00:00 imirkin: AndrewR: patch sent
00:27 AndrewR: imirkin, now it segfaults in different place
00:27 imirkin: bleh
00:27 imirkin: i thought i looked at the various stuff
00:28 imirkin: but i guess i missed a spot
00:28 AndrewR: http://paste.debian.net/918294/
00:28 imirkin: grrrrr
00:28 imirkin: ok
00:28 imirkin: i thought i could assume there was a fragprog.
00:28 imirkin: i guess i can't
00:29 imirkin: more changes are a-coming
00:45 imirkin: AndrewR: ok, try again :)
00:46 imirkin: this is a game of whack-a-mole
00:47 imirkin: nvc0 will require updates too i think - pretty sure it relies on fragprog somewhere
00:47 imirkin: i don't have my desktop handy to test, unfortunately
00:49 AndrewR: imirkin, it works! \o/
01:01 imirkin: yay
01:01 imirkin: now i get to audit nv30 and nvc0.
01:02 skeggsb: imirkin: on that particular issue, i've been hitting several things..
01:03 skeggsb: was wondering whether checks on cso existance be added everywhere they're used, or to plug in default ones at context init
01:04 skeggsb: in *most* cases, we'll have valid csos when the validate checks are hit, it's mostly just when stuff like clear() happens before all states have been set etc
01:05 imirkin: skeggsb: and only since recent times :)
01:06 imirkin: skeggsb: i dunno, seems easy enough to bail for those cases. the swr guys went with default values.
01:21 skeggsb: imirkin: one of the changes i attempted to make with this mt stuff made it a lot more common :) but, i'm doing stuff a different way now and will deal with making that smoother/simpler another time
01:23 imirkin: ok cool
01:23 imirkin: i'm not attached to any particular approach
01:25 imirkin: somehow i get the feeling that you've already rewritten everything 4 times, and are gearing up for a 5th... :)
01:26 skeggsb: i'm trying very hard to limit the scope of changes :) i *am* starting to think i should've taken your advice and started from scratch, and copy+pasting the hw logic into it
01:26 skeggsb: meh, live and learn
01:27 imirkin: don't remember giving that advice, but that does sound like a reasonable idea :) the tricky bits can largely be copied in any case
01:28 imirkin: the whole prog fixup is a little sketchy, esp in a multithreaded context i suppose
01:28 imirkin: but that'd be a sad situation, where 2 contexts are using the same program but with slightly different settings that cause fixups to be applied on and off :)
01:29 skeggsb: i think *eventually* we should give each context its own text area, i'll use locks there for now though
01:29 imirkin: real alternatives are probably required, but ... again ... feature creep
01:35 dboyan_: imirkin: did you receive the 2/3 of my patch series? I'm just curious because i didn't find it in my gmail inbox. It is on the ML archive though.
01:37 imirkin: dboyan_: as a reply to the 1/3 patch for some reason
01:37 imirkin: dboyan_: i haven't reviewed it. i think some of my comments will apply to it as well.
01:38 imirkin: but mostly i had other things i had to do and figured my initial review would be enough for now :)
01:38 dboyan_: oh yeah, there it was
01:39 imirkin: so your rcp was off by 2 ULPs from the cpu version in some cases?
01:39 imirkin: i wonder how far off the "real" result it was ... the cpu version isn't guaranteed 100% accurate either
01:40 dboyan_: I didn't record that, I'll test for it if needed.
01:41 dboyan_: I was just too lazy when writing test programs...
01:41 dboyan_: But I'm rather confident in my algorithms, at least in rcp
01:42 imirkin: yeah, i mean it seems reasonable. other than the part at the very end, i felt like i understood the whole thing.
01:42 imirkin: so ... well done with the comments :)
01:43 imirkin: i might do a pass after you're done and add in even more. but other than the things i pointed out, you don't have to do that.
01:44 dboyan_: The blob actually use more complicated algorithms for the last case of rcp, out of my understanding. That part was actually a simple trick that I invented.
01:46 dboyan_: And you can tell that I'm not good at naming things, especially labels :)
01:48 imirkin: well, no matter how bad you are, i'm sure you can do better than "L2" and "L3" :)
03:16 imirkin: skeggsb: btw, did you see that RSpliet got his nvc4 reclocking?
03:16 skeggsb: yeah, he PM'd me, i think he's asleep now though for my reply
03:17 imirkin: 3am in UK? seems likely.
03:17 skeggsb: who's asleep at 3am? ;)
03:17 imirkin: me, of late =/
03:18 skeggsb: i should be, but i've been keeping weird hours the last few weeks
03:18 imirkin: s/weeks/decades/?
03:18 skeggsb: btw, i'll be gone for 2.5 weeks as of wed (holiday to NZ)
03:18 skeggsb: haha, yeah, on and off for decades
03:19 imirkin: k, noted. have fun!
03:21 skeggsb: i'll try not to get killed jumping off things :)
03:23 imirkin: good plan
04:50 Horizon_Brave: Hi everyone..
04:59 Horizon_Brave: So...I have a question all... about the development of open source drivers..
05:00 Horizon_Brave: Does nVidia provide any help what so ever to the nouveau community of developers here?
05:00 Horizon_Brave: if so, in what way to they help and what do they provide you all?
05:11 airlied: Horizon_Brave: they provide some infomation and signed firmware modules and some code on a case-by-case basis and hardly ever a timely one
05:15 Horizon_Brave: airlied: Do you usually have to ask them for this information? Do they require anything from you guys when they know you're working on reverse engineering one of their cards/chipsets?
05:15 Horizon_Brave: I assume they'll never provide you stuff without actually being requested for it?
05:23 airlied: Horizon_Brave: depends on whether it is something they need for their tegra work
05:23 airlied: so mostly it's ask for stuff, and some stuff gets contributed because they want it upstream for something else
09:18 thehungus: am having some trouble with multihead (gasp), 3 monitors across two GT200 cards. X drives all 3 using two instances of nouveau; screen 0 with two monitors on one card, screen 1 with the other monitor. the mouse goes between all three, but the window manager will only run on screen 0, probably because xrandr doesn't list the other screen unless explicitly asked to with --screen.
09:19 thehungus: have previously had this setup work fine with the proprietry driver, but it crashes and causes all sorts of other grief (and won't update for me as my card is "too old")
09:20 thehungus: am pretty sure if i could get xrandr --listproviders to show more than one card at a time then all would be well :)
09:20 thehungus: any ideas?
14:41 dboyan: imirkin_: I forgot to mention the nan thing (setting the most significant bit in mantissa) in your comments in my reply. It might be related to "signal" bit in ieee 754
14:41 dboyan: iirc, i'm seeing the same thing on cpu
15:17 karolherbst_: mupuf_: it would be good if you look over my last series on the ML, because some/most commits still have your r-by tag on them
15:17 mupuf_: karolherbst_: yes!
15:17 mupuf_: I glanced at it, thanks a lot!
15:18 karolherbst_: I hope those changes make much more sense now than they did months ago
15:18 mupuf_: hehe, hopefully ;)
15:20 karolherbst_: mhh, in patch 7 is a random chunk
15:22 karolherbst_: and patch 8 is just crap... (well technically correct, but some of the changes belong elsewhere)
15:23 karolherbst_: ohh wait, not true. patch 8 is indeed correct :D
21:18 coolkid87: What is the difference between mesa and nvc0 here https://mesamatrix.net/ ? mesa = software emulated version?
21:19 imirkin_: coolkid87: mesa = nothing, nvc0 = nvc0.
21:19 imirkin_: i have no idea why "mesa" is tracked as a separate category there
21:19 imirkin_: i think it's the union of all the other things?
21:20 coolkid87: I used to think that too
21:20 coolkid87: before seeing that
21:21 imirkin_: if you want to know what's supported on what gpu, https://people.freedesktop.org/~imirkin/glxinfo/ should be a good indication.
21:22 coolkid87: nice, is it scripted?
21:23 coolkid87: i.e. updates itself?
21:23 imirkin_: no
21:23 imirkin_: i have to update it. as you can see, it doesn't have 17.0 yet
21:23 imirkin_: (because i've been super-busy with things)
21:23 imirkin_: others in the mesa group can update it too, but they don't
21:25 coolkid87: Is it possible to supported version getting reduced after installing a compiled version of mesa over system packaged ones?
21:25 imirkin_: can you rephrase your question? i'm having trouble parsing.
21:26 coolkid87: I just installed mesa 17 today, the nouveau module, and it started to report 2.1 instead of 3.3
21:26 imirkin_: you probably built without --enable-float-texture
21:26 coolkid87: it = glxinfo
21:26 imirkin_: (or your packager)
21:26 coolkid87: yes I didn't put that
21:26 coolkid87: okay I try
21:30 coolkid87: still 2.1, my configure line is `./configure --prefix=/usr --libdir=/usr/lib/x86_64-linux-gnu --with-gallium-drivers="swrast,i915,nouveau,virgl" --with-dri-drivers="swrast,i965,nouveau" --enable-dri --enable-glx=dri --enable-float-texture`
21:30 imirkin_: you have to clean after you do that.
21:30 coolkid87: Did I do something stupid?
21:30 coolkid87: okay
21:30 imirkin_: also i think you can drop --enable-dri and --enable-glx -- those will be on by default
21:31 coolkid87: okay
21:40 coolkid87: "make clean"ed and "make -j4 && sudo make install"ed still getting 2.1
21:41 coolkid87: Do I need to reboot, probably not, right?
21:42 _prg_: try --enable-texture-float instead
21:42 coolkid87: and clean?
21:42 imirkin_: oh bleh, sorry!
21:42 imirkin_: i'd definitely recommend cleaning
21:42 coolkid87: so?
21:42 coolkid87: okay
21:42 imirkin_: and yeah, check ./configure --help for the proper option name :)
21:42 imirkin_: that's what it's there for
21:43 glennk: i think mesa = core infrastructure, no driver back end
21:43 imirkin_: it should have given you an error
21:43 coolkid87: well it didn't
21:43 imirkin_: or a warning
21:45 coolkid87: I'm checking config.log for /warning, there are a few from gcc
21:46 imirkin_: try WARNING
21:47 karolherbst: texture-float is the right flag though
21:47 coolkid87: sorry, I should've checked that, yes, it's there
21:48 imirkin_: no worries
21:58 coolkid87: thanks, now it shows 4.3
21:58 coolkid87: core profile
21:59 imirkin_: cool
21:59 imirkin_: which GPU do you have?
21:59 coolkid87: GTX 660 Ti
21:59 coolkid87: nVidia
22:00 imirkin_: ah ok - i'd recommend getting kernel 4.10 since that will have working manual reclocking
22:00 coolkid87: okay should I downgrade?
22:00 coolkid87: I've installed 4.4 recently
22:00 imirkin_: 4.10 is the latest kernel.
22:00 imirkin_: 10 > 4
22:00 coolkid87: oops
22:03 coolkid87: any chance mesa will give 4.5 on my card? :)
22:03 coolkid87: maybe if I upgrade the kernel?
22:03 imirkin_: no - it's been requested that we not expose it
22:04 imirkin_: to avoid giving trouble to distros
22:04 coolkid87: okay
22:04 imirkin_: since they may be on the hook with khronos for distributing an unapproved thing with GL 4.4/4.5
22:05 imirkin_: or something.
22:05 coolkid87: you mean a feature that is patented?
22:06 airlied: the GL trademark
22:06 coolkid87: oh
22:06 imirkin_: no, they want you to (a) pass their conformance tests and (b) pay them money. having them waive (b) has been discussed, but we're nowhere close to (a).
22:06 coolkid87: But I read somewhere that mesa declares that it follows GL but isn't GL compliant, or something like that
22:08 airlied: well it's all blurry now that there are mesa compliant drivers
22:10 karolherbst: I would try to deal with (a) but somehow the test suite is a mess to run
22:11 coolkid87: found it https://www.mesa3d.org/license.html in 1st paragraph
22:11 karolherbst: or not really user friendly, maybe I just do something wrong
22:11 coolkid87: so this is the story behind it
22:11 karolherbst: airlied: any idea what is the lowest GL level the conformance test is supposed to run at? is it fine to start with 4.4 fix everything and try to get an ok exposing 4.4 and then deal with 4.5?
22:12 imirkin_: coolkid87: if you want you can patch your own mesa to report 4.5 for nouveau - just change 430 -> 450 in nvc0_screen.c
22:12 airlied: karolherbst: best just deal with 4.5
22:12 airlied: most of the bugs are probably 4.4 anyways
22:12 airlied: or older
22:12 imirkin_: 4.2 :)
22:12 karolherbst: airlied: well, 4.4 is less work I assume
22:12 karolherbst: ohh
22:12 karolherbst: so for exposing 4.5 we don't need to pass the 4.4 tests? :D
22:12 coolkid87: imirkin_: and I'll have DSA?
22:12 imirkin_: coolkid87: you have DSA either way
22:13 imirkin_: coolkid87: you'll have GL 4.5 in the version string though
22:13 coolkid87: no I mean without EXT/ARB suffix?
22:13 airlied: karolherbst: you do, but you may as well just go for 4.5, avoids bugs in the test suite
22:13 karolherbst: I see
22:13 airlied: some tests are dumb, and might use 4.5 features in 4.4 tests
22:13 karolherbst: ohh, understood
22:13 imirkin_: karolherbst: right now we fail at 3d images for fermi/kepler. that's kind of a big one.
22:13 airlied: the test suite is pretty much only aimed at 4.5
22:13 karolherbst: what's the best way to run the tests?
22:13 karolherbst: imirkin_: I see
22:13 airlied: I used to run them from piglit, but I haven't in ages
22:14 karolherbst: not that I have time to fix those issues, but maybe I would use this as an opportunity to do more on the mesa side
22:16 imirkin_: karolherbst: i've tried running the open-sourced thing
22:16 imirkin_: but i get tons of errors that are test suite bugs
22:16 airlied: imirkin_: which ones?
22:17 karolherbst: imirkin_: yeah, I tried that as well and it was super messy to use
22:18 imirkin_: airlied: i don't remember ... i filed a bug about one of them
22:19 karolherbst: imirkin_: by the way, if we/I/you have time, we/I/you should discuss/think about how to make this nice and clean: https://github.com/karolherbst/mesa/commit/96001592997c465450ee12514809676a5129e341
22:19 imirkin_: airlied: and then iirc nha fixed it. but there were tons of issues.
22:20 imirkin_: karolherbst: nah, we want that either way
22:20 karolherbst: ohh by the way, I will move into the center of hamburg over the next weeks, so whenever somebody of you want to come to hamburg, soon would be a good chance and I would have enough space to host a few people :D
22:20 imirkin_: karolherbst: if dst == src2 then nv50 gets to use the short encoding
22:21 karolherbst: well yeah, but you weren't happy about the rsult of the pass
22:21 karolherbst: https://github.com/karolherbst/mesa/commit/6fb3a66353f056fbf2d6702995b4dc29597ccb9b
22:21 imirkin_: oh, for nvc0
22:21 karolherbst: yeah
22:21 karolherbst: like I care about anything else
22:21 karolherbst: :D
22:21 imirkin_: for nvc0 it needs to be a totally different condition
22:22 imirkin_: i think
22:22 karolherbst: right
22:22 karolherbst: and the WIP commit was "good enough" to get the gpr changes in check
22:22 imirkin_: so ... right. there you need ... hrm.
22:22 karolherbst: but there needs to be a better check I assume
22:22 imirkin_: anyways, will think about it later.
22:22 imirkin_: busy now, sorry
22:22 karolherbst: :) k
22:23 imirkin_: airlied: https://github.com/KhronosGroup/VK-GL-CTS/issues/37
22:23 imirkin_: that kind of issue tells me that (a) no one's seriously run this and (b) no one cares about the bugs i file
22:24 imirkin_: so my plan was to give the test suite another 6 months or so until people who cared had ironed out the issues
22:26 airlied: imirkin_: oh that is just normal CTS behaviour
22:26 airlied: that test probably passes fine on nvidia
22:26 imirkin_: airlied: really? #version 330 and tess stuff?
22:27 airlied: imirkin_: never underestimate the power of binary drivers
22:27 imirkin_: either way, i don't have the time or, more importantly, desire to deal with BS like that.
22:28 airlied: imirkin: okay whatever, seems like finding the obviously nouveau fault failing tests isn't that hard a job
22:29 imirkin_: it felt like a pretty low SNR =/ some legit stuff too every so often - i've sent patches to fix it up when possible
22:29 imirkin_: unfortunately there's no test list and i haven't been able to generate one
22:29 imirkin_: and nouveau tends to die half-way through a run
22:29 imirkin_: without a test list i can't use the piglit runner
22:29 airlied: I'll see if I can get piglit to run it, since it used to work
22:29 imirkin_: (the test it dies on, invariably, passes just fine on its own)
22:30 imirkin_: there's def an issue since running a bunch of tests together i can repro the crash
22:30 imirkin_: anyways, i doubt we'll have an impl that passes everything
22:31 airlied: ah seems like a nouveau bug, maybe once ben fixes multithreads stuff will work nicer :)
22:31 imirkin_: doubtful - it's single-threaded
22:31 imirkin_: i couldn't rule out a test suite bug either
22:33 imirkin_: i tend to get discouraged pretty quickly by stuff like that ... the payoff seems quite low as well
22:34 airlied: other than having GL4.5 :)
22:34 airlied: which seems more useful than having GLES3.1
22:35 airlied:sees if I can build it
22:36 imirkin_: dEQP seems like it has much higher quality tests though
22:36 imirkin_: imho it's a much better use of time analyzing those failures
22:38 imirkin_: it does look like the intel/amd people are fixing it up though
22:38 imirkin_: so perhaps in a few months' time it'll be closer to workin
22:42 Lyude: where's that trello board you guys use for keeping track of nouveau's TODO? wanna figure something out to work on
22:42 imirkin_: https://trello.com/b/ZudRDiTL/nouveau
22:42 imirkin_: check with someone before grabbing an item
22:43 karolherbst: Lyude: allthough I always recommend to start with issues which annoy you most
22:44 Lyude: hehe, normally I would but I haven't actually used a nvidia card for my machines in a while now
22:44 Lyude: I ended up switching to amdgpu
22:44 imirkin_: wise move.
22:44 karolherbst: imirkin_: do you think I can mark this as done? https://trello.com/c/RcF2wkYI/11-fixing-kepler-engine-reclocking
22:45 karolherbst: there are still issues,but I assume they get triggered on <5% of all kepler/maxwells
22:45 imirkin_: i have no clue. feel free to close the card and reopen a more targeted one for the reamining issues. or just close it entirely.
22:46 imirkin_: oh, i guess i added the card. but i don't think i had something specific in mind.
22:46 karolherbst: k, will make more specilized ones
22:47 karolherbst: https://trello.com/c/E1w00aYi/175-weird-engine-clocking-issues-on-high-frequencies-kepler
22:48 imirkin_: too generic
22:48 imirkin_: and inactionable
22:48 karolherbst: I've added a description
22:48 imirkin_: are there specific bug reports, cards, etc?
22:48 karolherbst: no idea
22:48 karolherbst: only thing I know is, it happens on higher frequencies
22:48 imirkin_: what makes you think this happens?
22:48 karolherbst: because it happens
22:48 imirkin_: to who?
22:48 karolherbst: users here in IRC
22:49 karolherbst: I know of those issues
22:49 imirkin_: who are using kernel 4.10?
22:49 karolherbst: and they are related to the frequencies
22:49 karolherbst: or my branches, yes
22:49 karolherbst: I investigated enough to be sure, there are some issues
22:49 imirkin_: i'd recommend entering as much info as possible into that card
22:50 karolherbst: better now?
22:51 imirkin_: better.
22:53 karolherbst: it's even that odd, that adding 23MHz makes the gpu super unstable
22:53 karolherbst: even adding like 0.2V doesn't change anything
22:54 karolherbst: so I think this is related to the PLL code and not volting
22:54 karolherbst: or something else
22:55 imirkin_: yeah, could be running PLL out of rane
22:56 karolherbst: or something funky
22:57 karolherbst: pitty is, we don't have access to such a GPU :/
22:58 karolherbst: skeggsb: how much does it cost to send some GPUs from you to mupuf_? last time you told me you have some keplers with reclocking issues?
22:58 Lyude: imirkin_: mind if I take this? https://trello.com/c/Fc6BO64w/151-gm200-nv-fill-rectangle i'm down to work with rectangles
23:00 imirkin_: hehehe
23:00 imirkin_: someone else was eyeing that one
23:01 karolherbst: claim it before somebody else claims it!
23:01 mupuf_: Lyude: hehe
23:01 mupuf_: karolherbst: ok, about to start the review!
23:01 karolherbst: mupuf_: awesome
23:01 imirkin_: Lyude: how much GL experience do you have?
23:01 karolherbst: mupuf_: we need some GTX 650 and 750 Ti GPUs by the way, try to keep an eye open
23:02 mupuf_: Back home from the language club, torturing myself to keep a discussion going in Finnish.
23:02 Lyude: imirkin_: not much if any, was hoping to pick some up through this
23:02 imirkin_: good idea
23:02 imirkin_: that's how i learned about GL
23:02 airlied: imirkin_: got it to produce a test case list in tree, just not from piglit
23:02 mupuf_: oh, yeah, I was eyeing the 750 Ti, for the higher memory clock, but they were a little to expensive
23:02 karolherbst: mupuf_: they most likely cause some reclocking issues with nouveau
23:02 karolherbst: that's why I want one
23:02 imirkin_: airlied: it always crashed when i tried.
23:03 mupuf_: karolherbst: I really would like Ben to share some of his GPUs, along with everyone here in this community.
23:03 karolherbst: yeah
23:03 mupuf_: so as I don't have to have everything here
23:03 karolherbst: he told me, he has some with issues even using my branch back then
23:03 airlied: ./glcts --deqp-runmode=txt-caselist --deqp-case=GL45-CTS*
23:03 imirkin_: Lyude: anyways, i'd say - go ahead and take it. most people who show up here saying they want to do one or another thing end up never appearing again.
23:03 airlied: produces the GL45 list
23:03 karolherbst: airlied: executed from where?
23:04 imirkin_: Lyude: so i (a) am not going to hold on to the item, and (b) won't be surprised if you disappear never to be heard from again
23:04 Lyude: hehe
23:04 karolherbst: airlied: found it
23:04 karolherbst: airlied: "FATAL ERROR: Failed to initialize dEQP: Unsupported surface type at tcuX11GlxPlatform.cpp:655" ;)
23:04 imirkin_: karolherbst: force GL 4.5
23:04 karolherbst: hsw
23:05 karolherbst: or do I still have to force it?
23:05 imirkin_: airlied: can you post that list somewhere convenient?
23:05 karolherbst: glxinfo tells me 4.5 on my intel gpu
23:05 imirkin_: karolherbst: then it's something else.
23:05 airlied: https://people.freedesktop.org/~airlied/scratch/GL45-CTS-cases.txt
23:05 imirkin_: thanks
23:05 airlied: karolherbst: switch to x11-egl target
23:06 imirkin_: dunno how often you sweep, but mind keeping it there for at least a week or so?
23:06 airlied: or at least I always run it in that mode: )
23:06 karolherbst: hum
23:07 airlied:leaves it running against master
23:09 imirkin_: airlied: thanks. on kepler?
23:10 airlied: yes nve7
23:10 imirkin_: cool.
23:11 imirkin_: i did fix a bunch of stuff, but it was mostly one-offs
23:12 airlied: my life is currently paused on various cts runs
23:13 imirkin_: sounds like you've made poor life choices :p
23:13 airlied: or I need more computers
23:13 imirkin_: i think we all know the correct answer here :)
23:20 airlied: imirkin_: GL45-CTS.arrays_of_arrays_gl.InteractionFunctionCalls2 is the test that takes ages, seems to annoy the compiler
23:21 karolherbst: airlied: okay, and how do I start those tests now?
23:21 imirkin_: airlied: bleh. that also reminds me that i have real issues to fix in the compiler.
23:21 airlied: karolherbst: you can run it as-is or via piglit
23:21 imirkin_: might require a rethink of some very sad parts of the compiler
23:22 imirkin_: which in turn makes me sad.
23:22 airlied: as-is external/openglcts/modules; glcts --deqp-case=GL45-CTS*
23:22 imirkin_: --deqp-visibility=hidden
23:22 airlied: piglit get the test list hack tests/cts_gl45.py to use it, edit piglit.conf to point at gl_cts binary
23:22 airlied: glcts
23:22 karolherbst: airlied: cool, thanks
23:26 imirkin_: Lyude: feel free to ask questions here about various issues you run into. i know you have a lot of low-level knowledge about GPUs/displays and their APIs, but are probably missing some of the higher level bits.
23:26 Lyude: imirkin_: I was just about to ask which source file I should start looking at?
23:26 imirkin_: let me pull up what i told the other person
23:27 imirkin_: Lyude: https://people.freedesktop.org/~cbrill/dri-log/?channel=nouveau&date=2017-03-03
23:28 imirkin_: look at my conversation with vliaskov towards the top
23:28 Ako: Hi
23:29 Lyude: imirkin_: gotcha, thanks!
23:30 Ako: I have a question. Does nouveau driver support hdmi audio for Geforce 320m chipset?
23:30 imirkin_: Ako: yes. but there was a bug where it didn't work properly. should be fixed now.
23:31 Ako: Thanks! I'll try the latest drivers then
23:31 imirkin_: what kernel are you on?
23:32 Ako: Hmm... I have just installed Linux mint 18.1
23:33 imirkin_: Ako: run "uname -r"
23:34 airlied:restarts with -c this time
23:34 imirkin_: the fix i'm thinking of went into kernel 4.10, and should be in the process of being backported to various stable trees
23:34 imirkin_: airlied: you really want skeggsb's fixes if you do that
23:35 imirkin_: (don't think they made it into 4.10... did they?)
23:36 karolherbst: mupuf_: is it normal, that you look at patches in random order? Just curious, because it always seems that way, when you review stuff
23:36 airlied: don't think so, not sure what kernel this is even on, probably 4.8
23:36 imirkin_: Lyude: you might also look for changes that added a simple ext... i might recommend EXT_polygon_offset_clamp - it was a simple one that still managed to touch a bunch of things.
23:36 mupuf_: karolherbst: random order? Must be your email provided
23:36 karolherbst: mhhh
23:36 karolherbst: seems in order for me
23:36 imirkin_: Lyude: the gallium bits were already in place for it, but it should give you a good idea of the types of changes ot mesa core that need to take place to support a new ext.
23:37 mupuf_: well, I do make multiple passes, first an overview and then I dig into it
23:37 karolherbst: k :)
23:37 karolherbst: just wondering
23:37 imirkin_: airlied: ah yeah, you should expect fireworks with -c :)
23:37 karolherbst: mupuf_: comming to hamburg soon? I have enough space now in a few weeks :D
23:37 imirkin_: turns out locking is important. who knew.
23:37 mupuf_: karolherbst: that is tempting!
23:38 karolherbst: 10 minutes away from the center of hamburg :D
23:38 Ako: 4.4.0.53
23:38 imirkin_: Lyude: you might also talk to mareko, iirc he wanted to enable some NV_fill_rectangle-ish ext for AMD
23:40 mupuf_: karolherbst: perfect for a good reeperbahn night out :D
23:40 imirkin_: Ako: doesn't look like the 4.4 series has received that patch yet
23:41 karolherbst: mupuf_: exactly :D
23:41 imirkin_: Ako: er wait. what? looks like it's in kernel 4.4.48
23:41 karolherbst: at night you usually always get home, sometimes the time between 2 and 4 o clock might be hairy, but meh
23:42 imirkin_: Ako: not sure what "4.4.0.53" means - must be some distro-specific numbering
23:42 Ako: I have tried proprietary nvidia drivers and I can get hdmi sound
23:43 imirkin_: not sure how that's related
23:43 karolherbst: meh "glcts: /home/karol/Dokumente/repos/deqp/external/openglcts/modules/gl/gl4cPipelineStatisticsQueryTests.cpp:919: Unknown function: Assertion `0' failed."
23:43 imirkin_: karolherbst: yeah, that sounds familiar.
23:43 imirkin_: it's an error in the test.
23:43 imirkin_: like i said, clearly nobody's run this test suite.
23:43 airlied: karolherbst: is that from the VK-GL-CTS repo?
23:44 karolherbst: too late to get piglit running those tests
23:44 karolherbst: airlied: yes
23:44 imirkin_: airlied: yeah, that happens with VK-GL-CTS
23:44 imirkin_: the pipeline stats test doesn't handle compute queries properly internally iirc
23:44 imirkin_: it has a big switch for all of them and then misses stuff
23:44 imirkin_: or ... something along those lines.
23:45 imirkin_: airlied: wait, are you running the internal repo and not the public one? i'm sure the internal one works :)
23:46 airlied: imirkin_: no I'm on the public one now, hadn't ran either in a while
23:47 Lyude: imirkin_: just to make sure: I can use a GM204 for testing that right?
23:47 imirkin_: Lyude: yes
23:47 Lyude: just wanted to make sure
23:47 karolherbst: mupuf: I even ran the super awesome linux check_patch tool, and it didn't complain about those comments :(
23:47 Ako: 4.4.0-53-generic - that is the exact printout from "uname -r"
23:47 imirkin_: assuming it generally works with nouveau, it should be fine.
23:47 karolherbst: silly tool
23:48 imirkin_: Ako: hm ok. do you have your hdmi screen plugged in?
23:48 Lyude: it doesn't, but I've still got skeggsb's hack he gave me to cut off the 512MB of ram nouveau can't initialize :P
23:48 Ako: Yes. And I see video on it
23:48 imirkin_: Ako: ok. can you....
23:48 imirkin_: (p.s. is there some way to tell wtf kernel this is? http://packages.ubuntu.com/trusty/kernel/linux-image-4.4.0-53-generic)
23:49 imirkin_: Ako: find /proc/asound -name eld\*
23:50 imirkin_: Ako: by the looks of it, 4.4.0-53 does NOT include the fix i had in mind
23:51 imirkin_: assuming i'm reading this right, it roughly corresponds to kernel 4.4.30
23:52 imirkin_: Lyude: well you'll obviously need a generally-works-with-nouveau GPU in order to hack on mesa stuff :)
23:52 imirkin_: Lyude: any GM20x should have the feature in question
23:52 Lyude: no it does work! it was the gtx 970 with the 4GB of ram
23:52 Lyude: for now it works perfectly if I limit it to 3.5Gb
23:52 imirkin_: oh, goodie
23:53 imirkin_: can you post a patch to the bug where people are desperately trying bad things?
23:53 Ako: For the command above I get: /proc/around/card0/eld#3.0 4.0 5.0
23:53 imirkin_: Lyude: https://bugs.freedesktop.org/show_bug.cgi?id=94990
23:53 Lyude: haha, sure but I think skeggsb has an actual one working now
23:53 imirkin_: Lyude: yeah, but that's harder to backport
23:53 Lyude: ah, gotcha
23:54 imirkin_: keep in mind the average skill level of the people trying this.
23:54 imirkin_: Ako: ok... can you cat all 3 of them and pastebin the results?
23:58 Ako1: cat /proc/asound/card0/eld#3.0 monitor_present 0 eld_valid 0
23:59 imirkin_: pastebin the results of the one where monitor_present 1 :)
23:59 imirkin_: (pastebin.com)