03:05 mangix: does nouveau work with OpenCL?
03:06 mangix: i ask since it does not work on my GTX 980 but have heard iof it partially working under a 9500 GT
03:07 imirkin: it does not.
03:42 mangix: good know
03:42 mangix: *tto know
04:16 imirkin: not the same thing, but you do get compute shaders on fermi+
04:16 imirkin: [in OpenGL]
04:36 mangix: hrm ok
18:53 karolherbst: duh... I read the pstate file in the exactly wrong moment (going into suspend) and now nouveau is stuck
18:57 imirkin_: well done :)
19:30 karolherbst: okay, but I think I messed up anyway somehow
19:36 karolherbst: uhh right, I have to notify the lock
20:56 PyroSamurai: hey everyone :) Sorry, not here to help with the compute stuff today, that project was put on backburner. Anyways, I recently got an ultrawide monitor with a res of 2560x1080, but I can only display 1920x1080 :( So I'm wonder if you guys know where the problem lies for this (its not the hardware since the monitor and hdmi cord work fine on windows). So that leaves either the display software or driver. I've tried the xrandr suggestions on the
20:56 PyroSamurai: net and they don't work, so now I'm out to fix the code.
20:57 PyroSamurai: So should I be looking at the xorg code or the nouveau code?
20:58 karolherbst: imirkin: ohh, I found my mistake: reclocking while fini was triggered
20:59 karolherbst: imirkin: which woke the GPU up again, because I increase the refcount, when reclocking now
21:00 imirkin_: PyroSamurai: pastebin xorg log
21:08 PyroSamurai: imirkin_: joy, btw its a little messy from my attempts, not really sure what info you'll gain from it https://pastebin.com/CXwG2aV2
21:09 imirkin_: it'll gain me quick answers to a lot of diff questions :)
21:10 PyroSamurai: basically contains my attempts to do force 2560x1080 with xrandr
21:12 imirkin_: PyroSamurai: upgrade your kernel to v4.5 or later, or cherry-pick 1a0c96c075bb4517d4ce4fb6750ee0a3cf38714c into your build.
21:12 imirkin_: i fixed this issue 2 years ago :)
21:12 imirkin_: [ok, just shy of 2 years... commit date is nov 3 2015...]
21:14 PyroSamurai: okay, thanks. I'll see if I can find a repo for the kernel, don't much care for manual updating
21:15 imirkin_: while you look for one, ponder all the benefits of using a "stable" kernel.
21:16 PyroSamurai: like being 2 years behind on ultrawide resolutions?
21:16 imirkin_: like wasting hours of your time debugging an issue that was fixed ages ago
21:17 PyroSamurai: lol so true ;3
21:18 koz_: Main reason I'm not on a 'stable' kernel: Nouveau goodness.
21:19 karolherbst: :3 nouveau should be always the reason of not using legacy software
21:19 koz_: karolherbst: Yeah, definitely.
21:19 koz_: (although it means you have to use legacy hardware instead, sadly)
21:20 karolherbst: well, the most legacy of your system is probably the firmware anyway
21:20 koz_: karolherbst: I have a 680, which is pretty old.
21:20 koz_: Next step up I think is either 780 or 780Ti.
21:21 karolherbst: 780ti
21:21 imirkin_: K80!
21:21 imirkin_: you know you want it.
21:21 koz_: karolherbst: Is the 780 not significantly better than the 680?
21:21 karolherbst: imirkin_: :3
21:21 karolherbst: koz_: not worth the money
21:21 karolherbst: it's like a joke
21:22 koz_: karolherbst: I know the K80 is a joke.
21:22 karolherbst: allthough the 780 is faster than the 680
21:22 karolherbst: but the 780ti is even faster
21:22 koz_: Fair enough - that's my next goal.
21:23 karolherbst: the 780 should be around +33% perf, where the 780ti is like 70% more
21:23 koz_: Although right now, I really wish Nouveau could do OpenCL.
21:23 PyroSamurai: I'm fairly happy with my hardware, 760, we don't even have games that max it out, so why worry?
21:23 karolherbst: the 780ti is a monster
21:23 koz_: PyroSamurai: I'm hitting performance bottlenecks with my 680.
21:23 PyroSamurai: on what?
21:23 koz_: (${DEITY}-damned Minetest)
21:23 koz_: PyroSamurai: Minetest.
21:24 koz_: The blob can hit 60 frames consistently on max settings.
21:24 koz_: Nouveau struggles in dense areas.
21:24 karolherbst: koz_: how slow does it become?
21:24 koz_: As low as 20fps.
21:24 koz_: More commonly 40 or so.
21:24 orbea: koz_: maybe something apitrace could help debug?
21:24 PyroSamurai: i was talking about old hardware, yes the software needs lot of work still
21:24 koz_: orbea: What apitrace?
21:25 orbea: it traces GL
21:25 karolherbst: currently searching why this commit causes two reclocks on temperature changes: https://github.com/karolherbst/nouveau/commit/839439bdce60a5815cf2374f7c43395367311b28
21:25 koz_: orbea: How do I use it?
21:25 karolherbst: ohhh
21:25 karolherbst: I changed the frigging message
21:26 orbea: koz_: apitrace trace minetest
21:26 orbea: it will spit out a minetest.trace
21:26 koz_: orbea: OK, I will try that and see.
21:26 orbea: you can replay it with apitrace replay minetest.trace
21:26 koz_: Next time I'm on my rig anyway.
21:26 orbea: and then xz -9 minetest.trace to share it here
21:27 koz_: orbea: Thanks, will do!
21:27 orbea: warning, the traces can get pretty large fast
21:27 koz_: orbea: I can imagine.
21:27 koz_: My suspicion is that it's some kind of idiotsyncracy with Irrlicht.
21:28 koz_: (since that's the engine driving Minetest)
21:29 orbea: yea, could be some dumb code on part of irrlicht and/or nouveau
21:30 koz_: I would sooner blame Irrlicht than Nouveau.
21:30 koz_: Irrlicht is really not a very good engine where code quality is concerned...
21:45 PyroSamurai: hmm I'm gonna go with the "linux-generic-hwe-16.04" package, gives me a newer tested kernel (4.10.*), the equivalent of a unstable branch i suppose
21:48 PyroSamurai: gonna restart my computer an see if it worked.
21:48 PyroSamurai: brb
21:52 PyroSamurai: yup, worked beautifully
21:53 PyroSamurai: imirkin_: thanks for the advice :)
21:59 imirkin_: np
21:59 imirkin_: should also get you working reclocking on that board
22:00 karolherbst: pmoreau, mupuf: I currently found that in the therm subdev there is already some downclock threshold handling, I guess I will work on integrate my stuff into it, as it seems to use hw triggers
22:02 karolherbst: pmoreau, mupuf: uhm okay, seems to be prior gt215 only
22:02 karolherbst: the hw interrupt bit that is
22:04 pmoreau: karolherbst: If you need some testing, feel free to ask :-)
22:04 pmoreau: Not that I’m back, I have all my cards available.
22:04 pmoreau: *Now
22:04 karolherbst: pmoreau: you can use my thermal_throttling or stable_recloking_v7 branch
22:04 karolherbst: both works against 4.12
22:05 pmoreau: What hardware?
22:05 karolherbst: doesn
22:05 karolherbst: 't matter much
22:05 karolherbst: allthough the throttling might be kepler+
22:05 pmoreau: Ok
22:05 karolherbst: but there are many more changes which might need testing on older chips
22:05 pmoreau: I’ll try to test on my old laptop (Tesla) + the Kepler one at least.
22:06 karolherbst: :)
22:06 karolherbst: pmoreau: there are many things to test: restore of clock state after suspend. Reclock while suspended and expecting new state after resume...
22:06 karolherbst: temperature change only triggeres revolt without changing pstate/cstate
22:06 karolherbst: etc...
22:07 karolherbst:just noticed, that the word "revolt" also has a more common meaning stating something completly different
22:11 pmoreau: Noted
22:12 karolherbst: I think it is still possible to trigger a race condition and get the GPU to reclock allthough it is suspending, mhh odd
22:13 PyroSamurai: pmoreau: what was the current code repo for the compute work, again?
22:14 karolherbst: I have to call to pm_runtime_get_sync in nvkm_clk_update_work, but it is guarded by !pm_runtime_suspended(dev), but there is still a time frame to trigger the suspend the instruction exactly after this check
22:14 karolherbst: and then we would start to suspend, but get a pm_runtime_get_sync directly after
22:15 karolherbst: chance for this to happen might be 0.0001%, but mhh
22:18 pmoreau: PyroSamurai: https://phabricator.pmoreau.org/diffusion/MESA/
22:18 pmoreau: PyroSamurai: I haven’t made too many changes since last time, as I have been busy and getting the linker in shape for merging in SPIRV-Tools.
22:21 PyroSamurai: pmoreau: thanks, I've been really busy too lol. Really need other programmers to step up on the easier projects so the talented ones have more time for projects like these. ;3
22:22 PyroSamurai: lol wish I was one of those "talented ones"
22:27 PyroSamurai: pmoreau: do you have todo list specifically for the compute part? preferably with difficulty ratings?
22:29 pmoreau: Hum… no diffculty rating sadly :-/
22:29 PyroSamurai: but a todo list, yes?
22:30 pmoreau: I have https://phabricator.pmoreau.org/w/mesa/spirv_support/ which lists all the SPIR-V opcodes, which ones are supported, which ones aren’t, and which ones we don’t care (for now) because they are shader only.
22:30 pmoreau: The ones in relational, bit and arithmetic should be pretty easy.
22:33 PyroSamurai: okay, i'll take look when I get the time. Gotta go now though, got work that needs to be done now that I've got full res :)
22:33 pmoreau: :-)
22:34 pmoreau: Feel free to ping
22:34 PyroSamurai: will do :)
22:34 pmoreau: I’ll put up a proper todo list this week, and update the wiki to mention SPIRV-Tools being needed, and how to set it up.
22:35 PyroSamurai: that would be wonderful :)
22:45 koz_: pmoreau: What would SPIRV-Tools give us?
23:34 koz_: orbea: I'm just compressing that trace I obtained from Minetest.
23:34 koz_: I have a segment right at the end where my framerate was <30.
23:36 koz_: And you were 100% right about 'can get big fast'. I played for maybe a few tens of seconds, and the damn thing is over a gig.
23:36 orbea: they compress to a fraction of their size though :)
23:36 koz_: orbea: I assume they're text or something?
23:37 orbea: binary file, file says its data
23:37 koz_: Ah. OK.
23:37 koz_: Who should I beam it to?
23:37 koz_: (once it's compressed of course)
23:38 orbea: when I have a trace that might be a nouveau issue I link it here and usually someone like imirking is nice enough to take a look :)
23:38 koz_: Alrighty, shall do that then.
23:38 koz_: (and alert imirkin specifically)
23:39 koz_: Although this compression is taking its dear sweet time.
23:39 koz_: I guess that's the -9
23:39 orbea: yea
23:41 koz_: Lol, still like 120M.
23:41 koz_: But that's a lot better.
23:41 koz_: (10 times better or so)
23:43 koz_: imirkin (and anyone else who cares): This is a trace of Minetest exhibiting my <30fps performance issues. https://transfer.sh/12SdaX/minetest.trace.xz
23:49 orbea: koz_: in the future you probably could make the trace smaller by not making it full screen :)
23:49 koz_: orbea: Fair point, sorry. You're replaying it?
23:50 orbea: i replayed, but unfortunately I don't know how to debug traces much...
23:51 koz_: orbea: Well, thanks for doing so anyway. :)
23:51 koz_: (at least I know it can be replayed)