01:21pmoreau: Happy new year everyone!
01:21pmoreau: imirkin: Do you need testing for your latest patch on nv50?
01:24mupuf: pmoreau: happy new year!
01:24pmoreau: mupuf: Thanks! You too! I can't recall, are you already in Helsinki?
01:25mupuf: not yet, I'm at the airport of Montpellier, waiting for my flight
01:25pmoreau: So I was a few hours too early
01:26mupuf: yep, not bad ;)
01:30imirkin: pmoreau: yes, that'd be awesome... i've sent an updated version though
01:31imirkin: i have an updated version
01:31imirkin: but i haven't sent it
01:31imirkin: i was previously not copying modifiers
01:31pmoreau: Ok, I'll test it after breakfast
01:32imirkin: but it'd be a lot more useful to get a baseline test with latest git
01:32pmoreau: Test first with latest git, then with your patch applied?
01:32imirkin: i've updated the ones on my fdo page
01:32imirkin: yes, and send me both results
01:33pmoreau: Was it for a particular command?
01:33imirkin: see http://people.freedesktop.org/~imirkin/ for how to run piglit so that it avoids the more crashy tests
01:33pmoreau: Will do
01:34imirkin: also, if you're looking to get into some nv50 3d stuff, i just added an item in trello that should be moderately simple: https://trello.com/c/mlh4Cj5B/95-nv50-try-enabling-compression-again
02:15pmoreau: imirkin: For now I'm more into adding reclocking support for G80:GT200
02:15leilei: Hello everyone. I have an old laptop with GT240M(Family NV50, Chipset GT216 NVA5). Recently I switched form the proprietary driver to nouveau. Since then, the fan always runs at a high speed which is loud. What can I do?
02:18imirkin: leilei: pastebin dmesg
02:21imirkin: doesn't appear that there's a fan attached to the card. chances are it's just running hotter because there's no automatic reclocking
02:21imirkin: (it's common in laptops to have a shared cooling solution)
02:22imirkin: if you update to 3.19-rc2 then you should be able to reclock to the lowest level and hopefully reduce the temp/fan speed
02:22leilei: Thanks. I will have a try.
02:23imirkin: make sure to boot with nouveau.pstate=1
02:23imirkin: which should enable the pstate file in sysfs (/sys/class/drm/card0/device/pstate)
02:24leilei: OK, but I have to compile the mainline kernel first.
02:24imirkin: this will allow you (hopefully) to control the performance level of the card (the levels are listed in that file)
02:25imirkin: yeah. the patches for reclocking nva3/a5/a8 were only upstreamed for 3.19
03:13imirkin: pmoreau: if you have a nva0+ tesla available, would be nice to see result of https://github.com/imirkin/mesa/commit/bcb71ed0dc104cb245147ff210ee21685cb6fbec against tf2 and tf3 piglits
03:22pmoreau: imirkin: I could try on my nvac.
03:23pmoreau: What do you mean by tf2/tf3? I guess it's not team fortress, is it? :)
03:26imirkin: pmoreau: no, the ARB_transform_feedback2/3 piglits
04:52RSpliet: imirkin: I see you already pushed the patch
04:52RSpliet: sorry, I was too busy trying to get drunk ;-)
04:52RSpliet: (I failed :-( )
04:56Yoshimo: nothing wrong with a clear mind
05:00RSpliet: Yoshimo: everything with moderation
05:38pmoreau: imirkin: They are a few links missing on your page for NV50 results: all, skipped, changes and enabled.
06:34leilei: imirkin: I've compiled the mainline kernel and set the pstate, but I don't get a quiter fan. Maybe I should go back to the proprietary driver and make sure there's nothing wrong with my laptop.
08:17mew007: Happy New Year!
08:31mew007: Happy New Year!!!
08:33tobijk: thx, happy new year for you as well! :)
08:36mew007: how do I find Martin Peres?
08:40tobijk: his nickname
08:43pmoreau: mew007: He's currently moving to Helsinki, he should be back in a few hours (if he has internet).
08:44pmoreau: (And happy new year to you too :) )
08:45mew007: pmoreau: moving? He is changing his residence or just it's just a trip?
08:45pmoreau: imirkin: I have piglit results with latest git for NV96 and NVAC, sending them to you and retesting with b0e44e282206b15ac3469e66d95fbcaa1b40730b afterwards
08:45pmoreau: mew007: Changing residence, at least for 1/2 or 1 year
08:47mew007: I'm thinking about making a gift to him. I have Gigabyte Geforce GTX550Ti... And nouveau doesn't work with it at all
08:48pmoreau: Oh right! This bug
08:51Yoshimo: mew007: RSpliet could also need a Fermi card for reclocking
08:51mew007: Yoshimo: unfortunately I have only 550Ti
08:51mew007: and I'm still thinking about gifting or not =D
08:52Yoshimo: yea but if one of them doesn't want it, there is now an alternative ;)
08:52mew007: where do RSpliet live?
08:54pmoreau: In the Netherlands
08:54mew007: ohhh, too far from me...
08:55mew007: may I exchange my 550Ti with, say, GT320?
08:56pmoreau: What do you mean by "exchange"?
08:57mew007: I send you my 550Ti, you test it (does it work or not), then you send me working GT320
08:59mew007: funny idea? :)
08:59pmoreau: I don't have a GT 320 ;)
08:59pmoreau: Nor any NVA[3/5/8]
09:00mew007: I mean, maybe, developers are interested in such 'exchange'. Coz ifaik GT320 works perfectly with nouveau, it's old. Why don't you exchange 320 with 550?
09:01pmoreau: I see
09:01mew007: that 'exchange system' gonna work better than 'non-refundable' donations, imo
09:02pmoreau: Probably, let's see if other developers are interested. :)
09:02pmoreau: You could send a mail to the mailing list about it
09:02mew007: but, it's not really 'non-refundable', coz u'll get nouveau working with it
09:03mew007: pmoreau: may u ask Martin Peres about it when he come here?
09:04mew007: if I'm not here ofc
09:04pmoreau: Sure, I'll ask him
09:04mew007: ok, thx bro
09:04pmoreau: You're welcome
09:44imirkin: pmoreau: did you get a chance to do the follow-up tests?
09:47imirkin: pmoreau: gah!!! did bin/fs-textureOffset-2D -auto -fbo really fail on the nv96??
09:57imirkin: pmoreau: i don't like your nv96 results
09:57imirkin: all the texel fetch stuff seems retarded :(
10:00imirkin: on the nvac as well
10:00imirkin: RSpliet: can you run the above textureOffset test and see if i've messed it all up on nv50, or if it's just pmoreau?
10:01tobijk: any nv50 will do?
10:01imirkin: just not pmoreau's
10:05tobijk: (may take a while)
10:11imirkin: pmoreau: btw, the index/etc pages are missing coz they add no value and are huge and i'm trying to be respectful of fdo storage space
10:39tobijk: imirkin: a special branch to test? mesa/master is fine (pass)
10:40imirkin: tobijk: ok, thanks for confirming that i'm not a total jackass
10:40imirkin: tobijk: which gpu is that with btw?
10:41tobijk: my old G86
10:42imirkin: i have seen that before though... offset magically stops working
10:42imirkin: no idea how or why
10:43imirkin: perhaps some pgraph setup bit is missing :(
10:44tobijk: is the fail on the macbook? :/
10:44imirkin: yeah, but iirc i've randomly seen it on other people's stuff too
10:49tobijk: a failing nve7 :O
10:49imirkin: no, don't think so
10:49imirkin: i meant like on nv50 family cards
10:50imirkin: tobijk: if it's not too much trouble, mind doing a piglit run against master on your G86?
10:50tobijk: yeah i jumped onto the buglist, sorry for not making it clear
10:50tobijk: will do
10:51imirkin: tobijk: please update piglit too (and rebuild)
10:52tobijk: have done that already :)
10:52imirkin: ah ok. then just do a piglit run with the command at http://people.freedesktop.org/~imirkin/
10:55tobijk: my standard piglit settings :)
10:59imirkin: tobijk: btw, if you want to mess around with nv50, this may not be a bad project: https://trello.com/c/mlh4Cj5B/95-nv50-try-enabling-compression-again
11:32mew007: I'm back. News?
11:34mew007: pmoreau: hey
11:40mew007: is Martin Peres back?
11:40imirkin: mupuf: --^
12:22pmoreau: imirkin: The patch introduced one regression on my MCP79, but not on the G96
12:22imirkin: pmoreau: which regression (and which patch? heh)
12:22pmoreau: Patch: commit/b0e44e282206b15ac3469e66d95fbcaa1b40730b
12:23imirkin: oh. my MAD folding oen
12:23pmoreau: Regression on: spec/glsl-1.30/execution/texelFetch fs sampler3D 98x1x9-98x129x9
12:23imirkin: your texelFetch's are all messed up
12:23pmoreau: Changed from pass to dmesg-warn
12:23imirkin: what was the warning?
12:23pmoreau: No idea :D
12:23imirkin: look in the results file
12:23pmoreau: I'm sending you the results
12:23imirkin: no, just read the results file yourself
12:24imirkin: search for that texelFetch thing
12:24imirkin: should be easy enough
12:24imirkin: my guess is that it was an unrelated message in dmesg
12:25pmoreau: Right, seems unrelated: "[ 4765.076308] [sched_delayed] \u00014CE: hpet increased min_delta_ns to 11521 nsec"
12:26pmoreau: I'll test the other patch now, about arb_transform_feedback2/3
12:28pmoreau: That patch was only to test on the MCP79, not on the G96, right?
12:36imirkin: the g96 doesn't have the relevant functionality
12:36imirkin: pmoreau: but you only need to do the ARB_tf2 tests...
12:36imirkin: specifically these 2:
12:37imirkin: bin/arb_transform_feedback2-draw-auto -auto
12:37imirkin: bin/arb_transform_feedback2-draw-auto instanced -auto
12:37imirkin: i think the rest won't be affected by my change
12:39tobijk: imirkin: plain results or better the summary for the full piglit run?
12:39imirkin: tobijk: mail me the resulting json (xz -9'd)
12:41imirkin: cool thanks
12:41imirkin: and by nv50 i assume you meant nv86 right?
12:42tobijk: yeah it just determines the driver
12:43pmoreau: imirkin: Both failed :/
12:43imirkin: annoying things in dmesg too, or just failed?
12:44pmoreau: Just failed
12:44pmoreau: Color observed is always (0.2, 0.2, 0.2), no matter what
12:44imirkin: cool. well that's better than before
12:45pmoreau: Oh, I didn't realise! :)
12:45imirkin: before it also dumped things to dmesg
12:45imirkin: about invalid values
12:46pmoreau: Well, I didn't checked dmesg, but in the "out" line, there are no such erros.
12:47pmoreau: I can retry, this one is realy quick to complete
12:49adrgrg: hey, can anyone tell me where to find nouveau.h included from nouveau_local.h?
12:50adrgrg: i'm tring to compile on an obscure distribution and lost in figuring out dependencies?
12:51adrgrg: I know two nouveau.h's, one from libdrm and one from linux-headers
12:52imirkin: adrgrg: which nouveau_local.h? ./src/mesa/drivers/dri/nouveau/nouveau_local.h that one?
12:52imirkin: almost definitely the libdrm one
12:52imirkin: adrgrg: what's the compilation error you're getting?
12:54imirkin: ah. well, same answer :)
12:54imirkin: (and same question -- what's the error?)
12:55imirkin: adrgrg: for working out dependencies, you might consider cheating and just looking at what others have done. e.g. http://sources.gentoo.org/cgi-bin/viewvc.cgi/gentoo-x86/x11-drivers/xf86-video-nouveau/xf86-video-nouveau-1.0.11.ebuild?revision=1.1
12:56adrgrg: ah, good idea, thx
12:56adrgrg: the error: nouveau.h: No such file or directory :D
12:57imirkin: yeah, you need libdrm
12:57adrgrg: it's my first time trying to compile drivers and stuff
12:57imirkin: the configure script should have failed
12:57adrgrg: and I don't know c, so don't sweat it :)
12:57adrgrg: learning as I go (hopefully :) )
12:58imirkin: you may benefit from first using a distro that manages those compilations for you, e.g. gentoo
12:58imirkin: fwiw i switched from slackware to gentoo a decade ago due to precisely this dependency-disaster issue
12:59imirkin: you want to build package X, but that requires you to upgrade Y Z W, each of which depend on ...
13:00adrgrg: i'm used to binary distributions, arch was fine for me until now, but i'm hyper curios about this distribution
13:00adrgrg: it's morpheus :)
13:01imirkin: right. so with gentoo, everything is still built on your box, but the dependencies/etc are managed by portage
13:01pmoreau: imirkin: Confirmed, no errors in dmesg: good work!
13:02imirkin: (i think arch largely copied that system for their package system/etc)
13:02imirkin: pmoreau: ok. sad that it still fails though. i guess some investigation will be necessary
13:02imirkin: i have some ideas, but... gr
13:02adrgrg: ok, at least I know it's the libdrm one :)
13:02imirkin: my only card is that nva3+gddr5 which should boot now, but apparently still has pretty significant issues
13:03pmoreau: Oh, it can boot now? How was it fixed?
13:03imirkin: after i banged my head against the wall for like 6 months, ben looked at it for 1-2h and fixed it
13:04imirkin: turns out they conditionally appended something to the ctxsw script for those cards
13:04imirkin: to disable some bits somewhere
13:05imirkin: commit fd3e5b6194e in ben's repo
13:09mew007: am I here?
13:09pmoreau: Kinda hacky, but if it works. :)
13:09pmoreau: mew007: Not sure
13:10mew007: pmoreau: why?
13:14adrgrg: ok, found the answer, libdrm was compiled with "--disable-nouveau" :)
13:15imirkin: yeah, that's kind of a bummer
13:15imirkin: btw, you should use xf86-video-nouveau 1.0.11 -- that fixed a bunch of issues related to tearing
13:16adrgrg: i'm walking on eggshells over here, 1.0.9 is fine for now :)
13:17adrgrg: anyway, now I need sarea.h that seems to be in xorg-server, right?
13:28BenBB: Hey there. On literally every Linux distribution I've tried (Ubuntu/Fedora/Arch) I've been having issues with my EVGA 750ti SC graphics card.
13:29BenBB: Nouveau doesn't seem to support it and it always seems to stretch weirdly.
13:29BenBB: I have to use nomodeset to make the display clear.
13:29BenBB: Works okay with proper NVIDIA drivers
13:29BenBB: Card works flawlessly on Windows as well.
13:29imirkin: BenBB: yeah, that's expected
13:30BenBB: imirkin, Is there anything I can do about it?
13:30BenBB: Makes installing distros rather hard.
13:30imirkin: BenBB: http://nouveau.freedesktop.org/wiki/NVC0_Firmware/
13:30imirkin: actually it should modeset fine i guess
13:30imirkin: just no acceleration
13:30imirkin: that stretching is a bit weird... hm
13:31BenBB: If I use nomodeset it displays fine but at low resolution.
13:31imirkin: [you have a maxwell card, GM107, for which there's no nouveau-supplied context switching firmware. you have to extract it yourself from the blob.]
13:31BenBB: but then it'll stretch after a random set of time.
13:31imirkin: with vesa?
13:31BenBB: I'm assuming so.
13:31imirkin: that's not supposed to happen :)
13:32imirkin: and even with nouveau it should work fine with the max resolution
13:32BenBB: If I boot normally it'll just look horribly distorted and wrong. You can still use it :\
13:32imirkin: how is your screen attached?
13:32imirkin: that's one of the less-problematic ones...
13:32imirkin: if DVI is an option, that's even less problematic
13:33BenBB: I would use DVI but my computer case blocks off the DVI port on the card for some reason :\
13:33imirkin: so... do you have some linux distro installed on it now?
13:33BenBB: I've got Arch installed atm.
13:34imirkin: with nvidia blob or nouveau?
13:34BenBB: I'm on Windows now though.
13:34BenBB: With nvidia
13:34imirkin: k. well, for debugging nouveau you'd need nouveau.
13:34imirkin: i've never heard of your particular problem though
13:35mew007: pmoreau: is Martin still offline?
13:35BenBB: imirkin, thanks for the help anyway.
13:35imirkin: i don't think pmoreau has any special insight into that. be patient. wait a day. or two.
13:35pmoreau: mew007: Well, I can't tell. Just that he hasn't spoken on IRC since
13:35mew007: ahh, ok
13:37mew007: pmoreau: how cool in C should I be to participate in nouveau project?
13:38imirkin: BenBB: ideally start with uploading dmesg after a nouveau boot that shows the issues... maybe booted with drm.debug=0xe nouveau.debug=debug
13:38imirkin: BenBB: if you want to avoid the hassle of changing your install around, i think there's a livecd somewhere...
13:39pmoreau: I don't know how Mesa code looks like. In the kernel part, be prepared to deal with class hierarchies using structures and function pointers
13:39imirkin: mew007: can you define some units of coolness? :)
13:40pmoreau: imirkin: Something like the ones here? https://nouveau.pmoreau.org
13:40mew007: imirkin: maybe you know: my 550ti works when I run Linux Mint 17.1 Mate i386 in compatibility mode. What is difference between usual mode and compatibility?
13:40imirkin: mew007: dunno, depends on the distro. there's no special definition.
13:40imirkin: BenBB: see pmoreau's link.
13:41mew007: imirkin: 5 years in real system development - imo, it's cool. But I have no experience in C at all.
13:41imirkin: pmoreau: do those boot with liveusb as well?
13:42pmoreau: imirkin: They can be dd'ed to a usb stick and used to boot if that's what you mean
13:44pmoreau: BenBB: The WM is missing some configuration... So you don't have access to applications from it (yet)... You'll have to switch to a tty to get the dmesg
13:44yurikoles: is someone with maxwell here?
13:44yurikoles: I can't setup opensource drivers on my arch with 750 Ti
13:45imirkin: yurikoles: what seems to be the issue?
13:45BenBB: imirkin, Thanks ;)
13:45yurikoles: have done extracting FW and putting it to the right place
13:45yurikoles: installed mesa-git and ddx git
13:45mew007: and may one small burnt-out capacitor cause videocard problems with nouveau and let videocard work perfectly with nvidia drivers?
13:45imirkin: yurikoles: pastebin dmesg, Xorg.0.log
13:46yurikoles: but when I start X it shows welcome screen (sddm) in corrupted colors, then hangs
13:46yurikoles: and only reset can help me
13:47yurikoles: how to get latest boot's dmesg with journald?
13:47pmoreau: The dmesg command still works :)
13:47imirkin: dunno. journald is too complex for me to use.
13:47yurikoles: but it's for current session
13:48imirkin: but if you can get those logs, i can see if there's anything obvious
13:48yurikoles: how to get dmesg for previous boot?
13:48imirkin: when you say 'corrupted colors', what do you mean?
13:48imirkin: just general corruption, or some sort of color swap?
13:48yurikoles: yes it swap
13:48yurikoles: but on different parts of screen
13:49imirkin: take a photo?
13:49yurikoles: some parts are correct but some is swapped
13:49yurikoles: let me try again
13:49pmoreau: `journalctl --list-boots` returns a list of boots id
13:49yurikoles: is it fine to run stable mesa and ddx?
13:49yurikoles: or only -git?
13:49imirkin: xf86-video-nouveau 1.0.11 and mesa-10.3.x should be fine for maxwell
13:50pmoreau: then you can use `journalctl --dmesg --boot=bootid` to retrieve kernel logs relative to that boot
13:51yurikoles: ok, let me try again, i will return in few minutes (I hope)
14:28imirkin: yurikoles: what xf86-video-nouveau are you using?
14:29yurikoles: with git ddx and mesa it was different
14:29imirkin: 1.0.11 should be fine
14:31yurikoles: here is dmesg http://pastebin.com/qXsirHmV
14:31imirkin: i think you meant
14:32imirkin: otoh since there's no internal firmware, it tries to use external firmware anyways
14:32yurikoles: xorg log http://pastebin.com/CJ8QxcCU
14:34yurikoles: so I need to try with your flag?
14:34imirkin: no, you don't
14:34imirkin: but you should fix it in the future. won't change anything for now (other than removing the warning about the unknown option)
14:35imirkin: oh right... you're using glamor
14:35yurikoles: so how to make it work?
14:35imirkin: hold on
14:35imirkin: what mesa are you using?
14:38imirkin: yurikoles: fwiw, glamor had a bunch of issues when i played with it... iirc i did some thing where i forced it to use EXA and disabled the actual acceleration bits
14:39adrgrg: ok my xorgserver is compiled with --disable-dri --disable-dri2 --disable-aiglx --disable-glx-tls --disable-glx --disable-glx --disable-config-dbus --disable-config-hal
14:39adrgrg: can any of these affect nouveau?
14:40imirkin: yurikoles: unfortunately that config is on mupuf's machine which is presently in transit. something like EXAMask or something, i forget
14:40imirkin: adrgrg: yes, like all of those need to be enabled
14:40imirkin: or at least all the non-dbus/hal ones
14:40adrgrg: heh, good, thx :)
14:41imirkin: adrgrg: where are you getting such a config from? that's basically the "let's use X like it was in 1980" options
14:43adrgrg: morpheus linux :)
14:45adrgrg: told you, it made me super curious, it's from the suckless.org guys I think
14:46imirkin: yeah, but disabling dri means no accel. if that's what you want, that's fine
14:46imirkin: nouveau should work fine with that
14:46imirkin: but you won't get 3d accel
14:46adrgrg: it's fine :)
14:47adrgrg: so, do I still need to enable hal and dbus?
14:47imirkin: those are the ones you should keep disabled
14:47imirkin: the rest i would enable
14:49adrgrg: no accel is fine for now :)
14:49imirkin: you'd still get 2d accel
14:49imirkin: just no 3d accel
14:49imirkin: in that case all those --disable-dri/glx/etc things are fine
14:50adrgrg: good, thx a lot :)
14:55adrgrg: @imirkin are you a contributor to nouveau?
14:55yurikoles: did I miss smg?
14:55imirkin: yurikoles: yes, you did -- check logs
14:56imirkin: (but nothing serious)
14:56imirkin: adrgrg: yes
14:57adrgrg: cool, as part of your job or in you spare time, I always wondered how people contribute to open source?
14:58yurikoles: but in past it was working
14:59imirkin: adrgrg: spare time
14:59yurikoles: may be 2 months ago, all git
14:59imirkin: yurikoles: what version of mesa are you using?
14:59imirkin: in any case, it was most likely some kernel change which messed things up
15:00imirkin: you could find a kernel that worked and bisect
15:01yurikoles: I don't recall correctly but I think I used nouveau module from freedesktop.org
15:01imirkin: there were definitely big problems with using glamor in nouveau ddx in 1.0.11
15:01imirkin: so i'd try flipping it to EXA and then using an EXAMask of 0
15:01yurikoles: maybe just build it from git?
15:01imirkin: i don't know that the issues have been fixed
15:02yurikoles: did you get EXA to work on Maxwell?
15:02imirkin: no... i got close and then ran out of steam
15:04imirkin: had to decide how to do the stupid scratch allocations for the quads being drawn
15:04imirkin: since there's no inline vertex mode anymore
15:05yurikoles: all that you are saying now is magic for me
15:05yurikoles: maybe just try kernel from freedesktop or at least nouveau module?
15:06imirkin: i dunno, try to figure out what you were running when it worked
15:06imirkin: and then see if you can bisect to identify the breakage
15:06yurikoles: how to lurk trough the irc logs?
15:07imirkin: see topic
15:12yurikoles: ok, so I will just try to search with site:freedesktop.org for yurikoles
15:14RSpliet: imirkin: did you benchmark "nv50/ir: fold MAD when one of the multiplicands is const"? I'm curious what such an optimisation would give you in practice :)
15:14imirkin: RSpliet: very little
15:15imirkin: RSpliet: i think it's fairly uncommon for that pattern to occur
15:15imirkin: but i did see it a bunch with code generated by st/nine, not 100% sure why
15:15yurikoles: hi, Roy
15:15imirkin: and i dislike seeing crap code
15:15yurikoles: 03:57 < RSpliet> I'm glad it got resolved so quickly :)
15:15yurikoles: 03:57 < RSpliet> so we can get yurikoles to test my tree soon :p
15:15yurikoles: it was 2014-09-22
15:16yurikoles: so you get gt240 reclocking to work?
15:16RSpliet: yurikoles: haha, wow erm... yes, I thought that was you
15:16RSpliet: the GDDR3 one
15:16RSpliet: yes, should work with kernel 3.19rc1
15:17RSpliet: imirkin: hehe fair enough, s not 0,4fps extra with $Windows_game_X
15:17yurikoles: maybe my gt240 will be faster then gtx 750 ti?
15:17yurikoles: I mean opensource stack :)
15:17imirkin: RSpliet: could be for all i know, but i just saw crap code and wanted to not be generating it
15:17RSpliet: yurikoles: I doubt it, although I was surprised to see Portals framerate double
15:18RSpliet: 40->80fps (at given resolution X)
15:18imirkin: RSpliet: it's almost as if memory frequency somehow correlates to the speed at which data can be read and written :p
15:18RSpliet: imirkin: yes, isn't that extrordinary?
15:20yurikoles: and gpu unit itself?
15:20yurikoles: it's also get overclocked?
15:21RSpliet: not overclocked... just... well, clocked properly
15:21RSpliet: at the tip of your fingers that is, you have to manually tell nouveau to pick a set of clocks
15:22yurikoles: yes, I mean factory clocks :)
15:22RSpliet: but yes, all the relevant clocks are changed upwards if you tell it to
15:22yurikoles: so for now I have half year old hw that isn
15:22yurikoles: ins't supported in free software drivers
15:23yurikoles: and 5 year old card that is running well
15:23yurikoles: not bad :)
15:23RSpliet: hehe, yes, I hope maxwell gets some proper TLC soon
15:26imirkin: yurikoles: well, with a volunteer developer base, people tend to work on what they feel like working
15:26imirkin: maxwell doesn't seem to be high on the list
15:26imirkin: and the only paid developer on the project has _plenty_ of stuff to do s.t. maxwell, while a priority, is not necessarily #1
15:28imirkin: yurikoles: btw, mind doing 'ls -l /lib/firmware/nouveau/nv117*'
15:30RSpliet: imirkin: I was about to say... even if it is #1 priority, "that firmware thing" is not very motivating; not knowing where to code towards
15:31imirkin: yurikoles: hexdump -C /lib/firmware/nouveau/nv117_fuc409d | head -20
15:32imirkin: yeah ok, that's all good
15:32imirkin: just wanted to make sure it wasn't totally messed up data
15:32yurikoles: btw, where to read about all that autogenerating FW stuff?
15:33imirkin: i don't think a substantial effort has been made to work out the maxwell context switching stuff yet
15:35yurikoles: I mean kepler
15:36imirkin: there's probably a tiny bit about it in hwdocs
15:36imirkin: the rest... rtfs :)
15:36imirkin: i.e. not extremely friendly
15:36imirkin: i, for one, have no clue how any of it works
15:38yurikoles: all hope is for Maxwell gpu for android :)
15:38yurikoles: maybe next year, yes?
15:38yurikoles: as I know they are contributing code and some specs for Tegra TK1?
15:39imirkin: no specs, a little code
15:39imirkin: and a lot of glue
15:41RSpliet: and a lot of docs in the form of an android driver :p
15:41yurikoles: LINUX HAS NO GAEMS
15:41yurikoles: but Android does
15:42RSpliet: eh what? Linux has Steam with a good hundred+ games
15:42imirkin: not to belittle their efforts... but most of the stuff they're fixing is very platform-specific
15:42imirkin: and largely not gpu-specific
15:42yurikoles: so It will be better for me to run Android-x86 on my Maxwell box
15:43imirkin: but the platform stuff _is_ very important (and annoying!) to fix, so i'm very glad they're doing it :)
15:43RSpliet: imirkin: exactly :) and I bet it's their first interest, nouveau is good enough to work on the K1 because it's essentially just a kepler with a handicapped memory subsystem
15:44imirkin: well, handicapped by the fact that it's arm
15:44RSpliet: and it uses stolen memory
15:44RSpliet: instead of dedicated VRAM
15:44imirkin: it doesn't
15:44RSpliet: it doesn't?
15:44imirkin: it's UMA
15:44imirkin: not stolen
15:45imirkin: stolen means "i'm going to use vram, but it's actually going to be backed by system ram but you won't know about it"
15:45imirkin: uma means "just use sysram"
15:45RSpliet: hmm yes, you're right
15:47yurikoles: at last
15:47yurikoles: at last
15:48yurikoles: I finally get it to work on 3.19-rc1 :)
15:48yurikoles: now I can have a good sleep
15:48imirkin: the gt240 or the gtx750?
15:48yurikoles: all that relates to nouveau is from git
15:49yurikoles: how to disable vsync for gears?
15:50imirkin: vblank_mode=0 glxgears
15:51yurikoles: thank you
15:51imirkin: iirc with glamor, it didn't expose the GLX_ARB_create_context_profile extension, so you couldn't create core contexts
15:51yurikoles: already downloading heaven
15:51imirkin: that was the #1 reason i switched to EXA
15:52yurikoles: oh shi...
15:52yurikoles: speed is increasing!
15:52imirkin: double check with glxinfo
15:52yurikoles: now again decreasing
15:52yurikoles: so it
15:53yurikoles: and I will definetly try cs_go
15:54imirkin: mupuf reported some issues in xonotic... i made a few semi-random fixes which resolved a couple of them, but there's still a big one left
15:55imirkin: but piglits pass on gm107 about as well as they do on other cards, which is nice
15:55imirkin: otoh it's unclear where the remaining bug lies
16:04yurikoles: so heaven is hanging my gpu
16:04yurikoles: imirkin: do you have a Maxell GPU?
16:05yurikoles: maybe I can send you a gift
16:05imirkin: the stuff i've done i've tested on mupuf's box
16:05imirkin: nah, i don't really have the time for any of that stuff right now
16:05yurikoles: ok, let me know
16:05yurikoles: my email is email@example.com
16:06imirkin: unless you can make the kind of gift that could fund a full time developer for a prolonged period of time. (in which case it still wouldn't be me, but perhaps others here)
16:06yurikoles: I will send Maxwell to anyone who is want to do a driver work
16:07yurikoles: so it can be a guy in Eastern Europe I think
16:07yurikoles: we have good developers here in Ukraine
16:07imirkin: send 'em over here :)
16:07yurikoles: can you give me a vacancy description
16:07tobijk: pretty expensive, duh
16:08imirkin: like... what kind of developers would ideally join, or?
16:08yurikoles: I will talk with my HR girls (I have ~150 in my skype)
16:08imirkin: well, this is still a volunteer project, presumably those devs expect to get paid
16:08yurikoles: why not propose project for free desktop
16:09imirkin: again, there's no actual money to pay people
16:09yurikoles: I can give around $500/mo
16:09imirkin: a few EVoC's get funded
16:09imirkin: i've toyed with funding mechanisms to employ people on nouveau full time, but haven't come up with any genius ideas
16:10imirkin: only wishy-washy lots-of-hope-and-dreams type stuff, not something reliable
16:10yurikoles: so for begining we need a web page for this project
16:10imirkin: http://nouveau.freedesktop.org/wiki/ is the web page :)
16:11yurikoles: and maybe croudfunding compain
16:11imirkin: yeah, of course that involves lots of annoying things, like what is getting funded
16:11imirkin: is it one individual, or is it an organization
16:11imirkin: if it's an organization, then that requires some amount of administration
16:12yurikoles: so Illia why don't we create a croudfunding project?
16:12imirkin: what is the charter/legal foundation of thar org... etc
16:12yurikoles: or Freedesktop
16:12yurikoles: it may be FOP Kolesnikov Y. S. in Ukraine
16:13yurikoles: who cares?
16:13imirkin: the people giving money
16:13imirkin: and the governments that collect tax/enable these things
16:13yurikoles: ok, here in Ukraine I pay 5% for any income to my account
16:14yurikoles: I think in Europe there is better taxes
16:14imirkin: it would be interesting to test the waters of crowdfunding, to see how much interest can be generated in funding nouveau
16:14imirkin: my bet is -- not enough
16:14yurikoles: then another guy in Ukraine or Russia/Poland
16:15yurikoles: do you have contacts in freedesktop?
16:15imirkin: i'm not sure that freedesktop is a thing
16:15imirkin: X.org is a thing, but they're trying to merge into SPI
16:15yurikoles: it can be at least $100/year for Windows license
16:16imirkin: partly because of the administration/legal headaches
16:16yurikoles: so we can get at least this money from Linux funs with Maxwell
16:16yurikoles: what is SPI?
16:16imirkin: i dunno, some org
16:17imirkin: i might have the name wrong
16:17imirkin: Software In The Public Interest
16:18imirkin: i think i saw another one recently that did ~the same type of stuff
16:18yurikoles: so for example we have some developer from Eastern Europe, I think it will take about 2 weeks to find one
16:19imirkin: what kind of salary can such a developer expect to make?
16:19yurikoles: can you do the interviews?
16:19yurikoles: I need a vacancy description
16:19imirkin: the thing is that for a new person to learn about nouveau is a major thing... it's not like some well-defined task of "well just fix maxwell" or whatever
16:19imirkin: in order to do that, an enormous amount of understanding needs to be had
16:20imirkin: so it would have to be someone employed for a long time, not a short-term thing
16:20yurikoles: then I can talk with recruiters, they take about ~50% of first salay if person do at least 3 months work
16:20imirkin: i don't think it'd be hard to find someone
16:21imirkin: i think it'll be hard to come up with the funds
16:21yurikoles: for now let's calculate funds
16:21yurikoles: it's need to be Linux hacker with knowledge of C++ or what?
16:22imirkin: C more than C++
16:22yurikoles: ok, Senior Linux hacker, yes?
16:22imirkin: no particular graphics knowledge required, on the bright side
16:22imirkin: sure why not
16:26yurikoles: what is correct name of nouveau project?
16:26yurikoles: I'm creating profile on djinni.co
16:26yurikoles: it's good site
16:27yurikoles: and could you please give me time when you able to make a skype interview in GMT?
16:28imirkin: there is no official nouveau organization
16:28imirkin: and i'd definitely want to talk with the person over email before skype
16:28yurikoles: Бонус равен 10% от зарплаты в профиле кандидата, но не менее $100.
16:28imirkin: (and see their resume/etc)
16:28imirkin: monthly or yearly?
16:29imirkin: but again, how much would such a developer expect to make in eastern europe?
16:32yurikoles: ok, it's maybe know-everything kid
16:32imirkin: presumably per month?
16:33imirkin: (that's really cheap)
16:33imirkin: in US that sort of resume would fetch 10k/mo or more
16:35yurikoles: but we need also a lot of gpus for him
16:36imirkin: cost of hardware is insignificant
16:36yurikoles: can you write some vacancy requirements/skills somewhere on Google Drive?
16:36imirkin: compared to human cost
16:36yurikoles: my gmail is firstname.lastname@example.org
16:37yurikoles: or could you please send me an email?
16:37yurikoles: it's up to you
16:37yurikoles: so a develeloper is need to know Linux kernel and userspace, C, Git
16:38yurikoles: what else? reverse-engineering?
16:38imirkin: meh, more like 'attention to detail'
16:39imirkin: and experience working on open-source projects is probably good
16:39imirkin: or at least a desire to acquire such experience
16:39yurikoles: Illia, we need to sell this guy to Linux funs
16:40yurikoles: did you open my link?
16:45RSpliet: imirkin: CodeEmitterNV50::emitFMAD hints at having a "saturate" bit for the destination in the long encoding format
16:45yurikoles: I already searching for a person
16:45imirkin: RSpliet: ok...
16:45RSpliet: but the nv50_target doesn't define it as available... as a result the ModifierFolding pass doesn't fold a sat into the mad
16:45yurikoles: I just tell them that I search a Developer for Nouveau project
16:46imirkin: RSpliet: ah cool, fix it :)
16:46imirkin: RSpliet: but test it ;)
16:46yurikoles: They need to know the rest
16:46RSpliet: imirkin: sure! I'll see what I can do
16:46RSpliet: but I see this pattern recurring on UT2003
16:46yurikoles: and our friend Roy!
16:46yurikoles: Roy, do you plan to work on nouveau in future?
16:47RSpliet: yurikoles: long story...
16:47yurikoles: how about full time job?
16:47yurikoles: anyone of you guys, who has at least one commit?
16:48imirkin: yurikoles: probably a number of people here. however they'd want a stable job they could count on for some time, not a 1-month thing
17:03yurikoles: so guys, anyone who is interesting (has knowledge, video cards funds, PR experience)
17:03yurikoles: please join channel #nouveau-crowd-funding
17:05adrgrg: sorry for the noob questions, if I just copy sarea.h to src dir will it work? I see that in sarea.h there's only one include file which points to a known path (drm)
17:06imirkin: adrgrg: iirc the joke is that sarea isn't actually needed
17:06adrgrg: can I delete the iclude then?
17:07adrgrg: I'm curious
17:07imirkin: well, the code uses it
17:07imirkin: but not for any productive purposes
17:08adrgrg: anyway I copied the sarea.h file to the src dir, and tried to compile
17:08adrgrg: lots of errors :/
17:13adrgrg: ofc, I get errors on dri :/
17:14adrgrg: dooh :)
17:32imirkin: the other org, which wine is a part of apparently, is https://sfconservancy.org/ which apparently does a bunch of stuff to help projects too (the legal/admin aspects)
18:09yurikoles: does anyone know how to install and run d3dnine?
18:09imirkin: try #d3d9
18:10imirkin: there's a wiki page somewhere... https://wiki.ixit.cz/d3d9
18:29RSpliet: imirkin: just shoved a tiny patch to the mL
18:29RSpliet: I used http://hastebin.com/jivuqewayo.mel to test :)
18:29imirkin: haven't gotten it...
18:30imirkin: are you not subscribed to the list?
18:30RSpliet: not sure...
18:30RSpliet: whether I'm subscribed with this particular e-mail address that is
18:30imirkin: pastebin patch?
18:32imirkin: and pastebin the NV50_PROG_DEBUG=1 with that test?
18:35imirkin: looks good
18:35RSpliet: that's what I thought :)
18:38imirkin: ok, received the patch in mail now :)
18:39imirkin: btw, in the future, wrap commit descriptions to 72 chars
18:40RSpliet: oh yeah... sorry
18:40imirkin: np, i fixed it up
18:40RSpliet: the tiny things that I keep forgetting :p
18:41imirkin: well, by default many editors will auto-wrap when you do 'git commit'
18:42RSpliet: vim apparently doesn't
18:42imirkin: does for me...
18:44imirkin: i think i'm also going to push my mad folding opt
18:46RSpliet: I don't have a test case, but it looks fine
18:47imirkin: yeah, it's easy for that stuff to look fine
18:48imirkin: but i think i've tested all the various cases
18:51RSpliet: hehe, that's remarkably true
18:51imirkin: [and i had originally missed some of the various cases]
18:51imirkin: but pmoreau didn't see any regressions in piglit... so probably fine
18:51imirkin: maybe :)
18:53imirkin: anyways, pushed
18:54RSpliet:pushes himself into bed
18:54tobijk: i'm too late :( the compression is ready as well...
18:55imirkin: i can push many times
18:55imirkin: it's not like a once-per-year thing
18:55imirkin: and now you have to wait 'til 2016 :)