01:40gears_: java: ../../nouveau/pushbuf.c:727: nouveau_pushbuf_data: Assertion `kref' failed.
01:41imirkin: probably due to multithreading ... i have some patches that you can try, if you're interested
01:41gears_: imirkin: me?
01:42gears_: It's a crash that happens when I try to run a Minecraft client, hah.
01:44GEARSOMG: imirkin: Hey, back. That time it crashed my PC
09:02inglor: So is the reclocking for GTX 690 supported now? :))
09:26pmoreau: inglor: You can try https://github.com/karolherbst/nouveau/tree/stable_reclocking_kepler_v5
09:26pmoreau: Dunno if 690 is a special case or not, but Kepler cards should reclock nicely with that branch
09:52karolherbst: isnt 690 dual core
09:52karolherbst: dual gpu
09:52karolherbst: yeah, 2x gk104
09:53karolherbst: inglor: well you will only be able to use one of the "gpus" at once, because nouveau needs SLI support to use both
09:53karolherbst: inglor: but, maybe with prime offloading you might be able to use one for desktop one for gaming or something
09:53karolherbst: and like run one core at lowest clock (desktop) and the other on highest
09:55karolherbst: pmoreau: do you think this might work that way?
09:55karolherbst: but I fear that the stuff will be transported over pcie too in this case :/
09:56pmoreau: I have no idea how the two chips are advertised to the system.
09:57karolherbst: as seperated devices afaik
09:58karolherbst: they even get their own debugfs entries and stuff
09:58Yoshimo: SLI wasn't supported yet, was it?
10:00karolherbst: no, but I think the main reason is, that no nouveau dev has the card(s) for that
10:00karolherbst: I think SLI is mainly done in userspace
10:01karolherbst: and the SLI connection is just faster than doing it over PCIe
10:01karolherbst: I doubt there is a lot of thinks to do here
10:02karolherbst: well, speeding up prime offloading for sli connection gpus might be a good first task to SLI OpenGL support
10:38Lekensteyn: Should the video card always be powered up when a register is read from the HDMI audio controller? Or would it be better to leave the video card suspended and just fail the HDA register reads?
10:38Lekensteyn: I'm trying to figure out how vga_switcheroo should handle this, maybe l1k has an idea?
10:43Lekensteyn: The current implemented policy is that the video card dictates the power state and that the HDMI audio device has to adapt (even if it is seemingly in use)
10:49lawll: what prominent improvements do we find for nouveau in mesa 12.0?
10:49lawll: will it improve ppsspp and dolphin emulators?
11:02Yoshimo: higher OpenGL levels
11:09karolherbst: lawll: there are some patches to reduce CPU overhead, which got triggered in pscx2
11:09karolherbst: I think some of them landed for 12.0 already
11:09karolherbst: imirkin: how is the state with the work regarding pcsx2?
11:20gregory38: karolherbst: lawll: to be fair, PCSX2 will be faster when PCSX2 will push all gl commands into a separate threads
11:20gregory38: Otherwise, I think there is a bad performance hit for the upload of texture
11:20gregory38: My code is using PBO, and as far as I understand, it will setup a GPU draw call to do the transfer
11:21gregory38: I don't know if it is the way to program the DMA of the GPU or to do some parallel transfer
11:21gregory38: But sometimes, I transfer small textures (or even palette 1x256 or 1x16 texels)
11:22karolherbst: well, to be fair, I never had big performance problems with dolphin and nouveau
11:22karolherbst: never tried ppsspp though
11:23gregory38: because the "big" part is the multithread stuff
11:23gregory38: which I think dolphin does already
11:23karolherbst: they do
11:24karolherbst: but for pcsx2 the situation should be better on 12 compared to 11.x right?
11:26gregory38: give me 5 minutes
11:26gregory38: I will do a quick benchmark
11:27karolherbst: but somehow I get the feeling, that dolphin is the most active developed emulator anyway
11:28karolherbst: stuff like fullHD + 4x SSAA usually works with a lot of games and the binary driver without perf problems
11:28gregory38: I'm mostly alone on PCSX2, so yeah there are more active
11:29karolherbst: I compared to like every console emulator
11:30gregory38: yeah, don't know the others but lots of people are on dolphin
11:30gregory38: good for them, bad for others :)
11:30gregory38: I didn't compare the free driver is upscaling
11:30gregory38: with upscaling
11:30gregory38: but I'm feeling that it won't be too far of the nvidia blob
11:31karolherbst: nice :)
11:31gregory38: but I need to profile it :)
11:32gregory38: so bench 11.1 = 70, 12 (git actually) = 113 on test 1
11:32gregory38: test2, 11.1 = 64, 12 = 105
11:34karolherbst: but the newest 11 release is 11.2.2
11:34gregory38: so yes it is much faster and lighter on the CPU side
11:34karolherbst: but I doubt there is much work done there
11:34gregory38: dunno, if the reported silly validation for ssbo and image was fixed in 11.1.3
11:35gregory38: Note I have 11.1.3 (debian)
11:35karolherbst: no idea
11:35karolherbst: do you know if it is possible to profile waits?
11:35gregory38: what do you mean ?
11:36karolherbst: threads waiting on a lock for example
11:36gregory38: with vtune maybe ?
11:37karolherbst: lawll: in the end, you should just test it out ;)
11:38karolherbst: but chances are good, that for CPU limited usecase it might be better
11:38lawll: ppsspp is also under active development
12:05l1k: Lekensteyn: apologies, I was AFK and am reading your message only now. r u there?
12:06Akien: Running The Talos Principle on intel hd 4000 in Vulkan mode
12:06Akien: Never thought I could harness that from such an IGP :)
12:38Lekensteyn: l1k: yes here
12:39l1k: Lekensteyn: I've started to write you an e-mail with cc to dri-devel because that's an important question, please stand by
12:40Lekensteyn: if it is related to audio, maybe cc alsa-devel too (subscribe to post without being held in the queue)
13:16l1k: Lekenstey: you have mail ;)
13:16l1k: Lekensteyn: ^
13:18Lekensteyn: got it, reading :)
13:24Lekensteyn: l1k: btw, your clock seems off by 5 minutes, time to poke NTP?
13:24l1k: haha yes I know :)
13:25karolherbst: well my system syncs at least once a day :p
13:26l1k: better now.
13:33l1k: Lekensteyn: I'll be AFK for an hour or so but will keep lurking here...
13:34Lekensteyn: l1k: ok, I'll be checking the options and try to draft a reply to your mail
14:12Lekensteyn: hmm, hit an earlier deadlock again by writing "detect" to /sys/class/drm/card1-HDMI-A-1/status while runtime suspended...
14:14Lekensteyn: and apparently the sound subsystem likes to stay in an uninterruptable state (snd_power_wait) which makes speaker-test unkillable..
14:20karolherbst: my feeling is, that a lot of stuff stays uninterruptable, allthough they shouldn't
14:22karolherbst: in my opinion everything should be interruptable by default, except there is a very important reason not to (like significant performance overhead)
19:21mupuf_: imirkin: nice of you to have spent the time to open up the glxinfo page
19:22imirkin: not entirely altruistic... less work for me ;)
19:30Yoshimo: it also helps to cover a few more cards i guess
21:30orbea: imirkin: you said you have some patches that might help the concurrent vdpau and/or gl calls crashes (As with mpv)? Or did I misread?
21:31imirkin: you didn't misread, but you may have misunderstood
21:31imirkin: in actuality my patches don't help at all for vdpau
21:31imirkin: because i'm lazy
21:31imirkin: and trying to get the underlying model fully right
21:32imirkin: before sprinkling the video stuff with locks
21:32orbea: oh, well if you have anything you want tested I might be able to help :)
21:32imirkin: i did make a branch of that work at https://github.com/imirkin/mesa/commits/locking
21:32imirkin: but it's highly unlikely that will help with vdpau + gl
21:32imirkin: in fact it's semi-likely to hurt, since i never even tested that
21:32orbea: im not sure I have anything that is just gl + gl
21:33imirkin: well, a bunch of games are moving to it, which is what lit a fire under my ass
21:38orbea:imagines a bunch of stream games made by devs who did not even consider nouveau :/
21:41imirkin: well, e.g. warsow has an explicit force-disable for multithreading for nouveau
21:41imirkin: a bunch of steam games happen to time stuff so that it works out on nouveau most of the time but hardly all
21:42imirkin: like grid autosport
21:42vita_cell: steam games I player worked fine, except: Metro, Serious-Sam 3
21:43vita_cell: DIRT showdown, crashes after manu
21:43orbea: pcsx2 has a multithreading too, but its only enabled with nvidia drivers. Its supposed to have big improvements with some games and nvidia, I wonder if that would hold true for nouveau too in time
22:01karolherbst: orbea: I am sure it hasnt for OpenGL
22:01karolherbst: what you mean is hardware emulation
22:01karolherbst: most likely
22:41orbea: karolherbst: yea, I use the opengl hardware renderer with pcsx2
22:42karolherbst: thats not what I meant
22:45orbea: sorry, im not following then
22:48imirkin: pcsx2, while it might use multithreading, does not do GL things from multiple threads, or if it does, it interlocks those accesses.
22:49orbea: ah, I dont know much more than it being a nvidia only feature