00:00karolherbst: where I can disable the intel GPU :)
00:00karolherbst: I just want to see how much it actually improves things
00:01karolherbst: not quite sure what to do for kms based applications
00:01karolherbst: like the dracut stuff
00:02karolherbst: maybe hope that it never messes up the GPU?
00:06pendingchaos: why does the computer become unusable once a channel is killed btw? and why would killing the application which caused it help?
00:06karolherbst: *sigh*, now sddm doesn't display anything, but this could be just a mesa update
00:06karolherbst: pendingchaos: X is waiting on the applicaiton
00:06karolherbst: and because X sucks with that kind of stuff
00:07karolherbst: maybe it is better under wayland?
00:08karolherbst: *sigh*, now sddm really wants to disappoint me
00:16karolherbst: ohh uhm, my mistake
02:26imirkin: Lekensteyn: booo ... i hate the default rtfd scheme
02:27imirkin: anyways, that's all for mwk.
02:27imirkin: he's the one who got it all going
02:45icarious: Hi. Can anyone suggest me a latest (still available in the market) card supported by Nouveau before the Maxwell ones? I have shortlisted the GT 730. But wonder if I can get anything better that is still working and not available in the Nouveau Matrix
02:46HdkR: Check newegg and Amazon for any GTX 7xx series cards still available? :)
02:47icarious: HdkR: ARE ALL THE gtx 7* kepler?
02:47icarious: The matrix is a little confusion. like the 730 wasn't listed but the 730M
02:47HdkR: A couple are Fermi and a couple are Maxwell first gen
02:47HdkR: Wikipedia has the list
02:48icarious: Fermi is too slow. Maxwell isn't supported.. So I need something in the middle
02:48HdkR: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_700_series Look for GF, GK and GM in the start of the code names
02:48icarious: let me check
02:49icarious: HdkR: Thanks a lot mate
02:49icarious: You are a life saver.
02:49karolherbst: okay cool
02:50karolherbst: that reset stuff works
02:50karolherbst: X won't freeze anymore for me :)
02:50karolherbst: (given nouveau is able to kill the channel)
02:51karolherbst: imirkin: do you have some easy ways to trigger those gpu crashes running X on nouveau?
02:53imirkin: karolherbst: run mplayer -vo vdpau
02:53imirkin: icarious: you really want to go with amd, btw
02:53imirkin: icarious: GT 730 is fine, obviously low-end
02:54HdkR: That would be the better alternative if you want to stay in open source land yes :)
02:54icarious: imirkin: wait what? AMD?
02:54imirkin: much much much better support
02:55icarious: Oh I forgot to mention. I don't run firmware blobs. I use linux-libre. So ya, that blacklists the Radeons
02:55icarious: they need the loadable blob I suppose? or no X?
02:55imirkin: so no x86 cpu's either?
02:55karolherbst: ohh wow
02:55imirkin: anyways, if you want to run a distro which randomly deletes code, then you're on your own
02:55icarious: imirkin: don't get into the "has microcode / BIOS" argument. its becoming to exhausting. Just respect someone's preferences?
02:56karolherbst: lwjgl crashes when there is no primary display
02:56imirkin: i can't make recommendations given those constraints. best of luck!
02:56karolherbst: makes sense, doesn't it?
02:56icarious: I am someone who believes in using something "solvable" , while ignoring the rest. and not for "All or nothing" side of the binary thinking camp
02:57imirkin: like you said - it's a tiring debate. i'm not really interested in rehashing it.
02:57imirkin: best open-source support is on intel / amd gpu's.
02:58imirkin: if they don't meet your needs for whatever reason, and nouveau happens to, you're in for a lot of sadness if you're looking for a good end-user experience.
02:58icarious: imirkin: I am not looking for a good end-user experience as such. Not into CGI Development. just want a midrange card that can handle 4K Videos decently and bit of gaming
02:58imirkin: cool, yeah. definitely avoid nouveau.
02:59karolherbst: imirkin: mhh, can't install mplayer, because ffmpeg-4
02:59icarious: imirkin: let me rephrase. Games like Xonotic / 0AD
02:59imirkin: i just got a hang playing portal 2 yesterday.
03:00karolherbst: well, I also got tombraider to hang
03:00karolherbst: and with the abort on reset thing, X hangs for 4 seconds and goes back to normal after :)
03:00karolherbst: I just want to test different things, not just games
03:00karolherbst: so mplayer is good
03:00karolherbst: getting the channel killed of kwin would be nice
03:00imirkin: this was weird... i could do num lock, but couldn't ssh in
03:01karolherbst: with portal 2?
03:01karolherbst: curretnly playing some game which uses lwjgl
03:01imirkin: older mesa (18.1.x) -- didn't have pendingchaos's fixes for ubo flushing
03:02imirkin: so i was also getting some rendering corruptions
03:02imirkin: (dunno if they get fixed, but one can hope)
03:02karolherbst: I kind of get random rendering corruptions in my entire plasma session :/
03:08imirkin: airlied: skeggsb: what's the minimal thing i need to have set up in order to get DP-MST connects appear and disappear? e.g. if i have a hub, will its outputs appear even if nothing's plugged in? and separately, i assume a passive DP -> HDMI adapter can't work in a DP hub?
03:09airlied: imirkin: the hubs usually don't do anything unless something is plugged in
03:09airlied: and yeah passive dp->hdmi shouldn't work I don't think
03:10airlied:can't quite remember what the in-monitor hubs do
03:14imirkin: there are a few steps before i can get a proper setup, so just trying to figure out the minimum
03:14imirkin: (like i have to clean my desk...)
03:14airlied: skeggsb: please resent pull req cc' dri-devel
03:14airlied: imirkin: its very device dependent
03:15airlied: some of them need to be explicitly put into dp1.2 mode
03:15airlied: and some of them won't go MST unless they have to
03:15imirkin: yeah, these monitors have an explicit setting
03:15imirkin: but once it's on, it's on
09:30Lekensteyn: imirkin: https://kernel.org/doc/html/latest/ is using the rtd theme as well, so you'll get used to it (and possibly still hate it) :)
09:30Lekensteyn: the rtd theme is supposedly better for mobile/tablets where the sidebar gets collapsed
10:22jscinoz: Hi everyone. I was wondering, with regards to power management / reclocking of newer cards (particularly Maxwell and Pascal), if the status on the wiki is up to date (these pages say they haven't been updated since 2017), and if so, what the blocker is at present - does this require blobs from Nvidia that they're refusing to release, or is it technically possible, just not yet implemented?
11:49RSpliet: jscinoz: without NVIDIA PMU firmware we can't do thermal management or execute memory clock adjustment scripts. Both are showstoppers
11:56karolherbst: jscinoz: it would be technical possible on maxwell2 laptopt GPUs where the fan is controlled by the system EC, thing is, we can't know for sure (or at least we don't know how)
12:00pendingchaos: "2D/3D acceleration supported on all GPUs" should probably be changed to "2D/3D acceleration supported on all pre-Volta GPUs" or something (on https://nouveau.freedesktop.org/wiki/)
12:11jscinoz: RSpliet: Ah, and I'm guessing Nvidia aren't going to be forthcoming with that presumably because they want to push people to use their proprietary driver (for the usual anticompetitve and/or artificial segmentation reasons, presumably)
12:11jscinoz: karolherbst: Ah I see, thanks for the info :)
12:57karolherbst: ohh wow, my gm204 has actually 4 meaningful perf levels
12:57karolherbst: didn't know
13:00RSpliet: jscinoz: I'm convinced the reasons are more in the line of "publishing firmwares takes effort. Effort costs money. We don't have a reason to spend money."
13:00karolherbst: what a GPU... 60W idle power consumption, and tht's a laptop
13:01karolherbst: Lyude: was there NvPmEnableGating support for maxwell? I don't think so, right?
13:02diogenes_: Hey guys, what would be the mobile GPU for laptops that is best supported by nouveau driver?
13:04karolherbst: diogenes_: depends, but normally the situation is kind of crappy today
13:04diogenes_: karolherbst, crappy in what way?
13:04karolherbst: runtime suspend/resume is pretty much broken and nobody has any ideas why
13:05karolherbst: I mean, I have some clues, but still
13:05diogenes_: karolherbst, it won't be an issue because i intend to run it with optimus so the main workflow will be on intel and the heavy one on nouveau
13:05karolherbst: Lyude: this is insane, going from 0x7 to 0xa perf level: 20W -> 53W
13:06karolherbst: diogenes_: that's exactly the issue
13:06diogenes_: karolherbst, didn't get you
13:06karolherbst: diogenes_: currently on many laptops the GPU doesn't wake up anymore
13:06karolherbst: the GPU is shut off
13:06karolherbst: and doesn't came back
13:07diogenes_: oh you mean the newest ones?
13:07karolherbst: it might also affect maxwell ones
13:07karolherbst: and maxwell is kind of what you want to go for
13:07karolherbst: currently I have a laptop with a 970M and it seems like I am able to reclock the GPU
13:07karolherbst: so that's quite nice
13:07diogenes_: oh ok thanks, to maxwell would be good
13:07karolherbst: I have no idea on which laptopts we don't mess up
13:08diogenes_: ok but what about the laptop manufacturer, what would you advice?
13:08karolherbst: as maxwell/pascall is the time where the runtime d3cold thing was introduced
13:08karolherbst: seriously, I have no idea which one would work or which wouldn't
13:08diogenes_: maybe system76
13:08karolherbst: kepler might be a safe bet, but ... maybe there are keplers with the same runtime 3d thing
13:08karolherbst: they only care about nvidia drivers
13:09diogenes_: oh damn
13:09karolherbst: and nvidia doens't even turn the gpu off
13:10diogenes_: karolherbst, you know what, i've noticed with the current kernel updates the gpu doesn't turn off anymore
13:10karolherbst: if I would be system76 I would use AMD GPUs.... but hey, it isn't my business
13:10karolherbst: diogenes_: huh
13:21karolherbst: Lyude: enabling those 0x20200 registers: 60W -> 40W :/
13:21karolherbst: maybe I find some time to look into it at some point. power gating for maxwell that is
13:38imirkin: Lekensteyn: yes, the rtd theme is better on devices i don't care about and i doubt anyone will be reading envytools docs on. i hate it on rtd, i hate it on kernel, i hate it everywhere. just because other people use it doesn't mean i should. but like i said, this is mwk's baby - his call.
14:15mwk: I do read envytools docs on mobile sometimes, and I still hate the default theme
14:16mwk: that said, the current one ain't perfect either, and craps itself when large tables are involved
14:16mwk: [but the default one craps itself way way more on those, which is partly why I hate it]
14:17mwk: unfortunately, hardware docs tend to involve lots of large tables
14:28karolherbst: mhh, crap. The reset notification thing doesn't work within X as we end up getting the same fd inside glamor as we do for the vsync stuff :/ annoying
16:08pendingchaos: karolherbst, imirkin: any idea why using nve4_p2mf_push_linear/nvc0_cb_bo_push would be much faster than nve4_m2mf_copy_linear?
16:08pendingchaos: https://pastebin.com/raw/D4zbh4KZ gives a 7-10% improvement with Hitman
16:08pendingchaos: 2% or something improvement with Deus Ex: Mankind Divided and no change with Dolphin, 0 A.D and Tomb Raider.
16:09karolherbst: pendingchaos: no idea, I think pmoreau had a patch once which had some improvement on the talos principle
16:11karolherbst: I could imagine it slows down other things though
16:12pendingchaos: yeah, I'm sure there' some reason for NOUVEAU_TRANSFER_PUSHBUF_THRESHOLD
16:13HdkR: pendingchaos: Dolphin uses a ringbuffer to update its UBOs a full UBO at a time, which is why I'd guess this optimization wouldn't do much there
16:13karolherbst: anyway, would be nice to figure all that out
16:13karolherbst: and know what improves performance under with situations
16:13HdkR: +persistent+coherent mapping*
16:14HdkR: pbuffer configuration in the Vulkan blob keeps the size under 256...elements?bytes? 256 somethings
16:15pendingchaos: 256 bytes
16:16HdkR: Probably reasonable to keep the size around that limit :P
16:18pendingchaos: I think Hitman would still benefit with NOUVEAU_TRANSFER_PUSHBUF_THRESHOLD being 256
16:22karolherbst: pendingchaos: it might hurt on older GPUs
16:22karolherbst: anyway, I doubt we want that to be a hardcoded threshold anyway
16:38pendingchaos: karolherbst: so instead a different value depending on whether Kepler, Maxwell or Pascal is being used?
16:38karolherbst: I think so
16:38karolherbst: anyway, we would have to experiment with it
16:38karolherbst: and we don't track perf regressions
18:29Subv: is it known how blocklinear 3D textures are stored in maxwell?
18:30HdkR: block linear of course </s>
18:45karolherbst: Subv: I think they are aligned to 0x1 on the z axis, but y/x are configuereable
18:45karolherbst: well, z is as well
18:46karolherbst: but nvidia uses a 0x1 alignment for good reasons
18:46Subv: i mean, is the blocklinear layout for 3d textures the same as for 2d textures? ie, are they treated as a single large array of size width*height*depth or are they interpreted as 'depth'-number of 2D slices?
18:47karolherbst: there is some padding going on
18:47Subv: i see nouveau uses the m2mf engine to convert them into blocklinear format for rendering
18:47karolherbst: no idea if for 2d textures as well
18:47Subv: oh, what kind of padding?
18:47karolherbst: just the axis aligment
18:48karolherbst: so each "block" is aligned on each axis
18:48karolherbst: but in the end you end up with a compatible layout for 2d and 3d textures
18:48karolherbst: on maxwell that is
18:48karolherbst: I couldn't get a 0x1 alignment to work on kepler
18:49karolherbst: Subv: it makes gl easier where you access 3d images though 2d images...
18:49karolherbst: so you don't have to do any black magic on maxwell to support it
18:54Subv: mm, is there any CPU converter between pitch and blocklinear textures for maxwell?
18:54Subv: or has nouveau always used the copy engine to do that (without cpu fallback)?
18:56karolherbst: I don't know, but I don't think so
18:59Subv: alright, just to see if i understand correctly, converting a pitch 3D texture of dimensions W, H, D to pitch is equivalent to converting D pitch 2D textures with dimensions W, H to blocklinear?
19:00Subv: (equivalent with respect to the resulting memory layout)
19:04karolherbst: I don't know if that's always the case, but at least if you look from a shader perspective on it, you can use 2d texture operations on 3d textures if you configure the texture view correctly (offseting + layout mainly)
19:48pendingchaos: imirkin: I think nodep is liveOnly, not ndv? ndv seems to be something called derivAll
22:41pmoreau: karolherbst, pendingchaos: I’m pretty sure it wasn’t me who had a patch improving Talos Principle: :-)
22:42karolherbst: maybe hakzsam?
22:42karolherbst: pmoreau: it was about the flickering issue
22:42karolherbst: and I am sure you had a patch :p
22:43karolherbst: but imirkin complained about it, because it wasn't a fix, just a workaround
22:43karolherbst: (in the end the precise modifier stuff fixed it for real)
22:49pmoreau: Looking at my patches that were merged in Mesa, there is nothing that would match that.
22:50karolherbst: yeah, because it wasn't merged
22:51pmoreau: If still considered valuable, could someone merge https://patchwork.freedesktop.org/patch/191743/ please?
22:52karolherbst: pmoreau: ohh, right. I will do it
22:52pmoreau: Thank you! :-)
22:52pendingchaos: why is it prefixed with "nvir"?
22:52pendingchaos: I read it as "nir"
22:53karolherbst: there is the idea of renaming nv50/ir into nvir or something
22:55pmoreau: I have some patches somewhere doing that