01:32hakzsam: imirkin, 2D array images are fixed now :)
01:33hakzsam: unfortunately, I don't have my IRC password to answer on #dri-devel
01:33hakzsam: will do tonight because I'll be back at home :)
01:34hakzsam: so the next step is 3D images
04:28karolherbst: hakzsam: got some time to review the two newest patches from this branch? https://github.com/karolherbst/mesa/commits/to_upstream
06:28vedranm: pmoreau: clinfo now works fine (no missing capability) on G96 with Mesa 11.2.0 rc3
06:28vedranm: not sure if it is intended, but cool regardless
06:35tcermiat: what's new in 11.2?
06:36vedranm: tcermiat: no idea, but clinfo works
06:37vedranm: nothing else, and pmoreau told me earlier it is not supposed to anyway
06:37vedranm: tcermiat: try it yourself if you want
06:37vedranm: I get Error: Failed to build program executable!
06:37vedranm: invalid source
06:39RSpliet: vedranm: an OpenCL C to nouveau-IR compiler is a work-in-progress, steps are made but it's premature, non-functional and doesn't all sit in the mainline Mesa tree
06:39vedranm: RSpliet: thx for the info
06:40RSpliet: tcermiat: the release notes for 11.2 can be found in the Mesa git tree - as the WIP-release notes for 11.3 ( https://cgit.freedesktop.org/mesa/mesa/tree/docs/relnotes/11.3.0.html?id=1e8435ce0cce671024ebf9c5465ea8bdcb563b69 )
06:40tcermiat: 11.3 already out?
06:41RSpliet: 11.2 release notes: https://cgit.freedesktop.org/mesa/mesa/tree/docs/relnotes/11.2.0.html
06:42tcermiat: please speed up on fermi reclocking
06:43RSpliet: tcermiat: please give me an endless amount of time
06:43mupuf: tcermiat: that's a bit rude, isn't it? :s
06:44tcermiat: not at all
06:45tcermiat: it's not an order
06:46vedranm: tcermiat: dunno about you, in my books nouveau is going unrealistically fast
06:46tcermiat: i agree
06:46tcermiat: something happened last months
06:49vedranm: tcermiat: I am still amazed there is a tablet shipped with nouveau instead of nvidia driver
06:49vedranm: so, it is happening
06:49tcermiat: nvdia has no support for tablets?
06:49RSpliet: vedranm: don't be fooled though, I assume they did use a closed-source userspace but made it interface with the nouveau kernel component
06:50RSpliet: tcermiat: I take it vedranm refers to the Google Pixel C, which ships with Linux and nouveau *by default*, the *by default* bit is new and exciting ;-)
06:51tcermiat: maybe they did not get source of nvidia driver
06:51vedranm: RSpliet: I do
06:51tcermiat: or did not want to pay for it
06:51RSpliet: Google doesn't have to pay for that
06:51vedranm: RSpliet: but they do not ship NVIDIA's proprietary stack, but custom one?
06:52tcermiat: RSpliet: why?
06:52tcermiat: maybe they own nvidia
06:52RSpliet: tcermiat: because they already pay for the hardware. If NVIDIA doesn't provide the drivers they'd just as happily use Qualcomm HW
06:52vedranm: RSpliet: it's NVIDIA's OpenGL indeed
06:53vedranm: kinda like amdgpu stuff
06:53vedranm: but cool regardless
06:53RSpliet: vedranm: I think OSS kernel module, NVIDIA proprietary user-space. They did a similar set-up with L4T, but with a special "tegra" kernel space component
06:53RSpliet: the latter they likely want to drop support for on the long run because nouveau is nicer ;-)
06:53vedranm: of course :-)
06:54RSpliet: tcermiat: anyway, wrt fermi reclocking. It's a known issue, I take it you had no intention of offending anyone, so let me just inform you that there is work in progress
06:54RSpliet: nothing testable yet
06:55tcermiat: i was told it can require one year
06:56RSpliet: correct, but it depends mostly on volunteer effort (in this case, me). I'm a full-time PhD in a not-really related area, and as such I don't have a lot of spare time to work on it atm... it fluctuates
06:57tcermiat: nvidia ignored my request to give specs for fermi
06:57RSpliet: they're not very forthcoming with details on the memory controller. Possibly because it might contain 3rd party IP
06:58RSpliet: but that last bit is speculation
06:59RSpliet: it's not impossible to get working w/o specs as proven with our 2nd gen Tesla and Kepler/Maxwell implementations, it just requires a lot of time
06:59vedranm: RSpliet: not-really related area?
06:59RSpliet: vedranm: computer architecture, under the lowRISC umbrella
07:00vedranm: RSpliet: RISC-V niiiice
07:00RSpliet: I have an interest in parallel accelerators, but not reverse engineering NVIDIA GPUs for my research... well, not really
07:01vedranm: will lowRISC SoC have any GPU or FPGA?
07:02RSpliet: in it's current form don't count on a GPU/parallel accelerator. Nothing is set in stone, but it depends heavily on community effort
07:03RSpliet: we're currently manually pushing a fresh snowball, let's see how much snowflakes get on board during the ride :-)
07:04vedranm: RSpliet: fingers crossed
07:06RSpliet: nothing is unthinkable, parts of a GPU have emerged in the open source world recently (Nyanu, hwacha...)
07:07RSpliet: but as with fermi reclocking, it all depends on how much time individuals have to spend on actually making it happen
07:07tcermiat: that's good
07:12vedranm: RSpliet: you mean Nyuzi? or is there another one?
07:13vedranm: RSpliet: this hwacha thing is interesting, thx for the info
07:15RSpliet: vedranm: yes, that one. It changed names from "Nyami" recently apparently
07:16RSpliet: Nyanu is a weird made-up name that was stuck in the back of my head, sorry for the confusion :-P
07:28karolherbst: RSpliet: maybe I will find some time to hack on my fermi a bit after I am done with the volting stuff...
07:30karolherbst: but I would say it also depends on maxwell a bit, because I think it would be worth to get this stable as well. I doubt there are any differences to kepler, but who knows
07:30karolherbst: having kepler/maxwell working is somewhat higher priority than fermi for me
07:33RSpliet: karolherbst: that's fine
07:34karolherbst: yeah I know, but I also said like two months ago I will look at fermi next, but didn't really :D
07:34RSpliet: it's probably easier for the both of us if I first try and get " my" Fermi(s) working, and then try to fine-tune things with yours
07:34RSpliet: there's enough other great work you've done
07:35karolherbst: well mine is a DDR3 one though, I think it would be easier for you to get gddr5 ones working ;)
07:35RSpliet: despite my sometimes direct words - I'm merely trying to squeeze the last 5% of quality out of you :-P
07:37karolherbst: yeah I know
07:38karolherbst: it is just a bit sad, that ben has so little time :/ most of my stuff is just stalled until he has some
07:38RSpliet: one of my major frustrations with the OSS (and academic) community when I first started were the sometimes harsh sounding words from peers
07:39karolherbst: yeah, but I was aware of that before already
07:40RSpliet: they're hardly intentional, but still it took me a few FOSDEMs and an XDC to separate the tone of voice from the perception of peers
07:41karolherbst: ohh I know what I could do in the meantime, find a way to check for the highest supported pixel clock on fermis for hdmi
07:41karolherbst: I think there is a little piece? and my fermi laptop has the HDMI port connected to nvidia
07:47karolherbst: ohh mhh, might need a display which supports that high resolutions as well...
10:37hakzsam: karolherbst, maybe later or tomorrow
10:39karolherbst: no problems
11:27yoshimo: https://imgur.com/4o3QboJ could that be the drivers fault?
11:27yoshimo: all text is garbage
11:46karolherbst: yoshimo: this is wow right?
11:47karolherbst: hard to say, could be wine, could be mesa
11:47karolherbst: yoshimo: start it with LIBGL_ALWAYS_SOFTWARE=1 and see if something is different
11:47yoshimo: it is, although hardly to see
11:48karolherbst: yoshimo: also, did you start wow in opengl mode?
11:50yoshimo: no d3d9 as dx11 is not ready for production on wine yet
11:50yoshimo: your variable seems to fix rendering
11:51yoshimo: still looks not 100% but it is readable again
11:53karolherbst: then it is a issue within nouveau
11:53karolherbst: well mesa, but yeah
11:55yoshimo: so how do we proceed from there?
11:55Weaselweb: I did a piglit run on my NVAA on a 4.5 kernel all but one glean tests fail with stderr message: can't open display :0
11:56Weaselweb: but if I run that test manually afterwards it succeeds.any idea what's wrong here?
11:57karolherbst: yoshimo: so you run it explciitly in opengl mode?
11:57yoshimo: not yet, i have gxapi set to d3d9
11:57yoshimo: i can try opengl sure
11:57karolherbst: opengl will work most likely
11:58karolherbst: it also gives you much better performance
11:58karolherbst: but if you want you can apitrace wine in d3d9 mode
11:58karolherbst: and open a bug
12:00yoshimo: opengl crashes with access violation so looks like apitrace is necessary
12:01karolherbst: yoshimo: which version has your client?
12:02yoshimo: 6.2.4 as of todays patch
12:02karolherbst: maybe you find here something: https://appdb.winehq.org/objectManager.php?sClass=version&iId=32314
12:03karolherbst: but if it crashes you should check where and maybe find out why
12:04yoshimo: so far people have connection issues with the new patch, but i can't find people with garbled text
12:06karolherbst: yeah, most likely a nouveau issue then
12:06yoshimo: maybe i found one let me check
12:06karolherbst: I also had some corrupted textures with oblivion under wine...
12:06karolherbst: yoshimo: could you check if the corruption looks different each time and whether sometimes it looks okayish?
12:07karolherbst: you might have to try several times
12:07karolherbst: but also you should get the opengl renderer working, because that's usually what you want to use
12:10yoshimo: it looks kind of okay with just small pieces missing, but if i move the window or maximize it, it gets worse
12:10karolherbst: yeah okay, looks like the issue I also have
12:10karolherbst: maybe something in wine changed
12:11karolherbst: yoshimo: you also run wine 1.9.5?
12:11karolherbst: could yuo try out something older?
12:11karolherbst: like 1.8 or even older?
12:11karolherbst: mhh well I could to, but compiling takes forever
12:12karolherbst: but I could try to bisect it then
12:13yoshimo: i can only do a poor mans binary search if i don't end up in a dependency hell for the binaries
12:13karolherbst: well steps doesn't matter
12:14karolherbst: I just want to know if an older version works better
12:14karolherbst: if 1.8 works fine, this is all I need
12:14yoshimo: my crash btw is https://bugs.freedesktop.org/show_bug.cgi?id=91600#c5
12:15karolherbst: compat profiles
12:15karolherbst: what the....
12:15karolherbst: why even bother
12:15karolherbst: they can't even use this on mac os x
12:16karolherbst: yoshimo: k, then we cheat a little
12:17karolherbst: yoshimo: start in in opengl mode with those variables: MESA_GL_VERSION_OVERRIDE=4.1C MESA_GLSL_VERSION_OVERRIDE=410
12:17yoshimo: with 3.3 the game complains about deprecated stuff, let me try 4.1
12:18karolherbst: the C is important though
12:18karolherbst: mhh maybe not
12:19karolherbst: mesa might give a core profile either was
12:21yoshimo: i could go back to 1.7.55 and try there
12:22karolherbst: yeah sounds good
12:22karolherbst: but does forcing 4.1 work?
12:25yoshimo: no, crashes
12:34sarnex: karolherbst: still getting the lockups if nouveau isnt blacklisted
12:44yoshimo: karolherbst: wine-64 --version is 1.7.55 staging and wine64 wow-64.exe still gives broken text
12:46yoshimo: each window resize changes the broken text around
12:46karolherbst: mhh okay
12:59yoshimo: i tried an apitrace but it is 103mb in size after compression, can i shrink that somehow?
13:10yoshimo: so that is the wine one http://workupload.com/file/rmdaFRu0 , original coming later
13:46karolherbst: Yoshimo: yeah, I also have some corrupted stuff, but not as much
13:47karolherbst: sarnex: yeah well, we need output of dmesg for that, otherwise we have no clue what is going wrong
13:47Yoshimo: do you need the same thing from within windows too to compare?
13:48karolherbst: I doubt it
13:49Yoshimo: any idea what is going on there?
13:53karolherbst: gnurou: Is it currently possible to get the gpu load on tegras when running nouveau? I have no idea what kind of interface the nvidia driver uses for this on desktop GPUs and was wondering if there are any plans for tegra to do so on nouveau?
14:21sarnex: karolherbst: https://bpaste.net/show/0ee3f7d16db7
14:21sarnex: search for "stall"
14:21sarnex: the log is like 3 boots combined sorry about that
14:22karolherbst: there is nothing nouveau related
14:22sarnex: yeah but i have literally never seen it with nouveau blacklisted
14:22sarnex: and i unblacklisted it last night after our conversation and i got the stall today
14:22karolherbst: mhh I really need the entire dmesg outpuit
14:22karolherbst: not only parts
14:23sarnex: do you know what file that would be in after you reboot
14:23karolherbst: the stalls don't have any usefull backtrace too
14:23sarnex: yeah last time you told me when it happens i should try to dump a kworker
14:23karolherbst: sarnex: depends on your system logger
14:24sarnex: yeah it seems it has multiple days in one file :(
14:24sarnex: i have no idea how to try to debug this
14:25sarnex: but i had it blacklisted for months and it never happened, and on the first boot without it blacklisted i got the stall
14:25karolherbst: yeah well, but we really need to know where it stalls and why
14:25karolherbst: you don't even use the nvidia gpu, right?
14:27sarnex: nope i have 3 GPUs in my system, radeonsi,r600, nouveau
14:27sarnex: i only use radeonsi
14:27karolherbst: any idea what threaded-ml is?
14:27sarnex: nope sorry
14:29karolherbst: is there anything gpu related you do before that happens?
14:30sarnex: no i literally got back from class and tried to open a browser and it didnt work, and i tried to open a new terminal and it didnt work, so i used an existing terminal to dmesg and saw the stall messages
14:30sarnex: but in the past it's happened when i was using the system, so i dont think its power saving or anything
14:30karolherbst: sarnex: could you check if you have a thread running called threaded-ml?
14:31sarnex: can you tell me how sorry lol
14:31sarnex: also now nouveau is blacklisted again let me know if that affects it
14:31karolherbst: mhh you could use htop and display all userspace and kernelspace threads and filter
14:31karolherbst: doesn't matter
14:31karolherbst: this thing should run also without nouveua loaded
14:31sarnex: trying it now thanks
14:32karolherbst: maybe it is some kind of gpu accelerated thing and it uses the nvidia gpu too
14:33sarnex: no results for threadded-ml
14:33sarnex: i disabled hide kernel threads
14:34karolherbst: sadly I don't find whatthat is...
14:34karolherbst: yeah I hoped some binary has a string "threaded-ml" inside it ..
14:34sarnex: ah ok
14:35sarnex: yeah this is a very strange error
14:37sarnex: let me know if you have any ideas i appreciate the help
14:43pmoreau: vedranm: Weird… IIRC, hakzsam sent a patch to remove advertising compute on Tesla cards.
14:43hakzsam: what's the problem?
14:44pmoreau: vedranm: Hopefully, I should be able to work almost full time on SPIR-V to NV50 IR in two weeks, and try to get loops and conditionals to an almost working state.
14:45pmoreau: hakzsam: Apparently clinfo is happy on G96, with Mesa 11.2-rc1, which it shouldn't
14:45hakzsam: yes, it should not
14:46hakzsam: that's weird, but whatever opencl just won't work :-)
14:46hakzsam: btw, I did a bunch of progress with arb_shader_image_load_store
14:47hakzsam: (fermi only for now)
14:47pmoreau: Nice :-)
14:47pmoreau: I might have a look at plugging in atomics while working on the out-of-SSA pass.
14:48hakzsam: in a couple of weeks, we will have GL 4.2 on Fermi :-)
14:48pmoreau: Textures could also be interesting, but, most likely more work.
14:53karolherbst: yay 4.2
15:33vedranm: pmoreau: great :)
15:33vedranm: hakzsam: I don't mind ;)
15:34vedranm: hakzsam: yay 4.2 on Fermi, how's Kepler and Maxwell?
15:34hakzsam: I need time :-)
15:34hakzsam: but it's on my todolist
15:35sayih: hakzsam: do u mean you will hack reclocking fermi?
15:36hakzsam: sayih, nope, that's unrelated
15:37hakzsam: I'll just bump the GL version to 4.2 once the last extension is done
15:37hakzsam: (ie. arb_shader_image_load_store)