00:00 Leftmost: Is PFETCH a TGSI instruction, or is it nvidia-specific? (Still trying to get a handle on the code layout.)
00:04 Leftmost: Ahh, well, it's not TGSI.
00:08 imirkin: it's codegen-specific
00:08 imirkin: i think it might be a real op on nv50 actually
00:10 Leftmost: Ahh, more info on it in nv50_ir.h.
00:20 Leftmost: Okay, so I think I have an understand of what's required crystallizing: we're getting an IR representation of the tess shader, translating into IR for the chip, and then uploading. For Maxwell, the means by which the uploaded shader code needs to get addresses for its inputs has changed, and so I would need to figure out how the binary driver is doing address calculation in those cases and change the GM107 code to translate the codegen
00:20 Leftmost: representation of address calculation into what GM107 needs.
00:20 Leftmost: Does that sound about right?
00:21 imirkin: yes
00:21 Leftmost: Alright, great. Now I have a high-level understanding of the task at hand. :)
00:21 imirkin: note that "IR" = "intermediate representation", so "IR representation" is a bit redundant
00:22 Leftmost: Ahh, yeah, slip of the tongue.
00:23 imirkin: ;)
00:34 Leftmost: How is the address calculation represented in the codegen IR? Should I just set it up to do a print pass after it translates from TGSI and run the tess shader tests through?
00:34 Leftmost: I suspect that'd be the most instructive way, if not the most expeditious.
00:35 imirkin: NV50_PROG_DEBUG=255 will print it a lot more often if you wanted
00:35 imirkin: you can grab a sample shader's TGSI representation
00:35 imirkin: and then feed it into the nouveau_compiler binary, supplying different targets
00:35 imirkin: to see what happens
00:36 Leftmost: Cool, thanks.
00:39 imirkin: i.e. nouveau_compiler -a c0 foo.tgsi
00:39 imirkin: -a e4 foo.tgsi
00:39 imirkin: those are the two interestingly different ones
01:29 tajjada: my attempts at extracting nvidia firmware failed miserably; i give up :)
07:50 Tom^: karolherbst: nvidia has a patch for 4.6
07:51 Tom^: karolherbst: https://git.archlinux.org/svntogit/packages.git/plain/trunk/linux-4.6.patch?h=packages/nvidia
07:55 karolherbst: Tom^: ohh thanks
11:43 pmoreau: karolherbst: I rebased your stable_reclocking_kepler_v5 on top of Ben’s master, and reclocking works pretty fine.
11:43 karolherbst: pmoreau: ahh good
11:44 karolherbst: pmoreau: I rebased my stuff on 4.6 today anway
11:44 pmoreau: And no stability issue on 0f either, over an almost 10h period. (the only issue I had while playing was running out of VRAM… nothing you can really do about :-D)
11:44 pmoreau: Ok
14:14 l1k: karolherbst: your domain is no longer delegated, one can no longer send you e-mails
14:16 l1k: <nouveau@karolherbst.de>: Host or domain name not found. Name service error for name=karolherbst.de type=A: Host found but no data record of requested type
14:16 l1k: dig +short ns karolherbst.de @a.nic.de
14:16 l1k: => nothing
14:21 karolherbst: l1k: uhh, I thought I had fixed that some time ago. Let me check again
14:23 hakzsam: karolherbst, btw, I tracked down the UE4 rendering issues, there is a flushing issue which needs to be fixed too
14:23 karolherbst: hakzsam: ahh, good
14:34 aereaux: hello
14:35 aereaux: is anyone able to help with debugging reclocking a GK107M card?
14:37 hakzsam: karolherbst, I'm downloading all UE4 demos to check if I can pinpoint other issues
14:42 hakzsam: mmh, the effects cave demo is seriously buggy (even with our fixes)
14:43 imirkin: aereaux: step 1: grab karolherbst's branch
14:43 aereaux: on github under karolherbst/nouveau?
14:44 imirkin: aereaux: https://github.com/karolherbst/nouveau/commits/stable_reclocking_kepler_v5
14:44 imirkin: probably that one
14:44 imirkin: he has a number of fixes for a variety of issues with the current reclocking and surrounding logic
14:44 imirkin: (like voltage, etc)
14:44 hakzsam: karolherbst, time to check a little thing for me?
14:45 karolherbst: hakzsam: yeah
14:45 karolherbst: aereaux: it is based on 4.6 by the way
14:46 aereaux: yeah, I'm currently on 4.6
14:46 hakzsam: karolherbst, update mesa master, apply this patch http://hastebin.com/fineqapeqa on top, and run bin/arb_shader_image_load_store-invalid (from piglit)
14:47 aereaux: where to make?
14:48 imirkin: aereaux: drm
14:50 karolherbst: is there a good way to tell patch to not create those stupid orig files?
14:50 hakzsam: git am?
14:51 karolherbst: mhh yeah well, right
14:51 karolherbst: but sometimes I can't do git am
14:51 hakzsam: yeah, sometimes you need patch -p1 <
14:51 karolherbst: no, patch creates orig files when there is an offset mismatch
14:51 karolherbst: I know, it sounds stupid, but it does
14:52 karolherbst: or something like that
14:53 karolherbst: uhh
14:53 karolherbst: hakzsam: your patch file is missformated
14:53 hakzsam: uhu?
14:53 karolherbst: the last line is missing
14:54 hakzsam: weird
14:54 hakzsam: sec
14:54 karolherbst: well, it still applied
14:54 karolherbst: but patch was complaining
14:54 hakzsam: oh okay
14:54 aereaux: it says no rule to make target '$thedirimbuildingin/nouveau/drm/nouveau/nvif/object.o', needed by '$thedirimbuildingin/nouveau/drm/nouveau/nouveau.o'. Stop.
14:55 karolherbst: aereaux: odd
14:55 imirkin: aereaux: you should be building in foo/nouveau/drm
14:55 imirkin: is that where you're building?
14:55 aereaux: yeah
14:56 imirkin: odd indeed then.
14:56 imirkin: oh, there's probably an error higher up
14:56 karolherbst: aereaux: can you run make clean
14:56 karolherbst: aereaux: and then give me the entire output of make?
15:00 karolherbst: hakzsam: some tests are failing
15:00 aereaux: make -C /lib/modules/4.6.0-mainline/build M=/home/aidane/Downloads/nouveau/drm/nouveau KCPPFLAGS="-Iinclude/uapi/drm" modulesmake[1]: Entering directory '/usr/lib/modules/4.6.0-mainline/build' make[2]: *** No rule to make target '/home/aidane/Downloads/nouveau/drm/nouveau/nvif/object.o', needed by '/home/aidane/Downloads/nouveau/drm/nouveau/nouveau.o'. Stop. Makefile:1428: recipe for target '_module_/home/aidane/Downloads/nouveau/drm/n
15:00 hakzsam: karolherbst, &> log | grep "fail" log ?
15:00 hakzsam: to see which ones fail
15:00 karolherbst: hakzsam: https://gist.github.com/karolherbst/dcbeabc46dfd15368e768dfc3509285b
15:01 hakzsam: expected.
15:01 hakzsam: thanks for testing
15:01 karolherbst: aereaux: ahh I have an idea
15:01 imirkin: hakzsam: why do those fail? i thought you had limits in place, no?
15:01 imirkin: hakzsam: or are they broken for buffers?
15:02 hakzsam: it's broken for buffers yeah, I added a comment in the lowering pass about that
15:02 karolherbst: aereaux: make CONFIG_DRM_NOUVEAU=m
15:02 hakzsam: imirkin, sort of robustness thing
15:03 aereaux: Same output
15:08 karolherbst: aereaux: very odd
15:09 l1k: karolherbst: in the e-mail I tried to send you, I asked if you could test this branch: https://github.com/l1k/linux/commits/drm_runpm_fixes_v1
15:09 l1k: it fixes all runtime pm ref leaks I could find
15:10 l1k: you had complained a few months ago that runtime pm refs are leaked by nouveau and others
15:10 aereaux: http://www.jmad.org/tmp/test.log if you want proper formatting.
15:10 karolherbst: l1k: right, I will test it after I reboot again, thanks
15:11 l1k: karolherbst: thanks. the branch is based on drm-next but it should be possible to rebase on anything recent, so you can just cherry-pick my patches on your working branch and it should work
15:11 karolherbst: l1k: I tried to fix it myself once, but I encountered a sudden increase in kernel crashes on unload
15:13 l1k: karolherbst: the only thing that doesn't work yet is switching on muxed laptops, such as the MacBook Pro. a ref needs to be taken when switching to the discrete GPU and released when switching to the integrated GPU. I'll fix this in a separate series.
15:14 karolherbst: ahh yeah, makes sense
15:14 imirkin: aereaux: very odd... not sure what's going on in your setup to cause that =/
15:14 imirkin: aereaux: do you not have a .config maybe?
15:15 imirkin: aereaux: or does /lib/modules/4.6.0-mainline/build point somewhere it shouldn't?
15:16 aereaux: Where would I need a .config?
15:17 imirkin: aereaux: in the build dir... for linux
15:17 imirkin: otherwise the make system gets very upset
15:17 imirkin: or if the makefile there is messed up
15:17 aereaux: which build dir?
15:17 imirkin: the one with the linux source
15:17 imirkin: i.e. /lib/modules/4.6.0-mainline/build
15:19 aereaux: There's a config, the build dir looks fine to me. Anything I should look out for?
15:19 aereaux: I could try resinstalling it, but it would take me about 30mins because I'd have to recompile
15:20 imirkin: GAH! gentoo is finally defeating my clever plan to never upgrade past gdm-2.x :(
15:20 imirkin: "<gnome-base/gdm-3.8" is blocking gnome-base/gsettings-desktop-schemas-3.18.1
15:20 imirkin: grrrr
15:20 mlankhorst: hard to resist change ;)
15:21 imirkin: but i really like gdm :(
15:21 imirkin: ligthdm sucks
15:22 imirkin: and xdm... gr.
15:22 imirkin: why did they have to make gdm depend on stupid gnome-shell :(
15:23 orbea:just uses startx, works
15:25 imirkin: i don't like leaving a shell logged in as me
15:29 imirkin: ok. i guess icedtea-bin has to go. it depends on gnome-shell.
15:29 karolherbst: imirkin: well I am happy with sddm :D
15:30 karolherbst: imirkin: lol? icedtea depends on gnome-shell?
15:30 imirkin: not directly. but it wants something which wants something which wants something
15:30 karolherbst: mhh
15:30 karolherbst: use icedtea then
15:31 karolherbst: I don't have gnome-shell installed
15:31 imirkin: emerge -C seems easier
15:31 karolherbst: well
15:31 karolherbst: depends on if you need java or not
15:32 karolherbst: imirkin: or use slim
16:03 pq: imirkin, I suppose 'exec startx' won't do?
16:35 karolherbst: aereaux: well I have no idea what might be wrong, I am sure it is something silly
16:36 hakzsam: mesa master now exposes GL 4.2 on Fermi!
16:37 aereaux: karolherbst: Yeah, I tried reinstalling the kernel, and that didn't fix it. Anything else I should try?
16:37 karolherbst: hakzsam: awesome :)
16:37 karolherbst: aereaux: well, in the end you can just copy the nouveau sources into your kernel
16:37 hakzsam: karolherbst, well, now we have to fix the rendering issues with UE4 before bumping to GL 4.3
16:38 karolherbst: aereaux: replace drivers/gpu/drm/nouveau by nouveau/drm/nouveau
16:38 karolherbst: aereaux: and just rebuild your kernel
16:38 karolherbst: hakzsam: yep
16:44 aereaux: karolherbst: Just to make sure, the stable_reclocking_kepler_v5 branch right?
16:58 hakzsam: imirkin, the flushing issue seems to be related to constbufs, at least on fermi
17:04 aereaux: OK, so I tried the nouveau dir in my kernel tree, and then recompiling, and it installed fine.
17:04 aereaux: however, when I try echoing to pstate it still hangs the computer.
17:14 imirkin: hakzsam: huh ok
17:14 imirkin: hakzsam: can you always upload constbuf stuff via 3d?
17:15 hakzsam: no clue
17:15 hakzsam: I'll check
17:15 hakzsam: and try to track down what's the issue exactly
17:16 imirkin: i.e. right now you alternate between 3d and cp to upload constbufs
17:16 imirkin: i wonder if that causes trouble
17:17 hakzsam: yeah, maybe
17:29 imirkin: aereaux: are you positive you loaded the updated module?
17:29 aereaux: any way I can check?
17:30 imirkin: mmmmmm... i think there should be a "boost" file next to pstate if you loaded karol's module
17:30 karolherbst: aereaux: yeah
17:30 karolherbst: imirkin: right
17:31 aereaux: there is a boost file
17:31 imirkin: aereaux: congratulations on having a GPU that needs something we don't fully support :)
17:31 karolherbst: yay
17:31 karolherbst: maybe again the firmware or something like that
17:31 karolherbst: aereaux: I would like to see your vbios then
17:31 karolherbst: /sys/kernel/debug/dri/0/vbios.rom
17:32 aereaux: just upload it somewhere?
17:32 karolherbst: yeah
17:32 karolherbst: and also dmesg from the crash
17:38 aereaux: OK, the vbios.ROM is at jmad.org/tmp/vbios.rom.
17:39 aereaux: how can I get the dmesg after it has crashed?
17:39 karolherbst: uhh
17:39 karolherbst: pstore maybe
17:39 karolherbst: /sys/fs/pstore/
17:39 karolherbst: depends on what happened though
17:39 karolherbst: maybe it is in your journal or sys logger
17:40 imirkin: aereaux: often it doesn't actually crash... just display stops working. you can ssh in/etc
17:40 imirkin: other times the PCI bus locks :)
17:41 imirkin: netconsole can work on occasion
17:41 aereaux: imirkin: shouldn't sysrq still work in that case? it didn't for me.
17:41 imirkin: ah yeah, usually
17:43 karolherbst: aereaux: try pstore, that's the place where I look first
17:43 karolherbst: usually it works on modern enough systems
17:44 karolherbst: the vbios is nice by the way
17:44 karolherbst: some engineer really spend time on that one :D
17:44 karolherbst: wut
17:44 karolherbst: 5 pstates?
17:44 karolherbst: well
17:44 karolherbst: 2 invalids
17:44 karolherbst: but still
17:45 karolherbst: no, the vbios is nice
17:46 Tom^: ive never understood why they ship vbioses filled with stuff that is invalid cruft.
17:46 karolherbst: Tom^: nvidia tools
17:46 karolherbst: they don't write it by hand
17:46 karolherbst: and they don't really bother cleaining up
17:46 karolherbst: *cleaning
17:49 imirkin: Leftmost: http://hastebin.com/mewoxacofi.avrasm - sanity / quads tgsi dumps for tess ctrl/eval shaders. enjoy.
17:52 aereaux: karolherbst: the computer totally freezes up, can't ssh or sysrq. nothing in pstore or journalctl about the crash either. any suggestions?
17:53 imirkin: aereaux: boot with nouveau.runpm=0 ?
17:53 karolherbst: aereaux: is pstore mounted?
17:53 imirkin: aereaux: you have an optimus setup right?
17:53 aereaux: how do I check if pstore is mounted?
17:53 karolherbst: mount
17:53 aereaux: imirkin: yeah
17:53 imirkin: aereaux: try nouveau.runpm=0 then.
17:54 karolherbst: mhhh
17:54 aereaux: karolherbst: it's mounted.
17:54 imirkin: this will prevent nouveau from suspending the gpu
17:54 karolherbst: but if it is an optimus setup, it shouldn't freeze
17:54 karolherbst: ohhh
17:54 karolherbst: I thought I also fixed the freeze issues while the gpu is suspended on my branch
17:54 karolherbst: odd
17:55 karolherbst: but maybe I messed up
17:58 karolherbst: yeah, most likely I messed up
17:58 aereaux: still freezes with runpm=0
17:59 karolherbst: aereaux: okay, well let's debug this properly
17:59 karolherbst: aereaux: stop X completly
17:59 karolherbst: aereaux: and reclock from tty
17:59 karolherbst: this shouldn't cause any freezes
17:59 karolherbst: ohh
17:59 karolherbst: aereaux: is your intel GPU enabled?
18:00 mfrw: hi, guys any lead on this, i cant suspend my computer when nouveau is loaded .. Earlier i used to but i dont know i have messed up somewhere
18:00 karolherbst: mfrw: dmesg
18:01 Leftmost: imirkin, great, thank you.
18:01 aereaux: karolherbst: in case it matters, I'm running Wayland with xwayland. my igpu is enabled.
18:01 mfrw: where should is put the output, i am new to irc :( _newbie_
18:02 karolherbst: aereaux: mhh
18:02 karolherbst: aereaux: more odd
18:02 karolherbst: aereaux: well, maybe the kernel crashes for some stupid reasons :/
18:02 mfrw: karolherbst: where should is put the output, i am new to irc :( _newbie_
18:03 karolherbst: aereaux: well, I would try to reclock from tty and see what happens
18:03 karolherbst: aereaux: without anything graphical running at all
18:03 InquisitiveEdgar: so
18:04 InquisitiveEdgar: linux 4.6 has maxwell support right?
18:04 InquisitiveEdgar: cause i enabled maxwell in the kernel but nothing works :S
18:04 imirkin: InquisitiveEdgar: GM20x starting 4.6 for acceleration, GM107 starting 4.1 for acceleration
18:04 InquisitiveEdgar: i use a gtx 960
18:04 imirkin: InquisitiveEdgar: in the GM20x case you need to make sure you have the nvidia-signed firmware installed
18:05 InquisitiveEdgar: i mean, not even X works
18:05 imirkin: InquisitiveEdgar: dmesg
18:05 InquisitiveEdgar: :\ hmm
18:05 InquisitiveEdgar: brb
18:06 mfrw: karolherbst: https://gist.github.com/mfrw/e44a0c451218d60c56b9a20dd58377b4
18:07 karolherbst: looks like acer_wmi messed up
18:07 karolherbst: mfrw: try to unload acer_wmi before suspending and see if that changes anything
18:09 mfrw: karolherbst: it was there when i was able to suspend, when i dont compile the kernel with nouveau, the suspend works good
18:09 karolherbst: mfrw: was that dmesg after you tried going into suspend?
18:10 mfrw: karolherbst: if i try to suspend , the my computer freezes after i try to wake it up
18:10 mfrw: karolherbst: no it was not
18:10 karolherbst: mfrw: well we would need the logs, because we don't know what goes wrong on your side
18:11 aereaux: from tty I get some messages, and then it freezes for a bit. then I can do stuff again and the pstate is back to the lowest one ( which I believe is off ). nouveau portions of the dmesg are at jmad.org/tmp/nouveau_dmesg.log
18:11 mfrw: karolherbst: is there any way to get logs, after the computer freezes, as when i suspend it freezes, and then i have to manually power it down and restart
18:13 karolherbst: mfrw: mhhh, the mc again :/
18:13 karolherbst: mfrw: do you try to clock to 0f or 0a?
18:13 mfrw: karolherbst: i have no idea what you mean :(
18:13 karolherbst: mfrw: sorry, I meant aereaux
18:14 aereaux: karolherbst: I tried to clock to 0f
18:14 karolherbst: mfrw: mhh ask your system logger, but chances are it doesn't log it. you can also try to boot with nouveau.runpm=0 but than your GPU is always on
18:14 mfrw: karolherbst: I dont know anything on the graphics side, just a little bit of kernel programming
18:14 karolherbst: aereaux: does 0a work?
18:15 mfrw: karolherbst: I dont actually want to use my nvidia card, my only problem is that, i want to conserve as much battery as i can, is there an option of switching the device off somehow ?
18:15 karolherbst: mfrw: well if you don't plan to use the GPU, you can also just install bbswitch
18:15 karolherbst: mfrw: it's from bumblebee
18:15 karolherbst: and tell it to turn the GPU off
18:16 karolherbst: mfrw: options bbswitch load_state=0
18:16 karolherbst: mfrw: in modprobe.d
18:16 mfrw: karolherbst: Thank you .. will try this :)
18:16 karolherbst: mfrw: but the kernel crash information would help us finding out why it crashes
18:16 karolherbst: because it shouldn't
18:17 mfrw: karolherbst: any option of getting the logs of previous failure on the next restart ? , i mean can we get previous system logs after restart
18:18 karolherbst: depends on if the HDD controller is still working and the kernel driver for it
18:18 karolherbst: usually it should be there
18:18 karolherbst: except something really bad happens
18:18 mfrw: karolherbst: sure will try something :)
18:18 mfrw: karolherbst: thanks anyways :)
18:19 karolherbst: aereaux: well the GPU doesn't like us after nouveau tries to reclock, I guess there might be something we still do wrong.
18:19 karolherbst: aereaux: but first you should try if the other pstates work (07 and 0a)
18:20 karolherbst: aereaux: anyway, it would be good if you would create an mmiotrace with the binary driver after you reclock a bit
18:20 Animeking: imirkin: bpaste.net/show/27ef65824d38
18:20 Animeking: i'm edgar bt w :v
18:20 Animeking: what do you think of that imirkin :|
18:22 aereaux: karolherbst: seems to be the same result with 0a. I've uploaded the two dmesgs to jmad.org/tmp/
18:24 karolherbst: aereaux: yeah, then just try out 07, but I think it will also crash
18:24 aereaux: karolherbst: and again the same result with 07.
18:25 karolherbst: aereaux: anyway, then I would like to have two things: 1. mmiotrace of nouveau while reclocking to 07 and a second one while reclocking to 0f
18:25 karolherbst: aereaux: 2. an mmiotrace of nvidia reclocking through nvidia-settings
18:25 aereaux: any guide on how to do that?
18:25 karolherbst: aereaux: https://wiki.ubuntu.com/X/MMIOTracing
18:26 karolherbst: aereaux: for nvidia you would have to start a second X server, but you can simply use bumblebee for this
18:28 imirkin: Animeking: that all seems perfectly happy... can you grab xorg log?
18:28 aereaux: karolherbst: OK, I'll do this and get back to you, then. I have to take care of a couple things first. thanks for the help.
18:29 imirkin: Animeking: do you really have a 1280x1024 screen connected to it, or is that a misdetection of sorts?
18:29 aereaux: also, I have not regained control over the tty after trying reclocking to 07.
18:29 Animeking: imirkin: bpaste.net/show/f07b02e2f85e
18:29 Animeking: imirkin: that is indeed correct
18:29 Animeking: my big monitor died ;-;
18:31 imirkin: Animeking: update your Xorg and/or install xf86-video-modesetting
18:31 Animeking: :|
18:32 Animeking: so I had a package missing
18:33 Animeking: alright, let's see if this works now
18:34 Animeking: it works :D
18:55 Animeking: OpenGL version string: 3.0 Mesa 11.2.1 <--- does this mean I am using opengl 3.0
18:55 Animeking: o.O
18:56 imirkin: Animeking: it means you have a GL 3.0 compat profile. chances are if you scroll up you'll see GL 3.3 core profile
18:56 Animeking: ah
18:56 imirkin: mesa never provides anything higher than GL 3.0 for compat
18:56 Animeking: Oh
18:57 Animeking: I see it now
18:57 imirkin: (and you don't get GL 4.1 like fermi + kepler do because *someone* was lazy and never finished tessellation on maxwell)
18:57 Animeking: >.>
18:58 Animeking: who is this someone so i can get my pitchforks v_v
18:58 Animeking: kidding
18:59 imirkin: you're looking at that someone...
18:59 imirkin: catch me if you can :p
18:59 Animeking: >_> I WILL EXTERMINATE YOU!
18:59 Animeking: So, when can we expect decent performance like say the boost patches in the older drivers?
18:59 Animeking: or older cards not older drivers
19:00 imirkin: never
19:00 Animeking: damn
19:00 imirkin: the hw is super-locked down by nvidia
19:00 Animeking: Jesus freakin' christ
19:00 imirkin: so it's basically whenever they feel like letting nouveau change clock speeds
19:00 imirkin: which i assume is "never"
19:00 Animeking: >.O
19:00 Animeking: Why would they do this
19:01 imirkin: it took up until kernel 4.6 for them to allow nouveau to provide any sort of acceleration at all
19:01 Animeking: doesn't this mean nouveau is kinda dead
19:01 imirkin: pretty much yeah
19:01 Animeking: in the future
19:01 Animeking: :[ augh
19:01 imirkin: which is also why i'm not lifting a finger for maxwell
19:01 Animeking: so not only do they have shitty wayland support
19:02 Animeking: shitty as in nvidia trying to play differently
19:02 Animeking: they also won't allow the cards to perform well :[
19:02 imirkin: buy amd.
19:02 imirkin: vote with your wallet
19:03 Animeking: You can't really vote with your wallet when they already rake in billions, and most of their buyers are windows and binary drivers (who actually like using the binaries) people .-. it's like, a teensy bleep in their radar
19:04 imirkin: maybe your wallet's just not big enough :p
19:05 Animeking: lol
19:06 Animeking: So, any clue why it took them so much to release the acceleration pieces?
19:08 Animeking: if i recall they said they would in like, 2013? 2014 and then only released them like recently o.O
19:08 karolherbst: Animeking: they don't make money doing it
19:08 Animeking: blargh
19:08 imirkin: because it only got released due to an internal 'champion' who went through all the internal BS of releasing it
19:08 karolherbst: kind of sad though
19:09 karolherbst: like if they really don't release any fimrwares anymore...
19:09 karolherbst: well, at least we did a decent job with kepler
19:09 Animeking: .-. I hope this internal 'champion' releases the acceleration stuff...
19:09 Animeking: why did they lock it down this hard
19:10 karolherbst: "security"
19:10 imirkin: because people were selling fake hardware
19:10 Animeking: i know they made the chinese claim but this just seems overkill
19:10 imirkin: security is a bs argument and everyone knows it
19:10 karolherbst: yeah
19:10 imirkin: but it's the official one
19:10 karolherbst: never heard about the fake hardware claim though
19:11 Riastradh: It's not *user* security. It's NVIDIA's market security.
19:11 Animeking: probably doesn't happen too much over here
19:11 karolherbst: Riastradh: not even this
19:11 karolherbst: well
19:11 karolherbst: but also phoronix doesn't get any pascal samples , allthough I have no clue how honest he was
19:12 karolherbst: ...
19:12 karolherbst: still a dick move from nvidia PR
19:26 imirkin: karolherbst: feel like tracing some stuff for me on nvidia?
19:26 imirkin: karolherbst: dEQP-GLES31.functional.tessellation.invariance.outer_edge_symmetry.triangles_equal_spacing_cw_point_mode and dEQP-GLES31.functional.shaders.builtin_functions.common.floatbitstouint.vec2_lowp_tess_eval
19:32 Yoshimo: karolherbst: as long as there is no reclocking for maxwell cards i don't give a damn about any pascal tests
19:43 karolherbst: Yoshimo: :D
19:43 karolherbst: imirkin: yeah, no problem
19:43 imirkin: i never double-checked calim's RE of how all those bits are set
19:43 imirkin: but it seems like sometimes we're doing it slightly wrong
19:43 imirkin: those are two of the cases
19:43 karolherbst: imirkin: in which folder was the binary again?
19:44 karolherbst: ohh wait
19:44 karolherbst: I have it
19:44 Yoshimo: i guess Nvidia didn't give any updates on the firmware situation when they announced pascal right?
19:45 imirkin: nothing publicly at least
19:45 karolherbst: imirkin: uhh, right, EGL :/
19:46 imirkin: you've done it all before, no?
19:46 karolherbst: imirkin: "Warning: Unable to create native OpenGL ES 3.1 context, will use wrapper context."
19:46 karolherbst: is this important?
19:46 karolherbst: imirkin: http://filebin.ca/2hx0mnXlMeMF/dEQP-GLES31.functional.tessellation.invariance.outer_edge_symmetry.triangles_equal_spacing_cw_point_mode.log.xz
19:46 karolherbst: does it look okay?
19:47 imirkin: karolherbst: yep
19:47 karolherbst: good
19:47 karolherbst: then I do the other one too
19:48 imirkin: i am, of course, thoroughly confused
19:48 imirkin: but ... that's another matter
19:48 karolherbst: imirkin: http://filebin.ca/2hx1N112QU8P/dEQP-GLES31.functional.shaders.builtin_functions.common.floatbitstouint.vec2_lowp_tess_eval.log.xz
19:48 aereaux: karolherbst: to mmiotrace nvidia's reclocking what do I do. is there something simple to do to get it to reclocking the card?
19:49 karolherbst: aereaux: nvidia-settings
19:49 karolherbst: aereaux: with bumblebee: optirun -b none nvidia-settings -c :8
19:49 karolherbst: and in powermizer you can set it to perf mode
19:49 karolherbst: then it does upclock
20:04 imirkin: karolherbst: thanks, i think i got it
20:04 imirkin: karolherbst: don't set winding when you're in points mode.
20:05 imirkin: let that be a lesson to you :)
20:05 karolherbst: what?
20:06 imirkin: heh
20:06 imirkin: don't worry about it. thanks for those traces :)
20:06 imirkin: i fixed at least one issue
20:06 karolherbst: :D
20:06 imirkin: looks like there's another one though =/
20:29 imirkin: karolherbst: how about dEQP-GLES31.functional.tessellation.primitive_discard.triangles_equal_spacing_ccw_point_mode ?
20:30 karolherbst: http://filebin.ca/2hxE3BFRT8Zz/dEQP-GLES31.functional.tessellation.primitive_discard.triangles_equal_spacing_ccw_point_mode.log.xz
20:31 imirkin: thanks
20:32 imirkin: karolherbst: does it pass btw?
20:33 karolherbst: yes
20:33 imirkin: :(
20:35 imirkin: i object!
20:36 karolherbst: still passes
20:37 imirkin: i still object.
20:37 karolherbst: maybe nvidia lies
20:37 karolherbst: no idea
20:37 karolherbst: maybe dEQP always passes with nvidia
20:37 karolherbst: who knows :p
20:37 imirkin: this is my current list of fails with deqp tess... http://hastebin.com/xiseyetoxu.m
20:38 karolherbst: sure you tested with dEQP-GLES31.functional.tessellation.primitive_discard.triangles_equal_spacing_ccw_point_mode
20:38 karolherbst: and not dEQP-GLES31.functional.tessellation.primitive_discard.triangles_equal_spacing_cw_point_mode?
20:38 imirkin: karolherbst: they both fail for me :)
20:39 karolherbst: well I know that dEQP-GLES31.functional.tessellation.primitive_discard.triangles_equal_spacing_ccw_point_mode passes on nvidia :p
20:39 imirkin: also cw vs ccw is the same in point mode
20:39 karolherbst: at least for me
20:39 karolherbst: ahh
20:42 imirkin: looks like primitive discard somehow doesn't work for point mode???
20:45 karolherbst: aereaux: any issues so far?
20:45 karolherbst: aereaux: ohh, you wanted to do other stuff, sorry for the interruption then :D
20:49 aereaux: karolherbst: having issues installing the nvidia drivers.
20:51 karolherbst: aereaux: usually you should install it through the bumblebee packages, otherwise it messes up your mesa for intel :/
20:52 karolherbst: it is a mess, because usually the nvidia pacakge brick mesa
20:52 karolherbst: *packages
20:52 aereaux: karolherbst: does it matter if I use the nvidia beta, it looks like its fixed there.
20:52 karolherbst: nope, doesn't matter
20:52 karolherbst: aereaux: you are on 4.6, right?
20:52 aereaux: I don't need the libgl packages, those are what would conflict, right?
20:52 aereaux: yeah
20:53 karolherbst: okay, good
21:26 aereaux: karolherbst: so now I'm having trouble mmiotracing nouveau. I get some INFO task ... blocked for more than 120 seconds.
21:54 karolherbst: aereaux: :/
21:54 karolherbst: aereaux: well since the beginning or after reclock?
21:54 aereaux: after reclock
21:57 aereaux: zsh gets stuck after the reclock and I can't even switch to another vt.
21:58 karolherbst: aereaux: ahh, no that is fine
21:58 karolherbst: aereaux: as long as you have the log
21:58 karolherbst: aereaux: did you load nouveau with runpm=0?
21:58 aereaux: I think the process writing the log gets stuck too, though.
21:59 karolherbst: yeah, the gpu crashed
21:59 aereaux: karolherbst: is it possible to modprobe it with those options?
21:59 karolherbst: aereaux: well, you can giv me the output of tail log
21:59 karolherbst: or just compress all logs and send them to me :D
22:00 karolherbst: if something is missing I will coplain
22:00 karolherbst: *complain
22:09 aereaux: karolherbst: OK, so I forced a restart, and it looks like the nouveau log got cut off. for the nvidia one, I opened nvidia-settings and made sure it changed the frequency at least a couple times.
22:09 karolherbst: aereaux: well, as long as the reclocking script is inside the nouveau log, I am happy
22:12 aereaux: karolherbst: what do you mean?
22:12 karolherbst: doesn't matter, I will just take a look and see if everything important is there
22:13 karolherbst: I hope it is a stupid simple issue
22:15 aereaux: karolherbst: OK, the two logs are at jmad.org/tmp/mmiot.tar.xz
22:17 karolherbst: aereaux: uhh, the nouveau one is quite short :/
22:18 karolherbst: actually there isn't anything usefull in it
22:19 aereaux: karolherbst: yeah, I think a lot of it got cut off. I don't know how to get the rest of it, because for some reason my kernel stops writing to disk when I try to reclock.
22:19 karolherbst: mhh
22:19 karolherbst: did you try to run sync?
22:20 aereaux: I can't. zsh doesn't give me another prompt, and I can't switch vts
22:21 karolherbst: :/
22:22 karolherbst: ssh?
22:22 karolherbst: I am sure you can just pipe it through scp
22:24 imirkin: urgh. have to use LINKED_TSC mode for > 16 textures, at least on fermi :(
22:29 imirkin: er no. i'm incompetent...
22:29 imirkin: hrmph. it really looks like it should work :(
22:39 aereaux: karolherbst: OK, slightly more success. I tries something slightly different, and uploaded the new nouveau mmiiotrace to jmad.org/tmp/mmiot_nouveau.tar.xz. there's also the script I used to get it, although it was interrupted part way through when I used the sysrq keys.
22:39 hussam: hello.
22:40 hussam: is there a trick to use reclocking my gt 630 card?
22:40 hussam: it is stuck at a low clock
22:40 karolherbst: hussam: which gt 630 do you have?
22:40 karolherbst: (aka what is shown in lspci)
22:41 karolherbst: aereaux: at least the file is big enough now
22:41 hussam: 01:00.0 VGA compatible controller: NVIDIA Corporation GF108 [GeForce GT 630] (rev a1)
22:41 hussam: a fermi card
22:41 karolherbst: hussam: k, that's fermi, so you are out of luck
22:41 hussam: ouchies
22:41 karolherbst: sadly memory reclocking doesn't work yet
22:42 karolherbst: well, depends on for what you actually need more perf though
22:42 hussam: even tvtime makes it laggy. on closed source driver, I could run two sdl games and tvtime with no lag what so ever
22:43 karolherbst: aereaux: okay, the faulty stuff is also in the trace
22:43 karolherbst: aereaux: I think that's good enough
22:43 hussam: No patch or anything I can edit? I am comfortable compiling kernels
22:43 karolherbst: hussam: mhh
22:43 karolherbst: hussam: there is something you could do
22:43 aereaux: karolherbst: OK, good luck figuring out what's wrong.
22:44 karolherbst: hussam: but it won't change much
22:44 karolherbst: hussam: maybe 25% more perf is possible currently
22:44 hussam: It is worth the try
22:45 karolherbst: hussam: https://github.com/karolherbst/nouveau/commit/2048d5d760691a85ac04ea2ee51068fc134e4ecb.patch and https://github.com/karolherbst/nouveau/commit/1ebc30581dcb6bbbed88dc09baf475e314706fcf.patch
22:45 karolherbst: hussam: apply inside drivers/gpu/
22:46 karolherbst: hussam: and the just use the pstate interface of nouveau
22:46 hussam: It was suggested in #gnome-shell to remove xf86-video-nouveau so it uses the plain mesa driver. I did that + installed mesa from git and gnome-shell CPU dropped from 30 to 2
22:46 imirkin: hussam: that just kills perf
22:46 karolherbst: hussam: ... so, does glxinfo still lists nouveau?
22:46 hussam: this card is identified as NVC1
22:47 hussam: ok, let me check that.
22:47 karolherbst: seriously... some just give stupid advices
22:47 karolherbst: at least I didn't encountered the nomodeset "tipp" anymore
22:48 imirkin: otoh i guess i don't use gnome-shell, so for all i know it managers to trigger some retardo in nouveau
22:48 hussam: OpenGL vendor string: nouveau
22:48 hussam: OpenGL renderer string: Gallium 0.4 on NVC1
22:48 karolherbst: ahh okay, so modesetting ddx was picked up
22:48 hussam: OpenGL core profile version string: 4.2 (Core Profile) Mesa 11.3.0-devel (git-7dce479)
22:49 hussam: Ok, I will compile my kernel with those patches.
22:49 karolherbst: funny: 115.676786 MMIO32 R 0x137100 0xbad0011f PCLOCK.SRC_SEL => { CLK0 = PLL | CLK1 = PLL | CLK2 = PLL | CLK7 = DIV | CLK8 = PLL | CLK9 = DIV | CLK12 = DIV | CLK14 = DIV | 0xbad00018 :D
22:50 karolherbst: uhhh. W 0x088068 0x000000ff PPCI.MSI_HEAD <= { CAP_ID = 0xff | NEXT_CAP_PTR = 0 | QMASK = 0 | QSIZE = 0 }
22:50 karolherbst: and then R 0x000100 0xbad0011a PMC.INTR_HOST
22:51 karolherbst: ohh
22:51 karolherbst: the card crashed there already
22:51 karolherbst: ahhh
22:51 karolherbst: right
22:51 karolherbst: PMU script -> gpu crash
23:07 hussam: karolherbst: it fails to patch at drm/nouveau/nvkm/subdev/clk/gk20a.c
23:07 hussam: maybe it is easy to patch manually. I will check
23:12 hussam: the actual line is ret = nvkm_clk_ctor(func, device, index, true, &clk->base);
23:12 hussam: line in patch is ret = nvkm_clk_ctor(&gk20a_clk, device, index, true, &clk->base);
23:13 hussam: this is kernel 4.6 by the way
23:22 hussam: I just added ", true" now let us see if it builds.
23:32 hussam: Most of this is actually because NVIDIA is currently butchering Fermi stability :d
23:50 karolherbst: hussam: well, I will head to bed now. Don't expect too much from those patches, but maybe the biggest bottleneck is solved with this, maybe not
23:51 karolherbst: and maybe it won't work at all
23:51 hussam: karolherbst: It's fine. thank you :)
23:51 hussam: if it does a slight bit, perhaps I can ask for a patch rebased on next kernel if possible?
23:52 hussam: Good night.
23:52 karolherbst: well, shouldn't matter with those patches
23:53 hussam: I already had to edit one line. so it may diverge further in the future.
23:54 karolherbst: well there is a much easier way to enable this, but well...
23:54 karolherbst: actually maybe it might make sense to really upstream this
23:54 karolherbst: even if it doesn't reclock memory, the increased engine clocks might help for a few workloads