00:02 karolherbst: ugh, librtmp.so.0 is required
00:04 mgoodwin: http://paste.fedoraproject.org/360476/01821146
00:04 mgoodwin: no dice
00:04 mgoodwin: Three crashes with PGRAPH firmware loaded
00:04 mgoodwin: Back to proprietary :(
00:05 imirkin_: yeah, i think a lot of people have had issues with plasmashell
00:07 mgoodwin: no other magic options to try?
00:08 imirkin_: i assume you're not using libdrm 2.4.60?
00:08 imirkin_: and i assume you're on a recent mesa? (i asked you before, but i forget quickly)
00:08 mgoodwin: libdrm-2.4.66-1.fc23.x86_64
00:09 mgoodwin: mesa-dri-drivers-11.1.0-2.20151218.fc23.x86_64
00:09 imirkin_: well, like i think i mentioned, you could try 11.1.2
00:09 mgoodwin: You said there was a newer version of mesa
00:09 imirkin_: although to be perfectly honest, i can't remember if there are fixes there that could help
00:09 imirkin_: let me look it over real quick
00:11 mgoodwin: https://bugzilla.opensuse.org/show_bug.cgi?id=957378
00:11 imirkin_: yeahhhhh not sure. definitely a bunch of fixes, but i couldn't say
00:12 mgoodwin: someone has the same bug here but the link to the freedesktop bug doesn't seem related
00:12 imirkin_: so THAT issue is fixed with a mroe recent kernel
00:12 imirkin_: and is a totally different bug
00:12 imirkin_: that's the set_domain thing
00:12 imirkin_: i sent a patch for that which has been backported to a variety of kernels
00:12 mgoodwin: yeah I read you fixed it
00:13 imirkin_: if you're seeing any of that, update your kenrel
00:13 mgoodwin: Probably more likely for me to find more bugs with a message not using PGRAPH firmware
00:13 mgoodwin: I doubt many peopele are doing that
00:13 imirkin_: shouldn't matter
00:13 mgoodwin: Oh?
00:14 mgoodwin: I figured loading/notloading the firmware would change the outcome
00:14 imirkin_: i suppose it could
00:14 imirkin_: it's just ctxsw fw
00:14 imirkin_: i.e. some firmware that runs to read out a bunch of gpu state, and store somme old gpu state
00:14 imirkin_: ctxsw = context switch
00:14 imirkin_: in case that's not trivially obvious
00:15 mgoodwin: Well the facts are that without the firmware im stable for less than an hour, usually immediately crashes
00:15 imirkin_: the thing is i have no clue what your errors mean
00:15 mgoodwin: With the firmware I can go as long as 5-8 hours
00:15 mgoodwin: How do I make them more detailed?
00:16 imirkin_: you read nvidia internal docs about wtf it means :)
00:16 mgoodwin: :(
00:16 imirkin_: ultimately it seems like we're telling it to use a texture that's not there
00:16 imirkin_: and there aren't a TON of reasons i can think for that to happen
00:16 mgoodwin: o shit it came back
00:17 mgoodwin: O_o
00:17 imirkin_: but perhaps i fixed it in one of my various commits
00:17 imirkin_: so definitely trying a more recent mesa should be on the menu
00:17 mgoodwin: I just let it sit while I talked to you on my laptop and it started reacting again
00:17 mgoodwin: Oh wow and then it immediately locked up again
00:19 mgoodwin: thanksfully someone I know has a COPR setup for a F23 build of 11.2.1
00:19 mgoodwin: Let's give it a whirl
00:20 mgoodwin: imirkin_: do you want me to go firmware or no firmware
00:20 imirkin_: doesn't matter
00:20 mgoodwin: ok
00:20 imirkin_: there were also minor vdpau fixes btw, if you happen to watch super-high-bandwidth videos
00:20 mgoodwin: I choose no firmware because less time to negative result
00:20 imirkin_: (or high-resolution)
00:24 imirkin_: bbl
00:28 karolherbst: mupuf: yay, it is working
00:29 karolherbst: mupuf: just needed to remove the shiped libcurl.so linking against librtmp.so.0 and libgssapi_krb.so :D
00:32 karolherbst: k
00:32 karolherbst: post processing is broken with nouveau
00:33 alligatorjoe: hey has anybody seen crocodile lilly lately...i can't find her anywhere around this swamp
00:33 karolherbst: 20 fps in the main menu on ultra settings :D
00:34 alligatorjoe: whats up with the lack of 256 color support in the nouveau driver...that should have been the easiest to implement
00:34 karolherbst: "fifo: read fault at 015283f000 engine 00 [GR] client 00 [GPC1/L1_0] reason 00 [PDE] on channel 2 [00bf890000 TombRaider[7228]]" :D
00:34 karolherbst: yay
00:37 alligatorjoe: and what is the matter with xrandr that can't set bpp or the new depth syntax
00:37 karolherbst: mhhh
00:37 karolherbst: hakzsam: Tomb Raider uses GL_ARB_shader_image_size
00:37 karolherbst: and GL_ARB_shader_texture_image_samples
00:37 karolherbst: the heck is GL_EXT_shader_integer_mix
00:38 alligatorjoe: this nouveau driver needs a hell of a lot of work before it will ever be useful
00:39 karolherbst: hakzsam: but no compute_shaders used as it seems
00:41 karolherbst: awesome, that engine fault is reproduciable
00:41 karolherbst: imirkin_: mmt trace for that fault will help?
00:42 alligatorjoe: 256 color support should have been the first goal of nouveau
00:43 alligatorjoe: why have 16 million or whatever if you can't set up to only use 256
00:44 karolherbst: alligatorjoe: who wants ot use 256? seriously
00:44 alligatorjoe: karoll....anyone who needs to use a damn computer....how will they ever play myst for example
00:45 karolherbst: alligatorjoe: myst is through wine, right?
00:45 karolherbst: or is there a native port now?
00:45 alligatorjoe: karol...of course..but wine needs x to work adequately
00:46 alligatorjoe: karol...it is the nouveau driver that will not allow 256 color mode on x
00:47 alligatorjoe: karol...and there are thousands of 256 color games and apps
00:47 karolherbst: yeah I kind of get this already, you should try to stay focused on the issue and work on your phrasing a bit. We want to help, but you shouldn't just, you know.
00:47 alligatorjoe: karol..that is just only one obvious example
00:47 karolherbst: alligatorjoe: yeah, I get that, but nobody really uses the deskopt with 256 colors anymore
00:47 alligatorjoe: karol...of course not....that doesnt need to be the default mode.
00:48 karolherbst: I guess wine changes the 256 colors?
00:48 alligatorjoe: karol...but it should be an available option
00:48 karolherbst: or does it just fail in doing so
00:48 alligatorjoe: karol..no wine is set to use whatever x is set for
00:48 karolherbst: yeah so, where is the issue exactly?
00:49 karolherbst: or does the game just missrenders?
00:49 alligatorjoe: and xrandr needs a hell of a lot of work..but in the meantime you can use Xset...for depth options...and reboot....but nouveau does not permit only 256 colors
00:49 imirkin: karolherbst: someone would have to look at it. dunno i'd have the time.
00:50 alligatorjoe: imirkin...yea well if it doesn't get done soon it is no big deal..but a note should be made about its limitations
00:50 karolherbst: alligatorjoe: the question is, is it worth the effor fixing it
00:50 karolherbst: alligatorjoe: I think he meant the issue I found with tomb raider :/
00:50 alligatorjoe: karol...well....that depends on whether you puchased a lot of games and apps that you don't want to trashcan
00:51 karolherbst: alligatorjoe: well but they should run when your desktop is using 16 millions of colors, right?
00:51 karolherbst: I don't see the point of switching your desktop to 256 colors
00:51 alligatorjoe: karol...no some videos require the 256 color only mode...check myst on xp if you don't believe me
00:52 imirkin: alligatorjoe: not sure what you're trying to make... is it that amd gpu's have much better open-source support and everybody should dump nvidia and go amd? if so, i agree.
00:52 imirkin: what point*
00:52 karolherbst: alligatorjoe: okay, so if you start them through wine, you need to have your desktop setup with 256 colors?
00:52 karolherbst: alligatorjoe: I just want to understand this issue first, because you clearly need it (I assume?)
00:53 alligatorjoe: imirkin...well i certainly have an issue with nvidia ..but i am not sure amd is any better...i have no way to check it at this time....and plus i don't know if they are going to play the same game of telling everyone they need to purchase an $8999 video board for their computer
00:55 alligatorjoe: like telling them to purchase an $8,999 dollar pascal board or else settle for used clothes with everyone's fecal matter and urine all over them at a reduced price
00:55 imirkin: so... what *is* your goal?
00:56 alligatorjoe: imirkin....well my goal is certainly not to charge $8,999 to get something on the video monitor
00:56 karolherbst: alligatorjoe: do you plan on using nouveau with your GPU and encountered those issues and want them to be fixed?
00:57 alligatorjoe: karol..no i make the best of the situation..and just walk on by ..unresolved issues...until someone decides to scratch an itch..who is interested in promoting free software.
00:58 imirkin: which i assume is not you? you're being pretty hostile towards nouveau...
00:59 imirkin: i dunno what you're trying to achieve, but going around saying "your software sucks" isn't a great way to get helped.
00:59 imirkin: even if the software does suck
00:59 alligatorjoe: imirkin...why should i be hostile towards nouveau?...you need evidence to back up statements...my hostility is with all the damn hardware companies that owe me royalties for my model 1 ideas
01:00 alligatorjoe: imirkin...making a factual statement that the software is inadequate for lots of stuff...is not hostility...it is plain fact...and specific as to what issue is binding
01:01 karolherbst: alligatorjoe: yeah well, we don't get paid and we can usually have only time to fix issues, which are more important than running some old windows games through wine :/
01:01 karolherbst: sorry about that, but this is the situation
01:01 karolherbst: if we could, we would fix every issue
01:01 alligatorjoe: karol...fine like i say i just wait for someone to scratch an itch
01:02 imirkin: so perhaps step 1 isn't to alienate the people working on the project
01:02 karolherbst: alligatorjoe: well everybody is free to help
01:03 alligatorjoe: karol...well that is up to the free software people..if they want to give their ideas away for free...that is entirely up to them...after all there is biblical scripture to support it...but i certainly don't give my model 1 ideas away for free
01:05 imirkin: sounds like you'd be more satisfied with nvidia's proprietary driver
01:05 alligatorjoe: imirkin...well that is the rub...as they say...the nvidia driver no longer compiles with the newer kernels and the newer gcc
01:06 karolherbst: alligatorjoe: old gpu?
01:06 imirkin: so your solution is to go complain about highly non-specific issues to a community of a handful of volunteers?
01:07 alligatorjoe: imirkin...nvidia has been woefully inadequate in ensuring their products can be used by kernel developers
01:07 imirkin: so why do you keep buying them?
01:07 imirkin: actually, nevermind. i'm unsubscribing from this conversation. good luck.
01:08 alligatorjoe: imirkin...i don't...$8,999 for a video board is too pricey for me....and i don't enjoy use clothes that have been shit on and pissed on by many for a reduced price version
01:08 karolherbst: imirkin: the tomb raider issues seems like TressFX related :D :/
01:08 imirkin: karolherbst: well.... if you can make a mmt + dmesg errors i could perhaps look at it, but... no guarantees.
01:09 karolherbst: right
01:09 imirkin: karolherbst: not sure what L1_0 is. could be something ssbo/image related
01:09 imirkin: e.g. we're not flushing the cache enough
01:09 karolherbst: but at least we can tell people: disable tressfx with nouveau
01:09 karolherbst: :D
01:10 karolherbst: mhh
01:10 karolherbst: well it uses image in their fragment shaders
01:10 imirkin: plain images, or volatile? and/or atomic ops?
01:10 karolherbst: GL_ARB_shader_image_size
01:10 imirkin: that ext has net to nothing to do with images
01:10 karolherbst: ohh okay
01:10 imirkin: it's basically retrieving properties of images
01:10 karolherbst: ohh okay
01:10 imirkin: which are just in some constbuf
01:11 karolherbst: they also use GL_ARB_shader_texture_image_samples
01:11 imirkin: same thing
01:11 karolherbst: k
01:11 karolherbst: well I already uploaded the shaders
01:11 imirkin: (hard-coded to 1 for now)
01:12 karolherbst: GL_ARB_fragment_layer_viewport, GL_ARB_shader_image_size, GL_ARB_shader_storage_buffer_object, GL_ARB_shader_texture_image_samples, GL_ARB_texture_query_levels and GL_EXT_shader_integer_mix are used
01:12 karolherbst: no idea if they require something below GL 4.2
01:14 imirkin: ssbo is part of GL 4.3
01:14 imirkin: as is fragment_layer_viewport
01:14 imirkin: shader_texture_image_samples is from 4.5
01:15 imirkin: either way, i'm sticking to my "it's a ssbo/images issue" theory
01:15 karolherbst: k
01:15 karolherbst: well I try to trace it now
01:15 karolherbst: might take a while
01:16 karolherbst: god why does the game switch to fullscreen now...
01:22 karolherbst: yay 651MB apitrace uncompressed
01:24 karolherbst: yay!
01:24 karolherbst: imirkin: that apitrace triggers the fault :)
01:24 karolherbst: awesome
01:25 karolherbst: huh
01:25 karolherbst: glDispatchCompute
01:25 karolherbst: ohh
01:25 imirkin: which dispatches a compute program which uses images and ssbo's, i'm guessing
01:25 karolherbst: it uses compute shaders
01:26 karolherbst: why doesn't ST_DUMP_SHADERS dumps those then
01:26 karolherbst: ...
01:26 karolherbst: mhh
01:26 karolherbst: I guess TressFX is using compute stuff
01:27 karolherbst: well at least I should be able to do a mmt trace easily now
01:29 karolherbst: "[62215.464765] mmap: mmt-amd64-linux (3516): VmData 23916544 exceed data ulimit 0. Will be forbidden soon." :/
01:29 karolherbst: huh
01:30 alligatorjoe: man this is ridiculous...it has now been more than three days to try to compile freecad...and then i have to see if the nouveau driver can support at least some of it
01:30 karolherbst: alligatorjoe: is there no package you could just use?
01:31 alligatorjoe: karol..my interest is in developing gcc and the kernel...not in using someone elses packages
01:31 karolherbst: imirkin: hehehe, guess what happened with the mmt...
01:32 karolherbst: alligatorjoe: well here is not hte place to talk about compilation problems with freecad ;)
01:33 alligatorjoe: karol...i always have lots of things to do as many factors are constantly deprecated or upgraded or enhanced or have to be shelved when the compiler standards change or the kernel changes
01:34 alligatorjoe: karol and gcc has to change many times just to keep the ability to compile a newer kernel
01:35 alligatorjoe: karol..well i am not seeing compilation problems with freecad...it is going fine...it is just taking 3 days to complete
01:37 alligatorjoe: karol..it is of course to be made available to anyone who wants to use free software
01:38 alligatorjoe: as is ease....but i don't have a build on that yet
01:48 alligatorjoe: karol i am hoping it won't require another day....it seems to be nearing a finish where i can log any further limitations of nouveau if there are any
01:52 karolherbst: nice, nouveau above 60% performance in tomb raider :)
01:53 karolherbst: imirkin: https://drive.google.com/file/d/0B78S7GSrzebIMDRYdWJLUjNoWFk/view?usp=sharing here is the trace if you want to take a look
01:54 alligatorjoe: karol..well i would sure like to experience tomb raider....i would pay top dollar for tomb raider if they would support my choice of hardware
01:55 alligatorjoe: karol...and if those hardware bastards would pay me the royalties they owe me for using a bios or ram chips or rom chips or pixel graphics or sound , then financial constraints would not be so binding
01:57 imirkin: karolherbst: what's the associated dmesg error?
01:59 karolherbst: fifo: read fault at 010d67f000 engine 00 [GR] client 00 [GPC2/L1_0] reason 00 [PDE] on channel 2 [00bf890000 glretrace[3517]]
01:59 karolherbst: this was on replay
01:59 karolherbst: sadly I can't make a mmt trace
01:59 karolherbst: due to the ioctl thing ...
01:59 karolherbst: but the apitrace also triggers it
02:01 karolherbst: imirkin: what is the easiest way to force the old ioctl interface?
02:02 imirkin: karolherbst: flip it in mesa
02:02 alligatorjoe: karol...from the trailers i could see..Laura seems to wear her dual pistols well...a classic idea...and well worth top dollar.
02:03 karolherbst: imirkin: how?
02:04 imirkin: drm->nvif = false in ... the winsys thing
02:07 karolherbst: mhh
02:07 karolherbst: I don't find it :/
02:07 karolherbst: ohh wait
02:07 karolherbst: now I get it
02:07 imirkin: you have to add it.
02:07 karolherbst: in nouveau_drm_screen_create?
02:08 alligatorjoe: karol...everyone i talked to that said they bought the Laura Croft Tomb Raider Rises..said they had no regrets...they felt their money was well spent....and they would do it again if another great one comes along
02:12 imirkin: karolherbst: yeah sounds right
02:18 alligatorjoe: karol..it is just kind of sad that the only purchasers were women
02:20 alligatorjoe: karol..that is only a potential market of about half the world
02:24 karolherbst: imirkin: no idea why, but somewhat didn't work :/
02:26 alligatorjoe: well i am hoping another day won't be required...and of course it is only version 0.15
02:26 alligatorjoe: karol...but it should make a nice compliment to the 2d librecad and blender 3d attempts
02:27 imirkin: karolherbst: yeah, demmt needs some help
02:30 alligatorjoe: karol..if they would just keep all those gnome compositors and other crap out of the way so people can get some real work done.
02:31 alligatorjoe: karol it is basically why noone wants a desktop that uses sound...because some people are trying to do some real work with sound and dont want the desktop stuff to get in the way
02:56 alligatorjoe: karol...and i can't do much amd/ati examinations because i only have one agp radeon board ....and i really need to use it on the windows machine
03:05 alligatorjoe: karol..just to work on issues for people that don't want to use free software and like to pay to use paid-for software
04:09 alligatorjoe: karol..because clearly there are some people that want to spend every penny they have to buy paid-for software instead of trying to use any free software
04:42 alligatorjoe: karol..and then again..there are some people who are sick and tired of spending every penny they have for a few pieces of paid-for software
08:43 alligatorjoe: karol well ok i got a build on freecad and it has some documentation on the screen but i have no idea how to find a sample to see if the 3d issues will work on nouveau...
08:54 EdB: hakzsam, pmoreau, imirkin and Hans (I don't his nick ):I don't know if this information will be usefful but: for work_dim and other value of the same kind, as for radeon driver, clover pass them as implicit arg's to the driver. They are put in the buffer arg and then OpenCL runtime read them at know position.
08:56 pmoreau: EdB: (Hans is hansg, but he is really rarely hanging out on IRC.)
08:57 EdB: ok, thanks
08:57 pmoreau: EdB: Oh right, but wouldn’t it be better to have a separate buffer for those?
09:02 EdB: they can be see as constants args
09:06 EdB: soem driver may want to put them in some special register. I'm not sure how it work for nouveau
09:08 hakzsam: they are sticked into the driver constant buffer, so we can just load the information directly from the compute shader, it's like constant arguments except that they are not uploaded as user uniforms
09:08 hakzsam: global/local sizes are already sticked there too, so it makes sense for nouveau at least :)
09:14 EdB: and those value already are the kind of value read by a SV instruction ?
09:18 hakzsam_: not really a SV instruction because we use a load for those
09:19 hakzsam_: but it's a SV instruction for tidx (for example)
09:20 hakzsam_: okay, I can replay the TombRaider trace but my gpu hangs...
09:26 hakzsam_: cool, this game uses ssbo, images and compute shaders
09:26 hakzsam_: good example :)
11:28 hakzsam_: okay so the issue seems to be related to a compute shader
11:29 karolherbst: hakzsam_: tomb raider thing?
11:30 karolherbst: ohh wait I should read the backlog first
11:31 karolherbst: ahh
11:31 karolherbst: yeah your GPU hangs :)
11:31 karolherbst: mine too
11:31 karolherbst: hakzsam_: I think they use it for the hair sikulation, this TressFx (or how this is called)
11:31 hakzsam_: I'm looking at the issue
11:31 karolherbst: hakzsam_: when I disable it ingame, it works
11:31 karolherbst: *simulation
11:32 hakzsam_: okay
11:32 karolherbst: the trace is with minmal settings and only this enabled
11:33 hakzsam_: Mesa: User error: GL_INVALID_OPERATION in glDispatchCompute(no active compute shader)
11:33 hakzsam_: well, this is bad
11:33 karolherbst: I thought as much already
11:33 karolherbst: but the overall performance is pretty good compared to nvidia, so I think it would be good to support this game kind of .D
11:34 hakzsam_: yeah yeah, but finding the issue in the trace with millions of calls is not trivial :)
11:34 karolherbst: well I tried to do a mmt, but couldn't create one due to the nvif thing
11:35 pmoreau: If you want to dig in TressFX, don’t forget it is now open source: https://github.com/GPUOpen-Effects/TressFX/ (or at least part of it).
11:35 karolherbst: :D
11:35 hakzsam_: I don't think making a MMT is useful here
11:35 karolherbst: okay
11:35 karolherbst: but what is L1_0 and L1_1? Level 1 cache?
11:37 hakzsam_: don't take much attention of that, there is a read fault because something is not correctly bound and we need to figure what is going on
11:37 karolherbst: yeah I know, I just want to understand what the GPU is complaining about :D
11:39 karolherbst: hakzsam_: do you know which call upsets the GPU?
11:40 hakzsam_: not exactly
11:40 karolherbst: mhh okay, then I try to find out this
11:40 karolherbst: some range which it might be?
11:42 alligatorjoe: karol...well it appears that either the hardware or the driver itself does not support freecad at all
11:43 alligatorjoe: karol...there seems to be a lot of emphasis on stereo 3d which may require advanced shading language features...i am not sure
11:48 alligatorjoe: karol...i dont even have that stereoscopic 3d stuff working in blender at this time..so i am just putting this stuff on a backburner for awhile
11:50 alligatorjoe: karol...the last i checked there were some simple 3d objects manipulations in office....i could see if nouveau can support at least the simple office 3d stuff
11:55 karolherbst: hakzsam_: it hangs in frame 1398 for me
12:00 alligatorjoe: karol..ok nouveau seems to support some basic 3d objects as well as garoud shading and shadowing..and some lighting features...but the texture mapping features seem greyed out at this time
12:03 alligatorjoe: karol anyway that is a little 3d support for now even though freecad is a long way off i think
12:03 karolherbst: hakzsam_: between 2116395 and 2115050
12:08 karolherbst: 2116204-2116300 I am getting close :)
12:11 karolherbst: 2116271-2116289
12:11 karolherbst: hakzsam_: 2116272
12:12 karolherbst: it is a Draw call and with that the gpu hangs
12:13 hakzsam_: sorry I don't have for that right now actually
12:13 karolherbst: no worries
13:26 EdB: hakzsam_: is TGSI_FILE_MEMORY supposed to ack as memory buffer ?
13:26 EdB: to act
13:35 EdB: to be more precise are TGSI_FILE_BUFFERs deprecated?
14:27 p3rs3us: hey guys, I am getting segfault while attempting to run an application that uses OpenGL. A log of the segfault taken via dmesg is http://pastebin.com/CC7rk7rG I know it doesn't say much but you can let me know if you spot something.
14:42 dcomp: I see GM108 support has been included in skeggsb:master
14:43 dcomp: Still no improvement though I see :(
14:46 imirkin: EdB_: memory is just plain memory. buffer is relative to a bound buffer object.
14:48 imirkin: p3rs3us: what version of mesa are you using?
14:50 dcomp: "fifo: write fault at 0000102000 engine 05 [BAR2] client 08 [HOST_CPU
14:50 dcomp: _NB] reason 0c [UNSUPPORTED_KIND] on channel -1 [007fd76000 unknown]
14:50 imirkin: dcomp: file a bug
14:50 imirkin: (a new bug)
14:51 p3rs3us: imirkin, it is the one that comes with debian jessie I get Version: 10.3.2-1+deb8u1
14:51 p3rs3us: imirkin, i guess this is the package libegl1-mesa-drivers
14:51 imirkin: p3rs3us: ah i see. please check with 11.2.1 and let me know if it's still happening.
14:52 p3rs3us: imirkin, does it matter if both versions are installed?
14:52 imirkin: not sure what you mean... i doubt the package manager will allow that
14:52 imirkin: but i don't care what you have installed, as long as you're using the latest mesa :)
14:53 drgibbon: hi, I'm looking for a graphics card for Linux, and I want to use open-source drivers. basic acceleration would be nice, but I'm not looking to play games.
14:53 drgibbon: I heard that Kepler cards are the best for nouveau
14:53 p3rs3us: imirkin, fair enough i can try to get backport version
14:54 drgibbon: anyone know of a low power Nvidia card with good nouveau support? :)
15:01 imirkin: drgibbon: you're best off getting an amd gpu
15:02 drgibbon: imirkin: why do you think that?
15:02 imirkin: drgibbon: if you're *set* on nvidia, pretty much all keplers are equally supported.
15:02 imirkin: drgibbon: because amd has a team of full time engineers backing the amd driver
15:02 drgibbon: I have no special preference really, just something low power with basic 3D and open-source drivers
15:02 drgibbon: I've been reading online, but it's hard to determine what works and what doesn't
15:03 imirkin: the lowest power keplers are the GK208's... like 35W
15:03 imirkin: not exactly speed demons though
15:03 imirkin: (or 25W? i forget)
15:04 drgibbon: ok, good to know that all Keplers will be the same support wise
15:04 drgibbon: thanks, useful info
15:04 imirkin: they should be, but aren't. for some people, they work poorly. we have no idea why, or what separates those from the ones that work fine.
15:05 drgibbon: different vendors perhaps
15:05 drgibbon: doing weird things
15:05 drgibbon: do you know of how to find out if particular AMD chipsets are supported?
15:05 imirkin: well, it's nouveau not handling some odd condition
15:06 imirkin: you can go into #radeon and ask
15:06 drgibbon: will do
15:06 imirkin: i recommend that you get a gpu supported by the 'amdgpu' driver
15:06 drgibbon: stable?
15:06 imirkin: which means.... CIK+? not sure.
15:06 imirkin: nah, they're about the same - but amdgpu is going to be supporting more features going forward
15:07 drgibbon: ahh
15:07 imirkin: at least that's my impression
15:09 drgibbon: I actually have a Maxwell card running on the proprietary drivers, I assume that's not going to function so well with nouveau?
15:10 imirkin: should work fine... depending on the specific card you might need a very fresh kernel
15:10 imirkin: which one is it? GMxxx is what i'm looking for
15:14 drgibbon: hmm, GTX 750 Ti, is that what you mean?
15:15 imirkin: GM107. should be fine with any kernel 4.1 or later, iirc
15:15 urmet: i tried with a gm206 a few weeks back. kernel from latest upstream git. a bit too crashy and artifacty for use :)
15:15 drgibbon: ahh the codenames, ok I see
15:15 imirkin: urmet: i hope you reported a bug?
15:16 drgibbon: maybe I will just try nouveau with the current card then with some newer kernels and see how that goes
15:18 urmet: didn't have the time to investigate further :( -march=native compiled kernel and million other things that could mess it up from my own side. and the crashes were hard lockups - so no logs
15:19 imirkin: urmet: ah, sad
15:33 dcomp: https://bugs.freedesktop.org/show_bug.cgi?id=95188
15:35 RSpliet: dcomp: don't forget to add a description of the actual problem. Even if it's obviously "doesn't work after boot", just explain to what extend things do work
15:36 dcomp: well any time I try DRI_PRIME=1 ... it hangs for a couple of seconds and falls back to the intel
15:36 RSpliet: dcomp: include that in your bug report
15:37 RSpliet: now it's just a compilation of logs - which is nice and could be useful, but doesn't help attract the right people to your report ;-)
15:52 imirkin: dcomp: sounds like vram is just MIA =/
15:52 dcomp: would an mmiotrace from the nvidia driver be useful ...
15:53 imirkin: i suspect so, yes
15:58 karolherbst: k another issue within Tomb Raider: PostProcessing + Depth of Field Ultra gives bright white stuff in the foreground making the game unplayable: https://drive.google.com/open?id=0B78S7GSrzebIaExsb1FNSUNPTDA
15:59 karolherbst: I will open bugs for both issues
16:04 Wolf480pl: Hi. Is there anything (experimental?) that I could test for you on my GK104 in Optimus setup?
16:05 Wolf480pl: karolherbst?
16:05 karolherbst: Wolf480pl: well always my recent reclocking branch
16:07 Wolf480pl: is it for 4.5 or 4.6-rc ?
16:07 karolherbst: 4.5
16:07 karolherbst: but you can rebase on master_4.4 or master
16:07 karolherbst: master is always for newest rc
16:10 dcomp: arghhh... stupid modpost GPL enforcement
16:10 Wolf480pl: and master_4.5 is latest reclocking for 4.5 ?
16:13 karolherbst: Wolf480pl: no
16:13 karolherbst: Wolf480pl: master_4.5 is master just compatible for 4.5
16:13 karolherbst: Wolf480pl: stable_reclocking_kepler_v5 is the reclocking branch
16:25 mgoodwin: still have artifacts on DRI3 w/ Mesa 11.2.1 btw
16:28 mezo: any nouveau benchmarks for tomb raider?
16:40 imirkin_: mezo: i believe the report is that it hangs something in a compute shader that uses images and ssbo's
16:40 imirkin_: mezo: which are 3 recently added features. so somewhere in there :)
16:48 mezo: so it will run only with the latest mesa changes?
16:48 imirkin_: i'm not 100% sure. i think that might only be with tressfx
16:48 imirkin_: karolherbst has been trying it
16:49 karolherbst: mezo: tomb raider has like 60% perf compared to nvidia
16:49 karolherbst: mezo: TressFX causes the game to crash
16:49 karolherbst: mezo: PostProcessing + Depth of Field Ultra doesn't render the fog correctly, but that's working on currently
16:50 imirkin_: it's a feral-ported game, right?
16:51 imirkin_: hakzsam_: feral has been receptive to requests for keys from the radeon guys, and have indicated that they wouldn't reject requests from nouveau developers either
16:51 hakzsam_: imirkin_, pretty nice news
16:51 mezo: what are the main reasons for performance differences between nouveau und nvidia? does nvidia patch there drivers? the reason why im asking: many source games run nearly same on nouveau as on nvidia, but dota2 for example does not.
16:51 karolherbst: imirkin_: they also told me via twitter that they officially support Mesa :D
16:51 karolherbst: well
16:52 karolherbst: they kind of said any open source driver
16:52 imirkin_: karolherbst: i think that means "radeonsi"
16:52 karolherbst: I asked especially about nouveau
16:52 hakzsam_: imirkin_, seriously, I'm pretty sure that the hang is not compute-related (I will answer later for that bug)
16:52 mgoodwin: imirkin_: sorry I can't remember, did you say that the mesa update would have an effect on VDPAU too?
16:52 mgoodwin: Im on 11.2.1 now
16:52 imirkin_: mgoodwin: i said it could, for very-high-bitrate or very-high-resolution videos
16:52 karolherbst: I think
16:52 karolherbst: ..
16:53 mgoodwin: which to me is likely to mean UHD+?
16:53 mgoodwin: which I dont have
16:53 imirkin_: well, i don't think it's actually anything specific to high-resolution videos, just that those tend to be pretty high bitrate :)
16:53 mgoodwin: true
16:54 imirkin_: i think in practice, yeah, you'd need a good quality 4k video to hit the issues
16:54 imirkin_: which are rare, which is why we never ran into the issues for such a long time
16:54 mgoodwin: I don't know if mesa related but I only started noticing it after building and upgrade is I seem to be losing control of windows when moving them
16:54 mgoodwin: has that been reported....
16:55 mgoodwin: This is one of those issues that make it hard to articulate
16:55 mgoodwin: Also means it's hard to search for ;/
17:07 imirkin_: pmoreau: just noticed you're still pulling from skeggsb's old repo in your livecd thing
17:07 imirkin_: (or at least that's what the webpage claims)
17:08 mgoodwin: http://paste.fedoraproject.org/360760/63268146
17:08 mgoodwin: nope still bjorked
17:08 imirkin_: mgoodwin: never heard that before. i don't think a ton of people use plasmashell and nouveau.
17:08 imirkin_: probably because plasmashell invokes a ton of issues on nouveau
17:08 imirkin_: nice... a desync. oh, but that's from kodi. kodi + nouveau = fail
17:08 karolherbst: or Qt5 in general
17:09 imirkin_: or at least kodi's GL renderer + nouveau = fail
17:09 imirkin_: or maybe kodi + vdpau + GL = fail
17:09 mgoodwin: pretty much nouveau=fail
17:09 imirkin_: :)
17:09 mgoodwin: just buy amd right?
17:09 imirkin_: that's what i recommend to everyone who asks
17:10 mgoodwin: Why do you bother with this if that's the case
17:10 karolherbst: though intel iris is also a choice
17:10 imirkin_: mgoodwin: because it's fun
17:10 mgoodwin: What's the motivation
17:10 imirkin_: i like fun.
17:11 imirkin_: which is also why i don't worry about plasma not working
17:11 imirkin_: i don't use it, none of the developers appear to care (or at if they do, they haven't expressed such caring)
17:12 imirkin_: none of the plasma developers, that is
17:12 karolherbst: well I cared, but even running plasmashell via DRI_PRIME I didn't get any crashes
17:12 karolherbst: so :/
17:12 karolherbst: no idea if that is helpfull or not
17:12 imirkin_: i don't use kodi, and the thing it does totally doesn't work with nouveau.
17:12 mgoodwin: Maybe i'll mail you this stupid card if I get rid of it
17:12 imirkin_: i use mplayer, which works *great* on everything
17:13 mgoodwin: It's certain to guarantee an adventure
17:13 imirkin_: (not mplayer2, not mpv, not mplayer-but-not-really...)
17:13 mgoodwin: Just another case of you being stuck in the past
17:13 imirkin_: ok
17:13 imirkin_: i'm happy there :)
17:14 mgoodwin: Your users aren't
17:14 imirkin_: i don't have users
17:14 mgoodwin: Yeah you do, whether you like that or not
17:14 imirkin_: then where are they? why aren't they helping?
17:14 mgoodwin: Because they're users, and they... use
17:14 imirkin_: yeah, and most of them use just fine
17:15 imirkin_: and some don't and come in to complain
17:15 imirkin_: very few come in to say they're highly satisfied, even if they are
17:15 imirkin_: [but it does happen on rare occasions]
17:15 mgoodwin: I don't know a single person satisfied with nouveau
17:15 imirkin_: i think it's natural for you to assume that however you're using nouveau is the average use-case. but i think it's demonstrably not.
17:16 imirkin_: do you know a single person who uses nouveau?
17:16 mgoodwin: What you have is a group of people that use your driver just long enough to get the binary driver
17:16 mgoodwin: Yes, many I have a private linux channel
17:16 Riastradh: I'm satisfied with it on Linux. Not going near the binary driver with a 20-foot pole.
17:16 mgoodwin: The IE of drivers
17:17 imirkin_: so basically you have a bunch of people who don't contribute anything but their dissatisfaction
17:17 imirkin_: and you're expecting me, a volunteer who works on the project for fun, to care?
17:17 Riastradh: (Still working on it in NetBSD, though between i915, radeon, and nouveau, nouveau has been the easiest to get working, if lowest-priority.)
17:17 mgoodwin: So do you know what I'm going to say now? Lead nouveau dev says to get AMD and stay away from nouveau and nvidia
17:17 imirkin_: mgoodwin: feel free to quote me on that.
17:17 mgoodwin: I will
17:17 Riastradh: NVIDIA certainly doesn't seem very friendly to me.
17:17 imirkin_: mgoodwin: in fairness, i recommend intel over amd
17:17 mgoodwin: not ironically either
17:18 imirkin_: mgoodwin: but intel's range of gpu's ends at a fairly low point. if you want a higher end gpu (or add-in), definitely go with amd over nvidia
17:18 Riastradh: But I happen to have some NVIDIA hardware -- not because I chose to give money to NVIDIA but because I acquired it incidentally -- and it would be nice to take advantage of it.
17:19 mgoodwin: Also, lol, I just caught that quip
17:20 mgoodwin: How on earth am I not using nouveau 'averagely'
17:20 imirkin_: i dunno. but you seem to be hitting a ton more issues than most users who come in here with questions
17:20 mgoodwin: I want to browse some sites with firefox and watch some movies with a popular open source media center
17:20 imirkin_: yeah. kodi is just not supported. sorry.
17:20 imirkin_: or rather, disable vdpau. or GL.
17:21 imirkin_: i think if you use just one of those at a time, it should be fine.
17:22 mgoodwin: you're right though it does seem to be the interaction of both at the same time
17:22 imirkin_: nouveau doesn't interlock multi-threaded draws
17:22 imirkin_: (which is, ultimately, a bug in nouveau, albeit difficult to fix)
17:23 mgoodwin: Unfortunately it seems that what we have here is a driver made from fun instead of what's necessary
17:23 imirkin_: pretty much
17:23 mgoodwin: So i can imagine you just work on what you like or want to instead of what is hard or necessary
17:23 mgoodwin: hence the state of affairs
17:24 mgoodwin: I have projects like that too
17:24 Riastradh: A driver made by volunteers reverse-engineering the hardware of a secretive adversarial company, sure.
17:24 imirkin_: i do fix a lot of reported bugs
17:24 imirkin_: but they tend to be the easy-to-debug ones
17:24 Riastradh: I'm pretty impressed by the results, really, and I appreciate the work they've done.
17:24 mupuf: mgoodwin: this is a lot of work for very little benefit
17:24 mupuf: and it may regress the performance of all the other cases
17:24 mupuf: so...
17:24 imirkin_: nah, i really do feel bad about the multi-threaded thing
17:24 imirkin_: it's a totally legitimate use-case
17:25 imirkin_: i did try to address it at one point, but that ended in total failure
17:25 mgoodwin: I can't help though but think that if support was utterly dropped a la truecrypt style maybe a better team would pick it up
17:25 mgoodwin: Or it would force nvidia to do something because they would lose market share
17:25 imirkin_: i'd probably need a full week of full-time effort to work it all out
17:25 mgoodwin: This is the case of the 'good enough' band aid never being replaced
17:26 imirkin_: seems doubtful. there's a pretty steep learning curve, and most of the information is in people's heads
17:27 Riastradh: mgoodwin: Maybe if you offered constructive contributions the situation would improve. It is hard to imagine that entitled whining at volunteers will accomplish anything.
17:27 imirkin_: Riastradh: amusement for me :) and probably relief for mgoodwin.
17:27 mgoodwin: Take it behind the barn and kill it to force various actors to do something better etc
17:28 mgoodwin: even better
17:28 imirkin_: mgoodwin: anyways, feel free to put together your own team of volunteers and write a better driver. i'd even help them if they wanted.
17:29 mgoodwin: I can't do that unless you disappear the project as well as your domain knowledge
17:29 mgoodwin: Otherwise people will keep saying "nouveau is works kind good enough"
17:31 karolherbst: mgoodwin: and you would need a lot of time anyways
17:31 mgoodwin: Far more interested to see what happens if this project doesn't exist than wait out and 'fun' fixes from people that care more about cuda processing and tomb raider benchmarks rather than general desktop computing
17:31 mgoodwin: any*
17:31 karolherbst: well
17:31 karolherbst: I have a optimus setup
17:31 mupuf: imirkin_: don't get any ideas :p
17:31 karolherbst: so I usually only care about GL performance and kernel stuff :O
17:32 karolherbst: mgoodwin: well the point is, nobody really gets money for doing nouveau stuff (well except one) and this is the situation
17:32 mgoodwin: Exactly
17:33 karolherbst: mgoodwin: self motivation is like the only reason to do this. If you get bored with nouveau you just leave, nobody here has to stay or has to do anything
17:33 karolherbst: and everbody is free to help if they want to
17:33 mgoodwin: That's precisely my point
17:33 mgoodwin: Why does AMD have a team of developers for radeon?
17:33 karolherbst: yeah, but what is the conclusion from this?
17:33 mgoodwin: Probably because they want to gain what little market share they can from Nvidia
17:33 karolherbst: mgoodwin: because they pay?
17:34 mgoodwin: The conclusion is that nothing is going to improve with nivdia on linux while nouveau still exists
17:34 mgoodwin: because nvidia is not properly motivated to pay for it
17:34 mgoodwin: It's that simple
17:34 karolherbst: mgoodwin: and how you get there?
17:34 karolherbst: well why should nvidia care?
17:34 mgoodwin: Kill the project and wait months for people to be pissed at nvidia
17:34 karolherbst: and why would it change anything if nouveau wouldn't exist?
17:34 mupuf: mgoodwin: nvidia said VESA was good-enough to download the driver and install it
17:35 karolherbst: well
17:35 karolherbst: now we have efifb
17:35 mupuf: so, nvidia will never care, as long as VESA works
17:35 karolherbst: efifb is plenty enough
17:35 mupuf: right
17:35 karolherbst: efifb is quite awesome by the way
17:35 mjg59: mgoodwin: Radeon was only supported with reverse-engineered drivers between 2001 and 2007. And then AMD decided to add an engineering team to it.
17:35 karolherbst: native resolution support
17:35 Riastradh: mgoodwin: nouveau is what happened when people got pissed off at NVIDIA.
17:35 karolherbst: Riastradh: right :D
17:35 karolherbst: well
17:35 karolherbst: they want to help though
17:36 mjg59: mgoodwin: So the idea that a "good enough" driver is enough to prevent a company from deciding to put engineering effort on something is provably false
17:36 mupuf: Riastradh: yep, that was the reason indeed
17:36 karolherbst: it isn't as bad as mayn think, but they don't get money for supporting nouveau
17:36 karolherbst: and so, why should they spend money on nouveau?
17:36 mgoodwin: Then it was
17:36 karolherbst: usually they can't help due to IP issues
17:36 mgoodwin: Now is different with valve
17:36 karolherbst: and because it requires ressources on their end
17:36 mupuf: karolherbst: well, they need their code upstream in the kernel
17:36 mupuf: so, they need a userspace for it
17:36 mupuf: an open source one
17:36 karolherbst: tegra?
17:37 mupuf: that is the only reason why they are investing in nouveau now, AFAICT
17:37 mupuf: yes, for tegra
17:37 karolherbst: right
17:37 karolherbst: but why do they need an open userspace?
17:37 mupuf: "someone with a lot of money" asked for it
17:37 mupuf: karolherbst: rule of the kernel
17:37 karolherbst: :D
17:37 mjg59: mgoodwin: Uh you're going to have to explain the logic for that one
17:38 karolherbst: mupuf: well I thought in userspace you can do whatever you want
17:38 mupuf: karolherbst: you can, but an open source userspace is necessary to land a new interface
17:39 mupuf: otherwise, you can never reasonably maintain it
17:41 karolherbst: right
17:48 mgoodwin: My logic is that I think the project would be better off without a prick lead who keeps telling anyone that doesn't agree with the way he uses a computer to do less with it or go buy amd
17:49 mgoodwin: Clearly that's ineffective, plus he only does fun things
17:49 imirkin_: on the bright side, i don't prevent people from doing things i don't find fun
17:49 mgoodwin: Passively perhaps, yeah
17:52 p3rs3us: hey guys, I am getting segfault while attempting to run an application that uses OpenGL. I am using mesa version 11.1.2. A log of the segfault taken via dmesg is in http://pastebin.com/gG3vuYGA I know it doesn't say much but you can let me know if you spot something.
17:52 imirkin_: p3rs3us: remind me what GPU you have?
17:52 p3rs3us: imirkin, i could get version 11.1.2 from debian backports
17:52 mgoodwin: Can't use opengl + novueau it's not supported
17:52 mgoodwin: Buy amd
17:52 p3rs3us: imirkin, one sc
17:52 imirkin_: mgoodwin: you want to troll me, that's fine - please don't troll actual users.
17:53 mgoodwin: Like you did me?
17:53 p3rs3us: imirkin, GeForce GTX 770
17:53 imirkin_: mgoodwin: i didn't troll you, i stated specific facts. you perhaps misunderstood what was being said, or are over-generalizing.
17:54 imirkin_: p3rs3us: and what kernel are you on? looks like pre-4.3?
17:54 p3rs3us: imirkin, 3.16.0-4-amd64
17:54 p3rs3us: imirkin, yep
17:55 imirkin_: p3rs3us: oh hm... what's "liger" btw?
17:55 p3rs3us: imirkin, its an application that has been developed in my Uni .. like an interface
17:55 p3rs3us: imirkin, they used Qt for it
17:55 imirkin_: http://codem.group.shef.ac.uk/index.php/liger - this guy?
17:56 p3rs3us: imirkin, and after they migrate to Qt 5.6 I started getting segfaults
17:56 p3rs3us: imirkin_, yes
17:56 imirkin_: hm, interesting.
17:57 imirkin_: well, looking it over, there's maybe one potentially relevant fix in 11.1.3. however there have also been a lot of kernel fixes in the past year+ since 3.16 has been released
17:57 p3rs3us: imirkin_, the thing is we are using javascript libraries for showing up some graphs and that is where opengl gets called
17:58 imirkin_: i'd highly recommend updating to kernel 4.5 and mesa 11.2.1 and seeing if issues persist
17:58 mgoodwin: If you're not bleeding edge, you're wrong! (but don't use a compositor)
17:58 p3rs3us: imirkin, so at the moment i got 11.2.1 from backport of debian .. the one way to get higher is compile the source and so on
17:59 imirkin_: p3rs3us: oh, 11.2.1 should be fine. i thought you said 11.1.2
17:59 p3rs3us: imirkin_, sorry its 11.1.2
17:59 imirkin_: mgoodwin: plenty of people use a compositor successfully. plasma does appear to trigger extra-special problems that have not been diagnosed though.
18:00 mgoodwin: Surely, but your offer to me yesterday was to use windowmaker and make sure not to use a compositor because that's how you do it and you have no issues (of course you dont)
18:01 imirkin_: mgoodwin: doubtful. more probably i mentioned what i used and that i had no issues.
18:01 imirkin_: mgoodwin: rather than prescribing that you should do the same thing
18:04 karolherbst: noo, another issue :/
18:05 karolherbst: "gr: GPC0/TPC0/TEX: 80000049"
18:05 mgoodwin: Why are you making it a point to game on nouveau when it will never have any of the game-ready optimizations
18:06 mgoodwin: Never saw nouveau as being a contender if you wanted to actually game because of that - like saying "Yes please I would love to lose 30% in performance"
18:06 imirkin_: karolherbst: followed by a pte error of some sort?
18:06 karolherbst: well I get above 65% performance compared to nvidia with tomb raider, it is more actually, but never benchmarked carefully enough
18:06 karolherbst: imirkin_: fifo: read fault at 00058cb000 engine 00 [GR] client 01 [GPC0/T1_0] reason 02 [PTE] on channel 2 [00bf890000 TombRaider[6278]]
18:06 imirkin_: mgoodwin: what's your goal here?
18:07 mgoodwin: blow off some steam and eventually leave
18:07 karolherbst: mgoodwin: 30% perf less isn't much
18:07 imirkin_: karolherbst: sounds like we're trying to access a texture that's not mapped anymore? grrr...
18:07 mgoodwin: It's a genuine question though
18:07 karolherbst: ohh
18:07 karolherbst: vertex issues on lower quality settings
18:07 mgoodwin: I've been wondering it since he started tomb raider yesterday
18:07 Tom^: mgoodwin: why do you think there is a point?
18:07 Tom^: mgoodwin: he is having fun.
18:08 p3rs3us: mgoodwin, why don't you become a coder of nouveau and make it happen. What is it stopping you?
18:08 imirkin_: mgoodwin: i think people don't look at it as "30% less perf" but rather "a ton more perf compared to software rendering"
18:08 mgoodwin: $$$
18:09 mgoodwin: If you want to play games you don't use nouveau... everyone knows that
18:09 imirkin_: apparently some people didn't get the memo
18:09 p3rs3us: mgoodwin, i play nexuiz .. it plays perfectly well
18:10 karolherbst: imirkin_: not reproducable :/
18:10 karolherbst: well I play some games with nouveau
18:10 imirkin_: karolherbst: i wonder if there's still some weird shenanigans in the ctxsw logic =/
18:10 p3rs3us: karolherbst, and you had fun :)
18:10 mjg59: mgoodwin: Clearly there is nobody who (a) can do something about the problems you're facing and (b) cares enough to do so
18:11 mgoodwin: Clearly
18:11 pmoreau: imirkin_: You’re completely right!! Going to change that immediately. Thanks!
18:11 mjg59: mgoodwin: They're not suddenly going to start existing if novueau ceased development
18:11 mgoodwin: suddenly? no
18:12 karolherbst: well
18:12 mjg59: Meanwhile a large number of people whose needs *are* met by nouveau would be greatly inconvenienced
18:12 mgoodwin: i have zero hope of ever getting this hardware to work right, so I don't need suddenly
18:13 karolherbst: mgoodwin: I don't think you quite get that nouveau is getting competitive performance wise for some workloads
18:13 imirkin_: pmoreau: if you could kick off a build too, that'd be awesome - i want to point one of the recent bug filers at it
18:13 mgoodwin: "for the greater good" mjg59
18:13 pmoreau: imirkin_: Yup, going to kick one. :-)
18:13 mjg59: mgoodwin: For the promise of a greater good that probably doesn't exist
18:13 imirkin_: pmoreau: that's how i noticed that it seemed wrong :)
18:13 pmoreau: Hehehe! I was wondering about that :-D
18:13 mjg59: mgoodwin: There's no reason to believe you wouldn't just get an identical set of people working on a replacement
18:13 mgoodwin: karolherbst: Maybe? I've never viewed it in that light but it will still never have the game specific optimizations
18:14 Tom^: mgoodwin: and?
18:14 mgoodwin: There are two markets where people will always choose performance over open source: gaming and number-crunching
18:14 Tom^: you are missing the bigger picture, most of us spend our free time on doing things we enjoy doing. be it coding or driving snowmobiles.
18:14 imirkin_: i think it's fair to say that a project backed by a handful of part-timers without hw docs are never going to outperform a team of hundreds of full-timers with hw docs
18:14 Tom^: and coding drivers is simply just fun for some :p
18:15 karolherbst: right
18:15 karolherbst: and some prefer open source drivers over closed ones
18:16 karolherbst: in the end it comes down to what somebody wants to give up for using an open driver
18:16 Tom^: its all just a way to spend your time doing creative things ;)
18:16 karolherbst: and if it is just 20% or 30% of performance, the choice is easier, than having no OpenGL with an open driver
18:17 mgoodwin: That's exactly my point though
18:17 mgoodwin: I gave up gaming in general to use linux
18:17 mgoodwin: Turns out some of that's returning
18:17 p3rs3us: mgoodwin, are you in attention seeking mode?
18:18 mgoodwin: yes
18:19 karolherbst: mgoodwin: well wine also supports d3d10 and d3d11 games much better now
18:19 mgoodwin: Not at all yet iirc
18:19 karolherbst: since 1.9.8 they do
18:19 mgoodwin: oh?
18:19 karolherbst: TYPELESS formats were added
18:19 karolherbst: which are used by like every d3d10 and d3d11 game
18:20 p3rs3us: mgoodwin, i have some tinder ugly girls messaging me every 5 min .. i was wondering if you know them
18:20 mgoodwin: I don't do social media
18:20 mgoodwin: Sorry
18:20 p3rs3us: mgoodwin, it is OK. I was thinking they went to school with you.
18:21 imirkin_: let's stay on topic... i'd hate to start having to kick people
18:21 pmoreau: imirkin_: Build started, and the entries should correctly link to github commits
18:21 imirkin_: pmoreau: cool thanks!
18:21 mgoodwin: If you kick me my client is going to auto-rejoin iirc
18:21 mgoodwin: So be sure to ban ahead of time
18:22 imirkin_: duly noted.
18:22 mgoodwin: karolherbst: source on the wine thing?
18:22 mgoodwin: I feel like I would have heard if wine suddenly started supporting DX10,11
18:22 karolherbst: mgoodwin: https://bugs.winehq.org/show_bug.cgi?id=22762
18:22 karolherbst: but there are more games
18:22 karolherbst: many more
18:22 karolherbst: I hit this issue with like 4 games of mine and then I stopped trying
18:23 karolherbst: full d3d10 support is pretty near
18:23 karolherbst: and if you have d3d10, d3d11 isn't far away
18:23 mgoodwin: I don't even recall getting d3d10 installed properly via winetricks
18:23 mgoodwin: is that fixed now
18:23 karolherbst: well you don't need to install directx
18:23 karolherbst: it is just a workaround
18:23 Tom^: installing via winetricks is not what you were supposed to do
18:24 karolherbst: 27 fps :) https://i.imgur.com/RP6GbnN.png
18:24 Tom^: that just overrides the wine implentations in favour of trying to load the dlls themself with the other uh windows implentations
18:24 mgoodwin: It's not like there's a set path
18:24 mgoodwin: Everyone is mixing and matching shit to see what works
18:25 mgoodwin: argh I remember that level
18:25 mgoodwin: It gets so repetitive after that
18:26 mgoodwin:wonders if his copy of tombraider is upgraded to support linux now
18:27 mgoodwin: Let's see how quickly I can crash this before I go back to binary land
18:28 mezo: is there a way to force vsync on a game?
18:29 mgoodwin: karolherbst: whats your hardware on that laptop
18:31 mgoodwin: also what dx10 game did you get working with wine?
18:33 karolherbst: mezo: yeah, I think vblank_mode=3? I never know what number to put there
18:33 karolherbst: mgoodwin: 770m
18:33 karolherbst: mgoodwin: I didn't try yet, though
18:37 mgoodwin: welp hold onto your pants
18:37 mgoodwin:start tomb raider
18:37 imirkin_: make sure to disable tressfx - i think karol said that triggers crashes
18:46 mgoodwin: So far it's pretty playable for me on high
18:46 mgoodwin: I didn't touch or disable anything
18:46 mgoodwin: I am however using karolherbst's driver branch
18:46 mgoodwin: Don't know if that means it's reclocking
18:47 mgoodwin: So strange to me that a game like this works yet the desktop is utterly broken
18:49 hakzsam_: yeah, tressFx hangs the gpu :/
18:52 karolherbst: mgoodwin: well did you boot with something special or set the pstate to something?
18:52 karolherbst: mgoodwin: and what gpu do you have?
18:52 karolherbst: hakzsam_: I got a random crash in 1.5h playtime with nouveau
18:53 karolherbst: hakzsam_: could be the same as the tressfx issue though
18:53 hakzsam_: :/
18:53 karolherbst: but I also got this "gr: GPC0/TPC0/TEX: 80000049"
18:53 karolherbst: hakzsam_: well the crash was at a save point and the save did count, so I didn't mind about that one :D
18:53 hakzsam_: this sounds familiar :)
18:54 karolherbst: hang at fire place?
18:54 karolherbst: happend only once though
18:54 karolherbst: and I couldn't reproduce
18:55 karolherbst: mgoodwin: well, running one thing at the time should be pretty solid, having multiple things at once, isn't as much :/
18:55 mgoodwin: i boot to pstate 10 EVGA 660 SC
18:55 karolherbst: you might want to change to 0f
18:55 karolherbst: this should give you the real performance boost
18:56 karolherbst: well 15 with the boot parameter
18:56 karolherbst: 0f with the pstate file
18:56 karolherbst: should get you more than doubled performance
18:57 karolherbst: let me check how much it makes of a difference for me
18:58 mgoodwin: So it doesn't autoreclock yet?
18:59 karolherbst: mhh, 27-> 38 FPS in main menu for me
18:59 karolherbst: and only memory clock changed from 1620 tt 4008
18:59 karolherbst: mgoodwin: of course not. Before we can think of that, reclocking itself needs to be far more stable
19:00 imirkin_: mgoodwin: there's a certain (low) chance the gpu will hang on a reclock, so it's not something we can do automatically at this point.
19:01 mgoodwin: did you change the /sys/class path :/
19:01 karolherbst: mgoodwin: yeah, it is in debugfs now
19:01 karolherbst: /sys/kernel/debug/dri/
19:01 mgoodwin: !
19:01 imirkin_: mgoodwin: this is covered on the front page of the wiki - https://nouveau.freedesktop.org/wiki/
19:02 karolherbst: and you don't need the pstate boot parameter anymore
19:02 mgoodwin: That might seem like the best place for it but this is where I think to go to for pstate stuff: https://nouveau.freedesktop.org/wiki/KernelModuleParameters/
19:03 mgoodwin: and also the top search
19:03 karolherbst: yeah, this needs to be updated :D
19:03 karolherbst: mgoodwin: well it can also happen that you get screen coruptions on reclock
19:03 imirkin_: mgoodwin: ah right, thanks for the reminder. forgot about that page.
19:03 karolherbst: mgoodwin: but for this you can just reclock back to something else
19:03 karolherbst: mgoodwin: and this fixes it usually
19:03 imirkin_: [even though i added it]
19:03 mgoodwin: We're going to find out shortly
19:04 imirkin_: karolherbst: /sys/kernel/debug/dri/*/pstate - right?
19:05 mgoodwin: what are the others adjacent to 0
19:05 mgoodwin: I have 0, 128, 64
19:05 imirkin_: it's one for each drm node... card0, controlC64, renderD128
19:05 imirkin_: they're all the same underlying gpu though (in all likelihood)
19:06 karolherbst: imirkin_: yes
19:06 imirkin_: page updated.
19:06 mgoodwin: so setting the pstate for one is sufficient?
19:06 karolherbst: imirkin_: well the boot parameter doesn't exist though
19:06 imirkin_: yep
19:06 imirkin_: karolherbst: yes, i noted that
19:06 karolherbst: ahhh
19:06 karolherbst: yeah see it
19:07 mgoodwin: is there a real difference between 0d and 0f
19:08 mgoodwin: pstate shows them identical
19:08 RSpliet: mgoodwin: to verify you'd have to check the VBIOS
19:08 imirkin_: there are other domains that we don't show... probably *some* difference
19:09 karolherbst: mhh
19:09 karolherbst: yeah well, from my experience, usually there is none
19:09 karolherbst: some vbios have lower memory clocks (4GHz instead of 5GHz)
19:09 mgoodwin: it seems to have switched
19:09 RSpliet: karolherbst: that probably just means we haven't found the difference yet
19:09 karolherbst: RSpliet: yeah, maybe some engines are lower clocked
19:09 mgoodwin: screen flickered and cat shows * on 0f
19:10 karolherbst: yeah
19:10 karolherbst: that's normal
19:13 karolherbst: mgoodwin: and is the performance much better now?
19:14 mgoodwin: yes
19:14 karolherbst: if you are up to you can always increase the boost level too
19:14 karolherbst: but this might overheat your gpu
19:14 karolherbst: well when the gpu gets too hot it will just shut down anyway and refuse any work
19:15 mgoodwin: what was that thing about the benchmark not being a good test
19:15 karolherbst: I was sloppy running them
19:15 karolherbst: didn't exactly remember the numbers
19:15 karolherbst: but nvidia was significantly faster, but less then 50%
19:16 mgoodwin: what are you using for fps
19:16 mgoodwin: steam overlay?
19:16 karolherbst: which means nouveau is around 66% of the performance
19:16 karolherbst: mgoodwin: nope, ingame
19:16 mgoodwin: https://i.imgur.com/mx5zBUL.png
19:16 mgoodwin: switch?
19:16 mgoodwin: or ingame option
19:16 karolherbst: ingame option
19:16 karolherbst: well I disabled shadows, because they eat quite a lot of performance :/
19:17 karolherbst: but my gpu is also not fast enough for maxed out settings on nvidia anyway
19:18 karolherbst: and with nouveau every was maxed out except shadows, depth of field, post processing and object quality? no idea how that was called again
19:18 karolherbst: and antistropic filtering x16 has a really odd impact on perf, x8 has none, but x16 has something around 5%
19:21 mgoodwin: are you trolling me with the ingame option
19:21 mgoodwin: lol a little help :|
19:21 karolherbst: in the main menu
19:21 karolherbst: "start benchmark" :D
19:22 mgoodwin: Ohh I thought you meant a FPS overlay
19:22 mgoodwin: I did that too but I guess you have to wait for it to finish
19:24 mgoodwin: https://i.imgur.com/MVXFWY6.png
19:25 hakzsam: with nouveau? if so it's pretty nice :)
19:25 mgoodwin: agreed
19:25 karolherbst: mhh
19:25 karolherbst: more than I have :D
19:25 mgoodwin: Now if I could only play a video and work on my desktop at the same time
19:25 karolherbst: mgoodwin: everything maxed out?
19:25 mgoodwin: hell no
19:25 karolherbst: ohh right
19:25 karolherbst: no tressfx
19:25 karolherbst: well
19:25 hakzsam: imirkin_, well, it seems like that enabling PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS also fixes the TressFx issue with TB :)
19:25 mgoodwin: That was High
19:26 hakzsam: that's.. unexpected
19:26 karolherbst: well ssao is pretty intense
19:26 mgoodwin: but about the same settings I was using on windows when it came out
19:26 mgoodwin: playable is playable
19:26 karolherbst: right
19:27 mgoodwin: how do you turn off the compositor without logging out
19:27 mgoodwin: Never had to
19:27 mgoodwin: I guess just restart plasmashell
19:28 mgoodwin: kwin*
19:29 mgoodwin: Now let me turn off all this other stuff and set it to max
19:29 mgoodwin: but you're telling me no tesselation that will crash
19:31 karolherbst: nope
19:32 karolherbst: mgoodwin: no TressFX
19:32 karolherbst: mgoodwin: that's the fancy hair thing
19:32 karolherbst: mgoodwin: and no postprocessing + depth of field ultra
19:32 karolherbst: either alone is fine, but not both
19:32 imirkin_: hakzsam: can you grab the shader that fails to link?
19:32 imirkin_: hakzsam: or rather, programs i guess
19:32 hakzsam: imirkin_, for which issue?
19:32 imirkin_: the one where you're hitting a linker error
19:33 mgoodwin: https://i.imgur.com/xtqC2Gh.png
19:33 mgoodwin: This was ultra -tesselation
19:33 karolherbst: mgoodwin: shadows on max is something
19:33 karolherbst: mgoodwin: and 4x ssao
19:33 karolherbst: mgoodwin: everything else has nearly no impact
19:33 karolherbst: well depth of field ultra has a bit too
19:33 karolherbst: ..
19:33 karolherbst: I mean 4x ssaa
19:34 karolherbst: amd sao on ultra
19:34 karolherbst: hakzsam: you are right :/ with your patches the trace still hangs the gpu
19:34 gouchi: sorry to bother you I was wondering if somebody has time to make a little test http://www.hastebin.com/geyawicavu.hs
19:34 karolherbst: hakzsam: but without I couldn't enable tressfx ingame, with I can
19:35 hakzsam: karolherbst, that makes more sense
19:35 hakzsam: it's more likely unrelated
19:35 karolherbst: okay
19:35 gouchi: if it's crashed or not
19:35 mgoodwin: I can't tell if there's vsync on somewhere or the unlikely chance it's maxing out at 60.7 fps
19:36 karolherbst: hakzsam: okay, seems like tressfx is completly broken though
19:36 mgoodwin: Sorry that was unclear, TressFX = tesselation?
19:36 imirkin_: mgoodwin: by default, unless applications ask for non-vsync, they're vsync'd. you can force-disable this by sticking enable_vblank=0 in the environment.
19:36 mgoodwin: in the settings?
19:36 karolherbst: mgoodwin: nope
19:36 karolherbst: mgoodwin: hair quality
19:37 mgoodwin: imirkin_: hard coded huh?
19:37 hakzsam: karolherbst, yeah it is, but I don't care about that :)
19:37 hakzsam: imirkin_, should be this one http://hastebin.com/ojosadazir.avrasm
19:37 imirkin_: (they can ask for non-vsync by ... i forget. some glx thing.)
19:37 imirkin_: hmmmm
19:37 imirkin_: lame.
19:37 imirkin_: that's ken's thing :(
19:37 karolherbst: this is so insane.. disabling TressFX with nvidia: 19 -> 34fps
19:37 hakzsam: imirkin_, tomb raider shaders are just crazy btw :)
19:38 imirkin_: hakzsam: it's the lower_const_to_uniform pass
19:38 imirkin_: (or something like that)
19:38 karolherbst: hakzsam: why are they crazy? :D
19:38 imirkin_: hakzsam: lower_const_arrays_to_uniforms
19:38 imirkin_: basically it sticks that icb array into a uniform, and references it from there
19:39 hakzsam: will have a look at that pass
19:39 imirkin_: buuuttt... it's a huge array... 327 * 4
19:39 imirkin_: which is 5k
19:39 mgoodwin: Ah that's why im lost
19:39 imirkin_: which is well shy of 16k. weird.
19:39 mgoodwin: Hair Quality is grayed out for me
19:39 mgoodwin: must not be a gpu feature?
19:39 karolherbst: mgoodwin: ohh right
19:39 karolherbst: no
19:39 karolherbst: compute_shaders
19:39 imirkin_: mgoodwin: you'd need to use mesa from git for the various GL 4.3 features
19:39 karolherbst: it was just enabled like a few days ago
19:40 hakzsam: imirkin_, mmh
19:40 imirkin_: hakzsam: oh, but i bet every shader in that program is like that
19:40 mgoodwin: :/ lol I just built mesa
19:40 karolherbst: mgoodwin: when?
19:40 hakzsam: imirkin_, it seems like yes
19:40 karolherbst: mgoodwin: yesterday? :D
19:40 mgoodwin: But I did the 11.2.1 release
19:40 imirkin_: hakzsam: so if there are 5 stages, it's 5 * 5k = 25k
19:40 mgoodwin: this morning
19:41 karolherbst: wut
19:41 hakzsam: imirkin_, 5 stages, you mean, vert, frag, comp etc?
19:41 karolherbst: softpipe has compute shadersupport:D
19:41 imirkin_: vert, tessc, tesse, gs, frag
19:41 hakzsam: okay
19:41 imirkin_: can't have compute and graphics shaders in the same pipeline
19:41 hakzsam: sure, makes sense
19:42 hakzsam: but I don't see any tesselation shaders in the trace
19:43 imirkin_: man... such a dumb shader
19:43 imirkin_: i dunno how we can do a good job on it, sadly
19:44 imirkin_: without wildly complicating matters
19:44 imirkin_: are uniforms per stage, or are they per program?
19:44 EdB_: imirkin_: thanks, I was puzzled
19:45 hakzsam: imirkin_, how can I check that?
19:45 imirkin_: EdB_: sorry, i lost context on our conversation
19:45 mgoodwin:facepalm
19:45 imirkin_: EdB_: oh, the memory vs buffer thing?
19:45 mgoodwin: forgot to put export before the assignment
19:45 mgoodwin:logs out again
19:45 imirkin_: hakzsam: rtfs? (s = spec in this case)
19:46 hakzsam: :) okay
19:46 imirkin_: or we can observe that glGetUniformLocation takes a program argument, not a shader
19:46 imirkin_: so... it's per-program
19:46 imirkin_: whereas ideally this pass that ken wrote would only stick those uniforms for that one shader.
19:46 imirkin_: hakzsam: out of curiousity, could you dump the whole program? or is it all SSO'd?
19:47 hakzsam: seems like it's per-program
19:47 hakzsam: according to glGetUniformLocation()
19:49 hakzsam: imirkin_, SSO'?
19:49 imirkin_: separable shader objects
19:50 mgoodwin: seems like I also need vblank_mode=0
19:50 hakzsam: imirkin_, so, there are a bunch of different programs
19:50 imirkin_: hakzsam: each shader stage is in its own program, and then they're linked together later
19:50 imirkin_: rather than the thing where a single program has multiple stages inside of it
19:50 EdB_: imirkin_: that is :)
19:52 hakzsam: imirkin_, oh okay
19:52 imirkin_: hakzsam: basically you create a pipeline object and then bind the SSO programs to it... or something along those lines
19:52 imirkin_: i forget the exact details
20:05 hakzsam: yeah, and this has to be done before the program is linked
20:05 hakzsam: for the separable shader object thing
20:07 imirkin_: anyways... the deal is that we need to flip that pass to stick uniforms into only the relevant stages
20:07 imirkin_: dunno if the mesa internal frameworks support that particularly well
20:07 imirkin_: disabling the link-time warning just lets it proceed, but it will execute incorrectly
20:07 imirkin_: since ultimately the constbuf will be limited to 16k
20:07 hakzsam: I know that it's limited to 16k
20:08 hakzsam: and it's not a fix actually :)
20:08 imirkin_: some old nouveau code had support for an "immediates" constbuf
20:08 hakzsam: but something has to be improved around there
20:08 imirkin_: perhaps it was intended for precisely this thing
20:08 imirkin_: (search around for immd_size, etc)
20:08 imirkin_: i think it was a thing that calim was thinking of adding but never went through with it
20:09 imirkin_: it might be nice, instead of sticking these things into uniforms, to stick them into some shader-private area. dunno.
20:09 imirkin_: that may not "jive" nicely with the rest of mesa's internal logic though =/
20:09 mgoodwin: https://imgur.com/a/BECqh @ karolherbst
20:10 hakzsam: imirkin_, yeah, it doesn't even use this thing from the codegen part, so I guess this just doesn't work
20:10 imirkin_: hakzsam: not as-is, but ... it's an idea.
20:10 hakzsam: prog->immd_size
20:10 hakzsam: sure I get the idea
20:12 hakzsam: imirkin_, it's limited to 16k because it's the max size of a constbuf, right?
20:13 imirkin_: hakzsam: yeah, can only address up to 16k from within the shader
20:13 imirkin_: er wait
20:13 imirkin_: 64k!
20:13 imirkin_: it's blowing through 64kb in these constants???
20:13 hakzsam: NVC0_CB_USR_SIZE is 6 << 16
20:13 hakzsam: so yeah 64k per shader stage
20:14 imirkin_: so something's going wrong somewhere
20:14 imirkin_: no way they have 64kb of constants
20:14 imirkin_: and there's only a handful of other uniforms
20:14 hakzsam: it's the just the area size inside the driver cb, so why do we limit to 16k?
20:15 imirkin_: we don't - we limit it to 64kb
20:15 imirkin_: of course that's 16k ints/floats
20:15 hakzsam: oh okay
20:15 imirkin_: or 4k vec4's
20:15 hakzsam: makes sense :)
20:16 hakzsam: so, we could bump to 128K and hope it works :)
20:16 imirkin_: no, 64k is the addressable limit
20:16 hakzsam: but that's definitely not a correct solution
20:16 imirkin_: from within the shader
20:16 hakzsam: right
20:16 hakzsam: mmh
20:19 hakzsam: imirkin_, so, if 64K is the addressable limit inside a shader we have to load the uniforms from global mem? like what we do for compute UBO's on kepler?
20:20 hakzsam: or maybe we can reserver two constbufs for user uniforms, but that seems crazy
20:20 hakzsam: yeah, this won't work
20:21 hakzsam: oh no, immd_size ues c14 actually
20:22 imirkin_: right
20:22 imirkin_: so i think the deal is that every shader stage is getting the SAME set of (user) uniforms
20:22 imirkin_: which is an enormous waste to be uploading it 5x =/
20:22 hakzsam: so, the idea is to use one more constbuf for user uniforms?
20:22 karolherbst: mgoodwin: :D
20:22 karolherbst: mgoodwin: yeah, now you know what really needs performance :)
20:23 karolherbst: mgoodwin: but TressFX is broken with nouveau as it seems
20:26 hakzsam: imirkin_, are you sure that all constbufs are limited to 64k?
20:27 hakzsam: or maybe c14[] is a special one?
20:30 imirkin_: hakzsam: quite sure.
20:30 imirkin_: hakzsam: the ISA doesn't have more room to encode offsets :)
20:30 imirkin_: hakzsam: and the gpu will yell at you quite loudly if you try to set the constbuf size > 64k with CB_SIE
20:30 imirkin_: CB_SIZE that is
20:30 hakzsam: yeah okay
20:30 hakzsam: you are right
20:31 hakzsam: imirkin_, but this might be a serious limitation for recent games like tomb raider
20:31 hakzsam: so we need to find a solution :)
20:31 imirkin_: if you could make a piglit that repros this, would be hugely useful
20:32 imirkin_: the key is a large const array and indirect access into that array
20:32 hakzsam: make a piglit that uses more than 64k of user uniforms?
20:32 imirkin_: wellllll
20:32 imirkin_: no
20:32 imirkin_: i doubt that their shaders actually use that many uniforms
20:32 imirkin_: or rather, consts
20:32 imirkin_: that's the thing
20:32 pmoreau: constbuf are not paginated, are they?
20:32 imirkin_: they're not even uniforms
20:33 imirkin_: they're immediates in the shader
20:33 imirkin_: that we lower to uniform accesses
20:33 imirkin_: because otehrwise it's quite horrid code that has to be generated
20:33 imirkin_: (due to the indirect access)
20:34 hakzsam: yeah okay
20:35 hakzsam: like this shader http://hastebin.com/ojosadazir.avrasm a huge const array with indirect access :)
20:36 imirkin_: right... so try to figure out what's going on to cause the link error
20:36 imirkin_: it's a lot of consts, but... not 64k's worth
20:37 hakzsam: okay
20:47 Horrorcat: d’oh
20:47 Horrorcat: there goes karolherbst
20:51 Impyy: hi, is there any news on support for GM108 cards?
20:52 Horrorcat: I’m just now testing one
20:52 Horrorcat: people are working on it, i guess
20:57 imirkin_: Impyy: support was added to ben's repo
21:00 Impyy: imirkin_: where can I find ben's repo?
21:00 Impyy: also how soon could one expect that to be merged?
21:00 imirkin_: Impyy: https://github.com/skeggsb/nouveau/commits/master
21:00 imirkin_: i'm guessing it'll be included in kernel 4.7
21:01 imirkin_: Impyy: however note that at least one person reported issues with vram coming up, so ... expect issues :)
21:01 Impyy: alright, got some more waiting to do then
21:01 Impyy: nice to see it's being worked on though
21:01 Impyy: thanks :)
21:40 karolherbst: Horrorcat: and I am back :p
21:41 karolherbst: Impyy: well if you want to test reclocking on your gm108, I can give you some patches for that :)
21:41 karolherbst: just to verify it works in theory
21:41 karolherbst: I know it does for some maxwell cards (memory reclocking), but not sure if it will work for all
21:43 Horrorcat: wb karolherbst
21:43 Horrorcat: I have tried your branch
21:43 Horrorcat: i cannot get anything to render with it
21:44 Horrorcat: or, rather
21:44 Horrorcat: after I touched pstate in any way i cannot
21:44 karolherbst: Horrorcat: optimus?
21:44 Horrorcat: I... don’t know
21:44 Horrorcat: do you need context?
21:44 karolherbst: then you don't have optimus :D
21:44 karolherbst: Horrorcat: and what do you mean with you couldn't render anything with it
21:44 karolherbst: did your system froze?
21:45 Horrorcat: not the whole system
21:45 Horrorcat: glxgears and wine froze, in a state which did not allow me to SIGKILL them
21:45 karolherbst: k, but your desktop wasn't frozen
21:45 karolherbst: mhh
21:45 Horrorcat: then I did something™ (possibly writing a different value to pstate) which unfroze and killed them
21:45 karolherbst: do you have dmesg from that run?
21:45 karolherbst: ohh
21:45 karolherbst: k
21:45 Horrorcat: it was a few minutes ago
21:45 karolherbst: that may be just bad luck then
21:46 imirkin_: Horrorcat: when things freeze, it can take a few minutes for the driver to kill the processes
21:46 Horrorcat: mhm
21:46 imirkin_: it's difficult to detect the difference between a hung gpu and a very slow gpu batch
21:46 Horrorcat: in other news: I just modprobe -r nouveau; modprobe nouveau and let the pstate file alone. then glxgears renders with DRI_PRIME
21:46 karolherbst: ahh
21:46 karolherbst: optimus
21:47 Horrorcat: :)
21:47 Horrorcat: sorry i don’t know the terms
21:47 karolherbst: Horrorcat: well your desktop is rendered by intel
21:47 Horrorcat: yupp
21:47 karolherbst: right
21:47 karolherbst: that's usually called optimus these days
21:47 Horrorcat: i know what it is, but i don’t konw the term
21:47 imirkin_: karolherbst: i'm only aware of mobile GM108's
21:47 karolherbst: right
21:47 Horrorcat: I get a lot of those: Apr 28 23:43:55 sinistra kernel: nouveau 0000:02:00.0: fifo: SCHED_ERROR 08 []
21:47 Horrorcat: (in the "successful" glxgears run just now)
21:48 Horrorcat: and also two of these: Apr 28 23:43:55 sinistra kernel: nouveau 0000:02:00.0: fifo: write fault at 0000110000 engine 05 [BAR2] client 08 [HOST_CPU_NB] reason 0c [UNSUPPORTED_KIND] on channel -1 [007fd76000 unknown]
21:48 Horrorcat: where the first one had 0000158000 instead of 0000110000
21:48 karolherbst: Horrorcat: which branch did you use?
21:48 karolherbst: Horrorcat: the maxwell_reclocking one?
21:48 Horrorcat: yes
21:48 karolherbst: ohhh
21:48 karolherbst: right DDR3
21:48 Horrorcat: 36926942fa7af55013b73a6f1804196cbbc2a566
21:48 karolherbst: never tested the stuff on DDR3
21:48 Horrorcat: (^ commit sha1)
21:48 karolherbst: okay
21:48 Horrorcat: if you need anything more let me know
21:48 karolherbst: Horrorcat: can you switch to stable_kepler_reclocking_v5?
21:49 karolherbst: this hasn't the memory reclocking activation stuff for maxwell
21:49 Horrorcat: building ...
21:49 Horrorcat: is make clean in drm/ sufficient?
21:49 urmet: I think my gpu hung
21:49 urmet: http://pastebin.com/j22UEsU5
21:50 karolherbst: Horrorcat: you don't need to clean
21:50 Horrorcat: excellent
21:50 karolherbst: urmet: did you reclock?
21:50 karolherbst: ohh wait
21:50 Horrorcat: modprobe’d
21:50 urmet: no
21:50 karolherbst: gm206
21:51 karolherbst: Horrorcat: I hope higher pstate will be stable now
21:51 karolherbst: Horrorcat: but it won't give you good performance
21:51 karolherbst: Horrorcat: just maybe 30% more than stock clocks
21:51 karolherbst: Horrorcat: ohh glxgears will be boosted like hell though
21:51 karolherbst: Horrorcat: but that's because PCIe is the bottlebeck with prime
21:51 imirkin_: urmet: are you on a recent mesa? hrmph
21:51 karolherbst: Horrorcat: and we speed up the link on higher pstates
21:51 imirkin_: you must be because otherwise GM20x wouldn't be supported
21:52 urmet: mesa from git
21:52 Horrorcat: karolherbst: to be honest, I’m not doing this for the performance for me. I am happy with the intel, currently at least. I’d like to support the development with my testing though, as it seems that this hardware hasn’t received a lot of nouveau testing yet :-)
21:52 imirkin_: urmet: looks like a desync of some sort :(
21:52 karolherbst: Horrorcat: right
21:52 karolherbst: Horrorcat: but if the reclocking works now, this tells us that there is more magic with DDR3 as it seems
21:53 Horrorcat: as expected X3TC is slow, but menu renders
21:53 urmet: mouse cursor still moves :)
21:53 Horrorcat: no more odd error messages, but there’s this: https://paste.fedoraproject.org/360859/18804191/
21:53 karolherbst: Horrorcat: that means your card is going into suspend
21:54 karolherbst: like it nearly powers off
21:54 Horrorcat: that line was marked red by journalctl: Apr 28 23:52:39 sinistra kernel: nouveau 0000:02:00.0: priv: HUB0: 10ecc0 ffffffff (1b40822c)
21:54 karolherbst: to save power
21:54 karolherbst: yeah, well
21:54 karolherbst: may be important?
21:54 karolherbst: we don't know
21:54 Horrorcat: just thought I’d share that, in case this isn’t normal ;)
21:54 Horrorcat: excellent then
21:54 Horrorcat: I’ll lurk here, please ping me if I can do any testing
21:56 karolherbst: Horrorcat: well, I don't plan into looking into DDR3 reclocking on maxwell yet, because ... there aren't many GPUs and I don't have such one
21:56 karolherbst: and I don't think anyone of us has such
21:58 Horrorcat:didn’t even know GDDR5 was a thing until you mentioned it here
21:58 karolherbst: ...
21:58 karolherbst: well
21:58 karolherbst: GDDR5 is old news by now anyway :D
21:58 karolherbst: HBM comes now
21:58 Horrorcat: for me, opengl 3.3 is something new and shiny *hides*
21:59 RSpliet: pretty sure NVIDIA treats HBM like GDDR4
21:59 RSpliet: straight for the successor ;-)
21:59 urmet: so, is there anything i can try to get my video working?
22:05 karolherbst: Horrorcat: so with memory reclocking disabled it runs fine?
22:07 Horrorcat: yes
22:07 Horrorcat: although I only made one test
22:07 Horrorcat: and it’s slow
22:07 Horrorcat: I can do more tests if you think a single test might not give a reliable result
22:07 karolherbst: Horrorcat: well it will be faster than the stock clocks
22:08 karolherbst: but notmuch
22:08 karolherbst: memory speed is pretty important
22:08 Horrorcat: well, it’s slower than the intel
22:08 karolherbst: but good to know that at least engine reclocking is also kind of stable with my patches
22:08 karolherbst: Horrorcat: right
22:08 karolherbst: Horrorcat: because intel has much higher memoryspeed
22:08 karolherbst: :D
22:08 karolherbst: and to be honest
22:08 karolherbst: even with nvidia it will be a close call most likely
22:09 Horrorcat: if that is not an important datapoint, I’m not inclined to test with nvidia
22:09 urmet: got a slightly different dmesg this time http://pastebin.com/NF15phfv
22:09 karolherbst: I know, but in any case you will be hapier with intel in most cases
22:09 karolherbst: Horrorcat: what gpu do you have? 940M?
22:11 Horrorcat: 940 MX
22:11 karolherbst: Horrorcat: what intel gpu?
22:12 Horrorcat: OpenGL renderer string: Mesa DRI Intel(R) HD Graphics 530 (Skylake GT2)
22:12 karolherbst: :D
22:12 karolherbst: yeah well
22:12 karolherbst: that intel gpu is like half as fast as the nvidia one
22:12 karolherbst: so yeah...
22:13 Horrorcat: :-)
22:14 Horrorcat: well, I didn’t choose the device for the nvidia GPU. it just happened that all options had one.
22:14 karolherbst: :/
22:15 karolherbst: waste of money, seriously. I don't get why they _always_ put those slow GPU into the laptops
22:15 karolherbst: they should rather ship with an intel Iris one
22:15 karolherbst: less pain
22:15 Horrorcat: maybe
22:15 Horrorcat: I am not really up-to-date with GPUs, as you can imagine ;-)
22:15 karolherbst: Horrorcat: ohh do you know if all your video ports are connected to intel? :D
22:15 Horrorcat: is there a way to figure it out?
22:16 karolherbst: because sometimes, they wire them up with the nvidia one
22:16 karolherbst: and this is painful
22:16 Horrorcat: Provider 0: id: 0x4e cap: 0xb, Source Output, Sink Output, Sink Offload crtcs: 4 outputs: 11 associated providers: 0 name:Intel
22:16 Horrorcat: 11 matches with the number of outputs in xrandr
22:17 Horrorcat: and there is no other provider. is that sufficient?
22:17 karolherbst: Horrorcat: no, xrandr won't tell you without setting it up first
22:17 karolherbst: ls /sys/class/drm/
22:17 karolherbst: card0 is intel
22:17 karolherbst: card1 is nouveau
22:17 karolherbst: there might be more entries than ports
22:17 karolherbst: but you should be able to tell if nouveau has too many
22:17 Horrorcat: nouveau has just card1
22:18 karolherbst: Horrorcat: then you are lucky :D
22:18 RSpliet: karolherbst: they add these midrange GPUs to Intel laptops so that nouveau devs have a cheap way of reverse engineering the GPU without losing display
22:18 Horrorcat: intel has card0-{DP-1,DP2,...}
22:18 karolherbst: RSpliet: :D
22:18 karolherbst: Horrorcat: for me nouveau has "card1-VGA-1"
22:18 Horrorcat: RSpliet: :D
22:18 karolherbst: Horrorcat: but there is no port on my nvidia one
22:18 karolherbst: but I also have three HDMI entries for intel
22:18 karolherbst: :D
22:18 karolherbst: I am still looking for the other two ports
22:19 Horrorcat: do you have a docking port?
22:19 Horrorcat: maybe they’re wired there
22:19 karolherbst: could be
22:19 karolherbst: but I am sure I have none
22:19 karolherbst: well
22:19 karolherbst: sometimes there are more
22:20 karolherbst: and sometimes they also add entries for the DP port
22:20 hakzsam: karolherbst, so well, I figured out the white issue with Tomb Raider, should be fixed by the intel team :)
22:20 karolherbst: :D
22:20 hakzsam: karolherbst, and this might also fix the TressFx thing
22:20 karolherbst: awesome
22:20 karolherbst: how long do I have to wait for a fix?
22:20 hakzsam: no clue
22:21 hakzsam: I tried to make a fix
22:21 hakzsam: but I don't know so much this glsl compiler
22:21 karolherbst: ahh okay
22:21 hakzsam: so, I would prefer to assign the bug to someone else which knows this area
22:21 karolherbst: well it doesn't start with intel anyway I think :/
22:21 karolherbst: let me check
22:21 hakzsam: this affects all drivers
22:22 karolherbst: I heard
22:22 karolherbst: well when I start the game with intel it is all black :D and Iam sure those traces will also be black. But I am also sure there will be some tests written for that issue?
22:23 hakzsam: not sure
22:24 hakzsam: the issue only appears if the total size of uniforms is greater than 64k
22:24 hakzsam: pretty rare
22:24 hakzsam: but if you declare a huge constant array with a bunch of references, you will even get more than that
22:24 hakzsam: pretty easy to reproduce :)
22:25 karolherbst: okay :)
22:33 karolherbst: hakzsam: anyway, thanks for the help:)
22:34 hakzsam: np
23:15 karolherbst: mupuf: "nvidia-settings --assign GPULogoBrightness=100"
23:15 karolherbst: mupuf: if you have time, could you check if that changes the LED on the titan?
23:22 karolherbst: yay
23:22 karolherbst: "Attribute 'GPULogoBrightness' (reator:0[gpu:0]): 100."
23:23 karolherbst: this will be fun