08:13RSpliet: karolherbst: I stand corrected. Yes. I don't know whether I'm dreaming this, but I think technically someone could add shader support for one of those nv20s. In practice though, IIRC it has like an 8-instruction limit, so it's not worth the effort :-P
09:38glennk: vertex shader on gf3 was max 128 instructions if i remember correctly
09:39glennk: fragment processing was texture/register combiners, not really shaders
10:21karolherbst: that's a new one.. my nv43 just crashed after booting :(
10:21karolherbst: "nouveau 0000:01:00.0: timer: stalled at ffffffffffffffff"
14:56raket: RSpliet: it's not the same bug with rx560. tested. And i don't understand how bisect works :-(
15:45RSpliet: raket: the theory is : if you have a big anywhere between "patch 0 (good) and patch n (bad)", you try building and running mesa at patch n/2. If its good, you know the bug is in the top half, if not, it's in the bottom half.
15:46RSpliet: That way you can find the offending patch in log_2(n) iterations of rebase, build, test.
15:47RSpliet: Git can help you with the rebase part. You tell it a good and a bad point, and it will automatically rebase to the mid-point. Tell git whether it was good or bad, and it'll suggest and rebase to the next one until it's narrowed down.
15:48qeeg: karolherbst: sounds like your gpu might be dying. PTIMER is a rather central part of the card.
17:03raket: RSpliet: nice one! i'm bisect now :)
20:15raket: RSpliet: Package mesa-git-6005b86893-x86_64-1.tgz installed. <- that one works (6005b86893e4a388b037717742de3e5894b83c81) the xcb errors is not there. however, if using xport mesa_glthread=true ezquake segfaults. it's really a none-issue at it does exactly nothing to quake and it's not really used anyway.
20:15raket: it works fine to compile with or without llvm as well.
20:15RSpliet: I think glthreading is known-broken with nouveau
20:16RSpliet: That's still a WIP, karolherbst's moving mountains to right 15 years of in-hindsight-poor-design-decisions
20:16karolherbst: RSpliet: it's not WIP
20:16raket: it works with mesa-20.3.5 but.. it does nothing at all... besides it's enabled, like a switch that does nothing
20:16RSpliet: karolherbst: is it not?
20:16karolherbst: I am done with the patches :D
20:16RSpliet: oh! \o/
20:16karolherbst: even works on nv30 and nv50
20:16RSpliet: wen upstream much? :-D
20:17karolherbst: I just found some weird bug on nv50
20:17karolherbst: where some submissions are not worked on by the GPU if I submit too fast or something
20:17karolherbst: and.. well
20:17karolherbst: missing testing
20:17karolherbst: but at least the android emulator is happy until the channel crashes :)
20:17RSpliet: that sounds like there's work in progress :-P
20:17karolherbst: of course there are more multithreading bugs, but not on the CPU side at least
20:17RSpliet: But glad to hear there's light at the end of this tunnel :-D
20:18karolherbst: yeah... took me long enough
20:18karolherbst: people can review it though :p https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10752
20:18karolherbst: it's not a huge change if I might say
20:18karolherbst: most of the patches are trivial
20:19RSpliet: I'm going to have to leave that to someone smarter than me
20:19karolherbst: mhh sadly pierre isn't around
20:19RSpliet: yeah, he's not supposed to make time for anything, like at all
20:20RSpliet: also shame that calim isn't around :-P
20:23raket: ccr: thanks for the great support with the mesabuild.sh stuff, llvm can **** it self
20:26raket: karolherbst: can we git clone mesa and just apply your patches? :p
20:26karolherbst: you can also checkout my branch and rebase it :D
21:01sneil: Hey I just tried that (nvc0_threading_upstream branch rebased) on my GM107
21:01sneil: I can report success running piglit: [56917/56917] skip: 7177, pass: 49577, warn: 8, fail: 123, crash: 32
21:02sneil: and running 4 or 5 steam games, though I'm not sure if I'm really exercising any multithreaded paths
21:08raket: sneil: did you get any fps boost in csgo?
21:09karolherbst: sneil: you do when opening the overlay web browser :)
21:09karolherbst: raket: it shouldn't
21:09karolherbst: but I expect some weirdo glitches
21:09karolherbst: state tracking isn't really fixed, but that's for the next batch
21:10raket: when csgo works at > 150~160 fps stable nouveau i won't use the blob for that game. it work's kinda "good" with a 980ti (yeah with reclocking..) but in fights it goes down to 60fps :-)
21:14sneil: haven't tried that one raket, is it threaded?
21:15raket: no idea - that's why i asked :)
21:31sneil: oh is there a toggle to get fps in csgo?
21:32raket: sneil: net_graph 1
21:33raket: or.. cl_showfps 1 (i guess - haven't played in a while)
21:39sneil: hmm so it runs under system mesa (21.1.4) but not my custom built one
21:41sneil: I'll have to investigate more later, gotta run
21:59Svanto: Hello dudes
21:59Svanto: So I have that problem that I already mentioned on the 10th that month, and I'm pretty sure it's nouveaj since it happens in tty as well
22:00Svanto: My external screens in my setup are black
22:02Svanto: When I reconnect the screen in tty I get I get similar issues to the ones I saw in the dmesg, I'll transcribe them
22:02Svanto: [ 28.148709] nouveau 0000:08:00.0: DRM: core notifier timeout
22:02Svanto: [ 28.166462] nouveau 0000:08:00.0: bus: MMIO read of 00000000 FAULT at 61a804 [ PRIVRING ]
22:02Svanto: [ 28.176239] nouveau 0000:08:00.0: bus: MMIO read of 00000000 FAULT at 61a804 [ PRIVRING ]
22:02Svanto: [ 28.176514] nouveau 0000:08:00.0: MMIO read of 00000000 FAULT at 61a804 [ PRIVRING]
22:02Svanto: [ 28.276599] nouveau 0000:08:00.0: bus: MMIO at 616e18 [ PRIVRING ]
22:02Svanto: [ 30.276897] nouveau 0000:08:00.0: DRM: core notifier timeout
22:02Svanto: [ 32.276917] nouveau 0000:08:00.0: DRM: base-1: timeout
22:05Svanto: It doesn't seemlike a vbios issue
22:08Svanto: I'm sorry to just dump this here, but it seems like something
22:29RSpliet: Svanto: what kernel version is that? Can you paste the whole dmesg somewhere? Fedora and Ubuntu both have a paste command line option (Fedora's one is called fpaste, not sure what it is on other distros), you can just pipe dmesg output into there and copy-paste the resulting URL here
22:32RSpliet: Oh, btw, I'm sure other distro's have them too... just never had the pleasure of verifying that with my own hands
22:37Svanto: RSpliet Kernel version is 5.12.15-arch1-1
22:38Svanto: Unfortunately I can't give you my dmesg right now, I havent got that gpu on hand
22:39RSpliet: Svanto: ok, that's new enough to qualify for help here ;-) Which GPU is this?
22:43Svanto: GTX 970!
22:49RSpliet: Alrighty, that sounds like an issue Lyude may have some thoughts about. She's the residential display expert (or so I hereby proclaim)
22:49RSpliet: Although I suspect more info is required. Not all MMIO read faults are problematic. There's one or two that just kind of stick out like a sore thumb in the logs, but are harmless otherwise. Not sure if these fit the same bill
22:52Lyude: good timing as my plate is clearing up soon, but it may still take me a little time to get to this. if you could open a gitlab issue and get a full dmesg of the issue with `drm.debug=0x16 nouveau.debug=disp=trace` that would be appreciated
22:52Lyude: oh sorry - you add that last part to the kernel commandline without the backticks
22:52Lyude: also add log_buf_len=50M
22:53Lyude: Svanto: also do you know if this is a regression?
22:54Lyude: RSpliet: yeah the initial core timeout is almost certainly the issue here
23:24raket: Lyude: about the 2k issue with GK110B, i just sold the card and use NV120 which works fine with 165hz at 2k :-). What if i plug in a 2k monitor with 240hz? will it work? :)
23:28Lyude: raket: oh I have no idea on the last question lol
23:28Lyude: I would hope so :)
23:30raket: Lyude: in fact, the GK110B could do 2560x1440@165hz in windows 10 (i just tested for fun). However, it only worked at the session where the titan-driver was installed
23:30raket: kinda buggy drivers
23:31Lyude: oooh. it's definitely doing something funky then
23:31raket: Lyude: Install the drivers, 165hz works, after reboot or changing hz in the control panel and the monitor would go green (bad signal?)
23:46Svanto: Lyude: it was strange, it worked fine for about a week but then this issue happened
23:46Svanto: After the update
23:47imirkin: karolherbst: vieux has support for nv34, but only for the purpose of me testing vieux without an old card plugged in.
23:48imirkin: karolherbst: or rather ... vieux has support for loading nv34 which emulates the NV25_3D class
23:48Svanto: I will get the dmesg asap. Thank you very much, RSpliet, Lyude!
23:50karolherbst: imirkin: ahh.. interesting
23:51imirkin: karolherbst: https://cgit.freedesktop.org/mesa/mesa/tree/src/mesa/drivers/dri/nouveau/nouveau_screen.c#n134
23:51imirkin: you have to force the loader to load vieux though
23:52karolherbst: yeah.. interesting.. I didn't touch vieux with my mt fixes because there was no need and I don't have a matching GPUs.. but with this.. I might even check it out :D Although I highly doubt it's even worth it
23:53imirkin: well, unsurprisingly, it doesn't share any of the gallium architecture
23:53imirkin: it does use libdrm_nouveau of course
23:54karolherbst: but I don't see anybody using it with more than like xfce or whatever
23:54karolherbst: I mean.. even trying to use chromium or anything like it will already end in pain
23:54imirkin: i don't see anybody using it.
23:54imirkin: the accel in the ddx? sure. the GL driver? not so much.
23:54karolherbst: and given how broken nv30 is, I don't see many people using that one as well :)
23:54imirkin: first off, it's only GL 1.x
23:55karolherbst: saw my gnome experience?
23:57imirkin: i'm guessing it's one of the changes removing tgsi
23:57imirkin: but maybe not, and it's something silly? dunno
23:57karolherbst: it is also broken with like 20.1
23:58karolherbst: or maybe we switched earlier
23:58karolherbst: https://drive.google.com/file/d/1XrCObCHNlOxmZnq1mEsY9EyUtZrWHyuo/view?usp=sharing :)
23:58karolherbst: google has a few troubles, so you might have to download
23:59karolherbst: I think I might take another look next week, but that looks like seriously broken :)