00:26nyef: Stereo on gf119: Success.
00:26nyef: And without even breaking HDMI audio.
00:27nyef: Still to do: Stereo on g94, frame packing, patch series revision.
00:37imirkin: nyef: frame packing seems like it should be #1
00:38imirkin: although i guess probably unrelated to the arch bringup code
00:38imirkin: also, why g94? did you mean g84?
00:38nyef: Sure, the g84 disp, but my card is a g94.
00:39imirkin: ah heh
00:42nyef: Getting all of the arches working is fairly important, as it means that we don't have a loose end in terms of having to detect that such-and-such cards *don't* have working stereo support. Or not doing the detection, and leaving the hole in coverage as a semi-surprise.
00:43urmet: imirkin: hi. I had a problem yesterday with nouveau not loading with my GM108. now I did a looot of kernel compiling today and git bisect tells me that c2a6bbaf0c5f90463a was the commit that broke it.
00:43urmet: just a tiny change in acpi code :/
00:44nyef: Also, I don't quite have an angle on the frame packing, while I do on the rest, and having a v2 round before the end of the month would be good in terms of having a chance to incorporate feedback for a v3 or further before the v12 window opens.
00:44imirkin: urmet: contact the people in question. i haven't the faintest clue wtf any of that stuff is...
00:44nyef: Err... the 4.12 window opens.
00:45imirkin: urmet: file a bug, attach an acpidump, and most importantly, verify that reverting this patch on top of current "fixes" it for you.
00:46urmet: kernel bugzilla?
00:46imirkin: nyef: i don't suppose you've tried tracing the blob while enabling framepacking?
00:46imirkin: urmet: yeah, probably best - that's where rafael hangs out at least
00:47urmet: alright. time for more testing and kernel compiling. woo
00:47imirkin: hakzsam: did you have patches to dump ssbo contents in glretrace?
00:48hakzsam: I had
00:49imirkin: but they're gone?
00:49hakzsam: not on my local machine at least :)
00:49hakzsam: maybe I have a copy on reator, who knows
00:49hakzsam: you need them?
00:50imirkin: trying to figure out why dboyan_'s trace fails on fermi
00:51nyef: I don't even know how to enable framepacking with the blob.
00:52imirkin: nyef: do you know how to set any 3d modes?
00:52hakzsam: imirkin: I can check tomorrow
00:52imirkin: hakzsam: thanks, would appreciate it
00:53nyef: Not with the blob.
00:53nyef: I'm half-considering trying to do the qemu passthrough thing, though.
00:54imirkin: mwk: any clue what CONSTANT vs UNIFORM might be referring to in NV90C0_INVALIDATE_SHADER_CACHES_NO_WFI (ftp://download.nvidia.com/open-gpu-doc/Compute-Class-Methods/1/cl90c0.h)
01:02Lyude: BTW: I won't work on this immediately but it's always nice to have a not-hardware dependent project to work on when I get bored in some places, do we need a better tool for autogenerating register headers for nouveau?
01:02Lyude: when I was working on mesa I noticed the register headers got manually edited a lot
01:04imirkin: Lyude: well, the thing is that someone (mwk) changed some details of the generated headers by rnn, which made them incompatible
01:04imirkin: i did a resync once with sed applied
01:04imirkin: but haven't done it since
01:05imirkin: we could also update a few macros to be compatible with the newly generated headers
01:05imirkin: (basically NV50 -> G80, etc)
01:05imirkin: however the key files get changed so rarely that it's not really worth it
01:06imirkin: anyways, rnn/headergen is plenty good enough afaik
01:06imirkin: robclark didn't like it and wrote a headergen2 in the freedreno envytools clone
01:24nyef: g84/g94 stereo worked first try. I think that my next step is to revise the patch series.
01:27gnarface: after that, see if you can get h264 decoding working on the g92 :D
01:28nyef: I don't have a g92. d-:
01:29nyef: (I was hoping that I would end up with one when I ordered what ended up being my g94, though.)
01:31nyef: Okay, I might have a g92, but if I do, it's in a PPC Mac, and at that is probably in a *dead* PPC Mac.
01:32imirkin: it's just a pcie chip...
01:33imirkin: although it might lack a vbios, which could be inside OF only, in which case you'd be SOL
01:34nyef: Right, IIRC, the VBIOS stashed in OF is typically fairly truncated.
02:06imirkin: that's not a problem - if you can retrieve it, should be enough to post the card...
02:07imirkin: hm. maybe. not sure.
02:18imirkin: gnarface: i wouldn't be surprised if it were something silly, like doing a reclock fixes it
02:18imirkin: gnarface: i did log some of my observations at https://trello.com/c/2BgiX9VW/88-vp2-figure-out-why-g92-hangs
02:18imirkin: unfortunately i have no such gpu, so i can't really poke at it too easily
02:19nyef: I know that there were stray pointers to outside the available VBIOS image, including for things like the connector table.
02:19imirkin: nyef: that's not supposed to happen :)
02:21nyef: I distinctly remember having to hardcode connector layouts for some powermac cards, both nVidia and AMD.
02:22imirkin: so some super-old stuff required various quirks
02:22imirkin: and in fact, some of it never worked with nouveau
02:23nyef: Hrm... GF119 vblank?
02:25imirkin: hmmm... GTS 250 for $15. a little over my budget...
02:28nyef: The G70 or the G96 version of the GTS 250?
02:29imirkin: afaik GTS 250 is only G92
02:29imirkin: ooh. quadro fx3700 for $20...
02:29imirkin: might almost be worth it
02:30nyef: pci.ids says it's for G70, G92, and G96.
02:30imirkin: that's gotta be wrong
02:31imirkin: i think they were *mostly* G92's
02:31imirkin: either way, the one on ebay is definitely a G92
02:31imirkin: coz it's this one: 3842 1150 GeForce GTS 250 P/N 512-P3-1150-TR
02:32nyef: Fair enough.
02:37imirkin: all yours if you want it :)
02:38imirkin: no hdmi though =/
02:40nyef: Yeah, I think I'll pass, thanks. I have more than enough projects on my queue already. (-:
02:41imirkin: i put in a bid on the fx3700. we'll see, doubt i'll end up getting it.
02:42imirkin: (moral of the story: don't be like me and spend all your allowance on gpu's you have no possible need for.)
02:50nyef: Right. I'll spend it on other obscure obsolete computing hardware! (-:
03:33gnarface: imirkin: i'd send you this one, but i'm pretty sure it would cost more than 15$ for shipping
03:34imirkin: yeah, not a good use of time/money/effort
03:35gnarface: if you ever get bored enough, or if it does become worth the time and effort, i can arrange remote access for you to a system with it installed. in theory i could even set up a crappy low-end webcam to stream a view of the screen for you too
03:36gnarface: i know you turned that offer down last time, but it still stands, just so you know.
03:36imirkin: i think i've had close to that when playing with it with AndrewR's G92
03:36imirkin: and i was unsuccessful
03:36imirkin: i think having it local would make a difference. i guess i dunno.
03:37gnarface: it does make noises when it starts up...
03:37gnarface: the caps are real loud actually
03:37gnarface: you might be able to figure something out if you had some way of imaging the changes in the RF emissions
03:38gnarface: but it doesn't seem to do anything at all when the video fails
03:38imirkin: there is a chance that the issue is something REALLY dumb, like not setting the right memtype somewhere
03:38imirkin: yeah, the vdec engine hangs
03:38gnarface: anyway, it's a bummer
03:38imirkin: the xtensa rtos is fine, but the underlying hwdec engine never responds
03:39imirkin: mwk: is there a way to tell whether the underlying decoding engines are running from mmio?
03:39imirkin: [context: vp2]
03:39imirkin: i guess i could write my own little rtos... seems complicated...
03:40imirkin: at least there's gcc support
04:30Horizon_Brave: evening everyone
04:34Horizon_Brave: does the tool... mmiotrace mean anything to you guys here?
04:36airlied: it's the tool used to trace the binary driver
07:06mwk: imirkin: most of them are visible from MMIO, so yes
07:07mwk: should probably be enough to scan 0xf000, if it's writable you're probably fine
10:54karolherbst: wuhu, NetBSD 7.1 has initial Nouveau support
11:46karolherbst: imirkin: next patch of hitman should contain some fixes regarding nouveau, I guess we won't need to overwrite GL_VENDOR anymore then
11:49xen: hi there
11:49xen: i'm considering an aquisition of a 1070
11:50xen: any chance it will work with nouveau for day2day basic tasks (VM management, 2D video streaming/editing) ?
11:52karolherbst: xen: with 4.12? I guess there will be accell support
11:52karolherbst: before that only driving a display should work
11:53dboyan: karolherbst: Do you have any idea about "unable to copy vbios" under mmiotrace? now it happens on both of my machines.
11:54karolherbst: no, sorry
11:54karolherbst: you could try to debug the mmiotracer... but this is just painful
11:55dboyan: well, I guess I won't have spare time enough for kernel space debugging
11:56dboyan: btw, I do have a mmiotrace captured on my gk208 some time ago, if you want it, i'll upload it somewhere later today
11:56xen: karolherbst: well, i actually have 6 displays to drive, but if it can do that, i would still be happy
11:58xen: currently on 4.10.2, there are some issues with the offload rendering setup i have, but had no time to investigate it yet
13:42pq: dboyan, any error messages from mmiotrace itself in the kernel log?
13:43pq: dboyan, or UNKNOWN entries in the mmiotrace output?
13:57dboyan: pq: I haven't seen either of them.
13:57dboyan: But luckily, mmiotrace somehow succeeded again once on my gk208 laptop, so I may get something karolherbst needs.
13:57pq: too bad, no obvious clue to investigate then
14:00karolherbst: I think we need to write something from scratch or so
14:00karolherbst: we already have the problem regarding the repeat instructions
14:12dboyan: karolherbst_: https://people.freedesktop.org/~dboyan/mmio-gk208.log.xz
14:13dboyan: I have seen changes between 3 power states in nvidia-settings during trace
20:58kattana: I got disconnected but not before reading something truly disturbing
20:58kattana: virgl is not 100% compatible with nouveau?
20:58kattana: that was my last gaming option :(
21:00Lyude: imirkin: other then nvc0_3d.xml.h, what other files do I need to modify to actually get a register added? I added an entry for NVC0_3D_FILL_RECTANGLE but then I get undefined errors for NVC0_3D_0x0000113c
21:00imirkin_: sounds like you made a mistake
21:01imirkin_: pastebin git diff
21:01imirkin_: (ideally just the nouveau bits of that...)
21:01Lyude: imirkin_: https://paste.fedoraproject.org/paste/HuWghluecSAZajpCk4~gXF5M1UNdIGYhyRLivL9gydE=
21:02imirkin_: s/NVC0_3D_// in nvc0_state.c
21:02imirkin_: also don't forget to bump the state object size by 1
21:02Lyude: yep, already did
21:02imirkin_: in nvc0_state.h or nvc0_stateobj.h
21:02imirkin_: ok cool. it's an easy bit to miss :)
21:12Lyude: imirkin_: btw, how would I go about running a piglit test for this?
21:12imirkin_: [not sure i understand the question... i assume you know how to execute binaries from a shell though]
21:14Lyude: hold on
21:15imirkin_:is holding on... but fingers are starting to slip
21:16mupuf: ah ah
21:16Lyude: imirkin_: https://github.com/Lyude/mesa/tree/wip/NV_fill_rectangle basically I've got this so far, and I'm trying to figure out how I test it to see if it works orn ot
21:16Lyude: *or not
21:16imirkin_: Lyude: write a test, then see if it passes
21:16imirkin_: it's great - you get to basically decree how a correct implementation works by writing both the test and the implementation
21:17imirkin_: and don't let the fact that you know nothing about GL stop you -- i never did
21:17Lyude: things would be much less fun if I did :)
21:17imirkin_: i like to think i've picked some up over the years though
21:27Lyude: imirkin_: for that EXT_polygon_offset_clamp patch series did you end up writing a piglit test for that? wondering if I can see an example
21:49imirkin_: Lyude: sure
21:49imirkin_: just ... don't try to understand it. it'll make your head hurt.
21:49imirkin_: Lyude: https://cgit.freedesktop.org/piglit/tree/tests/spec/ext_polygon_offset_clamp
21:53imirkin_: all that depth stuff... very confusing. i still barely understand it.
22:00imirkin_: Lyude: you don't have to do anything fancy... just make sure that flipping the functionality on does the right thing
22:01imirkin_: is this core-only functionality? if not, good to make sure that the displaylist bits are all hooked up properly
22:01imirkin_: and that it's in the right attrib group
22:16BlubberBop: Hi there. I have a desktop with 3 ASUS VGA cards (VGA compatible controller: NVIDIA Corporation GT218 [GeForce 210] (rev a2)) 3 monitors, and Mint 18.1. If I try to select the nvidia 340 driver, my system won't boot anymore (hangs on the drive encryption password, which appears in 1024x786 or something), it seems I can only use nouvou driver.
22:16BlubberBop: When I use that drive, then at boot and drive encryption password I do see the three monitors working, but as soon as the sddm login comes up, only one monitor is working. after login I run xrandr and it only recognizes one card, though lspci shows the 3 cards.. Any idea how I can get the system to use the 3 monitors?
22:17BlubberBop: Also, some further investigation in xorg log shows "/dev/dri/card1: failed to set DRM interface version 1.4: Permission denied" and same for card2.. Google wasn't too helpful yet with the exception of some specific projects that were not X.. Anybody here who might have an idea?
22:17BlubberBop: And, does nouveau actually support multiple monitors at all?
22:17imirkin_: BlubberBop: you want https://nouveau.freedesktop.org/wiki/Optimus/ -- "Using outputs on discrete GPU"
22:18imirkin_: i would, however, recommend using both outputs on one GT218 and the one more output on a second GT218
22:18imirkin_: rather than sticking one monitor on each one
22:18imirkin_: perhaps that's not an option for you, but if it is, then you may find that things work better
22:30BlubberBop: imirkin_: I currently have the 3 video cards. They have dual output, but only 1 HDMI connector and all monitors are HDMI only
22:31BlubberBop: So I have to stick with that setup for the moment, but I'll check the optimus page, I hope that will help fix this..
22:31imirkin_: you can get DVI <-> HDMI adapters
22:32imirkin_: another avenue for this, btw, is to use Xinerama, which is the pre-xrandr solution to it. works well but loses you direct rendering, which a lot of applications rely on these days
22:32BlubberBop: Is there any outlook on dual monitor setups permanently working "normally" on Linux? NoFi! Just wondering, because I've been fighting with multi monitors for over a decade by now, and I've seen little to no improvement :(
22:32imirkin_: the issue you're facing isn't dual-monitor, it's dual-gpu
22:32imirkin_: (well, triple-gpu)
22:32imirkin_: this complicates matters greatly
22:33imirkin_: it'd Just Work (tm) if all the outputs were on the same GPU
22:33BlubberBop: imirkin_: well, to rephrase my question then.. Will multi GPU ever be solved at all? at least on nvidia side?
22:34imirkin_: erm... well... there's *a* solution, which is the current one
22:34BlubberBop: Optimus, you mean?
22:34imirkin_: that's just the name of the wiki page
22:34imirkin_: and gpu's marketed as optimus do use that approach as they're multi-gpu setups
22:34BlubberBop: imirkin_: well yeah, that is what I meant :)
22:34imirkin_: but it's generic to any multi-gpu setup
22:35BlubberBop: I'll definitely check it out!
22:35karolherbst: does this look right? slct_ne(b, c, set_ne(a, 0)) -> slct_ne(b, c, a)?
22:35BlubberBop: Thanks for the help so far!
22:35imirkin_: karolherbst: yes, it *looks* right...
22:35imirkin_: karolherbst: but looks can be deceiving ;)
22:36imirkin_: BlubberBop: the xinerama approach is documented here - https://nouveau.freedesktop.org/wiki/MultiMonitorDesktop/
22:37imirkin_: ugh. all the text on all those pages is so wrong. feels like we should just nuke it all and start over =/
22:37karolherbst: imirkin_: should I put an opt like this also where I put the other one?
22:37karolherbst: ohh wait, this had 2 imms
22:37karolherbst: this is somethign else here
22:38imirkin_: karolherbst: this is a AlgebraicOpt imo. the fact that it's comparing against 0 is slightly worrying, but i doubt there will be lots of cases where constfolding will change the outcome
22:38karolherbst: it just modifies the slct anyway
22:38imirkin_: also fact is that it triggers on the SLCT op, not on the SET op which has the imm
22:39karolherbst: and the set may be DCEed away
22:39imirkin_: which would make it inconvenient to stuff into ConstantFolding
22:39imirkin_: (you could look at uses, etc. but that's annoying.)
22:42karolherbst: there is a handleSLCT already, let's check what it does
22:42imirkin_: not much.
22:42imirkin_: if a == b
22:44karolherbst: it also does slct->cc(c, 0)
22:44karolherbst: that looks odd
22:45karolherbst: wouldn't that be more like constantFolding?
22:45karolherbst: if it already evalutes the immediate in c with the CC against 0?
22:45imirkin_: and and b may not be constant.
22:45imirkin_: a and b ...
22:46karolherbst: the other way around is missing
22:52karolherbst: imirkin_: interesting, it seems like that for newer games, the higher you pump the settings, the more load you get on the shader cores and memory bandwith is getting less important. At least this seems like the common situation with modern games today
22:52imirkin_: makes some sense... post-processing and whatnot
22:56BlubberBop: imirkin_: Looks like the geforce 210 (chip on my cards) are not even supported by nvidia driver.. yay! I guess that explains why boot fails with the nvidia drivers..
23:02imirkin_: BlubberBop: 340.x should support your GPUs afaik. but we tend not to play the nvidia support game here - you have problems with their driver, complain to them.
23:03BlubberBop: imirkin_: I'm not playing anything here, just saw that info on the nvidia driver page, is all. Sorry if that was insinuated
23:03BlubberBop: Let me ask this differently.. I need a 3 monitor work station that supports basic 3D graphics (including desktop 3D effects), nothing fancy, just enough that it's 3D. I need a *quiet* work station, not noisy ventilators. I have 3 HDMI monitors, I have a motherboard with integrated HDMI (ASUS Maximus VIII) and currently 3 cheap (but ventilatorless) ASUS geforce 210 cards.. What do I need to do to get 3 working monitors?
23:04BlubberBop: For example, would I need to buy a dual HDMI out and combine that with the integrated Intel? would that work?
23:04imirkin_: your setup should work as-is
23:05imirkin_: it's unfortunate that nvidia doesn't seem to produce any low-end gpu's with 3 digital outputs... lots with 2 and 1 vga
23:06imirkin_: and obviously higher-end gpus have this, but they're never fanless
23:06BlubberBop: imirkin_: Well okay, that is one good thing then :) But I'm still going over https://nouveau.freedesktop.org/wiki/Optimus/ but I havent been able to make much heads or tails of it.. I need to run "An updated graphic stack (Kernel, xserver and mesa), KMS drivers for both GPUs loaded, and DDX drivers for both GPUs loaded."
23:06imirkin_: BlubberBop: are you at the workstation in question? if so, pastebin your xorg log.
23:07BlubberBop: Now, I'm quite comfortable with Linux (mostly ubuntu) but I don't know what KMS and DDX driuvers are.. I thought KMS was for virtualization..
23:07BlubberBop: imirkin_: sure, hold on
23:07imirkin_: that's KSM - kernel-shared memory
23:07imirkin_: KMS = kernel modesetting
23:07BlubberBop: and yeah, no idea why they don't do more HDMI only since most monitors these days seem to be just HDMI only as well
23:08BlubberBop: imirkin_: ah, there you go, my bad :)
23:08BlubberBop: imirkin_: Any preferred pastebin site?
23:08BlubberBop: Depending on IRC channel, people want me to use a different one..
23:09BlubberBop: If not, http://pastebin.com/1UNLg7hW
23:09BlubberBop: imirkin_: ^^^^ There is the pastebin
23:10imirkin_: i don't care, as long as i can access it without logging into some dumb thing
23:10BlubberBop: imirkin_: As you can see in there, "/dev/dri/card1: failed to set DRM interface version 1.4: Permission denied", same for /dev/dri/card2... I'm not 100% sure, but I do think that there is my problem
23:10imirkin_: ya hold on
23:10BlubberBop: imirkin_: no login required
23:10imirkin_: so that's problem #1
23:10imirkin_: ls -l /dev/dri
23:11BlubberBop: imirkin_: http://pastebin.com/p4JxBsKK
23:12BlubberBop: I think I see 3 files each for 3 cards there.. I have no idea though if this is correct or not, this is beyond my linux gfx knowledge
23:13imirkin_: (yes, that is correct)
23:13imirkin_: oh, is this systemd nonsense?
23:13BlubberBop: imirkin_: oh crap
23:13BlubberBop: imirkin_: yeah it is... the nightmares had just stopped, please tell me they're not coming back
23:14imirkin_: i don't do systemd... check with a distro support channel to see how to operate i
23:14BlubberBop: imirkin_: Is this a systemd issue, then?
23:14imirkin_: i don't know
23:14imirkin_: but my suspicion is that for some reason it can't open those files
23:14imirkin_: i have no idea why that might be.
23:15imirkin_: however i know that systemd is nuts and impossible to use, so that's always a rich source of fail
23:16BlubberBop: imirkin_: Yeah, I had to reinstall this machine nearly a dozen times thanks to systemd, and got banned from #ubuntu for saying that systemd caused more problems than it resolved
23:17imirkin_: i thought the problem it solved was that things were too simple
23:17imirkin_: it does that quite admirably
23:17imirkin_: either way, much as i don't do nvidia support, i also don't do distro support
23:17imirkin_: you either know how to operate your distro, or you know how to get help operating your distro from someone who is not me.
23:17BlubberBop: imirkin_: Agreed! (please don't ban me, systemd rocks! its the best!)
23:18BlubberBop: imirkin_: Yeah, I'll check the ubuntu channels again.. sigh.. but so, in your opinion.. IF I were to fix this access issue, then in theory, things should work?
23:18imirkin_: things may not work *out of the box*, but you'd be able to link up the GPU's and have all the screens available in xrandr to configure as you see fit
23:19imirkin_: another favorite for weirdo failures like this is selinux
23:19BlubberBop: imirkin_: Oh if I get that up, I'd be so happy already.. worst case scenario, I'd have a boot script to fix that for me..
23:20imirkin_: some DE's auto-link-stuff-up, others don't
23:20BlubberBop: imirkin_: yeah, I didn't wanted to mention it, but I was thinking selinux also might be something.. Though I haven;'t seen or used selinux in years, and I'm not even sure if this machnie has it running.. I'll ckechk..
23:22karolherbst: imirkin_: total instructions in shared programs : 517683 -> 514168 (-0.68%) for hitman
23:22karolherbst: and I only detect 0 on src1 and just the ne/ne and eq/eq pairs
23:23karolherbst: but this is also nice: total local used in shared programs : 6881 -> 6865 (-0.23%)
23:23imirkin_: any tangible improvement in perf?
23:23karolherbst: I will check perf if I am at -10%
23:23karolherbst: or so
23:23karolherbst: otherwise I fear that I won't see much difference
23:25karolherbst: newest version: https://gist.github.com/karolherbst/fbfa391895cc8de8bc79541a6846e868
23:25karolherbst: 3554: slct u32 $r2 eq $r63 0xffffffff $r2
23:26karolherbst: this could be a set I guess
23:26karolherbst: ohh wait
23:26karolherbst: I messed up
23:26imirkin_: but it wouldn't gain that much
23:26karolherbst: it was "3568: set u32 $r2 ne $r2 $r63" before
23:29Lyude: imirkin_: guessing I'm going to have to figure out how to write a shader for this? (assuming that's what piglit_build_simple_program() does)
23:29imirkin_: i was kinda recommending you teach shader_runner.c about it
23:29imirkin_: and then write a shader_test
23:29imirkin_: since it's just a glEnable()
23:30karolherbst: okay, I think I know how I killed that opt
23:30imirkin_: for which there's already support, you just have ot add it to the list
23:30karolherbst: the slct turned out to be useless, so the defs were rewritten and the slct removed
23:30karolherbst: but now, because I move the set source in, this detection doesn't work anymore
23:31karolherbst: or something like this
23:51karolherbst: with all new opts: "total instructions in shared programs : 520322 -> 514004 (-1.21%)" for hitman :)