06:33hexedit: hello i have a problem with the nouveau driver on an nvidia quadro fx2000 card in the 800x600 mode
06:36hexedit: is this the correct channel to report nouveau bugs
06:41orbea: hexedit: go for it, someone will probably reply
06:41hexedit: orbea ok...i was able to get some decent video with another nvidia geforce chip but the quadro is my favorite board however the driver is not too good at this time
06:43hexedit: orbea and the nvidia 173 drivers are useless as they won't compile with linux4.1 headers
06:44hexedit: orbea...so uh basically nouveau is all there is at this time and it is not in good shape for 800x600 work
06:48hexedit: orbea..at this time i have simply been forced to retire the quadro as useless junk for now...and had to go back to a much older slower geforce.\
06:49hexedit: orbea...until such time as a better driver comes along if it does.
06:49martm: good morning karolherbst: i looked at trello, there are basically two entries there, which have been my primary focus, but i am not sure if i will be able to help there..i am a crank like nvidia dudes, taking the code explaining only a bit, and then go like, everyone implement it like they want
06:54hexedit: orbea...i can't see how nouveau people could overlook that the screen is viewable at 1024x768 but totally painted out at 800x600
06:55hexedit: orbea..it is beyond me how they could overlook that
06:56martm: karolherbst: but at this point i can congratulate the programmers, if one can think there is nothing to be afraid of anymore i belive, remember my expertease will be focused on programmable devices, so i have not been looking into reclocking on asics, i could help a bit on the scheduler stuff, like in concept
06:56martm: when you get stuck in times in pm, but i belive you are quite intelligent you can figure it out too
07:02hexedit: it is such a relief to finally have an os that is free from unix royalty liabilities
07:03hexedit: even though not too much is working at this time
07:03martm: karolherbst: multithreading may be easy if the mentioned patch would function if anyone has time to test it, following roughly the line count which might be primary indicator for complexity
07:04martm: then the scheduler falls into group of hard infact, it's not hard for the brain, but it totals thousound of lines indeed at least
07:05hexedit: i have been reading some of the logs mentioned in troubleshooting but mostly all i find is communication among programmers about shader language computations and i don't really see anything about unusable screens at 800x600 ...is there a way to search for this topic
07:05martm: karolherbst: do you know what means maxwell sched codes that phoronix and hakzsam have been talking about?
07:10karolherbst: martm: yeah, since kepler the scheduling stuff is encoded into opcodes in the uploaded binaries, nobody really knows what they mean though, I think, maybe there is somebody with a good idea about it though
07:15hexedit: it is such a relief to see full-blown new computers like raspberry pi zero and many others selling for only $5.00 with a free operating system....at least finally hardware and software will no longer cost more than $5.00....since if they can make brand new computers for $5.00 ...then people know that everyone can...and no one wants to be price gouged by any higher prices than that
07:17hexedit: and since smartphones are computers also...there is no need for their prices to be more than $5.00 either.
07:18pq: ...until you want a specific feature
07:18hexedit: pq...you mean you have to pay extra for extra features.
07:21pq: hexedit, yes - but that can often mean any of 5$ computers won't do at all, you need completely different design, and then the price goes to 10x - 100x
07:22hexedit: pq..you mean the sky is the limit on how much extra you have to pay beyond a simple $5.00.
07:23pq: say you wanted a different GPU for any reason. Or a SATA port. Or... requirements that are not too wild, but just invalidate the whole category of available 5$ units.
07:24hexedit: qp...well what if you had a free old unused sata drive...why couldn't you just use a sata to usb adapter and make use of it with the raspberry pi zero
07:25pq: because the usb on rpi is traditionally of crap quality
07:25hexedit: pq...ok so you think it might have data integrity problems
07:26pq: I mean it sucks, even when it works correctly.
07:26hexedit: pq..oh....well i have not tried it so i certainly cannot argue the point without any evidence.
07:27pq: hexedit, saying "they are all computers" is like saying "they are all trasports" when you refer to bikes, cars, and aeroplanes.
07:28hexedit: pq...yes well all computers have one thing and one thing only that they all do...and that is to compute...that is why they are called computers.
07:29hexedit: pq..and if you need to make a computation for any reason...why not make it on a $5 computer instead of a $100 million dollar supercomputer or mainframe
07:30pq: because you want the result today, not next year
07:30hexedit: pq...ok so teraflops is an important concept to you...i take it you believe time is money.
07:31pq: wow, you really are trolling. Goodbye.
07:32hexedit: pq...no i am not trolling...i have a specific problem with a nouveau driver and i am just waiting for some developers to show up that might know how to resolve it
07:33hexedit: pq..i just quoted some facts that are all over the internet...you can see prices for raspberry pi and others everywhere on the internet
07:33hexedit: pq..and you can make dogbone clusters with racks of $5 computers if teraflops is important to you at not much additional cost
07:34hexedit: pq...think beawolf clusters for example
07:35hexedit: for massively parallel computations where speed can be obtained in many cases.
07:36hexedit: and you are still talking less than a hamburger at mcdonalds
07:39pq: Sorry for picking up a completely off-topic bait.
07:40hexedit: pq...it is not bait...you can answer however you wish...i am just killing time waiting for some developers to show up
07:49hexedit: but it is clear to most people that the days of making money from hardware or software is over....when you can get brand new computer for $5 with a free operating system that supports 50000 free applications
09:23karolherbst: another deadlock in nvkm_pmu_send
09:57karolherbst: 55ms (nouveau) vs 43ms (nvidia) in pixmark_piano
09:58karolherbst: stock mesa: 57.8ms
10:04karolherbst: mupuf: with ezbench you can also find performance improvements in a collection of benchmarks automated?
10:38hexedit: any nouveau developers here today
10:50karolherbst: hexedit: well depends on what you wnat
11:20hexedit: karolherbst...ok thanks for responding...i have a quadro fx 2000 and at 1024x768 the screen looks ok...but at 800x600 it is just painted a couple of colors and you can't see any fonts or anything
11:21hexedit: karolherbst..so i was wondering if any developers had any ideas how to make nouveau a little more useful
11:21karolherbst: uhh that's like ancient hardware
11:21hexedit: karolherbst...yes it certainly is....old but not forgotten
11:21karolherbst: does dmesg tell you anything?
11:22hexedit: karolherbst...at 800x600 i can't read anything on the screen and at 1024x768 the screen is ok but my monitor can't handle it so i don't leave it on very long
11:22karolherbst: you could use ssh
11:23hexedit: karolherbst....you mean jump out of x back to a console
11:23karolherbst: no, use a second machine and connect via ssh
11:23hexedit: karolherbst...well i am using the second machine for communications until i get the first one working...it has been a long ordeal
11:24hexedit: karolherbst...i could jump to a console out of X and get some info...but most likely it would be in Xorg.0.log
11:24hexedit: karolherbst...rather than journalctl -k
11:24karolherbst: so the corrupted screen disappeasr when you switch back to a tty?
11:24karolherbst: well most likely it also changes the resolution or someting
11:26hexedit: karolherbst...yes when i change the resolution from 1024x768 to 800x600 the screen becomes unusable and when i change back using xrandr it becomes usable again....but not for long because my monitor can't support those ridiculous resolutions
11:26karolherbst: yeah well, you should get something in dmesg maybe or in the X log
11:26karolherbst: or maybe something else is odd
11:27hexedit: karolherbst...well i dont know if it is a modeline error or what....but the driver needs a little work for that chip
11:28hexedit: karolherbst...the driver does seem to work with a much older geforce fx5200 but it is of course much slower than the quadro
11:28karolherbst: we can't help you without the logs
11:28hexedit: karolherbst...at least for videos and having a useful screen
11:28hexedit: karolherbst...well i can provide any info you request but it takes time to change back to the quadro from the geforce
11:30karolherbst: yeah and we would need the complete X log and dmesg after the mess up and maybe the vbios too (/sys/kernel/debug/dri/0/vbios.rom)
11:30hexedit: karolherbst..ok i will get you that info ....where would i paste it so you can find it on the web
11:30karolherbst: some pasting site
11:31hexedit: karolherbst well i have not used the pasting sites in a long time so i am not familiar with the sites now
11:32karolherbst: well I don't care which one as long as it works
11:33hexedit: krolherbst...well it has been a long time...let me see what i can google up today
11:39hexedit: karolherbst..it looks like pastebin.com or justpaste.it might work these days
11:39hexedit: karolherbst...i will have to go get the data for you which will take a little time
11:40martm: dpaste.com is popular, as i remember
11:41martm: very nice and clean version, i like it
11:42martm: karolherbst: well fermi and kepler are beatiful cards, i really did an improved version of the schedulers idea on them, playin with the cache, i am not sure if that works below nv50 though
11:44martm: i think it would work on all cards, i really almost can forget all the previous ramblings, i thought about whole new scheme
11:48martm: as it is reordering the cachelines while executing, i am thinking it may have some coherency issues though
11:54hexedit: karolherbst...ok there will be some delay as the machine is still busy compiling firefox....it has been several hours so i hope not much longer.....and also my suspicion is that the nouveau driver is ignoring the nvidia chip...and a default vesa driver may be taking over and not working very well
11:55hexedit: karolherbst...so that is what i am thinking without giving you hard evidence yet
11:58hexedit: karolherbst..fortunately i don't have to change the driver or recompile the kernel..as both cards attempt to use the driver first...so it is just a matter of swapping the cards and getting the data after this firefox jazz finishes
11:59martm: now this one is highly simple method, i belive it would go as medium or easy in trello
12:03martm: we'll discuss this, well you leave the order the same as they appear coming from opt, but the header of the shader..recurses in cache, pinning branches one after another for every command in code, since cbranch takes an immediate, it starts to jump to the end of the cache, but slot before the end, the end goes jumps back
12:04martm: now when it goes into recursion, the branch will be changed by reordering a new data to that slot by index
12:06martm: very very simple hack, and probably should work, you just generate the correct masks in the header
12:06hexedit: karolherbst...as i recall the nouveau driver just seems to be looking for certain geforce chips and ignoring any quadro chips but the logs will have that info for you
12:08RSpliet: hexedit: false. nouveau doesn't care whether a chip is GeForce or Quadro
12:09hexedit: RSpliet...well it certainly puts out data regarding several geforce chips and the nv3,4x notations...and is silent on quadro modells
12:09hexedit: RSpliet...so it is not false as you will get a chance to see the reports when i can post them to the web
12:10RSpliet: hexedit: I know that the DDX driver outputs that kind of nonsense into the X.org log, but that can be safely ignored
12:11hexedit: RSpliet..well that is fine if the info can be ignored but my suspicion is that the driver is ignoring the chip...so i will get the evidence for one or the other as soon as the firefox compilation finishes
12:12hexedit: RSpliet...then there can be some hard facts to find out what is really going on
12:21martm: ouh dear, months of wasted time, this thing so easy
12:21martm: is so
12:40hexedit: SaveTheRobots..what a nick....you must be in agreement with the kara video demo
12:50SaveTheRobots: hexedit: that video was intense.
12:50SaveTheRobots: makes me wanna watch Ex Machina again
12:50hexedit: SaveTheRobots...you like the video did you...i hope to see fascinating things coming from quantic dream in the future if they stick with it
12:50SaveTheRobots: karolherbst: i think you highlighted me a while back, can i help with something?
12:52SaveTheRobots: hexedit: yeah, that was a pretty cool vid :p did it get turned into a game or was it just a tech demo?
12:53hexedit: SaveTheRobots..quantic dream says they want to work on a game with that basic idea as a launching point...i hope to see interesting things from them in the future.
12:54hexedit: SaveTheRobots...right now most of the attention is on Laura Croft TombRaider Rising...and a lot of people are waiting for the hidef remake of FF9 which is promised for later
12:55hexedit: SaveTheRobots..and Fallout4 just won 1st place in serious competition so i guess they are glad to finally receive recognition for their efforts to create substantial storylines with hopefully something to learn for the players
13:22hexedit: SaveTheRobots...if you think kara video was intense you might also see if you are alive by examining lirum's death demo in Lost Odyssey and related saving Sarah video in Lost Odyssey..
13:23hexedit: SaveTheRobots ...you should be able to tell right away if you are in fact a real human being.
13:25hexedit: SaveTheRobots...of course it is not about Robots per se...but still has some lessons to learn about life.
13:31hexedit: i should say lessons about life for the players to learn
13:35hexedit: karolherbst..ok it looks like firefox is finally concluding after many many hours of compilation....so i will start to get you that info now
13:55hexedit: karolherbst...ok it is booted into the quaddro..and i will try to post the files from that machine
14:01hexedit: karolherbst...ok someone suggested that i use dpaste.com so i gave it a try.
14:02hexedit: karolherbst...here is the xorg log file first for the quadro ...you should be able to view it at http://dpaste.com/2GZVCQ4
14:02hexedit: karolherbst...did you get that
14:05RSpliet: hexedit: looks like nouveau functions correctly, and is configured to display in a resolution of 848x480 as reported by your monitor
14:06hexedit: RSpliet..yes it is working at 848x480 at least for screen uses and possible hidef video..let me check a tough hidef video file
14:07hexedit: RSpliet..the monitor can easily handle 800 or so and this is quite close
14:07RSpliet: the monitor even report 1024x76860Hz through EDID
14:09RSpliet: but 800x600 doesn't work?
14:09hexedit: RSpliet..i don't have EDID capability...that is for expensive new monitors...not old junk
14:09RSpliet: see lines 159 to 165
14:09hexedit: RSpliet....ok this is a very old cheap packard bell monitor..in the days of windows3.1
14:10RSpliet: did you also paste your kernel log somewhere? that'd be helpful
14:10hexedit: RSpliet...no but i will paste that next if you think you can learn anything extra from it regarding video...but please don't steal my other hardware ideas
14:11hexedit: RSpliet...as you know journalctl -k gives out a lot of hardware detail
14:11hexedit: RSpliet...i will post that next as you wish
14:14RSpliet: hexedit: thinking one step ahead: you mention it's a pre-1995 monitor, yet the log mentions DVI-I. Are you using a convertor to connect the monitor to a DVI port?
14:14hexedit: RSpliet...just give me a minute for the hi def video to finish so i can tell if there are any skipped frames or audio pauses.
14:14hexedit: RSpliet...how wise of you to notice and yes i am using an adapter
14:15RSpliet: did you use the same adaptor for your GeForce FX5200?
14:16hexedit: RSpliet..no it does not need it ..it has a normal vga out
14:17hexedit: RSpliet...both are great boards....but i like the quadro better because on tough stuff it really is much faster than the geforce
14:19RSpliet: it is. The converter (possibly in combination with nouveau) might be the cause of these problems though. Did you happen to capture an Xorg log from the FX 5200 too?
14:19RSpliet: anyway, sorry, take your time to capture the kernel log first
14:26hexedit: RSpliet...ok here is the info you requested from journalctl
14:26hexedit: dpaste.com/2NEON8O...i think it is capital o not zero
14:27hexedit: RSpliet but i am not sure ...you might try both
14:27RSpliet: I tried various combinations, but none seem to work
14:27RSpliet: ah, both were zero
14:28hexedit: RSpliet...ok so you found it
14:28RSpliet: that's not a kernel log
14:28RSpliet: oh sorry
14:28RSpliet: it went on further
14:28hexedit: RSpliet...it is the output of journalctl -k ..it is even more than a kernel log
14:29hexedit: RSpliet..you should be able to see linux kernel 4.1
14:30RSpliet: interestingly, no EDID errors
14:30RSpliet: (which I would expect if your monitor doesn't support EDID)
14:32RSpliet: yes, I would like to see the Xorg log from your FX 5200 please
14:32hexedit: RSpliet...no i think it just bypasses that...it is a packard bell 1512sl if you need the specs for any reason
14:32RSpliet: you might be able to retrieve an old log rather than reboot with the other card, whichever is easier for you
14:33hexedit: RSpliet..i will see if any are still hanging around from the 5200
14:39hexedit: RSpliet..ok this should be the 5200 geforce...see dpaste.com/1RO3XA3...and try big o or zero
14:40hexedit: RSpliet...did you find it
14:41hexedit: RSpliet...ok i hope it has the info you are looking for
14:41RSpliet: interestingly, I also found an EDID for your monitor. Seems it is new enough to have one
14:42hexedit: RSpliet...well if the specs say it does....then i am quite surprised..but i accept the tecnical specs as accurate
14:42RSpliet: oh I wish I could find the specs, no I found a guy with a database of EDIDs that listed your monitor
14:43RSpliet: either way, your two GPUs detect the same modes
14:43RSpliet: and then you say the Quadro FX doesn't properly display 800x600, whereas the FX5200 can do this
14:45RSpliet: (or did I misunderstand your question there?) I'm inclined to point at your VGA->DVI converter as the culprit, given how none of the logs indicate any errors
14:45RSpliet: the only way I can think of verifying this
14:46RSpliet: is by plugging in this monitor using the VGA->DVI-I converter in your FX5200 and see if you get the same problems
14:47martm: my brain is getting tired very much, it's not entirely sure if the best version of the idea would work, I'll recap tomorrow
14:48hexedit: RSpliet..ok well your suspicion might be right on...but i don't see any way of using the adapter on the 5200...and yes the 5200 displays correctly with no problem
14:49hexedit: RSpliet..it is just a little slow compared to the quadro
14:49RSpliet: hexedit: the logs indicate the FX5200 has a DVI-I port, does it not?
14:50hexedit: RSpliet...yes it certainly does...i did not notice because i always use the vga out on it
14:50RSpliet: try plugging the converter on to that one, just for testing purposes
14:51hexedit: RSpliet..ok it will take some time to do some hardware swapping again...and this via computer is tied up with radeons and windows....so i don't like to change anything on it
14:52RSpliet: sure, in your own time!
14:54hexedit: RSpliet..i am just now getting the via system stable after 10 years of tinkering with it...so i don't like to change it very much.
14:54hexedit: RSpliet...so i do most of my hardware work on the intel chipsets
14:56hexedit: RSpliet....and sometimes i use the amd processor for older freedos board repairs
14:57RSpliet: I don't expect behaviour to change with the motherboard/CPU you use, just the GPU. You can try on any computer (provided you have a working Linux installation with nouveau)
14:58hexedit: RSpliet...well it usually is no problem to get a working linux install if you use linux from scratch or gentoo or slackware....it is just very time consuming to build all the code
15:03imirkin: hexedit: distcc :)
15:05hexedit: imirkin yes i thought about getting a few dogbones and stacking about 50 $5 raspberry pi zeros and cross-compiling from the cluster for a single processor computer
15:06imirkin: the common thing is to have a more modern box sitting around
15:06imirkin: but if you don't, then that's less viable.
15:07RSpliet: in fairness, a RPi is a more modern machine... and in all likelihood faster as well
15:07hexedit: imirkin...but the thing is i usually spend the most time tracking down compilation errors...and that really is the bottleneck...but once i get a compile i don't have to change it very often
15:08hexedit: imirkin..like this developer version of firefox38esr...i probably won't have to build another for some time
15:09imirkin: whatever works for you
15:10hexedit: imirkin...well actually i use several things...i also have a back up browser based on webkit-gtk3 called epiphany to try to maintain some compatibility with all the webkit-qt derived stuff
15:11martm: RSpliet: damn i like the cache stuff, i completed a very neat scheme, basically the issue is that cbranch on radeon takes an immediate, but actually when the stuff is in cache, the immediate can be overwritten
15:11martm: making the unconditional branch think that it had several immediates, and letting it to pin another branch from the header
15:12hexedit: imirkin ...i was going to build another plasma platform...but i get an error that qtdbusmenu for qt5 is missing an ebuild and i haven't figured out how to tell it to just use qt4
15:12martm: so first time you do that, you pin everything as NOP's to offsets that you bitwise in the header recursively
15:12imirkin: good luck. this is the wrong chan to be discussing that though.
15:12hexedit: imirkin..so i just built a mate desktop to get something goin
15:14hexedit: imirkin i am ok with mate or plasma...since i dont have the heavy duty video hardware to use gnome3
15:14martm: you take masks from cache and pin stuff as NOP's then remove the masks, use the cache for something else, and copy over the data at those offsets
15:15hexedit: imirkin i leave all that gnome3 stuff for redhat fedora fans
15:16imirkin: again... off topic. this chan is about nouveau. not your desktop preferences.
15:16hexedit: imirkin...well nouveau is the backbone of a lot of it
15:16martm:goes to kde5 sometime later, which is my favorite, but cinnamon is damn awesome too
15:17hexedit: imirkin..in some real sense nouveau is about as important as the kernel...you can't do much without either one
15:18martm: yes, i think i have it, but need to test...i belive we do not need those sched codes yet
15:19hexedit: imirkin because for example...these boards would use the nvidia 173 drivers which no longer compile with kernel 4.1 headers and gcc4.9
15:19hexedit: imirkin...so nouveau becomes a very critical backbone piece of software
15:21martm: hexedit: well fx5200 i have two of them, but actually that card little bit sucks
15:22hexedit: martm...no it does not...it supports zsnes,psemu, fceu, and thousands of other games as long as you paid for them
15:22martm: hexedit: the code for this can be improved somewhat too for the perf, but i don't think anyone has time to do such things anymore
15:22hexedit: martm...well the performance is outstanding as far as i am concerned
15:23martm: hexedit: yeah it's ok i used that card some time ago, and it was almost going back then too, but it's an old card, the performance can be raised for it lot further too
15:26hexedit: martm...yes no one argues against perf improvements...but i am saying the card is excellent for supporting thousands of games and many many types of graphics worlds....it is just that some people wanted to add some functions to gl..so they changed from fixed function to programmable functions using shader languages to implement various graphics features.
15:27martm: hexedit: yes of course, but the programmable pipeline if you know anything about it, works much faster too
15:27hexedit: martm ...i personally don't see the need for a couple of extra graphics functions....i think excellent stories with outstanding symphonies can be created using the fixed function platform for useful games that inspire learning and creativity
15:28hexedit: martm..but the gl people never seem to get tired of creating new graphics stuff
15:29martm: hexedit: ok ok, i see where you getting, fx5200 is ok, but my card is a lot better, fx5200 helped me too to learn and stuff
15:32martm: do they say that NOP does zero cycle execution or what, it's prolly then always at the top of the decoder right in hw?
15:35martm: hexedit: if i get my projects running at some point, i'd be almost willing you to donate a low end gt730, but sounds like you don't have the tower to put it into, but it would take years time for me to get there
15:39martm: allthough ideal would be to finnish the job on this card, and switch it against radeonsi:)
15:41martm: imirkin: are you still ignoring me? what you think about the review and testing of mlankhorst code that you refer to in trello?
15:48martm: RSpliet: i thought through a lot easier method, than that before, i am getting into peoples nerves i see with my ramblings i seem to be really happy, despite the fact that for me there is too much to do to get into some position
15:49martm: anyways, well have a convo about this later, i got tired, but i think i got it right in my head for easier scheduling csheme
15:55martm: i have some literature about how branches work in hw, so i estimate that my hack should work according to hose, maybe i post a theory tomorrow to dpaste.com
15:57martm: but in theory if that is l1 cache i play with, when iread and write to it using different threads, there could be issues, so...i try to do it prefectly
17:21karolherbst: SaveTheRobots: well you could try out my new patches
17:45SaveTheRobots: karolherbst: ok, i'll give them a go when i get chance (hopefully today)
17:46martm: http://arxiv.org/pdf/1509.03118.pdf i don't understand how one can transfer from memory to cacheline with 5.3clock cycles
18:46martm: TB/s BW shaMEM___ 4B * 32banks * 15 SMs * half 1.4GHz = 1.3 TB/s only on FERMI
19:08karolherbst: k, back to dynamic reclocking then
20:10szt: hello guys, I have a dmesg now
20:10szt: remember nouveau was freezing for me
20:14karolherbst: ohh so chromium does something odd
20:14karolherbst: szt: can you verify if it is always chromium?
20:14szt: no I can't
20:14karolherbst: but mmting chromium is a really painfull :/
20:14szt: but it might just be chroium
20:15karolherbst: could you try to narrow it down do something?
20:15szt: tell me how ._. I'm sorta lost
20:15karolherbst: szt: memory mapped tracing (or something like this)
20:15karolherbst: maybe it always crashes doing the same thing?
20:15szt: the crash feels sorta random
20:16karolherbst: how long does it take?
20:16szt: well sometimes it takes days
20:16karolherbst: I could try to run chromium on my gpu and see if something happens
20:16karolherbst: I see
20:16szt: then again sometimes it happens right after a reboot
20:16szt: It might as well be my gpu
20:16karolherbst: don't think so
20:17szt: well chromiums hw accelleration used to crash using the old driver
20:18szt: now the acceleration doesn't crash anymore but nouveau deadlocks my xorg
20:18szt: if only there was a way to get around having to reboot
20:18karolherbst: nouveau could try to restart the gpu and move on
20:18karolherbst: I think skeggsb worked on that recently
20:19karolherbst: szt: in chrome://gpu
20:19szt: yeah what about it?
20:19karolherbst: stikonas_: do you have Canvas, Native GpuMemoryBuffers, Rasterization and Video Decode set to software?
20:20stikonas_: karolherbst: me?
20:20stikonas_: are you sure it's not somebody else
20:20karolherbst: stikonas_: sry
20:20stikonas_: it's ok...
20:20stikonas_: don't worry
20:20karolherbst: I mean szt
20:21szt: karolherbst: https://i.imgur.com/3CMQpFJ.png
20:21karolherbst: szt: k, you don't have 50 yet?
20:22karolherbst: I have chromium 50.0.2661.66 installed
20:22szt: what distro?
20:23szt: I'm using 49
20:24karolherbst: yeah well I just run chromium a bit on my nvidia gpu and see what happens
20:24karolherbst: but in the end I don't really know how to debug this, cause it can be caused by everything really
20:24karolherbst: and having a mmt of a 2 or 3 days run is pretty insane
20:25karolherbst: stikonas_: do you have xorg-server 1.18.2 installed?
20:25szt: what gpu do you have?
20:26szt: never heard of
20:26karolherbst: it is a 770m
20:26karolherbst: the chip is gk106
20:26szt: mine is a 550 Ti
20:30szt: karolherbst: I disabled the gpu flag override for now
20:30szt: I'll see if that fixes anything
20:30karolherbst: szt: well
20:30karolherbst: szt: they blacklist stuff because the driver does something stupid
20:30karolherbst: szt: and we should rather fix it :D
20:30szt: so maybe the flag override causes the crash
20:31szt: it doesn't sound that far fetched
20:31szt: can I get into development?
20:31karolherbst: everybody can
20:31szt: well yeah.. it's free software after all lol
20:31karolherbst: I am only here like since mid last year?
20:37szt: karolherbst: I just cloned the repo
20:37szt: git://wiki.freedesktop.org/wiki/nouveau <-- that one?
20:38karolherbst: that's the wiki
20:38karolherbst: szt: https://github.com/skeggsb/nouveau
20:38karolherbst: but this is against 4.6 (well tehre are branches htough)
20:38karolherbst: and you have to run make inside drm
20:39szt: so it's not "easy"
20:39karolherbst: some stuff is easy, some is difficult
20:39karolherbst: depends really on what you plan to do
20:39szt: well I just want to look into it
20:40szt: and get an idea on how things work
20:40karolherbst: you should decide on a issue you want to fix on your gpu and we could try to guide you or maybe somebody already looked into that and got distracted with something else
20:40karolherbst: szt: well I never did this
20:40karolherbst: usually you will find the right places quite easy, so there is no need to have a bigger picture of the entire kernel module
20:41karolherbst: szt: ever developed for the kernel?
20:41szt: nope never
20:41karolherbst: then you will run into other issues first anyway :D
20:41karolherbst: not understand the code is the smallest issue
20:41szt: I've been using linux for about 3 years now but never got into development
20:41szt: I've mainly been a user
20:41karolherbst: well, it isn't that hard, but you can just randomly crash your kernel if you aren't carefull
20:42szt: s/(development)/kernel \1/
20:42karolherbst: usually you want to have a second machine or your main gpu being not driven by nouveau
20:42karolherbst: otherwise it is a bit risky
20:42szt: I imagine a VM with VGA passthrough would be a possible solution
20:43karolherbst: I have an optimus laptop, so even if I mess up nouveau, I always have the intel one
20:43karolherbst: and nouveau never gets loaded on boot
23:50karolherbst: mupuf: maybe we should really have a look at the engine clock gating. on 07 psatate: 9.8W => 9.0W 0a: 16.97W => 15.17W 0f: 21.11W => 19.30W
23:52RSpliet: karolherbst: do you mean the block-level clock gating (the blcg regs)?
23:53Jayhost: karolherbst built newest maxwell_reclocking on 4.5 Pstate change is different?
23:53karolherbst: RSpliet: PTHERM regs
23:53karolherbst: RSpliet: 20200-20260
23:54karolherbst: RSpliet: it is really trivial on my kepler to get it working, just one reg poke
23:54karolherbst: Jayhost: yeah, I updated the branch today
23:54karolherbst: Jayhost: and added a lot
23:54RSpliet: if it's equally simple on other cards, propose a little patch I'd say?
23:55karolherbst: RSpliet: well, I have a mobile gpu, maybe something is required on desktop ones, because it isn't turned on by default
23:55karolherbst: but I will check on reator
23:55karolherbst: now that the power sensors code is merged, we get reliable power readings
23:55karolherbst: Jayhost: you mean my nouveau repository, right?
23:56karolherbst: yeah, I think I am nrealy done with everything, just fixin comments if anybody else complaints :D
23:56karolherbst: RSpliet: sadly, this gating only gives us a benefit while fully idling. on high load the difference can shrink to below 0.1%
23:56karolherbst: and on fermi this is much more complex to deal with
23:57karolherbst: and with kepler+ the gpu is usually turned off
23:57karolherbst: anyway, dynamic reclocking will yield a bigger power savings than this
23:58karolherbst: anyway, I am off to bed
23:58karolherbst: Jayhost: have fun trying out my newest version oft the branch
23:58karolherbst: Jayhost: I am sure everything will work out fine, but maybe there is something missing somewhere