00:00imirkin: the crtc would need to be configured accordingly, obviously
00:00imirkin: with some way to indicate to it "scan out *now*"
00:01imirkin: and some way to indicate to the monitor that a frame is starting. or perhaps it's just a reduced vsync time and it just has a pll which can lock onto it.
06:22mupuf: imirkin: they usually prefer talking about how long vblank can be stretched
06:23mupuf: you need to program the HW so as it knows how long the vblank can be, then there is a bit that can cancel it
06:24mupuf: so if you do page flipping from the host side then just flip to the new fb, then toggle the bit and off you go
06:24mupuf: as simple as that
06:25mupuf: the problem is not there, it's the userspace that needs to know about this because that's a significant change in ABI
12:59imirkin: mupuf: stretching vblank == delaying next frame
13:07mupuf: imirkin: yep
13:11imirkin: mupuf: anyways, makes sense that hw should be aware of this up-front
13:11mupuf: imirkin: the sensible way is to have the HW be responsible for enforcing the timings, yeah
13:11mupuf: and just have an early-fire mechanism
13:12mupuf: and even the early-fire is not gonna trigger before the minimum vblank time
13:13mupuf: no idea how nvidia implemented it... but it is hard to design anything else but this
13:25imirkin: yeah, well, all we need is a GSync-capable monitor to test it out
13:25imirkin: i'm sure those will be popular in another 20-50 years :p
13:42RSpliet: mupuf: if we can stretch vblank long enough we can get all monitors in vblank, then reclock memory... surely that'll solve embarrasment in a live reclocking demo :-P
13:42karolherbst: ... I have such a display...
13:43karolherbst: that laptop I bought with a gm204 has a gsync displays as well
13:43mupuf: imirkin: it's possible the HW actually would support VRR
13:43mupuf: that would be a big rsk for nvidia not to make sure they are ready
13:43Sarayan: Is there a way to know if a laptop lcd panel is gsync/freesync or anything similar?
13:44karolherbst: Sarayan: check the marketing bs of that laptop
13:44karolherbst: if there is nothing advertised, it won't have it :p
13:45Sarayan: it's a business laptop, not a gaming laptop, so I suspect they may not advertise it
13:45karolherbst: imirkin: if you want to look at it, I could try to create an mmiotrace where I am more or less sure it does gsync stuff
13:45karolherbst: Sarayan: the nvidia-settings thing should also tell you
13:46karolherbst: there are several gsync modes you can use
13:46Sarayan: karol: pretty sure the nvidia doesn't see the outputs, only the intel, and it doesn't boot anyway :-)
13:47karolherbst: I had to put my laptop in nvidia only mode
13:47Sarayan: karol: rememeber, it's the one that timeouts in the pmu bootup
13:49karolherbst: well, the situation is different if you run the nvidia drivers
13:49karolherbst: nouveau won't be able to do gsync
13:49Sarayan: I guess. I must admit I haven't tried the nvidia driver on it yet
13:50imirkin: karolherbst: not really.
13:51karolherbst: yeah.. I am also not really motivated, because nobody really would benefit from it
13:52karolherbst: those few ms reduction in output latency is really not our biggest concern right now :p
13:52Sarayan: karol: you never worked on a video game emulator, some people are insane about that :-)
13:52karolherbst: Sarayan: that's why I said "our"
13:53karolherbst: and yes, older hardware usually has lower output latencies
13:53karolherbst: especially pre the time Os cared about security
14:02imirkin: Sarayan: feel free to send patches
14:03Sarayan: imirkin: As soon as I manage to make my card boot :-) And in any case I'm pretty sure there's no *sync in my laptop because everything goes through the intel gpu which does not support that
14:07imirkin: exactly - everyone's got bigger problems to deal with
14:37Armada: Sarayan, if it has G-Sync you can be damn sure they'll advertise it, the license for that is not cheap
14:37mupuf: RSpliet: hehe ;)
14:37Armada: Freesync is another story though
14:37mupuf: there are usually other features that allow synchrozing displays :)
14:38Armada: yeah, but those mostly fall in the Freesync category
15:07vliaskov: sorry to post this again, has anyone seen similar nouveau messages when using 2 gpus (nvidia+AMD, or 2xnvidia) and /dev/dri/card0 is a nouveau card https://pastebin.com/4p4nniUn
15:10simonpatapon: since last upgrade, i dont see msg but i lost 2 screen out of 4
15:11simonpatapon: hi, i used to have 4 screens on onboard i915 and Nvidia GPU with nouveau. I updated my debian SID and now only have 2 screens out of 4, here is xorg log, xrandr info and dmesg https://pastebin.com/TTMnvYpH https://pastebin.com/qRfuGA2i https://pastebin.com/EwKXRxSt thx for your time
15:12imirkin: vliaskov: fyi, "system-logind" is a lie
15:13imirkin: could be just about anything. the pid is misreported due to ... long story.
15:13imirkin: vliaskov: what's your desktop environment like?
15:14imirkin: simonpatapon: you're not linking the providers together. you have to run ... xrandr --setprovideroutputsource 1 0
15:14vliaskov: imirkin oh ok thanks. it's GNOME mutter/wayland, so it could be the compositor's fault. It doesn't happen with switched cards though, that sounds weird.
15:15imirkin: simonpatapon: oh, and it looks like GLAMOR is failing to load with modesetting, so you're in for a bad time
15:15imirkin: simonpatapon: i'd encourage you to use xf86-video-nouveau
15:15imirkin: vliaskov: switched cards as in ... where everything runs on the amd gpu?
15:15Sarayan: glamor is x-over-gl, right?
15:15imirkin: Sarayan: exa-over-gl
15:16imirkin: wellll... i guess not exa. i think it's more direct. so yeah, x-over-gl.
15:16Sarayan: isn't exa a memory/buffer management scheme?
15:16simonpatapon: i try this xrandr thing but it fails
15:17imirkin: simonpatapon: yeah, probably coz no acceleration
15:17imirkin: Sarayan: among other things... it also has various drawing api's that the X server makes use of
15:17imirkin: simonpatapon: should figure out why that is...
15:17imirkin: unfortunately i have to run
15:18Sarayan: imirkin: Oh ok, I didn't realize it had drawing apis too
15:18Sarayan: I thought it was just passing intel "bytecode" to the card
15:18simonpatapon: imirkin: thx, but how to user xf86-video-nouveau
15:18simonpatapon: juste apt-get install it?
15:19vliaskov: imirkin i think mutter tries to do glBlitFrameBuffer to copy to the primary GPU. When /dev/dri/card0 is the nouveau card mutter logs say "Failed to blit shared framebuffer: EGL failed to allocate resources for the requested operation."
15:19vliaskov: anyway thanks for the info on systemd-logind probably being irrelevant.
22:44tertl3: my video game is getting hot? did yall hack my drivers?
22:44tertl3: damn nerds
22:45HdkR: I'm alwaysh hacking your drivers
22:48tertl3: yeah, he who smelt it dealt it
22:52tertl3: Flatulations 4:20
22:52tertl3: in the Holy Book of the Bros
23:37mindofmateo: I'm sorry if this is a dumb question, but is it possible to use my GTX1050 with Nouveau? I want to try sway-wm with Wayland on Ubuntu18.04 (which cannot use Nvidia's drivers). I switched from the Nvidia drivers to Nouveau, but now my system shows I'm using Intel graphics. Is it possible to use my graphics card with the Nouveau drivers instead?
23:43karolherbst: mindofmateo: is this a laptop or maybe a CPU with an intel GPU?
23:44karolherbst: some distributions more or less hack around that and every solution sucks
23:44mindofmateo: Yes, it is a laptop, from the Nouveau Website the code is GP107m
23:44karolherbst: problem is
23:44karolherbst: you don't want to use the nvidia GPU all the time on a laptop
23:44karolherbst: cause battery
23:45mindofmateo: I don't really use it when it isn't plugged in, I'm not that concerned about battery
23:45karolherbst: and what nvidia "supports" is to use the intel driver to display whatever nvidia renders: nvidia GPU always on, intel GPU always used = bad battery lifetime
23:45mindofmateo: So does that mean the GPU will turn on when it's needed, or no?
23:45karolherbst: mindofmateo: sure, but, the proper solution is like how you expect this to work on windws, meaning intel does all the stuff, normally and only when you run applications with high demands you turn on the second GPU
23:46mindofmateo: OK. Does that happen automatically? I just want to make sure I understand.
23:46karolherbst: mindofmateo: well, there are several ways to force an application to use the nvidia GPU. Gnome has something on the context menu I think
23:46karolherbst: on the CLI you have "DRI_PRIME=1" as an environmental variable
23:47karolherbst: mindofmateo: doing that automatically would mean we would have to create a whitelist of all applications
23:47karolherbst: this is what nvidia does
23:47karolherbst: well, on windows that is
23:47karolherbst: mindofmateo: there is one problem though. there is a severe bug causing the GPU to mess up whenever we want to turn it on, on some laptops
23:47mindofmateo: That's unfortunate, haha.
23:47karolherbst: and it involves signed firmware and undocumented stuff :/
23:48karolherbst: really painful to fix that issue if you don't know how what you have to do
23:48karolherbst: on the 1050 we can't change the performance modes or clocks or whatever
23:48karolherbst: meaning that with the intel driver you should get more performance out of the GPU, sadly
23:49joepublic: But Nvidia totally knows that they are the problem, and they promise to totally to think really hard about what to do, last I heard. [this was not recently]
23:49karolherbst: the hw features are locked down behind some hw security feature where we need signed firmware from nvidia to actually do it (or do the same approach as with the video accel API)
23:49mindofmateo: So does that mean I need to choose between using sway-wm, or using the Nvidia drivers, but not both?
23:49karolherbst: why doens't sway-wm run on nvidia?
23:49karolherbst: ohh wayland?
23:50mindofmateo: What you're saying makes sense. That's what I've been reading. Correct, it's not because of the GPU per se, it's because sway-wm says specifically it will only work with Nouveau.
23:50karolherbst: I think your software stack is too old for that
23:50karolherbst: and uhm
23:50karolherbst: EGLstream support
23:50karolherbst: which sway-wm won't have I guess
23:50mindofmateo: Which part of the software stack?
23:50karolherbst: so nvidia is out of question
23:52karolherbst: mindofmateo: there was some work going on to support some nvidia only stuff inside XWayland, but that doesn't help if the compositor can't work with it
23:53karolherbst: and Xwayland you need for OpenGL applications written against X (which are currently nearly all (ignoring GUI stuff))
23:53mindofmateo: well bummer. So should I just use i3 with X11?
23:53karolherbst: I doubt there are many alternatives currently
23:54mindofmateo: I'm still a CS student and so much of this is beyond what I understand at a meaningful level.
23:55karolherbst: mindofmateo: well, except you take lessons especially targeting computer graphics, it doesn't really help if you finished your studies or not
23:55karolherbst: I've never had such classes
23:55mindofmateo: I'm in linear algebra right now.
23:55RSpliet: that's basically graphics
23:55mindofmateo: We're touching on its relationship to graphics. But not like hardware.
23:55joepublic: "linear" means "of or relating to lines"
23:56karolherbst: RSpliet: well, it's pretty much basics to everything
23:56mindofmateo: lol I don't have an A, but I at least got that much haha
23:56karolherbst: you need it for compressions as well
23:56karolherbst: and ... others things
23:57karolherbst: sounds stuff as well
23:57mindofmateo: Well darn. It sounds like I could use sway-wm on Wayland technically, but then I'd just have a useless GPU
23:57joepublic: slope, intercept, cartesan coordinates
23:57RSpliet: karolherbst: dense/convolutional neural nets are mostly linear algebra afaik
23:58mindofmateo: I don't have a deep understanding of ML either, but it seems very relevant as well, as far as vectors go
23:58karolherbst: if you think about it, a lot of things are based on that
23:58mindofmateo: We used it to categorize "images" based on the k-nearest means method
23:58karolherbst: that's more than I did
23:58RSpliet: mindofmateo: the secret in neural nets is the backpropagation bit, that's proper calculus. The rest is all matrix multiplications
23:59karolherbst: I seriously hope that ML hype is over in 5 years though
23:59mindofmateo: Yeah, vector or matrix representations. RSpliet: it sounds so interesting and fun, but it's hard to keep up with all of it
23:59karolherbst: better 2, but I know that won't happen :p