00:53 Tom^: koz_: good morning.
00:53 koz_: Tom^: Haiiii!
00:53 Tom^: oh meh i highlighted wrong person. :p
00:53 koz_: LOL.
00:53 koz_: I was wondering why you were greeting me out of the blue.
00:53 Tom^: k<tab>
00:59 snacks: Tom: good evening.
03:11 Tom^: karolherbst: good morning sir.
03:17 karolherbst: Tom^: there is something you could do, allthough it might be not needed: boot with the nvidia driver and Option "Coolbits" "28" set in the xorg.conf
03:18 karolherbst: check if you can control the fan through nvidia-settings in that case
03:18 Tom^: karolherbst: i found out nouveau can control the fan. "[ 941.444992] nouveau 0000:01:00.0: therm: temperature (90 C) hit the 'fanboost' threshold"
03:18 karolherbst: Tom^: ohh
03:18 Tom^: and it goes to 100% , until [ 948.441152] nouveau 0000:01:00.0: therm: temperature (87 C) went below the 'fanboost' threshold
03:18 karolherbst: Tom^: so after 90°C it boost it up?
03:19 Tom^: yea but goes back to idle 20%/1050rpm at 87C
03:19 karolherbst: okay
03:19 karolherbst: then the calculations is just odd
03:19 karolherbst: yeah okay, I will eat something first then :D
03:19 Tom^: i was trying to poke in the vbios with nvbios and checking the Temperature Table but its greece to me xD , because /sys/class/hwmon reports same values as lm_sensors
03:20 Tom^: so im just assuming the high,crit values is just way to high so it doesnt think its hot enough to ramp up the fans
03:28 Tom^: or does it read some voltage from a table that is foobared because of my change with info.min, info.max
03:36 karolherbst: Tom^: lm_sensors just reads hwmon out
03:37 Tom^: however it feels like something else too, https://gist.github.com/anonymous/dc6ad77ea448b58d1206 only therm debug info im getting is when i hit the fanboost mode. so where is all the other debug output that should come :<
03:37 karolherbst: Tom^: try debug=debug
03:37 karolherbst: Tom^: also those high/crit values are constants inside nouveau, they aren't yet parsed
03:38 Tom^: yea i found them, https://github.com/karolherbst/nouveau/blob/master_karol_no_touchy/drm/nouveau/nvkm/subdev/therm/temp.c :p
03:38 karolherbst: but 90°C is fine. nvidia clocks down to min at 97°C
03:40 karolherbst: there is a periodic print about the fan level then
03:40 karolherbst: also there is a temp => fan speed mapping somehwere
03:40 karolherbst: and there is the your issue
03:42 Tom^: now we are talking https://gist.github.com/anonymous/cd0a55ecfb126440396e now just to ramp up some usage see what it prints.
03:43 karolherbst: as you see, it maps all temps to 0% fan ;)
03:43 Tom^: mh
03:43 Tom^: yep keeps printing request: 0% when temps rises.
03:57 Tom^: karolherbst: maybe set liner_min_temp and max to something manually? and see if its the vbios parsing or the calcs?
03:57 Tom^: https://github.com/karolherbst/nouveau/blob/master_karol_no_touchy/drm/nouveau/nvkm/subdev/therm/base.c#L67 there
04:06 chillfan: so I found nouveau is working almost nicely for nouveau from linux 4.3
04:08 chillfan: is using an xorg driver version beyond 1.0.11 likely to improve upon that, or should it be done only to test hardware?
04:21 karolherbst: Tom^: well I have no clue how it works currently :D
04:21 karolherbst: chillfan: it depends on what your issues are
04:21 Tom^: i have no clue what i am doing.
04:21 Tom^: xD
04:21 karolherbst: Tom^: but there has to be a formular with some parameters
04:22 chillfan: ah alright :)
04:23 Tom^: we need mupuf_
04:23 mupuf_: Tom^: what for?
04:23 mupuf_: oh, your fan issue
04:23 karolherbst: Tom^: yeah well, when the FAN can be controlled through nouveau the issue is much simplier
04:23 Tom^: im having a bit of fan issues on 780ti
04:23 chillfan: i do notice mouse cursor and window dragging is slightly slow looking
04:23 karolherbst: mupuf_: yeah, nouveau has a 0% target all the time
04:23 karolherbst: and boost to max at 90°C
04:24 karolherbst: I bet it is just the vbios parsing goes wrong
04:24 chillfan: guess i'd have to compare to nvidia more exactly to know what's what
04:24 Tom^: mupuf_: https://www.dropbox.com/s/7gg91nlyb6t6911/vbiosprom.rom?dl=0 vbios,
04:24 karolherbst: chillfan: "slow" or "laggy"?
04:24 chillfan: laggy i'd say
04:24 karolherbst: mupuf_: nvf1/Tom^ ;)
04:24 Tom^: mupuf_: it just keeps requesting 0% , [ 2627.339064] nouveau 0000:01:00.0: therm: FAN target request: 0% , and only time it ramps up is when it hits fanboost mode at 90C , and then drops down back to idle 20% on 87C
04:25 mupuf_: yeah, looks like pwm_min and max is wrong, let's look at the vbios
04:25 karolherbst: nvbios parses 20/100 afaik
04:26 karolherbst: but "-- type: UNKNOWN, duty_range: [20:100]%, fan_div: 27000 --"
04:26 karolherbst: type UNKNOWN?
04:27 chillfan: i'd say hard to tell with window dragging as the edges look a little distorted
04:27 Tom^: chillfan: you are just getting tearing you mean? without a compositor that can vsync, that is to be expected.
04:28 chillfan: hm i think i don't have a compositor
04:28 Tom^: well depends if you run a DE or standalone wm.
04:29 Tom^: standalone wm's i would suggest looking at compton, or just live with the tear. ;)
04:29 chillfan: standalone, my compositor is missing for some reason
04:29 karolherbst: chillfan: what window manager are you using?
04:29 chillfan: openbox
04:30 chillfan: basically i notice it with iceweasel (firefox) when dragging the window, the scroll bars look pretty odd
04:30 chillfan: just enabled compton
04:30 chillfan: same
04:30 karolherbst: mhhh
04:30 Tom^: compton has bunch of various --vsync flags tho
04:30 karolherbst: could you make a x11grab with ffmpeg?
04:30 Tom^: which isnt on by default
04:30 karolherbst: I really want to see how it looks like :D
04:30 karolherbst: before we discuss different issues
04:31 mupuf_: hmm, who figured out this fan table? :D
04:31 mupuf_: let's see
04:31 mupuf_: I hope it is not me!
04:31 chillfan: okay, hm x11grab with ffmpeg, not sure how to do it, but i'll have a try
04:31 chillfan: speaking of vsync, is it on by default?
04:31 chillfan: for nouveau i mean
04:31 karolherbst: chillfan: a script I use for streaing purposes, but it also works if you want to save it locally ;) https://gist.github.com/karolherbst/68249022a5d58f4f4bb1
04:31 Tom^: mupuf_: ;)
04:32 chillfan: cheers, i'll take a look
04:32 karolherbst: chillfan: currently it will save to "/tmp/out.mp4"
04:32 karolherbst: but you may want to change some parameters
04:32 chillfan: required are x11grab and ffmpeg?
04:32 Tom^: i think obs-studio can record to a file instead of streaming to a site too.
04:32 karolherbst: x11grab is part of ffmpeg afaik
04:32 karolherbst: Tom^: I bet it is just using ffmpeg :D
04:32 chillfan: ah alright will have a try and post that later somewhere
04:33 Tom^: karolherbst: *shrug* :p
04:33 mupuf_: Tom^: darn it, it *was* me
04:33 chillfan: i notice as well game frame rates can't be capped greater than monitor refresh, is vsync on by default?
04:33 karolherbst: it took me like 3 hours to write this script :/
04:33 mupuf_: I guess it is new for maxwell and forgot about it
04:33 karolherbst: ffmpeg is just messy to figure out
04:33 karolherbst: chillfan: with mesa I think yes
04:34 Tom^: mupuf_: even the best makes faults at some point.
04:34 karolherbst: chillfan: if it caps to the display hz then yes
04:34 mupuf_: I can't relate to the best
04:34 karolherbst: best to check with glxgears
04:34 mupuf_: anyway, let's see what nouveau does with this table
04:34 mupuf_: and by the way, what card is this?
04:34 chillfan: hm says it is using vrefresh
04:34 Tom^: mupuf_: 780ti , gk110b
04:34 chillfan: any clues on how to change it from vsync?
04:34 Tom^: NV uh E0? iirc.
04:35 Tom^: mupuf_: yap NVE0 NVF1 (GK110B) an 780ti
04:36 karolherbst: chillfan: well, make the x11grab with ffmpeg
04:36 karolherbst: before that we don't really know what the issue is ;)
04:37 chillfan: alright :)
04:37 mupuf_: Tom^: Oh oh, it had impressive caracteristics :D
04:38 mupuf_: but hey, don't think I will put down this much money for it :p
04:38 Tom^: mupuf_: 77fps in unigine heaven on nouveau yesterday.
04:38 Tom^: =D
04:38 Tom^: just it went a bit high temp so had to abort
04:39 mupuf_: I can understand why :D
04:39 mupuf_: anyway, let's check the code in nouvea
04:39 mupuf_: after all, this is my fault if it does not work
04:40 Tom^: its funny how all of this bughunting began because of michael at phoronix which led imirkin to highlight me.
04:40 chillfan: oh it looks like i did disable vsync in the xorg.conf
04:41 chillfan: didn't think it had worked, option GLXVBlank to false
04:41 chillfan: forgot i'd done that
04:41 chillfan: brb then
04:43 mupuf_: Tom^: hehe
04:43 mupuf_: I do not parse everything on the nouveau side, but that should be good
04:44 Tom^: does the therm or fan use anything from subdev/volt/ ? because i made a tiny quirk that changes my gpu core volt
04:44 Tom^: so it isnt my fault :<
04:45 karolherbst: Tom^: no, your qurirk shouldn't effect this
04:45 karolherbst: this is just a temperature thing as far as i know
04:46 mupuf_: Tom^: what is the output of sensors?
04:46 mupuf_: I assume it is the right temperature
04:46 Tom^: mupuf_: https://gist.github.com/anonymous/419985cefd2dd4e0ce11
04:47 Tom^: and it dips down to around 35-40C when im on lowest pstate so yea i think its correct
04:47 mupuf_: that's pretty high
04:47 mupuf_: but we do not do power or clock gating
04:47 mupuf_: so, it is to be expected
04:48 karolherbst: ohh right
04:48 karolherbst: Tom^: ddo you have envytools installed?
04:48 Tom^: yep
04:48 karolherbst: nvapeek 0x20200 pls
04:48 Tom^: 00020200: 22722444
04:49 karolherbst: do nvapoke 0x20200 0x60 22722455
04:49 karolherbst: and tell me if the gpu is still stable
04:49 karolherbst: and if the temp is lower when idling
04:49 Tom^: seems stable on idle, and temp dropped 2C
04:50 karolherbst: I bet this makes a big difference on your gpu
04:50 karolherbst: uhh nice
04:50 mupuf_: hmm, I will need to check on my nve6 again and see. We should land that support for kepler since it looks trivial
04:51 mupuf_: we still would need to change the settings
04:51 karolherbst: yeah
04:51 Tom^: buut it died when i started glxgears :p
04:51 mupuf_: there is more than just this, and that should yield better savings in mixed modes
04:51 karolherbst: :D
04:51 karolherbst: okay
04:51 karolherbst: so it isn't that trivial
04:51 mupuf_: ah ah
04:51 karolherbst: at least it works for me that way
04:51 karolherbst: but I know there are some engines the blob does't touch
04:52 karolherbst: never bothered figuring it out though, because it just works for me
04:52 karolherbst: Tom^: I wonder why sensors doesn't show your power consumption though
04:52 karolherbst: ahh uhh
04:52 mupuf_: karolherbst: because we did not upstream the code?
04:53 karolherbst: mupuf_: he is using my branch
04:53 mupuf_: ack
04:53 karolherbst: INA3221 and MAX6649
04:53 karolherbst: mhh
04:53 chillfan: no change with vsync on, i'll erm see if i can produce more small issues of the same, where should i upload the ffmpeg grab?
04:53 karolherbst: chillfan: mhhh any file uploading site :D
04:54 karolherbst: just search for file upload on the wbe :p
04:54 chillfan: alright, will do after i wake up a bit more :)
04:54 chillfan: should i test with linux 4.4-rc2 before doing that?
04:56 chillfan: hm nvm i'll try both, using 4.3 right now anyway
04:57 karolherbst: Tom^: would youl like to check if it is stable just with nvapoke 0x20200 22722455
04:57 karolherbst: this just turns on clock gating for one engine
04:57 karolherbst: and I am sure the blob also does this
04:58 Tom^: seems so
04:58 karolherbst: temp also dropped?
04:58 mupuf_: Tom^: there is something really weird going on
04:58 karolherbst: you can revert it with vapoke 0x20200 22722444
04:59 mupuf_: can I see your kernel logs showing" FAN_UPDATE duty 0%
04:59 Tom^: karolherbst: doesnt seem to make much of a difference
04:59 karolherbst: mhh yeah well, we need to get your power sensors working duh :p
05:00 Tom^: mupuf_ https://gist.github.com/anonymous/974a4cc00c705de3e11d
05:00 karolherbst: ohh right it is nve6 only for now :D
05:00 chillfan: alright bbl :)
05:00 urjaman_: hi ppl, 1) nouveau works way better than i expected on my 9400GT, thanks (dual screen and KSP work (with lower settings) so i'm pretty much set) 2) I've had it rarely crash in a mouse moves (and ssh works) but otherwise frozen kinda way, dmesg: http://urjaman.dy.fi/nouveau-barf.txt
05:00 karolherbst: Tom^: okay, see this? https://github.com/karolherbst/nouveau/blob/master/drm/nouveau/nvkm/engine/device/base.c#L1826
05:01 karolherbst: add ".iccsense = gf100_iccsense_new," after ibus
05:01 Tom^: karolherbst: mm
05:01 karolherbst: it might work or might not
05:02 urjaman_: I tried searching for a bug that'd match that dmesg (description of Xorg freezes does match to one...) but didnt find one that'd match both description and dmesg, so i'm just popping in to see if this is known or should i make a bug? (its kinda very hard to reproduce since random and it runs fine for days...)
05:02 karolherbst: because you have more sensors, it may be that the value printed by sensors is wrong
05:03 karolherbst: urjaman_: uhh ohh, it seems like flash is doing something crazy?
05:04 mupuf_: Tom^: ok, so your fan is not disabled, it spins at 20%. So the code that is wrong is the one computing the needed speed
05:04 urjaman_: karolherbst: yeah that was the only thing that did anything at the moment of crash (playing flash video), and even that only had Xv or similar available to it (didnt have the fw installed yet for decoding)
05:04 Tom^: mupuf_: mmh indeed, and since it ramps up to 100% at fanboost or 90C it seems to be able to control it atleast
05:04 mupuf_: yes
05:04 urjaman_: i added a space in the dmesg to signify the point where i think it crashed
05:04 urjaman_: *line, enter, whatever
05:04 mupuf_: so, it is likely due to linear_min_temp and linear_max_temp being wrong
05:05 Tom^: karolherbst: \o/ power1 25.10W at lowest pstate.
05:05 karolherbst: ohh
05:05 karolherbst: okay, that value doesn't seem that wrong
05:05 karolherbst: do that a couple of times and check in which range it is
05:06 karolherbst: then do nvapoke 0x20200 22722455 and see how much it drops
05:06 Tom^: 99.96W on 0f with cstate 34
05:06 karolherbst: :D
05:06 Tom^: muh power bills ;_;
05:06 karolherbst: yeah, that is a bit hard to cool with a slow fan
05:06 karolherbst: well that is only at idle :p
05:07 Tom^: yea range is around 101.0 - 101.50 W on idle
05:07 mupuf_: ah ah
05:07 mupuf_: that is a beast!
05:07 Tom^: 98.04W after nvapoke
05:07 karolherbst: 275W peak :p
05:07 karolherbst: uhh
05:07 karolherbst: 2%
05:07 karolherbst: is it also stable with load?
05:07 mupuf_: you may be the first one to need fan management for sure :D
05:08 Tom^: haha
05:08 Tom^: karolherbst: sure runs unigine heaven :p
05:08 karolherbst: nice
05:08 Tom^: sensors showed around 220W
05:08 karolherbst: mupuf_: okay, so it is trivial if we take care of this engine
05:08 Tom^: D=
05:08 karolherbst: and this engine does the most power savings
05:08 mupuf_: karolherbst: 2 sample points are not enough
05:08 karolherbst: all other envinges together were like 0.1W for me
05:09 karolherbst: I know
05:09 mupuf_: but yeah, we can concentrate on pgrap
05:09 karolherbst: yes
05:09 karolherbst: but 2% are nothing :/
05:10 karolherbst: it is around 6% for me :D
05:10 mupuf_: clock gating helps for mixed mods
05:10 mupuf_: power gating is what brings the power down on real idle scenarios
05:10 karolherbst: yeah I know, it is still something
05:11 Tom^: mupuf_: cant i set the linear_max_temp and min_temp manually to some value just to see if its parsing the vbios wrong?
05:11 karolherbst: mupuf_: by the way: I got this kepler bios thing running under wine
05:11 mupuf_: karolherbst: nice!
05:12 mupuf_: care to share how?
05:12 mupuf_: Tom^: yeah, I think you can
05:12 karolherbst: mupuf_: download .NET 3.5 full package
05:12 karolherbst: install it in clean prefix
05:12 karolherbst: disable wine mono
05:12 karolherbst: :D
05:12 Tom^: mupuf_: what values should it be tho
05:12 mupuf_: cd /sys/class/drm/card0/device/hwmon/hwmon*/
05:13 Tom^: mh
05:13 karolherbst: mupuf_: winetricks .net35 messes up for some reasons, didn't figure out what exactly fails
05:13 mupuf_: Tom^: hmm, let me check something
05:13 karolherbst: Tom^: okay, I think your gpu has a target of 260W and 275W boosted
05:15 karolherbst: Tom^: 2x6 pins power supply?
05:15 Tom^: yes
05:16 mupuf_: Tom^: sorry, not, you cannot read those values
05:17 karolherbst: 75W pcie + 2x 75W 6pin port = 225W :/
05:17 karolherbst: Tom^: sure about 2x 6 pins?
05:17 Tom^: suspected as much, ive been poking in these files earlier today :p
05:17 karolherbst: isn't it 1x6 and 1x8?
05:17 Tom^: karolherbst: uhm perhaps.
05:17 karolherbst: would you like to check?
05:17 Tom^:drops down on the floor...
05:18 RSpliet: I've got some cards with 2x 6-pin
05:18 RSpliet: that's not uncommon
05:18 karolherbst: yeah, but it isn't enough for his card
05:19 karolherbst: 2x6 pins are just +150W
05:19 karolherbst: the Titen Z has 375W TDP, ... how
05:19 karolherbst: 2x8? 75 + 2x150 = 375
05:19 karolherbst: but this is a bit... risky
05:20 RSpliet: doesn't it have a nuclear power plant embedded? that glow from the "GeForce GTX" surely has to come from somewhere
05:20 karolherbst: :D
05:20 karolherbst: yeah but 2x8 isn't included in the pcie standard
05:20 karolherbst: so it isn't a pcie card officially ;)
05:21 karolherbst: yeah, two 8 pins
05:21 karolherbst: wow
05:21 pmoreau: Are there many dual chip GPUs out there?
05:21 karolherbst: well
05:21 pmoreau: (Apart from the Titan Z)
05:21 karolherbst: yeah some
05:21 karolherbst: every gen has one
05:21 karolherbst: 690
05:21 karolherbst: 590
05:21 karolherbst: ...
05:21 karolherbst: 295
05:21 Tom^: karolherbst: i thought i never had to see this mess again http://i.imgur.com/xdVkVVv.jpg
05:21 karolherbst: 9800 GX2
05:22 Tom^: karolherbst: but yes 1x6 1x8
05:22 pmoreau: Oh okay, didn't know that
05:22 karolherbst: GeForce 7950 GX2
05:22 karolherbst: :D
05:22 RSpliet: Tom^: hahaha, looks nice
05:22 karolherbst: pmoreau: I think the GeForce 7950 GX2 is the first
05:22 Tom^: RSpliet: 6 hdds, one massive 120" cpu cooler and this beastie in a tiny case.
05:22 Tom^: RSpliet: cables, CABLES EVERYHWERE
05:23 karolherbst: Tom^: okay, so a 25W buffer
05:23 karolherbst: that seems okay
05:24 mupuf_: Tom^: can you printk all the variables here: http://cgit.freedesktop.org/~darktama/nouveau/tree/drm/nouveau/nvkm/subdev/therm/base.c#n74?
05:24 RSpliet: Tom^: cable management is a myth...
05:24 RSpliet: my machine currently contains a card so large that I had to take the HDD bracket out
05:24 RSpliet: (including the HDD and card reader it is housing)
05:24 urjaman_: 120" ? o.O :P
05:24 Tom^: mupuf_: sure
05:25 RSpliet: Tom^: that's 10ft... how much is that in non-barbaric? 3m? :-D
05:25 karolherbst: ohh I like mine
05:25 karolherbst: no cables
05:25 karolherbst: :D
05:25 karolherbst: ohh two for the fans, but meh
05:26 karolherbst: Tom^: I think it is 120mm ;)
05:26 Tom^: ^
05:26 Tom^: i dont speak non barbaric RSpliet
05:26 Tom^: 120" is 120mm
05:26 Tom^: :P
05:26 karolherbst: :D
05:26 karolherbst: yeah, I bought a 120" display too :p
05:26 RSpliet: Tom^: I've been forced to at least understand barbaric measures since moving to the UK
05:27 RSpliet: horrible measures
05:27 karolherbst: wait, UK also uses them partly?
05:27 RSpliet: UK is inconsistent
05:27 karolherbst: I thought they only use punds, huhu :p
05:27 karolherbst: *pounds
05:27 RSpliet: works with metric for measuring everything except themselves and long distances
05:28 karolherbst: RSpliet: but don't they use sea miles then?
05:28 karolherbst: for long distances?
05:28 karolherbst: would figure
05:28 RSpliet: no, land miles
05:28 RSpliet: ~1,6km
05:28 karolherbst: yeah I know, but I alsways thought UK is so used to the sea
05:28 karolherbst: :D
05:28 RSpliet: converting from KM to miles is thus simply taking the previous number in the fibonacci sequence
05:29 karolherbst: :D
05:29 RSpliet: not nearly as used to it as NL
05:29 karolherbst: I see
05:29 RSpliet: but yes, the UK a funny country...
05:29 RSpliet: they used to understand the difference between a milliard and a billion just like everyone else
05:29 karolherbst: ohh wait, sea mile to km conversion isn't fixed :O
05:29 RSpliet: until they folded for the Americans in 1974, redefining one billion to be a thousand million
05:30 karolherbst: it depends on where you are
05:30 karolherbst: ohh interessting
05:30 RSpliet: fucked up, that's what it is :-D
05:30 karolherbst: :D
05:32 karolherbst: mupuf_: anyhow, I can confirm, that the base clock from the table is the clock the blob idles at if you prefer max performance in nvidia-settings, so I guess nouveau should clock up to that for now, until we implement boosting
05:34 mupuf_: I approve, but how did you check it?
05:34 mupuf_: shall I test it?
05:34 karolherbst: well I was faking the vbios on reator
05:34 karolherbst: and this was the clock I always saw in nvidia-settings --query directly after X start
05:34 karolherbst: it was always the clock I faked into the vbios
05:34 karolherbst: but you can confirm it if wou want
05:35 karolherbst: also the blob clocks to 705 MHz ony my gpu, and the vbios contains this clock as the base
05:35 karolherbst: but it would be nice to have at least one other person confirm this :D
05:35 mupuf_: no, I would just re-do the experiment you made
05:36 karolherbst: yeah, but I couldn't check the gui
05:36 mupuf_: good finding anyway
05:36 karolherbst: so I jsut guess, the blob idles at that if you set it
05:36 mupuf_: :)
05:36 mupuf_: there is only one 'base clock' right?
05:36 karolherbst: mhhhh
05:36 karolherbst: well
05:36 karolherbst: it isn't that simple
05:37 karolherbst: there is a list of clocks
05:37 karolherbst: but the header tells you which entry is the "base" one
05:37 mupuf_: so this one should be set only when ramping to the highest perf level
05:37 karolherbst: without boosting, yes
05:37 karolherbst: https://gist.github.com/karolherbst/28fcfc36013873249077#file-mine-rom
05:37 mupuf_: yeah, I don;t like the name baseclock then
05:37 mupuf_: I would say that it is the rated clock
05:37 karolherbst: the three entries are in the header
05:37 mupuf_: and we can have the boost clock
05:38 mupuf_: what do you think about this?
05:38 mupuf_: in a way, base also makes sense actually
05:38 karolherbst: yeah
05:38 mupuf_: pick whatever you like
05:38 karolherbst: there is also a tdp entry
05:38 karolherbst: but I never verified what tdp does
05:38 mupuf_: what do you mean by tdp?
05:38 karolherbst: it was just called this way in the bios mod tool
05:38 karolherbst: but
05:38 karolherbst: I think this clock is used when you hit the tdp
05:38 karolherbst: only a wild guess though
05:39 Tom^: mupuf_: https://gist.github.com/anonymous/01437884a40ba2642ed6 this didnt work, or atleast it didnt show anything in dmesg :<
05:40 mupuf_: oh oh oh
05:41 mupuf_:is an idiot
05:41 mupuf_: http://cgit.freedesktop.org/~darktama/nouveau/tree/drm/nouveau/nvkm/subdev/bios/fan.c#n86, add fan->fan_mode = NVBIOS_THERM_FAN_LINEAR;
05:44 Tom^: mupuf_: now some values finally https://gist.github.com/anonymous/f31fe2e64797541d6293 :p
05:45 Tom^: bit much copy pasta that % behind temp: value is a bit wrong ;)
05:46 mupuf_: well, and fan management works for you now ;D
05:46 mupuf_: My work here is almost done, just need to cook a patch
05:46 mupuf_: and ask for inclusion
05:46 Tom^: nice
05:47 mupuf_: Can you PM me your full name and email please? For the reported-by and tested-by?
05:47 Tom^: yep fan is ramping up
05:47 karolherbst: Tom^: nice :D
05:47 Tom^: full pstate, full cstate, and fan working. no freezes in unigine heaven. my work here is done.
05:47 Tom^: xD
05:48 karolherbst: Tom^: now remove your -V hack ;)
05:48 Tom^: but you have to fix my volts or cstates before :<
05:48 karolherbst: it should run stbale at 1.21V as long as the temp doesn't go too high
05:48 karolherbst: nah, doesn't matter for now
05:48 karolherbst: we will clock to base clock before for now later anyway
05:48 karolherbst: just have to write some patches for that
05:48 karolherbst: meh
05:48 karolherbst: adding vbios parsing in the nouveua module :/
05:51 karolherbst: Tom^: ohh wait
05:51 karolherbst: Tom^: maybe remove your entire hack and use info.min again
05:52 karolherbst: ohh no
05:52 karolherbst: it wasn't a temp issue because it hung from the start
05:52 karolherbst: :/
05:52 Tom^: it was simply to high clock
05:52 Tom^: and also to low volt
05:52 karolherbst: Tom^: there is no such thing as to "high clock"
05:53 karolherbst: *too
05:53 karolherbst: yes
05:53 karolherbst: too low volt for a clock
05:53 karolherbst: allthough there are some physical caps you can't go over, but with your clocks you are far away from them
05:54 karolherbst: Tom^: it goes like that: each cstate has a voltage entry mapped to it, and this entries have a min-max range
05:54 karolherbst: nouveau currently only uses the min valuue
05:54 karolherbst: mupuf_: because of temp issues?
05:54 karolherbst: but the min value might be too low in some situations, or the table isn't parsed right or whatever
05:54 mupuf_: karolherbst: because we do not understand how we are supposed to pick the right value
05:54 karolherbst: the safest way is to use the max voltage, but this has several disadvantages
05:54 karolherbst: mupuf_: k
05:55 mupuf_: the problem with a voltage too high is the power dissipation which may exceed the tdp
05:55 karolherbst: 1. faster cstates not available, because of volting ranges
05:55 mupuf_: that's the real limit here
05:55 karolherbst: 2. heat and tdp ;)
05:55 Tom^: http://i.imgur.com/1HjYuAA.png never went above 72C
05:55 Tom^: \o/
05:55 karolherbst: Tom^: yeah, but the power consumption is more important
05:56 karolherbst: it shall _never_ go above 300W
05:56 karolherbst: never ever
05:56 karolherbst: and it shouldn't go above 275W
05:56 Tom^: topmost i noticed was 237
05:56 karolherbst: and shouldn't exceed 260W when (insert reasons we don't know yet)
05:57 karolherbst: Tom^: yeah I think it will be higher if you also do nvenc stuff
05:57 karolherbst: but nouveua can't do that yet
05:57 Tom^: perhaps the 260w is the boost power budget?
05:57 karolherbst: nvenc is really nice tech by the way, we should really try to figure out how to use it
05:57 karolherbst: Tom^: no, 260W is the normal tdp
05:57 karolherbst: 275W is the boosting ones
05:57 Tom^: mh ok
05:57 karolherbst: I think you can use more than 260W if the temperature is low
05:58 karolherbst: but if you hit a specific temp (maybe 95, maybe 80, who knows) you have to stick with 260
05:58 karolherbst: but it can be something else
05:58 mupuf_: karolherbst: there are more components that are monitored by power sensors
05:58 karolherbst: what else?
05:58 mupuf_: Tom^: why disable tesselation? :D
05:58 karolherbst: :D
05:59 karolherbst: 8x msaa!
05:59 karolherbst: ultra
05:59 karolherbst: !
05:59 Tom^: 8x msaa just gives this LSD effect.
05:59 karolherbst: ohh right
05:59 karolherbst: 4x msaa!
05:59 karolherbst: and tess please to normal
05:59 karolherbst: :D
05:59 karolherbst: ot the highest one because that looks just painful
05:59 mupuf_: and full screen :D
06:00 mupuf_: should increase the score
06:00 karolherbst: I have an idea
06:00 mupuf_: and composition disabled
06:01 mupuf_: the 10 FPS min clearly indicates we have some real bottlenecks at some point
06:01 Tom^: mh
06:01 mupuf_: imirkin_: do we spill a lot in our shaders for heaven?
06:01 karolherbst: mupuf_: scene loading
06:01 mupuf_: ?
06:01 karolherbst: Tom^: hit enter until you get at the first scene again
06:01 karolherbst: then pres sbenchmark
06:01 karolherbst: mupuf_: shader compilations in between
06:01 karolherbst: there are 24? scenes in the benchmark
06:01 karolherbst: and between them the engine might compile shaders
06:01 karolherbst: but these are cached
06:02 mupuf_: well, nouveau indeed compiles right before usage, so that may explain big drops
06:02 karolherbst: yes
06:02 karolherbst: I tested that already
06:04 Tom^: :o i once went to 261W
06:04 Tom^: 269!
06:04 karolherbst: yeah
06:04 karolherbst: this benchmark is just crazy
06:04 karolherbst: Tom^: you are safe though as long as you don't go over 275W
06:05 karolherbst: above that, its a risk
06:05 karolherbst: near 300 => don't do that
06:05 mupuf_: wow, I did not expect nouveau to get that close to the TDP :o
06:05 mupuf_: mad props to the mesa devs!
06:05 karolherbst: mupuf_: well 65% binary performance
06:05 karolherbst: what did you expect? :D
06:06 mupuf_: not too bad, but I wonder how nvidia can be that efficient then :D
06:06 mupuf_: well, I am sure we have some really ugly-looking shaders :p
06:07 Tom^: karolherbst: yea that worked http://i.imgur.com/HLene7P.png
06:07 Tom^: no major dips
06:12 Tom^: http://i.imgur.com/WhgAhEi.jpg so it seems its only msaa 8x thats broken.
06:12 karolherbst: 13.3 fps here :/
06:12 Tom^: the rest works , quality ultra, tesselation extreme etc.
06:12 karolherbst: 9.5 to 26.6
06:12 Tom^: its time to benchmark the blob
06:12 Tom^: see how much it differs
06:12 karolherbst: Tom^: your gpu is just a beast :D
06:13 karolherbst: Tom^: yeah, we know that 8x msaa is broken
06:13 karlmag: Tom^: since I haven't caught up with it yet and I am a curious bastard; which gpu is that?
06:13 karolherbst: it has something todo with the flags though
06:13 Tom^: karolherbst: 780ti
06:13 karolherbst: Tom^: ohh you cheated
06:14 karolherbst: Tom^: ultra quality pls :p
06:14 karolherbst: and 4x msaa
06:14 Tom^: haha
06:14 mupuf_: lol, who designs roads with rocks like that :D
06:14 karolherbst: mupuf_: extreme tesseleation
06:14 Tom^: people with square wheels.
06:14 karolherbst: Tom^: and please normal tesselation
06:14 karolherbst: I said max tesselation hurts :p
06:14 mupuf_: What if I want to walk with high heels on this? Thought about the drags? :D
06:14 mupuf_: Mouahahahah
06:15 mupuf_: the drags of heaven, by mupuf :D
06:15 Tom^: fine il do an proper 1920x1080 tess, normal, qual ultra, msa x4. then il compare to blob
06:15 karolherbst: thanks :D
06:15 mupuf_: sorry, giggling way too much over here
06:15 mupuf_: so, next issue... What will I have fun with now?
06:16 karolherbst: mupuf_: I know, power sensor
06:16 mupuf_: glamor seems sort of broken on the maxwell I am running right now
06:16 mupuf_: power sensor...
06:16 karolherbst: :p
06:16 karolherbst: we need that
06:16 karolherbst: for boosting
06:16 mupuf_: can it be my joker?
06:16 karolherbst: mhhh
06:16 karolherbst: okay
06:16 mupuf_: hmm, seems related to buffer age
06:16 mupuf_: funky
06:16 karolherbst: but it may be, that in a month I say: we _need_ the powr consumption
06:16 karolherbst: :p
06:17 mupuf_: right, and we need it for the dvfs rig too :s
06:17 Tom^: im gonna order a major success pizza now.
06:18 mupuf_: lol
06:19 mupuf_: the thing with the power sensor is this ... I need more sample points
06:19 mupuf_:was wondering about buying an nvidia gpu for his machine
06:19 mupuf_: hence why I am testing the maxwell support right now
06:19 karolherbst: nice
06:19 karolherbst: mupuf_: but you know, you have some gpus there :p
06:20 karolherbst: ohh right
06:20 karolherbst: mupuf_: you know what is still missing within your patch?
06:20 mupuf_: right, but I also got a ton of games out of the last steam sales :D
06:20 mupuf_: which patch?
06:20 karolherbst: power rail mapping
06:20 karolherbst: mupuf_: yeah well, I stole your power sensor commit from your git :p
06:20 mupuf_: yes, this is the issue I have
06:21 mupuf_: I have no idea how to do the mapping correclty
06:21 mupuf_: on some gpus, I can
06:21 mupuf_: but the INA3221 ... not always
06:21 karolherbst: well on my vbios it is easy
06:21 karolherbst: because
06:21 karolherbst: I have only one extdev
06:22 karolherbst: and only one valid power rail
06:22 Tom^: karolherbst: http://i.imgur.com/UwOVi7k.png now just to compare to blob.
06:22 karolherbst: Tom^: ahhh better
06:22 karolherbst: then I am not that far away anymore
06:22 karolherbst: yay
06:22 Tom^: haha
06:22 mupuf_: yeah, there are easy cases
06:22 karolherbst: ony 33% perf compared to 780 ti
06:22 karolherbst: this is nice for a mobile chip :p
06:23 mupuf_: yep
06:23 karolherbst: but I didn't overclocked +135MHz
06:23 karolherbst: :D
06:23 Tom^: i havent either
06:23 karolherbst: Tom^: yeah, but you hit your tdp nearly
06:23 karolherbst: you shouldn't
06:23 Tom^: im on cstate 34 so im on 1087mhz
06:23 karolherbst: ohh okay
06:23 Tom^: which is what the blob boosted to anyways.
06:24 karolherbst: yeah
06:24 karolherbst: this is a fair comparison then with the blob
06:24 karolherbst: if I OC I have to go above my cstate table anyway
06:24 karolherbst: and well
06:24 karolherbst: this is a bit out of spec
06:24 karolherbst: :D
06:24 Tom^: :p
06:25 karolherbst: EXTDEV 1: type 0x4e [INA3221] at 0x80 defbus 0 unk02_5 2 unk03 0x02 => power rail 0: extdev_id/power_rail = 1, shunt resistor = 5 mOhm
06:25 karolherbst: mhh
06:25 karolherbst: extev.id == power_rail.exted_id
06:25 karolherbst: this is the only valid mapping I see here
06:26 karolherbst: mupufs kepler: EXTDEV 0: type 0x4e [INA3221] at 0x80 defbus 0 => power rail 0: extdev_id/power_rail = 0, shunt resistor = 5 mOhm
06:27 karolherbst: yeah maybe this works mupuf_?
06:27 karolherbst: same for Tom^
06:27 mupuf_: not always
06:27 mupuf_: why do you think I called it power_rail too? :D
06:27 karolherbst: mupuf_: where not?
06:27 mupuf_: I have more than one rail
06:28 karolherbst: I know
06:28 karolherbst: I jsut saw this: EXTDEV 2: type 0x02 [MAX6649] at 0x98 defbus 0
06:28 mupuf_: so, we cannot encode both the extdev_id and the power rail
06:28 karolherbst: and there is no power_rail with extdev_id 2
06:28 karolherbst: :(
06:28 mupuf_: so, guessed that if we use the INA3221, then there is only one anyway
06:28 mupuf_: so it is fine
06:28 karolherbst: ohh okay
06:28 karolherbst: so the MAX6649 would get all the others or just all?
06:28 mupuf_: but... some vbios do not work
06:29 mupuf_: MAX6649 is a temp sensor IIRC
06:29 karolherbst: okay
06:29 karolherbst: mupuf_: which vbios do you have which doesn't work?
06:29 mupuf_: give me a sec
06:33 mupuf_: ah, yes, I remember
06:33 mupuf_: the problem was: What should we do when there is an INA3221 and INA219?
06:33 karolherbst: ohh right
06:33 mupuf_: nve4/pecisk/vbios.rom for instance
06:34 karolherbst: at least there are two ina219 entries
06:34 mupuf_: then I wanted to trace what the blob is doing ...
06:34 karolherbst: I bet the 2mOHM thingies are for the ina219
06:34 mupuf_: but ... I can't
06:34 mupuf_: I would need to dump the content of the RAM of pmu
06:34 mupuf_: and find the power there
06:34 karolherbst: uhh
06:35 karolherbst: sounds ugly
06:35 mupuf_: yep...
06:35 mupuf_: short of having access to a really expensive card ... I can't
06:35 mupuf_: and I would need to fake INA219 from the bus if I wanted to reverse this case
06:36 mupuf_: and see, some cards count from 1 to 3 for the power rail
06:36 mupuf_: and some from 0 to 2
06:36 karolherbst: mhhh
06:36 karolherbst: I think if the pecisk one is figured out, everyone is figured out :D
06:37 karolherbst: this just looks like there is some logic we just can't see
06:38 Tom^: karolherbst: http://i.imgur.com/LDI1hLN.png
06:38 Tom^: :<
06:38 Tom^: blob ran around 10C hotter too
06:38 karolherbst: mupuf_: ohh
06:38 karolherbst: the extev id isn't part of the vbios
06:38 karolherbst: it is just the position in the list :/
06:39 karolherbst: yeah well, should be the same anyway
06:39 karolherbst: mupuf_: there is more in the header though
06:39 karolherbst: ohh no, it isn'T
06:40 karolherbst: meh, I thought I was already fully waken up now
06:40 karolherbst: Tom^: what did you expect? tiny difference? :D
06:40 mupuf_: I have NFC what all those fields represent
06:40 karolherbst: Tom^: yeah well, maybe the fan was also slower? who knows
06:40 Tom^: i expected noveau to run at 300fps
06:40 mupuf_: and it is hard without nvidia reporting the power usage
06:41 Tom^: ;_;
06:41 mupuf_: by changing the values of the resistors, you can make it hit the tdp
06:41 mupuf_: but even that is not good because we do not understand the power budget table either
06:41 karolherbst: right
06:41 karolherbst: there are several entries and some might look just sane :/
06:41 mupuf_: so, here we are
06:42 Tom^: karolherbst: but anyways its quite acceptable performance from nouveau :p i see no use of the blob
06:42 karolherbst: :D
06:42 karolherbst: with gallium nine there is none already
06:42 karolherbst: :D
06:42 karolherbst: Tom^: and there are games which run better with gallium nine than native linux
06:42 mupuf_: Tom^: very nice :)
06:43 karolherbst: *cough* post that under the phoronix article *cough* :D
06:43 karolherbst: ohh wait
06:43 karolherbst: he did test valley
06:43 karolherbst: nvm then
06:44 Tom^: yea this is still with quirks and manually setting cstate to lower so its not quite ready yet. :p
06:44 mupuf_: AGREED
06:44 mupuf_: sorry for the caps
06:45 karolherbst: also the generated gpu binaries aren't perfect either
06:46 karolherbst: mupuf_: want to look over those patches? http://lists.freedesktop.org/archives/nouveau/2015-November/023304.html
06:46 karolherbst: but I think I don'T want them to be landed until we cap to base clock
06:46 karolherbst: otherwise gpus might get hot
06:46 karolherbst: :D
06:48 karolherbst: mhh
06:48 karolherbst: I have some better ideas for the second patch anyway
06:53 karolherbst: https://github.com/karolherbst/nouveau/commits/kepler_voltage
06:55 karolherbst: mupuf_: the latter patch should fix up your pwm based kepler card
06:56 karolherbst: and nouveau won't go above the pwm max value aymore
06:58 mupuf_: I onkly have one pwm-based voltage mabagement gpu
06:58 mupuf_: and it is my maxwell
06:58 karolherbst: really?
06:58 karolherbst: mhh ohh right then
06:59 karolherbst: well then that one should be fixed
06:59 karolherbst: :D
07:01 mupuf_: what do you mean by fixed?
07:01 karolherbst: well
07:01 karolherbst: some voltage map entries go above 1.2V
07:01 karolherbst: and the pwm duty was set above 0x60 in such cases
07:01 mupuf_: oh, interesting
07:01 karolherbst: yeah
07:01 karolherbst: but I figured out whiy
07:01 karolherbst: becuase nouveau just clocked to much higher clocks
07:02 karolherbst: basically it is like this: there is a base clock, a boost clock and your cstates
07:02 karolherbst: there is always a cstate for the base clock and boost clock
07:02 karolherbst: and higher ones
07:02 karolherbst: if the gpu is cool enough and the power consumption under control, the blob clocks above the boost clock
07:02 karolherbst: for that it needs those higher cstates
07:03 karolherbst: but nouveua just clocked to the highest one
07:03 karolherbst: which basically means: max boost
07:03 karolherbst: but there seems to be also entries, which have voltage entries attached, which can't be handled by that gpu
07:03 karolherbst: the blob never clocks to them anyway
07:04 karolherbst: Tom^: 1300MHz was your highest cstate or something, right?
07:04 karolherbst: and the blob clocked to 1099 something?
07:04 Tom^: yep
07:05 Tom^: both blob and windows went to 1097 at most at 1.175V
07:05 karolherbst: yeah
07:05 karolherbst: mupuf_: so yeah, I think we should just drop those cstates
07:05 Tom^: at full load that is, stock idle was only like 980 or so
07:05 karolherbst: yeah
07:05 karolherbst: 980 is your base clock
07:05 mupuf_: karolherbst: agreed
07:06 karolherbst: so that fixes those volting issue which only occured on high cstates :)
07:06 mupuf_: I understand your patch for the headerless, but I would like to make sure it is not going to wreak havoc in the co,,on code
07:06 karolherbst: yeah I know
07:06 karolherbst: pmoreau tested the second one on his teslas
07:06 karolherbst: but I really want to have more testing there
07:07 karolherbst: mupuf_: actually, the second patch is only dangerous for gpus with no max/min values in the vbios
07:07 pmoreau: And my laptop melted after testing it… :'-(
07:07 karolherbst: so where we iteratore over the entries
07:07 karolherbst: :p
07:07 pmoreau: ;-)
07:07 karolherbst: but yeah, more testing
07:10 Tom^: forgot i had ordered pizza, came as a pleasent surprise.
07:10 Tom^: nom nom.
07:11 pmoreau: :-D
07:11 Tom^: pmoreau: did you add headers to your distro yet?
07:11 pmoreau: No… :-(
07:12 pmoreau: But I finally started rewriting my SPIR-V work so…
07:14 pmoreau: Tom^: While you are at testing, would you mind doing some Mesa testing, to check if I can get some kernels running on Fermi+, or if I'm still screwing up inputs on those?
07:14 Tom^: im running mesa-git :o
07:14 Tom^: or did you mean testing in general on your distro
07:15 Tom^: i thought you meant [testing] in arch.
07:15 Tom^: but sure i dont mind testing, just tell me what you want and il do it :p
07:15 pmoreau: Testing a branch of mine, which parses SPIR-V and generate the corresponding NV50 IR code (far far far from complete), which enables you to run some **really** basic OpenCL kernels on Nouveau
07:16 Tom^: sure handover your branch and tell me what you want tested. :p
07:17 pmoreau: Just need to update a few things in the archive
07:17 Tom^: sure thing
07:19 imirkin: Tom^: mupuf_: fyi the mesa 11.1 release will compile shaders at creation time, not use time. should hopefully decrease fps drops during the game, in exchange for slower load times
07:19 Tom^: imirkin: nice
07:19 mupuf_: imirkin: yep, very nice :)
07:19 karolherbst: imirkin: did it land already?
07:19 imirkin: yeah a long time ago
07:19 karolherbst: ohhh
07:19 imirkin: but i dunno what Tom^ is testing with
07:20 Tom^: mesa-git :<
07:20 imirkin: if it's with 11.0.x, then it won't have that
07:20 karolherbst: imirkin: but I think unigine creates those shaders not from the start :/
07:20 Tom^: imirkin: 11.2.0_devel.74740.ea3f394-1
07:20 Tom^: also the fps drops i think they were on the blob too
07:20 Tom^: so its more of a unigine problem i suspect
07:20 karolherbst: or
07:20 imirkin: ok that's pretty recent
07:20 karolherbst: it is made on purpose :p
07:20 karolherbst: but then
07:20 imirkin: should def have the changes in question
07:20 karolherbst: mhhh
07:21 karolherbst: bad benchmark design then
07:22 pmoreau: Tom^: I think this should work: https://phabricator.pmoreau.org/F13589 There is an INSTALL file in it explaining how things should be set up
07:23 Tom^: roger
07:25 imirkin: no compute on nvf0 yet
07:26 pmoreau: Oh…
07:26 Tom^: meaning, i cant test it? :|
07:27 hakzsam: please, provide a MMT trace if you want compute support on nvf0 :)
07:27 hakzsam: you can't
07:27 pmoreau: Tom^: Then, could you run the script in https://phabricator.pmoreau.org/F12234 and send me the resulting MMT files please?
07:27 imirkin: there's some prelim support in nouveau for nvf0
07:27 imirkin: but it does something funky which causes it to fail on nouveau ctxsw firmware
07:27 imirkin: however the suspicion is that it's just totally unnecessary in the first place
07:27 imirkin: but... need to investigate
07:28 Tom^: pmoreau: valgrind: failed to start tool 'mmt' for platform 'amd64-linux': No such file or directory
07:29 pmoreau: Oh, you need valgrind-mmt
07:29 pmoreau: Tom^: http://nouveau.freedesktop.org/wiki/Valgrind-mmt/
07:29 Tom^: its in the AUR, im fetching it.
07:29 Tom^: <3 archlinux
07:30 pmoreau: I didn't check AUR, I just built it :-D
07:31 karolherbst: mupuf_: any idea how this vbios parsing works in nouveau? I have a little hard time figuring out the function names
07:31 karolherbst: like what is Ep and Em?
07:31 mupuf_: ah, yes
07:31 mupuf_: it is a little confusing at first
07:32 karolherbst: Te seems to be the table header
07:32 karolherbst: and Ee seems to be the entry index or something
07:32 karolherbst: or entry header
07:32 karolherbst: but no clue about Ep, Em, Xe and Xp
07:32 mupuf_: are you trying to hack one existing code or writing your own?
07:32 karolherbst: writing my own
07:33 mupuf_: then write whatever makes sense for you
07:33 karolherbst: but I would like to follow the design
07:33 mupuf_: just follow the external interface
07:33 Tom^: pmoreau: uhm segmention fault on all uh "tests"
07:33 karolherbst: extev?
07:33 karolherbst: *extdev
07:33 karolherbst: uhhh
07:33 karolherbst: yeah that looks more nice :D
07:34 karolherbst: I don't like those type * name arguments in functions
07:34 karolherbst: somehow I always think a struct is better suited for that
07:34 Tom^: pmoreau: "./start_traces: line 6: 17699 Segmentation fault (core dumped) valgrind --tool=mmt --mmt-trace-nvidia-ioctls --log-file="${1}.mmt" "./${1}" 0"
07:34 pmoreau: Tom^: Sorry, forgot to specify you should run the script while running the blob
07:34 Tom^: oh xD
07:34 pmoreau: Otherwise it's not going to help us RE compute on your card. :-D
07:34 imirkin: karolherbst: te = table entry
07:35 Tom^: pmoreau: haha duh!
07:35 karolherbst: imirkin: yeah, I was already that far ;)
07:35 pmoreau: Tom^: But yeah, I should definitely have specified, it wasn't really clear.
07:35 imirkin: p = pointer
07:35 karolherbst: ohhh
07:36 karolherbst: yeah I don't like that kind of interface, I make something easy to use and easy to understand :p
07:41 Tom^: pmoreau: still segmention faults
07:42 pmoreau: :-/
07:42 pmoreau: What if you run the programs by hand?
07:42 Tom^: seg fault
07:43 pmoreau: Meh…
07:45 pmoreau: So `./get_global_id_3d 0` does segfault?
07:45 Tom^: there we go, gdb'it apparently it used my opencl-mesa-git libs.
07:45 Tom^: had to uninstall them before it used the nvidia ones
07:45 pmoreau: Interesting, it's the opposite on my laptop
07:46 Tom^: so you want all those .mmt then?
07:46 pmoreau: Every time it tries to use the blob ones, but can't modprobe the module, so it moves on the Mesa ones.
07:46 pmoreau: Yes pleas :-)
07:46 pmoreau: *please
07:48 Tom^: pmoreau: https://www.dropbox.com/s/wy4xtd3bc1jgt4a/mmt.tar.gz?dl=0
07:49 pmoreau: Thanks!
07:50 pmoreau: Let's see
07:50 hakzsam: Tom^, could you also please trace vectorAdd from the CUDA samples on your nvf0?
07:51 Tom^: you could just aswell have uttered latin there.
07:51 Tom^: excuse me, trace what on what now? =D
07:51 pmoreau: vectorAdd is a CUDA sample adding two vectors
07:52 pmoreau: and nvf0 is the chipset of your card ;-)
07:52 pmoreau: Tom^: The path to the sample: /opt/cuda/samples/0_Simple/vectorAdd
07:53 Tom^: just gotta figure out what packages that
07:53 hakzsam: pacman -S cuda
07:53 Tom^: cuda-7.0 from aur sounds old
07:54 pmoreau: cuda from extra is 7.5.18 IIRC
07:54 Tom^: O_o odd i searched for cuda and missed it entirerly.
07:54 pmoreau: s/extra/community
07:56 imirkin: hmmmm.... demmt is somehow failing on these traces =/
07:56 pmoreau: Yeah
07:57 pmoreau: It gives some output for get_local_id_3d.mmt, but completely fails on get_global_id_3d.mmt
07:57 imirkin: oh
07:57 imirkin: that was just Tom^ failing
07:57 imirkin: it's a tar file
07:57 Tom^: O_o
07:57 imirkin: but i never see a LAUNCH
07:58 imirkin: and i see obviously-pushbuf writes to an fd
07:58 imirkin: but demmt never picks it up
07:58 imirkin: oh well. a mystery for another day.
07:59 Tom^: https://www.dropbox.com/sh/br6ow3to21x2fnx/AAAV3rwcEUiQut9jS9nbsLA7a?dl=0
07:59 Tom^: retraced, and untared-
08:01 imirkin: urjaman_: btw, that DMA_PUSHER thing is something i've seen before, but no clue how it happens
08:02 karolherbst: mupuf_: do you know if I have access to nvbios in the nvkm_clock ctor?
08:03 Tom^: pmoreau: https://gist.github.com/anonymous/5919c6e10eb23451313e just a normal trace of it?
08:03 pmoreau: The get_global_id_3d works now, great! But some parsing missing… for some other time
08:04 pmoreau: Tom^: I think hakzsam wants an MMT trace of it, just like the ones you just did.
08:04 Tom^: i see
08:04 hakzsam: sure, thanks
08:08 Tom^: hakzsam: https://www.dropbox.com/sh/br6ow3to21x2fnx/AAAV3rwcEUiQut9jS9nbsLA7a?dl=0 vectorAdd.mmt
08:10 imirkin: interesting. looks like there might be 2 separate IB buffers?
08:12 hakzsam: Tom^, thanks
08:13 Tom^: np
08:17 hakzsam: it's a nvf1 not a nvf0 btw
08:26 mupuf_: karolherbst: you mean, access to the bios? Sure!
08:26 karolherbst: mupuf_: no, access to the nvkm_bios object ;)
08:26 mupuf_: yes
08:27 karolherbst: ahh okay
08:27 mupuf_: check every subdev, it will show you
08:27 karolherbst: but only after nvkm_subdev_ctor I suppose ;)
08:27 mupuf_: oh, I see
08:27 mupuf_: no, fan uses it in ctor
08:27 karolherbst: k
08:37 karolherbst: mupuf_: as nvkm_info? [29668.783709] nouveau 0000:01:00.0: clk: base: 0 MHz, boost: 0 MHz
08:37 karolherbst: just with valid numbers :D
08:37 karolherbst: or would you like some other print
08:37 karolherbst: or none at all?
08:37 mupuf_: bios should be initialized before clock
08:37 mupuf_: I don;t get it
08:37 karolherbst: I didn'T read the stuff out yet
08:38 karolherbst: just dummy functions
09:06 karolherbst: mupuf_: yay: [31413.952666] nouveau 0000:01:00.0: clk: base: 1411 MHz, boost: 1594 MHz
09:06 karolherbst: though maybe I want to / 2 them to get the "official" clocks
09:06 karolherbst: in the print
09:09 karolherbst: https://github.com/karolherbst/nouveau/commits/base_clocks
09:09 karolherbst: ohh have to edit header
09:13 john_cephalopoda: HEy
09:22 john_cephalopoda: I read the bug report about the flightgear bug and it doesn't seem to be exactly what I get.
09:34 karolherbst: mupuf_: NvBoost options, if disabled: drop cstates above the base clock. Does this sound sane?
09:34 karolherbst: I am not sure though it dropping cstates entirely in this case makes that much sense :/
09:36 mupuf_: why would you do that?
09:36 karolherbst: well if we disable boosting, the highest clock available should be the base clock from that table
09:36 karolherbst: and only enabling boosting will change that
09:37 karolherbst: but should it be possible to change that after the module was loaded?
09:38 karolherbst: have to go :/
09:38 Tom^: blob doesnt let you enable/disable it. nvidia simply stated this is the way it works now.
09:38 Tom^: xD
09:41 imirkin: john_cephalopoda: make a new one, apitrace, etc
09:52 karolherbst_phon: Tom^: yeah lol and then we get above the tdp
09:58 john_cephalopoda: imirkin: Just realized that mesa is maybe outdated.
09:58 imirkin: yeah, def test with a recent mesa... stuff gets fixed
09:58 imirkin: (and broken. heh.)
09:59 Tom^: interesting, whenever i play a movie it seems everything else gets vsynced towards its fps.
10:00 Tom^: or maybe not, hm 2k fps in glxgears while the flashmovie is running. it seems to just really chog thigns down.
10:07 pmoreau: mwk: Thanks a lot for all the reviews, especially for the renaming as it was the most tedious one I would say! :-)
10:07 pmoreau: imirkin: And also, thank you for the reviews ;-)
10:07 Tom^: yea disabling hw acceleration of chromium solved that
10:08 john_cephalopoda: imirkin: I have used archlinux for a long time and now on source-based crux - and somehow it seems like the most stable and bug-free software is git or latest release, while the old stuff and stable releases are broken all the time.
10:09 imirkin: john_cephalopoda: yeah but those bugs don't get in there on their own
10:09 imirkin: while fixing stuff, new problems are added, etc
10:11 john_cephalopoda: imirkin: But those bugs tend to be smaller than the previous ones - at least in the software I used :D
10:51 Tom^: this is just more and more awesome. cs:go 150+ stable fps. i simply cant find a single reason to use the blob any more
10:53 imirkin: Tom^: stability, opencl/cuda are probably the main ones
10:53 Tom^: i guess, and wouldnt hurt with dynamic reclocking but its in the works so. :p
15:24 pmoreau: "nv50_tls_realloc:745 - Unsupported number of temporaries (268435455 > 3929). Fixable if someone cares." O.O
15:24 pmoreau: I just created one function, a couple of BBs and one constant…
15:31 imirkin: pmoreau: you messed something else up... either that or it thinks you're using a crazy amount of l[] space
15:31 pmoreau: :-/
15:33 pmoreau: imirkin: The output with debug=255: https://phabricator.pmoreau.org/P67
21:35 tagr: RSpliet: official name for what? couldn't reconstruct the context from IRC logs, too many conversations going on =\
21:37 tagr: pmoreau: you might want to ask that question to the openlinux mailing list, because I suspect that gnurou will be just as unfamiliar with EVO than I am, given it's not present on Tegra