status: no timeouts in dmesg, but nothing happens
09:28karolherbst: gnurou_: here is the full debug log: https://pastebin.com/nKVtm8Wt
11:40karolherbst: can somebody review the nouveau related patches of my precise series? imirkin or pmoreau? It's the only patches with lack a r-by, thanks
11:41pmoreau: karolherbst: I’ll have a look right now
11:41karolherbst: awesome, thanks .)
11:43pmoreau: Regarding SHA2017, I won’t be coming: I won't be able to take vacation days, and I come back from a conference in the USA around the 4th August.
11:44pmoreau: karolherbst: v2 is the latest version of your series?
11:44karolherbst: pmoreau: v3
11:45karolherbst: pmoreau: k
11:51pmoreau: Patch 10: one of the two new conditions is useless: tryADDToMADOrSAD is only called from handleADD
11:52pmoreau: I would remove the test from handleADD, and move the comment to tryADDToMADOrSAD
11:57pmoreau: karolherbst: With -^ changed, patch 10 Rb. Rb for patch 8 and 9.
12:01karolherbst: pmoreau: we need both checks though, we have to check the add and the mul
12:01karolherbst: maybe I got it wrong though
12:02karolherbst: but mh, I think doing that OP_SAD thing is still valid for precise instructions? dunno
12:07karolherbst: yeah, I don't see how the results precision is affected by SAD in any way
12:14pmoreau: karolherbst: Ok, I was wrong for patch 10. I hadn’t realised you were checking two different instructions.
12:15pmoreau: So, Rb for patches 8,9,10.
12:15karolherbst: nice thanks :)
14:11imirkin_: karolherbst: SAD = integer thing. sum of absolute differences.
14:12imirkin_: (at least i think it's an integer thing)
14:12imirkin_: actually it's probably not
14:12karolherbst: it's integer
14:12imirkin_: ok :)
14:12imirkin_: so precision isn't an issue
14:12imirkin_: it's just like |a-b| + |b-c|
14:12imirkin_: or some permutation of those variables
14:43dboyan_: imirkin_: I added a latency-aware scheduling policy, but it doesn't seem to improve a lot beyond what I had done yesterday
14:43dboyan_: Pushed at https://github.com/dboyan/mesa/tree/wip/sched2
14:43imirkin_: well, it's conceivable that you did it wrong :p
14:44imirkin_: have you looked at the output and verified that it looks the way you want
14:44dboyan_: not really, i will do that
14:44imirkin_: an important step imho
14:45dboyan_: i agree
19:00ojn: imirkin: what the fuck are you on about? Accusing me of not giving a shit abotu a bug because I didn't do the dirty work of 2 days worth of bisecting?
19:36karolherbst: ojn: don't insult people
19:38karolherbst: your phrasing is more than questionable already, either you get back to a normal tone and behave or you nobody will help you
19:38karolherbst: ojn: imirkin does everything in his spare time and please tell us what obligations any of us here have in helping you out?
19:47ojn: karolherbst: Got it. Take personal abuse on the mailing lists but don't talk back on IRC. Clear.
19:48ojn: imirkin: anyway, i'll bisect it down when I get a few spare days.
19:51karolherbst: ojn: nobody abused you
20:07ojn: Yeah, ok. I'm out.
20:14karolherbst1: imirkin: I found a shader which sometimes compiles differently
20:15karolherbst: but only when shader-db runs in parallel
20:24karolherbst: okay, seems like my fault
20:28karolherbst: pmoreau: when somebody adds a field to the instruction class, the init method needs to be adjusted otherwise the fields may contain random values
20:46imirkin_: i guess someone doesn't like it when people respond in kind? o well.
21:04koz_: imirkin_: Thanks for all the hard work you and the other nouveau devs do in their spare time.
21:04koz_: (because I don't think you hear this enough)
21:05imirkin_: you're welcome =]
21:21Echelon9: imirkin_: thanks also for all the work you and other nouveau devs provide in spare time
21:23koz_: Echelon9: We're gonna start a thanking trend, aren't we? :P
21:23Echelon9: I should be able to resolve another user's bug report on GP107 support tonight; so saving you an easier bug triage
21:24mangix: karolherbst1: do you happen to know which commit fixes the GR init issue?
21:27Echelon9: relative to the noise in some other IRC channels, it's a welcome change =]
21:27mangix: building skeggsb's repo has the issue but yours does not
21:41mangix: hmm two extra commits by gnurou
21:41mangix: those must be it
22:34mangix: alright great success
22:34mangix: i cherry-picked some commits from gnurou and one of them got GR working again
22:34mangix: now to figure out which...
23:12mangix: that commit and the previous two are needed
23:12mangix: what a silly bug...
23:19imirkin: karolherbst: pmoreau: skeggsb_: hakzsam: is there like a list of stuff that requires my attention?
23:19imirkin: skeggsb_: i just noticed i still have your patches in my tree for rects
23:20imirkin: skeggsb_: i think i'll just push them out. iirc i tested them a bit, and nothing exploded particularly loudly
23:20imirkin: i'll triple check just in case
23:20imirkin: (a bit)
23:34skeggsb_: imirkin: i know piglit is fine, beyond that i didn't test much.. i have access to a lot more things to test with now though thanks to feral/valve :)
23:34imirkin: ok, well i'll run through its paces right now and push them out. if you run into bugs later, you can send additional fixes.
23:35imirkin: i think it's fairly straightforward and is the sort of thing that would totally not work if it were outright broken
23:35skeggsb_: true enough
23:35imirkin: (i.e. visibly so)
23:35skeggsb_: it's particularly nice in that it makes piglit not destroy my GPUs :P
23:35imirkin: an added bonus =]
23:36imirkin: although ... i rather like it destroying your GPU - incentivizes you to write better recovery code
23:36mangix: skeggsb_: can you backport those fixes for 4.11?
23:36skeggsb_: i don't believe nvidia can even recover from the error thius triggers
23:36imirkin: skeggsb_: so ... you're saying that pci unbind device + rebind doesn't fix it?
23:38skeggsb_: no, that works, it's just an uncorrectable error that requires the channel to be killed.. but yes, with some more plumbing, userspace should be able to do that itself just like page faults
23:38imirkin: for now, i'd be happy enough with the box not dying unrecoverably, forget userspace
23:39skeggsb_: i can't recall, but i *think* once the kernel detects the hang, it actually recovers the rest of the gpu already, piglit still hangs because userspace doesn't notice and waits forever on stuff
23:40imirkin: could we just auto-kill userspace
23:40imirkin: unless it has set the "i know what i am doing" flag
23:40imirkin: like a SIGGPU or something :)
23:42Plagman_: skeggsb_: and if you find you need need other testcases that aren't valve or feral games, just let me know the name of the game and send me your steam account login
23:43skeggsb_: Plagman_: cool. thank you :)
23:55imirkin: skeggsb_: btw, any opinion on the issues with MSI + PPC G5 platform? unclear what the precise problem is, yet... do you know if it ever worked?