09:32karolherbst: imirkin: same with dx9 eon games ;)
09:34karolherbst: and they do not use geometry shaders though
09:50manio: hi devs, if you are boring i think we have a problem with attached patch, which i just confirmed that works:
09:50manio: i think it may be related with: https://bugs.freedesktop.org/show_bug.cgi?id=39094 and https://bugs.freedesktop.org/show_bug.cgi?id=98383
10:17nado: i had a small question, is nouveau the same/better than proprietary driver for a Geforce 7600GT (NV40) ?
10:18nado: i had something similar with an old ati where the free driver was better than amd’s (and amd had locked me in old kernel & xorg)
10:26manio: nado: just give it a try :)
10:26nado: i will, this evening :)
10:26RSpliet: nado: better is rather subjective. In terms of performance and correctness it's almost certainly worse, but if you don't hit any of the existing bugs (normal desktop usage?) it's user experience might be better
10:27nado: but before then, im trying to gather informations
10:27RSpliet: if you happen to be a vec4 compiler expert you might be able to replace the current shader assembler with an actual shader compiler and improve performance ;-)
10:28nado: well, coming from a really really ord card (cant play 1080p video with mpv --vo=opengl), it should be better anyway
10:28mwk: mupuf: wrt the C++ fix commit: http://en.cppreference.com/w/cpp/language/static_assert
10:28nado: i dont have strong gl/shaders skills
10:29mwk: static_assert with a message is C++11, static_assert without a message is C++17 and needs gcc 5
10:29mwk: so I had to add an empty message
11:28mupuf: mwk: ahaha
11:42nado: hey, in case i wanna play/tinker nouveau for my old card, where could i have the basics to get into src code ? I dont even know in which repo i should begin (linux/nouveau/mesa)
12:00mupuf: imirkin: ok, I will check on my end what is going on with your patches
12:16mupuf: karolherbst: so, you found the entry that says which entry should be used?
12:16mupuf: for the power capping?
12:16mupuf: why are there many entries in the table then, if the index is hardcoded anyway?
12:17karolherbst: mupuf: because that field is optional
12:17karolherbst: not set on my gpu for example
12:17karolherbst: around 75% of keplers have it set though
12:18mupuf: ok, not really convicing though
12:18karolherbst: it is used when it is set
12:18karolherbst: but yeah
12:18karolherbst: it's something though
12:18karolherbst: nvidia-smi displays the avg as the power cap
12:18mupuf: yeah, but they could just as well not have set a power cap in a single entry
12:18karolherbst: nope, they use this entry
12:19karolherbst: if it's set, they use it for power capping
12:19mupuf: I trust you, I am just finding it stupid
12:19karolherbst: on your titan that entry says 150W/250W/265W
12:19karolherbst: you see exactly those numbers there as well
12:19karolherbst: min power limit, max power limit and enforced power limit
12:20karolherbst: 265 is the max one
12:20mupuf: well, very good
12:20karolherbst: still want to spend more time on that though
12:20mupuf: yeah, I still find it funky to have a *table* when you hardcode the index
12:20mupuf: there may be reasons for changing the index
12:21mupuf: AKA, sharing the vbios between different GPUs or different conditions
12:21mupuf: the latter seems plausible
12:22mupuf: would be nice to try increasing hte temperature and seeing if the value changes
12:22karolherbst: if the value isn't set in the header
12:22karolherbst: the thing starts to react funky
12:24karolherbst: like it uses entry where byte 0 is 0x1 or 0x3
12:24karolherbst: 0x5 crashes the module
12:26karolherbst: mupuf: but there is something even better, nvidia-smi tells us the shutdown temperature ;)
12:26karolherbst: so if we re this one and enforce it in nouveau + add the power capping stuff, we are much more safe than we currently are
12:27mupuf: ok, that seems funky, indeed ;D
12:27mupuf: but yeah, this is great news that we have a GPU where we ca play with this!
12:28karolherbst: nvidia-smi -q -d is the thing ;)
12:28karolherbst: d takes parameters
12:28karolherbst: -d ALL lists like everything
12:28pmoreau: nado: Which part of the driver do you want to play with? The OpenGL part? (see https://cgit.freedesktop.org/mesa/mesa/) The kernel part? (see https://github.com/skeggsb/linux/, or https://github.com/skeggsb/nouveau) The DDX part? (see https://cgit.freedesktop.org/nouveau/xf86-video-nouveau/)
12:28karolherbst: even something like "double precision: low precision mode"
13:28imirkin: hakzsam_: sounds like some issue with border colors + stencil sampling? not extremely surprised...
13:29hakzsam_: but would be nice to fix them :)
13:29imirkin: patches welcome :p
13:30hakzsam_: don't you want to look into those? :p
13:30imirkin: make a trello card for it
13:30imirkin: i doubt i'll have time anytime soon
14:34nad0: pmoreau, ty for the explanations
14:34nado: actually, i feel very dumb
14:35nado: i saw a brown input on the motherboard, so i thought it was agp
14:35nado: big mistake, it was a pcie >_<
14:35nado: so im now with a pcie graphic card without fan, and 2 agp graphic cards
14:56nado: pmoreau, temperature monitoring would be part of the kernel module then ?
14:56nado: coupled with acpi ?
14:58pmoreau: nado: That would be somewhere here I guess: https://github.com/skeggsb/nouveau/blob/master/drm/nouveau/nvkm/subdev/therm/temp.c
14:59pmoreau: Look at `nvkm_therm_sensor_event()`for example
14:59nado: wow, thank you, i didnt expect so much info in an answer :)
15:00pmoreau: you’re welcome
15:01pmoreau: (I don’t know how that code works. So if you have questions, better to see with mupuf or karolherbst)
15:02nado: yeah, i planned to take a while to read the src and get how it all works
15:02karolherbst: nado: uhh, you can simply use the hwmon interface
15:02karolherbst: depends on what you want to do though
15:02karolherbst: currently there is no way nouveau would notify userspace to report "too hot" events, because we don't know what "too hot" actually is
15:02nado: i broke the fan of a graphic card (gt610)
15:03karolherbst: you can read hwmon periodically
15:03nado: i repaired it the barbarian way, but id like to check it doesnt go nuts because of absent failsafe
15:03karolherbst: do you have lm-sensors installed?
15:03karolherbst: if you
15:03karolherbst: run sensors
15:03karolherbst: the hwmon stuff is located in /sys/class/hwmon/hwmon*
15:09nado: hmm, i cant see the gpu data with sensors (ive done the sensors-detect)
15:10nado: gonna explore the /sys/class/hwmon directory
15:38karolherbst: nado: mhh odd, did you configure the kernel yourself?
15:39nado: no, its a debian testing with default kernel
15:39nado: proprietary driver for now
15:40nado: i get the temperatures with nvidia-settings
15:40nado: not with sensors
15:40karolherbst: if you run nvidia, you can't use hwmon
15:40nado: aaah, thats why :)
15:41nado: i just disconnected the fan and im waiting to see how high it goes in temperature
15:41karolherbst: in nouveau we try to expose as much as we can through hwmon, though
17:57karolherbst: mupuf: in which subdev would you put the power budget stuff?
17:59karolherbst: mupuf: I think I will write a patch for 4.10 to expose the avg/max power consumption though hwmon, then users can complain if something is wrong
18:32Piwig: ya qqun?
18:57karolherbst: skeggsb: do you mind if I rework the vbios parsing inside nouveau? there is a lot of stuff we could share between each table like general header parsing and a lot of things are done multiple times
19:43karolherbst: mupuf: fun, already some tesla cards have this table...
19:44karolherbst: allthough empty
19:49RSpliet: I wonder why ZTE is sending us nouveau patches... reckon they're building Tegra stuff?
19:49karolherbst: RSpliet: most likely
19:49RSpliet: every patch is welcome of course :-)
19:49karolherbst: tegra cloud
19:49karolherbst: why not
19:50karolherbst: they don't like windows anyway and are scared of american backdoors anyway :p
19:50RSpliet: NVIDIA has it's roots in the US ;-)
19:50airlied: well techincally they are coming from linearo
19:50karolherbst: that's why they have high performance chips not based on intel or nvidia :D
19:51airlied: ZTE might just have linaro engineers
19:51karolherbst: airlied: ohh, that might explain it
19:51karolherbst: maybe they are evaluating and somebody got bored at work
19:51karolherbst: who knows
19:53RSpliet: yeah they seem to be prominent Linaro members :-)
19:53RSpliet: it's all just silly speculation anyway, but I do believe in people scratching their own itches esp. in corporate... anyway
20:01karolherbst: RSpliet: huh, pls, nobody does this ever
20:06karolherbst: mupuf: your kernel is strange on reator...
20:50karolherbst: :) https://gist.github.com/karolherbst/a555524e5820cde35fc110a2865b942c
20:57robclark: RSpliet, airlied, yeah, iirc zte is linaro member.. so probably some linaro assignee doing kernel cleanup work (if you are talking about the "mark symbols static where possible" stuff)
21:27Moguri_work: What repo and branch should I use for the Linux kernel if I want to stay up-to-date with Nouveau changes? master from https://github.com/skeggsb/linux/?
21:31karolherbst: Moguri_work: use rather the out of tree nouveau repository
21:31karolherbst: but it doesn't matter that much
21:34karolherbst: Moguri_work: do you hace something special in mind you wanna do?
21:37Moguri_work: karolherbst: Nothing special at the moment, just getting a nouveau dev environment setup. I have a version of mesa built from recent git sources, so I figured I'd get a kernel built too. I may try getting into some power management stuff.
21:38karolherbst: Moguri_work: well there are many little tasks you might be able to look into, depends on your gpu though
21:38Moguri_work: karolherbst: My machine that I'm setting up for this is using a GF 108 (GT 425M).
21:38karolherbst: ohh right, that one
22:50xavier1: Hi, I asked about this yesterday but still haven't sorted it out:
22:52xavier1: Running Awesome WM and Nouveau -- Scrolling in urxvt is horrendously slow but is fine in xterm, even with default configurations for everything. It works fine with nvidia drivers.
22:53imirkin_: urxvt probably performs some X action that we accelerate poorly or not at all
22:53imirkin_: are you using xf86-video-nouveau or -modesetting?
22:56xavier1: I was asked to check CPU usage while scrolling in urxvt yesterday but I had switched to nvidia drivers. I'm back with Nouveau now and the X process shoots up to 15%
23:01imirkin_: xavier1: get some call traces with perf maybe?
23:03xavier1: Wasn't aware of any profiling tools
23:03xavier1: I'll have a look after I try a few things
23:03imirkin_: use 'perf record -g'
23:23xavier1: hmm that's a kernel thing? gonna have to read up on it quick