04:34martm: i mean i just minorly looked at mutli2sim now too, to find some more blocking instructions, it can provide clues how kepler isa works for setting more masks for instructions
04:35martm: MichaelLong: in that case just dig more into multi2sim/trunk/src/arch/kepler/emulator/ThreadIsa.cc
06:18wvvu: imirkin: should I try again with the debugging of crazy error?
07:22bkeys: Hello I would like to contribute to Nouveau by assisting the documentation, specifically on the wiki. Could someone tell me where this is being hosted and how I should start?
07:35imirkin_: bkeys: if you edit a page, it should give you instructions on how to create an account iirc
07:36imirkin_: bkeys: is there a particular section you feel needs improving?
07:42bkeys: Which website is it hosted on? I see some documentation on github, but the documentation on there worded it like cleaning up documentation would be a good way to start contributing
07:42imirkin_: oh i see
07:43imirkin_: i thought you were talking about nouveau.freedesktop.org
07:43imirkin_: if you're talking about the hwdocs, those are at https://github.com/envytools/envytools/tree/master/docs
07:43imirkin_: these are then "compiled" and made available on envytools.rtfd.org
07:44bkeys: I just want a way to start contributing to this project; I have some background in OpenGL but I would like to get as familiar with Nouveau as soon as I can because I would want to help with Wayland support and especially Vulkan since that is not far away
07:44imirkin_: bkeys: what GPUs do you have available to you?
07:45Ycarene: Is it worth it to try gallium nine?
07:45imirkin_: if you're looking to do code contributions, i keep a trello at https://trello.com/b/ZudRDiTL/nouveau
07:45bkeys: Only an intel at the moment; but I will soon have my Nvidia 650 Ti boost available to me
07:45imirkin_: Ycarene: i've heard it can be quite a bit faster than wine + GL for d3d9 games.
07:45imirkin_: bkeys: ok cool, that should be a kepler chip. well supported.
07:46Ycarene: Faster than the proprietary drivers?
07:46imirkin_: Ycarene: if you have concrete questions about it, you may want to ask on #d3d9
07:46imirkin_: Ycarene: mmmm ... dunno. if nouveau supports reclocking on your gpu, maybe?
07:46imirkin_: what gpu do you have?
07:47Ycarene: It's a gt216
07:47imirkin_: Ycarene: ah ok. reclocking *ought* to be supported on it
07:48mwk: ah, Tesla
07:48imirkin_: although there have been some reports of failure
07:48mwk: an elegant GPU... for a more civilized age
07:48imirkin_: (esp on the GT216 i'm afraid)
07:48Ycarene: mwk - lol
07:48imirkin_: Ycarene: boot with nouveau.pstate=1 (no a 3.19+ kernel)
07:49imirkin_: and you should be able to switch between power states by echo'ing stuff into /sys/class/drm/card0/device/pstate
07:49mwk: imirkin_: you wouldn't happen to know anything about GK210, would you?
07:50imirkin_: mwk: i assumed it was a marketing stunt around GK110
07:50mwk: imirkin_: I thought so, but then I read the specs
07:50imirkin_: mwk: however my highest end chip is a GF108, so... i don't know too much about it :)
07:50imirkin_: [actually i have a GK208 now too, at work]
07:50mwk: but the MP dimensions are so different, it has to be a distinct chip
07:50mwk: it has 2× the amount of GPRs
07:50imirkin_: yeah, someone pointed that out to me too
07:50mwk: and shared memory, for that matter
07:51imirkin_: could be that they fused it off on "plain" GK110? dunno
07:51mwk: also compute capability 3.7 as opposed to 3.5, but that may not mean much
07:51mwk: doubtful, GK210 seems to be a very recent development
07:51imirkin_: oh? i think i heard about it a while ago
07:51mwk: why wouldn't they sell the extra fun before
07:52imirkin_: yield issues? who knows
07:52imirkin_: anyways, you could be righ
07:52imirkin_: but i think it's in the same pci id range...
07:53karlmag: not listed in the codenames list
07:54imirkin_: the codenames list is hardly authoritative :)
07:54karlmag: imirkin_: I know, but it should be... well, ideally I mean ;-)
07:55mwk: huh, same pci id range indeed
07:55karlmag: it's my go-to list
07:56karlmag: I guess we have at the very least 980M and and gk210 lacking from the list then
07:57mwk: I'll throw in GK210 just in case
07:57mwk: the pci id range already has NVF0 and NVF1, so...
07:58karlmag: mwk: cool :)
08:34karlmag: just to ask a possibly stupid question; the code names.. e.g "nv84" where do they come from? There must be an origin for them I imagine, but a bit of googling doesn't really turn anything up on that. It looks like it might have been a kind of system that got "bastardized " over time, but what do I know..
08:35imirkin_: karlmag: it's NV + the chipset code from mmio register 0
08:35karlmag: Case in point is that GK210.. How to find out which NVxxx it should have?
08:35karlmag: ah, right
08:35karlmag: so one actually must have the hardware to find out
08:35imirkin_: you can't, at least not without having one
08:35imirkin_: and GKxxx doesn't always map to a single thing
08:35karlmag: no kidding :-P
08:35imirkin_: e.g. GK208 might be NV108 or NV106
08:36imirkin_: GK110 might be NVF1 or NVF0
08:36karlmag: I know...
08:36imirkin_: (i believe those are the only examples)
08:36karlmag: I've gone through the whole CodeNames page, so I have noticed.
08:37imirkin_: the chipset is available since NV10 - https://github.com/envytools/envytools/blob/master/rnndb/bus/pmc.xml#L41
08:38karlmag:checks prices for the k80
08:38karlmag:decides he needs those months salaries for other things.. like food.. :-P
08:39karlmag: and power for his computers..
08:39imirkin_: only $4110 on amazon...
08:39karlmag: uhm.. yeah...
08:41karlmag: When hardware cost that much I would rather get it locally, even though that would most likely mean it being even more expensive.
08:42karlmag: We have very good warrenty and customer protection laws in this country.
08:42imirkin_: i doubt the average corner store keeps such items in stock :)
08:42karlmag: hehe.. there is that
08:45karlmag: Sounds like something noone would really want to keep in stock for too long.
08:45karlmag: just way too pricy
08:48imirkin_: and not a lot of people walking in off the street to get out of the cold and making an impulse K80 purchase
08:56karlmag: imirkin_: I guess there is that :-P I might.. If I had more $$$ than I knew what to do with, but I wouldn't be alone in that situation I think. :)
09:01Ycarene: so do I put options nouveau pstate=1 in my modprobe.conf?
09:01imirkin_: Ycarene: that should work
09:03imirkin_: Ycarene: assuming that nouveau is not loaded from initrd, where that modprobe.conf isn't available
09:03imirkin_: (or if you make it available in the initrd)
09:19bkeys: Sorry if I missed anything; but I missed how the best way for me to start contributing is?
09:21imirkin_: bkeys: are you looking to make docs contributions, or code contributions?
09:22bkeys: I am willing to do either; I just figured starting by doc contributions would make me more familiar and then ease my way into code contributions
09:22imirkin_: bkeys: well, you can have a look at https://trello.com/b/ZudRDiTL/nouveau where i keep a bunch of tasks
09:24imirkin_: bkeys: my experience is that it's easiest to get started in code by fixing some bugs
09:24imirkin_: unfortunately there aren't a whole lot of easy ones
09:25bkeys: This seems a bit overwhelming for a newbie; I am hoping/implying that we are using git?
09:27imirkin_: bkeys: mesa repo is at http://cgit.freedesktop.org/mesa/mesa
09:27imirkin_: (well, that's the web endpoint, but there are links to the git repo as well)
09:27bkeys: Alright; now is there a way at all for me to test my changes? I do not have my nvidia card available to me at the moment
09:27imirkin_: is there a particular feature you're interested in?
09:27imirkin_: not if you're making nouveau-specific changes
09:28bkeys: I would really like to contribute to implementing Vulkan in nouveau
09:28imirkin_: we don't have a simulator :)
09:28imirkin_: ah ok
09:28imirkin_: well the specs aren't out yet afaik
09:28imirkin_: i believe pmoreau has started working on a SPIR-V adapter -> nv50 ir
09:28imirkin_: none of that requires hardware though
09:29imirkin_: which is for the shader compiler aspect of all this
09:30bkeys: You said this does not require nvidia hardware?
09:30imirkin_: the shader compiler bits? definitely not
09:31imirkin_: it's all pretty standalone stuff
09:31bkeys: Are the people working on that in this IRC room? I am probably going to need some mentoring since I have never written a driver before
09:32imirkin_: afaik no one is working on vulkan yet, since there are no specs to work off of
09:33imirkin_: however the shader language of vulkan has been released
09:33imirkin_: and pmoreau has started on an adapter between SPIR-V and nv50 ir
09:33imirkin_: [and he is here as you can probably see in the member list]
09:33wvvu: imirkin_: should I continue debugging the problem?
09:33wvvu: should I use gdb as well?
09:34imirkin_: wvvu: up to you, but i don't know what's going that causes tracing to so badly fail for you
09:34wvvu: I am ready to whatever it takes.
09:35wvvu: it's alright, I know is just one goddamn emulator. dolphin game cube/wii games play nice, playstation 1 (psx), works nice. Is just the ppsspp.
09:36wvvu: imirkin_: what's the lowest level of debugging that exist in the universe?
09:36wvvu: a logic analyzer?
09:36imirkin_: electron scanning microscoped? :)
09:37bkeys: Is pmoreau a seperate project from nouveau
09:37imirkin_: pmoreau is a person
09:37aaronp: wvvu: http://blogs.nvidia.com/blog/2011/05/16/sneak-peak-inside-nvidia-emulation-lab/
09:37imirkin_: (at least i think he is)
09:38wvvu: imirkin_: oh man, electron to analyze electrons??
09:38mwk: imirkin_: wrt GK110 mapping to two codenames... there are also reverse examples, eg. MCP77/MCP78 are both NVAA
09:39imirkin_: mwk: ah yeah :)
09:39mwk: no idea wtf the difference between MCP77 and MCP78 is anyhow, seem to be the same chipset
09:39mwk: and there's a very... unclear situation with NV41/NV42 and NV40/NV45
09:39wvvu: imirkin_: woah!! what about tools for mortals?
09:39mwk: oh, and NV38 and NV35 are both NV35
09:39imirkin_: mwk: but that's just marketing right?
09:40mwk: *shrug* isn't GK210 just marketing?
09:40wvvu: on a more serious note, I tried with /dev/dri/card0, should I try with /dev/dri/xxxxD128?
09:40imirkin_: mwk: might be :)
09:40mwk: or GK110 for that matter?
09:40imirkin_: mwk: well, the point is that it's a diff chip
09:40wvvu: I need some guidance
09:42mwk: yeah... anyhow, it's a mess
10:17bkeys: So how do I contact pmoreau? I know he is here in the IRC but they appear to be afk
10:18imirkin_: he'll probably be around later, he's probably traveling back from FOSDEM right now. you could probably also track down his contact info from mailing list or git commit activity
10:19orbea: bkeys: msgserv?
10:21bkeys: Where would the git repo be for the SPIR V compiler?
10:24imirkin_: i believe pmoreau has been working on it here: https://phabricator.pmoreau.org/diffusion/MESA/history/spirv_1.0/
10:31martm: nvidia has some sort of tool that tells when suboptimal number of threads were launched
10:39martm: i think that would be easiest, though, i'd also know probably how to write one
10:42pmoreau: bkeys: Right, I was traveling back from FOSDEM, but now I'm home again (through, I'm having dinner right now, so…)
10:43bkeys: Alright; I sent you an email a few minutes ago
10:45pmoreau: I'll have a look in ~10mins. :-)
10:51martm: maybe would take a week to handle all the instructions that way, then make database, for all of them, only that branch prediction is sophisticated issue
10:55martm: i don't think there should be a problem though to get the perf, probably enough easy but (just some hair will fall) to the presiction as least as good as nvidia
10:57martm: though as a second thought, now that i remember interrupt would solve it anyways
11:13martm: so nouveau will perform well in the future, i see with correct method, is kinda would give false positive sometimes, are bank conflicts in memory, and possible unpredicted cache hits or misses
11:14martm: imirkin: how is that nv/ext_shader_load_store going, ask for help if some trace is badly needed for instance for kepler gt730
11:15martm: images was also somewhere in that ext stuff, so prolly it was ext_shader_image_load_store
11:18martm: ah heck, i leave, now that i was thinking basically intelligent code will handle the unpredicted bank conflicts and caches too
11:44pmoreau: bkeys: Sorry, took some more time than what I though: many things to catch up.
11:46pmoreau: So, as imirkin_ pointed out, my current work is here: https://phabricator.pmoreau.org/diffusion/MESA/repository/spirv_1.0/ . I have a wiki page explaining how to setup everything to have it running: https://phabricator.pmoreau.org/w/mesa/testing_opencl_through_spirv/
11:47pmoreau: The SPIR-V documentation can be found here: https://www.khronos.org/registry/spir-v/specs/1.0/SPIRV.html, and of course, if you have any questions: just ping me here or send an email
11:48pmoreau: bkeys: Do you have an NVIDIA card BTW?
12:46bkeys: I do not have one available to me right now but I will in the short future
12:48pmoreau: Ok. It's helpful to have some hardware at hand, since the code generated for Fermi and Kepler looks quite correct, but when running it, I don't get the results I expect.
12:57bkeys: What kind of hardware are you thinking about?
13:02pmoreau: Any NVIDIA GPU, to check that the code actually produces what you expect it to do.
13:02pmoreau: It could be there are still some pieces missing in the compute setup: some buffers not properly bound, or whatever.
13:09bkeys: I see, I have to get my workstation so I will have access to it in a week maybe
13:10imirkin_: you can still work on stuff, it can just be difficult to tell if your changes are correct
13:14pmoreau: You can already start poking at the code a bit
14:34Gizmokid2005: Hey all. I just upgraded my laptop from Fedora 21 to 22 via dnf's upgrade and now after I login I'm greeted with a black screen with only a cursor. I've managed to get the Xorg log here: http://ur1.ca/ogxup -- Any suggestions?
14:35imirkin_: i could be misreading
14:35imirkin_: but i think you want #intel-gfx
14:35Gizmokid2005: Hrm...that's possible. It is a dual-card laptop with a dedicated nvidia on board as well that could be being ignored...
14:35imirkin_: to be double-sure, boot with nouveau.modeset=0
14:35imirkin_: which will totally entirely disable nouveau
14:36imirkin_: that said, you do have some issues with nouveau that should be resolved with kernel 4.4 or so
14:36imirkin_: but i don't think those issues have anything to do with the blank laptop panel
14:36imirkin_: as that's connected to the intel gpu
14:37Gizmokid2005: Well that's good to know :) Ever since a bump of kernel 3.x my machine has lost the ability to successfully shutdown/reboot the laptop...so I'm sure there's other issues at hand. I'm really just trying to get into it at this point to make sure I've got nothing useful left before starting fresh.
14:37imirkin_: ok well that issue could be connected to nouveau
14:37imirkin_: your gpu only half-comes up
14:38imirkin_: and that could pose problems at suspend time
14:38imirkin_: should be better in kernel 4.4
14:38Gizmokid2005: Good to know. I don't actually recall any nouveau issues before, but that could just be ignorance.
14:38imirkin_: but i don't think it's connected to your other issues
14:38imirkin_: well, it's only a secondary gpu
14:38imirkin_: so unless you go out of your way, it doesn't get used
14:41mwk: seriously... someone bothered to report a resource leak in nva
14:41mwk: in the startup routine, no less
14:42imirkin_: doesn't hurt to fix
14:42Gizmokid2005: Thanks imirkin_, much obliged. Looks like a nouveau.modeset=0 doesn't actually change behavior
14:42imirkin_: Gizmokid2005: ok, so you can be sure that nouveau has no effect on your problems. that said iirc 4.2.x was a problem release for intel
14:42mwk: yeah, well, it's a trivial one-liner
14:42imirkin_: Gizmokid2005: really just upgrade to 4.4 :)
14:42mwk: still, a total waste of time
14:43Gizmokid2005: imirkin: I'm working on it :) That was part of the reason I was upgrading from F21, to get back on current release
14:46imirkin_: Gizmokid2005: dunno if you do gpu-intensive things, but if you do, come back when you get it figured out and we can help you start using the nvidia gpu
14:46Gizmokid2005: at some point that would be nice, but definitely a bit lower priority than getting things functional again :) I really appreciate it.
15:07ravior: imirkin_: Is there a way to get more useful debug informations for the acceleration lock-up issue?
15:08imirkin_: ravior: can you point me to your bug again? i've forgotten the details by now.
15:13ravior: imirkin_: https://bugs.freedesktop.org/show_bug.cgi?id=71659
15:13ravior: If I can help with any debugging informations, or anything that could shed light on the issue, just let me know.
15:14imirkin_: ravior: this was with kernel 4.2.3 right? have you tried updating? iirc a fix of mine went into 4.2.something > 3
15:15ravior: imirkin_: Yes. I've updated to the latest stable kernel (4.4.0) and the issue is still reproducing.
15:15imirkin_: ok can you include a kernel log from that?
15:16ravior: Sure thing. Do you want some exact cmdline arguments for the nouveau driver?
15:16imirkin_: can't think of any that'd be too useful =/
15:16imirkin_: but the full log from boot until lockup would be great
15:18ravior: Very well. I'll get that and update the defect log as soon as possible.
15:19imirkin_: but... it's a weird issue
15:19imirkin_: and i think it's out of my knowledge area
15:20imirkin_: perhaps skeggsb has some ideas for debugging steps
15:37Jayhost: Is there a place to get mmiotrace files. I'd like one for GK104 reclock
15:37imirkin_: ravior: so is it the INTR things which are the harbingers of doom for you?
15:37ravior: imirkin, I've attached the full dmesg log when using kernel 4.4.1
15:38ravior: Maybe. I'm not sure myself. Those 2 error logs always precedes the freeze.
15:38imirkin_: how long did you wait after the freeze?
15:38imirkin_: give it a few minutes, sometimes things can take a while to time out
15:39ravior: Depends. Now for example after the driver froze aproximatelly 40 seconds after the last warning message. So it wasn't an istant freeze.
15:39imirkin_: hmm ok
20:02orbea: any idea what this means? "mpv: pushbuf.c:727: nouveau_pushbuf_data: Assertion `kref' failed."
20:02orbea: seems inconsisetebt
21:28chrisb2244: Quick question (saw in a log but no response visible there) - does nouveau support DP 1.2?