04:09 reece: Hey is this a good place to ask for support with nouveau? (I'm a noob but not a total catastrophically annoying one, I think?)
04:09 HdkR: sure
04:13 reece: so I'm trying to play around with wayland on a nvidia machine, but the nouveau driver works terribly for my machine, I can't get 144hz running or change the screen resolution, is there a way I can fix it?
04:14 imirkin: reece: pastebin dmesg
04:15 reece: I'll go ahead and do that now, but give me a little bit because I've gotta uninstall the nvidia driver (I also don't get sound with nouveau so I had to re-install it)
04:15 imirkin: let me guess ... turing?
04:16 reece: turing?
04:16 imirkin: GTX 16xx / RTX 2xxx?
04:17 reece: GTX 1660 Ti, yeah
04:18 imirkin: ok, i expect that the hdmi audio (or even worse - dp audio) stuff should get fixed in the next release. there are patches you could apply if you wanted.
04:19 reece: In that case, I guess my best option is to wait honestly
04:19 imirkin: but i'm not sure why 144hz wouldn't work out of the box
04:19 imirkin: what resolution?
04:19 reece: 1920 x 1080
04:19 imirkin: and hdmi or dp?
04:19 reece: hdmi
04:20 imirkin: hmmm
04:20 imirkin: are you currently running that with nvidia driver/
04:20 imirkin: can you grab the edid for your monitor? want to see if it's a "high frequency" mode or not
04:21 reece: Yeah I'm currently running the nvidia driver with the same monitor, how can I grab the edid?
04:21 imirkin: not sure with wayland
04:22 imirkin: probably in the compositor's logs? not sure
04:23 imirkin: depending on how nicely nvidia plays with drm these days, it might be in /sys/class/drm/card0-HDMI-A-1/edid
04:23 reece: I'm on X11 right now, with Nvidia driver
04:23 imirkin: oh, should be in the xorg log then
04:23 reece: Where's that, I'll grab it for ya
04:24 imirkin: normally in /var/log/Xorg.0.log
04:24 imirkin: but if you have systemd, it dumps it somewhere funny
04:25 reece: https://paste.gnome.org/p5hxhifc5
04:25 imirkin: gah!
04:25 imirkin: i guess it doesn't get dumped in there for nvidia
04:26 imirkin: can you run "xrandr --verbose"
04:26 imirkin: and pastebin the results of that?
04:28 reece: Sure, and I don't know how much this affects it but I have to run pci=noaer to prevent logs flooding my system on Ubuntu
04:28 imirkin: probably unrelated
04:28 imirkin: but definitely odd
04:29 imirkin: i'd double-check that the gpu's plugged in ok
04:29 reece: https://paste.gnome.org/prce5hzmv
04:29 imirkin: AER = advanced error recovery
04:29 imirkin: you said it was HDMI
04:29 reece: oh, I thought it was, sorry >.>
04:32 imirkin: yeah, according to the EDID, it's DP all the way
04:32 reece: Regarding the GPU, it's a prefab and I'm assuming the GPU is fine because I game on it often on my Windows partition, idk. I remember googling the pci=noaer thing when I first ran into it and it was a known bug reported on launchpad a few years back, they just haven't fixed it yet :/
04:33 imirkin: anyways, looks like you have a 1920x1080@143.855Hz with a 332.88MHz pixclk
04:33 reece: What does that mean? (the EDID DP thing)
04:33 imirkin: afaik that should be no trouble for DP
04:33 reece: Oh, hmm, are there any other logs I can give that might be helpful?
04:33 imirkin: once you get nouveau running, yes
04:34 reece: (I'm out of my depth here, but trying to be as helpful as possible, not gonna lie)
04:34 reece: What logs can I check, I'll make a note of them on my phone
04:34 imirkin: this is how your edid decodes btw: https://hastebin.com/sefoyogopa.sql
04:35 imirkin: oh, well dmesg, see if there are errors. i'd be curious in what manner you're unable to get the 144hz mode going
04:35 reece: I'll check it out, from what I can remember when I selected 144hz in gnome settings, the display just goes blank and I have to reboot to fix it
04:36 imirkin: so it works then :)
04:36 imirkin: just not quite perfect
04:36 imirkin: well, would be useful to get logs from that
04:36 imirkin: and obv using a fresh kernel
04:36 reece: Alright, I'll try grab some for you, would me running a live Ubuntu usb be enough or does it have to be a fresh install?
04:37 imirkin: doesn't have to be a fresh install at all
04:37 imirkin: but ubuntu tends not to have very recent kernels
04:37 imirkin: so if we've fixed stuff in between, not sure how helpful it'll be
04:37 reece: which live usb can I set up that would be the most helpful
04:38 reece: I'm guessing something like Manjaro?
04:38 imirkin: mmm ... dunno. pmoreau -- is your liveusb thing still running?
04:38 imirkin: pmoreau: heh, last compiled oct 2017. i guess not!
04:38 imirkin: reece: probably arch live usb
04:39 imirkin: https://www.archlinux.org/download/
04:39 reece: Oh man, I'm too nooby for no gui
04:39 imirkin: i have no clue what it comes with
04:39 imirkin: i kinda assumed it'd have something dumb like xfce, but who knows
04:40 ericonr: imirkin: Arch ISO doesn't have anything
04:40 imirkin: ah ok
04:40 ericonr: quite a few recovery tools, but no GUI
04:40 imirkin: noted
04:40 ericonr: perhaps Pop!OS?
04:41 ericonr: or Pop!_OS, never know how it's written
04:41 ericonr: they have nvidia specific images
04:41 reece: popos is ubuntu based, same kernal as 20.04?
04:41 ericonr: oh wait they're using nouveau, mind fart
04:41 reece: (off topic) I wanted to switch to POP OS but I couldn't figure out if I could safely swap out their store for gnome's
04:42 reece: (back on topic) is there an arch spin with a gui these days?
04:42 reece: there used to be antergos I think
04:42 ericonr: Manjaro :p
04:43 ericonr: archiso too, perhaps, saw a post about it on unixporn
04:49 reece: Endeavour OS
04:49 reece: apparently
04:49 reece: looking into it now
04:54 reece: I've found an old and seemingly trusted, still updated PPA dedicated to latest versions of free drivers, including noveau, is that worth a try to see if the issue still persists?
04:55 soreau: reece: I think the first thing you want is a recent kernel
05:49 reece: Alright, finally got a live usb setup, brb with the log
05:56 reece: The plot thickens, I'm gonnna need to type a decent amount here so bare with me:
06:00 reece: There are two options in Endeavour OS (Arch with GUI), Nvidia Non-free and normal. I just picked the normal one, so I assume I was running Nouveau but I don't know how to check for sure. I COULD use 144hz, it was working fine, I'm going to test audio in a minute but I need to know how to check I'm running Nouveau driver. I tried to get dmesg output for you but it's spammed with PCI error with a resolved label, and I couldn
06:00 reece: 't figure out how to set pci=noaer boot parameter (when I pressed E to change this it was a bunch of text I don't recognise)
06:01 soreau: reece: `lsmod | grep nouveau` would be a good indication
06:02 reece: right, I'll jump back in and if I am using nouveau I'll test audio
06:17 reece: here's lsmod | grep nouveau https://pastebin.com/UyJLG8v0
06:25 reece: Can confirm 144hz is working on the live USB with Nouveau
06:43 headkinomehed: I once again mention, that this knowhow was partly taught in our schools in the years 1999-2000 if attended seondary school, that 32-bit has 1024 combinations which is nothing more than 32*32 combinations of bits in.
06:45 headkinomehed: *secondary school, and if you do not understand how to manipulate arithmetic based checksums, then you should not either deal with programming or quickly learn it.
06:46 headkinomehed: not on earth should you talk about what meds i need and how much all this does not make at all sense, cause this fucking truth morans.
06:46 headkinomehed: what i told you of course.
06:49 headkinomehed: cause twos complement adding is so fast in many ways either on different lanes or pipelined and whatnot many variations, then the adder procedures can unpack inside the queues also video decoded pixels rawly.
06:50 headkinomehed: or whatever or whenever you need faster storage, and can be used on any storage to multiplex more resources out from less resources.
06:56 headkinomehed: luckily adder is the most complex of them all but not that complex that it can not be comprehended, so of course i have this method purely under control how to multiplex or compress resources.
07:00 headkinomehed: cause it is unthinkable to patent pretty much due to so many variations like thousounds of them, everything is freely talked about by me to help you a bit.
07:01 headkinomehed: and even though i do not care myself either much the fact that i was again the author of that method, people still know that anyways.
07:52 headkinomehed: I mean i am not too angry , if you manage to behave well and having decent questions i would gladly help, any other than that i am very occupied at the moment, hence i am rarely coming to IRC!
11:31 sketchy_henry: hi all. I'm trying to install bumblebee and I'm hit with a host of errors.
11:31 sketchy_henry: These are two: Possible missing firmware /lib/firmware/i915/bxt_dmc_ver1_07.bin for module i915
11:31 sketchy_henry: and: Possible missing firmware /lib/firmware/nvidia/gp107/gr/fecs_data.bin for module nouveau
11:32 sketchy_henry: there are other but they're all along the lines of what's posted above. I've googled and I'm really not sure what to do
11:33 sketchy_henry: ^^ I shoul add that i'm running Debian and I'm trying to install bumblebee and primus
11:34 RSpliet: sketchy_henry: bumblebee is deprecated
11:35 RSpliet: Using GPU offloading should work out of the box with any modern distribution
11:36 RSpliet: No need for additional software
11:36 sketchy_henry: ah fantastic thank you
11:36 RSpliet: Missing firmware files are an issue that yo need to talk to your distribution people about. They should know where to find them
11:37 RSpliet: As for actually using your NVIDIA GPU: on the command line you can set DRI_PRIME=1 as an environment variable to launch any application on the discrete GPU
11:37 RSpliet: Gnome Shell has a right-click menu item "launch on discrete GPU" (or sth similar) in the launcher
11:37 RSpliet: Other desktop managers I have no idea abot
11:37 RSpliet: If this does not work, it is very likely that whatever Debian packages you have installed are simply outdated.
11:38 sketchy_henry: ok thanks a lot. I'll turn to #debian and ask about the missing firmware.
11:38 RSpliet: Be advised that when you use nouveau, your NVIDIA graphics card is probably slower or about as fast as the Intel. We can't bump the clocks to the levels advertised with the GPU
11:39 sketchy_henry: I understand. I just like the open source aspect
11:40 RSpliet: I wish NVIDIA appreciated that too. Would be nice if we weren't actively stopped from achieving good performance w/ nouveau!
14:57 pmoreau: imirkin: Yeah, I stopped building it as there was some bugs with the mesa build, and since I didn’t have time to solve them at the time, I just disabled the whole build.
15:00 pmoreau: I might look at getting it working again at some point.
17:58 imirkin: anyone know how to reach the endeavouros folks? they don't seem to list an irc channel
17:59 imirkin: they seem to provide very unfortunate advice for nouveau - https://endeavouros.com/docs/hardware-and-network/nouveau/
19:36 RSpliet: imirkin: what, they telling people to use it?
19:36 imirkin: they're telling people that xf86-video-nouveau is bad on older gpu's
19:37 RSpliet: oh yes, just reading through it. GLamor on GeForce 2, that'll be a good laugh
19:40 headkinomehed: RSpliet: by now you should be intelligent enough to understand that clocked circuits are full/long pipeline mode relevant only:), queues can be accessed async on most chips, dispatcher is the only clocked path, but this brings many instructions in during single clock cycle (fetch arbiter is async) and is not even changing state from execution and idle, only clocked buffer in miaow is in the SIMD arbiter, but this is only miaow limitation, don't
19:40 headkinomehed: get me wrong there are others clock sync logics on the GPU, but all of them will be bypassed in my method. And so were they one gtx580 for NVIDIA too, and every seriously designed chip, so acceleration path in queues is async anyways. TL;DR accessing queues is async.
20:49 pmoreau: OpenCL basic test results on NV50: “FAILED 31 of 80 tests”, with 11 more tests commented out due to crashes or asserts being triggered in Nouveau or NIR.
21:01 imirkin: almost perfect
21:06 RSpliet: pmoreau: is that *NV50*? Or like, gt21x? :-P
21:07 pmoreau: It’s for G96
21:07 imirkin: pmoreau: fyi i have a G84 plugged in, can look at stuff if you need.
21:07 RSpliet: Ah nice!
21:11 pmoreau: One thing that would be nice to look into, and might require some RE so I can not do that, is check emitCVT and see if we can do U8/S8 -> F32, U8/S8/U16/S16 -> F64, * -> U64/S64.
21:12 imirkin: well, none of the f64 stuff works outside of g200
21:13 pmoreau: Otherwise for the other Nouveau bit, it’s emitSTORE() for a 64-bit values to local mem. I’m assuming that is not possible, so we might want to split those stores before we hit the emitSTORE() function.
21:13 imirkin: probably... let me check
21:14 pmoreau: I see, so on G96 we would end up sending invalid opcodes to the GPU? Cause emitCVT those not seem to check for a chip version before emitting other F64 stuff.
21:14 imirkin: oh. maybe the cvt stuff is ok then?
21:15 imirkin: also, 64-bit st l[] should be fine too
21:16 imirkin: actually i dunno
21:16 imirkin: it's been a REALLY long time
21:16 imirkin: envydis should be quite accurate for G80 though
21:17 imirkin: so like look at "tabldstm"
21:17 pmoreau: Sorry, I meant shared (OpenCL’s local)
21:17 imirkin: all the cvt f64 stuff is G200-only though, afaik
21:17 imirkin: i just wouldn't advertise f64 support for non-G200
21:19 imirkin: and yeah, i don't see any 64-bit s[]
21:20 pmoreau: F64 is not even advertised for G200+, so alright, not an issue.
21:20 pmoreau: Still interested whether U8/S8 -> F32 is doable with CVT, or I need to do emulate it.
21:21 pmoreau: Okay, I’ll then add a pass to split 64-bit stores to shared.
21:24 imirkin: not G200+... it's *only* G200 that has it
21:24 imirkin: i.e. GTX 270/280 or whatever
21:24 pmoreau: Oh :-D
21:24 imirkin: (and Quadro FX whatever)
21:25 imirkin: and it only has add/mul, no sqrt/etc
21:26 pmoreau: Is there sqrt() & co for f64 on later hardware?
21:26 imirkin: hm, i only see [us]16 <-> f32 in the opcode map
21:26 imirkin: yeah, on fermi+ there's rcp and rsq helpers
21:27 imirkin: not 100%, but it's a good start, then you do newton-raphson
21:27 pmoreau: Oh wow
21:27 imirkin: nvidia implements sqrt() in some different way though, since 1/rsq is probably not precise enough with f64
21:27 pmoreau: Mmh, thanks for checking. I’ll emulate it then; shouldn’t be too hard.
21:28 imirkin: pmoreau: oh hm, i take that back
21:28 imirkin: { 0x44000000a0000000ull, 0xc4400000f0000000ull, N("cvt"), T(cvtmod), T(s35sat), T(cvtrnd), N("f32"), MCDST, LLDST, T(cvtiisrc) },
21:29 imirkin: so that converts to F32, afaik
21:29 imirkin: cvtiisrc includes an integer type, including all that stuff
21:30 pmoreau: Ah, thanks! I’ll look into that (though not now, cause it’s getting a bit late).
21:34 pmoreau: There is no opcode for (x+y)>>1 (where x and y are integer types), right? Or is that OP_VADD (which does not seem to be used anywhere in codegen)?
21:35 imirkin: on fermi+ there's a+b+1
21:35 pmoreau: Ok
21:37 pmoreau: Thanks for the help! I should get some sleep now.
22:20 karolherbst: pmoreau: I have some conversion stuff fixed
22:20 karolherbst: but I guess that's on top of my branch?
22:21 karolherbst: pmoreau: afaik 16 <-> 32 bit conversions are fully supported
22:21 karolherbst: but you need to get do some handling for 8 bit
22:55 imirkin: karolherbst: note he was talking about nv50
23:12 AndrewR: imirkin, hi! I see mesa 20.1 is in rc phase ...was this nv50 + old blender (2.79b) misrendering in some windows fixed there in some way?
23:13 imirkin: no
23:13 imirkin: i need to send something
23:26 karolherbst: ohh, I also looked into the issue
23:26 karolherbst: uhm...
23:26 karolherbst: what was it...
23:26 karolherbst: mhh
23:26 karolherbst: something stupid
23:27 imirkin: vbo flush fail
23:27 imirkin: or ... something fail
23:27 karolherbst: wait...
23:27 imirkin: with coherent mappings
23:27 imirkin: which the core vbo code is now using
23:27 karolherbst: but I think there was something stupid shader bug as well, no?
23:28 karolherbst: ohh wait, that was a blender 2.80 bug
23:28 imirkin: that's different then
23:30 karolherbst: ahhh.. why does plasma have to be so broken on wayland
23:30 karolherbst: or qt5