01:43 karolherbst:
01:49 AndrewR: imirkin, GLexcess demo works if I set X to depth 16. (with my hack reverted). But then ..I can live with broken demo for some time, obviously. At least bug in bugzilla and hopefully someday it will be solved properly.
02:26 karolherbst: shall I add me to the authors for each file I modify by the way?
02:32 pmoreau: Absolutely no idea. So far, I've only added myself to files I added.
02:34 karolherbst: k
02:34 karolherbst: okay, my second version of the patches will be much cleaner :D
02:35 karolherbst: pmoreau: I should include your stuff somehow
02:35 karolherbst: because I add a func.pcie.init functions
02:35 karolherbst: but well, my pcie_init is different for tesla and fermi
02:40 pmoreau: :-)
02:43 karolherbst: is there a nice way to add those v2 in git send-mail? :D
02:45 pq: karolherbst, usually you add oneself to the copyright holders of a file where you make a significant contribution. Not for every little thing.
02:45 pq: karolherbst, yes, -v2
02:46 pq: karolherbst, or for git-format-patch if you call that manually first
03:03 karolherbst: no, I usually call git send-mail directly
04:03 karolherbst: RSpliet: I've posted a new version and I hope I respected all your concerns :D
04:40 karolherbst: mupuf: how often should I poll the power consumption?
04:41 AnAkkk: hey
04:42 AnAkkk: RSpliet: I saw that reclocking now works on G94 and up, is there any chance to have G92 supported to make it work on 8800 GT?
04:43 karolherbst: AnAkkk: there always is :p
04:44 AnAkkk: of course, but you know what I mean :p
04:44 karolherbst: AnAkkk: "as it's only been tested on a single G94 and one G96." from the commit message
04:44 karolherbst: maybe it already works for you
04:44 karolherbst: you can replace the 0x94 with a 0x92 and test it out
04:45 AnAkkk: yeah..I don't know if it's safe enough
04:45 karolherbst: ohh wait
04:45 karolherbst: well, it shouldn't break your card if that's what you mean
04:46 AnAkkk: well, when I tried it once it made my screen flicker, and after a day or so, my screen died
04:46 karolherbst: RSpliet: your http://cgit.freedesktop.org/~darktama/nouveau/commit/?id=28a8611c331464367a638abbba38914dd883f1f1 patch seems toa lso touch G92
04:46 karolherbst: AnAkkk: died like it bricked or it just went off?
04:47 AnAkkk: wouldn't work at all, just stay black
04:47 AnAkkk: had to buy a new one
04:47 karolherbst: mhhh
04:47 karolherbst: strange
04:47 karolherbst: usually something like that should not happen
04:47 AnAkkk: yeah, well I don't know if it was related to reclocking but it happened not long after I started testing it
04:47 karolherbst: maybe the display was just not well made?
04:48 karolherbst: it could be that it was upset about the signals it got
04:48 AnAkkk: no, I tried everything I could to fix it
04:48 AnAkkk: even the monitor controls where no longer showing up
04:48 karolherbst: but usually dispalys shouldn't break from that, because they get something silly on the cable
04:48 AnAkkk: (it used to display a panel)
04:49 karolherbst: if this would be normal, some nouveau devs would have bought a lot of displays/laptops by now :D
04:49 karolherbst: AnAkkk: do you know how to build your own kernel or nouveau?
04:49 AnAkkk: yeah, well I don't know :p I just want to know what RSpliet think about the current G92 support before I try it again
04:49 AnAkkk: yes
04:49 karolherbst: then you could just try it out once, and if it doesn't work, well it doesn't work then
04:50 karolherbst: :D
04:50 karolherbst: k
05:02 pq: sometimes monitors do fail on their own, I RMA'd an Acer this year that still had a few months of guarantee left.
05:03 pq: I don't think a bad signal could fry a monitor since we stopped using CRTs
05:04 pmoreau: karolherbst: Patch send time? :-)
05:05 pmoreau: karolherbst: Patches are migrating to the ML due to winter coming?
05:05 karolherbst: well, don't know, ML seems like more official
05:06 karolherbst: I start to dislike MLs even more now :D
05:06 karolherbst: on github I can simply remove comments I've taken care of, but with a ML it is a bit more complicated to do that
05:06 pmoreau: Sure, it's just that you just sent, in 24h, like 3-4 patch series
05:07 karolherbst: right :)
05:07 karolherbst: and now I won't for another two months :D
05:07 pmoreau: :D
05:08 pmoreau: Well, 3-4 patch series every two months is still quite nice!
05:08 karolherbst: I need more tesla testing for the pcie stuff though :/
05:08 karolherbst: I am pretty sure it won't break anything, but I am not sure if it works for most of the cards either
05:09 pmoreau: :/
05:09 karolherbst: I tested it on the gt218 from mupuf
05:09 karolherbst: but this card did only 2.5
05:09 karolherbst: but had v2 support
05:09 karolherbst: so yeah
05:10 karolherbst: at least the v1 => v2 transition works
05:10 karolherbst: that's what I am sure about
05:11 pmoreau: IIRC, most of my Tesla cards only have one pstate, and the few that have multiple ones only do 2.5
05:13 karolherbst: :/
05:14 karolherbst: I don't think my patches do something usefull for g9x cards anyway, because their vbios doesn't state any pcie speeds
05:14 karolherbst: so the card will stay at the speed it booted
05:15 pmoreau: Maybe the GT200 could benefit from it
05:16 karolherbst: as long as the vbios states it, then yes
05:16 karolherbst: it needs PM_mode table version 0x40 or higher
05:17 pmoreau: I'll have a look
05:22 pmoreau: Don't have the vbios at hand... It is in the repo if you want to have a look.
05:26 karolherbst: nope, version 0x35
05:27 pmoreau: Too bad
05:28 karolherbst: yeah I think gt215 is the first one
07:24 karolherbst: yeah, setting all clock gates to auto: 10.4W => 9.6W in full idle
07:24 karolherbst: with no perf penalty
07:25 karolherbst: pmoreau: can you see your battery usage in W on your laptop?
07:25 karolherbst: in powertop for example?
07:27 karolherbst: checking my fermi card
07:29 karolherbst: wow
07:29 karolherbst: hard to tell if it is working if the gpu can't tell you
07:36 pmoreau: Let me see...
07:36 karolherbst: mupuf: the gates are also working on the fermi card I got here
07:36 karolherbst: putting all to on, increases power usage
07:37 pmoreau: Do you want specific conditions or just if I'm able to see it?
07:37 karolherbst: putting all to auto seems to be slightly better then eng=run/blk=auto
07:37 karolherbst: pmoreau: able to see it
07:37 karolherbst: mupuf: the difference is bigger when nouveau is loaded
07:37 pmoreau: Discharge rate of 17.6W
07:38 karolherbst: around 2W on this fermi from 0x44 to 0x55
07:38 karolherbst: pmoreau: nice
07:39 karolherbst: nvapeek 0x20200 0x60
07:39 pmoreau: Wow, and that's with the discrete powered off and the igd on lowest clock...
07:39 pmoreau: You meant poke I guess ;)
07:40 karolherbst: no
07:40 karolherbst: peek
07:40 pmoreau: Oh, you want a range, sorry
07:40 karolherbst: you don't want to poke there without knowing what you do :p
07:41 pmoreau: It could have funny effects! :D
07:41 karolherbst: very
07:42 pmoreau: Consumption down to 14.5W, seems to have stabilised
07:42 pmoreau: ... everywhere
07:42 karolherbst: as I thought
07:42 karolherbst: rnndb doesn't state they are gf100+, but the name indicates it
07:47 mupuf: karolherbst: nice!
07:47 mupuf: so my card is the weird one
07:47 mupuf: or my sensor is not accurate enough
07:47 karolherbst: maybe not
07:47 mupuf: because I could not see ANY change
07:47 karolherbst: yeah well
07:47 mupuf: well, I prefer it this way!
07:47 karolherbst: it was hard to see a change on the fermi card at first too
07:48 mupuf: so as I can debug the hard cards myself!
07:48 karolherbst: because I can't only check the battery
07:48 karolherbst: *can
07:48 karolherbst: but
07:48 karolherbst: the difference was bigger when nouveau was loaded
07:48 karolherbst: anyway, it seems to kind of work
07:48 karolherbst: and maybe setting every gate to auto/auto works out of the box
07:49 karolherbst: mupuf: maybe desktop cards have different default states on the card itself
07:50 mupuf: yes, this is very likely that the bios would set your card in a state that is easier to deal with :D
07:51 karolherbst: I could cook a patch up, which just sets everything to auto/auto when config=NvExpPwrSafe=1 is set or something
07:51 karolherbst: then it would be easier to find issues
07:52 karolherbst: it is awesome, that auto/auto has the same power consumption as stop/stop here
07:53 karolherbst: so somehow I don't see a reason why to care much more about those regs
07:55 pmoreau: Do you want me to test something on my laptop?
07:55 karolherbst: pmoreau: if I would knew the regs :D
07:55 karolherbst: but somehow I manage to do only sutff which doesn't help your laptop at all :p
07:55 pmoreau: :D
07:56 pmoreau: That's mean! :p
07:56 karolherbst: ohh wait
07:56 karolherbst: you've got a nice core voltage interface in hwmon
07:57 pmoreau: That's true
07:57 pmoreau: Thank you for that! :-)
07:57 pmoreau: (Not that I stare at it everyday, but... Still can be useful)
07:57 karolherbst: :D
07:59 karolherbst: mupuf: ptherm subdev?
08:00 mupuf: karolherbst: I am unsure right now
08:00 mupuf: sorry, meeting
08:00 karolherbst: k
08:24 karolherbst: wait, what
08:25 karolherbst: pmoreau: okay, that would be article nr. 2 :D
08:50 karolherbst: mupuf: what is the difference between ENG/BLK and CLK/PWR in the 0x20200 regs?
08:56 pmoreau: karolherbst: I would say that ENG (engine) would "power" off/on the whole engine, while BLK (block) would do it on a block level within the engine
08:56 karolherbst: mhh
08:56 karolherbst: kind of makes sense
08:56 pmoreau: Oh, just realisze that wasn't not exactly your question
08:56 karolherbst: partly
08:57 karolherbst: there are ENG_CLK and ENG_PWR also BLK_CLK and BLK_PWR
08:57 pmoreau: As for clock vs power gating, clock gating only "kills" the clock, so device still receive power, while power gating removes power
08:57 imirkin: engine vs block
08:58 pmoreau: IIRC what mupuf told me
08:59 karolherbst: mhhh okay
08:59 karolherbst: okay, then somehow the _PWR bits doesn't do anything yet
09:00 pmoreau: Well, if the engine is being used frequently, it might not power off
09:00 karolherbst: it is just crazy how the blob just uses the same value on like all cards
09:01 pmoreau: Could be that _PWR has a higher delay than _CLK
09:01 pmoreau: Or needs some more config
09:01 karolherbst: you can configure the delay in there
09:02 karolherbst: but no, I think you need to configure more stuff to get actually power off the engine
09:03 karolherbst: also ENG_PWR is always set to RUN
09:03 karolherbst: strange
09:03 pmoreau: On all engines?
09:04 karolherbst: anyway, setting ENG_CLK to auto helps a lot already
09:04 karolherbst: pmoreau: yes
09:05 pmoreau: It can make sense: if you need extra config to get _PWR working, you probably don't want to enable it by default
09:05 karolherbst: mhhh
09:05 karolherbst: maybe it doesn't matter
09:05 karolherbst: or isn't possible at all
09:06 karolherbst: or it is making the driver much simplier
09:06 karolherbst: who knows what you have to do after an engine was fully powered off
09:06 karolherbst: maybe you need to save/load the state
09:06 karolherbst: where only turning of the blocks would allow the engine to keep the state
09:07 pmoreau: Nouveau already supports powering down the card, so just an engine shouldn't be too difficult
09:14 karolherbst: pmoreau: if it is not in use
09:14 karolherbst: that means, no state
09:15 karolherbst: you can simply bring the card into a state like it was just booted and it would work
09:33 karolherbst: okay, the clock gate latencies are a bit wrong
09:34 karolherbst: 17k fps => 18.5k fps in glxgears
10:11 mlankhorst: length: 0.19999999999999998 m
10:12 pq: gnurou, I just read your 3dgraphics-talk slides. Very awesome slides, the presentation blew my mind. :-)
10:15 karolherbst: pq: is there a video somewhere? the html slides don't work for me
10:15 pq: I haven't seen
10:16 pq: first I thought the slides didn't work either, but tapping right arrow gets you stuff eventually
10:16 pq: but I do see some apparently blank slides too
10:16 karolherbst: yeah
10:16 karolherbst: me too
10:17 karolherbst: I think there is the most impressive part :D
10:17 pq: did you saw any of the 3d animations?
10:17 imirkin_: yeah, slides 6,7,8 are blank for me
10:17 karolherbst: pq: nope
10:18 karolherbst: I think there is some webgl stuff or something going on
10:18 karolherbst: maybe it works only on the blob driver, who knows :D
10:18 imirkin_: there's a lot of "503 Service Temporarily Unavailable" going on
10:18 imirkin_: the slide remains in stateInvisible until all the bits load
10:18 imirkin_: but i guess they don't
10:18 karolherbst: ohhh okay
10:18 imirkin_: http://rawgit.com/Gnurou/3dgraphics-talk/master/iframes/cube.html?7 & co
10:19 karolherbst: mhh
10:19 karolherbst: strange
10:19 pq: hm. I viewed with firefox 38.3.0 and I'm on nouveau. Not all slides work, but many do.
10:19 karolherbst: yeah, but it's no fun for me that way
10:20 pmoreau: Same with Firefox 41.0.1 and on the blob
10:20 pq: like slide 7 seems to be a 3d animation with a couple dozen steps
10:22 pq: ..or three dozen
10:22 karolherbst: would be nice if nvidia would actually get some paid devs for nouveau besides the tegra :p
10:24 karolherbst: allthough more docs would help a lot too
10:30 pmoreau: There was some pull request for merging doc into envytools, but the formatting was wrong. I don't think it as been resubmitted yet, but I could have missed it
10:31 karolherbst: there is your PR
10:32 pmoreau: I mean, a PR from NVIDIA guys
10:33 karolherbst: don't see one
10:33 pmoreau: https://github.com/envytools/envytools/pull/10
10:34 karolherbst: ohh
10:37 karolherbst: yeah, mhhh
10:37 karolherbst: don't know, I meant some more serious docs though
10:37 pmoreau: :-)
10:58 karolherbst: okay, so the gates doesn't seem to effect performance in performance critical stuff
10:58 karolherbst: I got no difference with unigine
10:58 karolherbst: checking desktop usage would be still interessting to check
11:02 pmoreau: Maybe with a lower delay you would see some perf changes?
11:03 karolherbst: I see one in glxgears
11:03 karolherbst: but I could try out how this behaves
11:06 karolherbst: yeah
11:06 karolherbst: I can improve perf again with higher delays
11:07 karolherbst: :D
11:07 karolherbst: I can cause engine faults
11:07 karolherbst: but glxgears is just running as nothing would happens
11:07 pmoreau: Eh eh :D
11:07 karolherbst: got a bunch of those: bus: MMIO write of 000000ff FAULT at 088068 [ TIMEOUT ]
11:09 karolherbst: ohh my power sensor is gone?
11:09 karolherbst: that's unfortunate
11:37 karolherbst: imirkin_: they don't like you :p
11:47 imirkin_: karolherbst: who?
11:47 MichaelLong: them phronix forum dudes
11:47 MichaelLong: but at least I learned I'm using the right distribution :>
11:48 idl0r: moronix
12:38 karolherbst: MichaelLong: :D
12:38 karolherbst: MichaelLong: arch?
12:40 MichaelLong: karolherbst, no the other one
12:40 karolherbst: fedora?
12:40 karolherbst: I really don't know what is hip these days
12:40 karolherbst: "back in my days" it was ubuntu
12:41 RSpliet: I don't think Ubuntu has ever been nice
12:41 MichaelLong: had fedora on my work laptop but I went back to gentoo there too
12:41 RSpliet: it's been widely used due to misperception
12:41 MichaelLong: karolherbst, I guess those mint thingy
12:41 karolherbst: MichaelLong: I thought gentoo < arch, at least on the phoronix forum
12:45 MichaelLong: not sure about the phoronix forum, I don't like forums in general and especially not the phoronix one
12:45 karolherbst: :D
12:46 karolherbst: oh I really like the discussions there. Everytime it just tells me: go away, don't waste time here
12:46 karolherbst: its a real motiviation to do something productive
12:47 Yoshimo: i think that is the issue with most online forums karol
12:47 karolherbst: :O
12:47 karolherbst: I was under the impression it wasn't as bad as today like 10 years ago
12:47 imirkin_: redhat was definitely the cool thing back in the day. but that day was like 1998.
12:48 karolherbst: mhh
12:48 karolherbst: then I didn't even thought about the idea of using linux at all
12:48 karolherbst: :D
12:48 imirkin_: really hard to get dialup going when you don't have an internet connection to look up instructions
12:48 imirkin_: pppd is *not* the most obvious application in the world
12:49 karolherbst: ohh, there were CDs alright
12:49 imirkin_: esp when you're a kid and don't have a full appreciation of, say, line disciplines.
12:49 MichaelLong: damn all my available nvidia cards here are the wrong ones, NVS300, 560Ti and a GTX980 :/ I'd like to test out your patches karolherbst
12:49 karolherbst: I had a linux boot CD in a mac magazine for macs
12:49 karolherbst: MichaelLong: :p
12:49 karolherbst: MichaelLong: why are they wrong?
12:49 karolherbst: MichaelLong: I have something for everybdy
12:50 MichaelLong: really? then I misread the phoronix headline
12:50 karolherbst: always
12:50 karolherbst: pcie stuff also works on maxwell
12:50 imirkin_: MichaelLong: s/mis//
12:51 MichaelLong: heh
12:51 karolherbst: also pcie stuff works on fermi
12:51 karolherbst: you just need a little hack
12:52 karolherbst: MichaelLong: https://gist.github.com/karolherbst/c730632ed5d1e115c59d
12:52 karolherbst: I just copied that from my terminal because of ssh
12:52 karolherbst: I doubt it will apply
12:52 karolherbst: but you should get the idea
12:53 MichaelLong: yes, this is based on what kernel?
12:53 karolherbst: it works since 3.19
12:53 MichaelLong: ah ok
12:53 karolherbst: but you need the rework
12:53 karolherbst: also the NVS300 is a GT218 card
12:53 karolherbst: pcie stuff should also work there
12:54 karolherbst: I just don't know if the card can actually do v2
12:54 MichaelLong: you mean the rework done in 4.3-rcX?
12:55 karolherbst: yes
12:55 karolherbst: you can use my branches though
12:55 karolherbst: MichaelLong: https://github.com/karolherbst/nouveau
12:56 karolherbst: these work for 3.19 up to 4.2 or 4.1
12:56 MichaelLong: hmm I've still some problem with the 4.3-rcX-series, monitor goes off when modesetting kicks in, if that's still happening at the weekend I wanted to bisect it
12:56 MichaelLong: but 4.2.x is fine
12:56 karolherbst: try to compile my trees then
12:58 MichaelLong: karolherbst, I'll try at the weekend and report back, thx!
12:58 karolherbst: nice
13:01 karolherbst: MichaelLong: but somehow for some strange reasons, it looks like the the gt218 doesn't support v2 :/
13:02 karolherbst: maybe it does though
13:04 MichaelLong: we'll see. not sure if it is noticeable at all, this card is certainly not a performance wonder
13:05 karolherbst: well, the perf difference should be rather small anyway
13:05 karolherbst: there are only a few usecases where this really makes a differece
13:05 karolherbst: like PRIME
13:05 karolherbst: and a handfull games
13:06 MichaelLong: for now it is a okish primary gpu for my host system driving two panels, gameing is done with the gtx 980 anyway
13:09 karolherbst: I see
13:11 MichaelLong: strange card anyway, but found it totally alone in our pc hw cabinet at work.
13:58 karolherbst: hakzsam: are you done for today?
13:58 karolherbst: on reator?
13:59 mupuf: karolherbst: when I am done with my finnish (it always is fun to say that :D), I will be back to the path series
14:00 karolherbst: :)
14:00 mupuf: it took me 4h to translate a 3-pages text... Bloody hell!
14:00 karolherbst: nice
14:00 karolherbst: wow
14:00 karolherbst: you are lucky though
14:00 karolherbst: it is not latin :D
14:00 karolherbst: I had to translate latin in school
14:00 mupuf: it is much worse!
14:00 mupuf: 16 cases
14:00 karolherbst: you need 1 minute per word there
14:00 karolherbst: you have no idea :D
14:00 mupuf: of course I do, I also studied latin :D
14:01 karolherbst: :D
14:01 mupuf: and gave up as soon as I could :D
14:01 karolherbst: nothing compares to latin here in europe
14:01 karolherbst: :D
14:01 mupuf: Finnish does
14:01 mupuf: look it up :p
14:01 karolherbst: :D
14:01 karolherbst: yeah I know, it is also a mess
14:05 mupuf: it is not a mess, it is just nothing alike any other language I know
14:05 mupuf: it is quite regular ... on good days
14:05 mupuf: but the flow chart to write a single word is .... it would not fit on a wall
14:05 mupuf: it is ridiculous
14:06 karolherbst: :/
14:06 imirkin_: mupuf: how a barn? :)
14:06 mupuf: probably would :D
14:06 imirkin_: https://en.wikipedia.org/wiki/Barn_(unit)
14:07 imirkin_: A barn is defined as 10^−28 m^2
14:08 mupuf: oh, I thought you meant the big buildings
14:08 imirkin_: ;)
14:08 karolherbst: my disc has a size of 15 Yb
14:08 mupuf: you damn joking laughing cow!
14:08 imirkin_: read the page for why it's called a barn
14:10 mupuf: imirkin_: excellent!
14:10 mupuf: it is just wonderful
14:11 mupuf: karolherbst: try making sense of this word using a dictionnary: Lentokonesuihkuturbiinimoottoriapumekaanikkoaliupseerioppilas
14:13 karolherbst: yeah, lol
14:13 mupuf: and you are lucky, most of the words are close to english because, you know, they did not have turbines 2k years ago
14:13 mupuf: but other than that, all the words different
14:13 imirkin_: other than turbines :)
14:13 karolherbst: I like atomiydinenergiareaktorigeneraattorilauhduttajaturbiiniratasvaihde more
14:14 mupuf: you may know kone though, machine. That's a escalator brand found worldwide
14:14 mupuf: yeah, german is full of fun too!
14:14 karolherbst: its not german :p
14:14 mupuf: wha'ts that?
14:14 karolherbst: you know
14:15 hakzsam: karolherbst, I don't use it
14:15 mupuf: oh, finnish again
14:15 karolherbst: :)
14:15 mupuf: well, hard to say when all are borowed words :D
14:15 karolherbst: :D
14:16 karolherbst: any idea about the subdevs of these gates though?
14:16 karolherbst: I had some fun time with mine
14:16 karolherbst: I managed to get the power sensor off the bus
14:16 mupuf: what? How?
14:16 karolherbst: wasn't rachable anymore though i2c
14:16 karolherbst: ohh
14:17 karolherbst: I just poked ffff2455 into the counters
14:17 karolherbst: got bus: MMIO write of 047b40ff FAULT at 088068 [ TIMEOUT ] fun in dmesg
14:17 karolherbst: but glxgears just rotated its gears
14:17 karolherbst: so, don't know
14:17 RSpliet: imirkin_: Other related units are the outhouse (1 μb, or 10−34 m2) and the shed (10−24 b (1 yb), or 10−52 m2), although these are rarely used in practice.
14:17 karolherbst: everything was fine :D
14:17 RSpliet: is that a thing?
14:18 karolherbst: I am too young for those
14:18 imirkin_: RSpliet: unlike barns, which come up in daily use for just about everyone :p
14:18 imirkin_: i prefer to just measure it all in eV
14:19 karolherbst: :D
14:19 karolherbst: allthough you have to be a bit creative to messure distances with that
14:20 imirkin_: why?
14:21 imirkin_: https://en.wikipedia.org/wiki/Natural_units#.22Natural_units.22_.28particle_physics_and_cosmology.29
14:21 karolherbst: as I said: you have to be a little bit creative
14:40 mupuf: oh, see you guys, nothern lights!!
14:41 mupuf: twice in 10 days, amazing
14:41 hakzsam: mupuf, you're lucky
14:41 karolherbst: :(
14:41 karolherbst: I did holidays in iceland once for this
14:41 karolherbst: and guess what
14:41 RSpliet:doesn't get any northern light, only the darkness from uranus
14:55 mupuf: yeah!! I saw ..... NOTHING!
14:55 hakzsam: haha
14:56 karolherbst: mupuf: :)
14:56 mupuf: I need to remember that what is a high value after the polar circle is notas good in Helsinki ÖD
14:56 karolherbst: mupuf: on the bright side, the northern lights are always there :p
14:56 mupuf: oopsie
14:56 mupuf: hehe
14:56 karolherbst: I actually saw them on the plain to iceland
14:56 karolherbst: but iceland itself was too cloudy
14:56 mupuf: oh, yeah, unlucky!
14:57 mupuf: there was only one cloud today, guess where it was?
14:57 mupuf: north!
14:57 karolherbst: :D
14:57 mupuf: so, can t see anything for sure!
14:57 karolherbst: maybe some hours later?
14:57 karolherbst: ohh wait, do you know this norther light site?
14:58 karolherbst: I just have to find the good one
14:59 karolherbst: there are like 100 bad one and one good one
14:59 karlmag: aka "somewhere north of the polar circle? :-P
14:59 karolherbst: I meant web site
14:59 karlmag: Figured as much ;-)
14:59 karolherbst: there is a good one, where you can like check for the next two weeks
15:00 karolherbst: http://www.gi.alaska.edu/AuroraForecast
15:00 karolherbst: this one
15:00 karolherbst: you just have to select your area :p
15:01 karolherbst: mupuf: seems like the next three days you shall see something :p
15:03 karolherbst: there is also this awesome site for iceland: http://en.vedur.is/weather/forecasts/aurora/#type=total
15:05 mupuf: it was at 6 when I left
15:06 karolherbst: :D
15:06 mupuf: http://www.aurora-service.eu/
15:07 karolherbst: oh yeah, I really dislike that site
15:07 karolherbst: and not only because the style is bad :D
15:08 karolherbst: it has a lot of data though
15:09 mupuf: looks good in svalbard islands: http://polaris.nipr.ac.jp/~acaurora/aurora/Longyearbyen/latest.jpg?1444774073
15:09 mupuf: and what a beautiful milky way!
15:42 karolherbst: skeggsb: what about the gddr5 patch? I was a little bit hoping it would land for 4.4
15:43 skeggsb: karolherbst: i'm just about to leave some comments on that, and the pcie patches
15:43 karolherbst: okay, nice :)
15:43 karolherbst: I like comments
15:43 imirkin_: "this change is bad" :p
15:44 skeggsb: not in this case :P
15:44 karolherbst: fine by me
15:44 karolherbst: :p
15:44 karolherbst: I maybe the pesonal attacked for a short time, but I know if I write bad code, it is not a good idea to include it :D
15:44 karolherbst: but I always do something wrong, so
15:44 skeggsb: don't we all :)
15:44 karolherbst: and most of the time I know it :D
15:45 imirkin_: i was jk btw
15:45 karolherbst: skeggsb: but the gddr5 patch can be included as a bugfix later on anyway, can't it?
15:46 skeggsb: i'll pull it for 4.4, sorry it missed 4.3. i don't think it's much of a candidate for -fixes or -stable due to it being for a codepath we don't enable/support right now anyway
15:46 karolherbst: mhh
15:46 karolherbst: why isn't it supported?
15:46 karolherbst: with pstate=1 you hit the code path
15:47 karolherbst: and because pstate is in sysfs, isn't it a little bit more on the supported site?
15:47 skeggsb: because none of that code is "ready"
15:47 karolherbst: ahh I see
15:47 skeggsb: when it's in a situation where we can change clocks by default, without requiring users to risk their lives, it'll be different :P
15:47 karolherbst: :D
15:47 karolherbst: k
15:48 karolherbst: yeah I am totally fine with 4.4
15:48 karolherbst: I just meant as a bugfix for drm to get it in 4.4 if the deadline for drm pulls is over
15:48 skeggsb: nah, it can make 4.4 easily, i haven't even finished my own stuff for 4.4 quite yet :P
15:49 imirkin_: sitting next to the drm maintainer has some advantages, huh
15:49 karolherbst: :D
15:49 skeggsb: not when he can hit you :P
15:49 imirkin_: goes both ways, heh
15:49 hakzsam: skeggsb, like nvif ;)
15:51 karolherbst: skeggsb: I think I will also work on fermi+ clock gates, but I have no clue where to put the code (I got like 10% on my kepler and 20% on my fermi power reduction)
15:52 karolherbst: 17W on the fermi laptop, with both gpus on, I was pretty impressed
15:52 karolherbst: with i7 cpu and everything
15:56 karolherbst: skeggsb: :O are all your mails that messed up or did I something wrong?
15:56 karolherbst: http://lists.freedesktop.org/archives/nouveau/2015-October/022656.html
15:57 karolherbst: the patch part bellow
15:57 karolherbst: or was it pgp :D
15:58 skeggsb: ugh, it's probably pgp... sorry
15:59 imirkin_: and you can't claim someone else sent the mail... it's got your sig right there!
15:59 karolherbst: :D
16:02 karolherbst: skeggsb: I don't know about the pll code, I mean I don't really care where it is, but I also don't care enough to actualy move it somewhere else :D it's basically your call, if you say fb is better, I move it, if you don't care, I won't care, too
16:02 skeggsb: yeah, move it, just stick that whole function in ramgk104.c directly i say, it's the only user
16:03 karolherbst: k
16:04 imirkin_: skeggsb: what all are you planning for 4.4?
16:04 imirkin_: another rewrite? :)
16:06 mupuf: moar refactor! :p
16:06 mupuf: anyway, bed time. Cheers!
16:08 karolherbst: skeggsb: I am also somehow unsure about this calc_pll_clock function :/ the name is bad
16:08 karolherbst: maybe gt215_pll_calc_output?
16:09 skeggsb: it's like, one line, copy it :P
16:10 karolherbst: yeah well, the calculation isn't that trivial, so that I would just copy it where it is used
16:19 karolherbst: skeggsb: you meant the set_version and set_link func pointers in the pcie struct?
16:19 karolherbst: here: http://lists.freedesktop.org/archives/nouveau/2015-October/022657.html
16:26 gnurou: pq: argh, I wish I had cleaned up the code a bit before you guys found these out
16:27 gnurou: pq: as you noticed many slides don't work when viewing online... and the talk is simplified to the point where it is borderline inaccurate sometimes
16:27 gnurou: btw, if you want to see nice webgl slides, this is the reference as far as I am concerned: acko.net/files/gltalks/pixelfactory/online.html
16:29 imirkin_: where was this when i was trying to understand all this stuff
16:32 karolherbst: gnurou: wow, these webgl slides are something :D
16:35 karolherbst: okay, my intel card isn't fast enough for this 127m calls /s example :D
16:38 gnurou: his talk is also prettty amazing, look it up on Youtube
16:39 imirkin_: gnurou: are yours supposed to not be working quite yet? getting a lot of 503 service unavailable's
16:40 imirkin_: gnurou: in state.js: function startFrame(independant = false) {
16:40 imirkin_: that's not a thing in javascript
16:41 gnurou: imirkin_: oh wow, Javascript interpreters are quite resilient then :P
16:41 imirkin_: you can just leave off the =false... it'll be false-ish
16:41 karolherbst: gnurou: independant = typeof independant !== 'undefined' ? independant : 42;
16:42 imirkin_: gnurou: as long as you don't do anything like ===false
16:42 karolherbst: ohh false
16:42 karolherbst: not 42
16:42 imirkin_: karolherbst: common typo :p
16:42 gnurou: imirkin_: actually it seems to be correct from a syntax point of view
16:42 imirkin_: gnurou: default param values aren't a thing in JS
16:42 gnurou: but please... just don't look at these right now, they are horrible
16:42 karolherbst: :D
16:42 gnurou: I threw most of it in a hurry
16:42 karolherbst: ohh back in my early days
16:42 karolherbst: I did a lot of JS
16:42 karolherbst: I even wrote a window system in javascript
16:42 karolherbst: :)
16:43 imirkin_: gnurou: at least not in ES3... maybe ES5 but you don't have the 'use strict' thingie to flip it on
16:43 karolherbst: *window manager
16:43 gnurou: even "independant" is not spelled correctly >_<
16:43 gnurou: this is my first real contact with JS FWIW
16:43 imirkin_: gnurou: welcome to the jungle
16:43 karolherbst: :)
16:44 karolherbst: ducky duck
16:44 gnurou: anyway, I will fix that some day... after the Maxwell FW is released *cough*
16:44 imirkin_: gnurou: you'll know you're a master when you get more than 1/4th right on a quiz like http://perfectionkills.com/javascript-quiz/
16:44 karolherbst: I used to like duck typing a lot :D
16:44 gnurou: I have having crappy code of mine online, even though there is a bunch of it already
16:44 imirkin_: [i.e. better than random guessing]
16:44 gnurou: s/have/hate
16:44 gnurou:gets some coffee
16:45 karolherbst: imirkin_: the first question :D
16:45 karolherbst: I know at least which of both
16:45 karolherbst: mhhh
16:45 karolherbst: nvm
16:47 gnurou: what JS lacks is a good fencing mechanism, all this async stuff that you cannot control... for instance, how do you know an iframe has been loaded, etc.
16:47 imirkin_: gnurou: it's all solvable, trust me
16:47 gnurou: we need a buffer of JS commands and the ability to signal when a point of this command buffer has been reached... oh wait
16:47 karolherbst: gnurou: events
16:48 gnurou: yeah, through dirty hacks mostly
16:48 imirkin_: gnurou: actually js is wholely single-threaded
16:48 gnurou: right, another annoying thing :)
16:48 imirkin_: if you just have a while (asdf) {} -- it'll never complete
16:48 imirkin_: cross-frame stuff is a pain, esp if you're trying to do cross-browser
16:48 gnurou: yeah, so you need to resort to the much cleaner setTimeout() :P
16:49 imirkin_: and esp when they're not on the same domain
16:49 karolherbst: I like how you can reference variables in javascript, which you are declaring later :)
16:49 imirkin_: karolherbst: nah, it's about scope.
16:49 gnurou: hopefully I will be able to remove all these horrible iframes... it was just the simplest way to put stuff where I wanted :D
16:49 karolherbst: imirkin_: :D
16:49 imirkin_: karolherbst: asdf; var x = 5; is the same as var x; asdf; x = 5;
16:49 imirkin_: and only functions have scope
16:49 karolherbst: that's not what I meant :p
16:50 karolherbst: x; /* not undefined */ var x;
16:50 imirkin_: function hoisting is fun too
16:50 imirkin_: exactly.
16:50 imirkin_: all the var decls get hoisted to function top
16:51 karolherbst: I think I miss the dirty trick in this quiz
16:51 karolherbst: :D
16:51 imirkin_: the?
16:51 imirkin_: every question is a dirty trick
16:52 karolherbst: typeof typeof x; :D
16:53 imirkin_: good one
16:53 karolherbst: that's easy
16:53 karolherbst: :D
16:53 imirkin_: 'string' right?
16:53 karolherbst: yeah
16:53 karolherbst: eval(typeof typeof x; + '(' + 5 + ')')
16:53 karolherbst: :P
16:54 karolherbst: ahh forgot something
16:54 karolherbst: ohh there is no string() funciton?
16:54 karolherbst: :(
16:54 imirkin_: that'd just be crazy
16:54 imirkin_: but you can use String() to intern a string
16:55 imirkin_: as opposed to getting a fresh object with new String()
16:55 karolherbst: wait
16:55 imirkin_: (you always want the former)
16:55 imirkin_: String("x") == String("x")
16:55 karolherbst: eval((typeof typeof x).charAt(0).toUpperCase() + (typeof typeof x).slice(1) + '(' + 5 + ')')
16:55 imirkin_: but iirc, new String("x") != new String("x")
16:58 karolherbst: 12 is tricky
16:58 karolherbst: ...
16:59 karolherbst: ohh 5 right :)
16:59 karolherbst: that's more than 1/4 :D
16:59 imirkin_: hehe
16:59 imirkin_: well done! :p
16:59 karolherbst: I think I got the easiest ones wrong: #1, #2, #4, #5, #6, #7, #9, #11, #14)
17:00 karolherbst: ohhh
17:00 imirkin_: if they were so easy, why did you get them wrong? :p
17:00 karolherbst: arguments is an object
17:00 karolherbst: boah
17:00 karolherbst: I selected array
17:00 karolherbst: :D
17:01 karolherbst: in 2 I searched for the trick....
17:01 karolherbst: my first thought was right though
17:02 karolherbst: what
17:02 karolherbst: 9 is wrong!
17:02 karolherbst: v8 gives me 1number
17:02 karolherbst: what the
17:02 karolherbst: that's cheating :D
17:04 karolherbst: yeah, with firefox I got that right
17:04 karolherbst: imirkin_: okay, I have 6 right :)
17:04 imirkin_: after revising your answers :p
17:04 karolherbst: nope
17:04 karolherbst: the answer is checked with your browser js engine
17:04 karolherbst: and v8 did that different than firefox
17:04 imirkin_: heh
17:05 karolherbst: I check 1function for the 9th one
17:05 karolherbst: but v8 says 1number
17:05 karolherbst: stupid browser can't trick me! :p
17:05 imirkin_: huh, that's weird
17:07 imirkin_: i actually get 1undefined for some reason that's not altogether clear to me.
17:08 karolherbst: browser?
17:08 imirkin_: chrome
17:08 karolherbst: mhh
17:08 karolherbst: maybe there is a bug there
17:08 karolherbst: :D
17:08 imirkin_: f should get hoisted out there
17:08 imirkin_: and typeof should see it. weird.
17:08 karolherbst: yeah
17:08 karolherbst: typeof f should be function
17:09 karolherbst: ohhh
17:09 karolherbst: I get now undefined too
17:09 karolherbst: ohhhh
17:09 karolherbst: I know why I got 1number
17:09 karolherbst: because I was playing alredy in the console
17:10 karolherbst: I bet I defined f already
17:11 imirkin_: tsk tsk tsk
17:21 karolherbst: skeggsb: I fixed the g84/g86 thingy. I think I leave the set_link and set_version thingy this way, because it is usually like this: pcie[._](set_)stuff
17:21 karolherbst: and somehint link_set and version_set sound kind of strange to me :/
17:22 karolherbst: maybe other names are better?
17:23 karolherbst: ohh I will remove the pcie2 accessors
21:55 phire: So, slightly off topic question, but do you have tools for reading the compiled shaders out of nvidia's Windows Drivers (both directx and opengl)
21:58 phire: ok, I admit. It's completely off topic. I really want to know why Nvidia's OpenGL shader is so much slower than their DirectX shader.
21:58 phire: and their profiling tools aren't helping
22:04 imirkin: phire: if you can repro on linux, you can use valgrind-mmt
22:04 imirkin: (in conjunction with demmt from envytools)
22:05 imirkin: phire: also if you can get the actual shader binary, that's easily disassembled
22:05 phire: yeah, I think I'll find the 'directx compiler' part hard to reproduce on linux
22:05 phire: oh, I have the shader binary
22:05 imirkin: the actual shader binary
22:05 imirkin: not some idiotic NV_gpu_shader5 bs
22:05 imirkin: er, NV_gpu_program
22:06 imirkin: wtvr
22:06 imirkin: anyways, if you make that available, i can easily disasm
22:06 imirkin: you can use 'nvdisasm' from cuda tools, or envydis from envytools
22:07 imirkin: nvdisasm -b SM30 foo should do the trick (SM20 for fermi, SM30 for kepler1, SM35 for kepler2, and SM50 for maxwell)
22:07 phire: yeah, NV_gpu_program_5 output is next to useless
23:26 pq: gnurou, even if it was borderline inaccurate, I can see it being a gold mine for someone new to 3d gfx. If you explain things accurately, thing often become complicated enough that it's not useful for newcomers. :-)
23:33 gnurou: pq: indeed, there is a compromise to be made - I'd have liked to re-run it a couple of times and made a commented online version before it goes public though... damn Phoronix :P
23:34 pq: haha
23:36 gnurou: I love how the very obscure last bullet point of some slide made it into a headline though
23:36 imirkin: gnurou: i'd ignore such things
23:37 gnurou: oh I do - I just hope my boss does, too :P