01:22P1ro: hey, is there anoyne around that could help me setting up xorg with nvidia ? i have nouveau drivers installed, but is not working here is xorg.conf and the error log. http://dpaste.com/24XW1W6.txt
01:58prg: P1ro: how to properly install the blob depends on the distro, there might be a wiki/other documentation available somewhere explaining how to do that
01:58prg: if you're interested in maybe getting nouveau working, pastebin dmesg
01:59P1ro: prg here is the dmesg http://dpaste.com/1QG6S4P.txt
02:00P1ro: prg, i did try the arch linux docs but all tree drivers failed, so far, nvidia, nvdia304 and nouveau
02:00prg: and what GPU is in there?
02:01prg: grep nouveau -r /etc/modprobe.d/
02:01P1ro: grep nouveau -r /etc/modprobe.d/
02:02P1ro: #nouveau grep nouveau -r /etc/modprobe.d/
02:02P1ro: grep nouveau -r /etc/modprobe.d/
02:02P1ro: -/etc/modprobe.d/nouveau_blacklist.conf:blacklist nouveau
02:02P1ro: -/etc/modprobe.d/nvidia-installer-disable-nouveau.conf:blacklist nouveau
02:02P1ro: -/etc/modprobe.d/nvidia-installer-disable-nouveau.conf:options nouveau modeset=0
02:02P1ro: got nouveau blacklisted?
02:02prg: yah, so that disabled nouveau
02:03prg: try without nomodeset
02:03P1ro: wich one i should rm ?
02:03prg: er, remove the modeset=0 line
02:04prg: (and the blacklist ones if you want the module to be loaded automatically)
02:04prg: then rmmod nouveau; modprobe nouveau
02:05P1ro: still dont start
02:05prg: anything new in dmesg?
02:06P1ro: wait box is lagging
02:06P1ro: no nothing new
02:06P1ro: [31560.020584] [drm] Initialized drm 1.1.0 20060810
02:06P1ro: [31560.069765] wmi: Mapper loaded
02:06P1ro: last two lines
02:07P1ro: others are wifi
02:07P1ro: Xorg -configure give this error: Number of created screens does not match number of detected devices. / Configuration failed.
02:08prg: you're not that far yet
02:08prg: first get the kernel module to do something
02:08prg: (also you probably shouldn't need an xorg.conf once that works)
02:10P1ro: yeas i hear about that
02:11P1ro: but at this rate i might end up installing again :(
02:13prg: you did rmmod nouveau before modprobe nouveau?
02:13P1ro: i did modprobe it
02:13P1ro: atfer start
02:14prg: to change module options you need to rmmod it first
02:14P1ro: dummy of me
02:15P1ro: this should keep like this without problems right ?
02:16prg: it shouldn't re-blacklist itself if you don't install the blob again
02:17P1ro: well i think that should be auto with the nv driver
02:17P1ro: but i wonder whi nvidia304 didnt worked
02:21P1ro: prg well thanks alot, did worked :D
02:23P1ro: prg another question, this might be unrelated to nouveau
02:23P1ro: i use that machine mostly remote, ssh/teamviewer
02:25P1ro: is there a way to set a custom resolution to team viewer? the monitor its old samsung 14inch and if i dont plug monitor to start when ill teamviewer it, resolution its like 480x240 or something really low
02:25P1ro: i could setup a force resolution with no monitor to use with teamviewer?
02:29prg: nfi about teamviewer. you can have a look at what resolutions are offered by xrandr
02:30prg: but no idea how to force an unsupported resolution, maybe you could use a fake EDID
02:31P1ro: fake EDID ?
02:31P1ro: could try
02:32P1ro: dont really want to have a monitor plugged into that box
02:34prg: how about NX in that case?
02:42P1ro: wao, atfer a short search there alot of option not this emulate teamviewer :D
03:08karolherbst: does anybody know when skeggsb will be back?
03:14RSpliet: is he gone?
03:15RSpliet: or just locked up in a dungeon cooking up a massive rewrite (like atomic modeset? :-P)
03:16RSpliet: complete radiosilence while he and airlied estafette-hack a Pascal to avoid any risks of breaking an NDA... could be many things :-D
03:20karolherbst: mhh, yeah still
03:21karolherbst: he could still answer though
03:43karolherbst: RSpliet: nve7 was the 650 card?
03:43RSpliet: karolherbst: no
03:44RSpliet: the NVE7 in the vbios repo is my DDR3 GT640
03:44RSpliet: and yes
03:44RSpliet: the GTX 650 is an NVE7 as well
03:44karolherbst: right :)
03:44karolherbst: that's what I wanted to know
03:44karolherbst: I don't like putting two cards into the same directory though
03:45RSpliet: I don't, but I'll sort that out over the weekend
03:45karolherbst: ahh so you will upload the vbios and stuff?
03:45RSpliet: yeah, I'll take care of that
03:45RSpliet: I'm slightly behind on my uploads at the moment, sounds like a nice choir
03:47karolherbst: great, more vbios to check stuff
03:49karolherbst: ohh god no... I have to check that
03:50karolherbst: I thought I found a big error, but it's okay
03:51karolherbst: this unk entry in the voltage table is clearly some kind of index: 1024 - 1148, step size 4
04:02pmoreau: imirkin: Hum, ok. That might be conflicting with my handling of globals, since I get (and will most likely always get, at least for arrays) the offset from the kernel arguments.
04:02pmoreau: I have to check how the blob does that.
04:06pmoreau: imirkin: Ping, if you have some time to review: https://lists.freedesktop.org/archives/mesa-dev/2016-February/108477.html
04:09karolherbst: mupuf: I splitted up the unk in the voltage table: https://gist.github.com/karolherbst/e52d39c3ac07ba386fe5
04:09karolherbst: mupuf: this looks kind of right
04:10karolherbst: but funny how some indices are just missing
04:18karolherbst: RSpliet: do you have some time to test reclocking again on the 650? I think I figured out the missing piece
04:22karolherbst: ohh wait, it shouldn't change a thing though...
04:32pmoreau: imirkin: Right, so the blob doesn't have this fetching logic in the OpenCL samples I just looked at, which makes sense since we directly get the pointer from the user.
04:33pmoreau: (again, talking about arrays in global space, not images or textures)
06:54karolherbst: it is working! :O
06:54karolherbst: I got divinity to the loading screen :)
06:55karolherbst: yeah divinity needs shading_include
06:55karolherbst: and now I tried to get the parser accepting those shaders
06:56karolherbst: now the linker fails, because I don't include the shader actually, but that should be the easy part
06:59karolherbst: that bison/flex stuff isn't that intuitive sadly :/
06:59karolherbst: but I think I got that figured out now
07:27imirkin: pmoreau: yeah, i wouldn't expect the blob to do anything like that -- this is just the way that i expressed buffer objects... you can decree that fileIndex == -1 means that you shouldn't do any of that. or some other indication of the base address needing to be loaded... perhaps that should be done right at the from_tgsi level for buffers
07:30pmoreau: fileIndex is different from the X of cX, or is it the same?
07:31imirkin: i know that gmem on nv50 can have a legit fileindex though
07:32imirkin: so that might not be the best thing to do =/
07:32imirkin: but this needs to be communicated *somehow*
07:32imirkin: perhaps i should have created a new nv50/ir file
07:32imirkin: and then lowered it away
07:33pmoreau: I'll keep that in mind for now, and rethink about it when I'll have a look at how images are handled in SPIR-V.
07:36imirkin: remember that none of this stuff is set in stone
07:36imirkin: it is whatever we say it is
07:40pmoreau: I think, by ignoring this lowering for now, and fixing NTID, that I should be able to run the OpenCL examples that work on Tesla, on Fermi+.
07:41pmoreau: And I'll have a look at 64-bit constant propagation. Cause mad 8 0 0 gets optimised away when using 32bit imm on Tesla, but not when using 64bit ones on Fermi+ :-D (well, it does reduce it to mul 8 $r63, but still)
07:42imirkin: why are you using 64-bit imms??
07:42imirkin: or are these doubles?
07:42imirkin: if so, i don't think anyone's really looked too carefully at doubles
07:43pmoreau: 64bit imms for 64bit offsets computation, when accessing elements in an array or structure
07:43imirkin: ah. integer mad.
07:44imirkin: probably some dumb thing that checks for type == s32 || u32
07:44danofsatx: I've been doing some reading about nouveau support for 970m and 980m GPUs. The outlook isn't promising, but all the stuff I've found is from last year. Has nvidia finally released the firmware for nouveau to use, or are they still offering nothing but emtpy promises?
07:45imirkin: danofsatx: they have released firmware for GM20x, the kernel support for it should be available in kernel 4.6
07:45danofsatx: GM20x is the 900 series?
07:45imirkin: however that firmware won't allow us to reclock, so if you're looking at using nouveau for anything other than basic desktop tasks, it's unlikely to work out the way you want.
07:45danofsatx:doesn't know how to figure that out
07:45imirkin: GM20x is GM20x.
07:45karolherbst: danofsatx: there is no strict mapping
07:45imirkin: 900 series is a marketing term that combines everything imaginable.
07:46karolherbst: but yeah, all desktop 900 _so_far_ are gm20x
07:46imirkin: but mobiles can be kepler... maybe even fermi for all i know.
07:46danofsatx: I don't game, I just can't find a skylake based laptop with a 4k display and 32GB of RAM with anything BUT an nvidia GPU
07:46karolherbst: 800 had fermis
07:46karolherbst: but there are 900m keplers, right
07:46imirkin: danofsatx: in that case, the display/etc are driven by the intel gpu
07:47imirkin: and nouveau will happily power off the nvidia gpu.
07:47karolherbst: danofsatx: pro top: don't care about ram and buy those seperatly
07:47danofsatx: awesome. Now to figure out if the intel GPU can drive the 4x4k display... thanks folks!
07:47karolherbst: it can
07:47karolherbst: not 4
07:47imirkin: it can do 1
07:47imirkin: it can probably even do 3
07:47karolherbst: only 1 4k?
07:47imirkin: but i doubt it can do 4.
07:48karolherbst: yeah 3 is max
07:48imirkin: check with #intel-gfx
07:48danofsatx: that's the channel I was looking for. thanks imirkin
07:48karolherbst: yep, skylake is still limited to three displays
07:48imirkin: and remember that the panel counts as a display.
07:49danofsatx: yes, I'm aware of that.
07:49imirkin: if you're looking for open-source friendly options, i highly recommend amd.
07:50karolherbst: using reverse prime is still a mess though :/
07:50danofsatx: This is more or less dreaming, anyhow - I don't own any external 4k displays yet, and I'm looking at a Clevo P-650 based laptop.
07:50karolherbst: or amd only laptop, uh well
07:52karolherbst: well why a such a laptop when you don't need the perf?
07:52imirkin: presumably needs the perf... just not the gpu perf.
07:52karolherbst: mhh right
07:52imirkin: karolherbst: believe it or not, but people do other things than play games, on occasion.
07:53karolherbst: yeah, but those laptops are an overkill for that too
07:53danofsatx: I do QA for the Fedora project. I want the memory to support multiple concurrent VMs.
07:54karolherbst: danofsatx: so you basically need a i7 + a lot of ram
07:55karolherbst: I am not saying that those clevos are bad, just mine has really loud fans and they are usually overpriced if you do not need the gpu ;)
07:56danofsatx: I don't need the GPU. I need the CPU and RAM.
07:56karolherbst: yeah then only concentrate on the CPU
07:56karolherbst: as long as the manufacturer is no asshole, you can just buy RAM cheaper and replace it
07:56karolherbst: (and better RAM)
07:58danofsatx: I like my current Asus ROG laptop, but it has the NVIDIA in it also.
07:58danofsatx: I did see a rather nice Lenovo Thinkpad with a skylake, 16GB, and Intel video at Best Buy yesterday, though.
08:17karolherbst: uhh this shadeing_language_include dependency begins to be messy. The glsl parser needs to fetch stuff out of the opengl context :/
08:47kugel: karolherbst: I've tried your new kepler reclock branch (the _to_uptream one). I've got a lockup upon echo 0f > [...] while I was in-game already. It didn't lockup in a second attempt after reboot, however I wasn't in-game at that point (I played the game without problems afterwards)
08:47karolherbst: yeah there might be still some voltage issues left I need to investigate
08:48kugel: I'm still using the "return info.max" patch alongside
08:48karolherbst: ohh waot
08:49kugel: I saw you added voltage related work in the _v2 branch but it I didn't try that one
08:49karolherbst: kugel: can you use the stable_reclocking_kepler_v2 branch without this info.max hack?
08:49imirkin: kugel: you're driving displays off your board?
08:49imirkin: kugel: or is it accelerator-only?
08:50kugel: there's a separate display on the card
08:50imirkin: if you're driving displays, it's expected that reclocking will hang every so often.
08:51kugel: expected in what way? can't be fixed?
08:51karolherbst: imirkin: I get the feeling, that nouveau also sets the wrong voltage than expected on some gpus :/
08:52imirkin: kugel: never is a long time. but definitely isn't presently fixed :)
08:52karolherbst: like nouveau wants to set 1.1V, but it did set 0.8V instead
08:53kugel: imirkin: okay
08:53karolherbst: kugel: I want to verify something, can you switch to the nvidia driver, checkout my stable_reclocking_kepler_v2 branch and then report back when you are running on nvidia?
08:53karolherbst: kugel: I wrote a tool to read the current state out of the hardware and compare it with nouveau would do
08:53kugel: I just reported because I got this hang immediately after switching to the newer branch, been using the original recockling branch (+ "return info.max" patch) for a while without issues
08:54karolherbst: yeah maybe I messed something up :/ or just bad luck
08:54karolherbst: yeah no idea
08:54kugel:giggles at "recockling"
08:55karolherbst: wtvr :D
08:55kugel: karolherbst: I'll try to not forget, if you could pastebin the commands I should use?
08:56karolherbst: kugel: well just update the repository and checkout origin/stable_reclocking_kepler_v2
09:00kugel: karolherbst: alright, I'm not at my PC now :-)
09:00kugel: (s/PC/desk with egpu/)
09:52mooch: hey guys. i'm having a bit of trouble determining how pfifo works
09:55imirkin: mooch: http://envytools.readthedocs.org/en/latest/hw/fifo/index.html
09:57mooch: imirkin: well, i'm trying to figure out what registers i should implement first
09:58imirkin: iirc someone tried to emu the xbox gpu (nv2a), you could start there
09:59mooch: is it mostly the same between nv4 and nv2a?
09:59imirkin: pfifo? yeah, i think so.
10:00mooch: k good
10:01imirkin: G80 really switched it up
10:01imirkin: but before then it's basically the same... i think.
11:03guest65443: hi, having an issue with nouveau and ubuntu 15.10 new installation
11:05Jayhost: karolherbst , You asked how much more perf do I get with memory reclocked. Do you mean FPS in-game?
11:08guest65443: Dual monitor works wuth ubuntu live cd, but 2nd monitor always asleep after install.
11:26karolherbst: Jayhost: yeah
11:26karolherbst: Jayhost: and with vsync disabled
11:44guest65443: Any help for why dual monitor works with ubuntu live cd but not after installation (nouveau driver in use for both, no nvidia installed).
11:48Tom^: guest65443: which kernel, which gpu and which version of xorg and xf86-video-nouveau
11:48Tom^: since you said ubuntu 15.10 im fairly sure they are all quite old
11:50guest65443: Kernel 4.2.0-32 & xorg 1:7.7
11:50guest65443: xserver-xorg-video-mouveau 1:1.0.11
11:51guest65443: VGA is nvidia geforce 8400 GS
11:52Tom^: ah, well then it kinda depends what you mean with "asleep", have you tried actually setting it on or giving it a resolution with xrandr ?
11:53guest65443: xrandr shows it's configured and connected, but the display says 'no signal' and goes ino power safe mode.
11:54guest65443: However, when booting from live dvd, all works well.
11:55guest65443: The display shows up perfectly in ubuntu display control panel, and can be configured. But it seems no signal is reaching the monitor (except when the live dvd is used)
11:57guest65443: xrandr says DVI-I-1 connected primary, DVI-I-2 connected
11:59guest65443: The nvidia 8400 has a single DVI with a y cable to split to two monitors.
12:00Jayhost: karolherbst. I don't know. I was able to run RE4 on Dolphin at 30 fps 2x native [opposed to native] but I can do more tests with instructions
12:23karolherbst: Jayhost: you could run the unigine heaven benchmark for example
12:26Jayhost: phoronix says they have an open source version of unigine
12:26karolherbst: if you pay most likely
12:26karolherbst: it's not an uncommon license model, pay and you get the source under NDA and can tweak it for yourself
12:31Jayhost: I was using a test called glmark the other day
12:33Jayhost: What I meant was if there was an easy way to reclock without memory
12:35Jayhost: also that sounds a whole lot like a paywall
14:48mooch: imirkin: on nv3:nv30 and nv31:g80, how do the plls in pramdac work?
14:48mooch: i saw xqemu's code, which said (crystal * n) / (1 << p) / m
14:48mooch: but that doesn't give me the right frequencies when i use 13.5 mhz as the crystal
14:49mooch: also, wtf is vip in vpll_source
14:49mooch: what's mpll_prog?
18:43Jayhost: karolherbst glmark2 - no change:freq change: memory change: http://pastebin.ca/3388312 http://pastebin.ca/3388314 http://pastebin.ca/3388317
21:28espes__: mooch: very likely xqemu's wrong for that stuff. if you figure out what it should be please let me know
21:28mooch: espes__: actually, the calculations are correct, it seems
21:28mooch: it's just that there's a bug in the svga code that calculates this shit
21:29mooch: though if you want to emulate variable refresh rates, vpll's what you're looking for
21:33mooch: espes__: i'm still dealing with that real mode access issue