00:21daemoz: Hello all.
09:25pdobrogost: Hi all!
09:30pdobrogost: I'm on Fedora 25 KDE with nvidia card and xorg-x11-drv-nouveau.x86_64 1:1.0.14-2.fc25. Recently I witness strange behavior; GUI "lags" but only on non-main display. Is it a known issue?
09:32karolherbst: pdobrogost: do you have xorg-x11-server 1.19.3-4 installed?
09:33pdobrogost: I have 1.19.3-1.fc25
09:33skeggsb: karolherbst: that change didn't get made in f25
09:33skeggsb: but, nice try ;)
09:34pdobrogost: What change? Any way to fix this in F25?
09:34skeggsb: what card did you have, and what other updates did you pull in?
09:36pdobrogost: I'm fully updated as of today, the card is GeForce GTX 750
09:38skeggsb:thinks we can blame 1.0.14 potentially
09:38skeggsb: can you create an /etc/X11/xorg.conf/d/00-modesetting.conf with:
09:38skeggsb: Section "Device" Identifier "modesetting-device" Driver "modesetting"
09:38skeggsb: err, that messed up the formatting :P
09:39skeggsb: but, it'll switch you back to the modesetting driver, which you were using before 1.0.14
09:40skeggsb: Section/Identifier/Driver/EndSection should each start on a new line btw
09:41skeggsb: my copy+paste messed up
09:46pdobrogost: 1.0.14? That's pretty old comparing to 1.19, no?
09:46skeggsb: talking about xorg-x11-drv-nouveau
09:47skeggsb: 1.0.14 added acceleration support for your board, whereas previously it'd fallback to the xserver's built-in 'modesetting' driver
09:47skeggsb: (which is accelerated using the GL driver)
09:48pdobrogost: Currently I have "Device" section with Option "DRI" "3" option. Should I leave it?
09:48skeggsb: just add: Driver "modesetting" to that
09:51pdobrogost: skeggsb: Thanks, this fixed issue. Is there any bug for this?
09:51skeggsb: no, not really... it's a tricky issue
09:51skeggsb: in f26/rawhide, i recently changed the xserver's default driver for new nvidia GPUs to modesetting
09:52skeggsb: which is what karolherbst was referring to when he asked your xserver version :P
09:53skeggsb: i'm undecided whether doing the same for f25 is a good idea or not yet
09:53skeggsb: when we did that switch for intel, we left it for >=f26
09:55pdobrogost: Is the problem with acceleration in nouveau diagnosed?
09:56skeggsb: pdobrogost: no, and it's not really worth the effort, -modesetting accelerates a lot more stuff than the nouveau 2d driver does, and is a lot better maintained
09:56skeggsb: which is why i made fedora default to it instead
09:57skeggsb: we can't really accelerate any more than we do in the nouveau 2d driver anyway, as we're limited to what EXA (an acceleration architecture in the xserver) allows. the modesetting driver replaces all that
09:58pdobrogost: Does it mean nouveau should not be used anymore?
09:58skeggsb: nouveau is more than the 2d driver (you're still using the nouveau kernel module, and 3d driver). but yes, i personally believe it should be dropped in favour of modesetting
09:59skeggsb: some others disagree :)
09:59skeggsb: mostly the argument is that the 3d driver is more complicated, hence more bugs, and the 2d driver is simpler
10:04pdobrogost: What uses 3d driver? Web browsers? Anything else? I guess not Plasma?
10:04skeggsb: if you're using the modesetting driver, your 2d rendering is accelerated with the 3d driver instead
10:05pdobrogost: It seems like 2d/3d separation should long be gone and there should only be 3d...
10:06skeggsb: a lot of hardware has gone that way already
10:06skeggsb: fwiw, we used the 3d engine in the 2d driver :P
10:06pdobrogost: Used? Does it mean you no longer do?
10:06skeggsb: bad choice of words
10:30bloblo: some help for 9800gtx+
10:39karolherbst_: skeggsb: but nvidia hw still has dedicated 2D engines? I don't really know anything here
10:41skeggsb: karolherbst_: yeah, it does. though, not terribly useful for the sorts of stuff people want to do today
10:41skeggsb: we use it for solid fills and straight blits (sometimes) still
10:42karolherbst: ahh I see
10:46bloblo: hi, i have gts 250, i have 1k fps glxgears with nouveau and 20k with nvidia 340, 0f is working if i change value g84.c 94 to 92, but i have 1,5-1,9k fps only, and pci link stay 2.5GT
10:46bloblo: nouveau support 9800gtx+ (gts 250) ?
10:47karolherbst: bloblo: check glxinfo
10:47karolherbst: but, I think reclocking doesn't really work on g92
10:47karolherbst: I tried it once as well and the results were bad
10:48bloblo: Vendor: nouveau (0x10de) Device: NV92 (0x615) Version: 13.0.6 OpenGL renderer string: Gallium 0.4 on NV92
10:48bloblo: i am currently on 4.9
10:49bloblo: but test result gived is from 4.1x for fps
10:49karolherbst: yeah well, reclocking for g92 is disabled for a reason and I doubt you actually reclocked the GPU by going to 0f
10:51bloblo: by default is 0f not work but with change value 84.c "94" to "92" 0f is working
10:51bloblo: but pci stay in 2.5gt
10:52karolherbst: you think it works, but you can't just enable stuff for a chipset and assume it is working magically
10:52karolherbst: pci might stay 2.5, because you need a newer kernel
10:52pq: also not good: a) using glxgears as any sort of indication of "performance", and b) framerates beyond display capabilities, say, 200.
10:52karolherbst: 4.11 or 4.12?
10:53bloblo: 4.10.13 and 4.11
10:53karolherbst: then 4.12 it is
10:53karolherbst: or 4.11? dunno, I'll check
10:53karolherbst: but then your board also needs to support something higher
10:54pq: bloblo, actually that does show a bug: glxgears should report your monitor refresh rate as fps. It's going way too fast.
10:54karolherbst: okay, should work with 4.11
10:54karolherbst: bloblo: you should see some messages regarding pci in dmesg maybe
10:54bloblo: @pq is without vertsync
10:54karolherbst: there are many reasons why the GPU might stay at 2.5 with 4.11
10:55karolherbst: bloblo: on an X level or application level?
10:55bloblo: i need compil 4.11 for test
10:55pq: bloblo, are actually interested in the swapbuffers() overhead rather than GPU performance?
10:57bloblo: i go build 4.11, for files g84.c i change 94 to 92 or i change nothing ?
10:57skeggsb: last i checked, the biggest difference in us vs nvidia with glxgears is that they do the SwapBuffers() in the rendering channel, we do it from X - which, leads to lots and lots of GPU context swtiches
10:57skeggsb: that are slow..
10:57karolherbst: but we should still get more fps than 1k :p
10:57skeggsb: not a real issue, for real apps, the difference there would be rather minimal
10:58bloblo: with gtx 770 i have 17k with nouveau and +-22k i think with nvidia
10:58skeggsb: it means: render; wait_for_idle(); swap(); wait_for_idle(); render...
11:01bloblo: @pq i dont know about swapbuffers
11:01karolherbst: with 4.10 there is something odd with the rendering syncing anyway...
11:01bloblo: @pq how enable/disable
11:02pq: bloblo, nevermind. I thought that you though the framerates you got with glxgears were an indication of... well, anything. They mean nothing really.
11:03karolherbst: well reclocking with G92 simply doesn't work... so
11:04karolherbst: no idea why G92 are so different compared to G94 here, but they seem to be
11:04pq: unless you are very specifically looking at a very tiny portion (swap buffers) of the graphics stack, which a) not comparable between nouveau and nvidia, and b) not a problem outside microbenchmarks designed to measure exactly that (glxgears).
11:05karolherbst: well, the fps of glxgears still gets drastically affected by the perf state in nouveau
11:05bloblo: i can test in game difference
11:05karolherbst: pq: on Prime system glxgears is only usefull for a PCIe bandwith benchmark though
11:58eau3x6: what might be the reason why nouveau fails to poweroff discrete card on optimus system? in my case this doesn't work reliably from boot to boot, i.e. sometimes the nvdia card is powered on despite no program makes use of it.
11:59eau3x6: 1:DIS: :DynPwr:0000:01:00.0
12:00eau3x6: often this doesn't change to DynOff on idle
12:01eau3x6: (debian testing, no bumblebee)
12:02karolherbst: eau3x6: something in dmesg?
12:02karolherbst: and you should check the clients file in debugfs /sys/kernel/debug/dri/1/clients
12:02eau3x6: karolherbst: yes nouveau is logging in dmesg when it activates/deactivates gpu
12:03karolherbst: I meant more like in the case when it doesn't
12:03eau3x6: clients file is empty (only header line)
12:04eau3x6: command pid dev master a uid magic
12:06eau3x6: dmesg just shows that nouveau "resumed" card when I started user Xorg session
12:07eau3x6: it was deactivated prior to that (some time after booting but before starting graphical session)
12:07eau3x6: i have some ACPI errors too
12:18eau3x6: the key point is that dynamic power management does work *sometimes* for me, but I have no idea what is the reason for that. my card is nvc0 (gf108) btw
12:18eau3x6: is it normal that card is activated on starting Xorg session?
12:19eau3x6: something is poking it...
12:25eau3x6: also xrandr --listproviders shows only intel card "Provider 0: id: 0x8f cap: 0xf, Source Output, Sink Output, Source Offload, Sink Offload crtcs: 2 outputs: 4 associated providers: 0 name:modesetting"
12:26karolherbst: eau3x6: I guess some applications polls some status or so
12:28eau3x6: shouldn't there be two providers in xrandr --listproviders output?
12:33eau3x6: can't find anything in google about this acpi errors. i'm afraid i might be the only person using this 2010 laptop :)
12:36karolherbst: eau3x6: the providers are just there for displays
13:14eau3x6: brb rebooting
13:18eau3x6: i've did a warm reboot and now it works properly: card is deactivated on boot, then it's activated when starting Xorg session and then (after ~5 sec) deactivated again.
13:52bloblo: for gts 250 i have build 4.11 with and without modify here log: https://pastebin.com/7Bj53QFB
14:16bloblo: yop, gts 250 here log nvidia https://pastebin.com/QkzBk3zm
17:38lesderid: what might be causing DRI3 and DRI2 to be unavailable? glxinfo shows the direct rendering and nouveau (gallium) renderer
17:47lesderid: nvm, was an Xorg config problem
17:51jamm: dboyan_: congrats!
17:52gnarface: lesderid: what did you have to change to fix that?
17:52lesderid: I'm not sure, I was asking for a friend
17:53lesderid: he said he used his original nvidia proprietary driver Xorg config, but with the nouveau driver
17:53lesderid: I imagine a mininal config would probably have worked too
21:38jamm: hakzsam: 6/8 shaders with sched codes: https://hastebin.com/abiciyirol.bash please reviewer whenever you could. I'll continue working on the other two. Cheers :)
21:38hakzsam: jamm: hi
21:39hakzsam: can you send me a patch instead?
21:39jamm: ah, sure
21:39jamm: i'm trying to figure out how to test these before i send it over
21:39hakzsam: okay cool
21:39jamm: i compiled and installed envyas, and assembled the shaders
21:40hakzsam: I can't have a look right now, during the weekend for sure
21:41jamm: no worries :)
21:41hakzsam: but it will be easier to review with an email :)
21:45karolherbst: imirkin: you know why the egl context creation fails in the khronos CTS with nouveau?
21:46karolherbst: ohhh, I am silly
21:47karolherbst: nice, seems like the CTS runs without problems
21:47hakzsam: karolherbst: btw, I'm going to push a fix for the heaven regression
21:47karolherbst: hakzsam: I saw
21:47karolherbst: hakzsam: saw my SHA email?
21:48karolherbst: I really want to do something Nouveau related there, because I am sure there are a lot of people interested into that project outside the "normal" Linux/X11 community
21:48hakzsam: karolherbst: yes
21:49hakzsam: not sure if I will attend though
21:49hakzsam: are you going to give a talk?
21:50karolherbst: well, there are different kind of talks
21:50karolherbst: I actually plan to do self-organized session(s)
21:50karolherbst: and that can be a talk or a workshop
21:50karolherbst: stuff like that
21:50karolherbst: if you are interested, I can add you to the trello board I've created to organize everything
21:50hakzsam: okay cool
21:51hakzsam: yeah, please add me
21:53karolherbst: I want to register the "village" soon (1-2 weeks), so that at least this is done
21:53karolherbst: and maybe we get others wanting to help out as well
21:53karolherbst: ohhh and I need that live image from pmoreau :p
21:56karolherbst: huh, don't we have transform_feedback_overflow_query implemented?
21:57karolherbst: ohhh, that isn't transform_feedback
22:04karolherbst: hakzsam: feel free to do anything on that board, I am sure I already forgot important things