00:09 Lyude: RSpliet: last time I had things working everything seemed to work perfectly on kepler+
00:09 Lyude: erm, s/kepler+/kepler/
00:09 Lyude: reclocking worked fine as well
00:09 Lyude: and heavy loads, etc.
00:20 imirkin: Lyude: is this in relation to your current issues?
00:20 Lyude: imirkin: no, they were asking earlier about the power gating stuff I'm planning on picking up again once I get the time
00:20 imirkin: oh
00:20 imirkin: yeah, that'd be nice
07:20 jamm: hakzsam: hi
07:21 jamm: i was away for a while (moved) XD
07:22 jamm: now i'm back, i'm wondering if you could take a look at my patches again :)
10:48 karolherbst: imirkin: what do you think about this? I really don't like "casting to an array to 'optimized' code, but nobody knows what the hell is going on" type of code. https://github.com/karolherbst/mesa/commit/2715bc616eb301d190e25db71a11708e5e982dbd.patch
10:49 karolherbst: I could also leave it out for now and only set pipeline_statistics.cs_invocations to 0
12:49 wivs: how can i debug visual errors on nouveau that are been produced with kde plasma?
12:50 wivs: i would like to make a propper bugreport for this case
12:59 karolherbst: I've updated the CTS trello card a bit: https://trello.com/c/qtJ5jkZh/13-cts-master-45-status
12:59 karolherbst: wivs: usually apitrace works well
12:59 karolherbst: wivs: problem with plasma is, that you most likely doesn't want to trace _everything_ so first you should pinpoint which binary is affected
13:00 wivs: karolherbst: how can that be done?
13:00 karolherbst: no idea, guessing I guess. It's a little bit problematic with plasma, because it has like hundreds of components
13:01 karolherbst: wivs: maybe you should open a bug report in the kde bug tracker, chances are those aren't nouveau issues
13:01 karolherbst: I also see artefacts with intel
13:04 wivs: i also have problems on gnome with wayland
13:04 karolherbst: also gnome isn't bugfree, neither is wayland
13:05 karolherbst: it's just super hard to pinpoint in big desktop environments
13:05 karolherbst: usually the devs will tell you what to test or which component might be affected
13:06 wivs: what and how should i do that all? i am a user and just seeing those bugs. I can make a youtube video if that is required
13:06 karolherbst: and if you are able to create an apitrace of that component, which shows the bug, we can look into it
13:06 karolherbst: a youtube video won't help
13:06 karolherbst: I would ask for support in the plasma channel first or open a bug there first
13:07 karolherbst: maybe they can even give information on how to launch components directly to ease debugging
13:10 wivs: karolherbst: what desktop environment are you using. i can try that one and then report if i have the same issue there
13:10 karolherbst: plasma
13:10 karolherbst: but with intel, because I have a laptop
13:15 karolherbst: wivs: what version of mesa are you using?
13:18 wivs: 13.0.6
13:18 karolherbst: that's quite old
13:18 karolherbst: what GPU?
13:18 wivs: its latest stable release: https://packages.debian.org/search?keywords=mesa&searchon=names&suite=stable&section=all
13:18 wivs: GPU: nv42
13:19 karolherbst: for debian yes, but we are at 17.1.7 and 17.2 is near to be released
13:19 karolherbst: well 13 is "just" a year old though, but a lot of things change
13:19 karolherbst: maybe not for nv42
13:19 karolherbst: let me check
13:20 wivs: thanks
13:21 karolherbst: mhh, nv42 doesn't get much love regarding super modern desktop environments and it's pretty much seperated from newer chipsets
13:22 karolherbst: I think imirkin has such a GPU, but I doubt he would run plasma just for fixing bugs.
13:22 karolherbst: wivs: best thing you could do is, to try to pinpoint the component causing the issue and record the issue with apitrace
13:23 wivs: karolherbst: you are a mesa developer?
13:24 karolherbst: kind of, but currently a rather minor one
13:25 wivs: could you push this bugfix to mesa stable release (17.1) and to recent development version (17.2)? https://bugs.freedesktop.org/show_bug.cgi?id=102349
13:25 karolherbst: I can't push to mesa
13:26 wivs: you probably know the people who can. would be nice if you could ping those
13:27 karolherbst: imirkin can
13:27 karolherbst: question is, if the patch already has production quality
13:28 tobijk: wivs: the one who wrote the patch should advertise it (mark it for stable), yet he is not satisfied with his own patch, so give the process time
13:28 tobijk: you can use a patched version in mesa for that time, with the half-ready patch :)
13:29 wivs: karolherbst: could you ask him if the patch has production quality?
13:29 karolherbst: wivs: I know that his isn't a proper solution to your issues, but plasma is a pretty heavy desktop environment and might be start to run worse on older GPUs from release to release. Lxqt may be a good replacement for older parts and it uses a lot of KF5/Plasma5 components as well. Never tried it myself though. I only know Xfce4, which I prefer on older machines
13:29 karolherbst: wivs: I am sure it hasn't
13:29 karolherbst: he called it "hackpatch2"
13:30 tobijk: wivs: it is stated that the patch is not production ready in the bug you linked
13:31 karolherbst: wivs: don't get me wrong, we want to fix all the bugs, the biggest problem is just, that nouveau doesn't have the man power for it, because it's mostly voluntarily and most devs only care about things affecting themselfes or interesting stuff
13:33 wivs: in this case its probably a mesa and not a nouveau bug, right? Are those visual errors also mesa problems and not nouveau errors?
13:33 karolherbst: well
13:34 karolherbst: parts of the entire nouveau stack are partly within mesa
13:34 karolherbst: mesa is an OpenGL implementation which contains a nouveau backend
13:36 wivs: karolherbst: when i start supertuxkart in window mode the whole screen goes crazy with visual artefacts. at some point the whole os then crashes and i cant change to tty any more
13:37 karolherbst: wivs: okay, this is easier to debug
13:39 karolherbst: wivs: for this you can create a bug report on the freedesktop buzgilla against nouveau directly
13:39 karolherbst: best if you could attach an apitrace, but this shouldn't be necessary, because supertuxcart is easy to get
13:52 wivs: karolherbst: https://pastebin.com/XKqM8xJa
13:54 wivs: https://pastebin.com/NTj1n6EA
13:57 karolherbst: mhhh
13:58 karolherbst: might be simply due to being out of memory
13:58 karolherbst: but I don't know anything about those old GPUs, so I can't say for sure
14:01 wivs: karolherbst: https://bugs.freedesktop.org/show_bug.cgi?id=102430
14:08 karolherbst: wivs: thanks
14:08 wivs: karolherbst: https://bugs.freedesktop.org/show_bug.cgi?id=102430#c1
14:08 karolherbst: mhh nvidia-drivers 384.69 seems to be untraceable with mmt :(
14:13 karolherbst: 384.59 works perfectly though
14:14 karolherbst: wivs: how much VRAM does your GPU have, 256 or 128MB?
14:16 wivs: 256
14:21 wivs: i wrote the ram size to this comment here: https://bugs.freedesktop.org/show_bug.cgi?id=102430#c2
15:31 karolherbst: mhh, intel does x = x*1+0; y=y*1+0, for the initial adjustments, sounds pretty boring
15:32 karolherbst: the other parameters are in global memory :/ this is gonna be fun
15:39 karolherbst: and it uses the one patch we don't use :(
15:40 imirkin: karolherbst: btw, i investiaged a lot of those varying locations failures
15:40 imirkin: i sent a couple patches, and a lot of bugs to VK-GL-CTS :)
15:41 karolherbst: yeah, I saw :)
15:41 karolherbst: I cherry-picked both of your patches already
15:42 imirkin: karolherbst: you could also just add the result->pipeline_statistics.cs_invocations = 0; at the end there
15:42 imirkin: doesn't really matter
15:42 karolherbst: yeah, I think I will just do this for now
15:42 imirkin: i kinda liked the old code, but wtvr
15:42 karolherbst: well... I see the point, but....
15:43 imirkin: karolherbst: untraceable, or demmt fails with 384.69?
15:43 karolherbst: ohh, demmt fails
15:44 imirkin: you need to specify a -m -- i assume you know that already
15:45 karolherbst: yeah
15:45 karolherbst: it wasn't the issue
15:45 imirkin: could be they made additional changes
15:45 karolherbst: more like it failed to decode certein regions
15:46 karolherbst: something with fail and a memory region of some sort
15:46 karolherbst: can give you the details in a minute
15:48 imirkin: nah, don't worry about it
15:48 imirkin: demmt (and mmt) require active maintenance, and there presently is none
15:49 imirkin: you can always convert the binary mmt trace to the old "dedma" format
15:49 karolherbst: I put it on my trello todo list :)
15:49 imirkin: which allows for much less accurate decodes, but is also a lot more fool-proof
15:50 karolherbst: ohh, okay
15:51 imirkin: the stuff you can see with demmt, you couldn't even begin to see with dedma. but dedma handles this kind of situation much better.
16:05 karolherbst: that 3d image bug is getting annoying. Nvidia indeed takes the false path, which we don't implement and I have no idea how important it is to have that path
16:05 karolherbst: it starts to look like a "lock yourself in a room for two weeks" kind of issue
16:05 karolherbst: probably the reason it wasn't implemented yet
16:05 imirkin: i doubt the path it takes matters
16:06 imirkin: i guess i dunno. hakzsam might remember what the diff between them is
16:06 karolherbst: well, it makes debugging just a little harder
16:07 karolherbst: at least I know that if we properly adjust out paths, we don't need de-tiling. I got pretty close to get it right for one layer, but this was more like turning on knobs and see what they do
16:08 karolherbst: knowing what those instructions do would help probably
16:08 imirkin: :)
16:08 karolherbst: also, that we pass always 0 as src2 to subfm might be suboptimal
16:11 imirkin: all that stuff was originally RE'd by calim
16:11 imirkin: before the dawn of time
16:12 imirkin: i believe hakzsam had to make very minor fixes to it when he finished up the images stuff
16:12 karolherbst: ohh, I see
16:12 karolherbst: I thought hakzsam did all himself
16:12 imirkin: he did do the maxwell stuff
16:12 karolherbst: I see
16:12 imirkin: which admittedly was vastly simpler, since you just give it coordinates ;)
16:13 karolherbst: yeah
16:13 karolherbst: I noticed that
16:13 imirkin: except it doesn't support 2d ms images, for that you need to fall back to manual
16:13 karolherbst: I found this: https://stackoverflow.com/a/30440589
16:13 karolherbst: I doubt this helps though
16:13 imirkin: and he just never implemented that because nobody cares about ms images
16:17 karolherbst: but I got interesting results toying around with the c15 cb
16:17 karolherbst: ohhh wait
16:18 karolherbst: mhh interesting
16:18 karolherbst: I think the last sueau might be some kind of offset
16:18 karolherbst: source
21:20 imirkin: karolherbst: by now i have either bugs filed or spec clarifications requested for all the varying locations failures
21:21 imirkin: [and i pushed one WIP commit to fix something else related]
21:51 anEpiov: any improvements in 4.12?
21:52 anEpiov: I wonder what
21:53 tobijk: anEpiov: working acceleration for pascal gpus
22:06 karolherbst1: imirkin: nice. I think I'll work on the direct_state_access next. Those issues seems easy enough as well
22:09 karolherbst: a test regarding gl_VertexID fails
22:09 karolherbst: switch(gl_VertexID) ... test various cases against values
22:14 karolherbst: mhh, a[0x2fc] is VertexId, so nouveau assumes
22:18 karolherbst: hum