00:01 imirkin: which, coincidentally, is why i care about ES - nobody in their right mind is going to run nouveau on a device that *actually* needs an ES driver
00:02 imirkin: [not like there are a ton of such devices to begin with]
00:08 imirkin: is there a fermi plugged into reator?
00:12 karolherbst: imirkin: I think so
00:13 imirkin: are you using it?
00:13 karolherbst: imirkin: last time I check it was fermi and maxwell
00:13 karolherbst: imirkin: no
00:13 imirkin: k, thanks
00:13 karolherbst: imirkin: but if you check, tell me what is currently plugged in :D
00:13 imirkin: hehe ok
00:13 karolherbst: imirkin: mupuf said this the last time: "karolherbst: I plugged another fermi along with the maxwell2"
00:13 imirkin: :)
00:14 imirkin: karolherbst: GM206 and GF114
00:16 karolherbst: imirkin: k, when you are done, I think I still wanna check something :)
00:16 imirkin: gr, need to set up a blob version on there that works...
00:16 imirkin: i hate computers.
00:17 karolherbst: and I hate smart software :D
00:41 karolherbst: pmoreau: I get a build error without the checks
00:41 karolherbst: pmoreau: "../../../../include/c99_math.h:221:12: error: ‘constexpr bool std::isinf(double)’ conflicts with a previous declaration"
00:41 karolherbst: " __MATHDECL_1 (int,isinf,, (_Mdouble_ __value)) __attribute__ ((__const__))"
00:41 karolherbst: pmoreau: same for isnan
00:55 imirkin: karolherbst: you going to be around for a little bit? i have a patch to test on kepler + deqp
00:55 karolherbst: imirkin: yeah, I guess so
00:55 imirkin: karolherbst: http://hastebin.com/ogosivesiw.pl - try running deqp with --deqp-case='dEQP-GLES3.functional.transform_feedback.*'
00:56 karolherbst: against mesa master=
00:56 imirkin: ya
00:56 imirkin: actually just run dEQP-GLES3.functional.transform_feedback.basic_types.interleaved.points.lowp_float
00:56 imirkin: and see if that one's failing
00:56 imirkin: if it is, then ignore
00:56 imirkin: (i mean, don't bother running the rest)
00:57 karolherbst: well it applies
00:57 imirkin: this is the full cmdline: ./deqp-gles3 --deqp-visibility=hidden --deqp-case='dEQP-GLES3.functional.transform_feedback.*'
00:58 imirkin: from modules/gles3
00:58 imirkin: Failed: 0/1260 (0.0%)
00:58 imirkin: that seems nice.
00:58 imirkin: [on fermi]
01:00 karolherbst: uhhh, kernel rounds down by default
01:00 imirkin: ?
01:00 karolherbst: unrelated
01:01 karolherbst: imirkin: how many tests do you expect to fail when it fails=
01:01 karolherbst: ?
01:02 imirkin: 0
01:02 imirkin: er
01:02 imirkin: when it fails, like 100% :)
01:02 imirkin: or... a large %
01:02 karolherbst: ahh okay
01:03 karolherbst: seems like nothing fails for now
01:03 imirkin: let it finish if you don't mind... shouldn't take too long
01:03 imirkin: a few minutes at most
01:03 karolherbst: yeah, it seems fast enough
01:03 imirkin: i assume that one test passed?
01:03 imirkin: or did you go for the whole group at once?
01:04 karolherbst: you never send me which one you want to have
01:04 karolherbst: ohh wait
01:04 karolherbst: you did
01:04 imirkin: i did, but no worries.
01:04 karolherbst: didn't run it
01:08 karolherbst: imirkin: 8.6% not supported
01:08 karolherbst: others passed
01:09 imirkin: excellent, thanks for checking
01:09 imirkin: i'll push that out then
01:11 karolherbst: done with reator then?
01:11 imirkin: mmmmmmmm yeah
01:11 imirkin: go ahead, i won't turn it off
01:12 imirkin: but i logged out
01:13 karolherbst: k
01:14 karolherbst: anyway, I think I finally found out how the blob calculates the voltage on fermi and newer :)
01:17 imirkin: oh crap
01:17 imirkin: that still fails on fermi
01:17 imirkin: i forgot to install
01:27 karolherbst: imirkin: I am done with reator if you still need it
01:27 imirkin: nope
01:27 karolherbst: mupuf: the gm206 has a really odd clock reading by the way
01:27 imirkin: i have the trace
01:27 imirkin: i just need to think
01:27 imirkin: about why it works for them and fails for me
01:30 karolherbst: mupuf: okay, same error on all 9 cards :)
01:46 martm: i have dumped on current git, bunch of errors having updated my kernel too, on Warthunder
01:47 martm: it seems like there is a problem with general thread-safety in gallium level, but sometimes the game runs on lower resulutions
01:50 martm: http://dpaste.com/04QP161
01:52 martm: i getting oibafs debug builds , and post another dump
01:59 martm: now the famous freeze, on cinnamon i managed to understand how i would not get a lockup, doing alt+f1 to make the wwarthunder into windowed mode
01:59 martm: http://dpaste.com/36GWT9D
01:59 martm: this is i normally almost always receive a freeze, could add a picture too
02:00 martm: it's a stalingrad map
02:07 martm: but there happen randomly all sort of mesa state tracking issues and segfaults
02:09 martm: http://dpaste.com/3TF806D that is another where there is a problem
02:15 martm: http://dpaste.com/2TRJBKX
02:15 martm: first time was able to run the game 6seconds with hight settings
02:18 martm: imirkin: i belive this can debugged under Xephyr even better, but i use cinnamons alt-f1 which is kinda exposee feature
02:43 martm: http://dpaste.com/1Q499Z5
03:11 martm: http://dpaste.com/0DFY3HV
03:24 martm: imo i seem to get most of the deadlocks on cpu not on gpu, and randomly...there are some of the functions to blame, i am thinking since radeonsi uses gallium, can't some of state manipulation be copied from them, RSpliet: this really is too complex indeed
03:39 martm: I belive i should probably try valgrind or some dynamic IR stuff, on this game if the cpu locks up i can not longer get any sort of stacktrace
03:41 martm: reentrancy problems in the state tracker, because functions are not reentrant it gives a stale data and program segfaults
04:06 martm: http://valgrind.org/docs/manual/drd-manual.html
05:06 martm: http://clang.llvm.org/docs/ThreadSafetyAnalysis.html
05:42 martm: mupuf: unfortunently the cpu code seems to be not thread-safe in very many places, i should forgive upfront that people who think it's non trivial are right, like RSpliet
05:45 martm: you guys really want to run Warthunder with valgrind or threadsanitizer, yes also the drm and kernel and all of them
08:31 bozhan: setting 0f in pstate of my GK107 i get : nouveau 0000:01:00.0: fifo: CHSW_ERROR 00000001 in dmesg, when i try to run steam
08:34 martm: i have somehwere stored how to do lockless algorithms, it was some casting mechnism, but i am hoping to do it automatically with this tool
08:34 martm: https://github.com/jaingaurav/ReserveTM/blob/master/doc/sigplanconf-template.tex
09:15 linuxguy: Hi
09:15 linuxguy: I don't see 970M and 980M on the list https://nouveau.freedesktop.org/wiki/CodeNames/#nv110familymaxwell - does it mean that these cards are not supported?
09:22 linuxguy: ?
09:23 Wonka: linuxguy: people knowing the answer might have a client here but not currently paying attention to it. Just wait some more.
09:23 Wonka: it's still nighttime in the US, for example.
09:24 linuxguy: It's evening in Japan.
09:26 huehner: linuxguy: maybe nobody got around to either check or just update the wiki. I assume both fall either into GM100 or GM200 categories like the other maxwells
09:27 huehner: linuxguy: in general both of those are supported (100 + 200) however 3d accelleration has been added only very very recently (mesa 10.2 + kernel 4.5 i think)
09:27 huehner: linuxguy: also no reclocking so most likely 3d performance will be very low as by default most cards run on there lowest speed
09:29 linuxguy: I found on wiki https://en.wikipedia.org/wiki/GeForce_900_serie that mobile versions of 970 and 980 have the same code name so maybe that's the reason? But if so then it would be good to update https://nouveau.freedesktop.org/wiki/CodeNames/#nv110familymaxwell to avoid confusion. These are expensive cards and for many ppl ability to handle linux is crucial.
09:30 linuxguy: huehner: thanks for clarification.
11:27 karolherbst: mupuf: nvidia made a lot use of the new clocking stuff on your maxwell2, the maxwell1 didn't use it though (or it never triggered for me yesterday) :/
11:29 karolherbst: bozhan: still there?
11:33 bozhan: yep
11:33 karolherbst: bozhan: I assume you use the stock nouveau module?
11:33 bozhan: are these REnouveau dump still usefull?
11:34 karolherbst: mhh not quite sure, I think not for reclocking on kepler and maxwell
11:34 karolherbst: we are pretty far now with how to reclock properly on those
11:34 karolherbst: it's just not mainlined yet
11:34 karolherbst: bozhan: do you build your kernel yourself?
11:34 bozhan: karolherbst: 4.6.0-rc3
11:34 bozhan: yes
11:35 karolherbst: bozhan: would it be possible for you to run nvidia for a simple test?
11:35 karolherbst: I think I got the last missing pieces yesterday and just need confirmation from what nvidia does
11:35 bozhan: karolherbst: nvidia driver? i think yes but need time to see how to install it
11:36 karolherbst: which distribution are you running?
11:36 bozhan: debian
11:36 karolherbst: uhhh
11:36 karolherbst: then you better not mess with it there :D
11:36 bozhan: sid :)
11:36 karolherbst: it is quite ugly on debian based systems (I tried to install both on an ubuntu machine, and it was messy)
11:37 karolherbst: bozhan: well you could just try out an out of tree nouveau module
11:37 karolherbst: but you would need to reinstall it for every kernel upgrade
11:37 bozhan: karolherbst: whatever i'll try install
11:37 karolherbst: no idea if it compiles against 4.6.0-rc3 though :D
11:38 karolherbst: ohh wait
11:38 karolherbst: gk107 it was?
11:39 bozhan: yep
11:39 bozhan: optimus
11:39 karolherbst: ahh good
11:39 karolherbst: bozhan: dri3 xorg server setup?
11:41 bozhan: not now...tried it once and it doens't work well but maybe becouse of some bug in kernel 4.4.x
11:41 bozhan: i can try it again but first binary driver there is version 355 ? or 352?
11:42 karolherbst: bozhan: git clone https://github.com/karolherbst/nouveau.git -b 4.6_reclocking
11:42 karolherbst: bozhan: don't bother setting up nvidia, it will be a mess and I don't want to waste your time to repair your system again :D
11:42 karolherbst: on my system even the login manager didn't want to start anymore, allthough it was an optimus system
11:43 bozhan: i;ve done this several time ... i have no choice :) whateve i'll clone you tree
11:45 bozhan: karolherbst: compiling :)
11:45 karolherbst: bozhan: inside drm?
11:45 bozhan: yep
11:45 karolherbst: k
11:46 karolherbst: so I suppose you also know how to install a kernel module? :)
11:47 bozhan: karolherbst: just read that it isn't recomend to install in it... just insmod ?
11:47 karolherbst: mhh
11:47 karolherbst: replace your nouveau.ko(.xz) file in /lib/modules
11:47 karolherbst: and regenerate initramfs
11:48 karolherbst: make install puts nouveua into a directory called extras so you end up with two nouveau modules in the end
11:49 bozhan: karolherbst: first i need to restart because of echo 07 > pstate , still getting hangs
11:56 martm: int a,b,c,d;
11:56 martm: int bar(int& x) {
11:56 martm: return x+1;
11:56 martm: }
11:56 martm: by compiler does not compile such code
11:57 martm: a heck this reserveTM i have tried to generate something with it, but it does not seem to function
11:57 martm: diff shows, that everything is the same allthough i run the pass
12:00 martm: mart-B85M-D3H-A ReserveTM # /usr/local/bin/opt -load /usr/local/lib/LLVMReserveTM.so -ReserveTM < teine1.bc > teine2.bc
12:00 martm: mart-B85M-D3H-A ReserveTM # diff teine1.bc teine2.bc
12:01 martm: on a modified version of the example code, it does fucking nothing
12:11 bozhan: karolherbst: now what?
12:11 karolherbst: bozhan: nouveau is loaded?
12:12 bozhan: yes
12:12 karolherbst: bozhan: check debugfs if there is now a entry called "boost" besides the pstate one
12:12 bozhan: karolherbst: *1: 1019 MHz
12:12 karolherbst: okay
12:12 karolherbst: now try to go to pstate 0f
12:14 bozhan: karolherbst: dmesg is full of red messages ;) like : nouveau 0000:01:00.0: pmu: wait on reply timed out
12:14 karolherbst: uhh
12:14 bozhan: and : drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request.
12:14 karolherbst: how could that happens
12:14 karolherbst: bozhan: even before trying to set to 0f?
12:15 bozhan: karolherbst: i think after : echo 0x > pstate
12:15 bozhan: karolherbst: after
12:16 karolherbst: bozhan: mhh I would need to know what happens at the beginning though
12:17 bozhan: karolherbst: http://pastebin.com/MbYvNP3K
12:19 karolherbst: okay, so something fedup the GPU
12:19 karolherbst: there were new commits in master which I pulled but not tested yet, let me try on my machine
12:20 karolherbst: mhh
12:20 karolherbst: for me it works
12:20 bozhan: k
12:21 karolherbst: bozhan: could you reboot and check if the pstate 07 is fine?
12:23 bozhan: ok
12:35 oeuvre: is this the art channel
12:35 oeuvre: i was looking for some art nouveau
12:35 karolherbst: oeuvre: I really doubt it
12:35 mwk: oeuvre: nope, it's about a GPU driver
12:36 oeuvre: does it support my S3 ViRGE?
12:37 mwk: sorry, nvidia only
12:37 oeuvre: LAME
12:37 karolherbst: ...
12:38 mwk: heh.
12:39 karolherbst: are those still used?
12:39 mwk: nope, long dead
12:39 mwk: I mean, well, I suppose you could find an operational one somewhere
12:39 karolherbst: odd
12:39 bozhan: karolherbst: 07 even bad. have to push power button
12:40 bozhan: badder
12:40 bozhan: haha worst :)
12:40 mwk: I think I have a working S3 Trio somewhere, which is... ancienter
12:40 karolherbst: actually it is worse
12:40 bozhan: yea ... :)
12:41 karolherbst: bozhan: mhhh, I have no idea why the memory controller messes up, but maybe I did something in the patches which are really bad or somebody else did something else
12:41 karolherbst: bozhan: ever used git bisect?
12:42 karolherbst: bozhan: well let's try the oldest commit you can test
12:42 karolherbst: bozhan: git checkout 3b3ec4e10ade69abcf0e10f46eb63293c7949698
12:42 bozhan: karolherbst: nope but willing to try :)
12:42 karolherbst: then compile nouveua again and install it
12:43 bozhan: where i have to run that in nouveau or in nouveau/drm or outside?
12:43 karolherbst: doesn't matter
12:43 karolherbst: well you still have to run make inside drm
12:44 karolherbst: but git checkout can be executed from everywhere I think
12:44 bozhan: no it have to be in nouveau, whatever, now: HEAD is now at 3b3ec4e... drm-next 67d1c0a25c05e2105d12abd9c0172d2d5c0e7654
12:45 karolherbst: ohh maybe your git version is a bit old then
12:45 karolherbst: I have 2.8.1 installed
12:45 bozhan: karolherbst: git version 2.8.0.rc3
12:45 karolherbst: mhh odd. well at least for me it works
12:45 karolherbst: k, the HEAD is good
12:45 karolherbst: try to reclock to 07 on this commit and see what happens
12:45 bozhan: ok now ... nouveau/drm$make?
12:46 bozhan: k
12:46 karolherbst: yeah and overwrite your nouveau module in /lib/modules again and update initramfs
12:46 karolherbst: and the boost file should disappear
12:46 karolherbst: mhh maybe we want to include the git version somewhere in nouveau...
12:47 bozhan: karolherbst: make install places module in /lib/moudulse/kerne/extra.... but module.dep doesn't update? do i have to place new copiled module in kernel/drivers/drm/nouveau....
12:47 karolherbst: bozhan: yeah, just overwrite the old one
12:47 bozhan: k
12:52 bozhan: i have to write one disk.... and then i'll reboot
13:00 G33KatWork: hey guys. I bought a new notebook: a lenovo T460p with optimus and a GTX940m. lenovo says in a readme all external display outputs are connected to the nvidia gpu, but that's wrong. they work fine with the nvidia gpu turned off in vgaswitcheroo. I saw that my chip wasn't recognized by nouveau on boot, so I googled and created a small patch which copies the GM107 function pointer table and uses it for the GM108. I can render on the chip now using DRI_P
13:01 G33KatWork: i'd like to see that show up in mainline maybe if somebody else is able to test this more thoroughly
13:07 bozhan: karolherbst: now ? 07 > pstate?
13:08 karolherbst: bozhan: yeah
13:09 bozhan: now i don't see boosts?
13:09 karolherbst: bozhan: right
13:10 bozhan: karolherbst: [чт апр 14 16:09:47 2016] WARNING:[чт апр 14 16:09:47 2016] WARNING: CPU: 1 PID: 0 at /home/bozhan/nouveau/drm/nouveau/nvkm/subdev/mc/base.c:51 nvkm_mc_intr+0xff/0x110 [nouveau] CPU: 1 PID: 0 at /home/bozhan/nouveau/drm/nouveau/nvkm/subdev/mc/base.c:51 nvkm_mc_intr+0xff/0x110 [nouveau]
13:11 karolherbst: bozhan: hah, it isn't me who messed itup :D
13:11 bozhan: k
13:11 karolherbst: bozhan: mhh the issue is now though, that earlier commits need an older kernel
13:11 bozhan: am
13:12 karolherbst: it could be though that the pcie link changes messes up the GPU a little
13:12 karolherbst: though I doubt this
13:12 karolherbst: well lets try it out though
13:12 bozhan: and now ... maybe i just have to wait
13:13 karolherbst: bozhan: https://github.com/karolherbst/nouveau/blob/3b3ec4e10ade69abcf0e10f46eb63293c7949698/drm/nouveau/nvkm/subdev/clk/base.c#L191
13:13 karolherbst: could you remove this line locally?
13:14 bozhan: remove or comment?
13:14 karolherbst: doesn't matter
13:14 karolherbst: with git checkout $path you can always reset files
13:14 bozhan: end now
13:14 bozhan: and now
13:14 karolherbst: compile and install again :)
13:15 bozhan: k
13:15 bozhan: ok
13:15 bozhan: CC [M] /home/bozhan/nouveau/drm/nouveau/nvkm/subdev/clk/base.o
13:15 bozhan: /home/bozhan/nouveau/drm/nouveau/nvkm/subdev/clk/base.c: In function ‘nvkm_pstate_prog’:
13:15 bozhan: /home/bozhan/nouveau/drm/nouveau/nvkm/subdev/clk/base.c:179:19: warning: unused variable ‘pci’ [-Wunused-variable]
13:15 bozhan: struct nvkm_pci *pci = subdev->device->pci;
13:15 bozhan: sorry
13:15 karolherbst: yeah, doesnt matter
13:15 karolherbst: it is just a warning
13:32 bozhan: karolherbst: first, for some time just: ouveau 0000:01:00.0: fb: ramcfg data for 0MHz not found
13:32 karolherbst: huh...
13:32 bozhan: and nouveau 0000:01:00.0: clk: unable to find matching pll values
13:32 karolherbst: something is messed up, but I have no idea what
13:33 bozhan: than hang. crash ... :)
13:33 karolherbst: bozhan: yeah, the gpu just "disconnects" from the bus and doesn't really reply to anything
13:33 bozhan: and now : kernel:[ 876.466551] NMI watchdog: BUG: soft lockup - CPU#4 stuck for 23s! [kworker/4:1:129]
13:33 karolherbst: ohh
13:33 bozhan: reboot:)
13:33 karolherbst: they really should add the PID of those kworkers
13:40 bozhan: karolherbst: no luck :)
13:41 karolherbst: bozhan: yeah, something is odd, and nouveau does something wrong, but I have no idea what
13:41 karolherbst: bozhan: I think a mmiotrace of the nvidia driver would be needed now
13:41 karolherbst: bozhan: ohh wait
13:41 karolherbst: bozhan: maybe just trace nouveau
13:42 karolherbst: bozhan: we should see there where the gpu gets fed up with us
13:42 bozhan: k thanks have a nice day. tomorrow if i have time again :)
13:42 karolherbst: okay
14:07 karolherbst: pmoreau: got my message from yesterday about removing those glibc checks?
14:38 pmoreau: karolherbst: Yep, I saw them! Not working, as expected :-)
14:38 pmoreau: Thanks for testing!
14:38 karolherbst: k
14:54 pmoreau: I can take testing requests on my dekstop cards now, since I finally compiled a custom Linux + Nouveau to get the GM206 going.
14:55 karolherbst: :)
14:55 karolherbst: awesome
14:56 karolherbst: pmoreau: well there is an issue with reading out the clocks on nouveau on nvf0, nvf1 and gm107+ , but all the tests I have include the nvidia driver :D
14:56 pmoreau: :-D
14:56 pmoreau: It should be do-able to install the blob ;-)
14:56 karolherbst: and also includes that nouveau reads out the clocks right
14:56 karolherbst: there is something funky with the gpc clock
14:57 karolherbst: pmoreau: wait a sec
14:57 pmoreau: I want to test my code on the GM206 and the GF100, then plug in a G84 to check an old bug report, and then I should be able to test whatever you want.
14:58 karolherbst: pmoreau: with nouveau loaded: nvapeek 0x132000 0x20
14:58 pmoreau: On which card?
14:58 karolherbst: gm206
14:58 pmoreau: K
14:58 karolherbst: 13200c should be not 0
14:59 karolherbst: maybe 0x137018 is funny too
15:00 pmoreau: Need to get envytools first :-D
15:00 karolherbst: :D
15:05 pmoreau: 0x13200c is 0
15:06 karolherbst: pmoreau: and the other regs?
15:08 pmoreau: 132000: 10010000
15:08 pmoreau: 132004: 000100801
15:08 pmoreau: Sorry: 132000: 18010000
15:09 pmoreau: 132010: badf5040 249 0 badf5040
15:10 karolherbst: odd
15:10 karolherbst: ohhh
15:10 pmoreau: 137018: 1e030038
15:10 karolherbst: 132 is the mem thing
15:10 karolherbst: right
15:10 karolherbst: I meant 137
15:10 karolherbst: nvapeek 0x137000 0x20
15:12 karolherbst: 0x137014 4-7
15:13 karolherbst: 13700c 0-15
15:13 karolherbst: and something in 13701c too
15:13 pmoreau: 0x137000: 0010020 00014201 00000000 0200f0f0
15:13 karolherbst: these affect the gpc clock on some way...
15:14 pmoreau: 0x137100:40000024 e601e00 1e030038 4100372a
15:15 karolherbst: I meant 0x137000 0x10 though
15:15 karolherbst: ...
15:15 karolherbst: 0x20
15:15 pmoreau: Sorry, it was 0x137010
15:16 karolherbst: k
15:16 karolherbst: on my kepler:
15:16 karolherbst: 00132000: 18010000 00010801 00000000 00000000
15:16 karolherbst: 00132010: 00000000 00000249 00000000 00000000
15:16 karolherbst: so those new regs have to be investigated...
15:17 karolherbst: I plan to do this on a nvf0 because I am sure with maxwell there is even more new there
15:21 karolherbst: and I also think nvatiming is a bit broken for the GPC frequency :/
15:24 pmoreau: Hum, interesting
16:00 karolherbst: mupuf: yeah, adding 0.1% to the voltage seems to do the trick for now
18:26 karolherbst: uhhh saints row 2 and 3 for Linux too
18:27 karolherbst: I hope the nouveau performance doens't suck as bad as on IV
18:27 karolherbst: :)
18:32 karolherbst: mupuf: do we want to limit the input of NvVoltOffsetmV in any way? Like only allow +-100mV or shouldn't it matter, because we have out voltage limits anyway?
18:33 karolherbst: and I think I won't exceed the info.min and info.max ranges, because they seem to be right
18:33 karolherbst: ...
18:33 karolherbst: yeah
18:33 karolherbst: it doesn't make sense to limit the input, because we will limit the possible voltage anyway
19:06 Yoshimo: can someone tell me where this installation is messed up? https://pastee.org/2capd i can't figure out why the screen stays empty although i am using oibafs current drivers and mesa builts
19:09 pmoreau: Yoshimo: Do you have anything in /dev/dri?
19:10 Yoshimo: anything else you could need to help me? I have to reboot to get anything usefull
19:11 pmoreau: The dmesg as well, and maybe check that you are in the video group
19:11 pmoreau: > "[drm] Failed to open DRM device for pci:0000:01:00.0: -19"
19:11 pmoreau: I’m sure I’ve already seen that one, but I can’t remember where and why… :-/
19:13 Yoshimo: bbl
19:28 Yoshimo: dev/dri has 3 files, a control file a card0 and third one
19:37 pmoreau: Ok
19:37 pmoreau: And do you have a dmesg?
19:37 Yoshimo: yes , in a minute, Firefox is updating right nwo
19:37 pmoreau: No problem
19:49 Yoshimo: https://pastee.org/2jd6f
19:50 karolherbst: Yoshimo: looks okay I think
19:51 karolherbst: Yoshimo: how do you start X?
19:51 Yoshimo: well, except for the fact that the graphic login screen doesn't appear and i am doomed to use the terminal and i have no idea what went wrong
19:51 Yoshimo: well usually the system starts the gui login itself, so something went wrong in last weeks update
19:51 karolherbst: ahh okay
19:51 Misanthropos: karolherbst, regarding gk106 voltage issues...
19:52 karolherbst: Yoshimo: lightdm or something else?
19:52 Misanthropos: did i get that right, that you cant do anything about that atm?
19:52 karolherbst: Misanthropos: no, it is done nearly
19:52 Misanthropos: oh cool :D
19:52 karolherbst: Misanthropos: did you mean the issue that we are a little bit below nvidia?
19:52 Yoshimo: sddm it should be as it is kubuntu 15.10
19:52 Misanthropos: i think so.
19:53 karolherbst: Misanthropos: https://github.com/karolherbst/nouveau/commit/791a920a0bc3d77d2fdcb4d0eae2b4cdf65ff184
19:53 Yoshimo: i think they switched with 15.04
19:53 karolherbst: :D
19:53 Misanthropos: at least i know its not stable without info.max
19:53 karolherbst: Yoshimo: ahh okay
19:53 karolherbst: Misanthropos: yeah, this is a bit troublesome
19:53 karolherbst: Misanthropos: you could try out my newest branch though
19:53 karolherbst: Misanthropos: I doubt it will change much, but who knows
19:54 Misanthropos: can i use it on 4.6.x?
19:54 karolherbst: Misanthropos: if you rebase on master
19:54 karolherbst: ohh wait
19:54 karolherbst: Misanthropos: 4.6_reclocking branch
19:55 Yoshimo: how often do you rebase your branches karol? with each new official kernel release?
19:56 karolherbst: Yoshimo: more often
19:57 karolherbst: Yoshimo: usually whenever I notice that skeggsb pushed new stuff on his branch
19:57 karolherbst: so they usually contain all the newest code
19:57 karolherbst: even all the 4.7 stuff
19:59 Yoshimo: so who wants to live on the bleeding edge and the wild side, should use your master?
20:03 karolherbst: you forgot the testing part. I don't test...
20:03 karolherbst: I just remove some drm-next commits, see that there are no build errors and that's it :D
20:04 Yoshimo: any other idea for my login issue tonight? Otherwise i don't have to care about which branch to use
20:04 karolherbst: Yoshimo: check logs of your login manager
20:05 karolherbst: Yoshimo: and see if the login manager has the right access rights
20:05 karolherbst: Yoshimo: running systemd?
20:05 Yoshimo: default these days on kubuntu
20:05 karolherbst: allthough sddm doesn't allow non rooted X yet
20:05 karolherbst: ...
20:05 karolherbst: Yoshimo: idea
20:05 karolherbst: Yoshimo: go into a tty and just start X
20:05 karolherbst: maybe even use startx
20:06 pmoreau: karolherbst: You should have a look at his Xorg.log: X fails to open /dev/dri/card0
20:06 karolherbst: pmoreau: yeah, which makes no sense if the X server is started as root
20:06 Yoshimo: startx fails after a while, last time it said something about the graphic driver not being installed properly
20:06 karolherbst: mhhh
20:06 karolherbst: I had a long time debugging an issue I had on my ubuntu machine
20:06 karolherbst: in the end it was a missing directory for lightdm
20:07 Yoshimo: so i could open single applications like muon but the kdeinit failed
20:07 pmoreau: Yoshimo: You should remove xf86-video-nouveau, if that wasn’t already done.
20:08 Yoshimo: pmoreau: i'll check that, otherwise i will be back tomorrow with more time
20:09 karolherbst: nomodeset ....
20:10 karolherbst: pmoreau: see that?
20:10 karolherbst: "ro recovery nomodeset"
20:10 pmoreau: I looked for it in the dmesg, but not in the Xorg.log… uhu
20:10 karolherbst: yeah well the modesetting driver won't work with nomodeset right?
20:11 pmoreau: Given the names, I guess not
20:11 karolherbst: now I noticed it like a minute too late
20:12 karolherbst: if there is an error while starting X or a login manager, it is always bs like that
20:12 karolherbst: ...
20:12 karolherbst: always
20:12 karolherbst: pmoreau: by the way, michael was able to run at full clocks with my recent kernel tree on the 780 ti :)
20:13 pmoreau: Nice! :-)
20:13 pmoreau: Should we expect updated benchmarks?
20:13 karolherbst: he is still doing the benchmarks
20:13 pmoreau: K
20:14 karolherbst: he just pinged me on IRC to confirm his setup is right
20:15 karolherbst: pmoreau: he will upload data after he is finished with boost set to 2
20:15 pmoreau: Oh, he even got that working? Great!
20:16 karolherbst: well he just started with boos 2
20:16 pmoreau: K
20:16 karolherbst: pmoreau: and now I work on the first OC stuff for nouveau :) though I lack a proper safe idea how to implement a volt offset for the fermi and newer :/
20:17 karolherbst: this option should allow a user to modify the mapped voltage
20:17 karolherbst: to increase stability on low voltages and increase performance if it's stable enough
20:17 karolherbst: all acording to the vbios though
20:18 karolherbst: so it will just reduce/increase the voltage for each cstate, so that more/less cstates are available
20:18 pmoreau: Why try to get OC before dynamic reclocking? Because it’s manual?
20:19 karolherbst: pmoreau: well, the idea was that a user can increase the voltage if nouveau isn't stable enough
20:19 karolherbst: but
20:19 karolherbst: with the same option the user can also just decrease teh voltage
20:19 karolherbst: which makes no difference in the code
20:19 pmoreau: Ok
20:19 karolherbst: this just happen to make higher cstates available on most high and mid end GPUs
20:20 karolherbst: so if a cstates requires 1.22V but the GPU can only do 1.2125V, the user can specify -8mV
20:20 karolherbst: and the former cstates requires 1.212V
20:20 karolherbst: and fits into the hardware limit
21:14 hakzsam: sadness, vshl on gk110 is not documented in envydis
21:16 imirkin_: hakzsam: you can easily create an atlas of all the instructions with a bit of shell scripting and nvdisasm
21:16 imirkin_: hakzsam: i have one for sm35 on my other comp, can check where vshl is later
21:20 hakzsam: imirkin_, no worries, I have a MMT trace :)
21:20 hakzsam: 000000c8: 021c040a b9f3ff01 $r2 $r1 $r4 0x0 ??? [unknown: 00000000 b9f00301] [unknown instruction]
21:20 imirkin_: that works too
21:21 hakzsam: imirkin_, well, about maxwell it seems like that only the ISA has changed
21:21 hakzsam: but we need tess before
21:21 imirkin_: and SUATOM? :)
21:21 imirkin_: before what?
21:22 imirkin_: compute/images? not really...
21:22 hakzsam: I mean, having images without tess is a bit useless because we can't bump to GL 4.2
21:22 hakzsam: and yeah SUATOM is a pain
21:22 imirkin_: meh. you could bump to ES 3.1
21:22 imirkin_: and games could use it
21:23 hakzsam: didn't know that
21:23 imirkin_: e.g. unreal engine doesn't use tess
21:23 hakzsam: but anyway, focus on fermi and gk208 first
21:23 imirkin_: but it does use compute/images/etc (when available)
21:23 hakzsam: okay
21:23 hakzsam: gk208 should not be hard to have
21:24 hakzsam: mostly need to RE some surface instructions and write code emission
21:24 imirkin_: ops tend to be fairly standard... use the emitForm_21 & co helpers
21:24 imirkin_: or at least read through them
21:24 hakzsam: yeah, will do
22:43 karolherbst: hakzsam: I heard that saints row usees compute shaders :O
22:44 hakzsam: yep
22:44 karolherbst: I would like to try them out then :D
22:45 karolherbst: so that I can see if this really makes a difference in performance
22:45 karolherbst: just this series? https://lists.freedesktop.org/archives/mesa-dev/2016-April/111522.html
22:50 imirkin_: i think he has a tree at https://cgit.freedesktop.org/~hakzsam/mesa/log/?h=gk104_images
22:50 imirkin_: that should enable compute as well, since compute depends on images
22:50 karolherbst: yeah I am currently searching the right think
22:51 karolherbst: *thing
22:51 hakzsam: karolherbst, or you can override with MESA_EXTENSION_OVERRIDE
22:51 karolherbst: hakzsam: so it is in master already, but just not enabled by default?
22:52 hakzsam: it's not enabled because compute is useless without images
22:52 karolherbst: ohh okay
22:52 imirkin_: basically all compute shaders use images
22:52 imirkin_: you can feel free to prove me wrong
22:52 karolherbst: hakzsam: do you know what saints row needs to make use of compute shaders then?
22:53 imirkin_: but i've never seen one outside of testcases
22:53 hakzsam: karolherbst, apply my images series for gk104 and override ARB_compute_shader
22:53 hakzsam: this should be enough
22:53 imirkin_: shouldn't even need to do that
22:53 karolherbst: hakzsam: so your gk104_images branch?
22:53 imirkin_: it should auto-enable itself
22:54 hakzsam: this one https://cgit.freedesktop.org/~hakzsam/mesa/log/?h=gk104_images
22:54 karolherbst: yeah
22:54 karolherbst: that's why imirkin_ and I meant :)
22:55 hakzsam: imirkin_, you need to force-enable arb_compute_shader
22:55 imirkin_: why
22:55 imirkin_: has the patch to do it based on images && counters not been pushed?
22:55 hakzsam: because it won't be exposed otherwise :)
22:55 hakzsam: no :)
22:55 imirkin_: oh sad
22:55 imirkin_: i thought that went out already
22:56 hakzsam: it's part of compute shaders series for radeonsi
22:56 imirkin_: should get bas to just push it
22:56 hakzsam: yeah
22:57 karolherbst: :)
22:57 karolherbst: I hope the game runs better with it then
22:58 karolherbst: because on lowest settings it is unplayable on my gpu :/
22:58 hakzsam: I have never tried
22:58 imirkin_: karolherbst: set the GL/GLSL version oerrides to 430 while you're at it
22:58 karolherbst: right
22:59 hakzsam: karolherbst, does this game is available from steam?
22:59 karolherbst: yeah
22:59 hakzsam: I guess not free?
22:59 karolherbst: 2 not so much yet
22:59 karolherbst: no
22:59 hakzsam: sad
22:59 imirkin_: hakzsam: the feral guy has said he'll give out keys to people trying to fix feral-ported games
23:00 karolherbst: is it feral ported?
23:00 imirkin_: (of which saints isn't one)
23:00 karolherbst: yeah
23:00 karolherbst: it is an eon port I think
23:00 hakzsam: imirkin_, cool
23:00 imirkin_: VP i think
23:00 karolherbst: yeah
23:00 karolherbst: with the funny glsl shader code :)
23:01 karolherbst: I think it will be nice to optimize those compute shaders
23:01 karolherbst: or maybe just the compute_shader fallback stuff is really bad
23:01 karolherbst: and with it we get a 50% perf boost or something
23:02 imirkin_: well, the uintBitsToFloat stuff doesn't really affect noueau
23:02 imirkin_: it's just a "mov"
23:02 imirkin_: :)
23:02 karolherbst: right :D
23:02 imirkin_: makes it impossible to read tho
23:02 hakzsam: wil try shadow of mordor which uses compute shaders too
23:02 karolherbst: and I still need to take care of those BB which are optimized away and leave a lot of instruction in post RA
23:02 karolherbst: imirkin_: but isn't the branch emptied while doing RA?
23:03 imirkin_: karolherbst: before i think
23:03 karolherbst: mhh
23:03 hakzsam: 51053MB, what the fuck?
23:03 imirkin_: or perhaps we need to do a bit of flattening pre-ra
23:03 imirkin_: dunno
23:03 karolherbst: hakzsam: normal :D
23:03 imirkin_: hakzsam: welcome to the future
23:03 hakzsam: I don't have enough free space on my hdd :)
23:03 karolherbst: hakzsam: you don't want to know how much my games need here :D
23:04 hakzsam: I don't, should be crazy I guess :)
23:04 karolherbst: 147GB...
23:04 karolherbst: on one of my partitions :D
23:04 imirkin_: it doesn't help that steam doesn't work with large XFS partitions
23:04 hakzsam: well, I can't even install this game
23:04 imirkin_: i have a large XFS partition for precisely this issue
23:04 imirkin_: and i can't use it with steam
23:04 imirkin_: coz they don't handle 64-bit inodes
23:04 karolherbst: 281GB on my main....
23:04 karolherbst: linux games^
23:04 imirkin_: (> 2T i believe is when this issue hits)
23:05 karolherbst: mhh only 86GB on my wine stuff
23:05 imirkin_: anyways, bbl
23:05 karolherbst: hakzsam: so yeah, but 50GB is really much though
23:06 hakzsam: karolherbst, well, I need to buy a new hdd if I want to use steam :)
23:06 karolherbst: I think I have one game needing so much?
23:06 hakzsam: 50G is just crazy
23:06 karolherbst: even bioshock only needs 41.3GB
23:08 karolherbst: I think max payne 3 needs alos 50GB?
23:08 hakzsam: no ideas
23:08 karolherbst: but it only gets worse :D
23:09 hakzsam: karolherbst, anyway, please report if you have serious issues with that game
23:09 karolherbst: I have without computer_shaders at least
23:09 karolherbst: performance being the serious issue :D
23:09 hakzsam: I was not talking about performance :) but yeah it's not cool to have a few number of FPS
23:10 karolherbst: well there were no visual issues so far
23:10 karolherbst: but I will try with compute shaders and see what is different
23:11 hakzsam: okay
23:11 karolherbst: hakzsam: GL_ARB_shader_image_load_store was it?
23:12 hakzsam: what?
23:12 karolherbst: the image extension
23:12 hakzsam: it is
23:12 karolherbst: k
23:13 karolherbst: then it is here with your branch :)
23:13 karolherbst: good
23:13 hakzsam: and GL_ARB_shader_image_size
23:14 karolherbst: ahh right
23:15 hakzsam: btw, what about performance with blob and that game? I guess it's much better but how?
23:16 orbea: does this look familiar to anyone? I got like 3 gbs of it spammed in my logs, but haven't been able to reproduce it yet... http://dpaste.com/1RA7QJ1
23:17 hakzsam: not really
23:19 karolherbst: hakzsam: well on nouveau I get like 9-13 fps on lowest settings in one scene
23:19 hakzsam: orbea, which card?
23:19 orbea: gtx 780 ti
23:20 hakzsam: gk110b, well try to reproduce the issue and open an issue on our bugtracker :)
23:20 hakzsam: I have no clue right now
23:21 karolherbst: orbea: is it reproducible?
23:21 orbea: not yet...
23:21 karolherbst: hakzsam: same scene, 60 fps nvidia plain
23:21 karolherbst: hakzsam: nouveau was fully reclocked
23:22 orbea: doesn't seem to be any games I played recently or mpv /mplayer
23:22 hakzsam: karolherbst, huge difference
23:22 karolherbst: yeah
23:22 karolherbst: this is too huge
23:22 hakzsam: yeah
23:22 karolherbst: I disable vsync now on nvidia
23:22 hakzsam: well, time to sleep, see ya!
23:23 karolherbst: hakzsam: 64-75 fps
23:23 karolherbst: now I try images
23:31 karolherbst: hakzsam: mhh no difference