02:26imirkin: if we're ever to pass CTS, i think we need to make a directed attempt at triaging all the failures across all chips and figuring out why they happen
02:27imirkin: i've created a new trello board - https://trello.com/b/lfM6VGGA/nouveau-cts - for precisely this task
02:28imirkin: let me know if you're interested in being part of that effort, and i'll add you to the board
02:29airlied: imirkin: you can pass CTS on one chip :-)
02:30imirkin: airlied: perhaps i phrased that wrong - what i meant is that we need to do the analysis on each chip, not just on one chip. i.e. if the same test fails on 2 chips, that probably needs to be treated as 2 failures
02:30imirkin: depends on the situation
02:31gnarface: fun patches?
07:30librin-thinkpad: are there any known issues related to running nouveau on UEFI? (UEFI on the mainboard, not the card itself)
07:35mlankhorst: uefi runs before nouveau, so don't see how..
07:39mjg59: mlankhorst: Card tends to come up in a different mode, legacy VGA addresses may not be mapped
07:39mjg59: Nothing that *should* matter, but
07:44librin-thinkpad: I traced back my flickering problem on higher refresh rates (which I found to actually be triggered at >=75Hz, not >60Hz, I tested modes with 74.9hz, even) to be UEFI-related
07:44librin-thinkpad: if I do a UEFI boot with CSM (Compatibility Support Module) disabled, I hit the problem
07:44librin-thinkpad: if I enable CSM back, even though it's not even used, it's all fine
07:44mjg59: librin-thinkpad: It's not UEFI as such, it'll be that the UEFI driver has configured the card slightly differently
07:45mjg59: If CSM is enabled it's probably running the legacy video BIOS instead
07:48librin-thinkpad: And in the end, Nouveau fails to properly handle that. The way it gets configured when CSM is not present, I mean
07:49librin-thinkpad: imirkin: any 2¢ on this?
08:52RSpliet: librin-thinkpad: that's kernel stuff presumably, I... guess you could boot your kernel with nouveau.NvForcePost=1 and see if it makes a difference?
08:52RSpliet: Report your results along with anything else to the freedesktop bugzilla (https://nouveau.freedesktop.org/wiki/Bugs/)
08:53RSpliet: and expect to be asked for mmiotraces of your kernel on boot
08:54librin-thinkpad: RSpliet: thanks a million!
09:43karolherbst: imirkin: why a new board?
10:37AndrewR: Hi all... may be I have another nouveau OpenGL-related bug .... At least I tried to run cinelerra (video editing software, open-source) on nouveau, and while one dev branch worked with OpenGL eacceleration, it had strange behavior when paused with one or more effects, and default branch simply crashed with opengl output/acceleration. But both worked with LIBGL_ALWAYS_SOFTWARE=1 https://git.cinelerra-cv.org/gitweb . I finally compiled new apitrace, and
10:37AndrewR: can try to trace this application (seems to be one rare case when anyone uses pbuffers).
10:42AndrewR: right now it tend to crash like this (main branch, https://git.cinelerra-cv.org/gitweb?p=CinelerraCV.git;a=summary) https://pastebin.com/0gHQgxdt
10:58RSpliet: AndrewR: you might need to install debugging symbols for nouveau_dri.so to get a more meaningful backtrace
10:59RSpliet: (and for others who can't be bothered to fire up bing translate: that signal is segfault)
10:59AndrewR: RSpliet, thanks ..
11:07zabell: Hey, I see that nouveau have support for GP107 but trying the mobile version GP107M seams to fail, is that intentional?
11:08RSpliet: zabell: no. Are you using a brand new kernel? what's failing?
11:09RSpliet: Also I don't think GP107 can be used for anything other than display modesetting and software rendering until NVIDIA releases their firmwares
11:10RSpliet: which in a laptop is rather useless, considering 99% of all displays are driven by the Intel GPU instead
11:10zabell: RSpliet: I have a Dell XPS 15 with Nvidia GTX 1050, the firmware loads now on Fedora 25, but I can only run wayland on the Intel graphics. And running lspci from terminal crashes the whole system.
11:11RSpliet: sounds like problems with power management
11:11zabell: Okay, I have read that it's no implemented yet, just wondered if there is a difference between GP107 and GP107M
11:11RSpliet: no there isn't
11:12RSpliet: Well, not as far as nouveau is concerned
11:12zabell: So just use the Nvidia driver until the kernel gets some more updates? :)
11:12zabell: Thought maybe my hardware was faulty
11:12RSpliet: Also, I don't think NVIDIA has released firmware for GP107
11:13RSpliet: so I highly doubt anything is loading at this point
11:13zabell: They have some time ago
11:13RSpliet: Oh that went completely past me. Cool
11:13RSpliet: you can try booting with disabled runpm
11:14RSpliet: and a 4.11 kernel from Koji
11:14zabell: okay to disable power management?
11:15RSpliet: well, I don't quite know what bits it disables to be fair. Probably something with suspend/resume, so it'll probably impact battery life and power consumption
11:15zabell: I guess I will wait for power management to work, becuase without that I might as well use Nvidia
11:15RSpliet: but with boot clocks I doubt it can overheat
11:15RSpliet: if you feel like experimenting, boot with nouveau.runpm=... I always forget. Either 0 or 1
11:16zabell: Though nvidia-drm modesetting it really hard to get working on new hardware, they have quite a lot of bugs.
11:16zabell: runpm is default 1 on optimus systems.
11:16zabell: So I could try setting it to null
11:16RSpliet: according to the wiki it's -1
11:16zabell: ohh yea sorry :D
11:17zabell: Is that a typo, or does -1 do something special i wonder.
11:17RSpliet: it's probably a semi-clever "auto"
11:17RSpliet: don't know the code well enough tbh
11:18zabell: yea okay
11:19zabell: Atleast now I know it is probably not my hardware.
13:55imirkin: karolherbst: dunno, seems a little orthogonal. this is basically a bug tracking and investigation board, broken down by family.
13:56karolherbst: well, we could have the same lists in the nouveau board as well
13:57imirkin: could be. but then it'd get confusing.
16:39orbea: So slackware current updated gcc-5.4.0 to gcc-7.1.0 and now pcsx2 kind of just broke... As in the OpenGL gsdx will render nothing, broken images, or severe graphical issues (Epileptic flashing). The software gsdx still works however. Here are two traces that show the issue, although the second one mostly replays correctly. http://ks392457.kimsufi.com/orbea/stuff/trace/PCSX2_gcc-7.1.0_ggx2.trace.xz and
16:39orbea: http://ks392457.kimsufi.com/orbea/stuff/trace/PCSX2_gcc-7.1.0_tota.trace.xz Here is an pcsx2 issue as well: https://github.com/PCSX2/pcsx2/issues/1937 So what happened? Is this nouveau specific? Mesa? gcc? pcsx2?
16:40imirkin_: let's see what happens replaying it on i965
16:40orbea: also an example of 'broken images', it varies though. http://ks392457.kimsufi.com/orbea/stuff/pics/scrots/pcsx2/gsdx_20170509134534.png
16:42imirkin_: 5801: message: application error 57005: ERROR: Source and Target are the same!
16:42orbea: I get the same
16:42imirkin_: seems like something bad is happening.
16:43orbea: I guess maybe it might be better to ask in #dri-devel?
16:43imirkin_: it does tend to render ok on i965 once you get into the game
16:43imirkin_: before then i get lots of blinking
16:44orbea: yea, without apitrace it will render nothing until the fmv which blinks a lot
16:45orbea: the other trace shows just garbage like I see without apitrace
16:45orbea: and prints a lot of: 129: message: application error 57005: ERROR: Update RenderTarget 0x0 bw:10 (0,0 => 640,480)
16:45orbea: 2218: message: application error 57005: OI_GsMemClear (0,0 => 639,479)
16:46imirkin_: that's something put in by the application
16:46imirkin_: so actually i think it's pcsx2 that's inserting those messages
16:46imirkin_: not mesa
16:50orbea: well, its fairly disconcerting for it to go from working fairly well to nothing. Well, I guess it was probably a time bomb waiting to go off...
18:09imirkin_: fp64 is back...
18:28karolherbst: Tensor cores... interesting
18:32karolherbst: :O " Volta GV100 is the first GPU to support independent thread scheduling, which enables finer-grain synchronization and cooperation between parallel threads in a program."
19:35RSpliet: "Transistors: 21.1 billion"
19:36RSpliet: That beast just got 30% bigger than its predecessor
19:36imirkin_: each one's smaller though
19:37RSpliet: well, with processes named after the smallest feature, that's not guaranteed
19:38imirkin_: still smaller than these guys, i bet: https://www.parts-express.com/electro-harmonix-el84-eh-vacuum-tube--072-880
19:38RSpliet: and cheaper, presumably
19:39imirkin_: at least on a per-unit basis :)
19:39RSpliet: yeah... guess we'll have to request a quote for 21.1 billion of them to verify
19:55gregory38: imirkin_: orbea: I'm sorry
19:55gregory38: those errrorsa are "normal"
19:56gregory38: I used them to monitor crappy emulation corner case
19:56imirkin_: well, sounds like there's still some real issue though =/
19:56gregory38: mesa seems to be fineon my side
19:56imirkin_: with the new gcc?
19:57gregory38: yes butit could be related to gcc 7
19:57gregory38: I need to check the apitrace on my sied
20:00karolherbst: Lyude: I have an idea: let's work on having Nouveau running on switch :O
20:01Lyude: seriously? that sounds kinda neat
20:01Lyude: granted though I am not sure if I've got the time between nouveau, working on <redacted> at work, and probably reverse engineering the mali gpu series
20:02Lyude: or the t600 at least
20:02karolherbst: nah, I don't care about the CPU side, I just want to replace nvidias blobs with nouveau :D
20:02karolherbst: and why mali?
20:03karolherbst: my phone has a mali as well
20:03Lyude: I haven't checked but I don't think we have a re'd driver for it
20:03Lyude: at least from what robclark told me
20:03Lyude: i would very much like to actually have my own drm driver. would help make me feel cool and also teach me a lot of stuff
20:03karolherbst: I think somebody wanted to work on mali?
20:03Lyude: yeah, some person apparently did work on it already and wrote a driver but, didn't release the source code for it?
20:04karolherbst: sounds about right I think
20:04karolherbst: I don't know the details though
20:04karolherbst: I think there was some sillyness going on, that's why it never happen, but I could remember wrongly
20:04Lyude: i am going to try to poke them at some point but if I don't end up getting much info from them hey, just means more re work for me
20:05Lyude: the most important part of all of this is that I decided to name the driver something along the lines of "flogging mali"
20:07karolherbst: Lyude: why mali though?
20:08Lyude: karolherbst: cause rob had a chromebook with it, and I just kinda want the experience of writing a set of drivers for a gpu from scratch
20:08karolherbst: I see
20:08imirkin_: Lyude: it's a lot of work
20:09imirkin_: but you get to know how GL works real well
20:09Lyude: imirkin_: yeah, I figured it was going to be
20:10Lyude: of course though i'm still working on nouveau as well :), tbh that is on a higher priority atm
20:12Echelon9: karolherbst, Lyude: Re getting nouveau working on the Switch (Tegra T210), I've been working on getting nouveau working on the SHIELD TV (Tegra T210)
20:13karolherbst: Echelon9: well, on the shield it should "just work" (tm)
20:13Echelon9: Unlocked bootloader and Android base is a starting point
20:14Echelon9: Device tree shenanigans aside etc
20:14karolherbst: and that switch is using FreeBSD
20:16Echelon9: No one appears to have fastboot / bootloader access on the Switch yet
20:16Echelon9: The SoC's should be broadly equivalent
21:58karolherbst: imirkin: it sounds like that with the tensor cores there will be a more vector centred ISA for a special core type?
21:59karolherbst: 120 TFLOPS of "matrix-multiply-and-accumulate"
22:01karolherbst: especially this f16 inout anbd f32 output thing sounds interesting
22:03RSpliet: karolherbst: it's an answer to Google's TPU
22:06RSpliet: (see https://drive.google.com/file/d/0Bx4hafXDDq2EMzRNcy1vSUxtcEk/view )