01:01 mlankhorst: RSpliet: dims in general, but easy to just add WARN_ON near every -ERANGE
01:03 mlankhorst: oops
01:03 mlankhorst: afk :)
01:26 Justice: does nouveau have issues with nvidia 7000M ?
01:26 Justice: seems im getting graphical corruption in unity.
01:41 karolherbst: Justice: 7000M or 700M?
01:41 Justice: 7000M the old one
02:07 karolherbst: and with unity you mean this fairly modern game engine?
02:08 Justice: desktop
02:08 Justice: ubuntu unity
02:09 Justice: gonna try mate since the laptop just cries when I tried legacy nvidia drivers.
02:10 karolherbst: ohh that unity
02:10 karolherbst: mhhh
02:10 karolherbst: maybe better go with xfce then?
02:10 karolherbst: though I don't know the footprint of mate in general
02:11 Justice: I heard mate is quite small "based on gnome 2"
02:12 Justice: xfce is my next bet
02:13 Justice: after that i will just install arch with openbox or something.
02:14 Justice: laptop came with Vista.....and had 800MB RAM...
02:15 karolherbst: :D
02:15 karolherbst: I never unterstood this, why save RAM if RAM is really cheap
02:16 karolherbst: even nowdays some laptops get like 2GB ram
02:16 karolherbst: Justice: but I would really go with xfce, because this is intented to run on old systems
02:17 karolherbst: mate is just gnome 2 only modern and fancy, because some disliked the gnome 3 style
02:17 karolherbst: so don't expect mate to run well on old hardware, cause that's not the target
02:17 karolherbst: even compositing with xfce runs pretty well on ancient hardware
02:19 Justice: hm
02:20 karolherbst: I once installed xfce on a old power mac g4 with dual 867MHz, but I also got a radeon 9700 Pro :D so that's not so fair to compare with yours
02:20 karolherbst: but it still run pretty good
02:20 gordan: XFCE is overrated.
02:20 karolherbst: gordan: howso?
02:20 gordan: I have run KDE reasonably comfortably on Toshiba AC100 with only 500MB of RAM.
02:20 karolherbst: it's not a problem of RAM
02:21 karolherbst: what gpu do you have?
02:21 gordan: On the AC100? Framebuffer.
02:21 karolherbst: yeah well
02:21 karolherbst: so no hw accell at all
02:21 gordan: The graphical performance will be affected by the app, not the desktop environment.
02:21 karolherbst: not true
02:21 Justice: if the desktop is composited it will
02:21 gordan: Depending on how many fancy effects you are running (I always disable them all anyway, they're disabled by default).
02:22 karolherbst: because the window manager is important
02:22 karolherbst: gordan: with a fb you can't do something else anyway
02:22 karolherbst: which is exactly the point why kde is not good for yours :D
02:22 gordan: As I said, KDE on AC100 with FF works just fine.
02:22 karolherbst: yeah, with a crappy kwin config
02:22 gordan: I tried XFCE at one point and it made no perceivable difference.
02:23 karolherbst: its like saying: yeah my gpu does run bioshock infite, just on lowest settings
02:23 karolherbst: I know what you mean
02:23 karolherbst: but having a composited window manager makes a difference
02:23 karolherbst: and I won't tell a user to use kwin, but just disable everything cause your gpu won't make it otherwise
02:23 gordan: I have no drivers for my GPU for soft-float distributions anyway (I maintain a soft-float ARM port of EL). :p
02:23 karolherbst: :D
02:24 Justice: EL?
02:24 gordan: But even the dumb FB has reasonable-ish basics covered, e.g. it can do video scaling. I can play videos on it just fine.
02:24 karolherbst: I would also try out kde on low end systems first just to see how they will drive
02:24 gordan: EL as in RHEL without the RH.
02:24 karolherbst: but tearing is really a bad thing for me
02:24 karolherbst: can't stand tearing
02:24 karolherbst: when windows are tearing, it's a nogo for me
02:24 Justice: hm but your CPU is better to take the load I guess
02:24 gordan: If you are sensitive to things like that, get a fat laptop.
02:25 gordan: Horses for courses.
02:25 karolherbst: yeah I have
02:25 Justice: this one is Athlon 64 X2
02:25 karolherbst: i7 + GTX 770M :D
02:25 Justice: pretty akward when my phone has 4times the memory of this laptop
02:25 gordan: I am typing this on an i7 with a 970M, but I mainly use that for the 4K screen as it makes having tons of terminals open for work more pleasant.
02:25 Justice: RAM*
02:26 karolherbst: Justice: well I would advise you to try out xfce and if you are disappointed with the effects you can also try out mate or KDE (if you want to have a full solution)
02:26 gordan: For when I want something light, I use my ARM laptops, fraction of the weight, multiple of the battery life.
02:26 Justice: heh effects :P
02:26 karolherbst: :D
02:26 karolherbst: yeah I know
02:26 Justice: Just want somewhat smooth desktop, not 2 minutes to open unity menu and have graphical corruption =)
02:26 karolherbst: the window shadow effect is pretty good though with xfwm
02:27 karolherbst: :D
02:27 karolherbst: yeah I know
02:27 karolherbst: I installed unity once (not on ubuntu)
02:27 karolherbst: it was pain
02:27 karolherbst: got also corruptions
02:27 karolherbst: sometimes
02:27 Justice: This laptop is for a child so as long as firefox and vlc plays somewhat smooth im good with it.
02:28 karolherbst: yeah then go with xfce
02:28 karolherbst: its easy, small
02:28 Justice: Running my desktop with 9590 + 970 on arch and kde5
02:28 gordan: I can highly recommend Chromebooks for that sort of thing. :)
02:28 karolherbst: maybe lxqt, but I never tried that
02:28 karolherbst: I would never use lxde anymore though :D
02:29 karolherbst: Justice: or install kde + xfwm
02:29 karolherbst: and just don't use kwin
02:29 karolherbst: but the only reasons for this would be to get all those applications and the desktop
02:30 karolherbst: xfwm doesn't even need OpenGL for compositing
02:30 karolherbst: which is I think a big advantage
02:30 Justice: hm
02:32 karolherbst: I think xfce is also the oldest compositor generally
02:34 karolherbst: mupuf: wanna take a look at this? https://github.com/karolherbst/nouveau/commit/6663eecc837a3840d5a07ee78945f05298f25ab1
02:36 karolherbst: I don't know if I also should check the fsrm div set in cfg_5, but we got a cut by 80% already even when the value is 0
02:41 karolherbst: gordan: ohh, do you want to try to reproduce an issue I have with plasma5? when the screen is locked and I (un)plug an external monitor, my plasma session sometimes goes boom
02:42 gordan: I don't use external monitors so don't have one handy.
02:42 gordan: Most likely RandR interaction issue.
02:42 karolherbst: yeah most likely
03:40 Justice: just got mate running
03:40 Justice: it runs pretty well
03:44 BanHammor: Hello people. Does anyone know what part of Nouveau was realized as xf86-video-nouveau
03:44 BanHammor: ?
03:50 karolherbst: BanHammor: the ddx
03:54 BanHammor: Just found xf86-video-modesetting. It finally makes sense.
03:55 BanHammor: thank you
04:31 karolherbst: weird, my fans can't even cool down my cpu beneath 50°C even when the package just consumes 2W :/
04:31 karolherbst: at full speed
04:42 karlmag: karolherbst: fan with thermal isolation? :-P
04:59 karolherbst: no clue
05:00 karolherbst: even when I manually speed up my fans, the temperature doesn't differe much
05:00 karolherbst: yeah like no change
05:00 Justice: btw is reclocking enabled by default now or?
05:01 karlmag: hence thermal isolation theory.. (ie. cooler is thermally isolated from what it's supposed to cool down)
05:02 karlmag: Have heard stories... Stuffz stops working for some (overheating) reason...
05:02 karlmag: Not really saying that's the thing for you though. :-)
05:12 karolherbst: it works okayish on higher temps/loads
05:12 karolherbst: Justice: not yet, we are working on it though :p
05:13 Justice: sweet maybe this 7000 will be able to run atleast some mobile games well :P
05:13 karolherbst: well
05:13 karolherbst: what do you want to reclock on this gpu?
05:14 Justice: the gpu?
05:15 karolherbst: yeah well, you can only reclock when you have different performance states
05:15 karolherbst: you can check in your vbios
05:15 karolherbst: but I don't think those old cards are reclocked in general
05:18 karolherbst: Justice: or can you switch the pstates with nouveau.pstate=1 booted?
05:21 Justice: havent tried
05:22 karolherbst: then you should :)
05:22 karolherbst: because when manually reclocking doesn't work, automatic reclocking won't either
07:34 Tom^: freeeedooooom, no more work until the 4th.
07:34 Tom^: karolherbst: still need those uh regs checked?
07:35 karolherbst: Tom^: yes
07:36 Tom^: https://gist.github.com/karolherbst/1854da8068b510d4cc14 so uh nvapeek 204c0 , etc ?
07:36 karolherbst: yes
07:36 Tom^: im learning \o/
07:36 karolherbst: all of them D:
07:36 karolherbst: :D
07:49 Tom^: karolherbst: https://gist.github.com/anonymous/c41562ea9a21ce094d0b
07:54 karolherbst: Tom^: I think i need the url again, my kernel just crashed
07:54 Tom^: =D
07:54 Tom^: https://gist.github.com/anonymous/c41562ea9a21ce094d0b
07:56 karolherbst: Tom^: the values are values, not regs :D
07:56 karolherbst: okay
07:57 karolherbst: so it is different for nvf1 cards
07:57 karolherbst: a big meh :O
07:57 karolherbst: mupuf: the fsrm stuff is different for Tom^s nvf1 card
07:57 Tom^: i was about to ask about it but meh just peeked it all :P
07:57 Tom^: or am i using to old envytools?
07:58 karolherbst: no
07:58 Tom^: oki
08:09 karolherbst: Tom^: its not good if that what you expect is not helpful :D
08:09 Tom^: :<
08:24 imirkin: gnurou: well, i combined the fact that it didn't seem to work on GK208 or GK106 with the fact that it was unlikely that you were maliciously lying -- that pointed to the formats only being on GK20A. also they're mobile-focused formats, and GK20A is mobile-focused...
08:26 imirkin: gnurou: all this has caused me to resuscitate my TK1 -- booted it last night, but into an older kernel and the older firmware was gone. rebuilt the new one which loads the diff fw but didn't have time to test anything.
08:44 mlankhorst: well, my radeon card died on me
08:44 imirkin: back to nvidia? :)
08:44 mlankhorst: nah, its a macbook
08:44 mlankhorst: modprobing causes a panic or something
08:44 imirkin: ouch
08:44 mlankhorst: but it boots with efifb on radeon
08:45 mlankhorst: but as soon as I modprobe radeon it crashes
08:45 imirkin: i think you want #radeon :)
08:45 mlankhorst: shrug, older kernels that used to work crash now
08:46 mlankhorst: did a hack in apple-gmux to disable radeon on probing, then modprobe i915 after so it detects the internal display
08:46 mlankhorst: battery live improved so its not all bad :)
08:47 imirkin: hehehe
08:50 karolherbst: mlankhorst: you are so lucky, that it is easy now :p
08:51 mlankhorst: try running links on a macbook, the orientation sensor or w/e is used as mouse cursor
08:51 mlankhorst: that one was fun to find out
08:54 imirkin: gpm can't be taught to look at other devices?
08:54 imirkin: you could also use lynx...
08:55 mlankhorst: i just rmmod joydev
09:26 josla972: RSpliet: I looked at the patches you submitted in september: "[Nouveau] All-round reclocking improvements". Could the reclocking part work if enabled for G84 by changing 0x94 -> 0x84 as per the last patch or would such an attempt only be stupid and riskful?
09:26 imirkin: josla972: what kind of vram do you have?
09:26 imirkin: ddr2 or gddr3?
09:27 josla972: [ 0.647364] nouveau 0000:01:00.0: fb: 256 MiB GDDR3
09:27 josla972: [ 0.697996] nouveau 0000:01:00.0: DRM: VRAM: 256 MiB
09:27 josla972: according to dmesg
09:27 imirkin: well, there's a chance it could work, but unlikely
09:27 imirkin: (if it were ddr3, then it def wouldn't work)
09:27 josla972: imirkin: ok
09:28 josla972: imirkin: risky?
09:28 imirkin: shouldn't be dangerous though... it'll just hang your box if it fails
09:28 josla972: famous last words
09:28 josla972: i'll try it
09:28 imirkin: i'm assuming you have a way of rebooting the box, i.e. this isn't some idiotic mac situation where you have to wait for the battery to drain :)
09:29 josla972: no no just a regular hp laptop
09:29 josla972: i will monitor the temperature with gkrellm or something as well
09:29 josla972: (should the reclocking work)
09:29 imirkin: most of the time there's not a dedicated fan on laptops
09:29 imirkin: i.e. it'll just be your system firmware that decides on fanspeed
09:30 josla972: that is my impression too of this laptop
09:31 josla972: only one fan that sucks air and blows it over the heatpipes
09:31 josla972: ok kernel is built
09:31 josla972: brb :)
09:31 imirkin: if you don't come back, we'll know why :)
09:32 josla972: imirkin: lol
09:36 josla972: imirkin_: one setting hanged the system, two did not, and one did not work
09:37 josla972: imirkin_: I think I managed to set the most extreme setting though.
09:37 josla972: ~ # cat /sys/class/drm/card0/device/pstate
09:37 josla972: 20: core 169 MHz shader 338 MHz memory 100 MHz
09:37 josla972: 21: core 275 MHz shader 550 MHz memory 301 MHz
09:37 josla972: 22: core 475 MHz shader 950 MHz memory 702 MHz AC DC *
09:37 josla972: AC: core 475 MHz shader 950 MHz memory 280 MHz
09:38 josla972: setting 20 and 22 works, 21 hanged the system, and AC does not work.
09:52 imirkin_: josla972: AC = current (well, when plugged into AC)
09:52 imirkin_: josla972: note that 22 didn't quite work though
09:52 imirkin_: the core reclocked, but memory stayed put
09:52 josla972: imirkin_: Hmm I do not think any option really worked though. I noticed this output from dmesg nouveau 0000:01:00.0: fb: invalid/missing rammap entry
09:53 imirkin_: josla972: put up your vbios somewhere, maybe RSpliet will have a chance to look at it
09:53 imirkin_: vbios available from /sys/kernel/debug/dri/0/vbios.rom
09:54 imirkin_: he might also be able to provide mmiotrace instructions to get the necessary info
10:00 josla972: imirkin_: I am sorry, I had to fix some file system errors as a result of my system hanging :P
10:00 josla972: imirkin_: you said something about vrom
10:00 imirkin_: http://people.freedesktop.org/~cbrill/dri-log/?channel=nouveau&date=2015-12-18
10:04 josla972: I dumped it at http://79.136.89.252/vbios.rom
10:09 josla972: imirkin_: thanks for the guidance
10:20 imirkin_: josla972: sounds like your core/shader clocks change ok... iirc you can boot with nouveau.config=NvMemExec=0 to disable the actual memory reclocking piece
10:21 imirkin_: but obviously the memory clock is pretty important to performance :)
10:39 josla972: imirkin_: I'll try that and see if things are stable
10:40 imirkin_: what kernel are you on?
10:40 imirkin_: actually doesn't matter -- looks like that's still around in the latest too
10:40 josla972: 4.3.3 with all the 9 patches submitted by rspliet
10:40 josla972: (I applied them manually)
10:41 imirkin_: ah ok. i think they should be in 4.4-rc5 as well
10:43 josla972: imirkin_: yeah probably, I just picked the latest unstable kernel available in the gentoo portage system since I do not know if 4.4-rc5 would run ok.
10:43 imirkin_: no worries
10:43 josla972: as long as I am not missing out on other dependencies it should be fine I guess
10:43 imirkin_: not a lot of changes in 4.4 which would affect your situation... conceivable that roy's patches were massaged a bit though, dunno
10:44 josla972: I actually picked them with git format-patch :)
10:44 imirkin_: right, but either roy or ben might have futzed with them without resending to list
10:44 imirkin_: iow i'm not 100% sure that what's on-list is what ended up in 4.4-rc
10:44 imirkin_: but... it very well could have.
10:45 josla972: yes but I picked the patches from the git master branch
10:45 imirkin_: ah ;)
10:45 josla972: seemed easier than downloading the patches from the mailing archive
10:46 imirkin_: sure
10:47 imirkin_: mwk: i'm inclined to just merge gnurou's new TIC stuff and deal with any fallout later. what do you think?
10:47 josla972: anyway, I'll try a reboot with your last suggested option and see what it does when changing pstate
10:47 josla972: brb
10:47 imirkin_: josla972: should be that the core clock changes, but mem stays put
10:48 josla972: imirkin_: how I will I know if I succeed? Just by outputting the contents of pstate?
10:48 imirkin_: josla972: look at the AC line of pstate
10:48 imirkin_: that indicates the current state
10:48 josla972: imirkin_: ah!
10:48 josla972: :)
10:48 imirkin_: josla972: also shader-limited things should go faster
10:49 imirkin_: (probably *not* glxgears)
10:49 josla972: I'll fire up dota 2 and see if it runs equally badly
10:49 josla972: lol
10:49 imirkin_: hehe
10:49 josla972: brb
10:49 imirkin_: you might want the non-reborn version
10:50 imirkin_: i think reborn wants more than 256MB of vram
10:50 josla972: :(
10:54 josef_: imirkin_: I still got the dmesg output: nouveau 0000:01:00.0: fb: invalid/missing rammap entry
10:54 imirkin_: hmmm right
10:54 josef_: imirkin_: core and shader are reclocked though
10:54 imirkin_: did it change the shader clock though?
10:54 imirkin_: k :)
11:04 josef_: I think performance is somewhat better
11:04 imirkin_: placebo effect? :)
11:04 josef_: but I got some errors after fiddling around with graphic settings
11:04 josef_: could be
11:04 josef_: [ 759.697189] nouveau 0000:01:00.0: gr: 00100000 [] ch 9 [000f2e4000 dota2[3355]] subc 3 class 8297 mthd 1510 data 00000001
11:05 josef_: [ 759.894389] nouveau 0000:01:00.0: dota2[3355]: nv50cal_space: -16
11:05 imirkin_: that means either gpu hung, or you're submitting commands a lot faster than it can process them
11:05 josef_: the only sensible way I can think of comparing before and after is to measure fps
11:05 josef_: ok
11:05 imirkin_: perhaps you got that on the lowest perf level?
11:06 josef_: yeah I have pretty much the lowest settings
11:06 imirkin_: btw, i've made a number of modest improvements to nv50 lately, i'd def recommend trying mesa-git and seeing if it's any faster
11:06 josef_: not merged into linux yet?
11:07 imirkin_: mesa is the piece of software that provides libGL
11:07 imirkin_: and it is all pushed out... just not in a release
11:07 josef_: ah, I am running mesa-11.0.6
11:08 josef_: the latest "unstable" mesa available in gentoo portage is 11.1.0 (but one could run even more experimental versions of cours
11:08 imirkin_: there should be a 9999 version, no?
11:08 josef_: exactly
11:09 imirkin_: which iirc pulls from git
11:09 josef_: it probably does :)
11:09 josef_: I can try that and see if it improves things
11:09 imirkin_: anyways, i doubt it'd be really noticeable
11:09 imirkin_: just a bit faster
11:09 imirkin_: like... a few percent at most
11:09 imirkin_: not 2x :)
11:09 josef_: worst case scenario is that it pulls in a lot of boring dependencies or doesn't work
11:09 josef_: ok
11:10 imirkin_: oh, and we precompile all shaders now, so hopefully less stuttering
11:10 josef_: I suppose memory reclocking would improve things much more
11:10 josef_: that llvm stuff?
11:10 imirkin_: nouveau doesn't use llvm
11:10 josef_: ah, only radeon does?
11:10 imirkin_: and llvmpipe (surprise surprise)
11:10 josef_: I read briefly about this stuff
11:11 imirkin_: llvmpipe uses llvm as a JIT
11:11 imirkin_: to compile shader code
11:11 imirkin_: into (optimized) CPU instructions
11:12 imirkin_: it's unbearably slow for any actual game, but performs fairly well for regular desktop use
11:12 josef_: I see
11:12 imirkin_: oh, and be sure to let me know if you see any misrendering
11:12 imirkin_: i don't actually play many games, so tend not to notice issues
11:12 josef_: imirkin_: sure
11:13 josef_: I do not know very much about GPUs
11:13 imirkin_: don't need to notice anything to notice that something's clearly not looking the way it ought to :)
11:13 imirkin_: s/notice/know/
11:13 josef_: nah, just interesting
11:14 josef_: I have a fairly good picture of how CPUs work, but not GPUs
11:14 imirkin_: well, if you're interested in learning more by contributing to nouveau, i could get you started :)
11:15 josef_: do you ah[ 759.697189] nouveau 0000:01:00.0: gr: 00100000 [] ch 9 [000f2e4000 dota2[3355]] subc 3 class 8297 mthd 1510 data 00000001any good material to read?
11:15 josef_: [ 759.894389] nouveau 0000:01:00.0: dota2[3355]: nv50cal_space: -16
11:15 josef_: sorry
11:15 josef_: do you have any good meterial to read?
11:15 imirkin_: wrong key? :)
11:15 josef_: I think so
11:15 imirkin_: mmmm... not REALLY. but i tend not to be a "let me read a book" kind of guy...
11:15 imirkin_: more of a "let me try to do this" style learner
11:16 josef_: I suppose I am more that kind of guy too
11:16 imirkin_: there's http://envytools.readthedocs.org/en/latest/ but it'll be VERY difficult to understand
11:16 josef_: learning by doing
11:16 josef_: ok
11:16 imirkin_: i still don't get like 50% of it :)
11:17 imirkin_: and remember being totally mystified when i first looked at it
11:17 imirkin_: do you know compilers? or alternatively, h264?
11:18 josef_: I know how to compile and program
11:18 josef_: h264 is just a codec to me
11:18 imirkin_: how about programming compilers? :)
11:18 josef_: I have never done that
11:19 imirkin_: is that something you're interested in?
11:19 josef_: it could be interesting to try out
11:19 imirkin_: do you know any GL?
11:19 josef_: nope
11:19 imirkin_: k
11:19 imirkin_: so... you're basically where i was 2 years ago.
11:20 imirkin_: there's a moderately simple task available herE:
11:20 imirkin_: https://trello.com/c/8F6Obtdu/115-evicting-shader-code-does-not-cause-currently-bound-shaders-to-be-reuploaded
11:20 imirkin_: i've purposefully not done it because it's not that big of a deal, and it's a good starter task
11:21 imirkin_: the basic idea is that there's a code segment
11:21 imirkin_: where we store shader code
11:21 imirkin_: but every so often we've uploaded too many shaders and there's no more room
11:21 imirkin_: we could be all clever and use some sort of LRU dealie
11:21 imirkin_: but instead we just say "fuck it" and evict everything
11:21 imirkin_: and then upload the shader we were trying to upload
11:22 imirkin_: HOWEVER
11:22 imirkin_: a single pipeline is composed of multiple stages
11:22 imirkin_: and so if we've uploaded the first 3 stages
11:22 imirkin_: and then evict everything
11:22 imirkin_: and upload the next 2 stages
11:22 imirkin_: then... the first 3 stages are gone
11:22 imirkin_: oops :)
11:22 josef_: hmm
11:22 imirkin_: it's reasonable to assume that the whole pipeline can fit
11:23 imirkin_: so we just have to go back and reupload the current pipeline on eviction
11:23 imirkin_: the relevant logic is here: http://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv50/nv50_program.c#n430
11:24 josef_: I'll have a look at it and see if I can learn something
11:24 imirkin_: sgtm. and feel free to ask questions
11:24 josef_: ok :)
11:26 imirkin_: i haven't *fully* thought about this, but i think the simplest thing is to have something in nv50_state_validate after going through the dirty state that forces the upload of all programs
11:59 josef_: imirkin_: hmm, another thing I notice is, once the eviction is done, one tries the heap alloc again, and if it fails the shader is considered too large... but is it not possible for the shader to be too large in the first place?
11:59 josef_: (thus causing unnecessary eviction)
11:59 imirkin_: i mean... impossible is a strong word
12:00 imirkin_: but in practice, it's highly unlikely
12:00 josef_: I am a total newbie here, so bear with me
12:00 josef_: :)
12:00 imirkin_: error handling is generally sorta fubar in nouveau too
12:00 imirkin_: it's not your task to fix it
12:34 mwk: imirkin_: you have my blessing
12:37 imirkin_: ok
12:37 imirkin_: i'm going to do it when i have time to deal with the fallout
12:37 imirkin_: probably over the weekend
12:38 imirkin_: and i'll take the opportunity to teach demmt about TIC/TSC on kepler+
12:38 imirkin_: mwk: btw, how do 32-bit gmem addresses work on kepler?
12:38 imirkin_: [or fermi for that matter]
12:41 karolherbst: ...
12:41 karolherbst: slowly I begin to get annoyed by nvidia changing stuff the entire time :D
12:42 karolherbst: now the question why that doesn't work on my kepler card
12:53 josef_: imirkin_: nv50_state_validate?
12:54 imirkin_: josef_: yeah, in nv50_state_validate.c
12:54 imirkin_: all the way at the bottom
12:54 josef_: ok
12:58 josef_: hmm no such file in v4.4-rc5?
12:58 imirkin_: mesa
12:59 mwk: imirkin_: uh... I think they're just zero-extended?
12:59 mwk: doesn't sound like rocket science
13:00 imirkin_: mwk: yeah, you're probably right
14:25 imirkin_: hakzsam: btw, i think we can just do "and b16 $c0 $r0l $r0l $r0l" for that first opcode
14:34 imirkin_: hakzsam: and something like st b128 g15[$r5] $r0q
14:34 imirkin_: oh, but it might not be aligned... doh
14:34 imirkin_: can we align it?
14:52 imirkin_: hakzsam: why not make the stride 0x20?
15:00 etnk: hello, i've got a machine (hp touchsmart iq775 w/ geforce go 7600) that has a dual-link display and when using nouveau i get image corruption, the freedesktop faq states i should come here for help/advice. the proprietary nvidia driver doesn't have this prob, but it's really outdated for this card and pretty buggy
15:04 imirkin_: etnk: can you describe the image corruption?
15:05 etnk: yeah looks almost like inverted colors, a light thru the screen (vert not horizontal) and like almost snow flickering
15:05 imirkin_: snow tends to indicate underclocking
15:05 imirkin_: can you pastebin your dmesg?
15:05 etnk: sure
15:05 etnk: i've added drm.debug=14 and log_buf_len=10M to my kernel
15:06 imirkin_: (i.e. memory can't keep up with scanout engine, or vice versa)
15:08 etnk: http://pastebin.com/wLnFdgQu
15:08 imirkin_: that.... did not have what i was expecting
15:09 etnk: is it due to the kernel lines i added?
15:09 imirkin_: skeggsb: did you kill the clk print in 4.3?
15:09 imirkin_: ah yeah, looks like it's gone
15:09 imirkin_: etnk: no, not your fault... mine =/
15:10 imirkin_: etnk: can you boot with nouveau.pstate=1
15:10 imirkin_: etnk: and pastebin /sys/class/drm/card0/device/pstate
15:10 etnk: sure, i'm on the machine right now so lemme just cut and paste this and get back in a moment
15:15 etnk: http://pastebin.com/AGXq6bbt
15:15 etnk: there you go
15:16 imirkin_: can you try to, as root, echo 22 > /sys/class/.../pstate ?
15:17 etnk: sure
15:18 imirkin_: does the AC line change after you do that?
15:18 imirkin_: oops, apparently not
15:19 etnk: yeah that hard locked the machine
15:19 imirkin_: i guess that didn't work out so hot?
15:19 imirkin_: :(
15:19 etnk: ha no
15:20 etnk: i should be on another machine for chat..
15:20 imirkin_: bleh, so... i thought reclocking for nv4x used to basically work :(
15:21 etnk: screen started to fade white and the boot screen was kinda flickery when it rebooted, seems ok now tho
15:21 etnk: i'd like to not kill the display if possible, but i wouldn't be like totally heartbroken if that happened
15:21 imirkin_: that's extremely unlikely to happen
15:22 imirkin_: hangs, on the other hand... definitely possible, as you've witnessed
15:22 etnk: oh ok cool, i have little understanding of driving display hardware etc
15:23 karolherbst: imirkin_: "Mesa: User error: GL_INVALID_ENUM in glTexImage2DMultisample(internalformat=GL_RGB9_E5)" any reason to take care of that?
15:23 imirkin_: karolherbst: that's wine doing stupid shit iirc
15:23 karolherbst: well
15:24 karolherbst: I get that with bioshock :D
15:24 imirkin_: o
15:24 karolherbst: but that's also not "real" native
15:24 karolherbst: eon I think
15:24 imirkin_: hmmmm
15:24 imirkin_: actually RGB9_E5 should be legal
15:24 glennk: as a render target in GL? don't think so
15:25 imirkin_: really?
15:25 glennk: texture format only afaik
15:25 imirkin_: ah yeah. def texture-only with nouveau
15:25 imirkin_: irrespective of anything GL might say :)
15:25 glennk: radeon i think supports it as a RT
15:26 glennk: but it also allows weird stuff like bptc as vertex data
15:26 imirkin_: looking at GL 4.5, rgb9_e5 is not required for rendering
15:26 imirkin_: so what nouveau/mesa is doing is fine
15:27 imirkin_: and what bioshock is doing is fine
15:27 imirkin_: karolherbst: so ignore the warning and move on
15:27 glennk: wonder if its a typo and they meant the other 3 component 32 bit float format
15:28 imirkin_: etnk: unfortunately i'm not entirely sure what to say about your issues... you could try an older kernel and see if we broke reclocking somehow
15:28 glennk:looks up GL_R11F_G11F_B10F
15:28 imirkin_: etnk: or the issue could be entirely unrelated :(
15:28 imirkin_: glennk: yeah, r11g11b10 is required for rendering
15:28 glennk: don't remember if it supported texturing
15:28 imirkin_: well, it's required
15:28 imirkin_: according to GL 4.5
15:29 imirkin_: and i suspect according to GL_EXT_packed_float
15:29 etnk: what version would you recommend? i'm on 4.3 now, 4.2 was same
15:29 imirkin_: etnk: well, there was a big rewrite for 4.3, so if 4.2 worked...
15:29 imirkin_: er
15:30 imirkin_: if 4.2 *didn't* work
15:30 etnk: maybe i should file a bug report with a pic of screen
15:30 etnk: do you think that should be my next course of action?
15:32 imirkin_: etnk: pic of the screen could be good
15:32 imirkin_: bugs.freedesktop.org xorg -> Driver/nouveau
15:32 imirkin_: but this is the sort of thing that's difficult to debug
15:32 imirkin_: since it requires very specific hw to trigger
15:33 imirkin_: and i can't remember last time i tried reclocking a nv4x...
15:33 imirkin_: i've def done it, but it was with desktop gpu's, so... more tested
15:34 etnk: i understand, i appreciate any help and am willing to do whatever i can for others who may have the same hardware/issue
15:34 etnk: thank you for your time on this
15:35 imirkin_: if you have the blob set up, you could try to get an mmiotrace of what it's doing
15:35 imirkin_: this is a general guide: https://wiki.ubuntu.com/X/MMIOTracing
15:36 etnk: yeah that would be no prob
15:36 etnk: tho i prob should do that in a bug report and not over irc here, would prob be more efficent?
15:37 imirkin_: the mmiotrace? yeah. if it's too big, send it to mmio.dumps@gmail.com
15:38 imirkin_: it should be on the order of 10-100MB uncompressed
15:38 imirkin_: [please compress before sending it. xz -9 works quite well]
15:41 etnk: gonna reboot w/nvidia now
15:43 etnk: ahh my eyes, they don't burn!
15:44 etnk: (screen corruption gone on nvidia, will get on looking up how to mmiotrace)
15:51 imirkin_: etnk: another thing to check if whether the issues go away at lower resolutions
15:51 imirkin_: (should have thought of that up front...)
15:51 etnk: for the mmiotrace, so i assume i should do the steps FOR RESOLUTION / DISPLAY BUGS, that is xinit "sleep 10"
15:51 imirkin_: yes
15:52 etnk: i don't remember if i tried lower res' on nouveau
15:52 etnk: so i'll def do again
15:53 etnk: does it matter if i use xrandr for res change or does it have to be a kernel line
15:53 imirkin_: doesn't matter
15:53 imirkin_: btw... another stupid question...
15:53 imirkin_: you're not using something like gnome-shell are you?
15:54 etnk: openbox atm, gnome-shell over x is same tho
15:54 imirkin_: well, something that's not a GL compositor would be best with that hw :)
15:56 etnk: i'm hoping to set this machine up as kodi a media center for my family, would really love to a. use nouveau over nvidia driver as its a bit buggy b. help others who might be stuck w/same hardware
15:56 imirkin_: errrr
15:57 imirkin_: it doesn't have hw h264 (or mpeg4) decoding
15:57 imirkin_: and the cpu inside is probably not too speedy
15:57 imirkin_: and while you can use xvmc for mpeg2 decoding accel, i think kodi has likely dropped that by now
15:57 imirkin_: (works fine with mplayer though)
15:57 etnk: yeah i no, but i'm hoping for the best, and won't be too disappoint if it don't work out
15:58 etnk: if it can play/rip dvds and take svid in alright i'll be happy
15:59 imirkin_: yeah... dvd decoding should work fine. you can use the accel or just say screw it and decode on cpu
15:59 imirkin_: the accel becomes more useful for ATSC/DVB streams
15:59 imirkin_: where the cpu's of the day couldn't keep up but the hw could
16:01 etnk: for me that'd just be a plus if it works, that'd only work for ota chans as i have cable am in the us where all the providers encrypt chans now
16:02 etnk: anyway i need to figure out how to boot w/o x on arch for the mmiotrace, i haven't done that since pre systemd
16:02 etnk: looking that up..
16:11 karlmag: Hmmm.... GTS 250... nv92?
16:12 karlmag: I guess the older stuff I find is really a bit older than I might hope.. Oh well :-P
16:12 imirkin_: karlmag: what's the question?
16:13 karlmag: not really sure there was one. I guess "usefulness" might be one though.
16:13 imirkin_: usefulness to...
16:13 imirkin_: getting reclocking working well on G92 would be nice
16:14 imirkin_: they were pretty beefy cards
16:14 karlmag: in general and to this project I suppose.
16:18 karlmag: Have a christmas to get through and a bit of act to get together, but hope to eventually "set up shop" so that I (sanely) can help do some testing at the very least.
16:21 karlmag: I guess having a range of cards might be helpful then.
16:41 imirkin: karlmag: yep
16:50 etnk: imirkin_: if you're still around, i couldn't get x to run via nvidia with mmiotrace, it hard locks the machine and i'm left with a log with only 33 lines, figure there should prob be more.. when i enable mmiotrace i did get some message about irq not being avail, i can't remember, trying to look thru journalctl for the msg but not sure if it ever got logged
16:51 imirkin: hmmmm weird
16:51 imirkin: you did start out without nvidia loaded right?
16:52 etnk: yes
16:53 etnk: tried several time, if i didn't enable the tracer x would start fine, with it hard locks with a cursor on the screen (seems to change the screen mode iirc)
16:53 imirkin: if you're willing to forego all acceleration, you could try using xf86-video-nv
16:56 etnk: well i'm still willing to be a monkey for you and other nouveau devs if you guys are interested in working on this with me, not now i've done enough for tonight i think but i'll def file a bug report (what do you think?)
16:56 imirkin: go for it
16:57 etnk: figure i'm capable of doing this kind of thing where other ppl with the hardware trying to run linux might get discouraged
16:57 etnk: anyway, thank you so much for your time on this
16:58 etnk: i'll try to post a bug report with some info in a few days, though imma try to swith res real quick before i go
16:58 imirkin: what resolution were you running again?
17:00 etnk: 1440x900
17:00 imirkin: errrr that should def have been fine
17:01 imirkin: as for discouraging people from trying linux... not a lot we can do
17:01 imirkin: there's probably 200-500x more manpower that goes into the nvidia proprietary driver, and they have documentation
17:02 imirkin: (not to mention hw access)
17:03 etnk: yeah i remember when amd merged with ati way back when, such a shock as back then it was always nvidia+amd that ppl prefered for linux
17:04 etnk: i dunno if true but iirc they approached nvidia first but due to company/ceo politics and/or some deal with microsoft
17:08 etnk: ok just checked, switching res does nothing, anyway will file a bug report
17:08 etnk: ty for your time
17:10 imirkin: gnurou: btw, i see [ 12.257617] nouveau 57000000.gpu: DRM: failed to create ce channel, -22 with kernel 4.4-rc5, otherwise seemingly-loaded firmware
17:11 imirkin: aaaan there's some kind of watchdog
18:03 sarnex: dumb question, does this mean the card isnt supported by nouveau at all? [ 12.889] (EE) Unknown chipset: NV124
18:03 sarnex: have a user having trouble with gallium nine, turns out hes using llvmpipe
18:33 imirkin: sarnex: GM20x doesn't have any accel with nouveau for now
18:34 imirkin: that's most of the GTX 9xx series
18:35 sarnex: imirkin: thanks alot