02:45 FreeFull: I have a laptop with Optimus, and I'm trying to get PRIME to work, but xrandr --listproviders only lists one provider, and I can't figure out why
02:46 FreeFull: Both the i915 and nouveau kernel modules are loaded, and I have both xf86-video-intel and xf86-video-nouveau installed too
02:46 FreeFull: mesa is configured for DRI 2 AFAIK
02:47 airlied: if you have an xorg.conf try without it
02:50 FreeFull: There is no xorg.conf
02:50 airlied: pastebin the xorg log file then
02:50 FreeFull: I have a 20-intel.conf that specifies uxa as the AccelMethod, because exa is buggy
02:51 airlied: that'll do it,
02:51 FreeFull: Ok
02:51 airlied: try without that
02:51 airlied: I assume you mean sna
02:51 airlied: since intel doesn't do exa
02:51 FreeFull: Oh, yeah, sna
02:51 FreeFull: sna is buggy
02:52 FreeFull: So, how would I deactivate something that's in xorg.conf.d, other than moving it elsewhere?
02:54 airlied: do that, move the file out
02:55 FreeFull: airlied: Thanks, that was it
02:56 FreeFull: How do I go about disabling sna then?
02:56 FreeFull: Hm, maybe I don't have to disable sna now, seems the bug I was having with it got fixed
02:59 FreeFull: airlied: Either way, PRIME seems to be working now, thanks
03:18 FreeFull: Ok, one big problem: Significantly lower FPS when offloading, compared to using the Intel graphics
03:21 imirkin: FreeFull: what gpu?
03:21 imirkin: [and what kernel]
03:22 FreeFull: imirkin: Intel Haswell + GeForce GT 745M
03:22 FreeFull: Kernel version 4.6.3
03:22 imirkin: you can try reclocking the nvidia gpu
03:23 imirkin: cat /sys/kernel/debug/dri/1/pstate
03:23 FreeFull: My /sys/kernel/debug/ directory is empty
03:23 imirkin: you need to mount debugfs
03:24 FreeFull: Oh, nevermind, it's not empty, I just couldn't see what files were in it as non-root
03:25 FreeFull: imirkin: 1 seems to be the intel graphics
03:25 imirkin: then try 0 :)
03:25 imirkin: oh, you might also want to have DRI_PRIME=1 glxgears running while you play around with this
03:25 imirkin: it's not super-duper reliable when the gpu's suspended
03:27 FreeFull: 07: core 405 MHz memory 810 MHz 0a: core 405-1045 MHz memory 1600 MHz 0f: core 405-1045 MHz memory 1800 MHz AC: core 405 MHz memory 810 MHz (with glxgears running)
03:29 FreeFull: Also another problem, if I have a fullscreen program with DRI_PRIME=1, and switch workspaces, the program graphics keep getting displayed on the screen. Wondering if that's xcompmgr's fault somehow
03:29 imirkin: ok, so ... this might hang your box
03:29 imirkin: so save your work first
03:29 imirkin: and then
03:30 imirkin: echo 0f > /sys/kernel/debug/dri/0/pstate
03:30 imirkin: the AC line should update itself to look a lot more similar to the 0f line
03:31 imirkin: ideally identical
03:32 FreeFull: My box didn't hang, but the echo itself seems to be stuck
03:33 FreeFull: Maybe I should have had glxgears still running when I tried it..
03:34 FreeFull: Trying to run glxgears with DRI_PRIME=1 doesn't get anywhere now either
03:35 FreeFull: imirkin: Yeah, anything trying to access the nvidia graphics now seems to be hung in an unkillable way
03:35 FreeFull: The panel is handled by intel so everything else still works
03:43 FreeFull: imirkin: Should I reboot?
03:47 imirkin: probably... yeah, any messing with that file should have the gpu active
03:47 imirkin: esp writing
03:48 imirkin: i guess i didn't make that clear enough... sorry
03:52 orbea: Will vulkan and nouveau be at all usable eventually? Vulkan seems like it will be useful for using emulators via the libretro api in the future.
03:52 imirkin: as for wrong things displayed... that's definitely not nouveau's fault.
03:52 imirkin: orbea: maybe? who knows. i have no immediate plans to work on vulkan, but perhaps others do?
03:53 orbea: cool :)
04:23 FreeFull: imirkin: I'm pretty convinced by now that 0f is not a good value to write into pstate :)
04:23 imirkin: did you have DRI_PRIME=1 glxgears running?
04:23 imirkin: you can also try with 0a ... on your box that should be a pretty comparable level
04:24 imirkin: there's also a branch which has much better reclocking support
04:24 FreeFull: I did have glxgears running with the nouveau offloading
04:24 imirkin: although mostly for gddr5 vram, which i don't think you have
04:24 FreeFull: I ended up with a *ton* of "nouveau 0000:04:00.0: fifo: SCHED_ERROR 06 []" errors in my syslog
04:25 FreeFull: I'll try 0a then
04:25 imirkin: https://github.com/karolherbst/nouveau/commits/stable_reclocking_kepler_v5
04:26 imirkin: skeggsb_: you should really take a minute and catch up on all those patches... a ton of people have tested them by now
04:26 FreeFull: Ah, yeah, I am on kepler
04:27 FreeFull: Using that branch might be a good idea then
04:27 FreeFull: I'll try 0a first
04:27 imirkin: well, like i said... most of the benefit it brings is for gddr5 vram
04:27 imirkin: but it also fixes some core reclocking issues, etc
04:27 imirkin: including fixing up voltage
04:27 imirkin: which might be the issue you're having
04:27 FreeFull: The graphics chip seems to have 2GB of dedicated RAM, but I'm not sure what kind
04:28 FreeFull: How long do you think it'll take for that branch to be mainlined, and then end up in a release?
04:28 imirkin: well, karol's been building it up for quite a while already
04:29 imirkin: so... a while longer, based on past history
04:29 imirkin: not sure why ben just doesn't merge it wholesale, but ... meh
04:30 FreeFull: I'll probably have to compile that myself then, or switch back to the propertiary nvidia drivers for the time being
04:31 imirkin: if you're looking for optimal performance out of the hw, nouveau won't come close
04:31 FreeFull: I wonder how difficult it is to get the propertiary drivers to work with PRIME
04:31 FreeFull: With bumblebee I was getting good performance, but bad latency
04:31 imirkin: afaik they've been working in integrating with it... but it'll still be annoying with diff libGL's
04:32 imirkin: too late now, but if you want the best open-source experience, stick to intel + amd
04:33 FreeFull: My previous laptop (which I still have) is just AMD (One of those CPU+GPU APU combos)
04:33 FreeFull: Hardware definitely isn't as powerful
04:33 imirkin: yeah, APU's aren't designed to compete with dGPU's
04:34 FreeFull: imirkin: I'll keep it in mind when purchasing a laptop in the far future
04:34 FreeFull: Assuming it won't be outdated advice by then ;)
04:34 imirkin: nah, it'll be much more actual advice
04:34 imirkin: since starting with the GM20x series, we basically can't do anything useful with the GPUs without nvidia's explicit OK
04:35 FreeFull: Well, maybe in 40 years nvidia will be out of business or something
04:35 imirkin: they don't build laptops like they used to...
04:35 imirkin: i doubt this one will last 40 years :)
04:35 FreeFull: True
04:36 FreeFull: I guess I can try propertiary with PRIME right now
04:40 FreeFull: I'll have to write a xorg.conf, it seems
05:23 FreeFull: Seems the PRIME support in the propertiary nvidia drivers is limited to "nvidia chip always draws everything, intel chip is just used to get access to the laptop's panel"
05:23 FreeFull: At least that'll give me good performance, I guess
05:28 FreeFull: Shadertoy seems a ton more responsive
09:00 pmoreau: karolherbst_work: `git log --oneline --grep="drm-next" | sed -e 's/^[a-z0-9]\{7\} drm-next //' | head -1` there you go
09:00 pmoreau: Now, let’s insert that in the pulling script of linux drm-next
09:06 karolherbst_work: pmoreau: you are the best
09:08 karolherbst_work: pmoreau: this returns the commit hash of the last commit with drm-next or the drm-next hash?
09:09 pmoreau: This returns the commit hash of the last supported drm-next version
09:16 karolherbst_work: mhh, I would need the hash of the nouveau commit , so that I can rebase drm-next.next on drm-next^
09:21 pmoreau: Well, then that would be `git log --oneline --grep="drm-next" | sed -e 's/ drm-next [a-z0-9]\{40\}$//' | head -1`
09:56 pmoreau: karolherbst_work: Alternatively, retrieving only the hash can be done using git, so: `git log --grep="drm-next" --format="%h" -s | head -1` which is a bit shorter, and most likely more efficient.
09:59 karolherbst_work: well, I will play around with it on the weekend then :)
09:59 karolherbst_work: or maybe today
09:59 pmoreau: :-D
10:03 karolherbst_work: anyway, we need to plan XDC at some point :p can't say that I will fine time for that the month before it :p
10:04 karolherbst_work: well planning the monday before might be a bit too short on time
10:04 karolherbst_work: who will come by the way?
10:07 karolherbst_work: skeggsb_, imirkin, imirkin_, pmoreau: will any of you come? (and I also asked those I forgot here :p )
10:08 pmoreau: karolherbst_work: I will, but you already asked me: remember the friend of mine defending on the Friday? :-)
10:09 pmoreau: I don’t remember if Samuel is coming or not.
10:11 karolherbst_work: pmoreau: yeah, I know :p I just asked again for stupid reasons
10:11 karolherbst_work: well not so stupid in my head
10:11 pmoreau: ;-D
10:13 karolherbst_work: anyway, now everybody knows it who isn't in dri-devel
10:14 karolherbst_work: ohh right, mwk I ask you too :p
10:17 hrw: hi
10:19 karolherbst_work: ahh RSpliet I ask you too…. why do I suck so much in creating a list of the most important nouveau people ...
10:20 karolherbst_work: pmoreau: you could add you to the list if you can: https://www.x.org/wiki/Events/XDC2016/Attendees/
10:21 hrw: which nvidia card would you recommend to buy when main requirement is ability to drive 4K monitor over displayport? Diablo III/wine or Civ V/linux would run good probably anyway in fullhd mode
10:24 karolherbst_work: hrw: well if you plan to use nouveau, then the 780 ti should be fine :p (hopefully, not guarentees though)
10:24 karolherbst_work: *guarantees
10:24 hrw: karolherbst_work: 750/760 would be too slow or not supported?
10:25 karolherbst_work: well 750 is maxwell, and that one is a bit bad supported
10:25 karolherbst_work: and the 760 fully reclocked might be too slow for 4k
10:26 karolherbst_work: well, if you are fine using nvidia as an alternative until we figure problems out, you can always get a nvidia gpu, but usually you should prefer intel/AMD due to official support
10:26 karolherbst_work: we can't guarantee you anything
10:28 hrw: karolherbst_work: after few years of radeon I want to get rid of them
10:29 hrw: it is not good when you start game and have to reset system after some (10-120) minutes because monitors went black
10:33 Calinou: 780 Ti is the best Nouveau card, IIRC
10:34 Calinou: but, it doesn't play lots of games at 4K@60 Hz playablt
10:34 Calinou: playably*
10:34 hrw: Calinou: I am fine with 1280x 1920x in games
10:34 hrw: not that I play a lot. It is more 'lets kill monsters, do a quest while software is building'
10:57 karolherbst_work: maybe a 770 might fine too
12:38 hrw: re after few radeon crashes
12:49 hrw: is gtx 950 one to avoid?
12:50 Yoshimo: GM206, yes
12:50 Yoshimo: https://nouveau.freedesktop.org/wiki/CodeNames/ might help
12:51 hrw: Yoshimo: so Kepler is a family to go for?
12:53 Yoshimo: at least the second maxwell gen aka gm2xx, i am not sure how well the first maxwells work at the moment.
12:53 Yoshimo: https://github.com/karolherbst/nouveau/commits/maxwell_reclocking doesn't work for the second gen
12:54 JosefR: I am experiencing very low frame rates when running weston with nouveau on the Tegra K1 with linux 4.7-rc6 and mesa 11.1 (weston-simple-egl demo: 1.4 fps). Any ideas what to check?
12:54 dcomp: Yoshimo: GM108M is currently hit and miss
12:55 Yoshimo: understandable
13:24 karolherbst_work: josef64: did you reclock?
13:24 karolherbst_work: ohh wait
13:24 karolherbst_work: tegra has automatic reclocking :D
13:24 karolherbst_work: josef64: there should be something like eglinfo? Check that you get the hardware renderer and no software one
13:25 josef64: karolherbst_work: Hm, wrong Nick?
13:25 josef64: yes :)
13:26 josef64: [15:01:26] <-- JosefR (b2ef4932@gateway/web/freenode/ip.178.239.73.50) hat das Netzwerk verlassen (Ping timeout: 250 seconds)
13:28 karolherbst_work: josef64: ohh right
13:28 karolherbst_work: the other is gone already...
13:28 josef64: np :)
13:29 karolherbst_work: those are the best, hanging around for like 20 minutes and are gone for good
13:29 karolherbst_work: 10 in this case
13:31 gruetzkopf: < better than those who don't have any nvidia cards at all *cough*
13:32 karolherbst_work: :p
13:32 karolherbst_work: I doubt that, cause those usually don't ask for support
13:33 karolherbst_work: but those who just stay for 10 minutes are the worst, cause they say that nouveau is shit, cause they got no support :p
14:58 imirkin: karolherbst_work: i have no plans on attending xdc2016.
15:00 karolherbst_work: :/ if that continues I won't see you in person ever :p
15:01 imirkin: i've seen a grand total of ... 4 developers involved in open graphics
15:02 imirkin: and if you'd asked me a month ago, that number would have been 2 :)
15:07 imirkin: unfortunately the majority are a continent (or two) away, so it doesn't happen too often
15:09 karolherbst_wor1: well I saw a lot already :D
17:49 dmj_s76: skeggsb_: How robust/reliable is the gp104/gp100 support in master at least as far as booting into a graphical DE?
17:49 imirkin_: dmj_s76: you won't get acceleration until nvidia posts firmware. afaik they've only sent a pull request for gp100, not gp104
17:50 imirkin_: and gp100 is a purely mythical gpu at this point, at least for end-users
17:51 dmj_s76: imirkin_: But for gp104, you would be able to get to a software accelerated desktop?
17:51 imirkin_: dmj_s76: yeah. expect it to be unbearably slow
17:51 imirkin_: esp if you use anything resembling a modern desktop environment
17:51 ajax: "software accelerated" is sort of a contradiction in terms
17:52 dmj_s76: imirkin_: That would be a major improvement over a blank screen unless you set 'nomodeset vga=791'
17:52 imirkin_: dmj_s76: meh :)
17:53 dmj_s76: imirkin_: Mostly thinking about people running a graphical installer
17:54 imirkin_: anyways, afaik you should get modesetting with gp104 with skeggsb_'s master branch
17:55 imirkin_: that said, it hasn't gotten a *ton* of testing, so ... ymmv
17:56 dmj_s76: imirkin_: Any chance the new gp104 support would introduce regressions for non-pascal hardware or is hardware enablement pretty well isolated?
17:56 imirkin_: dmj_s76: always a chance
17:56 imirkin_: but ... unlikely
17:56 imirkin_: not a ton of code shared between gp104 and nv4 (riva tnt)
18:54 hakzsam: imirkin_, can you give me your Rb for this one https://lists.freedesktop.org/archives/mesa-dev/2016-July/122764.html ?
18:54 imirkin_: oh ... right. sec.
18:56 imirkin_: wtf. that code used to be there... where did it go?
18:56 hakzsam: nope, it wasn't there actually
18:56 hakzsam: imirkin_, btw, https://people.freedesktop.org/~hakzsam/piglit/
18:57 imirkin_: it was there, then i updated my repo, then it was gone
18:57 imirkin_: looks like it was removed by ed3a284382b194292460a99c0dfe7fd3feccca40
18:58 imirkin_: oh i see. the indirectR was never adjusted.
18:58 hakzsam: yes
18:58 hakzsam: this was buggy for a long time ago :)
18:58 hakzsam: [not related to my refactor patch]
18:58 imirkin_: right
18:59 imirkin_: hakzsam: r-b: me
18:59 hakzsam: thanks
19:01 hakzsam: imirkin_, not sure if you are aware of that, but almost all arb_shader_draw_parameters tests fail on gm107
19:01 imirkin_: i do think i was aware of that.
19:01 hakzsam: okay
19:01 hakzsam: did you investigate?
19:01 imirkin_: happy for you to just "fix" it by disabling if you like
19:01 imirkin_: nope
19:02 imirkin_: i mean, i think i glanced at it, and it wasn't trivially obvious why it didn't work
19:02 hakzsam: does this extension is really used?
19:02 imirkin_: i assume not
19:02 hakzsam: if not, we can disable it on gm107
19:02 imirkin_: go for it
19:03 imirkin_: although whatever it is, should be easy to fix
19:03 hakzsam: I can have a quick look right now
19:03 imirkin_: and *could* be unrelated to shader draw parameters in the first place
19:04 hakzsam: okay
19:06 hakzsam: imirkin_, well, I still need to figure out the last bits of SUATOM, but I'm a bit lazy :)
19:07 imirkin_: i know the feeling
19:07 hakzsam: :)
19:09 hakzsam: ./piglit-run.py -1 --dmesg tests/gpu.py -t arb_shader_draw_parameters gm107-shader_draw_parameters
19:09 hakzsam: [10/10] pass: 1, fail: 9 -
19:09 hakzsam: okay...
19:09 imirkin_: which one passes
19:09 imirkin_: actually, just pastebin the full pass/fail list
19:09 hakzsam: spec/arb_shader_draw_parameters/drawid-indirect-vertexid: pass
19:09 imirkin_: of course.
19:10 imirkin_: and the fails?
19:10 hakzsam: http://hastebin.com/oqoloyofih.sm
19:10 imirkin_: curious
19:10 imirkin_: i disagree with these results.
19:10 imirkin_: go get more more logical results :)
19:10 imirkin_: let me glance at the code again to see if something obvious jumps out
19:11 imirkin_: ohhhhhhhhhhhh
19:11 hakzsam: enjoy! :)
19:11 imirkin_: i think i know what might be going on
19:11 imirkin_: hold on, let me read some code.
19:11 imirkin_: nope.
19:11 hakzsam: no ideas?
19:12 imirkin_: not the thing i was thinking of.
19:12 imirkin_: erm
19:12 hakzsam: okay
19:12 imirkin_: BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
19:12 imirkin_: PUSH_DATA (push, 512);
19:12 imirkin_: didn't we bump that size up a bunch?
19:13 imirkin_: yeah, it's 2048 now
19:13 imirkin_: can you fix that there
19:13 imirkin_: as well as in the mme
19:13 hakzsam: 2048 nox
19:13 hakzsam: *now
19:13 hakzsam: sure
19:13 imirkin_: when you have trouble compiling the mme, let me know
19:15 hakzsam: I think I know how to compile it ;)
19:15 imirkin_: i'm sure you do [think]
19:16 hakzsam: ./piglit-run.py -1 --dmesg tests/gpu.py -t arb_shader_draw_parameters gm107-shader_draw_parameters
19:16 hakzsam: [10/10] pass: 10 -
19:16 imirkin_: yay
19:16 hakzsam: I didn't look at the mme yet
19:16 imirkin_: might not need fixing
19:16 imirkin_: let me check
19:17 imirkin_: gtg
19:17 imirkin_: might not need it
19:17 hakzsam: see you
19:22 karolherbst: hakzsam: hey, wanna come to XDC?
19:23 hakzsam: karolherbst, not sure, but I will try to be there :)
19:24 karolherbst: mhh, k
19:24 karolherbst: mupuf_: seems like the nouveau talk needs to be on the first or second day :p
19:24 karolherbst: hakzsam: well, I guess at XDC are usually fewer people than on fosdem?
19:24 hakzsam: the "nouveau status update" talk?
19:24 hakzsam: sure
19:24 karolherbst: I guess
19:25 hakzsam: less than 100 people
19:25 karolherbst: yeah I know, but I meant more like graphics related
19:25 hakzsam: yeah
19:25 karolherbst: it isn't like entire Fosdem consisted of graphics related people :p
19:25 hakzsam: :)
19:26 karolherbst: though I somehow had the feeling there were hardly 100 graphics related guys at fosdem, but I might be wrong, because I hardly knew anybody back then
19:26 hakzsam: imirkin_, patch sent. I like when fixing bugs takes few minutes :)
19:27 karolherbst: but I have great ideas already for the talk :)
19:27 hakzsam: yep, ton of new things
19:28 pmoreau: hakzsam: Couldn’t it help to have the CB_SIZE as a define, constexpr, whatever, rather than just hardcoded?
19:30 hakzsam: well, maybe
19:31 pmoreau: Didn’t you even add such a define some months ago, while you were fighting magic constants?
19:31 hakzsam: it wasn't the same define
19:31 pmoreau: Ah, ok
19:32 hakzsam: feel free to do it if you think it will improve the thing
19:35 hakzsam: imirkin_, btw, with that fix we don't have any differences with piglit between gk107 and gm107 :)
19:36 hakzsam: [except that gm107 still doesn't have images and tess, but soon]
19:55 hakzsam: pmoreau, ilia did the same suggestion as you ;)
20:14 mupuf_: hakzsam: back home, any wishes for which cards I should plug?
20:24 hakzsam: mupuf_, sure, can you replace the kepler by a tesla?
20:24 hakzsam: I would like to run piglit
20:24 mupuf_: ok, late tesla or early one?
20:25 mupuf_: nv92 is probably the most common tesla
20:25 hakzsam: fine by me
20:25 hakzsam: but all variants are going to be useful :)
20:26 hakzsam: mupuf_, reator is off
20:26 mupuf_: ok
20:26 mupuf_: let me go dig in the pile of GPU
20:27 hakzsam: hehe
20:27 mupuf_: nv92 plugged
20:28 hakzsam: thanks
20:29 hakzsam: mupuf_, [ 76.463263] nouveau 0000:01:00.0: fb: trapped write at 0000102840 on channel -1 [1fea0000 unknown] engine 06 [BAR] client 04 [PFIFO_WRITE] subclient 00 [FB] reason 00000004 [PAGE_READ_ONLY
20:29 hakzsam: outch
20:29 hakzsam: when loading nouveau
20:29 mupuf_: lovely
20:29 hakzsam: probably related to your g92
20:29 imirkin_: hakzsam: grab a glxinfo -l -s from it, if it's a g92
20:29 mupuf_: let's see with the nv96 then
20:29 imirkin_: (or a pre-g200)
20:29 hakzsam: imirkin_, sure :)
20:30 hakzsam: imirkin_, and piglit
20:35 hakzsam: mupuf_, done?
20:35 mupuf_: so0rry ,no
20:35 mupuf_: stop it
20:36 mupuf_:got distracted by EVoC-related business
20:36 hakzsam: okay
20:37 mupuf_: hmm, it is still running, I guess it is failing to stop
20:37 mupuf_: I will force it
20:37 hakzsam: yeah, sometimes I need to force shutdown too
20:38 mupuf_: yeah
20:38 mupuf_: it is booting up
20:38 karolherbst: lalala
20:39 hakzsam: mupuf_, works fine now, thanks
20:41 mupuf_: hakzsam: ok, cool
20:41 mupuf_: I will have to try a different nv92 now, to see if this one is faulty or not
20:41 hakzsam: mmh, I'm going to launch piglit, do you need reator?
20:41 mupuf_: not sure I have another nv92 though :s
20:42 mupuf_: nope, go for it
20:42 hakzsam: cool
20:42 mupuf_: hopefully, I can plug another gpu before going to bed
20:42 mupuf_: and one more when I leave for work
20:42 hakzsam: well, I will be off tomorrow
20:42 hakzsam: mupuf_, anyway, please keep the gm107
20:43 mupuf_: well, tell me what command you used and I can start more runs
20:43 mupuf_: I will be gone this week end too
20:43 hakzsam: me too :)
20:44 hakzsam: imirkin_, http://hastebin.com/doletuxuwo.pl
20:45 imirkin_: thanks
20:45 imirkin_: i'll process it tonight, unless you do it first
20:46 hakzsam: I will let you do the thing
20:46 hakzsam: okay, piglit is running on g96
21:34 karolherbst: wut?
21:34 karolherbst: https://git.kernel.org/cgit/linux/kernel/git/firmware/linux-firmware.git/commit/?id=777480f043ff0d955209e274127bfcd64f6e8edc
21:34 karolherbst: ...
21:35 karolherbst: how fast
21:35 karolherbst: gnurou: what did you do? :D
21:35 karolherbst: ohh wait
21:35 karolherbst: only gp100
21:35 karolherbst: ....
21:35 karolherbst: okay, I am stupid
21:38 mupuf_: karolherbst: HEHE
22:08 t-ask: btw. bugreport done :) https://github.com/iXit/Mesa-3D/issues/224
22:10 mupuf_: t-ask: ?
22:10 mupuf_: the bug should be sent to bugzilla
22:10 mupuf_: not sure who you submited the bug to
22:11 mupuf_: ah, I see, it is Axel's branch
22:11 t-ask: last week we had a chat about those issues and the result was reporting it as mesa bug issue
22:12 imirkin_: well, it sounded like a gallium-nine bug
22:12 imirkin_: so you actually reported it in the right place.
22:13 t-ask: :)
22:14 t-ask: If you remember and you want to add something, feel free :)
22:41 RSpliet: karolherbst: yes I intend to attend
22:42 karolherbst: nice
22:42 t-ask: https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-GP100-Firmware
22:42 karolherbst: so we are three already!
22:43 karolherbst: you, me and pmoreau (intentionally leaving _somebody_ out) :p
22:43 imirkin_: t-ask: what that article doesn't mention is that GP100 is not available to end-users today
22:43 t-ask: :(
22:43 skeggsb_: the article also fucks up and says the kernel work doesn't include graphics, which is lies
22:44 imirkin_: t-ask: on average, there's at least one major factual error in those articles
22:44 imirkin_: and usually at least one significant factual omission
22:44 t-ask: hehe
22:44 dmj_s76: darn, that's not the gp104
22:44 skeggsb_: if he did what journalists usually do, and talk to the people they're talking about before writing shit, he might do some good
22:45 imirkin_: skeggsb_: that'd require effort
22:45 imirkin_: and caring
22:45 karolherbst: well, he talked to me before he did the reclocking benchmarks :D
22:45 skeggsb_: yep
22:45 karolherbst: but that's more like because he didn'T want to mess up and do everything for nothing :p
22:45 karolherbst: anyway, .. gtg
22:47 t-ask: skeggsb_: which journalists are you talking about? *gg
22:47 skeggsb_: t-ask: the decent ones :P
22:47 skeggsb_: not the usual type, of course
22:47 t-ask: well, not easy to find nowadays ;)
22:48 skeggsb_: indeed
22:48 dmj_s76: would the gp104 modesetting be fairly safe to backport to kernel 4.4?
22:49 imirkin_: dmj_s76: do you work on a distro?
22:49 imirkin_: or is this for personal consumption?
22:49 skeggsb_: various bits of preparation work for gp10x (the bits that were relevant on other GPUs too) sneakily went into 4.6/4.7
22:50 skeggsb_: it's hard to say if you'd be successful at applying to 4.4
22:50 imirkin_: simplest thing would be to grab nouveau from master and massage it to compile against 4.4
22:51 imirkin_: which would basically be reverting all the "drm-next" commits
22:51 dmj_s76: imirkin_: Sort of, I started working at System76 recently, so I've been working to make Ubuntu work better on our hardware.
22:51 imirkin_: dmj_s76: ah. that explains your handle :)
22:52 dmj_s76: right now it's a bit of a pain for our customers to do a fresh install with Pascal cards
22:53 imirkin_: dmj_s76: just stop shipping nouveau, should work fine no?
22:53 imirkin_: (or otherwise blacklist it)
22:53 dmj_s76: imirkin_: Actually we ship with nvidia-367 installed
22:54 imirkin_: oh right. you said "fresh install"
22:54 imirkin_: sorry, didn't click
22:54 imirkin_: oh, coz nouveau steals the fb from efifb, and then fails to load?
22:54 dmj_s76: So getting nouveau working enough in the Ubuntu release so users don't have to mess with boot options for the live image / subsequent first boot would be a win.
22:55 imirkin_: that'd actually be the better bug to fix - to either give the fb back to efi (probably not possible), or to make it drop efi later on so that it won't if it'll fail loading
22:56 imirkin_: maybe nouveau should just fail to load on unrecognized pci ids
22:56 imirkin_: (pci id ranges, that is)
22:57 skeggsb_: ah, i've actually got that bug on my list to look into (one day..)
22:58 imirkin_: dmj_s76: that'd be a much simpler "fix" - just patch nouveau to bail before dropping efifb if it's a pci id that's in the GP104 range
22:58 skeggsb_: actually, that should be easy
22:58 imirkin_: iirc the ranges are 0x40-sized for maxwell+
22:58 imirkin_: (or 0x80?)
22:59 skeggsb_: i think i can fix that one easily actually, without resorting to pciids
22:59 skeggsb_: i'll attempt that today
22:59 imirkin_: cool
22:59 imirkin_: obviously not fixable if nouveau fails "later" in the load
23:00 imirkin_: but should be easy to just test mmio 0 for it
23:00 dmj_s76: skeggsb_: That would be great, mind pinging me with the commit when you have it?
23:00 skeggsb_: dmj_s76: yeah, i've already made a note to ping you
23:00 dmj_s76: I can try testing it tomorrow on the 1070 and 1080
23:20 wjx: how can I search through nouveau irc log?
23:20 imirkin_: wjx: use google :)
23:20 imirkin_: (and site:)
23:24 wjx: ok, this works
23:24 imirkin_: airlied: btw, i've made a very half-hearted attempt at implementing int64 on nvc0: https://github.com/imirkin/mesa/commits/int64
23:25 imirkin_: it's basically 90%-non-functional
23:25 imirkin_: although with a small bit of work, should be able to bring that up to 90%-functional :)
23:28 t-ask: some new findings: http://storage9.static.itmages.com/i/16/0711/h_1468279700_6758131_9f56283c7c.png
23:28 imirkin_: t-ask: are you testing the thing i asked?
23:29 t-ask: imirkin_: I tried, but it doesn't start
23:29 imirkin_: weird
23:29 imirkin_: i wonder why not
23:29 t-ask: This is gallium again and it looks like a vertical bit shift
23:30 imirkin_: you mean nine? (it's all gallium....)
23:30 imirkin_: gallium is just an API
23:30 t-ask: might be in horizontal way, too. I can barly see a checkered pattern
23:31 t-ask: yes nine
23:31 imirkin_: bbl