00:01chillfan: what is the kernel module option to enable nvboost? I know how to do it via command line, just don't know the options to set when modprobing
00:01jouboyzandgals1: overall it's pretty good, well full month i was running nouveau, and i got one freeze , where i likely knew the cause of it
00:01chillfan: kernel command line*
00:03jouboyzandgals1: overall i should not be that sarcastic about programmers activity , but then again the performance issue is not a real issue either
00:07jouboyzandgals1: chillfan: i take it as there is one person , who seeks fot better performance:)
00:09jouboyzandgals1: you can tell others though, what card you had, but generally what ever you through me at, i could help
00:10jouboyzandgals1: throw though
00:12jouboyzandgals1: chillfan; there is probably no particular boost option, aside from nouveaus power management options, but i belive we could get you the perf is very much needed
00:14chillfan: the nouveau power management options is what I'm talking about :p
00:16jouboyzandgals1: yeah i do remember that some boost patches were circulating, but just feed your fingers
00:17jouboyzandgals1: in particular i belive karol had some tree even , but youd have to mention your card marketing name to me at least
00:18chillfan: ok i'll BBL
00:24jouboyzandgals1: i gonna head off to sleep, i stress out, don't be too embarrassed to listen to me, that is why converstions are about, or monologues:)
00:28jouboyzandgals1: lets say for even , and conversations even, but you are overly too much afraid about me, where as i never hurt anyone
00:31jouboyzandgals1: you are not talking with a troll, but one of the most talented ...
00:32jouboyzandgals1: i've had my downsides in life but in the end the stonger ones go through, who have internal content, hopefully i will manage to do that too
00:33jouboyzandgals1: i have never particularly shined that i do not regret the past, or have existance problems because of the minus i got, but...hey
00:34jouboyzandgals1: do not
00:35jouboyzandgals1: at some point all this needs to be flushed and we need to progress forward
00:36jouboyzandgals1: i'd literally break down and cry, but everyone has their own story
00:58jouboyzandgals1: probably a good mix when some trace the blob, and some bring some science into play, i would say that team cooperates, but you wanna ban me
06:42mlankhorst: imirkin_: eventually you'll hit the banlist limit ;)
08:49Walther: Hello! Anything happened recently with the nouveau drivers ref. gtx 970 support? Seems my earlier issue has reappeared, can't get to a TTY without setting `nomodeset` as kernel param
08:49Walther: I do have the `nouveau.noaccel` etc fixes in the kernel params, but those don't seem to do the trick now
09:20Walther: The same EDID checksum is invalid error as before shows up in logs
09:20Walther: disabling my display
09:30karolherbst: Walther: did you update your kerle or something?
09:32Walther: probably - the main kernel is 4.8.something, whatever arch pushed recently, I also have a 4.9 and a 4.10 installed but a) 4.9 shows the same issue b) 4.10 just kernel panics and i haven't debugged that further
09:34karolherbst: could you try older kernels?
09:35karolherbst: ... sry
09:37Walther: sure, I'll try and install an older kernel later - got to work on something now. I'll report back if it's of any help :)
09:41karolherbst: Walther: allthough, if it doesn't work with 4.8... :/ well maybe ping skeggsb, he usually knows what to do
09:49pmoreau: Walther: Is that a GTX 970 with 4GB of VRAM?
09:49Walther: yep, the infamous one
09:50karolherbst: ohh wait
09:50karolherbst: did you patch your kernel?
09:51pmoreau: I don’t think there has been much progress on that front, was there?
09:51karolherbst: no, but we have a working workaround
09:54karolherbst: mupuf: did you put the a maxwell2 card into reator?
09:54karolherbst: I want to work on that this weekend
10:20Yoshimo: anything specific?
10:28mupuf: karolherbst: hey, came back home yesterday. Will plug it now
10:29karolherbst: mupuf: awesome, thanks
10:30mupuf: you've got my two maxell gpus
10:31karolherbst: why two?
10:31karolherbst: mhh, let me think what would make most sense
10:31karolherbst: maxwell1 and maxwell2?
10:32karolherbst: okay, that would be perfect actually
10:32karolherbst: I can work on power budget stuff this way as well
11:00mooch2: mwk: what are ya workin on?
11:04padanoku: ouned: karolherbst: so you are a technical earth magic boy or what, with your PLL clock multiplier tasks, how such vision developed to such an idiotic dude though? we call that hallucinations
11:09Walther: I mean to be honest on the linux side I don't frankly even care about the 4G or 3.5G usable vram
11:09Walther: perfectly honestly most of the time i don't really need even hw acceleration
11:09karolherbst: Walther: yeah, there is this hack to limit it to 3G
11:09Walther: on linux my main concern is to get x running so i can get my terminals + editors + a browser
11:10padanoku: karolherbst: i would be saying on technical side take it up to tarjan and nvidia designers, that they are technical nothing, and you show greasefully how big of a joke you are
11:13mwk: mooch2: lots of things
11:14mwk: I've mapped out RDI areas for NV20, NV25, NV34 and some of their PIPE commands
11:15mwk: and attempted to do the same for NV4x, but it turns out that they got rid of RDI
11:16mwk: so I started looking at the ctxprogs instead, with partial success
11:17padanoku: he is nvidia employee, i only read and reported that, he is not technially nothing , i also showd the way how to implement this sw version, what ele do you need
11:18mwk: ie. I can save/restore the NV63 hidden state, and even managed to figure out what half of the areas do, but it turns out not to allow as much access as NV20/NV30
11:19mwk: and I've figured out most of the remaining unknowns about ctxprogs... but not all, the nvidia ctxprogs still use some commands that the card seems to ignore
11:20mwk: I figured I should pick another card, since NV63 seems to have simpler ctxprogs than some other cards, so I plugged in the actual NV40
11:20mwk: and, to my surprise, discovered it supports the Rankine 3d class
11:20mwk: which is going to be *very* interesting
11:22mwk: anyway, I've decided to finish mapping Celsius state first before dealing with all that crap in hwtest
11:22mwk: so I've plugged in another PIPE area to hwtest, and now I'm trying to figure out wtf is happening to it
11:23padanoku: techinically i go now, but i know how to implement this feature on every gpu, including nvidia and in several ways, in asm, also on opencl , what is your problem that i do not see
11:28padanoku: anyways we do not comment on your crap, you may schedule the threads in any ways you wish with minor code, SIMD offers it via cross-lane communication
11:31padanoku: there are more and several methods on different hw generations, you can grab absolute control of how the hw scheduler performs on all of them, so....
11:35padanoku: so there is no problem, only that you do not behave according to normal standarts, and tend to confuse others too, once you forget your crap belive me things will go well
11:35padanoku: who should i trust when not science from google done by nvidia employees, who would otherwise belive my theories?
11:41mooch: mwk: ah
11:41mooch: that's pretty interesting!
11:41padanoku: so please tell them , i am a german wonderboy, i tell hw in speech form what to do, and that is how it works, i care not about science and never read google materials about what actually goes on
11:42padanoku: it is purely idiotic what are you trying to say all the time
11:43mooch: padanoku, what are you trying to achieve by spamming this channel with incomprehensible gibberish?
11:44padanoku: you do nonsense, you talk nonsense and i have been sanktionized on top, get yourself together dudes
11:45mooch: sanktionized is not a word
11:45mooch: also, your grammar is laughably awful
11:47padanoku: it does not matter, call sanctioned
11:47padanoku: call it
11:47mwk: mooch: yep, so the good news is that there's an unbroken line of emulation going from NV3 to NV4x
11:47padanoku: i am not an english teacher, my point is also take care of the cache misses and multithreading if you want to ban me
11:48mwk: which means, once we figure out any common bit of functionality, it's going to be trivial to port that knowledge to all other GPUs in that range
11:49padanoku: otherwise i would be willing to help if you can not implement this, but i can't when you ban me, and users will suffer little because of that
11:49mwk: oh, and Kelvin/Rankine/Curie even have shared state bundle numbers [but Celsius does not], so yay
11:49mwk: and in related news, I've been trying to decipher rules.xml
11:49mwk: it's a *strange* file
11:50padanoku: do not overtry that much, usual code by design is thin and simple
11:50mooch: welp, it also means that i might have to reimplement EVERY SINGLE CLASS ON EVERY CARD TYPE
11:50mooch: mwk: what's rules.xml?
11:51mwk: it's a... file
11:51mwk: it describes nvidia registers
11:51mwk: and it was inspiration for the rnndb
11:52mwk: which I came to regret, but eh
11:52mooch: ah, i have one of those for nv1, nv3, nv3t, nv4/nv5, nv10, AND nv20
11:52mwk: noone seems to know where that file came from
11:52mooch: probably nvidia internal lmao
11:52mwk: the thing is, rules.xml doesn't really correspond to any GPU
11:53mwk: it has bits of NV4, NV10, NV20 in a weird mix
11:53padanoku: karolherbst: let's do that publicly i treated all of you with respect, as i always do, how have you responded? why are you so ashamed to follow the path what people write in reports?
11:53mooch: yeah, that's a problem
11:53mwk: I now think it's a WIP dump from when NV20 was being developed
11:54padanoku: this is not ego game, but the fastest and most efficient ways the code has to be landed
11:54mwk: also contains a few things not found on *any* GPU
11:54mwk: PDVD, for one
11:55mooch: yeah, that theory would make sense then
11:59padanoku: AFAIK. you intentionally are causing troubles to me, i should no longer deal with it, but hope that in my country people take me more seriously then your tried some sort of ruinage, it's a big ask
11:59padanoku: unfortunently those people who are doing the sancions they are incapable to do any sort of thinking
11:59padanoku: what you tell over irc they blindly seem to belive
12:03padanoku: and mooch: always the same thing as responce, i am estonian, try to talk couple of estonian words with me, i do mistakes in grammer in all this madness, but i speak quite fluently as been told to me
12:09padanoku: in my opinion as this is grown to be so enroumous trouble what goes around me daily basis you should at least try to avoid pushing false or incorrect information to he public only becaue you hate me that much
12:09padanoku: that yeah means more troubles to you and me also
12:18padanoku: you should had ever put the perf path on hold though, same wise actual capable persons should never sit locked up, for insence if hakzam was allredy hired, he will have enough work on stability issues too
13:31funfunctor: karolherbst: https://cgit.freedesktop.org/~funfunctor/envytools/commit/?h=decklink
13:32karolherbst: funfunctor: did you find the tools usefull for your work? :D
13:32funfunctor: I have not had enough time yet to attempt much yet
13:33funfunctor: I just wanted to point you to the patch and make note that I am following your advice
13:33karolherbst: I see
13:34funfunctor: I'll let you know if I get any further with them
13:34funfunctor: I am quite interested in the scan feature, hopefully it reduces down the state space
13:35funfunctor: Does it have a feature where I can scan for any strings?
13:39karolherbst: mot really
13:45funfunctor: karolherbst: the device is based off a FPGA
13:45funfunctor: its essentially a FPGA and a SPI flash were I assume it stores the bitstream compiled Verilog/VHDL/other
14:07karolherbst: I see
14:08funfunctor: I think I should backup the SPI flash externally to avoid bricking it
17:25nyef`: ... Thus far, if I poke MMIO registers to pass a 3d structure infoframe to my playstation 3d display, the display tells me "not supported" and turns itself off. I've tried half of the possible 3d structure packets so far, with none of them being supported. Still, this is progress. (-:
17:33nyef`: What's odd, though, is that I can force the side-by-side or top-and-bottom modes via the panel controls.
17:48nyef`: Ooh. Max TMDS clock in this EDID is 165MHz.
17:48imirkin_: which is expected for HDMI 1.2
17:49nyef`: Yeah, but the panel itself supports frame-packing.
17:49imirkin_: i understand
17:49imirkin_: but it's a HDMI 1.2 source. perhaps it detects that and reports an appropriate EDID?
17:50nyef`: Hrm. That could be unfortunate.
17:50imirkin_: also, nouveau will limit the TMDS to 165mhz unless you override it
17:50imirkin_: nouveau.hdmimhz can override it
17:51nyef`: ... how would it detect that it's an HDMI 1.2 source?
17:51imirkin_: no clue
17:52imirkin_: i have no idea how HDMI works
17:52imirkin_: [at that level]
17:54nyef`: It'd more-or-less have to be an i2c channel transaction, as it needs to occur pre-modeset, doesn't it?
17:54nyef`: That seems rather implausible.
17:54nyef`: Not to say that it couldn't happen somehow, using that mechanism or another, just that it seems implausible.
17:54imirkin_: so ... you're probably right. ignore me.
17:55imirkin_: anyways, if your expectations is that monitors, esp TVs, conform to standards, you're in for a surprise
17:57nyef`: Yeah, I'm fully aware that the EDID could be lying.
17:59imirkin_: also disregarding various infoframe things, etc
17:59imirkin_: does the edid have 3d modes in it?
17:59imirkin_: (can you pastebin the hexlified edid somewhere maybe?)
18:00nyef`: Would hexdump -Cv be sufficient?
18:00imirkin_: something i can undo easily
18:01imirkin_: which hexdump is not i think. xxd has a mode to parse right?
18:02imirkin_: so just xxd output
18:03nyef`: I don't seem to have anything called xxd.
18:03imirkin_: wtf?? appears to be shipped as part of 'vim-core' on gentoo.
18:04nyef`: That'd be why I don't have it. I'm an emacs user. d-:
18:04imirkin_: thanks. that's what i wanted it for anyways ;)
18:05imirkin_: oh. but 165mhz should be enough for 1920x1080@30 dual-frame
18:05imirkin_: so i think you have to send it the specific VICs for it to support the 3D-ness
18:07imirkin_: and i think the "VIC index" number is in reference to list of VICs it gives
18:07imirkin_: so like VIC index 14 is in reference to 1920x1080@24Hz
18:08imirkin_: while 1920x1080@60/50 supports top-and-bottom it seems
18:08imirkin_: (but not side-by-side)
18:08imirkin_: oh. top-and-bottom meaning alternating lines
18:09imirkin_: i.e. 1920x1080@60, but the top line being one eye, bottom line being another eye
18:10imirkin_: is what i'm saying making sense and/or new information?
18:15nyef`: That's about as far as I'd gotten.
18:15imirkin_: so then when you're dicking around with the infoframe, make sure it's one of those modes that's set
18:16nyef`: Okay, so I need to figure out which mode *is* set, or force one, and then also set up the AVI InfoFrame (because right now it's hardcoded).
18:17nyef`: Umm... And do a cross-check with a 3D PS3 game, to see what display mode it sets for frame packing, if I can get the display to tell me. (-:
19:11nyef`: Okay, AC3 is setting 1280x720 60p(3D). Guess that answers that.
19:12imirkin_: ok, so it does alternating fields. that makes sense.
19:13nyef`: Where do you get that?
19:14nyef`: I read that as VIC 4, index 4, supports side-by-side.
19:15imirkin_: oh yeah. right.
19:15imirkin_: i got that from assuming that things were logical. that'll show me!
19:57nyef`: ... Okay! Got the panel to switch to a side-by-side mode.
19:58nyef`: The side images seem to shift a bit relative to each other, though.
19:58imirkin_: 'side images'?
19:58nyef`: Probably something to do with the specific subsample whatever being used for side-by-side.
19:58nyef`: The left and right side images.
19:59imirkin_: are you sending it actual copies on left/right?
19:59nyef`: No, this is just the normal desktop view.
20:00imirkin_: how can you tell that it's shifted?
20:00nyef`: Because it's shifting /over time/.
20:00imirkin_: oh. that's surprising :)
20:00nyef`: Every so often, the text moves slightly.
20:01nyef`: Still, this is a decent win: I managed to kick the display into a 3D mode.
20:01imirkin_: job well done!
20:02nyef`: Yup. Horrid hack, but it works.
20:02nyef`: And I'll have to rebuild it in a few days because I'm going to switch panels.
20:14nyef`: So, major precursor for HDMI 3D is fixing the InfoFrame support in nouveau.
20:14imirkin_: and figuring out how to expose it to userspace, and making userspace do something reasonable
20:14nyef`: More or less, yes.
20:15imirkin_: so ... you're almost done ;)
20:15nyef`: Well, other drivers already expose HDMI 3D to userspace, so following that precedent is almost certainly in order.
20:22nyef`: Of course, this only covers HDMI 3D, not 3D Vision 2 Kit 3D.
20:39imirkin_: what is that? an api? or a literal device?
20:40nyef`: The 3D Vision 2 kit, from nVidia, contains some LCD shutter glasses and a USB IR emitter.
20:40nyef`: And a couple of other bits, but the emitter and the glasses are the important bits.
20:41imirkin_: "vesa stereo cable port"
20:41nyef`: Yeah, that's one of the optional bits.
20:41imirkin_: oh. that's the 3-pin din thing
20:41imirkin_: available on ancient quadro gpu's
20:42imirkin_: i have one, in fact ;)
20:42imirkin_: quadro fx 3450 or so (NV42)
20:43nyef`: So, there's a special driver that can be kicked in on windows to run a sufficiently fast (110Hz or faster?) display in sync with the glasses by way of the IR emitter so that alternating frames go to each eye. Or to both eyes on one set of glasses and then both eyes on another set.
20:43nyef`: (For two-player fullscreen games on one monitor.)
20:44nyef`: And then they built the emitter and a compatible panel into certain laptops.
20:45nyef`: There are *four* different types of emitter, and the timing parameters are largely unknown and almost certainly depend on the refresh rate and panel latency.
20:48nyef`: There's also a couple projects out there to get the emitter running, but the base code is pretty grotty, and it doesn't work on my setup. And it's fixed to a specific refresh rate.
20:48imirkin_: sounds like you've got plenty of work ahead of you
20:51nyef`: I'm currently thinking that the HDMI side might be the easier to get working, but the 3D Vision side is the one that I'm more interested in having working.
20:51imirkin_: well, the ultimate issue will be not feeding the 3d content to the applications
20:52imirkin_: well, the ultimate issue will be not feeding the 3d content to the output devices
20:52imirkin_: but rather of getting applications to create it in a usable manner
20:53nyef`: I'm going to presume that glxgears produces usable 3d content, and then I have an application of my own to hack 3D output onto.
20:53imirkin_: not sure that it does
20:54imirkin_: but it should be easy to hack to do whatever you need
20:54imirkin_: as i mentioned, i think the current thinking is alogn the lines of using the *multiview* extensions
20:54imirkin_: although the original GL 1.x solution was to have GLX_STEREO visuals
20:54nyef`: I'm perfectly happy with GLX_STEREO.
20:54imirkin_: well, neither is supported in mesa ;)
20:55imirkin_: [or glxgears]
20:55nyef`: glxgears supports GLX_STEREO.
20:55nyef`: I've been using it with glQuadBufferEmu for a while now.
20:55imirkin_: oh. interesting.
20:55nyef`: The problem being getting the emitter timing worked out.
20:55imirkin_: so it does.
20:56nyef`: And it *just* occurs to me that I now have a usable fullscreen 3D option over HDMI.
20:56imirkin_: well again, mesa doesn't expose stereo
20:56nyef`: You've been wrong about the userland once already. d-:
20:56imirkin_: yeah, but i'm pretty sure about that one :p
20:56karolherbst: have fun implementing it in mesa then :p
20:57imirkin_: and it'd have to somehow work over DRI2/DRI3 as well, dunno if those would support it easily or not
20:57imirkin_: i guess they ought to
20:57imirkin_: and then something would have to take those images and composite them in the "right format" onto the fb
20:58imirkin_: be it side-by-side or alternating lines
20:59nyef`:points to AddrLib::ComputeQbStereoInfo in mesa src/amd/addrlib/core/addrlib.cpp as an indication of possible support.
20:59imirkin_: you'd be mistaken
21:00imirkin_: that is effectively an external lib, shipped as part of mesa
21:00imirkin_: it's just taken from amd's other driver
21:01nyef`: There's also various hits in src/glx/.
21:01nyef`: Anyway, worst-case scenario is that I have to write it all myself.
21:02nyef`: I guess it being an external library from another amd driver would explain the lack of HDMI InfoFrame stuff in the kernel amd driver.
21:04nyef`: So, glQuadBufferEmu in side-by-side mode, fullscreen, and the hack to set HDMI InfoFrames for a side-by-side format, and user applications should be somewhat functional.
21:06nyef`: The real trick would be multiple windowed 3D contexts going at once.
21:07imirkin_: that's what i was talking about...
21:07imirkin_: something will need to create the two eyes
21:08imirkin_: and composite one eye's image onto one desktop, and the other eye's image onto the other desktop
21:08imirkin_: while keeping the "other" bits the same
21:08imirkin_: anyways, none of this is nouveau-specific
21:08imirkin_: you may want to discuss such design issues with people more competent than myself, and who have thought about such matters
21:09nyef`: Sure... But it can wait until I have the hardware functioning properly. (-:
22:33marduk: karolherbst: yeah i am sorry for brutal talking, i am not good programmer too, but about this simple thing i tried to explain the best i could
22:35marduk: there is reasonably simple path which i wanted other coders to know, it is mostly simpler then any sort of hard cpu coding, that is why i am so confident about this
22:41marduk: there would not a problem if i get a job, then i would explain this with slides and with a minor talk in addition somewhere
22:45marduk: currently i just don't yet have money for travelling and participating there on those events, surely it would be easier to answer to questions with a speech and minor prperation though
22:46karolherbst: marduk: thanks for the apology
22:53marduk: yeah, you know i did not mean to be harch, cause most userspace and kernel space programming is difficult , but i chose this field, actually with year of preperation there, this boils down to connecting only regs together in a slim manner, which is absolutely normally not the case in other coding areas
23:01marduk: ok i will be back some other time, so aplogies forwarded, i delt with pipelining research in mord then a years time, picked it caue it is easy, i can imagine that others are occupied just with more complex tasks
23:14nyef`: "Where do drm_modes come from?" "When a drm_connector and an edid love each other very much..."
23:22chillfan: karolherbst: are there any specific tests you are interested in to test reclocking on the 940m (maxwell/nv110)?
23:24chillfan: so far each reclocking level works and shows unique clock speeds once the discrete gpu is active
23:24imirkin_: if you want moar perf, you can apply hakzsam's patches locally
23:24imirkin_: [to mesa]
23:27chillfan: I guess I could benchmark it with phoronix, unigene etc
23:49karolherbst: chillfan: reclock and check how stable it is
23:49karolherbst: ohh yeah
23:49karolherbst: hakzsam patches as well
23:49karolherbst: to put more load on the gpu
23:49karolherbst: that helps with stresstesting
23:51chillfan: okay will do