00:01 mupuf: yep
00:52 karolherbst: I wrote to virtual proramming if they know anything which is really important for getting a good performance :)
00:52 karolherbst: I know it might be that I don't get any answers at all, but maybe they can provide us with an information like: "We try to use this extension, but the fallback is really slow"
00:55 mupuf: yeah, you can always ask :D
00:55 karolherbst: well
00:55 karolherbst: they helped me with my witcher 2 issue
00:55 karolherbst: the steam version should artifacts, but the gog not
00:55 karolherbst: so they gave me access to those files
00:55 karolherbst: ...
00:56 karolherbst: but they didn't updated the steam stuff...
00:56 karolherbst: ...
00:56 karolherbst: I should poke them because of this
00:58 karolherbst: mupuf: well in the end we have to try
00:59 karolherbst: mupuf: somebody might want to help if we just ask
01:00 karolherbst: anyway, I will head to bed now
01:00 karolherbst: night
09:14 louisdk: I'm trying to ditch the closed sourced nvidia in favor of nouveau on my ion330 chipset based mediacentre, however I don't know if it supported yet or not. Running Ubuntu 16.04.
09:17 karolherbst: louisdk: is it a 9400m?
09:17 karolherbst: MCP79 in lspci?
09:29 louisdk: yup
09:30 louisdk: karolherbst, it is
09:30 karolherbst: louisdk: then I would say it should work
09:34 louisdk: karolherbst, my max screen 640x480 and no sound via hdmi. Using the nvidia driver my hd-ready tv is detected as 1377x768.
09:34 karolherbst: louisdk: mhhh odd. could you provide me with your dmesg and Xorg.log?
09:35 karolherbst: maybe something just went wrong in the transition
09:39 karolherbst: louisdk: please just pastebin both on some site (fully)
09:42 louisdk: dmesg: http://pastebin.com/mdPZx08Q
09:42 karolherbst: louisdk: yeah, as I thought, nouveau isn't loaded
09:42 karolherbst: mhh
09:42 karolherbst: k
09:43 karolherbst: louisdk: next the output of this: find /etc/modprobe.d/ -exec grep nouveau {} +
09:43 karolherbst: there may be some "blacklist nouveau" entries and maybe even "alias nouveau off"
09:44 louisdk: and the xorg log: http://pastebin.com/SncLarWM
09:45 louisdk: As i see the installer disables files is missing an # in the beginning: http://pastebin.com/uBZH5Vf0
09:46 karolherbst: blacklist nouveau
09:46 karolherbst: yeah
09:46 karolherbst: /etc/modprobe.d/nvidia-installer-disable-nouveau.conf
09:46 karolherbst: this file has to go
09:46 karolherbst: or just add a #
09:46 karolherbst: and then regenerate initramfs
09:46 karolherbst: and reboot
09:47 karolherbst: "options nouveau modeset=0" this is also ciritcal
09:47 karolherbst: best is if you just add a # in each line there :D
09:50 louisdk: Scumbag nvidia. Just have to disable nouvaeu at all cost. Doing it in a single file is not enough for them :P
09:54 louisdk: karolherbst, success. Thank you :)
09:56 karolherbst: :D
09:56 karolherbst: louisdk: so you even get audio over hdmi?
09:56 karolherbst: but I assume the boot resolution was much higher already
09:59 louisdk: It's to high to be precise. The system thinks I have a 1080p screen which I don't, so I'll have to do some xorg hacking ;)
10:00 karolherbst: :D
10:00 karolherbst: well usually it is enough to just set the resolution in your desktop manager
10:00 karolherbst: but mhh
10:00 karolherbst: usually it shouldnt think so
10:00 karolherbst: you could open a bug on the freedesktop bug tracker for this
10:02 louisdk: I'll look into this :)
10:36 louisdk: karolherbst, it doesn't like my chipset supports vdpau with the nouveau driver: http://pastebin.com/D9wAjcm0 :/
10:36 karolherbst: louisdk: let me check
10:36 karolherbst: louisdk: okay, I think you just need to install the nvidia firmware for video decoding
10:36 karolherbst: louisdk: do you have any firmware load errors in dmesg?
10:40 karolherbst: louisdk: https://nouveau.freedesktop.org/wiki/VideoAcceleration/
10:40 karolherbst: "Firmware" section
10:40 karolherbst: there is a shell script
10:41 karolherbst: you need to execute it line by line
10:43 karolherbst: louisdk: sorry that the process is that messy, but we can't just violate the license...
10:55 louisdk: I'll try to install the firmware and see how it goes :)
11:01 karolherbst: hoho
11:02 karolherbst: started my plasmashell via DRI_PRIME on my nvidia gpu
11:04 karolherbst: imirkin_: any idea why kwin doesn't find GLX_EXT_texture_from_pixmap via prime offloading?
11:05 louisdk: karolherbst, installed the firmware. Same output running vdpauinfo: http://pastebin.com/4waGhHWq
11:06 karolherbst: louisdk: dmesg
11:07 louisdk: karolherbst, http://pastebin.com/DYSmkZVr
11:07 karolherbst: odd
11:07 karolherbst: ohh idea
11:08 karolherbst: louisdk: do you have mesa-vdpau-drivers installed?
11:09 karolherbst: and if so, check that there is a libvdpau_nouveau.so file inside /usr/lib/x86_64-linux-gnu/vdpau/
11:09 louisdk: karolherbst, yes
11:09 louisdk: ok. ill do
11:10 louisdk: karolherbst, yes there is.
11:10 karolherbst: mhh okay
11:10 karolherbst: so this should be fine
11:10 louisdk: Should I try to extract firmware from a newer release like 3.40?
11:11 karolherbst: I have to verify that it is really supported
11:11 louisdk: karolherbst, thank you.
11:11 karolherbst: odd, VP3 should be supported
11:15 karolherbst: louisdk: what kernel are you running?
11:15 karolherbst: ohh wait
11:15 karolherbst: 4.4
11:16 karolherbst: louisdk: VDPAU_DRIVER="nouveau" vdpauinfo
11:16 karolherbst: does this change anything?
11:20 louisdk: karolherbst, doesn't seem to change anything: http://pastebin.com/WRhJt31w
11:20 karolherbst: then I am out of ideas sadly
11:20 karolherbst: I am sure somebody else can help you here
11:20 karolherbst: I don't use vdpau with nouveau anyway, because I have an optimus system
11:22 louisdk: karolherbst, thanks for your help so far :)
11:59 karolherbst: imirkin_: does this tells you anything why eon games might be slow with nouveau? https://github.com/virtual-programming/witcher2-linux/issues/153#issuecomment-210793972
12:18 Celti: If anybody can figure out who fixed the issues I was having in bug 82152, I will totally buy them a beer!
12:19 karolherbst: Celti: let me check :D
12:19 Celti: Hadn't touched it in a year and ten days, decided to try again today, and *bam*, PRIME is working!
12:19 karolherbst: Celti: this was somebody :)
12:30 karolherbst: Celti: those messages are normal by the way
12:30 Celti: They are?
12:30 karolherbst: this is just nouveau telling you that the card got suspended
12:30 karolherbst: and the ACPI message appears like on every system I know off
12:30 Celti: Yes, but the MMIO write fault...
12:30 karolherbst: just board engineers misimplemented some ACPI stuff
12:31 karolherbst: mhh
12:31 Celti: Good to know that the ACPI message isn't a problem
12:31 karolherbst: they aren't as important
12:31 karolherbst: could be a delayed write or read or something
12:31 karolherbst: I wouldn't worry much about it except you encounter issues
12:36 hakzsam: imirkin_, yeah okay, I'll change it to g[] instead
12:40 karolherbst: hakzsam: did you read my comment about compute_shaders?
12:40 hakzsam: no
12:40 karolherbst: seems like there is a lot of garbage talk on the internet
12:40 karolherbst: saints row doesn't use compute shaders
12:40 karolherbst: nouveau just sucks...
12:40 hakzsam: are you sure about that?
12:40 karolherbst: yes
12:41 karolherbst: I traced nvidia
12:41 karolherbst: no compute shaders
12:41 karolherbst: but
12:41 karolherbst: I got a little comment out of a eon dev about a possible bottleneck: https://github.com/virtual-programming/witcher2-linux/issues/153#issuecomment-210793972
12:41 karolherbst: maybe
12:41 karolherbst: this could be it
12:41 hakzsam: well, they probably use compute shaders in a few number of scenes
12:41 hakzsam: and you won't trace all the game :)
12:42 karolherbst: yeah
12:42 karolherbst: but the scene I traced I got 15% perf of nvidia
12:42 karolherbst: ...
12:43 hakzsam: not cool
12:43 karolherbst: right
12:43 karolherbst: but one part of this comment interessts me
12:43 karolherbst: "There is a fallback code to use when permanent mapping isn't available, but it isn't well maintained, mostly because we expect that without the ability to use permanent mapping to avoid having to pass buffer maps and updates to our OpenGL background thread the fps would go down by a factor of 10."
12:44 hakzsam: yeah
12:44 karolherbst: so is this the case in nouveau or not?
12:45 hakzsam: no clue, imirkin_ ^
12:52 hakzsam: karolherbst, btw, if you want to find scenes where compute shaders are used, just add an assert in nvc0_launch_grid() and wait for a crash :)
12:52 karolherbst: well I care more about the bad performance currently :D
12:52 karolherbst: hakzsam: but ST_DUMP_SHADERS should dump those, right?
12:53 hakzsam: yeah,I think so
12:54 hakzsam: karolherbst, maybe, you can make a trace with apitrace and share it?
12:54 karolherbst: hakzsam: of what?
12:54 hakzsam: saint rows
12:54 hakzsam: because I don't have this game
12:54 karolherbst: ahh k
12:54 karolherbst: I have a 6GB one
12:58 hakzsam: that's a bit huge :)
13:02 rardiol: what is glxinfo supposed to be like for an updated system on nouveau? I have "OpenGL renderer string: Gallium 0.4 on NVA8 and OpenGL version string: 2.1 Mesa 11.1.2", shouldn't OpenGL be at 4 or something?
13:04 karolherbst: hakzsam: :)
13:04 karolherbst: hakzsam: well I didn't compress it yet
13:04 karolherbst: rardiol: did you compile mesa yourself?
13:04 rardiol: karolherbst: no, i'm on nixos
13:05 karolherbst: rardiol: odd, because it seems that texture-float is disabled
13:05 karolherbst: rardiol: ohh in the nix file optional enableTextureFloats
13:05 karolherbst: https://github.com/NixOS/nixpkgs/blob/master/pkgs/development/libraries/mesa/default.nix
13:06 karolherbst: seems to be disabled by default
13:07 rardiol: huh, would that cause awful perfomance in a game?
13:07 karolherbst: what would cause aweful performance?
13:08 karolherbst: without texture-float you don't get anything above 2.1
13:08 karolherbst: simply as that
13:12 Celti: the texture float patent is currently held by RPX, a sort of "anti-patent-troll" that buys up controversial patents with the promise of never asserting or litigating them
13:12 karolherbst: yeah well
13:12 karolherbst: well using it for yourself is still legal
13:13 karolherbst: but usually everybody enables texture-float by default now
13:13 karolherbst: except some crazy guys...
13:13 rardiol: I see. I'm asking because I'm trying to play a game I compiled ( freeorion) and it's very slow for some reason and I suspected it's because of being opengl2.1. is that possible?
13:13 karolherbst: mhh
13:13 karolherbst: maybe
13:13 karolherbst: maybe because of s3tc
13:14 karolherbst: rardiol: does "glxinfo | grep -i s3tc" show anything?
13:14 rardiol: hm, I did enable s3tc already
13:14 karolherbst: ahh okay
13:14 karolherbst: yeah, then most likely because you can't reclock your card
13:14 karolherbst: ohh wait
13:14 karolherbst: it is nva8
13:14 rardiol: karolherbst: GL_EXT_texture_compression_s3tc, GL_EXT_texture_cube_map, GL_OES_EGL_image, GL_OES_read_format, GL_S3_s3tc, "
13:14 karolherbst: you should be able I think
13:14 karolherbst: mhh GT218
13:14 karolherbst: I really don't know
13:14 Celti: FreeOrion has a minimum of OpenGL 2.0, so I don't think that's it.
13:15 karolherbst: I think it is the reclocking thing
13:15 rardiol: Celti: there are a few ifdefs involving texture floats. I'm no graphics programmer so I can't be sure, but I think it's used
13:16 rardiol: https://github.com/freeorion/freeorion/blob/master/GG/GG/glext.h#L927
13:17 karolherbst: well at least enabling it won't hurt
13:17 rardiol: karolherbst: I can play other games with better graphics than that, and freeorion is already being slow on the menu
13:17 karolherbst: ohh
13:17 karolherbst: yeah well, try out texture-floats then
13:22 karolherbst: hakzsam: 244MB compressed :)
13:22 hakzsam: upload it somewhere?
13:22 karolherbst: yeah
13:23 karolherbst: in 2 minutes
13:24 hakzsam: cool
13:25 hakzsam: I probably won't have a look right now because I'm writing code emission for surfaces instructions on gk208
13:25 hakzsam: but later today maybe :)
13:25 karolherbst: k :)
13:26 karolherbst: hakzsam: https://drive.google.com/open?id=0B78S7GSrzebId1pzOTQ4Y0FKRjg
13:26 hakzsam: thks
13:33 hakzsam: mupuf, reator is *terribly* slow today
13:33 karolherbst: network?
13:33 hakzsam: yes
13:33 karolherbst: slower than a few days ago?
13:33 karolherbst: not possible :D
13:33 hakzsam: right now? oh yes :)
13:33 karolherbst: I had like 2 second lags every 2 seconds for a minute
13:33 hakzsam: I was just looking at a MMT trace, won't use reator for now because it's really too slow
13:34 karolherbst: well
13:34 karolherbst: but yeah, it sucks...
13:34 hakzsam: yep
13:34 hakzsam: anyway, I can write code emission without testing ;)
13:34 karolherbst: :D
13:35 hakzsam: imirkin_, btw, some bits of IMADSP are unknown
13:35 hakzsam: I'm going to RE them
13:35 karolherbst: but we really should take a look why all eon based games suck so much on nouveau
13:35 karolherbst: even bioshock infinite is 70% slower
13:36 hakzsam: sure
13:36 Yoshimo: quite a margin
13:36 karolherbst: and I thought a 780 ti on 4K Very High 30fps is good
13:36 karolherbst: ..
13:36 karolherbst: but nvidia just pushed out 99fps
13:36 karolherbst: :D
13:37 karolherbst: Yoshimo: yeah, it sucks
13:37 karolherbst: well at least nouveau is faster running team fortress
13:38 Yoshimo: world of warcraft with maxxed settings on a 980 with nouveau sucks in a similiar way ;)
13:38 karolherbst: maxwell2 sucks for completly different reasons
13:38 karolherbst: Yoshimo: but did you reclock ? :D
13:38 Yoshimo: of course not, i wouldn't know how ;)
13:38 karolherbst: increasing the engines clock might increase performance by a bit
13:38 karolherbst: pstate file
13:38 karolherbst: like with any other card
13:39 Yoshimo: but 10fps with a medium game on a highend card is fun ;)
13:39 karolherbst: even furmark is faster with nouveau
13:39 karolherbst: but that's no surprise
13:42 karolherbst: basically nouveau sucks here: bioshock infinite(30%), DIRT (60% on 780 ti), valley (64%), gputest_triangle (61%)
13:42 karolherbst: allthough the eon stuff is the most important issue
13:42 karolherbst: being above 50% is good though
13:50 hakzsam: imirkin_, madsp on gk110 is totally wrong in envydis
14:46 tincman: Hey nouveau crowd. I had some questions on nouveau support for Tegra K1/gk20a. It seems most of the last bits needed are application support regarding render nodes, correct?
14:48 tincman: I know gnurou has some patched versions of weston/xf86-video-modesetting that implement this. But, I've also seen some gk20a commits appearing in xf86-video-nouveau
14:48 karolherbst: tincman: well yeah somwhat. First you need working glamor acceleration in the X server to be able to use GLX I think
14:49 karolherbst: tincman: applications don't need to support it, because it is already handled inside mesa
14:49 karolherbst: it's like PRIME offloading in the end
14:49 karolherbst: but I might be wrong in the details
14:49 karolherbst: tincman: why are you asking?
14:51 tincman: I've been trying to get an upstream kernel running on my chromebook (Acer CB5-311), and trying to get GPU acceleration working as well. Another person and I have had some luck with various hacked/patched versions following gnurou's example
14:52 tincman: and more wondering what's the best way to continue, or what direction efforts are being focused
14:52 gnurou: tincman: you are correct, there are workarounds in my X/Wayland/glmark2 repositories to support render nodes (and tiled buffers for display), but these are just hacks - we need a more generic and sustainable solution but none of the attempts made so far seems to make everyone happy
14:54 gnurou: lynxeye sent another promising attempt last month ([PATCH RFC 0/2] GBM API extension to support fusing KMS and render devices), but it got stuck during discussion again...
14:55 karolherbst: always those discussions ...
14:55 tincman: hahaha, I was worried it was something like that....
14:56 gnurou: it's really sad, render nodes are around since a while already, and the model they implement (separating display and render functions) is sane even for devices that implement both functions
14:56 gnurou: I mean, there is nothing wrong with exporting a buffer and re-importing it into the same driver
14:56 karolherbst: render nodes are GPL only right?
14:57 gnurou: but there are other issues, like knowing which buffer formats are supported by both devices
14:58 gnurou: tincman: if you are interested in helping with this, maybe you can revive the discussion from Lucas' patch
15:01 tincman: Yeah, let me take a look at that thread :] Just the hint I was hoping to find.
15:03 gnurou: tincman: know that there is interest in finding a solution for this
15:06 tincman: gnurou: thanks, makes me optimistic :]
15:07 tincman: Oh, which reminds me. I had seen some commits in xf86-video-nouveau regarding gk20a. However, I had to patch this lightly to get proper buffer allocation, but it still failed trying to initialize a screen. My hope was this was an easier interim (loading nouveau and using it as a PRIME offloader...)
15:08 gnurou: tincman: for gk20a, isn't it just simpler to use modesetting + PRIME?
15:08 gnurou: tincman: I have patches to support render nodes and DRI2 (hack!) in my github tree
15:09 karolherbst: I always though there are two drm devices in the end one 2d device driven by modesetting and a 3d (gk20a) driven by... no ddx...
15:09 karolherbst: gnurou: why not go straight for dri3?
15:09 gnurou: karolherbst: mmm I think some libraries only supported DRI2 or something like that... DRI3 should work out-of-the-box, by design
15:17 tincman: haha, that might explain it. I was trying to force DRI3 and still had xorg reporting the gk20a as a "GPUdevice" which made me hope this was closer to working with minimal patching, but I suppose still needing DRI2 would have prevented this...
15:18 gnurou: ah maybe GLamor itself required DRI2 - that's why I had to enable it
15:18 tincman: gnurou; Yes, I think I recall that was one of your xorg-server patches that had me scratching my head
15:19 gnurou: this patch is horrible by the way, I don't even remember how I came with it
15:20 gnurou: probably a fine example of random programming
15:20 gnurou: (try random things until it works)
15:21 karolherbst: mhh
15:21 karolherbst: odd
15:22 karolherbst: I know that video acceleration and reverse PRIME needs DRI2 :/
15:43 imirkin: karolherbst: hakzsam: nouveau supports ARB_buffer_storage, which is what brings coherent / persistent mappings
15:43 karolherbst: okay, so that means something else is the issue
15:43 imirkin: karolherbst: GLX_EXT_texture_from_pixmap really should be provided, no idea why that'd be missing
15:43 imirkin: any other questions i missed?
15:46 karolherbst: don't think so
16:04 Daif: Anyone here?
16:04 Daif: Will someone reply me? LOL
16:07 karolherbst: lol...
16:07 karolherbst: the hell
16:08 karolherbst: uhh
16:08 karolherbst: the article is out
16:32 imirkin: nice
16:32 imirkin: now try to get skeggsb_ to take your patches :)
16:32 karolherbst: I hope we can land them for 4.7
16:32 karolherbst: but it will be 4.8 most likely
16:53 Celti: ...man. I tried enabling reclocking. It crashed my system so hard it, I kid you not, *forgot it had wifi*.
16:54 Celti: Rebooted, kernel totally didn't find a device to go with iwlwifi, and no amount of digging around in sysfs, rescanning the bus, or anything could make it realise yes, there is a wifi card installed.
16:54 karolherbst: cold reboot
16:55 Celti: Yep, I eventually got there
16:56 Celti: Was mind-boggling, though
16:56 karolherbst: I can imagine
16:56 karolherbst: do you have the dmesg from the crashing boot?
16:56 Celti: I ought to
16:57 Celti: journald is good for that, at least
16:57 karolherbst: :D
16:57 karolherbst: sure?
16:57 karolherbst: I had fun with it already, messed up the logs, because my kernel hard crashes
16:57 karolherbst: and like the last few minutes are completly gone
17:02 Celti: mine didn't actually hard crash immediately
17:03 Celti: I modprobed nouveau with config=NvClkMode, it hung for a bit and spewed a ton of errors but recovered
17:03 Celti: aaand then I tried running gloss with DRI_PRIME=1, because I'm stupid
17:03 Celti: but that didn't actually crash either, just went into uninterruptable sleep
17:04 Celti: after that I did a bit more fiddling and tried to reboot, and *then* it crashed
17:05 Celti: not so hard I couldn't REISUB my way clear,, though
17:06 Celti: but cutting it short, https://ilmatar.celti.name/~celti/esoterica/dmesg-nouveau-reclock-blah.txt has everything from boot to sysrq, with the NvClkMode modprobe happening at 09:02
17:07 Celti: and here's the one from the no-wifi boot, if that helps: https://ilmatar.celti.name/~celti/esoterica/dmesg-post-crash-no-wifi.txt
17:09 karolherbst: ahh infamous SCHED_ERROR
17:09 karolherbst: I think I know where the issue is, but know what exactly.
17:09 karolherbst: But I don't get this error so much so I really can't help debugging it
17:09 karolherbst: but I would if I could
17:24 Celti: and just because it's different, here's a crash from trying to reclock using the debugfs inteerface: https://ilmatar.celti.name/~celti/esoterica/dmesg-debugfs-pstate-change-crash.txt
17:25 karolherbst: ohh right
17:25 karolherbst: you shouldn't reclock while the GPU is off :/
17:26 Celti: no, no, I didn't
17:26 Celti: I ran `DRI_PRIME=1 gloss` off on another window before I tried to reclock
17:27 karolherbst: ohh okay
17:28 karolherbst: ahh
17:28 karolherbst: "[TTM] Buffer eviction failed"?
17:28 karolherbst: no idea what that means
17:28 karolherbst: but can't be good
19:30 Celti: karolherbst: If I was to try one of your nouveau branches to see if reclocking worked any better, should I build your linux/nouveau_4.5_clocking tree or try building one of the standalone nouveau branches?
19:30 karolherbst: Celti: you have a nva8 right?
19:31 karolherbst: oh
19:31 karolherbst: GK107
19:31 Celti: NVE7
19:31 karolherbst: well
19:31 karolherbst: you are fine with an out of tree module anyway
19:31 karolherbst: Celti: can you unload nouveau while X is running?
19:31 Celti: I can, yes
19:32 karolherbst: okay
19:32 karolherbst: this makes it easier
19:32 karolherbst: git clone https://github.com/karolherbst/nouveau.git -b stable_reclocking_kepler_v2
19:32 karolherbst: cd drm
19:32 karolherbst: make
19:33 karolherbst: this is my current development branch for that stuff and there is much more done than on the kernel tree
19:45 karolherbst: Celti: when you have any troubles with this nouveau module, please tell me
19:45 karolherbst: Celti: it is really bad tested yet, but this is the foundation of what might gets merged in 4.7 or 4.8
19:46 Celti: well right now I'm getting: fatal error: soc/tegra/fuse.h: No such file or directory
19:46 Celti: trying to build it
19:47 karolherbst: ohhh
19:47 karolherbst: let me guess, you run arch?
19:47 karolherbst: no idea why
19:47 karolherbst: but Arch is the only distribution shipping linux headers without those ARM headers...
19:47 karolherbst: or whatever
19:48 karolherbst: Celti: you can just remove this include
19:48 Celti: Indeed, I'm on Arch
19:48 Celti: and I think it's because they've got a bizzare rivalry going on with ArchLinux-ARM
19:48 karolherbst: maybe yes
19:48 karolherbst: :D
19:48 Celti: god forbid you go into #archlinux and tell them you're running AL-ARM!
19:48 karolherbst: ohh wait
19:49 karolherbst: I know
19:49 karolherbst: just tell them about your build issue :O
19:49 karolherbst: and say that soc/tegra/fuse.h is part of the kernel-headers stuff
19:49 Celti: heheh
19:49 karolherbst: and why it isn't there
19:49 karolherbst: seriously :D
19:49 karolherbst: ohh wait
19:49 karolherbst: it is part of the linux sources package
19:50 karolherbst: well
19:50 karolherbst: I try to talk sense into them
19:51 karolherbst: Celti: could you check if your linux source package contains this file?
19:52 karolherbst: Celti: pkgfile -v fuse.h
19:53 karolherbst: mupuf: taking care of this stupid arch issue now... :D
20:05 karolherbst: mupuf: https://projects.archlinux.org/svntogit/packages.git/tree/trunk/PKGBUILD?h=packages/linux#n166
20:05 karolherbst: it is this line
20:06 karolherbst: they just copy "some" (32MB out of 33MB) from include into the lib/modules/$kernel/build/include location
20:10 karolherbst: pmoreau: weren't you also using Arch?
20:10 pmoreau: I’m still
20:11 karolherbst: I found out why the compiles fails
20:11 karolherbst: ...*compile
20:11 pmoreau: What compile?
20:11 karolherbst: nouveau out of tree
20:11 karolherbst: the "fatal error: soc/tegra/fuse.h: No such file or directory" error
20:11 pmoreau: Never been a problem for me that I can recall of
20:11 karolherbst: ever build nouveau out of tree with arch provided kernels?
20:12 pmoreau: Oh, no. Always used drm-next from airlied
20:12 karolherbst: right
20:13 karolherbst: but basically they try to outsmart themselves
20:13 karolherbst: and they only copy a selection of header files from include into build/
20:14 karolherbst: Celti: anyway, there ya go :D
20:14 Celti: Yep
20:20 Celti: Well, it doesn't make things hang anymore... but unigine-valley froze then segfaulted when I tried to switch the pstate.
20:25 karolherbst: mhh
20:25 karolherbst: how did you load nouveau?
20:27 karolherbst: also I would like to see the dmesg in that case
20:30 karolherbst: the heck?!?
20:30 karolherbst: imirkin, hakzsam: I get better performance in saints row 3 on 0a than on 0f...
20:31 imirkin: i guess we're not clocking something right and it ends up tripping over itself?
20:31 Misanthropos: progress! ;D
20:31 karolherbst: imirkin: or something memory related is messed up
20:32 imirkin: that'd be the "not clocking something right" bit :p
20:32 karolherbst: imirkin: with faster I meant like a tiny bit, and my engine clocks are the same on 0a and 0f
20:32 karolherbst: just memory changes
20:32 karolherbst: but somebody on the phoronix forums also said that after 15 minutes it performed better
20:45 karolherbst: Celti: so, anything in dmesg?
21:04 karolherbst: imirkin: so there is just no memory load
21:04 karolherbst: no idea why
21:06 karolherbst: huh
21:06 karolherbst: saints row 2 needs libvpx.so.1 ...
21:06 karolherbst: that's like ancient
21:17 karolherbst: imirkin: saints row 2: https://i.imgur.com/bC99ko5.png
21:17 karolherbst: https://i.imgur.com/nzwXwVO.png
21:24 imirkin: i like it
21:24 imirkin: i'm gonna go ahead and guess it's some stupid coherency thing i fail at
21:27 karolherbst: maybe
21:27 karolherbst: I am sure it would be worth to investigate it
21:31 karolherbst: imirkin: in the first screenshot the texture in the bottem left corner also rotates with the mini map, but I assume you may already thought this
21:31 Celti: karolherbst: sorry, I managed to completely lock myself out of my system by accident, still cleaning up the mess from that
21:31 karolherbst: uhh
21:32 Celti: Secure Boot plus upgrading things plus /boot being on a USB stick.
21:32 karolherbst: ...
21:32 karolherbst: how did you load the new nouveau library?
21:32 karolherbst: and was there a boost file?
21:32 Celti: There was a boost file, I loaded it by dumping it in /lib/modules/blah/updates/ and modprobing.
21:33 Celti: I'll get a dmesg for you in a moment
21:33 karolherbst: ah okay
21:33 karolherbst: I am sure we have issues inside the PR firmware
21:33 karolherbst: like 90% sure
21:53 karolherbst: at least saints row 2 runs on intel without issues
21:53 karolherbst: imirkin: want an apitrace? it already shows in the intro
21:54 imirkin: not particularly
21:54 imirkin: i guess someone should investigate
21:54 imirkin: feel free to
21:55 karolherbst: I am sure I will fail, but well maybe it is something obvious
21:55 imirkin: karolherbst: step 1: NV50_PROG_OPTIMIZE=0
21:56 karolherbst: right
21:56 imirkin: i don't think we have a lot of compiler errors left. a few, related to RA around integer divisions...
21:56 karolherbst: mhh
21:56 karolherbst: does this work on release builds?
21:56 imirkin: no
21:57 karolherbst: good to know
21:57 imirkin: if you send me a trace, i guess i could see if it happens on GF108
21:59 Celti: karolherbst: https://ilmatar.celti.name/~celti/esoterica/dmesg-nouveau-boost-pstate-crash.txt
21:59 karolherbst: :O
21:59 karolherbst: MESA_EXTENSION_OVERRIDE=-GL_ARB_buffer_storage fixes it?
21:59 karolherbst: ...
22:00 karolherbst: imirkin: there ya go :D
22:00 imirkin: was that a question or an answer?
22:01 karolherbst: I was questioning myself
22:01 karolherbst: mhhh
22:01 karolherbst: this will be troublesome to apitrace then
22:02 karolherbst: yep
22:02 karolherbst: MESA_EXTENSION_OVERRIDE=-GL_ARB_buffer_storage fixes it
22:03 imirkin: super. so it's some coherency thing.
22:05 karolherbst: and witcher 2 just crashes
22:05 karolherbst: :/
22:08 Celti: karolherbst: and here's one more dmesg log, this time with the bleeding-edge module and attempting to load it with config=NvClkMode=10
22:08 Celti: https://ilmatar.celti.name/~celti/esoterica/dmesg-nouveau-boost-nvclkmode-crash.txt
22:09 karolherbst: uhhh
22:09 karolherbst: this mc error
22:10 karolherbst: well
22:10 karolherbst: for some reasons the nvidia GPU doesn't like us anymore
22:10 karolherbst: and just stops doing anything
22:10 karolherbst: no idea why
22:11 Celti: The nvidia GPU in this laptop has never liked anyone, ever
22:12 Celti: It's even buggy using the binary blob as the main graphics driver in discrete-only mode
22:13 karolherbst: Celti: well we could do something
22:13 karolherbst: Celti: /sys/kernel/debug/dri/1/vbios.rom
22:13 karolherbst: I would need file
22:14 karolherbst: let's just increase the voltage a bit and see what happens
22:14 karolherbst: but I have to check the vbios for something
22:28 Celti: karolherbst: https://ilmatar.celti.name/~celti/esoterica/lenovo-ideapad-z710-geforce-gt745m-vbios.rom
22:29 karolherbst: thanks
22:40 karolherbst: Celti: can you give me the content of pstate?
22:40 karolherbst: mhh
22:40 karolherbst: Celti: well the 0f line is enough
22:41 karolherbst: I wonder
22:41 karolherbst: ...
22:41 karolherbst: maybe we just don't increase the memory voltage
22:43 karolherbst: mupuf: uhhh which GPIOs are used to change the memory voltage? MEM_VREF and MEM_VOLTAGE ?
22:44 karolherbst: now I wonder
22:44 karolherbst: what happens if neither is there
22:45 karolherbst: but maybe that's not as important for ddr3
22:45 karolherbst: Celti: okay, let's try something
22:48 Celti: karolherbst: 0f: core 405-1045 MHz memory 2002 MHz
22:48 karolherbst: uhh
22:48 karolherbst: that's high
22:49 Celti: the other two are:
22:49 karolherbst: but okay, seemsright
22:49 Celti: 07: core 405 MHz memory 810 MHz
22:49 Celti: 0a: core 405-1045 MHz memory 1600 MHz
22:49 karolherbst: mhh
22:49 karolherbst: let's try 0a first
22:49 karolherbst: ohh wait
22:50 karolherbst: you tried that
22:50 Celti: I've tried both 0a and 0f with much the same results.
22:51 karolherbst: is this site also broken for you? https://nouveau.freedesktop.org/wiki/KernelModuleParameters/
22:52 Celti: No
22:52 karolherbst: ...
22:52 karolherbst: the hell
22:53 karolherbst: anyway there is an option to disable memory reclocking
22:53 Celti: config=NvMemExec=0 ?
22:54 karolherbst: right
22:58 Celti: pstate switching works flawlessly with that flag, though it doesn't seem to exceed 915Mhz even at 0f
22:58 karolherbst: normal
22:59 karolherbst: your max voltage is 1.18V
22:59 karolherbst: if you call sensors
22:59 karolherbst: you should see that nouveau set the voltage to either this or something a bit lower
22:59 karolherbst: the next higher clock would require a voltage above 1.18V
22:59 karolherbst: but
23:00 karolherbst: it might be, that if your card gets hotter, nouveau can clock up a little due to changed voltage requiernments :D
23:01 karolherbst: mhh
23:01 karolherbst: allthough
23:01 karolherbst: ahh
23:01 Celti: doing 'echo 2 > boost' lends credence to this idea
23:01 karolherbst: yeah, I forget that it defaults to 1
23:01 karolherbst: but for 2 the same applies
23:01 Celti: as when I do that it gives me the 'umable to find matching pll values' error
23:01 karolherbst: you should be still below your max?
23:02 karolherbst: again?
23:02 karolherbst: ...
23:03 karolherbst: Celti: next time load nouveau like this: debug=volt=debug config=NvMemExec=0,NvVoltOffsetmV+50
23:03 karolherbst: ..
23:03 karolherbst: debug=volt=debug config=NvMemExec=0,NvVoltOffsetmV=+50
23:06 imirkin: karolherbst: fyi 400/500 are fermi... 200/300 are still tesla
23:06 karolherbst: ohhh :O
23:06 karolherbst: right
23:06 imirkin: GTX 260 is G200 iirc
23:07 karolherbst: yeah, you are right
23:15 karolherbst: meh :/
23:15 karolherbst: I should debug runpm issues now then
23:24 karolherbst: Celti: I think I never tested my newest reclocking code while runpm was enabled...
23:24 karolherbst: there could be still some issues left
23:37 Celti: karolherbst: https://ilmatar.celti.name/~celti/esoterica/dmesg-volt-debug-offset-nomemreclock.txt
23:37 Celti: Didn't get any pll errors, but it still crashed.
23:38 karolherbst: mhh
23:38 karolherbst: not that hard though
23:38 karolherbst: k
23:39 karolherbst: then set the volt offset to 100 instead of 50
23:39 karolherbst: Celti: from the GR errors you can usually recover
23:39 karolherbst: by just killing the clients
23:40 karolherbst: and wait until it goes away and the nouveau use count drops to 0
23:40 karolherbst: then your rmmod nouveau and wait a few minutes
23:46 Celti: karolherbst: https://ilmatar.celti.name/~celti/esoterica/dmesg-moar-volts.txt
23:46 Celti: more crashes
23:47 Celti: and unfortunately for some reason the clients *don't die*
23:47 Celti: I try and kill then and then all of xorg and everything locks up
23:47 karolherbst: it takes a while
23:47 karolherbst: ohhh
23:47 karolherbst: ohhhh
23:47 karolherbst: right
23:47 karolherbst: because Xorg grabs the gpu for you
23:47 karolherbst: Celti: https://gist.github.com/karolherbst/1f1bdd1a3822df74097f
23:48 karolherbst: this is what I use
23:48 Celti: hmm!
23:48 Celti: I'll give that a shot, then
23:49 karolherbst: well I develop with this xorg.conf applied and be able to unload nouveau usually
23:49 karolherbst: sometimes my kernel crashes while unloading though, but that has different reasons
23:49 karolherbst: mhh
23:49 karolherbst: something is funny though
23:51 karolherbst: Celti: try 150 next time
23:51 karolherbst: but more won't make sense