01:25hexedit: i had an opportunity to try out the nouveau drivers...and so far glxgears seems to work but that is about it...and wglxgears works a little but has a lot of noise in the picture
01:25hexedit: basically everything else crashes...and some of it locks up the gpu
01:26hexedit: so it is better than nothing but still needs a lot of work in my opinion to be useful
01:31hexedit: linus says he has been working now on the kenel for 25 years ...and he says he is not tired yet...and he thinks he can go for another 25 years to keep providing a free gnu system...for people who like free software...but nouveau may also need 25 more years of work
01:38airlied: hexedit: you know it works pretty well for other people, so it's probably just specific hw that isn't as well cared for as some other hw
01:39hexedit: airlied...well it could be bad intel and nvidia hardware but if intel and nvidia is no good who else is there
01:39hexedit: airlied..i prefer to think that driver improvement can solve some of these problems...especially to guard against gpu lockups
01:40airlied: no I'm not saying the hw is bad, just that nouveau isn't a consistent behaviour across all hw
01:40airlied: so for you gears is all you can getworking, others run a lot more than that on it
01:41hexedit: airlied...well i got some of abiword to work but the print preview does not work..however wine now supports the printing from it ok and it also is free software
01:41hexedit: airlied...but those are not state of the art 3d work
01:43hexedit: airlied..i am thinking another 25 years might be necessary to get a usable free system
01:45airlied: hexedit: as I said you are extrapolating a lot from a sample size of one
01:45airlied: don't think just because you have some issues that everyone else will have the same ones
01:46hexedit: airlied...that is very correct and as you know small sample sizes can cause large errors in conclusions...but as one point...i would say a lot more work is necessary.
01:46airlied: there is always more work to b done
01:46airlied: that isn't something that needs a proof
01:48hexedit: well Richard Stallman, Linus Torvalds, Bruce Perens, and those who pick up the ball from Ian Mudock and several other key players are all committed to work diligently for many years to come to try to create a free system that is usable
01:49hexedit: airlied...no one knows what happened to Miguel Icaza..but some say he will come back also to work on free software
01:55hexedit: airlied but one problem that slashdotters talk about is ..."Who wants to try to use version 0.1 of something that hasn't been updated in 10 years?" and how will people ever decide to quit using propietary software like windows with this kind of issues.
01:56hexedit: airlied...it is just going to take persistence, patience, and a lot more effort
04:13martm: for god sakes the people in my list to be treated are not innocent at all what they play to the world, that i am a bastard and they did not commit insanities, some of them aferised me 5years some 15 etc. i will reach to there at some point first court as said, if leftovers with law can not continue their activity and if seen theyd be charged continuously etc.
04:16martm: the probability to screw the chanche for me to communicate with any person was of course wanted to be acheived as said
04:17martm: its tough but i go for a payback knowing even if it produces more new problems, i don't tolerate certain people with false motives doing stuff in my life
04:26martm: now individual sports are things where sabbotaging looks impossible it was possible and i was injured the very most oawfully too, if i cure it out in 75degree, the only chanche would be to play again, about those idiots who caused those issue "intelligently" well guess what , they can not play or do anything that is legal
04:28martm: you may think by the looks they can, but after trailing 8:9 you may start to think that scamming aferism and piketations after that playing a victim is the only thing they can do
11:24ggherdov`: Hello. I'm seeing symptoms hinting at some nouveau problems with my graphic card (Nvidia Quadro). These include junk on the screen (see http://imgur.com/SWMHVSW and http://imgur.com/WAapMcf ) and unexpected logouts.
11:24ggherdov`: I am looking for help to produce a meaningful bug report. What logs should I look into?
11:24ggherdov`: Some system info: https://bpaste.net/show/965ebc8ebb55
11:34RSpliet: ggherdov`: that first one has some quite pretty junk.
11:35RSpliet: but start by updating your software stack - kernel 4.5 and mesa 11.2 are already going to make quite a difference I expect
11:52ggherdov`: RSpliet: ok thank you for the hint
12:04hexedit: ggherdov...yes some cards are not supported well by the driver but the information is kept private...they don't want secrets going into the chatroom
12:05hexedit: ggherdov....you have to respect their rights to protect what they consider is their property
12:05hexedit: ggherdov...which quaddro do you have?
12:07hexedit: ggherdov..they may contact you in secret if you hang around
12:10hexedit: ggherdov...it is clear that many people will support free software to an extent....but some things are considered to valuable to release from private ownership
12:12karolherbst: hexedit: which secrets?
12:14hexedit: karolherbst..i can't reveal certain things about jiffies time slices or cache coherency or other topics the driver programmers want kept secret.
12:14ggherdov`: hexedit: sure, I don't mean to get information I am not entitled to. My request was for what logs can I report in a possible bug entry that are more specific that "junk on screen - doesn't work"
12:14ggherdov`: by the way, Quadro K620.
12:14karolherbst: hexedit: which driver programmers?
12:15karolherbst: ggherdov`: ohh that's a maxwell one
12:15karolherbst: ggherdov`: maxwell still has a lot of issues, but we try to work them all out
12:15hexedit: karolherbst...they contact you in secret if you hang around...they dont' want some information going into the channel where it can be stolen and mass distributed...
12:15karolherbst: hexedit: now I am worried :O
12:17hexedit: karolherbst...but there are issues like above and syncronization primitives going to the pipeline and others...that they don't want to give to their competitors
12:17ggherdov`: karolherbst: do you suggest me to fallback to another driver, like nv or the proprietary nvidia one, until my card have better support?
12:17karolherbst: hexedit: yeah well
12:18karolherbst: hexedit: but the competitors will also look into the hardware
12:18karolherbst: hexedit: so in the end, they may know even more about the really secret stuff than we do
12:18RSpliet: ggherdov`: no, just try a new software stack. GM107 has had quite some work since kernel 4.1 was released
12:18karolherbst: ggherdov`: I would suggesting helping us :)
12:18karolherbst: ggherdov`: if it is an misrendering issue, create an apitrace and report it with a bug
12:19hexedit: ggherdov...no i would just use what you can get out of it....unless it freezes things where you can't use it...if so then fallback to vesa drivers or nv
12:19karolherbst: is nv nvidia or something else?
12:19karolherbst: ohh wait, it is that old fb driver for nvidia cards?
12:19hexedit: karolherbst..no nv is basically a 2d driver only
12:19RSpliet: karolherbst: nv is the ancient non-maintained driver
12:19karolherbst: yeah I just wanted to say
12:19karolherbst: this may have more issues than nouveau
12:19RSpliet: it does
12:20ggherdov`: RSpliet: ok copy that. karolherbst: that sounds fun. I'll google for what an " apitrace" is. Yes, by nv I meant the old driver.
12:20karolherbst: ggherdov`: just an application to trace the gl calls
12:20karolherbst: really usefull for debugging
12:20ggherdov`: karolherbst: ok.
12:20karolherbst: ggherdov`: like you can replay it on different gpus and different drivers and we can just check what we do differently or what nvidia would do
12:20ggherdov`: i see.
12:21karolherbst: but that's for opengl issues
12:21karolherbst: if the gpu just crashes, then you might have to check somewhere else
12:21hexedit: ggherdov...nv is quite stable and doesn't need much work...neither does the vesa drivers...but they are basically 2d only
12:21ggherdov`: but first I need to find some reproducible case. My thing doesn't happen consistently.
12:21ggherdov`: it comes and goes.
12:21karolherbst: ggherdov`: what's your issue by the way? Just a simple gpu crash?
12:22karolherbst: ggherdov`: and, did you reclock the card?
12:22ggherdov`: karolherbst: letters on the screen are sometimes rendered badly: http://imgur.com/SWMHVSW and http://imgur.com/WAapMcf
12:22mwk: hexedit: nv has the minor problem of not supporting any GPU since Fermi...
12:22karolherbst: ggherdov`: that's ony gnome, isn't it?
12:22ggherdov`: karolherbst: I didn't try anything else than gnome
12:23ggherdov`: What is "reclock" btw?
12:23karolherbst: ggherdov`: pstate file
12:23hexedit: mwk...why does nv need to support any gpu...it is 2d only...can't it just use the old vga and svga specs
12:23karolherbst: ggherdov`: you could try if running gtk software only changes anything
12:23hexedit: mwk...are you saying the new hardware is not backward compatible with vesa
12:24mwk: hexedit: it is compatible with vesa, but nv uses the hardware registers directly
12:24karolherbst: ggherdov`: mhhh strange issue though. There might be something weird with the glamor acceleration here. Mind updating mesa to the newest version?
12:24mwk: nv is actually much more advanced than vesa allows you to be, supporting 2d acceleration for one
12:24hexedit: mwk...oh ok...well i don't have any experience with the newer stuff....so i cannot offer any reliable info in that regard
12:25ggherdov`: I also have another symptom, which is logout at some random time. Like, I click on a link and bum, logout.
12:25ggherdov`: karolherbst ok will update mesa/
12:25karolherbst: ggherdov`: ohhh, I have an issue with plasma5, where when I open the context menu, it immediatly selects close .... :D
12:25karolherbst: but I run intel
12:26karolherbst: I would say the logout comes from a crashed session daemon or something like that
12:26karolherbst: like when gnomeshell would crash
12:26hexedit: karolherbst...well doesn't intel maintain some amount of backward compatibility
12:26ggherdov`: karolherbst: I see.
12:27hexedit: karolherbst..i am not familiar at all with intels' graphics chips but i would think they support some amount of backward compatibility
12:28mwk: hexedit: ... why would that matter?
12:28mwk: for what it's worth, no GPU manufacturer that I know of cares about backwards compatibility
12:29mwk: on the hw level
12:30hexedit: mwk...well maybe they don't ...that is their priviledge...after all itis their product....but some people like backward compability so they may lose some customers with their product choices
12:31karolherbst: hexedit: why, why should they care?
12:31karolherbst: you don't write your software against a specific GPU, just against an API, and this API usually stays the same or is based on an abstract feature set
12:31hexedit: karolherbst...that is a tough question...i suppose it must relate in some way to their self-interest
12:31karolherbst: but you usually don't write your binary code yourself
12:32karolherbst: they don't care about this
12:32karolherbst: and the one who do care, are <0.1% of all users
12:33hexedit: karolherbst...these things are not completely abstract...as some gpus have 32 threads per warp or 64 threads per wavefront
12:34hexedit: karolherbst...but what i am saying is that a tough question generally leaves me wondering if self-interest is a factor
12:35mwk: hexedit: backwards compatibility is fundamentally at odds with making reasonably performant GPUs
12:36mwk: every new GPU generation changes a lot of things in the interest of speeding things up
12:36hexedit: mwk...well i certainly don't tell people how to run their business....that is entirely up to the gpu mfgrs
12:37mwk: the only way to make eg. a Tesla-generation GPU compatible with Curie on hw-level would be to pull a Macintosh and include a copy of the old chip
12:37hexedit: mwk...what i do say is a lot of hardware people owe several of us model 1 pioneers a lot of money for our ideas we never released to the public
12:37hexedit: mwk...why bring up the word McIntosh....that ends the interest right there
12:38hexedit: mwk...some people associate Mac with Apple..and no one likes them
12:39hexedit: mwk...they are the antithesis of free software and free hardware
12:53Wonka: weeeell... the company Linus T. told "fuck you" is not apple, but nvidia ;)
12:55hexedit: Wonka...yes well Linus says lots of things and he is free to do so...but most people still respect him for his work even if everyone is not in agreement with his philosophy
12:56Wonka: I have a Tegra and a Tegra 3 device lying around which don't work as good as they could because of nvidia not releasing data...
12:56Wonka: my apple notebooks work better..
12:56Wonka: but yes, apple wants the sheeple in their walled garden
12:57hexedit: Wonka....yes well but i bet you had to pay a pretty penny for the Apple stuff....some of their iphones cost 800$...i certainly don't have one
12:57Wonka: hexedit: got a macbook pro 3,2 and a macbook pro 8,1 - no phones.
12:58Wonka: hexedit: the phones are LG Optimus 2X and Google/Asus Nexus 7 3G (tilapia)
12:58hexedit: Wonka...well what do you use them for ....how do you cost-justify they expense?
12:59hexedit: Wonka...what can you do with a MacBook
12:59Wonka: the mbp8 I bought summer 2011 and it lasts up until now, running Debian GNU/Linux. the mbp3 I used from 2007 to 2011, also with Debian.
13:00hexedit: Wonka..oh..so at least you don't have to cost justify the software expense.
13:00Wonka: and all the notebooks before, from hp, IBM, and a noname, only lasted less than 3 years each.
13:00hexedit: Wonka....yes i know...notebooks, laptops, watches..and so on are ridiculously short-lived
13:00Wonka: currently I wouldn't buy another apple book though.
13:01Wonka: there's just nothing on the market that I'd like to use.
13:01hexedit: Wonka...well some people have to be careful with their pennys or else they will lose their dollars.
13:01Wonka: I know.
13:03Wonka: I'd prefer Intel-only or Intel/AMD hybrid graphics, but iff nouveau does support low power stuff for the GPU in some candidate notebook, I might consider it
13:04hexedit: Wonka...well the choice is all yours...in a free society you are free to allocate your scarce resources as you see fit...not under the orders of someone else
13:27karolherbst: Wonka: well there is an easy way to reduce power consumption on kepler and newer GPUs, but if it's a hybrid, then the dedicated GPU is turned off anyway
13:27karolherbst: except when you want to use it while on battery
13:28imirkin: ggherdov`: definitely try updating your stack. other users were seeing stuff like that on maxwell's, but apparently it's fixed now with newer glamor versions (which are bundled as part of Xorg)
13:29ggherdov`: imirkin: ok let's do it
13:31ggherdov`: I have a little resistance against it as I like to live within the boundaries of my distribution (openSUSE) and what's packaged for it. But in this case it's so annoying that it might be worth the exception
13:31ggherdov`: I might as well report the bug to openSUSE, tho.
13:32imirkin: ggherdov`: afaik glamor changed the way it did some things to no longer tickle the bug in nouveau
13:33imirkin: tbh i'm not sure of all the parameters required to trigger the issue - i don't have the hw in question
13:34imirkin: ggherdov`: i'd definitely grab xorg 1.18.1 or later, and mesa 11.1.2 or 11.2.0
13:34imirkin: i think those are going to be the "active" components here
13:34karolherbst: and skip xorg-server 1.18.2
13:34ggherdov`: ok imirkin
13:36imirkin: ggherdov`: i believe that there are some (user-provided?) package repos for opensuse you can use to get updated stuff
13:36imirkin: however i don't know too much about distros
13:36ggherdov`: uhm.. probably
13:43karolherbst: there is also tumbleweed
13:43karolherbst: but I don't know how new the software is there
13:48karolherbst: did anybody looked into the D table yet?
13:48karolherbst: It seems to point to two tables in my vbios
13:48karolherbst: first begins with 0x10 0x0b 0xd0 0xb4 though
13:49karolherbst: ohh and there are also entries
13:52RSpliet: karolherbst: there's plenty of tables unparsed I believe
13:52karolherbst: I try to find some of them, well I try to find the one I need
13:52karolherbst: but well
13:55karolherbst: this looks like something: https://gist.github.com/karolherbst/d279f97e458d193f8886a17051df6bfc
13:56karolherbst: no clue what though
13:58karolherbst: RSpliet: I think I try to at least add all the tables I find like the unkowns in the P table, that they get at least displayed with -v
13:58RSpliet: karolherbst: good effort I'd say. Keep in mind some stuff is interpreted in nvkm but not in envybios
13:59RSpliet: could save you some trouble ;-)
13:59karolherbst: I don't think there will be unknown BIT tables being parsed in nvkm?
13:59RSpliet: skeggsb tends to not touch envytools too much, rather codes stuff up in the kernel straight away
14:00karolherbst: yeah I know
14:00karolherbst: I already backported some of his stuff to envytools :D
14:00RSpliet: it's a bad habit of his, but given the eternal work pressure I can't really blame him
14:01karolherbst: yeah I know
14:01karolherbst: well there is no bit table D parsed within nvkm
14:01karolherbst: so I am good here
14:01karolherbst: but I don'T get this table...
14:01karolherbst: there seems to be no header
14:01karolherbst: and only entries
14:02karolherbst: uhh the second table is a big one
14:03RSpliet: don't rule out the possibility of it being scripts
14:04karolherbst: yeah, but it doesn't look like one
14:04karolherbst: and it really looks like entries
14:19Misanthropos: are these entries related to my gk106?
14:20Misanthropos: if so: it has hdmi audio controller on it too
14:23RSpliet: Misanthropos: not sure what exactly you are trying to say, but HDMI audio with nouveau on Kepler should work since several years ago
14:23Misanthropos: just wild speculation about the table
14:42karolherbst: no clue what this table is
14:42karolherbst: except the letter D
14:43mwk: aren't they already nearly out of letters? might not mean much
14:46karolherbst: those two tables are in the D bit table referenced: https://gist.github.com/karolherbst/d279f97e458d193f8886a17051df6bfc
14:46karolherbst: the second makes actually more sense
14:46karolherbst: but looks a bit odd
14:48karolherbst: mwk: well the D bit table is also in a nv50 gpu
14:48karolherbst: also with two entries
14:48karolherbst: nv40 too
14:49karolherbst: I am not sure about the first table, but the second one looks valid
14:50karolherbst: 58 55 65 44 <= but maybe the first two bytes mean something else
15:17slayerjain: hello everyone!
15:20slayerjain: never really wrote a driver exempt a very minimal VGA driver on rust for a hobby OS
15:21slayerjain: can anyone help me get started? :)
15:21karolherbst: slayerjain: what do you want to do?
15:23slayerjain: make use of my GTX 970 for something more interesting than gaming :)
15:24slayerjain: and if that fetches me a stipend in X.Org Endless Vacation of Code, nothing like it! :)
15:25RSpliet: slayerjain: you'd have to be a bit more specific than that, nouveau should already support GTX970
15:26RSpliet: and I can't possibly imagine anything more interesting than gaming!
15:27hexedit: here is a link with free software tips to help you get started
15:28imirkin_: slayerjain: tess on maxwell would be a good project (GTX 970 is maxwell)
15:28slayerjain: so have to implement tessellation ?
15:28imirkin_: mmmm... not really
15:28imirkin_: have to implement fetching of tess control and eval inputs
15:28imirkin_: and fetching/storing of tess control outputs
15:28imirkin_: everything else should be done
15:29slayerjain: where should i look to get started?
15:29imirkin_: it'll be a bunch of tracing work and a bunch of work to plumb the logic through the compiler
15:29imirkin_: what's your level of familiarity with tessellation shaders?
15:29slayerjain: apparently only heard if it
15:30imirkin_: what's your level of familiarity with C and C++?
15:30slayerjain: I have done some competitive programming in C++
15:30slayerjain: and yeah i'm acutely well versed in C++
15:30imirkin_: this is a *very* quick writeup i wrote on how tessellation shaders work: https://trello.com/c/ZPMl5Oa0/46-a4xx-gl4-tcs-tes
15:31imirkin_: however it might not make too much sense if you don't know about the rest of the pipeline
15:31RSpliet: slayerjain: you still didn't really specify: are you trying to improve the current nouveau driver, or get support for "graphics" in a different operating system? Are you mostly after an X.org EVoC stipend, or is there something specific you want to do with your GPU?
15:32imirkin_: basically a graphics pipeline shovels data from start to finish, and each shader processes some input data, and outputs some data, and then there are some fixed function bits which transform/massage the data a bit as it gets shoveled between pipeline stages
15:32slayerjain: RSpliet : Apparently i want to learn about the current driver
15:32imirkin_: you might want to read up on how the graphics pipeline works in general a bit... or you don't have to and try to dive into the tess stuff directly
15:33slayerjain: so that i could make a similar driver for the OS dev thing going on in the rust community
15:33imirkin_: hah, that'd be quite the undertaking.
15:33slayerjain: RSpliet : these days when i game, a guilty monster screams all the interest i have in it xD
15:34RSpliet: slayerjain: ah okay! Generally what works best is finding a little itch of your own to scratch. The trello page imirkin_ linked to has a list of itches that we have... some smaller, some larger, touching various aspects of the graphics driver They can all be a source of inspiration for problems you might want to play with
15:34slayerjain: imirkin_ : thanks for the info! :) going though all the stuff
15:34RSpliet: (the full page: https://trello.com/b/ZudRDiTL/nouveau )
15:35imirkin_: RSpliet: actually i linked to the freedreno page since that's where i did my writeup about tess
15:35RSpliet: imirkin_: whoops, I should compare more than one character when comparing hashes it seems :-D
15:35slayerjain: i have never used trello before, looks a lil overwhelming :D
15:36imirkin_: slayerjain: it's just a thing that lets you have lists of cards, and cards with content in them
15:36imirkin_: it's a convenient "todo" list
15:36slayerjain: sounds fun!
15:36imirkin_: [at least that's how i use it... some people use it to implement more formal processes]
15:40Weaselweb: i'd like to get HDMI hotplug working on NVAA, but I lack time currently and also understanding what needs to be touched
15:41imirkin_: Weaselweb: what about it doesn't work? never gets the hpd?
15:42Weaselweb: imirkin_: I need to power on my A/V receiver before my system starts (presumable before nouveau gets actually loaded) or I don't get any X screen later
15:43Weaselweb: I would need to recheck for actual kernel output
15:43imirkin_: Weaselweb: i'd be curious what 'grep . /sys/class/drm/card*-*/status' and 'grep . /sys/class/drm/card*-*/modes' say when this happens
15:45Weaselweb: imirkin_: ok, set my a reminder, so I can show you this evening (CEST)
15:45imirkin_: i may or may not be around, but i read logs
15:45Weaselweb: yeah, no problem
15:48Weaselweb: imirkin_: any debug settings I should enable?
15:48imirkin_: assuming it doesn't Just Work (tm) with a new kernel
15:48imirkin_: i would do...
15:48imirkin_: nouveau.debug=debug drm.debug=0x1e
16:43hexedit: RSpliet...well video pipelines have always been about the mundane job of shoveling...but you should respect a man with a shovel...as you will need him one day.
16:43hexedit: RSpliet...the day you cannot shovel your own grave
16:52martm: lets do a minor joke, repeat after me, I HAVE TWO sides of brain, on the right side there is nothing right, and on the LEFT side there is nothing left!
16:53martm: obviously you want to talk about russian mickey mouses, i said i loved them, there is a basic error my dad thinks that his roof is in russia, but i think that my roof is in estonia
16:54hexedit: martm...i don't joke about psychobabbling bs...i think all psychobabblers shoud be destroyed or incarcerated for life....their institutions should be shut down and money returned to the taxpayers
16:54hexedit: martm...those sons of bitches should not be allowed to even live in my opinion
16:55martm: that means basically when we do not like russians, i.e one thinking about doing greater speeches how i suck, then obviously we will bury those
16:55hexedit: martm...they try to use knowledge as a weapon....speaking nicely but oppressing with force and drugs and restraining suits everywhere they can
16:56martm: hexedit: obviously you are an intelligent fellow from the past, but i dunno what the hell
16:56hexedit: martm..you don't know what the hell?
16:57martm: i dunno what are you talking about, that bigger companies should pay you, that never happens, you should be proud if they overtook some of your ideas though
16:57hexedit: martm...to sum it up the filthy rich capitalist pigs owe us a lot of money
16:57martm: yeah ...well..
16:58hexedit: martm...instead of using their bulldogs to claim we are mentally inferior...and depriving us of our jobs, our money, and our wives and children
16:59hexedit: martm...you don't know what the hell?...well it is a long story!
17:01martm: ouh...i am like intelligent or so, i can think of some versions how you were robbed:D
17:01martm: its not about ideas but how you can put it in practise
17:01hexedit: martm just because some goddamn president named bill clinton...was using power and authority to take advantage of and rape innocent women.....and with a wife that tried to support him and lay all the blame on the women....and say she stood for women's rights
17:02hexedit: martm...back around 93.
17:02hexedit: and the monica lewinsky headlines for one
17:02martm: ouh my grandmother knows that woman:)
17:03martm: the so called wife, not monica
17:03hexedit: martm...and with the crooked ass niggers that mostly were democrats...and the crooked nigger dick suckin jews that sided with them
17:04hexedit: martm..but it is a 20year+ long story
17:04martm: processing the last sentence , before your last
17:05martm: so democrats sucking jews?
17:05martm: ouh...that is pretty bad
17:06martm: but if you were to choose something, would you suck an arabian instead?
17:06martm: like hell no
17:06hexedit: martm...no i am talking about all the goddamn Jews...that suck the crooked niggers dicks and side with them against the greeks
17:06hexedit: martm...beginning with their involvement with the Queen of Sheba
17:07martm: ouh shit..man
17:07martm: so jews and niggas against the greeks?
17:07hexedit: martm...look up the old testament...the sorry nigger fuckin and nigger dick suckin Jews
17:08imirkin_: i'm told that parties who i've (and by now, i assume lots of others) ignored are loudly talking in here about topics highly unrelated to #nouveau. please stop.
17:09martm: imirkin: yeah
17:09martm: imirkin: but you hadrly talk about programming those days, but no worries i figured it out once aggain
17:10imirkin_: i _really_ don't want to go around banning people. if you want to chat about things not related to this channel's topic, feel free to do so in other chat rooms. you can even create your own.
17:12martm: imirkin_: well i don't want to, but... i just figured that to avoid deadlocks one would need a scheduler too
17:13martm: imirkin_: actually i talked about programming with hexedit than i realised that he is like two times older then me, and does not understand the hell i talk about
17:16martm: the recap is like so, nvidia has 32threads in a warp, and 8sm's on kepler
17:17martm: when a fast instruction is proccessed it is the max what hw can do 8*32threads
17:17martm: rest of them will skip the instruction and process the next one
17:19martm: but if it was a memory instruction and you were to launch 8*32 threads...7*32 would skip the instruction but, from the one
17:19martm: 28 of them would wait
17:19martm: that is the deal what scheduler solves actually
17:22martm: imirkin: i developed a scheme, where it is impossible to hit a deadlock, yet it outperforms everything, but does not matter, i know you are quite an intellifent fellow you could think one yourself too
17:33martm: you just have to pin the branches in correct order and with correct masks, on radeon it can be done two ways, either reorder and positive offsets to jump, to or same order and negative branch offsets, it goes back to handler in both cases
17:36martm: imirkin: as i said, on asics the job is done too, i needed a weekend to get it right
17:37martm: imirkin: so now you can go out with karolherbst and mwk: i think at least so
17:38martm: i can't come. cause my money is low
17:39martm: imirkin: it would be like programmers reunion
17:47martm: to be honest in contrast to..well i don't care i have no demands when you understand it's ok , it's to the free world, i can handle my my own
17:48martm: like no patents when you understand what i say i am glad
17:53hexedit: martm...well like i said...patents discriminate against the poor working class or proletariat....they are inherently unfair as they only protect the filthy rich capitalist pigs that can afford them or buy them for favorite "yes men" employees
17:55martm: but that is a 130nm processor you talk about, we can optimize those too, but i have like 20nm one, lots of more transistors
17:57hexedit: martm...so you know the procedures for shrinking die sizes without excessive leakage...and you claim to not even know the purpose of pl1 when you were caught laughing...in the logs
17:58martm: hexedit: man it's allright
17:59martm: hexedit: i am all the way drunk it's just i know the importance of the bootstrapping code, i never developed any
17:59hexedit: martm...it might be alright for you ...but where are my wife and children you motherfuckers stole
18:00hexedit: martm...well drinking alcohol never helped anyone solve a stochastic partial differential equation.
18:00martm: i am an outsider i never steal, never harm anyone..i give everything away
18:01martm: you sound like worse then a nagging russian!
18:01hexedit: martm...well if you give everything away ...you may find you have nothing in store for a rainy day...or a day that you become crippled...or a day you need to retire...or a day you won't have the funds to be a rainbow in someone elses cloud.
18:02hexedit: martm...if you cannot even help yourself...how in the world do you expect to help others?
18:03martm: good point, but.. my energy is bigger the theory
18:04hexedit: martm..well quite often energy dissipates...so you may want to prepare ahead of time with some form of energy replenishment or energy storage for future uses.
18:04martm: i can allow some of the leeches to suck my blood, and still function
18:05Tom^: imirkin_: no remorse, no mercy. we would understand :p
18:05hexedit: martm....fuck the leeches and parasites....you might sing a different tune if you get infected with drug resistant microorganisms
18:05imirkin_: hexedit: martm: please stop talking about random bs here. this is your last warning.
18:06hexedit: imirkin_...why waste time on a last warning....the delay could cost you your life if it were a gun....i don't need this channel..it is the people stealing my model 1 ideas...that need to pay me royalties.\
18:07martm: palun pahandust!:) it means please troubles!
18:08martm: imirkin: no i am ok, it does not matter it's not like i have not been banned before, but i am turning into more civlized person every day
18:10Weaselweb: imirkin_: here's the dmesg output on NVAA when HDMI is hotplugged (as in A/V is turned on) starting on line 1283: https://bpaste.net/show/e4af07d05444
18:11Weaselweb: media-libs/mesa-11.0.6, x11-libs/libdrm-2.4.66, kernel 4.5.0-gentoo
18:13imirkin_: it notices that it got plugged in
18:14imirkin_: and then immediately thinks it got disconnected
18:14imirkin_: and then shortly thereafter thinks it's ocnnected
18:14imirkin_: so the final state is "good", no?
18:15Weaselweb: grep . /sys/class/drm/card*-*/status: /sys/class/drm/card0-HDMI-A-1/status:connected and /sys/class/drm/card0-VGA-1/status:disconnected
18:15Weaselweb: grep . /sys/class/drm/card*-*/modes lists a bunch of resolutions for HDMI-A-1
18:16Weaselweb: same as the modlines in dmesg
18:16martm: ouh this one was bigger expert on hdmi then pekka paalanen, yeah imirkin prolly knows that stuff
18:16imirkin_: so ... it sees your HDMI-A-1
18:16imirkin_: so what's the problem again?
18:16Weaselweb: X is not shown on my screen (TV)
18:17Weaselweb: same as when the PC is off
18:17imirkin_: do you have 2 screens?
18:17imirkin_: no, just the 1, right?
18:17imirkin_: i assume you're ssh'd in?
18:17imirkin_: is X running?
18:17Weaselweb: yep, want Xorg.log?
18:17imirkin_: please. and the output of 'xrandr'
18:18imirkin_: (remember to do like DISPLAY=:0 xrandr)
18:18Weaselweb: Xorg.0.log: https://bpaste.net/show/054f424989a8
18:18Weaselweb: DISPLAY=:0 xrandr: https://bpaste.net/show/0a92f1e2abec
18:18imirkin_: so the screen's not on
18:18imirkin_: if you do
18:18imirkin_: xrandr --output HDMI-1 --auto
18:19imirkin_: i bet things will work better
18:19Weaselweb: imirkin_: yay \o/, got my X screen back. so what does this command actually do?
18:19imirkin_: it configures the output to be enabled
18:20imirkin_: when you hotplug a new output, X doesn't magically turn it on
18:20Weaselweb: is there a way to get this done automatically?
18:20imirkin_: you could have a thing listening to hotplug events... various desktop managers tend to do this
18:20Weaselweb: mh, I just have fluxbox for mythtv. any keywords I can google for?
18:21imirkin_: not sure offhand, sorry
18:21imirkin_: perhaps ask the mythtv guys?
18:22Weaselweb: well, I would rather ask fluxbox guys, mythtv is just an application running with fluxbox.
18:22martm: you know i'd say when you give stuff to free world, people will recognize you when you have trouble
18:22Weaselweb: but thanks so far, you're my here of the day (tm) ;-)
18:22imirkin_: you could also ask in #xorg-users
19:13Weaselweb: imirkin_: just FYI: is setup something myself using http://portix.bitbucket.org/srandrd/
19:14hexedit: martm...well i don't know about you...but i don't give away anything....resources are too scarce for waste...the problem is how to get all these thieves to pay what they owe...when all this amounts to trillions of dollars
19:15hexedit: martm...i don't see the benefits of martydom
19:16hexedit: martm...martydom and altruism...is self-annihilating philosophies
19:17hexedit: martm...and it is especially important that you don't give away scarce resources to people who will squander them for nothing and lose the benefits, societal or otherwise, of wisely allocated resources
19:18hexedit: martm...do you want to force a whole society to starve
19:20hexedit: martm...if you don't strive to see that resources are wisely allocated the pizza that feeds the entire society is lost to waste and inefficient use of scarce resources
19:23hexedit: martm...unless you want to play the imperialist capitalism game where the technologically superior conquer and destroy all opposition to aquire resources as they need them...but this is just barbarism
19:24hexedit: martm ...what chance did the indians with bows and arrows have against gattling guns brought to this land for example
19:25martm: ouh dear o will catch up
19:28hexedit: martm...its too late for the indians to catch up with the wealthier capitalist pigs that invested in superior technology with their wealth...
19:28hexedit: martm...they were all shot to hell by the gatling guns and cannons and carbines
19:29martm: in that sentence three words i have to translate, hold on:)!
19:30hexedit: martm...ok you don't understand superior technology vs inferior technology and the potential to resort to barbarism in rough times
19:31karolherbst: if you want to discuss such stuff, could you do that via private messages?
19:31imirkin_: they're back at it?
19:31martm: i am not even sure, why am i on the ...you know
19:35mwk: good riddance...
19:39SaveTheRobots: i seem to be having issues with DPMS and the nouveau driver, when running "xet dpms force off", the screen goes black for 10 secs then comes back on of its own volition
19:40SaveTheRobots: [john@voidmusl ~]$ grep DPMS /var/log/Xorg.0.log
19:40SaveTheRobots: [ 4320.421] (II) modeset(0): DPMS capabilities: Off
19:40SaveTheRobots: [ 4320.518] (==) modeset(0): DPMS enabled
19:40imirkin_: are you using DP?
19:40SaveTheRobots: HDMI atm
19:41SaveTheRobots: i can test DP if you like?
19:41SaveTheRobots: i'm using the modesetting DDX driver btw, not xf86-video-nouveau
19:41imirkin_: DP used to have such issues in the past (well, dpms totally didn't work)
19:41imirkin_: shouldn't matter, i think
19:41SaveTheRobots: ahh, ok
19:44martm: now i understand ... hardly matters but i have been struck back with spam
19:46martm: yeah i am sorry, quite an ...
19:47martm: uncfortable, there you see
19:47martm: that went wrong, unc...
20:03karolherbst: mwk: this looks valid to you, too? https://gist.github.com/karolherbst/82884b90b0a18098bf71ccb3626209e7
20:04b0s3d: hello :)
20:04karolherbst: no idea what it is though, but there are some patterns in the data
20:05mupuf: karolherbst, hakzsam: I am back, btw. So, I take card-swapping requests
20:05karolherbst: mupuf: no worries. I am currently searching my vbios for unknown tables
20:05mupuf: and I have very good news for piglit on ezbench :)
20:05karolherbst: what news?
20:05hakzsam: mupuf, I'm working on gk104
20:05mupuf: well, piglit can be run with ezbench, either the full run or a list of tests
20:06mupuf: I am currently limited by how long the command line can be, so I am adding a way to give a file as an input for the tests to run
20:06b0s3d: how do i change the graphic card driver on debian 8 jessie gnome3?
20:06karolherbst: mupuf: it does work for shader-db already?
20:06mupuf: and then, I plugged that into the smart ezbench mode
20:06b0s3d: i want to change it since folding@home does not agree with some due to gpu settings
20:06karolherbst: or is shader-db still to do?
20:06imirkin_: b0s3d: "change the graphic card driver"? what do you mean?
20:07mupuf: and boom, we are ready for auto-bisect of piglit regressions/improvements
20:07karolherbst: mupuf: any idea what this might be? https://gist.github.com/karolherbst/d279f97e458d193f8886a17051df6bfc
20:07b0s3d: in ubuntu/linux mint you have this option to change it
20:07karolherbst: mupuf: awesome!
20:07imirkin_: b0s3d: in any case, if you want to change your system configuration in a way that plays nice with your distro, you might want to consult a distro-specific support channel
20:07mupuf: karolherbst: what do you want me to do with shader-db?
20:08karolherbst: mupuf: track which commits adds a lot of instructions ofr example
20:08b0s3d: good answer
20:08karolherbst: if one commit increases instruction/gpc count by $input complaint
20:08mwk: karolherbst: what do you mean by valid?
20:08mupuf: karolherbst: I guess that could be integrated sort of easily
20:08mwk: looks like a pattern for sure
20:08karolherbst: mwk: well, if this is indeed a valid table
20:08martm: kinda sad about this greek guy , but i did not understand either
20:08karolherbst: mwk: the D bit table isn't parsed yet and I am adding it
20:08karolherbst: and the D bit table only has 4 bytes
20:09karolherbst: offsets to two? tables, but the first one is odd
20:09mwk: well, what you've just shown me looks like an usual bit table
20:09karolherbst: that's all I need
20:10karolherbst: mwk: the first table is here: https://gist.github.com/karolherbst/d279f97e458d193f8886a17051df6bfc
20:10karolherbst: do you have any idea about this one?
20:10karolherbst: there doesn't seem to be any header, but those lines look like entries indeed
20:10karolherbst: wow, this table is old...
20:10karolherbst: version 20 even for nv50
20:10mwk: sorry, no real idea
20:11karolherbst: then I skip it and just go for the second table
20:11mwk: maybe print first 0x100 bytes anyway?
20:11karolherbst: is something like that usualy for old cards?
20:12mwk: something like what?
20:12karolherbst: static sizes
20:12mwk: oh man, yes
20:13mwk: you haven't see the wonder that is BMP table, have you...
20:13mwk: if anything, bit is crazy about stuffing pointless sizes everywhere
20:14imirkin_: mwk: whereas bmp is fixed hard-coded offsets that vary based on ... who knows.
20:14mwk: which is good for us, not so much for nvidia
20:14karolherbst: mupuf: by the way, there are different coefficients used for voltage calculation on both our cards, both gk106
20:14mwk: pretty much yes
20:14karolherbst: I currently try to find a vbios with the second table version 0x20
20:14mupuf: karolherbst: hence why you are looking for stuff in the vbios?
20:15karolherbst: but even nv40 has a 0x20 one
20:15karolherbst: mupuf: sort of, yes
20:15mwk: you could try figuring out how to reach your table from BMP
20:15mwk: but that's not exactly fun
20:15mupuf: karolherbst: that is for old cards
20:15mwk: it's the thing that was before BIT
20:15mwk: before NV40
20:16karolherbst: do I want to?
20:16mupuf: karolherbst: anyway, shader-db tracking sounds like a good idea
20:16karolherbst: then I leave it to be bit only
20:16mwk: basically it's a huge blob, where things are arranged by offset
20:16karolherbst: except when this table is usefull for, something?
20:16mwk: and it grows a bit with every version
20:17mupuf: and the good thing is, we do not need to run it on cards. So, my main machine may be doing all the work every day to keep track of the results
20:17mupuf: sounds about good. How long does it take to run shaderdb?
20:18karolherbst: but at least I parse the tables like in power.c :)
20:18karolherbst: mupuf: uhhh
20:18karolherbst: mupuf: a while
20:19mupuf: karolherbst: yeah, good, mwk did a painful work of moving to this new separation between parsing and printing
20:19mupuf: but I never took the time to move the old stuff to it
20:19mupuf: I only added new stuff
20:20mupuf: anyway, define: "a while"
20:20karolherbst: two newest commits: https://github.com/karolherbst/envytools/commits/D_table
20:20mwk: yeah, it wasn't exactly fun
20:20karolherbst: I hope that's fine this way?
20:20mwk: I still hope to finish it some day
20:20karolherbst: mupuf: more like 30minutes+
20:21karolherbst: but the D bit tables is small though. only two entries really
20:21karolherbst: no idea what it is about
20:21karolherbst: and I was like: yeah, maybe we should just parse every table, then it is easier to find new stuff if you search something
20:21mupuf: karolherbst: 30 minutes on your machine, should be better on my desktop. In any case, it is nothing.
20:21mupuf: karolherbst: agreed
20:22mupuf: that was why I added all the UNK table in bit P
20:22mwk: karolherbst: I thought the same, but gave up on the first table I tried to handle properly [the I table]
20:22karolherbst: ohhh by the way
20:22karolherbst: unk40 is something close
20:22karolherbst: I know it is something we kind of would need
20:22karolherbst: well at least it looks like it
20:23mupuf: how many ISA versions do we need to track?
20:23karolherbst: there are the entries: https://gist.github.com/karolherbst/ae0d6c5dca8dd96808b17109f66d0868
20:23karolherbst: unk05 and unk0d
20:23karolherbst: and unk00 looks like a mode switch
20:23martm: you mean now it's my time to get some spam:), heh, iam just about to head out
20:23mupuf: imirkin_, karolherbst: if you have wishes for me to track shaderdb, just give me clear instructions and which chipsets I should run on. I will take time but I will make it happen
20:24martm: i am really sorry guys...
20:24karolherbst: mupuf: I only care about kepler, but you should know that already :p
20:24mupuf: kepler has at least two ISAs
20:24mupuf: but if only codegen changed, then maybe it is useless for me to track it
20:25karolherbst: well, that was the idea, but it would be nice to be able to ohhhh yeah
20:25mupuf: but yeah, tracking performance, shaderdb and piglit sounds like something quite useful
20:25karolherbst: ezbench could maybe just dump the shaders automatically and run them trhough shader_db and tell us: 10% perf impact, those shader changed.
20:26karolherbst: diff here
20:27imirkin_: mupuf: not sure what you mean by ISA's... SM10, SM20, SM30, SM35, SM50 are the main ones
20:27imirkin_: there are minor additions throughout the SM10 series
20:27mupuf: imirkin_: what I meant is, what chipset should I emulate?
20:28mupuf: I guess an nv50 and nvc0 would be the minimum, since they do not share the same compiler
20:28karolherbst: mupuf: is there any point in anything besides the current one?
20:28mupuf: well, the point is: Make sure we don't trash performance by accident
20:28imirkin_: mupuf: for what?
20:28karolherbst: yeah well, this is hard to tell by just looking at the shaders
20:29imirkin_: mupuf: nv50 and nvc0 do share the same compiler...
20:29martm: kinda shaking my head
20:29martm: need some cold shower almost
20:29mupuf: imirkin_: ok, so only the backend is different, the codegen, right?
20:29imirkin_: mupuf: they have diff lowering logic, and a different emitter for each ISA
20:29imirkin_: (well, SM20 and SM30 are done as one emitter)
20:30mupuf: if so, do we expect the codegen to have any significant impact on the output of a shaderdb run
20:30mupuf: that's the question I have, I never looked at the ISA and nouveau's codegen
20:30mupuf: nor have I run shaderdb even once
20:30mupuf: hence why I am asking
20:30imirkin_: isn't codegen the *only* thing which impacts shaderdb runs?
20:30karolherbst: yeah, it should be the only thing
20:31mupuf: ok, so, I would need to test the first chipset that introduced each SM version?
20:32karolherbst: mupuf: ohh so you want to check each chipset together? mhh well shader-db uses piglit for running stuff
20:32imirkin_: g80, g84, g200, gt215, gf100, gk104, gk110, gm107
20:32karolherbst: and piglit somewhat checks which gpu you have
20:32imirkin_: those are the cutoffs for various isa bits
20:32karolherbst: well it uses shader_runner
20:32mupuf: karolherbst: on intel, we can run the compiler for any gen, no need to have the actual hw
20:32karolherbst: ohh okay
20:33imirkin_: mupuf: there's no facility for faking that on nouveau... not through the GL pipeline like there is on intel
20:33mupuf: maybe it is not the case for nouveau. or the default is just to run it for your GPU
20:33imirkin_: mupuf: the compiler itself can run on whatever
20:33mupuf: ok, I see
20:33imirkin_: mupuf: but there's currently no way to access that through GL as you'd need to for shader-db
20:33mupuf: then shaderdb will be yet another benchmark to add to each machine
20:34mupuf: imirkin_: ack. I wonder if intel has an override env variable to force the detected GPU
20:34imirkin_: mupuf: they do
20:34imirkin_: mupuf: INTEL_DEVID_OVERRIDE
20:34mupuf: and since it won't send anything to the GPU, it does not matter
20:34imirkin_: can fake any pci id
20:34imirkin_: and i suspect flips on the "no hw" mode
20:34mupuf: great, should we have the same then?
20:35mupuf: yeah, very likely
20:35imirkin_: well, step 1 is... make a "no hw" mode
20:35imirkin_: right now that's not a thing.
20:35karolherbst: why not LIBGL_DEVID_OVERRIDE, LIBGL_VENID_OVERRIDE :)
20:35mupuf: it is very practical when measuring CPU consumption
20:35karolherbst: ohh right
20:35mwk: karolherbst: GL doesn't really care about the PCI ids though, it relies on chipset id from kernel
20:35mupuf: at least the mesa side
20:36karolherbst: mwk: yeah, but maybe you want to run your shader-db for nouveau while having an intel gpu
20:36karolherbst: then I wouldn't need to power up my gpu every time :D
20:36mupuf: karolherbst: ah ah
20:37mupuf: it would indeed be nice
20:37karolherbst: or maybe we should just hack shader_db to use nouveau_compiler
20:37mupuf: ok, gtg, I have a sick gf to take care of. I will read the backlog
20:37karolherbst: and make nouveau_compiler eat glsl shaders
20:37martm: i have no demands to be honest, glad you "tolerated" me at all!
20:37mupuf: this shaderdb output bisecting sounds like a fun plan
20:37karolherbst: yeah, with a lot of noise
20:38mupuf: is it supposed to be?
20:38karolherbst: kind of
20:38karolherbst: changes in the glsl -> tgsi can change some bits
20:38imirkin_: karolherbst: the simplest way to do that would be to implement the no hw mode :)
20:38martm: well cheers i gonna head off
20:38mupuf: imirkin_: you will still need to tell the LIBGL to forceload your driver
20:38karolherbst: mupuf: also by just looking at the shader-db output you can't really say if it increases perf or decreases it
20:39mupuf: of course you can't
20:39mupuf: you track metrics which can be related to perf
20:39karolherbst: I think it would be good to be able to look the diffs if ezbench found a regression
20:39karolherbst: and then just dumps it
20:39mupuf: the tool will just help you by tracking down changes to who introduced it
20:39mupuf: yeah, having the before/after shader would be nice
20:40mupuf: anyway, ttyl!
20:40karolherbst: but then again, the commit should tell you why the perf changed, mhh maybe it doesn't make so much sense to have it, allthough the idea sounds good at first
20:40mupuf: I will get piglit running nicely on the tk1 before moving on to this idea
20:41mupuf: karolherbst: as if all changes happened intentionally ;p
20:41karolherbst: but if you change codegen
20:41karolherbst: you would look at this yourself anyway
20:41karolherbst: but it would safe some time, that's right
20:42mupuf: yeah, but do you run it for all chipsets?
20:42mupuf: in any case, let's keep it in the back of our minds
20:43mupuf: and work on it if we feel the need to
20:43mupuf: in the mean time, a NO_HW mode would be great!
20:43mupuf: c y
21:02karolherbst: mwk: okay, this table seems to have 10 entries sinve g84 and 8 before that :/
21:02mwk: karolherbst: any differences in header?
21:03karolherbst: yeha I think so
21:03karolherbst: but I have no clue if this is the header: 10 0b d0 b4 20 90 31 10 12 6c c2 00 00 00 00 00 <= 10 entries
21:03karolherbst: a0 0f 20 00 31 58 1c 20 28 80 14 00 00 00 00 00 <= 8 entries
21:03mwk: doesn't look like one
21:04karolherbst: yeah, I know
21:04karolherbst: but it is super obvious if the entries are valid or not
21:04karolherbst: kind of
21:05karolherbst: ohh wait
21:06karolherbst: this is before the first entry: b2 00 56 8d 76 f1 e8 1b e9 5e 66 5f 66 59 8b e5 5d c3
21:25karolherbst: mwk: can it happen, that a table gets referenced multiple times?
21:27karolherbst: because it looks like the C table references it
21:27karolherbst: *reference the PLL table
21:28mwk: what table and from where?
21:28mwk: I know that the x86 machine code refers directly to various tables in places
21:28karolherbst: on two vbios
21:28karolherbst: on mine and martins vbios, the C bit table has a reference to the PLL table
21:29mwk: but I haven't heard of stuff reachable via two paths from BIT
21:29karolherbst: from where is the PLL table usually referenced?
21:29mwk: I doon't remember
21:29karolherbst: mhh same on my nvc1
21:30karolherbst: u16(C+8) references to PLL
21:31karolherbst: the former reference is from the C table
21:32karolherbst: just done within main in nvbios.c
21:32karolherbst: I C and U seems to be handled there..
21:32karolherbst: good to know
21:35karolherbst: L has a valid table
21:35karolherbst: version 0x40...
21:35karolherbst: and a small one
21:35karolherbst: funny table though
22:02karolherbst: mhh https://gist.github.com/karolherbst/fec46e17bd95e1063556e19fa62b9181
22:06karolherbst: it seems to have always 16 entries
22:07karolherbst: and the first byte seems to be some kind of switches