01:35 bloblo: hello
01:35 bloblo: someone help me
01:36 bloblo: nouveau support gtx 970 ?
02:18 krutonium: Hi, I'm just curious - Is there still work underway to get Video acceleration on the 750 Ti and newer cards?
02:18 krutonium: (Decoding, Mostly)
02:20 krutonium: I ask mostly because Nouveau is so much better than nVidia's binary blobs, but I need accelerated video decoding :(
02:25 imirkin: krutonium: sure. step 1: send patches to implement this. step 2: enjoy :)
02:26 imirkin: [afaik work was never underway for maxwell+]
02:27 imirkin: krutonium: in case you play video games, grab kernel 4.12, that should let you reclock your GPU
02:28 krutonium: I'm not capable of providing patches, I was simply wondering if it was being worked on :(
02:28 imirkin: yeah, sorry =/
02:29 krutonium: I'm already running 4.12.8-2 (Arch)
02:29 krutonium: Do I have to change anything or should it just work?
02:30 imirkin: well, it's manual
02:30 imirkin: controlled via /sys/kernel/debug/dri/0/pstate
02:30 imirkin: should show you a list of available states, echo one of them in to switch. note that this isn't super-duper solid, so box may hang
02:31 krutonium: Roger that
02:33 krutonium: Didn't hang, and instant, massive, improvement in FPS
02:34 krutonium: Question: Will this hang around post reboot, or do I have to do this every time?
02:35 imirkin: it's entirely manual
02:35 imirkin: on boot, you get whatever the VBIOS does
02:35 imirkin: which is "lowest clocks" usually
02:35 imirkin: "AC" tells you the current state
02:37 krutonium: >AC: core 405 MHz memory 810 MHz
02:37 krutonium: A little less than optimal for gaming. Thanks for telling me!
02:38 krutonium: Question: My other option shows a range, does that mean the GPU will clock up and down inside that range as needed?
02:38 imirkin: a theoretical driver might
02:38 imirkin: nouveau will pick the top core clock of the perf level that it can actually get to
02:39 krutonium: So this is just running it at full tilt then?
02:39 imirkin: (there are various reasons why the highest one might not be usable)
02:39 imirkin: you can clock back down if you like
02:39 imirkin: but within a level, it'll pick the highest end of the range
02:39 krutonium: Of course, I am just making notes... And debating if I should make a script to toggle GPU states...
02:41 imirkin: i believe some users do that
02:42 imirkin: hotkeys to switch between high/medium/low
02:44 krutonium: Is there somthing I can change to make /sys/kernel/debug/dri/0/pstate read/writeable without having to `sudo su` first?
02:45 imirkin: it's root-writable only
02:45 imirkin: you can use various cleverness like echo 0f | sudo tee /sys/.../pstate
02:45 krutonium: Even `sudo cat path` couldn't read it
02:45 imirkin:just likes having root shells around
02:45 imirkin: uhhhh
02:46 imirkin: well, that should certainly work
02:46 krutonium: If I sudo su then do it it works fine
02:46 imirkin: dunno why that'd be any different
02:46 imirkin: could be some helpful sudo configuration
02:46 krutonium: "helpful"
02:47 imirkin: i just use "su". no sudo, no troubles :)
02:48 krutonium: http://paste.ubuntu.com/25386828/
02:48 krutonium: 10/10 Would say "Wut" again
02:49 imirkin: sudo echo 0f > /sys/kernel/debug/dri/0/pstate
02:49 imirkin: why would you expect that to work?
02:49 imirkin: (think about what that does)
02:49 imirkin: and then compare that to the command i suggested and think about what *that* does
02:50 imirkin: [hint: my command will work]
02:50 imirkin: i.e. echo 0f | sudo tee /sys/.../pstate
02:50 krutonium: fuuuuuck I'm an idiot >.<
02:50 krutonium: Thanks for pointing that out lol
02:51 imirkin: you're not the first to get confused
02:51 imirkin: but like i said, i just do "su -", type in the root password, no sudo required.
02:51 imirkin: and no funny business
02:52 imirkin: to each their own.
02:52 imirkin: bbl
02:52 krutonium: Alright, thanks for the help!
12:54 imirkin: woo-hoo!! https://github.com/KhronosGroup/VK-GL-CTS/issues/51
12:54 imirkin: karolherbst: --^
12:56 RSpliet: what an unfortunately misleading nickname this person has... for a second it seemed like you were dealing with ARM legal, turns out to be ALExander GALazin
12:57 RSpliet: but that stuff is good news, glad there's a serious feedback loop between khronos and the OSS community :-)
12:59 imirkin: i'm just glad we don't have to deal with the stupid mod() thing
14:30 drathir: mornin/evenin...
14:33 drathir: small question, im wonder if anyone occur too whole os freeze playing fhd with gt630 + mpv + --hwdec=vdpau ?
14:37 RSpliet: drathir: stutter, or complete hang?
14:41 orbea: drathir: yea, known issue, nouveau can't do multithreading well, either do -vo=vdpau and -hwdec=vdpau, use MPlayer instead or use software decoding
14:41 drathir: RSpliet: complete hang restart needed...
14:42 drathir: orbea: i will try that solution too, and weak hw that why gpu used to decode...
14:44 drathir: im not think card damaged bc xonotic working flawlessly... that would be really bad luck if only that damaged...
14:45 drathir: btw is cuda working under linux? bc saw last time mpv get cuda support too...
14:45 orbea: its just mpv doing something beyond what nouveau can currently handle...
14:46 drathir: there any recommendations of gpu-s well supported which handle x264/x265?
14:47 orbea: something made by amd or intel
14:47 khronosschoty: I would recommend intel reluctantly
14:48 drathir: orbea: hmmm.... probably not worth of investing i guess if intel than new(more recent) platform build will get it integrated...
14:49 drathir: no idea if fine ask there but closed one drivers should doing any better or no chance?
14:49 orbea: probably
14:51 orbea: that is what the mpv dev told me when I reported that crash to them btw, to use the blob drivers instead. Personally not acceptable to me, but I shouldn't of got a nvidia card...
14:52 drathir: and one more thing i noticed/watched that looks like fan sometimes get on high speed and stuck no idea if that known issue too... just fresh getting card thats first thing i noticed so far...
14:52 drathir: orbea: lol thats ekhm strange sugestion...
14:52 imirkin_: drathir: GF108 or GK107 or GK208?
14:53 imirkin_: drathir: there's lots of hostility towards nouveau for some reason
14:53 orbea: i've seen people get literally irate when I mentioned I use nouveau in other channels :P
14:54 drathir: imirkin_: gf108
14:55 imirkin_: drathir: there are some known issues with fan fail on GF108
14:55 imirkin_: drathir: mupuf looked into them, but i haven't heard anything from him on the topic in QUITE a while
14:56 imirkin_: orbea: it's somewhat understandable... the thinking all too often is "if it works on nvidia, and doesn't work on $other thing, then $other thing must be broken"
14:56 drathir:also always try first open one drivers if possible...
14:58 imirkin_: well, i'm sure you'll find that nouveau has a number of various issues
14:58 mupuf: imirkin_: yep... still no updates on this...
14:58 imirkin_: the first of which for your gpu, is the lack of reclocking capabilities
14:59 mupuf: I can't figure out the equation
15:00 drathir: hmmm... fan at full speed have kind of sense if gpu stuck at full performance mode all the time...
15:01 RSpliet: drathir: unfortunately, nouveau isn't stuck at full performance mode. Rather at 10% performance mode
15:02 imirkin_: depends on the gpu
15:02 imirkin_: actually that's one of the redeeming qualities of GF108 - most boot to a middle perf level :)
15:05 drathir: that xfx one but from security side that 10% is made with head to not damage hw, but from user side neat to hear...
15:05 imirkin_: drathir: cat /sys/kernel/debug/dri/0/pstate
15:05 imirkin_: check the "AC" line
15:12 mupuf: drathir: unlike cars' engines, the noise level is not directly related to the performance :D
15:12 mupuf: it should... but not always.
15:13 imirkin_: or vacuum cleaners :)
15:13 mupuf: oh, for this one, nowadays, it is possibly an opposite correlation :D
15:13 imirkin_: [iirc someone made a quiet vacuum cleaner that failed in the market due to people thinking it wasn't working well enough]
15:13 mupuf: Ha ha ha ha
15:13 imirkin_: [this was some time ago]
15:35 siro__: will there be dynamic reclocking on NVE in the near future ?
15:56 imirkin_: siro__: define 'near'
15:57 siro__: next month ? next year ?
15:57 karolherbst: imirkin: nice!
16:01 imirkin_: siro__: not next month. probably next 12 months.
16:01 imirkin_: up to Karol though, for the most part. (and ben to actually merge stuff)
16:15 drathir: imirkin_: k checking...
16:15 drathir: mupuf: sadly yea especially vacum cleaners ;p
16:18 drathir: imirkin_: ac is middle one vaule...
16:19 imirkin_: some of the high-end GPUs boot into the lowest level
16:19 imirkin_: which can be VERY low, like 50mhz core clock
16:19 drathir: imirkin_: core 405/memory 324 MHz
16:19 imirkin_: yeah
16:20 drathir: ~50% looks like...
16:20 drathir: imirkin_: and yep lowest is 50/135
17:00 karolherbst: imirkin: can we also increase MaxNativeAttribs to 128 or only MaxInputComponents?
17:02 imirkin_: nfc what MaxNativeAttribs is.
17:02 karolherbst: no clue either
17:02 karolherbst: it gets the same value
17:02 karolherbst: aka PIPE_SHADER_CAP_MAX_INPUTS
17:02 imirkin_: yea dunno
17:03 imirkin_: the deal is that the max number of things is 128
17:03 imirkin_: however position takes up 4 of those.
17:03 karolherbst: ohh right, I do it the wrong way :(
17:04 imirkin_: so one of the caps is "how many varyings can you have"
17:04 imirkin_: and the answer to that is 124 (components)
17:04 imirkin_: another cap is "what is the max number of frag inputs", and the answer is 128
17:04 karolherbst: yeah
17:22 karolherbst: "KHR-GL45.limits.max_fragment_input_components" Pass :) https://github.com/karolherbst/mesa/commit/4ae904085b9543481b34720c0385a32fa1c4770c
17:27 karolherbst: and it seems like TGSI will be abandoned at some point for real
17:38 imirkin_: karolherbst: dunno, if that happens i'm probably done with mesa.
17:54 karolherbst: imirkin_: k
17:58 karolherbst: 4.4 status: Failed: 33/5989 (0.6%)
18:30 karolherbst: 4.5 Failed: 41/6516 (0.6%)
18:37 tobijk: karolherbst: what are the fails about?
18:38 karolherbst: tobijk: https://gist.github.com/karolherbst/ed497e99e5a0940d80ed2ead8dba2261
18:39 imirkin_: are all the enhanced layouts ones failing on intel as well?
18:39 karolherbst: the intel run isn't finished yet, but I think so, yes
18:39 karolherbst: in fact, I am sure intel has more fails than nouveau
18:41 karolherbst: KHR-GL45.CommonBugs.CommonBug_PerVertexValidation also fails on intel
18:41 imirkin_: have you updated?
18:41 imirkin_: (VK-GL-CTS)
18:41 karolherbst: yeah
18:41 imirkin_: there was a semi-recent commit directly addressing that test
18:41 karolherbst: I updated it today
18:42 imirkin_: k
18:42 imirkin_: this was a few weeks ago i think
18:44 tobijk: ah for KHR-GL45.direct_state_access.queries_functional i have https://git.thm.de/tjkl80/mesa/commit/d38aeca9d7f12c773b50eeba535f01271121330b and https://git.thm.de/tjkl80/mesa/commit/6eaee86334bd7dd4b6decddbb0d7ae880ad41acb
18:45 imirkin_: tobijk: i don't think can check hq in there
18:45 imirkin_: tobijk: have to add a is_ready callback
18:45 tobijk: mhm, i guessed you would have that idea, yet i was too lazy to start that :D
18:47 tobijk: imirkin_: a callback is better, of course, but why shouldn't i check hq there?
18:47 imirkin_: i'm surprised you have hq in the first place
18:47 tobijk: we use it alter on anyway
18:47 tobijk: 'later
18:47 imirkin_: that should be hidden behind the abstraction
18:47 imirkin_: well, hakzsam can rule on it
18:49 tobijk: hq->nesting that is to be precise
18:50 imirkin_: ah =/
18:50 tobijk: yet hding those is better :)
18:51 karolherbst: duh.. intel hits an assert on one test
18:51 karolherbst: KHR-GL45.arrays_of_arrays_gl.InteractionArgumentAliasing1
18:52 imirkin_: file a bug =]
18:52 tobijk: it should pass it, most likely a regression
18:52 imirkin_: although their track record for addressing issues filed in bz is ... varying
18:53 tobijk: heh, just whine a bit in #dri-devel then ;-)
18:53 imirkin_: i try to be constructive
18:53 imirkin_: and with sufficient pushing, the issues do get addressed
18:53 tobijk: *just kidding*
18:53 karolherbst: well
18:53 karolherbst: they have to pass 4.6 anyhow
18:55 tobijk: but that is from the 4.5 list, which the driver is "certified" for already :>
18:55 karolherbst: 4.6 should contain everything on the 4.5 list as well
18:55 karolherbst: at least 4.5 contains all 4.4 tests afaik
18:57 karolherbst: I think after I am done with my run, I just drop the list in the channel and then they can do whatever they want with it
18:59 karolherbst: tobijk: with your patches "KHR-GL45.direct_state_access.queries_functional" doesn't pass
19:00 tobijk: huh?
19:00 tobijk: it did for me
19:00 tobijk: well, work in progress then :D
19:41 tobijk: karolherbst: ah my patch was for KHR-GL45.conditional_render_inverted.functional, i rememberd wrong
19:41 tobijk: you have a always wait fox for that imho
19:42 tobijk: https://github.com/karolherbst/mesa/commit/a69fd7572c03b58dc95b6a2534a0e6c85f00cb1f
19:43 karolherbst: yeah, true
19:43 karolherbst: mhh
19:43 tobijk: karolherbst: would be nice to know if that patch works for you with my two fixes without that wait thing
19:44 karolherbst: yeah, will test after I am done with intel here
19:44 tobijk: patch -> test
19:44 tobijk: karolherbst: yeah, no rush :)
19:47 karolherbst: it doesn't look good for intel though
19:48 tobijk: :o
20:00 drathir: orbea: i guess both vdpau hwdec and vo in mpv does the trick preety smoothly working ad not yet crashed...
20:00 imirkin_: yeah, that should work
20:01 imirkin_: do note that there are some issues with H264 decoding. some videos get hit but not others.
20:03 orbea: i really think it has to do with how they were made
20:03 imirkin_: yep
20:03 imirkin_: some day, when i get 10 years of free time, i'll have time to do all the things i don't have time for.
20:03 imirkin_: figuring this out will be one of those things :)
20:04 orbea: heh :)
20:08 drathir: imirkin_: im guess a lot depend on compression level profile too... bc in past old 10y+ amd gpu takes fine x264 1080p and x265 720p with support reported in vainfo and vdpauinfo, sadly support for x265 taken away /
20:08 imirkin_: drathir: it has nothing to do with compression level
20:08 karolherbst: imirkin_: did you saw the last comment regarding the "last regression"?
20:08 imirkin_: drathir: we don't handle some condition in the video stream
20:08 imirkin_: karolherbst: hm?
20:09 karolherbst: imirkin_: https://bugs.archlinux.org/task/55244
20:10 tobijk: karolherbst: what has this to do with nouveau?
20:10 drathir: on amd it has from my testing with high compresion level profiles much slower processing and artefacs occured, but that was at VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV710/M92 [Mobility Radeon HD 4530/4570/545v]
20:10 karolherbst: tobijk: https://bugs.freedesktop.org/show_bug.cgi?id=102368
20:11 drathir: tobijk: yes its kind of ot a little, but just start playing with configuration of nvidia gpu which have for now upon my hand...
20:11 karolherbst: I am not dun with the run and already 65 fails with i965
20:12 karolherbst: *done
20:12 tobijk: :o
20:12 karolherbst: so yeah, nouveau passes more tests with my mesa branch
20:12 tobijk: lets advertise 4.5 ;-)
20:12 karolherbst: ;)
20:12 karolherbst: I would do it if we pass more than radeonsi
20:13 airlied: you also have to submit from a stable branch
20:13 drathir: may i ask how long takes rebuild with that test patches applied?
20:13 airlied: if memory serves
20:13 imirkin_: karolherbst: cool
20:13 airlied: so make sure fixes go into 17.2
20:14 karolherbst: airlied: yeah
20:14 karolherbst: there is one important patch we have to land: https://github.com/karolherbst/mesa/commit/175147269edcc0ed092ae35b7edb39eb96308aba
20:15 karolherbst: this kind of let us run a full CTS run
20:15 karolherbst: *do
20:15 imirkin_: yeah that's not happening anytime soon
20:15 karolherbst: I am quite sure I don't break anything with that anymore
20:15 karolherbst: it got a lot simplier than the previous version
20:15 imirkin_: that's what concerns me
20:16 karolherbst: me as well
20:16 imirkin_: the metric isn't "does this break tests"
20:16 imirkin_: the metric is "does it make full sense"
20:16 karolherbst: *sigh* 83 fails, I already exlucded 9 tests, hit another assert
20:18 karolherbst: imirkin_: well I am sure that with unordered_set we don't have duplicates inside the defs anymore
20:18 airlied: imirkin_: by that metric we'd have a no code
20:18 karolherbst: which reduces runtime by a lot
20:18 tobijk: imirkin_: we should rediscuss that patch now, as it is knida important, restating your concerns, which may or may not prove wrong later on :)
20:18 karolherbst: imirkin_: with the aoa we have single def list with 10k+ entries
20:18 karolherbst: per value
20:19 karolherbst: sure we could try to be smarter to not add duplicats or so
20:19 imirkin_: yeah i totally get that.
20:19 imirkin_: the question isn't "is it ok to make it a set"
20:19 karolherbst: but at least it let us do a full CTS run, which is nice
20:19 imirkin_: the question is "did the surrounding code change correctly"
20:19 karolherbst: mhh yeah
20:19 karolherbst: dunno for sure
20:19 karolherbst: but I think so
20:40 airlied: I also expect you fail some of the closed source crap pile still
20:40 karolherbst: most likely
20:41 airlied: horrible packed_pixels tests
20:41 karolherbst: but I start to wonder why intel "passed" the CTS
20:41 karolherbst: or they have simply tons of regressions
20:41 karolherbst: or only fail the new tests
20:41 karolherbst: currently at 170 fails
20:41 airlied: they run the GL cts in CI
20:41 airlied: just not the open source one
20:41 tobijk: karolherbst: maybe you have simply the wrong generation if intel hw :>
20:41 karolherbst: maybe
20:42 imirkin_: haswell's pretty solidly supported for GL
20:42 imirkin_: not so much for VK
20:42 karolherbst: 179/6497 (2.8%)
20:42 airlied: karolherbst: they've passed the requirements for GL4.5
20:42 karolherbst: fails
20:42 karolherbst: mhh, yeah most likely
20:42 airlied: that isn't anything like passing CTS master
20:42 airlied: not sure why people keep conflating those
20:43 airlied: GL4.5 CTS and open-source CTS are a long way divergenet
20:43 tobijk: :/
20:44 imirkin_: airlied: because people don't have access to the closed GL4.5 CTS and thus forget about its existence
20:44 airlied: imirkin_: they might soon :-)
20:45 karolherbst: well it looks like intel mainly fails new tests, so it looks alright
20:45 imirkin_: and in any case, fixing stuff for the open source repository, even if it's not a 100% correlation to the official tests, is still useful
20:45 airlied:can't get GL45 cts to build on f25 due to c++
20:45 airlied: imirkin_: yup it's a better use of time imho
20:45 imirkin_: it's ending up fixing real bugs, even if they're hard to hit
20:45 airlied: just don't be surprised you might have to fix stuff for intel and radeonsi as well in passing
20:46 imirkin_: a bunch of the bugs i've fixed were in common code
20:46 tobijk: why they don't make the tests publically available is beyond my horizon, the aspect of cheationg to pass is no apsect i can take seriously
20:46 imirkin_: i don't mind.
20:46 airlied: tobijk: they've made the tests they legally can publically available
20:47 tobijk: mhpf, ip can be a pain in the *** :/
20:47 airlied: tobijk: and GL4.4/4.5 had a lot of issues with timing of things
20:48 karolherbst: as long as we are getting closer with nouveau I won't mind as well
20:49 airlied: everyone should work on vulkan where things are less dysfunctional :-P
20:49 karolherbst: :D
20:49 karolherbst: not enough time for that right now
20:50 karolherbst: but sounds like a nice challange
20:55 tobijk: a nice thing to have would be offloading capabilities for wayland (opengl es)
22:37 karolherbst: KHR-GL45.pipeline_statistics_query_tests_ARB.functional_non_rendering_commands_do_not_affect_queries fails, interesting
22:38 imirkin_: i tried shutting some of those down
22:38 imirkin_: but i was unsuccessful
22:39 karolherbst: valgrind is pretty vorbose with that one
22:39 karolherbst: conditional jumps on uninit values
22:39 karolherbst: get_query_object (queryobj.c:839)
22:42 karolherbst: I don't see how value could be not initiliazed there, but still
22:43 karolherbst: ohhh
22:44 karolherbst: it makes sense
22:44 karolherbst: okay
22:44 karolherbst: imirkin_: I found it
22:45 karolherbst: imirkin_: q->Result was never written to
22:55 karolherbst: duh
22:56 karolherbst: okay, will send a fix out soon enough
23:01 karolherbst: imirkin_: what is the "best" way to init an union? union X x = {};?
23:02 imirkin_: sure
23:02 imirkin_: (sorry, dunno)
23:02 karolherbst: or { 0 }?
23:02 karolherbst: I will post a patch with the latter
23:02 karolherbst: is more explicit
23:09 karolherbst: it seems like = { 0 }; is quite common
23:13 karolherbst: two tests fixed :)
23:15 karolherbst: sent
23:24 karolherbst: next statistics fail is KHR-GL44.pipeline_statistics_query_tests_ARB.functional_primitives_vertices_submitted_and_clipping_input_output_pimitives
23:24 karolherbst: Invalid default QO BO uint64 query result value: found [7], expected:[15], query type:[GL_CLIPPING_OUTPUT_PRIMITIVES_ARB], GL_PRIMITIVE_RESTART mode:[disabled], draw call type:[glDrawElementsInstancedBaseVertexBaseInstance()], primitive type:[GL_TRIANGLE_STRIP].
23:24 karolherbst: but all fails give us 7 as the result
23:41 karolherbst: and now KHR-GL44.pipeline_statistics_query_tests_ARB.functional_fragment_shader_invocations passes for whatever reason
23:50 karolherbst: imirkin_: well, I'll try to figure out what exactly went wrong in nouveau, but I still think that setting a valid default value makes sense
23:50 imirkin_: all it'll do is hide bugs in the driver
23:51 imirkin_: e.g. you wouldn't have gotten this valgrind warning about it
23:51 karolherbst: right
23:53 karolherbst: mhhhh
23:53 karolherbst: I think I found it kind of
23:53 karolherbst: we shouldn't do those casts
23:57 karolherbst: okay.. I see it now