00:38mupuf: imirkin: thanks for the link!
00:39mupuf: to the new repo of Ben
05:39Tom^: imirkin: is a piglit run anything of interest on the 780ti?
05:41Tom^: hm "skip: 8442" it skipped it all, something didnt go as planned.
05:49pmoreau: Look at one of the output file
05:50pmoreau: It could be it didn't found the tests, or failed to create an OpenGL context
05:52Tom^: yea i just missed the fact that it didnt only have python scripts, it had things that required compilation. :P
05:52pmoreau: Oops! O:-)
06:01hakzsam: Tom^, yeah, it's useful. Please use the piglit command line as described here http://people.freedesktop.org/~imirkin/
06:01Tom^: yea i did, im a third through it, i got 1 dmesg warn, 1 dmesg fail so far.
06:01Tom^: im gonna let it go through the entire before posting the results
06:03hakzsam: Tom^, cool, are you using mesa master?
06:05hakzsam: the results should be pretty similar to GK104 but that's fine to have them anyway
06:20sooda_: use this for gm* development: http://www.ebay.com/itm/lollersson/171971035032
06:42Tom^: imirkin: hakzsam , few crashes, few warns and errors in dmesg. https://gist.github.com/anonymous/b8041d127b7ec59ce463 (dmesg) . https://www.dropbox.com/s/6cjyoifwcyoqs9v/results.json.bz2?dl=0 this is the result.
06:43hakzsam: thanks, I'll have a look
06:43Tom^: total 30683 tested, 27535 passed, 414 fail, 4 crash, 2701 skip, 4 warn, 24 dmesg-warn, 1 dmesg-fail.
06:44hakzsam: those crashes are interesting and should not be hard to fix
06:45Tom^: cool :p
06:45Tom^: i could build mesa with debug symbols too if thats of any help.
06:46hakzsam: that will be definitely useful
06:46Tom^: will do and rerun then.
07:29hakzsam: Tom^, those crashes are expected actually and you won't see them with a mesa debug build ;)
07:29Tom^: aw :(
07:30hakzsam: you will get assertions instead
07:31Tom^: these mem out of bounds things is known too?
07:32hakzsam: these are frequent errors but they should be fixed
08:30jarnos: Does VDPAU and hardware deinterlace work by nouveau driver?
08:31jarnos: I tried nouveau with Mythtv, and at least hardware deinterlace does not work there unlike with Nvidia driver.
08:33jarnos: Though the Nvidia driver is not perfect either: some dark colors seem to show video behind them.
08:34jarnos: like they are transparent
08:38orbea: jarnos: assuming your card is compatible see the instructions under firmware https://wiki.freedesktop.org/nouveau/VideoAcceleration/
08:48jarnos: orbea, Chipset: G84 (NV84), NVIDIA Corporation G84GLM [Quadro FX 570M]
08:51orbea: It might work, im not a dev, just know where the link with the info is. You could try it, if hardware acceleration doesn't work just remove the firmware again
08:52orbea: and make sure you have vdpau
08:59jarnos: Does the G84 mean the same chip as NV84 in dmesg?
08:59imirkin: jarnos: the DEINTERLACE_TEMPORAL should be supported with nouveau-supplied vdpau
09:00imirkin: jarnos: in my experience that is more than sufficient. others have felt otherwise.
09:00imirkin: jarnos: NV84 = G84
09:00jarnos: orbea, I have not loaded any firmware manually. I am using Xubuntu 15.10.
09:01imirkin: you need firmware if you want vdpau
09:01jarnos: orbea, also the "Using VDPAU" test fails: it does not show any playback window.
09:01imirkin: the wiki page that orbea linked to should explain it all
09:01orbea: jarnos: listen to imirkin, he knows better than I do
09:31jarnos: imirkin, do I have to reboot after installing the firmware?
09:31imirkin: jarnos: no
09:53jarnos: I installed the firmwarea using the instructions in the www page, and installed mesa-vdpau-drivers 11.0.2-1ubuntu4ackage from standard ubuntu repository. No it plays by vdpau in mythtv, even if I put on advanced 2x hw deinterlace mode.
09:54jarnos: Or maybe it uses tempral 2x hw mode, if it can not use advanced, I do not know.
09:54jarnos: Some artifact are visible in playpack.
09:55imirkin: DEINTERLACE_TEMPORAL is supported, but DEINTERLACE_TEMPORAL_SPATIAL is not
09:55imirkin: you can check this with 'vdpauinfo'
09:55imirkin: h264 will show artifacts on some videos but not others
09:56imirkin: it seems like movie trailers are especially problematic, no idea why
09:56jarnos: imirkin, I am playing mpeg2
09:56imirkin: ah... mpeg2... was fine i thought. hm. maybe there were some final little bits i never fixed?
09:57imirkin: is this a dvd, or a dvb stream?
09:57imirkin: i do sorta remember some artifacts at various points in development, but iirc i fixed them. maybe not?
09:58imirkin: unfortunately vdpau is not well-suited to mpeg2 decoding on G84-G96
09:59jarnos: imirkin, dvb
09:59imirkin: xvmc is a much better match for the hw capabilities
09:59imirkin: but... no built-in deinterlace with xvmc
09:59imirkin: can't win 'em all :)
10:04jarnos: imirkin, oh, I don't know what is xvmc. I guess I can not use it with mythtv.
10:05jarnos: imirkin, my card is capable of A feature set: https://en.wikipedia.org/wiki/Nvidia_PureVideo#Nvidia_VDPAU_Feature_Sets
10:06jarnos: imirkin, at least with VDPAU, that is.
10:08Yoshimo: does nouveau support SLI with multiple cards?
10:10karlmag: I don't think there is sli support at all..
10:11karlmag: Unless someone did something lately and I didn't notice
10:11Yoshimo: so where is the difference between multicard and sli?
10:13karlmag: If I have understood SLI correctly; it's a way to make multiple identical (or near identical) cards work as one card (with the outputs of only one card actually giving any output), but the compute power of all cards going out out to those outputs. Multicard is just that. Multiple independant cards. Each with their own outputs.
10:14karlmag: Please someone correct me if what I'm saying is wrong.
10:15sooda_: afaik sli is more low-level than that. with two cards, you could issue render commands to both just fine (as you described), but sli at least historically worked at even the scanline level. the cards would have (nearly) identical memory contents, and one would render the upper half of the screen and the other would the lower half
10:16sooda_: with just two cards without sli support, you'd need to do this all the way from the application to use both cards. sli works transparently
10:17karlmag: sooda_: I guess you're right. I didn't take it in detail, admittedly.
10:17sooda_: the problem is that "work as one card" doesn't work just as-is
10:17sooda_: or it does, but then that's called sli :P
10:18karlmag: I guess it's oversimplifying it.
10:18jarnos: imirkin, what do you mean by it is not well-suited to mpeg2 decoding? By VDPAU, it can do partial acceleration for MPEG-2: "Partial acceleration means that VLD (bitstream) decoding is performed on the CPU, with the GPU only performing IDCT, motion compensation and deblocking. " Can it do full acceleration by xvmc?
10:18karlmag: Still, the driver needs to have support for doing it.
10:19karlmag: the FeatureMatrix page says; "SLI or even multicard setups are very rare among developers. You should start hacking with us, if you have such a setup." I think that still pretty much applies.
10:20karlmag: (note #6)
10:29imirkin: jarnos: vdpau provides the full stream. however G84 can't decode the full stream. so mesa has to the bitstream (VLD) decoding. mesa's core competency is GL, not mpeg2 stream decoding. so it doesn't do such a fantastic job at it.
10:29imirkin: jarnos: otoh xvmc is designed to take the output of VLD directly, which is precisely what the HW takes, so it's a much better match.
10:30imirkin: Yoshimo: SLI (nowadays) is a connector that allows multiple cards to "link up", and know which SLI ID they have (or master/slave, i forget). The command stream, in turn, can specify that a particular command should be executed on a particular SLI ID, or on all GPUs.
10:30imirkin: Yoshimo: nouveau has no support for such a thing
10:31imirkin: sooda: scanline is the original voodoo2 SLI
10:32sooda: yea, wasn't the name scan-line interleave or something?
10:32imirkin: yeah something like that
10:32imirkin: Yoshimo: the nvidia sli link protocol has been figured out, i believe, but there remains the large question of wtf to do with it. splitting a workload among multiple GPUs is not easy.
10:33sooda: i've heard internally that it's goddamn complicated
10:33sooda: and several odd places in the official driver source have these "loop over all sli cards and do magix"
10:35jarnos: imirkin, I installed this PPA https://launchpad.net/~oibaf/+archive/ubuntu/graphics-drivers
10:35imirkin: right... i mean you could split every draw into 2, or you could alternate draws (but then you have to track dependencies), and the more sophisticated GL stuff has all this coherency bs, etc
10:36imirkin: auto-splitting of workloads was a lot easier when all the rendering just happened on one framebuffer directly
10:36imirkin: but now it's next to impossible
10:36imirkin: jarnos: ok
10:36imirkin: jarnos: he talks about how to use xvmc in there, good
10:37imirkin: jarnos: most people see it as an outdated api, and many video players have chosen to drop support for it.
10:37imirkin: jarnos: mplayer, as always, works fine though :)
10:38jarnos: imirkin, I have 4 files /usr/lib/dri/libXvMCnouveau.so* ; Should I add them all to /etc/X11/XvMCConfig ?
10:38imirkin: jarnos: no, just one
10:38jarnos: imirkin, which?
10:38imirkin: the libXvMCnouveau.so one
10:39imirkin: i don't think an X server restart is required, as he claims...
10:42karlmag: Sounds like regular multicard support would be more useful (not sure if it would be easier though)
10:43imirkin: karlmag: regular multicard support exists
10:44karlmag: imirkin: to which degree? (aka; which limitations?)
10:44imirkin: maybe you and i have diff understanding of "multicard"
10:44karlmag: that could very well be
10:44karlmag: Maybe that should be defined first
10:44imirkin: but they come up, they're separate GPUs in your system, you can use them...
10:44karlmag: That is true.
10:44karlmag: I have done that.
10:45imirkin: with Xorg you can use prime or reverse-prime to offload rendering or "offload" crtc's between them
10:45karlmag: I guess I'm thinking; "will every card I put into my system automagically come up when I start X?"
10:46imirkin: "come up"?
10:46karlmag: be found and used
10:46imirkin: precision is hard :)
10:47sooda: how would you like them to be used? you can't, e.g., get a perf boost for a single app just because there are two processing units
10:48sooda: they have separate memories and all
10:48karlmag: If I put -say- three cards in my machine and connect monitors to all outputs (assuming there is a 1-1 of usable outputs and physical outputs), will I get picture on all my monitors when I start X without having to do anything with xorg.conf (or similar)
10:48karlmag: still not precise enough? :-P :-D
10:49imirkin: precise enough.
10:49imirkin: only monitors attached to the primary gpu will light up by default
10:49imirkin: you can then use xrandr to slave the additional crtc's
10:49sooda: i guess you could run separate X sessions on the cards too?
10:50karlmag: imirkin: so.. "no" (as the short answer to my question)
10:50imirkin: sooda: sure. or even on each crtc with zaphod heads :)
10:51sooda: i've got two gpus and three displays because the gpus can only drive two (but the binary driver, sorry about that :D)
10:53imirkin: sooda: given your workplace, seems reasonable :)
11:28phomes: No news about pstate on the G92 card. However, I have a FX 3800 (GT200GL) which hangs and shows a pattern of dots on the screen when going to the highest level. The two first levels work fine.
11:29imirkin: RSpliet: --^
11:29imirkin: phomes: he's going to want that card's vbios
11:35phomes: imirkin: I got one from /sys/kernel/debug/dri/0/vbios.rom. Should I mail it or put it online somewhere?
11:35imirkin: phomes: either :) or both :)
12:05karolherbst: now I will only send Pull requests through github :p
12:11karolherbst: on a side note: ben merged my sysfs removal stuff so nouveau.pstate=1 is not needed anymore with current master and the pstate file is inside debugfs now ;)
12:16pmoreau: karolherbst: PR for the kernel?
12:16karolherbst: pmoreau: well nouveau master is on github now, soo :D
12:17pmoreau: Yeah right
12:18pmoreau: But, some like having patches in their mailbox, and I'm not sure you can have that with GitHub PR.
12:22karolherbst: mhh you can actually
12:22karolherbst: you can tell github to send stuff to a ML
12:23karolherbst: but I doubt you can configure github to send only pull requests :/
12:24karolherbst: but you could add a webhook on pull requests and do a POST request against an url
12:24karolherbst: pmoreau: but I wasn't _that_ serious about it ;)
12:25karolherbst: but it would be actually nice to be able to make a pull request and the same patches land on the mailing list
13:05pmoreau: karolherbst: That would be great, I'm definitely in favor of that. :-)
13:05pmoreau: It's true that there is an API you can use to setup hooks and so on, I had forgotten.
13:06karolherbst: it is too messy currently anyway :/
13:06karolherbst: because you can't just tell github to send only for new/updated pull requests
13:07karolherbst: "Pull Request opened, closed, assigned, labeled, or synchronized." that's the farest you can go :/
13:07karolherbst: maybe you could read out the POST request and do stuff then, who knows
13:08karolherbst: we could add a travic-ci hook and let PRs be autocompiled :D
13:08pmoreau: Well, wouldn't opened and synchronized be enough?
13:08karolherbst: and compiled it with different compilers at once or something
13:08karolherbst: we would just get noise about closed ones
13:08karolherbst: so this also means a notification about merged ones
13:08karolherbst: you get either everything or nothing
13:09karolherbst: that's what I meant
13:09pmoreau: You can't put ifs in the hook?
13:09karolherbst: you can check what you want to have, and "pull requests" is one of them ;)
13:10pmoreau: Or else we add a man-in-the-middle which only let go the one we want
13:10pmoreau: (Maybe that's what you suggested with the travic-ci hook?)
13:10karolherbst: pmoreau: do you know if coverity can analyze kernel modules?
13:10karolherbst: noo, travis-ci is just a compile test
13:11karolherbst: it would set a status on a pull requests
13:11pmoreau: I have never used coverity
13:11karolherbst: either good or bad
13:11karolherbst: mhh k
13:12karolherbst: I once included travis-ci, coverity and coveralls in one project, was pretty nice
13:13karolherbst: pmoreau: ohh wait, there is also an IRC service which would notify about pull requests
13:13karolherbst: or new issues
14:43jarnos: imirkin, I could not play a file with te xmcp or something interface, even if I put the line in the file and rebooted.
14:49imirkin_: jarnos: did you use mplayer?
14:49imirkin_: jarnos: mplayer -vo xvmc foo.mpg should work
15:06jarnos: imirkin_, yes, it told "Error opening/initializing the selected video_out (-vo) device."
15:11hakzsam: imirkin_, hey, you did not bind the shader buffers for COMPUTE, that might explain the PAGE_NOT_PRESENT error I see in dmesg :)
15:33imirkin_: jarnos: maybe mplayer was compiled without xvmc support?
15:33imirkin_: hakzsam: maybe ;)
15:34imirkin_: hakzsam: all the compute stuff used funny paths, many of which i commented out
15:34imirkin_: hakzsam: note that there's a separate cp bufctx
15:34imirkin_: hakzsam: and basically a separate compute pipeline in nve4 as well
15:34imirkin_: in the nouveau driver
15:35imirkin_: so make sure you're binding stuff to the appropriate bufctx
15:36jarnos: imirkin_, I have no idea. I am using the mplayer2 from ubuntu universe repository
15:37imirkin_: jarnos: ah, not mplayer2. mplayer.
15:39jarnos: imirkin_, that is only virtual package on recent ubuntu releases: http://packages.ubuntu.com/search?suite=all§ion=all&arch=any&keywords=mplayer&searchon=names
15:55jarnos: imirkin_, VLC does not seem to have XVMC output support, but it has color ASCII art video output. It does not seem to be able to play video properly with any more usual video output with the nouveau driver.
15:55jarnos: usual > useful
16:20duelle: Hi, I currently try to get a multi screen setup running. I have a Lenovo T530 with Optimus (Intel HD4000 + Nvidia NVS5400). Trying to get it to work with 2 external screens (DVI @ nvidia) and the buildin Notebook screen. My first approach was to use reverse PRIME. This works fine until I want to arrange the screens using xrandr.
16:21imirkin_: what's the issue with arranging the screens using xrandr?
16:21duelle: Then both external screens show one identical part of the workspace, and the internal screen one. The third part is not displayed but there. I can move my mouse there.
16:22imirkin_: pastebin your xrandr command + the output of xrandr
16:23duelle: Additionally I get the following error message: xrandr exited with -1. Configure crtc 5 failed.
16:28imirkin_: duelle_: having trouble with pastebin?
16:29duelle_again: imirkin_: Sorry for the delay. Had entered a wrong command which restarted X.. Here are the pastes: http://pastebin.com/KrTKSW4t
16:29imirkin_: no worries
16:30duelle_again: That is the state before arranging the screens with xrandr
16:30imirkin_: duelle_again: ok, please provide the xrandr command you run to "rearrange" as well as the xrandr output afterwards
16:31duelle_again: Just one thing that irritates me a bit: The two ports that are "reverse-primed" are displayed as DP1-2 and DP1-3 although the screens are connected to DVI ports.
16:31imirkin_: probably active DP -> DVI converters
16:31imirkin_: or maybe even inactive ones
16:32imirkin_: consider yourself lucky you don't have one of the MST docks
16:32imirkin_: since MST isn't currently supported with nouveau :)
16:36duelle_again: Here's the next one: http://pastebin.com/FQDXi5QT
16:36duelle_again: What does MST stand for?
16:36imirkin_: multi-stream transport
16:37duelle_again: tbh there is a dock in place - but there is a DP/DVI port in parallel twice.
16:39imirkin_: okkkk.... i don't think you can do that :) my guess is that the --pos stuff is messing it up
16:39imirkin_: if you want to do that, start with --fb 1234x1234 or whatever
16:40imirkin_: also... those are some funny positions you set it to
16:40duelle_again: That is the part from arandr. So I didn't intentionally do such stuff. But when I tried to do it manually (with leftto and so on) it gave the same error message.
16:40imirkin_: i would highly recmomend just using --left-of and --right-of
16:41imirkin_: so here's what i would do
16:41duelle_again: ok, I will restart X and create a manual setting.. brb
16:41duelle_again: Oh kk
16:41imirkin_: turn everything off except LVDS1
16:41imirkin_: then turn on one screen
16:41imirkin_: then turn on the second screen
16:41imirkin_: as separate commands
16:41imirkin_: makes sense?
16:41imirkin_: i wonder if it's trying to allocate crtc's on intel
16:42duelle_again: Sounds good. If I understood it correctly before one should use the --auto command to use the max resolution possible?
16:42imirkin_: what version of xf86-video-intel do you have?
16:42imirkin_: xrandr --auto will do _some_ stuff, but i usually don't like it
16:42duelle_again: xf86-video-intel 1:2.99.917+519+g8229390-1
16:43duelle_again: I'll be right back trying to do it step by step.
16:45imirkin_: i actually gtg, not sure if i'll be on later
16:47duelle: Ok, back again, I'll post the pastebin in a minute.
16:47imirkin_: i actually gtg, not sure if i'll be on later
16:48duelle: imirkin_: Ok, np. Thanks for your help so far.
17:11imirkin: i'm guessing this is an issue with xf86-video-intel
17:11imirkin: ... and duelle is gone. o well.
17:37orbea: I tried to play metro 2030 with nouveau, got as far as starting a new game and then my whole screen went red, killed the game and then everything (Keyboard+mouse) froze. I was able to ssh in, but it got stuck on being rebooted, the mouse unfroze and the screen went black though.... I think this is the relevent Xorg log. http://dpaste.com/0TT3AKE
20:50karolherbst: yeah meh, nvidia-drivers-361 now depends on libglvnd :/
20:54skeggsb: why is that a bad thing?
20:58karolherbst: it doesn't work yet :D
20:58karolherbst: and libglvnd deploys its own libGL stuff and so on
21:09skeggsb: that's not really any different situation than before nvidia used libglvnd though ;)
21:09skeggsb: and when mesa has been updated to use it too, then, things will be better :)
21:10karolherbst: yeah, but I can't use the driver yet
21:10karolherbst: because the xserver also needs some changes
21:10karolherbst: currently glXChooseFBConfig just returns 0 :/
21:10skeggsb: ah, well, that's not ideal
21:10karolherbst: anyway my entire bumblebee setup doesn't seem to work yet
21:10karolherbst: even if I use the two required libraries from libglvnd
21:11karolherbst: ahh wait
21:11karolherbst: I didn't install the xorg extension
21:13karolherbst: okay, still I don't get any visuals or fbconfigs
21:14karolherbst: I bet I need those libglvnd libgl stuff too :/
21:14karolherbst: oh well
23:51hakzsam: imirkin_, nv50 has a sort of compute pipeline as well