01:12 chomwitt: hi, in kubuntu 17.10 with nouveau (unloaded: modesetting, fbdev, vesa) on a gtx960 i cant play video (i see no image,just sound)
01:12 MichaelP: When running xf86-video-nouveau Plasma desktop freezes... Is a plasma or xf86-video-nouveau issue ?
01:13 imirkin: MichaelP: most likely the mesa driver
01:13 imirkin: (so ... none of the above)
01:14 imirkin: chomwitt: mmmm ... odd. sounds like a video decoding issue more than a nouveau one.
01:14 imirkin: MichaelP: basically a lot of recent software thinks it's hilarious to use opengl for regular 2d rendering. the nouveau opengl library is found ... wanting.
01:15 MichaelP: imirkin: running liri-shell with wayland same driver but no freezes...
01:15 imirkin: ok
01:16 MichaelP: so then guess i have to run the slower nvidia driver.
01:18 chomwitt: imirkin: thank for the tip
01:19 MichaelP: imirkin: nouveau opengl library is found ... wanting. .. Is that where the freeze is at ?
09:48 chomwitt: some steam games play in nouveau but the 'game' windows is like smaller or a portion of it is out of view
09:48 chomwitt: i tested that in kubuntu 17.10 ..
09:49 chomwitt: with GTX960
09:58 chomwitt: i checked the feature matrix, it seem that some basic functionality is in wip state.
10:00 chomwitt: video decoding accel is TODO for NV110 , maybe thats why vlc must be set to 'opengl' in video settings to see 'video' ?
10:04 aphirst: chomwitt, video rendering and video decoding are not quite the same thing
10:07 chomwitt: aphirst: is there a bot to ask about that ?
10:07 chomwitt: so not seeing video is not nouveua fault?
10:25 chomwitt: Why in feature matrix there is a NVFO column when in codenames page is displayed as subcategory of NVE0 ?
10:26 chomwitt:thinks i should better buy a used kepler gtx780 ..
10:27 chomwitt:...or Titan! wow
10:32 chomwitt: i've websearching.. has nvidia released the signed firmware for 900series ?
10:55 iterati: chomwitt: why buy nvidia card at all ? If you stuck with one already so be it, but if you want a new one better go for AMD, if you want to use it on Linux.
11:08 pmoreau: chomwitt: Still not for the PMU engine (allowing us to change the fan speed), but they have for the graphics engine (for hardware graphics switching).
14:20 chomwitt: i see. bit by bit..
14:20 imirkin: chomwitt: i'd recommend using 'xv' for video output
14:20 imirkin: (although perhaps it doesn't work with xf86-video-modesetting? i'd recommend using xf86-video-nouveau though)
16:23 Lekensteyn: imirkin: I have no new information to work with (and solve that power on problem), but karol might be on something
16:28 imirkin: Lekensteyn: ok
17:51 dagb: good evening.
17:52 dagb: Lyude: will the new clock-gating patches have any effect on Quadro K1000M (GK107)?
17:59 imirkin: skeggsb: interesting. mpv full-screen playback with opengl is very glitchy with depth=30. no such issues with a depth=24 screen.
18:12 orbea: mpv usability has regressed recently somewhere, it tends to crash the system even without hwdec...
18:12 orbea: seen apitrace do the same crash too...
18:13 orbea: dmesg for the apitrace crash http://dpaste.com/3663CVK
18:17 imirkin: well, it's smooth with depth=24, and super-glitchy with depth=30
18:17 imirkin: for a specific video
19:37 Lyude: dagb: yep
19:37 Lyude: they should work on any kepler, but if they don't I'd be interested to know :)
19:47 RSpliet: Lyude: I look forward to giving them a spin myself, but unfortunately my GK107 has never been super stable in the first place. I might not get anything useful out of it...
19:51 imirkin: Lyude: doh, i even have your patches in-tree, but forgot to enable them. should it work on a GK208?
19:51 imirkin: (not saying it doesn't, just asking)
19:51 Lyude: imirkin: yep
19:54 dagb: Lyude: great! Thank you!
19:55 dagb: My HP Battleship (8570w) will make good use of those patches, I hope.
20:00 dagb: Lyude: you don't happen to have a patch around that applies cleanly to vanilla 4.15?
20:00 imirkin: dagb: i'd recommend grabbing skeggsb's linux-4.16 branch
20:00 imirkin: it's 4.15 + all the drm stuff for 4.16
20:01 dagb: imirkin: can do. thanks
20:01 Lyude: mhm; which isn't a whole ton afaik other then powergting + fixes you probably want anyway
20:01 imirkin: dagb: https://github.com/skeggsb/linux/commits/linux-4.16
20:04 dagb: pulling it now
20:05 imirkin: should be a minor delta on top of a pre-existing tree
20:05 imirkin: just add it as a remote
20:06 dagb: imirkin: I don't have a local git repo of the kernel
20:06 imirkin: o
20:06 dagb: But no problem.
20:07 dagb: It is unclear to me if Bens tree is synced up to 4.15 as released
20:07 imirkin: it's not
20:07 imirkin: it's on top of airlied's drm-next tree
20:08 imirkin: which in turn has only had 4.15-rc8 merged in
20:09 dagb: ok
20:14 dagb: is it possible to pull in Linus' tree on top of Ben's? I have only scratched the surface of git.
20:14 imirkin: git remote add linus ....
20:15 imirkin: git fetch linus
20:15 imirkin: git merge v4.15
20:15 imirkin: or something
21:07 dagb: Lyude: built a kernel with all the goodies. running with nouveau as a module, not built-in. Is 'nouveau.config=NvPmEnableGating=3' the right boot parameter?
21:07 Lyude: dagb: replace 3 with 1
21:08 dagb: will do
21:09 Lyude: dagb: you should also see a message in your dmesg that says clockgating was enabled if it works
21:10 dagb: Lyude: yes, was looking for that, don't see it
21:10 Lyude: mind showing me your dmesg?
21:10 dagb: hence me asking if the syntax is right for a standalone module
21:10 dagb: Lyude: sure. stand by
21:11 dagb: https://pastebin.com/bVqQNwsV
21:12 dagb: Lyude: ^
21:12 Lyude: dagb: that should be right, let me double check and make sure I didn't do anything wrong in that patchset that might not make it get enabled
21:13 dagb: Lyude: thanks. No rush on my side.
21:13 Lyude: sanity check: you're sure you're running the right kernel?
21:14 dagb: Lyude: fair question. Provided imirkins instructions were correct *and* I managed to follow them, yes.
21:14 imirkin: uname -a
21:14 imirkin: should say 4.15.0-rc8+
21:14 Lyude: also just double checked, everything in nouveau's code looks fine
21:14 dagb: imirkin: not after I merged Linus 4.15...
21:14 imirkin: ;)
21:14 dagb: Linux hp8570w 4.15.0+
21:15 dagb: Linux hp8570w 4.15.0+ #1 SMP Sun Feb 4 21:26:23 CET 2018
21:16 imirkin: Lyude: is something supposed to get printed?
21:16 Lyude: dagb: mind checking the git tree you're on and making sure you can see the commits for clockgating? the only think I can think of is that you're running a kernel without the changes
21:16 Lyude: imirkin: yep
21:16 imirkin: how are you loading the nouveau module?
21:16 imirkin: is it autoloaded, or are you using modprobe, or insmod?
21:17 imirkin: cat /sys/module/nouveau/parameters/config
21:17 dagb: https://pastebin.com/bQY7Hp8T
21:17 dagb: imirkin: autoloading
21:17 imirkin: ok, then it should be fine
21:18 dagb: NvPmEnableGating=1
21:18 imirkin: yeah. good.
21:19 dagb: what is a quick git command to list the relevant configs?
21:19 dagb: commits
21:19 imirkin: git grep NvPmEnableGating drivers/gpu/drm/nouveau
21:21 Lyude: btw, make sure you have the 5th patch which actually enables the
21:21 Lyude: *actually adds the NvPmEnableGating parameter
21:22 dagb: Lyude: I pulled Ben's tree
21:22 dagb: imirkin: that command quickly scrolls a blank page
21:22 imirkin: no results? that means you're missing the commits
21:22 imirkin: oh
21:22 imirkin: git clone https://github.com/skeggsb/linux.git
21:22 imirkin: you missed a spot.
21:22 imirkin: use the linux-4.16 branch
21:23 imirkin: the master branch is set to something silly
21:24 dagb: imirkin: meh.
21:25 dagb: OK. The pillow is calling. Will continue tomorrow. Thank you for all your assistance so far.
21:25 dagb: Lyude: and for the work leading up to me being here tonight. :-)
21:25 Lyude: np! :)
21:25 Lyude: thank mupuf as well
21:26 Lyude: they did a lot of the initial research
21:42 imirkin: skeggsb: you may be interested to know that my tv's display_info.bpc works out to 12.
21:52 imirkin: skeggsb: might be worth adding 12 and 16 as possible bpc depths
21:52 imirkin: although should probably check how those shake out on pre-GF110 display blocks
21:52 skeggsb: imirkin: yeah, i've got code pending that adds additional stuff to newer display classes
21:53 skeggsb: i haven't added that, but, i can
21:53 skeggsb: if you want to test it afterwards :P
21:53 imirkin: currently i'm adding 10bpc dither
21:53 imirkin: since i think right now my 30bpp output is still getting dithered to 8bpc
21:54 skeggsb: yeah, that's likely, i don't think 10 was added until 0x907d
21:56 imirkin: indeed
21:56 skeggsb: will probably won't some extra HeadSetControlOutputResource stuff too
21:56 skeggsb: want*
21:56 imirkin: nah, that stuff's there
21:56 imirkin: (well, should get it for the depth > 10 stuff, but ... meh)
21:57 skeggsb: i need to see if we can/should do it for tmds too under some circumstances
21:57 imirkin: well, hdmi == tmds
21:57 imirkin: so yes.
21:57 imirkin: i'm connecting via hdmi btw
21:57 skeggsb: tmds is more restrictive in what it accepts vs DP, but i'll check it out
21:58 imirkin: i'm a bit ... unclear ... on all the details of pixel clocks vs bpc
21:58 imirkin: you should just get RH to buy you a (small) deep color TV - i doubt they're too expensive
22:17 imirkin: skeggsb: well that is just surprising
22:18 imirkin: i added 10bpc support
22:18 imirkin: (for dithering)
22:18 imirkin: and flip it on when display bpc >= 10
22:18 imirkin: end result? the 10-bit gradient looks like it's 8-bit
22:18 imirkin: when i flip the dithering to 8bpp, the 10-bit gradient looks properly displayed
22:19 imirkin: i guess you're right - i need to mess with the controls
23:03 imirkin: skeggsb: looks like you need something more than just setting the depth
23:03 imirkin: no errors, but the screen never came up
23:03 imirkin: (and then i had the genius idea of trying depth = 6. that went REAL well.)
23:05 skeggsb: imirkin: gimme a sec, i want to check something
23:05 imirkin: skeggsb: also gf119_something had to be fixed up
23:05 skeggsb: yeah, head_state() ?
23:05 imirkin: it was trying to look up based on depth, and depth 8 being unrecognized, it was real happy
23:05 imirkin: something like that
23:06 imirkin: i wiped the changes, but i could find it if you need
23:06 skeggsb: yeah, it wouldn't have mattered except for avoiding the WARN() in this case anyway :)
23:06 imirkin: and i'll hold on my 10bpc dithering thing - that only makes sense when we have a more grown up way of looking at this stuff
23:06 imirkin: a *lot* of WARN
23:10 RSpliet: skeggsb: I erm... will see if I can look into your request without too much trouble. I'll have to figure out whether I can toggle NvMemExec during run-time w/o unloading the module (we should do right?)
23:11 skeggsb: RSpliet: does the machine survive? if so, just leave it set and do boot->0x07->0x0f and let it crash, then collect the logs :P
23:11 RSpliet: It doesn't unfortunately...
23:12 RSpliet: and journalctl took away my cheeky option of tail -f /var/log/messages over SSH
23:12 skeggsb: or... "LD_LIBRARY_PATH=/path/to/nouveau/lib ./bin/nv_init -c NvMemExec=1,NvClkMode=0x07", and repeat for NvMemExec=0,NvClkMode=0x0f
23:13 skeggsb: that's what i do, personally
23:13 RSpliet: That's nifty, but never played with that so I suspect it might take a tad longer to get that built and ready to go
23:16 skeggsb: as long as you've got envytools installed, it should just be a matter of running make :P
23:17 skeggsb: (it uses envyas to build ucode)
23:21 RSpliet: Well, installed is a strong word. I have it residing on the HDD somewhere, bet it won't take more than an env variable to make it build then
23:22 imirkin: usually it's specified so that sticking ENVYAS into the env will do the strick
23:22 imirkin: trick*
23:22 imirkin: or make ENVYAS=foo
23:22 imirkin: Makefile:ENVYAS ?= envyas
23:22 imirkin: so yeah, should work
23:48 Lyude: so this happened https://www.golem.de/news/nouveau-nvidia-versucht-alles-zu-verstecken-1802-132571.html
23:48 Lyude: i guess fossdem must have went well
23:54 imirkin: in case anyone's curious, 30bpp support summary / update: https://lists.freedesktop.org/archives/nouveau/2018-February/029742.html
23:56 airlied: imirkin: I assume you know ville has a bunch of infoframe stuff for hdr
23:57 airlied:isn't really sure how it all slots together yet :)\
23:57 imirkin: airlied: i was not really aware, i was surprised at the lack of stuff in drm_edid.c
23:57 airlied: https://github.com/vsyrjala/linux/commits/hdr_poc has some core drm patch in there
23:57 imirkin: airlied: but i don't know if we're hitting those problems right now in nouveau
23:58 imirkin: oh wait, yeah, i did know about those. but i totally forgot :)
23:58 skeggsb: imirkin: it's potentially why your monitor won't come up with 36bpp sor depth?
23:59 imirkin: skeggsb: mmmmmaybe. i think it's more likely that it's because the pixclk is wrong