06:02fling: AccelMethod none
06:02fling: NoAccel true
06:02fling: everything works with these options and I have no crashes
06:02fling: but client resizes are really slow
06:02fling: especially if I have mpv running with x11
06:02fling: how to fix?
06:06fling: imirkin: Hello.
07:45AndrewR: ... Seamonkey displaying 3d photo from facebook tend to hang my display reliable - yet so far I avoided turned acceleration off .....
09:38juri_: morning everyone. thanks for all the hacking.
11:33fling: AndrewR: I disabled it because rxvt-unicode hangs X for me after some time :D
11:34fling: AndrewR: and box crashes right away if I export mesa llvmpipe variable in .xinitrc
11:34fling: juri_: Hello! Happy hacking to you!
11:36juri_: thanks. :D
11:40fling: Is not it possible to keep the accel for the server and disable for clients?
11:40fling: Or any other filthy trick to speedup client resizes?
12:31AndrewR: fling, no, I only have rxvt, and not use it much .... Did you use EXA (as in xf86-video-nouveau) or glamor (x modesetting inside X server)? I think general recommendation is to try and use EXA because it causes less load on 3d part of GPU
12:34fling: AndrewR: but resizing clients is not 3d
12:34AndrewR: fling, if you use glamor (X acceleration architecture) it forwards all 2d work into OpenGL .....
12:35fling: I'm using xf86-video-nouveau
12:36AndrewR: fling, then sorry ..:( I mean I don't think I have any idea how to debug/combat this
12:37fling: can't I just somehow keep the accel for X and disable for everything else?
12:37fling: AndrewR: what would be equivalant for AccelMethod none and NoAccel true but for .bashrc instead of xorg.conf ?
13:42AndrewR: fling, I think if we talk about 2d accel you can't? (all this routed via X server)
14:28fling: I think I can
14:28fling: I will try running urxvt with llvmpipe
14:28fling: just need a script for this haha
15:04fling: I actually should somehow hack awesome into running everything with custom env
15:28fling: AndrewR: ok so I hacked awesome into setting LIBGL_ALWAYS_SOFTWARE to true and GALLIUM_DRIVER to llvmpipe to every client
15:28fling: AndrewR: but this is not enough right? I still have accel?
15:29fling: what other env variables should I set?
15:38AndrewR: fling, I think this disables 3d acceleration ... now wait and see if hang returns? (I'm not aware about more fine grained 2d accel control for EXA - you probably can disable Upload to Screen/Download from screen and Render accel ... and not much besides this. And this may add bugs :/ because few humans test those params)
15:40AndrewR: fling, https://www.x.org/releases/current/doc/man/man4/exa.4.xhtml
15:41fling: AndrewR: but performance is soo good with llvmpipe compared to noaccel in xorg.conf haha
15:41fling: I get no stuttering in opengl like superslicer
15:41fling: ok, will try to hang it!
15:43fling: AndrewR: are you saying there is no other env varibles to try?
15:44fling: I will try setting these xorg.conf options later if it hangs
15:44fling: thanks for help
16:45fling: AndrewR: should I also try disabling dri3, egl, gbm and gles2 in mesa?
16:55AndrewR: fling, I don't know, I prefer to keep those compiled .....
17:00fling: ok, will keep them
17:00fling: disabled opengl everywhere though
17:00fling: except mpv :D
17:06imirkin: fling: i think you just want to remove nouveau_dri.so -- that should get you what you want.
17:07fling: imirkin: tell me more please
17:07imirkin: there's a command called 'rm'
17:07imirkin: it can be used to remove files.
17:07fling: no I know
17:07fling: I can add it to INSTALL_MASK
17:07fling: what will it do for me actually?
17:07imirkin: no nouveau GL accel
17:07imirkin: regular X accel
17:08fling: Is it only achievable by removing said file instead of tuninng any build flags?
17:08imirkin: oh sure - you can also not build nouveau in mesa.
17:09fling: ah ok
17:09imirkin: just remove it from the list of gallium drivers
17:09imirkin: wasn't sure what kind of setup you were running
17:09fling: and what use instead? like vesa?
17:09imirkin: huh? no, it'll go to llvmpipe
17:09fling: no, tried llvmpipe in .xinitrc and it crashes on awesome exec
17:09imirkin: vesa would be for display. you still use xf86-video-nouveau.
17:10fling: llvmpipe is the only thing to reboot my box shortly after the crash
17:10imirkin: ah ok. i thought you just wanted to stop using nouveau for GL.
17:10fling: imirkin: nah
17:11fling: imirkin: I did this AccelMethod none and NoAccel true
17:11fling: in xorg.conf do workaround crashes
17:11imirkin: that kills the xorg "2d" accel
17:11fling: but client resize became painfully slow especially with mpv running
17:11imirkin: which is super-stable. it ALSO disables glx though.
17:11imirkin: so you basically lose all accel when you do that
17:11imirkin: 2d and 3d
17:11fling: so I wanted to also workaround this too
17:12imirkin: alright, sounds like you just want something super-custom which is a bit beyond my understanding. good luck!
17:13fling: and I ended up with using the default xorg.conf for awesome to be able to use the accel BUT also exporting gallium driver llvmpipe and gibgl always software true in awesome to every client
17:13fling: imirkin: now everything except awesome is using llvmpipe here! ;D
17:13imirkin: i see. so you want one application using accelerated GL, but everything else using swrast
17:13fling: I'm waiting for a crash though since opengl things are super fast
17:14fling: maybe llvmpipe is too good
17:14imirkin: that reminds me ... i really need to do that xf86-video-nouveau release
17:14fling: And I started using llvmpipe because rxvt-unicode causing hangs
17:14imirkin: every time i try it's a rabbit-hole of remembering this or that
17:14imirkin: hopefully my pgp key hasn't expired yet
17:14fling: tried taskwarrior?
17:14imirkin: dunno what that is
17:15imirkin: so i'll go with 'no'
17:15fling: can do multiple projects, easily add tasks and subtasks with deps and dues and tags etc
17:15fling: gtd friendly app
17:15imirkin: and it does them for me?
17:15fling: you just add things you forget to do as tasks there…
17:15imirkin: is it different than trello in a material way?
17:16fling: idk, not tried trello, I will check later
17:16imirkin: the problem is too many tasks, too many things to do, not enough time
17:16fling: this is why I'm using this taskwarrior
17:16fling: it helps with figuring out what to do first
17:16imirkin: plus a giant mega-case of procrastination
17:17fling: because I can set weights per project, per tag, per any other thing like dependency etc
17:17fling: ahhh right
17:17fling: imirkin: coming from depression?
17:17imirkin: unless i've been depressed all my life
17:18fling: I should write a nice deps viewer for task
17:18fling: something with arrows :D
17:19imirkin: but there's a certain paralysis which comes with "too many things to do", almost like the paradox of choice
17:21fling: with default gtd workflow in taskwarrior you spend up to 15 minutes a day on managing your tasks first
17:21fling: you make sure all the tasks are weighted properly
17:21fling: you delete stale ones add new ones
17:22fling: then you just do whatever task is on the top
17:22fling: taskwarrior is what helps you to choose
17:24fling: imirkin: I'm going to try disabling these https://www.x.org/releases/current/doc/man/man4/exa.4.xhtml if it even crashes again
17:24fling: am I still using exa if everything exept awesome-wm is running on llvmpipe?
17:25imirkin: no exa = no GL accel
17:25imirkin: yes, you're using EXA
17:25fling: ok this is why opengl is so fast probably
17:29fling: imirkin: are there any more env variables I could use for disabling more accel here and there?
18:47karolherbst: imirkin: honestly.. with all the issues nouveau has with HDMI speed limitations and everything, I don't feel that bad about when I see how intel fails with that as well
18:47karolherbst: also on DP
19:00_171_: Hello, I have more questions regarding power management.
19:01_171_: So, my GPU would only work properly if I set runpm=0 and config=NvClkMode=7 in my kernel parameters.
19:02_171_: Now, since nouveau doesn't have any kind of dynamic power management besides turning the GPU on/off, runpm=0 only makes it so it's always on, correct?
19:06_171_: If I don't set these parameters and launch something with DRI_PRIME=1 so that it runs on the primary GPU, then the card boots up and uses clock mode 7 anyway according to pstate.
19:07_171_: What I don't understand, is why these kernel parameters make any difference at all if it's still using the same clock frequencies regardless.
19:12imirkin: _171_: correct - runpm=0 makes it always-on
19:13imirkin: _171_: pstate indicates a set of parameters to be configured ... maybe like 30-40 parameters. the high-level clock speeds are a fraction of that. so you don't really know how it comes up on its own.
19:14_171_: Oh, so the config string doesn't have the same meaning as the pstate mode.
19:16_171_: Alright, thank you!
19:52imirkin: AndrewR: finally pushed the release
19:52imirkin: mattst88: friendly request to bump xf86-video-nouveau in gentoo (to 1.0.17)
19:54AndrewR: imirkin, thanks :}
21:04mattst88: imirkin: will do, thanks
23:05karolherbst: imirkin: mhhh.. :/ 4K is quite laggy with nouveau when doing prime.. and the PCIe speed is already maxed out :(
23:11karolherbst: but fun.. my GPU starts in 8.0 mode
23:27karolherbst: ehh.. and some display related bugs