00:24wolter: pq: where could I get a list of all the modules I need to blacklist?
00:24wolter: pq: and so a systemd service is fine for this task?
00:29pq: wolter, I'm not talking about modules. I'm talking about libraries.
00:30wolter: pq and so what do you mean with this switch?
00:30pq: wolter, the basoc problem is that NVIDIA installs libGL.so, which cannot work with Nouveau, and vice-versa. There are many more than just libGL.so.
00:31wolter: pq and do these libraries replace nouveau-installed libraries?
00:32pq: so you have to take a look at what all libraries the NVIDIA driver package overwrites or installs, and make those switchable vs. the nouveau (Mesa) provided ones (if your distribution hasn't done that work for you).
00:32pq: there isn't a single package for the Nouveau versions of the libraries, NVIDIA will overwrite files from many packages.
00:33pq: I assume ubuntu has some tool to manage proprietary driver installations, but I do not know if it supports switching rather than install/uninstall.
00:34wolter: pq but now I'm very puzzled since I am able to boot with nouveau, my only problem now seems to be that I have to blacklist modules before I reboot, so what is happening really?
00:34mlankhorst: it supports switching, update-alternatives --config + ldconfig + regenerate initramfs..
00:34pq: mlankhorst, generate initramfs? do you have to boot twice to switch on boot?
00:35pq: wolter, sure, there libraries are mostly for 3D acceleration, so games and stuff. You can boot otherwise, but games won't run.
00:35mlankhorst: no, just to update the module blacklist..
00:35pq: mlankhorst, ah.
00:35pq: *these libraries
00:36pq: so anything that uses OpenGL or maybe some video acceleration would need the libraries switched, other stuff may work without correcting the libraries
00:36wolter: pq yes I have to boot twice to switch on boot
00:36wolter: pq but I have ran games :-.
00:37pq: maybe something already does some swtiching?
00:37pq: like that ubuntu tool
00:38imirkin: can someone with a nv50 plugged in test https://github.com/imirkin/mesa/commit/c48ae2692c9367f91ee85e3f72079e6d9b33b02b.patch against the latest piglit arb_pipeline_statistics_query tests?
00:38wolter: pq but I see where you're getting at. In the first place I got this idea from the gentoo wiki, where they list some methods to keep both nvidia and nouveau drivers installed, do you say this is flawed?
00:38wolter: pq: yeah, I was hoping not needing to use it though
00:38pq: wolter, I'm not saying anything is flawed, you just need to take all the relevant things into account.
00:39imirkin: wolter: eselect opengl should work just fine
00:39pq: gentoo also has a special tool to switch all relevant libraries
00:39wolter: imirkin: what would that command be fore?
00:40imirkin: switching between gl libraries
00:40imirkin: eselect opengl set nvidia, or eselect opengl set xorg-x11
00:40imirkin: used to just update some symlinks, but now i think they're copying actual files around for some reason
00:40wolter: pq but it does seem, because what mlankhorst said about update-alternatives, that nvidia installs its libraries where they don't replace others
00:41mlankhorst: they don't :P
00:41wolter: imirkin: it seems I don't have that application in ubuntu, are you sure it works for ubuntu, or is it analogous to update-alternatives that mlankhorst mentioned?
00:41wolter: mlankhorst: :D
00:41mlankhorst: library and xorg path is updated to find the nvidia libs first
00:41imirkin: i am sure it doesn't work on ubuntu -- it's a gentoo tool.
00:41pq: wolter, that again depends on how you install the NVIDIA drivers. If you take nvidia's original package and install it by default, I think it will overwrite stuff. Distro packages do differently.
00:41wolter: mlankhorst: so according to you, I only need to run some update-alternatives commands?
00:42wolter: pq: luckily I have distro package :D
00:42wolter: mlankhorst: (for my system to boot with nvidia or nouveau drivers)
00:42pq: right, but this is not a distro-specific channel, so we usually talk in general terms :-)
00:43pq: and if you want to replicate what your distro tools do, it helps to understand what the underlying problem is
00:43wolter: pq yes :) I've been trying to get to the bottom of it but it has taken me hours of research
00:44mlankhorst: 2 things really, 1. updating loader path to find opengl libs first.. 2. add a file to xorg.conf.d to prepend nvidia to the module path
00:44mlankhorst: and 3. adding blacklist nouveau to modprobe.conf
00:44wolter: mlankhorst: http://paste.ubuntu.com/10287337/ this are my alternatives containing the word nvidia in it (update-alternatives --get-selections | grep nvidia)
00:45pq: 0. making sure installing the drivers packages does not overwrite anything ;-)
00:45mlankhorst: I have some tools to do it by hand but its tedious :p
00:46wolter: mlankhorst: I currently have set up a script that systemd launches that 1) blacklists the unwanted module and 2) sets the wanted module/driver for the default Xorg device (gpu). All I'm missing then is updating the loader path?
00:46wolter: pq nice zeroth law
00:47mlankhorst: just 2 and 3
00:47wolter: mlankhorst: what do you mean with "prepend nvidia to the module path"?
00:48mlankhorst: it's a xorg.conf snippet that updates module path
00:48wolter: In fact, my current problem seems to be that my system "prefers" the nvidia driver. In the xorg log I see that nvidia is an "autoconfigured driver 0" and nouveau comes as 1
00:49wolter: and another thing, is it true that this all has to occur before the kernel is loaded and not before X is started?
00:49mlankhorst: oh that's always the case in ubuntu, just ignore it
00:49mlankhorst: if you update xorg.conf path so it won't find the nvidia module and the nvidia glx module it will run nouveau
00:50wolter: mlankhorst: can you englighten me more in that last thing?
00:50wolter: how do I trick X into that?
00:51mlankhorst: [ 58.500] (==) Using config file: "/etc/X11/xorg.conf"
00:51mlankhorst: [ 58.500] (==) Using system config directory "/usr/share/X11/xorg.conf.d"
00:52wolter: mlankhorst: you mean having X not use the xorg.conf.d directory?
00:53mlankhorst: no, just add or remove a file for nvidia there..
00:59wolter: mlankhorst: oh well, there are no nvidia-installed files there
00:59mlankhorst: there might be if nvidia is active..
01:11wolter: sorry I had to restart
01:11wolter: mlankhorst: did you say anything to me that I didn't reply to?
01:11wolter: or anyone else :)
01:13mlankhorst: 09:59 mlankhorst: there might be if nvidia is active..
01:15wolter: oh I thought I replied that; nvidia installed no files in xorg.conf.d
01:16mlankhorst: no but there is a file there if the current alternative is nvidia
01:21wolter: mlankhorst: only one I created, that lists the nouveau driver for device0
01:22mlankhorst: there's a symlink there if nvidia is the current alternative for x86_64-gnu-gl_conf or w/e
01:29wolter: not in /etc/X11/xorg.conf.d :-.
01:31mlankhorst: 09:51 mlankhorst: [ 58.500] (==) Using system config directory
01:33wolter: mlankhorst: :-O !!
01:34wolter: Although none of my files there seem to mention nvidia or nouveau
01:36wolter: only glamoregl
01:39mlankhorst: does it mention extra-modules ?
01:44wolter: this is the ls output http://paste.ubuntu.com/10288022/
01:51mlankhorst: in that case just do apt-get source nvidia-304 and figure out what it's doing :p
01:51wolter: thanks mlankhorst
01:59mlankhorst: ah it was changed between versions
05:23rhn_mk1: joi: do you think fglrx IOCTLs may depend on the chipset? what GPU did you analyze?
06:13imirkin: can someone with a nv50 plugged in test https://github.com/imirkin/mesa/commit/07d431702c8e91036e2751bc7fbc7eb9d8426afb.patch against the latest piglit arb_pipeline_statistics_query tests?
06:20mlankhorst: nv50 or any tesla?
06:41imirkin: mlankhorst: any tesla
08:50jiji: could I ask some questions about GPU ?
09:08imirkin: ask away
09:18joi: rhn_mk1: heh, that would be annoying; I used HD5450
09:18jiji: Ok, OpenGl is a library to order the GPU, such as print me a cube with green texture. Is it possible to control the GPU without OpenGL ?
09:19imirkin: jiji: sure! the opengl backend driver is written without using opengl :)
09:19imirkin: (there are also other api's like OpenCL which can execute on the gpu)
09:19jiji: But how ?
09:21imirkin: if you don't want to use one of the defined api's and their driver
09:21imirkin: you can just submit commands to the gpu
09:21imirkin: each gpu family has its own commands, its own everything
09:21imirkin: you can look at how the relevant mesa drivers do this... it's not particularly easy
09:22jiji: Just one question : It is easier to program with assembly than command gpu ?
09:24imirkin: not sure what that means
09:24imirkin: please rephrase your question
09:25jiji: Ok : What is the easiest ? Programming in ASM32 or to use a gpu without OpenGL ?
09:26imirkin: you mean like x86-32 assembly?
09:27imirkin: i'd say x86-32 asm is easier -- you can debug things, instructions follow a documented spec, and there's not a whole ton of diff things you can do.
09:27jiji: Yes, sorry i am not accurate because i don't know programming with others architectures in asm :)
09:28imirkin: and you'd still be stuck writing code in the gpu's instruction set for the shader programs in its (undocumented) binary isa
09:29imirkin: moral of the story -- if you're even thinking about asking that question, you should be using OpenGL/OpenCL/etc
09:29jiji: Ok, so I is harder than I thought.
09:29jiji: I ask it because this :
09:30jiji: Go to 4 min
09:31imirkin: #1 - you are not john carmack.
09:31imirkin: i.e. not the guy who personally wrote drivers for his games for early 3d accel hardware
09:32imirkin: it's not impossible (hey, the opengl driver does it _somehow_, right), but it's not a quick & simple project
09:32imirkin: while i could whip up a program that drew a 3d textured cube in like 5 minutes
09:33imirkin: the same thing would probably take me at least a full day to do if i had to create a command stream (without cheating and capturing the command stream output by the gl driver)
09:33imirkin: _and_ that would only work on one specific gpu family
09:35imirkin: what he's talking about is surfacing api's to better expose features like HSA
09:36jiji: Ok : So I linked a video where I understood nothing ! :)
09:37imirkin: i didn't watch the whole thing, just a short segment after minute 4
09:38imirkin: i think it's easy to have this idealized view of what the gpu does... but when you pull back the curtain, it's a lot lower level of an api than you might think
09:40jiji: Yes, OpenGl gives this feeling of simple gpu, (high level), but it is not so simple.
09:41imirkin: some gpu's are more opengl-y (nvidia), while others have relatively little to do with the api (powervr from my understanding)
09:41imirkin: amd and intel are kinda in between
09:42mlankhorst: well they're opengl-like but they tend to do more in sw
09:43mlankhorst: except powervr which is a big wtf from what I gathered :P
09:43imirkin: mlankhorst: well, e.g. the r600 shader pipeline bears little relation to the opengl one
09:43mlankhorst: oh that :p
09:43mlankhorst: didn't look that closely into radeon, touched most of the driver though
09:44imirkin: they have a shader that fetches vertices... the way geometry is implemented isn't... the same as gl :)
09:44imirkin: i dunno all the details
09:44imirkin: but i know it's got a lot of things that GL doesn't
09:45imirkin: mlankhorst: did you have a chance to try my pipeline stat nv50 impl?
09:47jiji: imirkin: So your advice (after Me =/= Carmack), do OpenGL and when you will master OpenGL maybe come back here.
09:47imirkin: jiji: when you master opengl, you will realize that you were asking the wrong question
09:48mlankhorst: sorry no, tomorrow probably :P
09:49jiji: I am just fearing about one thing : I must learn the up to date OpenGL or older like OpenGL 1/2.1
09:50imirkin: GL1 was *very* different
09:50imirkin: GL2 has a bunch of stuff that's like "well, let's take GL1 and add shaders", so still lots of legacy cruft
09:51imirkin: GL3.1+ core profiles did away with a lot of the stupidity (far from all, of course)
09:51imirkin: but now instead of there being 10 ways to do everything, there are only 5
09:53jiji: I must avoid the GL1/2/3 tutorial and learn GL4 ?
09:53imirkin: i would definitely start with a GL4 tutorial
09:53imirkin: GL4 also *adds* a bunch of fancy features, but hopefully a tutorial won't start off with combining tessellation and compute shaders to generate vertex lists for later draws :)
09:56jiji: It is not his point of view -> http://lazyfoo.net/tutorials/OpenGL/00_introduction/index.php
09:57imirkin: yeah, might not be a bad way to go about a tutorial
09:57imirkin: but that's definitely _not_ the direction that the graphics api's are taking
09:58imirkin: kinda how like cpu's have an "add" instruction, but not a "compute a 64-tap fft of this data" instruction
09:58imirkin: (dunno, maybe not the best analogy... sorta fits)
10:00jiji: I don't understand you example :<
10:01imirkin: yeah don't worry about it
10:04jiji: Ok - I will see the linked tutorial and I will come back here when I will be able to understand what you write :)
10:05imirkin: jiji: what's your goal btw?
10:07jiji: Understand how graphics work.
10:07buhman: nobody knows
10:08imirkin: jiji: ok, opengl is probably a good start
10:10jiji: imirkin: I will follow your advice, and maybe maybe i will be able to help projects like tesseract, panda3d
10:11jiji: buhman: That means that it is a too high goal ?
10:11buhman: jiji: never; it's just a lot of stuff to chew if you jump all the way down the rabbit hole
10:13buhman: 'understand how graphics work' is a pretty broad goal
10:14jiji: Ok, I must be more accurate : I want to understand how a 3d engine like tesseract work.
10:15imirkin: you should speak with the tesseract guys
10:17jiji: Ok it is not the place. But now i thought that they will not accept a noob !
10:18imirkin: most opensource projects are looking for capable and interested contributors, even if they don't necessarily have the exact required skillset
10:21jiji: I doubt : My aim it is to understand an algorithm like that https://www.youtube.com/watch?v=dHBLPNNjkrg. It is so nice. or the new ssao in blender viewport !
10:23imirkin: yeahhhh.... that's not step 1 :)
10:23imirkin: step 1 is to understand how a triangle is drawn
10:23imirkin: then you can move up to fancier things :)
10:25jiji: Thx speak with you help me to give me realistic aims, so forgot global illumination, and focus with simple things ok !
11:24rhn_mk1: joi: it might be interesting to you that some data seems to be exchanged over X in case of OpenCL, like mmap offsets. with the X-less setup, ioctls are different
11:50joi: well, I don't intend to improve it much - it was an interesting experiment/proof it's possible
14:51AnarquistaLibre: Hello all! I hope someone has the knowledge to help me. What would be the newest Nvidia graphics card with 3D acceleration that I could use on Trisquel?
14:52AnarquistaLibre: Or what would be the best card in general? Doesn't need to be the most modern.
14:52imirkin: GKxxx for both of those answers
14:53AnarquistaLibre: Not familiar with old Nvidia lines of cards, Could you tell me the full name of one of them so I could look into it?
14:54AnarquistaLibre: There is also a GeForce 8400GS on ThinkPenguin I was thinking of.
14:54imirkin: they're the almost-newest line of nvidia cards
14:54imirkin: not maxwell, but the ones right before that
14:55AnarquistaLibre: And they are well supported? No power management issues and such?
14:56imirkin: sure, there are issues
14:56imirkin: but reclocking is moderately supported on keplers
14:56imirkin: manual, and they generally won't reclock to the highest speed
14:56imirkin: GeForce 8400GS should work just fine as well
14:57AnarquistaLibre: So something as new as a GTX 780 Ti would work? :O
14:57imirkin: it's an older Tesla family card
14:57imirkin: yeah, 780 Ti should work fine. Tom^ has one, iirc he hasn't reported any issues of late.
14:57imirkin: except for the fact that it won't clock to the highest setting
14:58AnarquistaLibre: Wonder what the best kepler card would be. I'm assuming the 800m line is weaker than the 700.
14:59imirkin: well, be careful -- some mobile optimus gpu's don't end up coming up with nouveau at all
14:59imirkin: which is unfortunate
14:59imirkin: there's no great way to predict which ones will have the issue, it's on a per-laptop basis, not per-chip
15:00imirkin: and i think some of the 8xxM's are actually maxwell, which won't work with trisquel at all right now
16:52rabidfurby1: could someone help me troubleshoot a quad monitor setup? I'm able to get 3 monitors working using http://nouveau.freedesktop.org/wiki/MultiMonitorDesktop/ and this config (http://ix.io/gpH) but it's suboptimal because I'm using one monitor with a VGA output, and across my two cards I've got 3 DVI ports plus 1 HDMI, so it seems like I could do all-digital output for 4 monitors
16:52rabidfurby1: lscpi: http://ix.io/gpI and /sys/class/drm/: http://ix.io/gpJ
16:53imirkin: rabidfurby1: neither of your cards supports 3 outputs
16:53imirkin: rabidfurby1: oh, you should be able to plug into hdmi + dvi on the card you have with vga on it
16:58rabidfurby1: yeah, that's what I thought, but even trying a dual-monitor setup with DVI+HDMI, or the two DVIs on the 9500GT, I can't seem to make it work
16:58imirkin: what happens?
16:58imirkin: i don't know of any reason that would happen
17:00rabidfurby1: right now I have 2 monitors hooked up to the 2 DVIs on the 9500GT, and with this xorg.conf: http://ix.io/gpK I get output on one monitor, blank on the 2nd, and this Xorg log: http://ix.io/gpL
17:00imirkin: that's expected
17:00imirkin: you only have 2 screens defined
17:00imirkin: oh wait
17:01rabidfurby1: yeah, I'm trying to narrow it down, just to prove to myself that 2 digital outputs from either monitor works
17:01rabidfurby1: and so far no luck with either one
17:03imirkin: i wonder if the "gpu" thing is messing things up
17:03imirkin: add an option "AutoAddGPU" "off"
17:03imirkin: in the ServerFlags section
17:03rabidfurby1: in ServerLayout?
17:06rabidfurby1: same results. here's the latest log: http://ix.io/gpM
17:07imirkin: very odd
17:08imirkin: did someone break xinerama?
17:08imirkin: with just the one gpu, you don't have to do xinerama
17:08imirkin: maybe there's something wrong with that connector or monitor or cable...
17:08imirkin: try removing the xorg.conf entirely and see what happens
17:09imirkin: you can then use xrandr to do things
17:09rabidfurby1: good point about the cables too, let me see if I have a spare somewhere too
17:09imirkin: just move the xorg.conf out of the way
17:09imirkin: and try again
17:09imirkin: you should be able to use xrandr
17:10rabidfurby1: just moved it and restarted slim - and now no output at all
17:10rabidfurby1: I wonder if it's defaulting to the other GPU?
17:10imirkin: could be yea
17:10rabidfurby1: since that one is "first" in lspci
17:10imirkin: you can achieve all this stuff with xrandr btw, without xinerama
17:11imirkin: using "reverse prime"
17:12rabidfurby1: yeah, I'm fine with whatever works. I was in here several months ago, getting help with my 3-monitor setup, and whoever helped me back then recommended the ZaphodHeads setup, and it worked great, albeit with the VGA cable
17:15imirkin: yeah... xinerama has the benefit that it's simpler and just works out of the box
17:15imirkin: once you configure it
17:15imirkin: but you lose direct accel
17:15imirkin: with xrandr, i don't know if it's possible to make it so that the desired screen config happens on start
17:16imirkin: since you have to set up the provider source/etc
17:16rabidfurby1: this is my work computer, so I never need 3D accel, so that's a OK trade-off
17:16imirkin: i can't think of a reason why the vga vs dvi would matter
17:16imirkin: try swapping the dvi ports
17:16imirkin: and see if the same monitor keeps working
17:16rabidfurby1: good idea
17:16imirkin: or if they switch
17:17rabidfurby1: I just put the xorg.conf back and swapped the cable that wasn't working and that didn't change any output either
17:18imirkin: i.e. the same monitor stayed on?
17:18imirkin: that means the other monitor's cable/connector/something is broken
17:18rabidfurby1: oh, nvm, I mean I just swapped the cable with a different cable
17:18imirkin: what happens if you flip them on the card
17:18rabidfurby1: I'm testing with a different output now
17:19imirkin: coz the edid works -- you can see the two monitors returning diff serial numbers
17:20imirkin: so... it's gotta be something funky
17:21rabidfurby1: so now I have Device0 as DVI-I-3 and Device1 as DVI-I-2, where before they were reversed
17:21rabidfurby1: and that didn't change anything, same monitor (my leftmost) is getting output
17:22rabidfurby1: the other weirdness is, if I move my mouse off the left edge of the single monitor that has output, the cursor vanishes
17:22rabidfurby1: as if it's going to another part of the screen that just isn't being displayed
17:23rabidfurby1: I'm going to switch to trying the HDMI + DVI combo - if I can get that working, I'll toss the 9500GT and bring in a spare 210 that I have in my HTPC at home
17:24imirkin: well if you swapped the connectors
17:24imirkin: and the same monitor is still displaying
17:24imirkin: it must be some property of the "other" monitor
17:24imirkin: either the cable or the connector on it...
17:24imirkin: try plugging it into the other card, see what happens?
17:25rabidfurby1: yeah, that's what I'm going to try
17:25rabidfurby1: because this has all been with 2 DVIs from the 9500GT
17:25rabidfurby1: and it's so old it wouldn't surprise me too much if it's half-supported
17:25imirkin: should be fully supported afaik
17:32rabidfurby1: so 1 monitor on DVI and 1 on HDMI, I get X output on the DVI monitor, and initially nothing on HDMI - but I Ctrl-Alt-F2'd to a console, logged in, and then Alt-F7'd back to X11 - and then I had the console output leftover on the HDMI monitor
17:33imirkin: very odd
17:33rabidfurby1: I also switched from awesome to fluxbox in my .xinitrc, just to rule that out as a cause, and nothing changed as a result of that either
17:33imirkin: that shouldn't matter
17:33rabidfurby1: yeah, just wanted to make sure
17:34imirkin: very weird.
17:34rabidfurby1: with the DVI+HDMI pair, it does seem like the HDMI monitor is displaying the console, rather than X
17:34rabidfurby1: I just rebooted and all I get on that monitor is an _ at the top left
17:35rabidfurby1: hmm: http://ix.io/gpN
17:35rabidfurby1: near the end: "xf86: remove device 0 /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card0"
17:36rabidfurby1: followed by "failed to find screen to remove"
17:36imirkin: [ 3.871] (WW) NOUVEAU(1): No outputs definitely connected, trying again...
17:36imirkin: is hdmi-a-1 connected?
17:36imirkin: you can check in /sys
17:37rabidfurby1: yes, /sys/class/drm/card0-HDMI-A-1/status is "connected"
17:37imirkin: and modes has things in it?
17:38rabidfurby1: yes, including 1920x1080, which is its native resolution
17:38imirkin: what's the xorg.conf right now?
17:39imirkin: yeah i got nothing... someone must have killed xinerama... or zaphodheads
17:40imirkin: could be an X 1.17 thing
17:40imirkin: anyways... just switch over to the xrandr method
17:40rabidfurby1: I'm on an up-to-date Arch, so it is rather bleeding-edge
17:40imirkin: get rid of your xorg.conf
17:41imirkin: and restart
17:41imirkin: the first card's monitors should be accessible from xrandr
17:41imirkin: to get the second card, you can do
17:41imirkin: xrandr --setprovideroutputsource 1 0
17:41imirkin: which will hook up the second card's outputs to the first card
17:42rabidfurby1: OK, then normal xrandr foo --left-of bar stuff?
17:42rabidfurby1: ok, I'll give that a try
17:42rabidfurby1: thanks for all your help
17:44rick_: Sorry to jump in here but, speaking of old hardware, I'm attempting to get dual monitors working on a 6600 LE in a Power Mac Dual Core G5 running Debian Wheezy. Either monitor works if I plug it into the port closest to the mothboard. I've built the latest nouveau drivers from source, but I couldn't get X to do anything other than blink a cursor with the new drivers. The stable nouveau version in Wheezy works, but so far only with one
17:44rick_: monitor. Ideas?
17:45imirkin: that was a lot of words
17:45imirkin: you have 2 connectors
17:45imirkin: one consistently works, one consistently doesn't?
17:45imirkin: or you can't get it to work at all in X?
17:46rick_: The stable version of nouveau in Wheezy works on one monitor at a time with X
17:46imirkin: rick_: can you provide some logs? dmesg + Xorg log
17:47rick_: sure. As soon as I figure out how to post them...
17:47imirkin: or hastebin
17:47imirkin: or your favourite service
17:48imirkin: or your least favourite one... all the same to me, really
17:59rabidfurby1: well fuck a duck with a truck
17:59rabidfurby1: imirkin: my output names were wrong the whole time
17:59rick_: dmesg http://pastebin.com/CyjPbA2A
17:59rabidfurby1: I moved my xorg.conf aside, and started tinkering with xrandr
18:00imirkin: xrandr output names are diff
18:00imirkin: the zaphodhead names are the kernel's names
18:00rick_: xorg http://pastebin.com/NXfhvTw2
18:00rabidfurby1: yeah, but using them in my xorg.conf works
18:00imirkin: xrandr names are... different
18:00imirkin: *that* is a bug
18:00rabidfurby1: I've got 3 monitors working perfectly right now
18:00imirkin: airlied: ---^
18:00rabidfurby1: using HDMI-1 instead of HDMI-A-1
18:00imirkin: do you know where ZaphodHeads could have gotten messed up?
18:01imirkin: rick_: that's a really old kernel
18:02airlied: no xrandr and kernel names always differed
18:02imirkin: airlied: yes
18:02imirkin: airlied: but it sounds like he has to use the randr names for zaphodheads
18:02imirkin: whereas it used to be kernel names
18:02airlied: it should have been randr names all along I thouht
18:03imirkin: coz how do you address a second gpu's connectors?
18:03imirkin: they never get randr names...
18:03imirkin: (do they?)
18:04imirkin: airlied: http://nouveau.freedesktop.org/wiki/MultiMonitorDesktop/
18:04imirkin: that guide, which has been around for quite a while, certainly thought it should be the kernel names
18:05imirkin: rick_: what happens when you run 'xrandr -q'
18:05rabidfurby1: airlied: my /sys/class/drm: http://ix.io/gpJ and my xrandr output before xrandr --setprovideroutputsource 1 0: http://ix.io/gpP and after: http://ix.io/gpQ
18:06rick_: I thought 3.2 was fairly recent. uname -r 3.2.0-4-powerpc64
18:06imirkin: rick_: it was released in 2012 or so
18:06imirkin: Wed Jan 4 15:55:44 2012 -0800
18:06rabidfurby1: if I use HDMI-A-1, DVI-I-2, or DVI-I-3 in my xorg.conf, I get nothing
18:06rabidfurby1: I tried HDMI-1, and it worked perfectly
18:07rabidfurby1: and I'm setting it up with DVI-I-1-2 and DVI-I-1-3 now
18:07rick_: Oh.. That is fairly old.
18:07rick_: xrandr -q Can't open display
18:07imirkin: rick_: uhm... is X running?
18:08imirkin: and did you run it in X?
18:08airlied:isn't sure zaphodheads work with secondary gpus
18:08imirkin: airlied: that was the whole point of zaphodheads :p
18:08imirkin: (or one of the points)
18:08airlied: no it is for using one GPU in two screens
18:08imirkin: and it used to work fine... someone made like a 20-screen video wall
18:08imirkin: using that multimonitordesktop guide as a template
18:09airlied: oh okay not slave gpus then secondary ones
18:09imirkin: (with 10 gpu's)
18:09imirkin: but the config has it as a secondary gpu
18:09imirkin: take a look at that wiki link
18:09imirkin: it's not using that GPU fanciness
18:10airlied: the driver uses randr names
18:10airlied: the code hasn't changed since it was written I don't think
18:10imirkin: where do the randr names come from in a secondary gpu situation?
18:10imirkin: does it just keep incrementing the counter, similar to what the kernel does?
18:11rick_: ran from within X www.pastebin.com/1GJicsAM
18:11airlied: imirkin: the driver gets loaded multiple times for secondary gus
18:12imirkin: rick_: and the monitor's not displaying anything?
18:12airlied: so the xrandr names are local
18:12airlied: and xrandr doesn't work with xinerama
18:12imirkin: airlied: ok, so it'll get DVI-I-1 multiple times?
18:12rick_: DVI 1 is working great. Nothing on 2
18:12imirkin: rick_: DVI-I-1 seems to be a fixed-mode panel... is that right?
18:12imirkin: rick_: anyways, i'd try a newer kernel
18:13airlied: imirkin: yes the xorg.conf would
18:13imirkin: airlied: so... why did it work with DVI-I-3 and 4?
18:13rick_: IDK Apple cinema display. I'll give a newer kernel a shot. Thanks.
18:14airlied: that sounds like the slave gpu loading path
18:15airlied: though maybe nouveau just reuses the kernel names
18:15airlied: or close enough so then you'd get 3/4
18:15airlied: I thought it would number them from scratch each time, like other drivers
18:23jrayhawk: http://www.omgwallhack.org/home/jrayhawk/img/hovel/20140725_003.jpg i might be qualified to help
18:23jrayhawk: AFAICT, of the three of multi-device, xinerama, and xrandr, you get two choose any who.
18:24jrayhawk: Zaphodheads is good, but you'll want to disable xinerama.
18:24jrayhawk: So you'll wind up with a multiple screen server.
18:24imirkin: without xinerama you end up with separate screens
18:24imirkin: which most people can't deal with
18:24imirkin: you're probably a rare exception. even WindowMaker barfs on it now, although it definitely used to work.
18:26jrayhawk: Yeah, I've looked at what I believe to be every tiling window manager and found two that even support multiple screens, and one of them doesn't count because its author went crazy years ago.
18:29jrayhawk: Also a lot of many-output cards only support a limited number of CRTCs; I have four outputs on this G94GLM, but only get to use two at any given time.
18:29imirkin: with nvidia you have to get to kepler for 4 CRTC's
18:30imirkin: i think a bunch of radeons support 6
18:30jrayhawk: Yeah, I started adding radeons to this laptop.
18:31jrayhawk: http://www.omgwallhack.org/home/jrayhawk/img/hovel/100_2631.JPG you can see them in the lower-left here
18:32jrayhawk: s/two choose any who/to choose any two/
18:34airlied: radeon with an MST bug :)
18:37imirkin: jrayhawk: is that a minipcie <-> pcie slot adapter?
18:38imirkin: didn't know you could get that
18:38buhman: of course
18:39jrayhawk: Electrically they're just funny-shaped 1x slots.
18:39buhman: 1-lane → 16-lane PCIe adapters are also fun
18:40jrayhawk: You can also combine lanes somewhat arbitrarily.
18:42buhman: I've wanted to combine (mPCIe→1-lane PCIe)→(1-lane PCIe→16-lane PCIe) adapters to connect a gigantic GPU to a tiny netbook
18:44imirkin: jrayhawk: does this support hotplug?
18:44jrayhawk: On expresscard slots, at least. I don't know if MiniPCIE ports would be electrically provisioned for that.
18:45imirkin: jrayhawk: what about regular motherboard pcie with their little passthrough adapter?
18:45jrayhawk: You'll want to manually echo 1 > /sys/bus/pci/devices/$device/remove and echo 1 > /sys/bus/pci/rescan as appropriate.
18:45imirkin: basically... could i use this to avoid having to reboot my comp to swap gpu's
18:47jrayhawk: Not sure I've ever seen desktop boards provisioned with hotplugable PCI-E; generally I only see that on high-end servers.
18:48jrayhawk: I guess I don't know enough about what's needed on the hardware side to say if that can be hacked in by an adapter.
18:49imirkin: but... me wantee!
20:04imirkin: skeggsb: did you end up getting that gk208 going on the be box?
21:00rick_: Trying to X working with nouveau on a GeForce 6600 LE. Hardware is a PowerPC G5. The card is erroring out right now with nouveau. xorg: http://pastebin.com/uE7ymPkh dmesg: http://pastebin.com/XtJ1NbGr
21:00rick_: I'm running Debian Jessie
21:00rick_: Any ideas?
21:01imirkin: are both screens workign now?
21:01rick_: Neither screen works :(
21:02imirkin: i see that gnome dies
21:02imirkin: but both screens are powered off now?
21:02imirkin: even before the hang happnes?
21:02rick_: One monitor is displaying a funky digital color pattern. nothing on monitor two.
21:04imirkin: sorry, dunno =/
21:06rick_: Well, thanks for trying. I wish there was at least a closed source driver for ppc :/
21:07rick_: I'm going to try a Quadro 4500. Maybe a more mainstream GeForce 7 based card will work better...
21:08Tom^: imirkin: well 780ti only works with skeggsb repo :p
21:09Tom^: imirkin: so yea i guess it works, as fast as this enters the kernel http://cgit.freedesktop.org/~darktama/nouveau/commit/?id=6280db95d596859c2f9a12638ba315b14a02d32b
21:09imirkin: hopefully the relevant patch gets cc'd
21:09imirkin: [to stable]