00:08Yoshimo: how often do patches get merged to mesa-master, every 24h?
01:46martm: was still looking for the cache aware version of the scheduler during this weekend, as the code seems to be available in several places, it seems to be quite an easy port
01:48martm: this one only affects memory operations, with a minor effort the cache is handled too, it would be static analysis in the compiler , best done with a passthrough shader
01:50martm: https://github.com/tue-es/gpu-cache-model code is roughly here, would need to be integrated in the driver very similarly, but blocks/dims etc. lifted to 3d pixel fetch fragment iterations
01:51martm: it just traces the IR's ld and tex operations and dumps the info to the file, and feeds it to the model, and gets back the masks
01:53martm: so rouhgly when it's cache hit it masks fewer threads, when it's a miss it masks most of them
03:18martm: i sincerely thank you all for the live participation in da subjecto:) i can not witdhstand all the feedback and questions i get:D:D
03:22jazzpi: i'm trying to change back from nvidia to nouveau drivers, but X fails to start with "Unknown chipset: NV117", but according to https://nouveau.freedesktop.org/wiki/FeatureMatrix/ it should be supported (at least partially), right?
03:22jazzpi: Xorg.0.log: https://goo.gl/WGHukj
03:27martm: jazzpi: what you mean back, did you ever get that chip running with nouveau?
03:29jazzpi: i think so, but i'm not sure...
03:29martm: https://nouveau.freedesktop.org/wiki/CodeNames/ seems to be a maxwell card, so it ought to run as phoronix would suggest
03:29martm: but probably with something new in linux kernel front
03:29jazzpi: i believe it worked before i installed nvidia drivers
03:29martm: samewise with the mesa and stuff prolly
03:30martm: jazzpi: what distro?
03:31martm: right, so make sure you have libGL.so that of mesa in correct path
03:31jazzpi: do i need mesa-libgl to start X?
03:32martm: ouh, prolly not..
03:32martm: i think the glamor was removed, not quite sure though
03:32jazzpi: i installed it and it still fails with the same error :/
03:33martm: can you post all the Xorg.0.log?
03:34jazzpi: unless there are logs somewhere else than ~/.local/share/xorg
03:34martm: 19.268] (II) NVIDIA GLX Module 361.28 Wed Feb 3 15:10:57 PST 2016
03:34martm: there seems to be a problem that it loads a glx.so from nvidia ...
03:35martm: go twiggle with the package management where is the glx.so from Xorg of nouveaus driver, and then it should load
03:35jazzpi: hm, i installed mesa-libgl instead of nvidia-libgl
03:35jazzpi: still fails
03:37martm: yeah i can see, that it loads correct glx.so, sucks..i am a dummy
03:38martm: googling a bit, if that chip infact should work
03:40martm: seems like it should, it needs a firmware from aur repos, well go ahead and post dmesg too
03:41jazzpi: no nouveau errors in dmesg
03:43martm: yes, but you see searching the firmware from the dmesg logs does not show anything
03:48jazzpi: from what i gather from http://www.phoronix.com/scan.php?page=news_item&px=MTY2NDI you should only need to extract the firmware for 3D acceleration
03:52martm: jazzpi: maybe yeah..
03:52jazzpi: huh, i removed this file from /etc/X11/xorg.conf.d and now it works https://gist.github.com/jazzpi/69a234454c010c2df4c5
03:52martm: jazzpi: not quite sure, then try booting with nomodeset option
03:52martm: which driver works now?
03:53martm: well that is nice , good luck!
03:54martm: jazzpi: though i would like to see the Xorg.0.log again, what ddx it needed
04:01jazzpi: suddenly got a lot more complicated
08:17imirkin: Yoshimo: patches are pushed to mesa whenever the person pushing them ... pushes.
08:19Yoshimo: so until your proposed patch for the wow ui corruption will land in my ppa is unknown. Compiling it myself was too complicated. Too many dependencies
08:21imirkin: i'll push it out whenever i hear someone confirm it fixes the issue for them
08:29Yoshimo: after this video i might go back and try to get it tested
08:29imirkin: i'm in no rush :)
09:55martm: i am not quite sure, if the execution of the trace could be overall skipped, if it would be possible to get the same physical address of the data twice into cache
10:02martm: imo bitwise shift calculated from a copy of texcoord for instance and fed to texture2d or image load could produce duplicated addresses in data cache
10:05martm: aaah heck, appareantly not, should be safe to assume, that addresses are unique in data cache if varyings used, or basically same throughout the geometry, well compiler needs some analysis passes to get things right
10:14martm: i.e when one don't want to execute the ocelot kinda ir, which is undeniably overheadish, compiler rules are needed to take care of duplicates when coords are read from texture for instance it needs to search for duplicated values
10:17martm: then adjust the index etc. looks to be possible to make some smaller loops instead of executing, in case of instruction cache, compiler would give the addresses anyways, alltogether some work still yet to be done
10:43Yoshimo: cloning mesa, applying changes, autogen.sh , make &make install and a reboot should be enough to test stuff before it is merged upstream , shouldn't it?
10:44imirkin: not exactly the series of steps i'd take, but i guess so yeah
10:44imirkin: i'd install into a prefix, and then just run the application in question with a LD_LIBRARY_PATH
10:44imirkin: note that if the application in question is wine, it'll probably want a 32-bit version of your libs, not 64-bit
10:47Yoshimo: something must have gone wrong there because it still looks as ugly as before
10:52Yoshimo: why 32bit libs? i run wine64 on a wow-64.exe
10:52imirkin: ok, then 64-bit is fine. most windows games are 32-bit, that's why i mentioned it
11:42mupuf: imirkin: hey, an arch user has been asking the nouveau-fw to fix some of the filenames generated by extract_firmware.py and linked to this comment: https://bugs.freedesktop.org/show_bug.cgi?id=93629#c9
11:43mupuf: I also believe that Ben is right
11:43mupuf: and we should not support NVIDIA's PGRAPH firmware
11:43mupuf: but what about the video firmwares? Do we need to move them too?
12:04mooch: hey, uh
12:04mooch: i need some help figuring out why my emulation is hanging
12:05mooch: i'm attempting to emulate an nv4 on nt4, but it hangs
12:05mooch: i have logs for anybody interested
15:45imirkin: mupuf: blob firmware makes things work for some users
15:46imirkin: mupuf: it's sad and unfortunate that it got renamed across linux versions... i can update instructions and/or create more symlinks
15:55imirkin: mupuf: it's also your prerogative to not ship those in arch, but i do think my script should continue to allow them to be extracted.
16:39rardiol: Folks, what does a lot of messages like "nouveau: ch0: psh 00000001 0000000000 000000003c" or "nouveau: 0x00010000" on the terminal and a broken program mean?
16:45Jayhost: imirkin did you write that Vfetch handler for gm107 tesselation?
16:50imirkin: Jayhost: nope
17:03imirkin: Jayhost: oh, do you have nouveau loaded on your maxwell board?
17:04imirkin: if so, could i trouble you to grab mesa 11.2-rc3 and grab the output of 'glxinfo -l -s' for me?
17:57Jayhost: imirkin alright
18:01Jayhost: I havn't been able to help much recently. Been refreshing adrenal glands.
18:02imirkin: no worries
18:07Jayhost: installing llvm
18:11Jayhost: I had noticed that envyas was not as up to date as nvdisasm but I was glad to be able to use the 8mb nvdisasm without dependencies
18:14imirkin: heh. of course nvdisasm comes as part of a 700mb cuda tools package :)
18:16Jayhost: haha :D
18:16Jayhost: looks like I have to uninstall old mesa
18:16imirkin: sorry, i didn't mean for this to be a ton of trouble for you
18:17imirkin: i had assumed you'd just do a build and then look at glxinfo... no need to overwrite your system install...
18:17imirkin: it's nothing urgent
18:18Jayhost: glxinfo after install still says 11.1
18:18imirkin: i guess you're not familiar with the usual build procedures?
18:25Jayhost: well I'll try to figure it out here
18:30Jayhost: got it
18:31Jayhost: tell me if this is what you need http://hastebin.com/xafoqopezo.pl
18:32imirkin: Jayhost: close. please run 'glxinfo -l -s'
18:32imirkin: OpenGL renderer string: Gallium 0.4 on llvmpipe (LLVM 3.6, 256 bits)
18:32imirkin: also that's a fail =/
18:32imirkin: sounds like you didn't build with nouveau support
18:34Jayhost: I just did ./configure make & make install
18:38Jayhost: Let me know
18:39Jayhost: I take it you were making fun of my debian install
18:42Jayhost: imirkin LLMV too old?
18:49imirkin: you're still on llvmpipe
18:50imirkin: you need to --with-gallium-drivers=nouveau
19:11imirkin: Jayhost: still swrast
19:11Jayhost: ya killin me
19:12Jayhost: imirkin I did ./configure --with-gallium-drivers=nouveau
19:12Jayhost: maybe dependency issue with gallium
19:12imirkin: you must have done something odd
19:13imirkin: Jayhost: https://nouveau.freedesktop.org/wiki/InstallNouveau/#mesa
19:16Jayhost: okay next paste will be ./configure then next will be autogen
19:25imirkin: you appear to have missed --enable-texture-float
19:25imirkin: did you look at the wiki page i pointed you to?
19:30Jayhost: Ya the autogen version is building now
19:33Jayhost: Alright will assume this one is it http://hastebin.com/oyirorujis.pl
19:34imirkin: nope =/
19:34imirkin: still at gl 2.1
19:40Jayhost: What the hell hahaha
19:42Jayhost: okay I guess I gotta do the git clone thing
19:42Jayhost: I was using the tar off the mesa download page
19:44imirkin: shouldn't matter
19:47Jayhost: What does the issue seem to be
19:53imirkin: you forgot --enable-texture-float?
19:57Jayhost: Yeah I guess because the way it's formatted on the page I have to figure out how to reformat will take a few mins
20:02imirkin: errrr what?
20:02imirkin: it's formatted fine
20:09Jayhost: Idk I'm assuming the whitespace screws it up
20:09Jayhost: building now
20:10Jayhost: Or I added whitespace. Hopefully it's good now
20:23Jayhost: Maybe you can tell me what I'm missing
20:24imirkin: that seems right
20:26imirkin: maybe we busted something?
20:26Jayhost: Maybe, what might that mean?
20:35Jayhost: I guess I could downgrade and check glxinfo
21:46imirkin: mupuf: what's in reator nowadays?
22:01Jayhost: imirkin when I do a new make install it doesn't overwrite glxinfo. Idk if there's a make uninstall
22:02Jayhost: okay there we go. I've got 11.1 at opengl 3.3
22:11Jayhost: new mesa building
22:21Jimi: Hey, I have a problem. I'm trying to get a dual-card, dual-monitor setup, both NVidia cards, both under the Nouveau driver, with a standard ServerLayout section in xorg.conf.d. In Xorg.0.log, the only reference to the GTX 960 at all is "(EE) Unknown chipset: NV126". The other card, a GT 740 (low profile), is working fine right now.
22:21Jimi: I checked the website and it says NV126 should be supported?
22:21Jimi: Both cards work in the proprietary driver, but I'm trying to avoid it because it's annoying and buggy.