08:34 rmrfchik_: java crash in nouveau http://pastebin.com/mGXkRJCk
09:19 karolherbst: rmrfchik: without a backtrace with debug symbols, we can't do anythihng about that
09:20 karolherbst: rmrfchik: you should create a but with all needed information
10:17 kloofy: what i was saying was, i had all my humanity manners violated some vandalism involved with stalking when i really needed a spare time with my future girlfriend all nervous and bullying stuff to block that , plus lies about that situation.. it seems like such persons are pushing some social conventions when they should be flushed in the toilet, obisouly only crap can come out from their mouths
10:18 kloofy: i do not need social high standarts when they terrorise me so i can not have single minute in life with being people who i'd like to be
10:19 kloofy: i simply ignore those high standarts especially when they are made by obvious terrorists
10:20 kloofy: in any event it sounds like i am always the victim who have special rights, in other words rights taken away, it's better to ignore this shit
10:23 rmrfchik: karolherbst: I switched back to nvidia. Too many issues for me now :(
10:24 kloofy: rmrfchik: those issues can be solved, if ones would not talk about social standarts daily basis which does not anyways exist
10:25 kloofy: people know that i am naturally a class act behving human, they do not keep any of their liabilities to act humanly anyways
10:25 kloofy: lets better not waste time for such issues
10:27 kloofy: rmrfchik: well all those issues are easily solvable, but i am the underdog here, who seems to be responible for the old effecient team collapses it seems, or made responsible
10:29 kloofy: rmrfchik: actuually nouveau and other mesa&gallium drivers is very elegant code still
10:30 kloofy: you have like issues but when you are able to solve them you feel like in a batter position when doing a masters degree
10:30 kloofy: it teaches you a lot
10:32 kloofy: Calinou: sorry for highlighting any specific thing you want to know about amd's firmware?
10:33 kloofy: as you have probably read it's mostly what you called a pmu that is ran via a special microcontroller called latticemicro32 or something similar
10:34 kloofy: for all the work i intend to do it's basically totally irrelevant hunk, as all the needed stuff is splattered across the documentation, 3docs and code
10:35 kloofy: Calinou: there are some things missing what you get from the code, like how to treat the memory controller with it
10:35 kloofy: it's special mmio regs and program it, alex has written about that in the blog, i do not know that very well myself
10:36 kloofy: but to be honest, my for my projects the firmware is not stopping anything at all
10:40 kloofy: bye...
10:45 rmrfchik: what i just read
10:58 kloofy: Calinou: clocking the fpga circuit which should be the correct way to go, happens all on the chip, the microcontroller hence is not needed at all and can be ignored
10:58 rmrfchik: kloofy: i happy to help, but as it work computer I have to postpone experiments till friday
10:59 kloofy: there are many ways to do the circuit, but getting the core right will do that, if not there are tools
10:59 kloofy: http://www.opensocfabric.org/status.php
10:59 kloofy: https://github.com/LBL-CoDEx/OpenSoCFabric/wiki/Architecture
10:59 rmrfchik: btw, choppy scrolling with nvidia in libreoffice can be cured with libreoffice-gtk3, -kde.
11:01 kloofy: https://www.youtube.com/watch?v=828oMNFGSjg this is the popular one, the last link would do what this guy presents there
11:03 kloofy: i.e the two cycle arbitration and routing of commands to noced cores
11:04 kloofy: they call it warmhole routing among/between PE's there, but it can be done manually in verilog also
11:05 kloofy: making the timing correct and adding phases on kintex ultrascale it would do around 600k-700k mips
11:06 kloofy: and all that with a tiny board and more importantly a simple programmable single core looking manycore of the CPU
11:24 kloofy: Calinou: i hope that it's something you know right, that the microcontroller is in between cpu and real hw i.e gpu, primarly controlles how hw is feed with correct clock signals , something that is only readon asic specific, and known allready elsewhere how it works etc.
11:34 kloofy: rmrfchik: do what you gotta do, someday you can get back to nouveau it's no crime to use nvida propr. it's good driver too, but nouveau if worked abit yet not that much could even turn out to be better one
11:40 karolherbst: rmrfchik: we can't fix those if you don't provide us with the needed information though
11:58 rmrfchik: karolherbst: sure, I'll reproduce
11:59 karolherbst: thanks
15:33 siro__: is it save to use two pipe_contexts within the same process, but different threads without any further syncronization on nouveau ?
15:34 siro__: I'm getting engine crashes trying to upload buffers on a second thread
15:34 karolherbst: no, it isn't
15:34 siro__: do they use the same fd to communicate the kernel ?
15:34 karolherbst: threadting+nouveau = bad
15:35 siro__: even with one pipe_context on each thread ?
15:36 imirkin: siro__: https://github.com/imirkin/mesa/commits/locking safens it up a bit
15:37 imirkin: siro__: we only allocate 1 hw context per screen, so each context has to share it. there's no sync on it right now, so it's a pile of fail.
15:37 siro__: I see
15:37 siro__: does it affect shader compilation, too ?
15:38 imirkin: siro__: no
15:38 imirkin: shader compilation is fully self-contained
15:39 imirkin: however if you then try to upload those shaders ... :)
15:39 mebus: Maybe this helps somebody: https://bugzilla.redhat.com/show_bug.cgi?id=1368666 I replaced the card by an ATI card.
15:40 kloofy: bravo finally a bit techincal stuff too, now this was new info to me , but i generally do not know much about gallium
15:40 siro__: imirkin: is there the same restriction on r600 / radeonsi ?
15:40 imirkin: siro__: no, just nouveau.
15:46 kloofy: again, bravo
15:47 kloofy: anyways i am starting to design various chips and decided to hang on the channel, because information like this for gallium for various drivers i really need to know about, cause we need directx later
15:50 kloofy: i
15:50 kloofy: ve been putting together lots of information for you, i'd really appreciate some cowork in the feature to help me with maintaining the driver code
16:03 kloofy: Calinou: well, there is going on some in kaveri's hsa firmware that may happen behind the scenes, that is only maybe cause i have not tested if it generates some hw code in the firmware, but not for opengl conventional firmware, in hsa world the finalizer from hsail is closed source hunk too
16:05 kloofy: Calinou: but in general, the firmware in larger hunk is burnt on the asics and appears as hw, where as in that sence i't would need to be reimplemented according to the iSA, where as there pmu is irrelevant in the work that i soon be doing
16:12 kloofy: where the NOC is way is building a powerful gpu, same can be done for gpu as well, but there are many different options to redesign fully functional and performing circuit ontop of the drivers
16:21 kloofy: Calinou: in that sense don't even bother, no matter what anyone does it to you to block your progress, if you think things well through you will progress ahead regardless and can probably coordinate in the level you need, either in design , or programming , just take and area and manage it
16:25 kloofy: Calinou: i when there is left the chanche do mostly capitilize on it, that is why several people don't like me, i may had lost some time but if person is full of potential, in the end they normally progress out of it
16:55 kloofy: i am going to order lenovo laptop today, with core i7 6th gen + chosen r7/r9 series radeon, i there is possible to get southern/sea/volcanic islands, but couple of months occupied with my job..then will redecide how to coordinate cpu and gpu design world on programmable devices
17:02 mslusarz: envytools new logo is lovely :D
17:03 mslusarz: looks like a work of 5-year old ;)
17:04 imirkin_: i think karol was drunk
17:08 hakzsam: ahah
17:27 Tom^: where is the logo?
17:27 Tom^: oh there.
17:27 Tom^: haha <3
17:30 urmet_: where?
17:31 Tom^: urmet_: https://github.com/envytools
17:32 urmet_: oooh
17:32 urmet_: I was looking around https://github.com/envytools/envytools
17:33 urmet_: that's awesome
17:33 Tom^: if its karols work, you should leave all future and current logos to be worked on by him-
17:33 Tom^: =D
17:34 karolherbst: I wasn't drunk
17:34 karolherbst: mwk asked for something, I delived exactly as he wanted it to be
17:34 Tom^: and its priceless. :)
17:34 karolherbst: sure
17:34 karolherbst: that's why I don't do design stuff, cause nobody would be able to pay
17:35 karolherbst: and before nobody belives me: "< mwk> let's just take the nouveau logo, strike through the nouveau text, and scribble "envytools" on top of it in some crappy mismatched font "
17:42 urmet_: where can I download that font? :D
17:43 karolherbst: well from me if you pay up
17:43 karolherbst: it needs its own font rendering engine though
17:44 karolherbst: I added nifty features like every letter actually has different representations and so on
17:44 ajax: port freetype to run on the falcon
17:44 Tom^: the infiniherbst fork.
17:45 karolherbst: sadly I had to code from scratch, cause freetypes wasn't flexible enough :/
18:39 kloofy:is out of steam
18:40 kloofy: friggin tooth broke , totally different feeling, hard to enjoy different therapies
18:41 kloofy: karolherbst: that is a bit funny that you coded a font like this
19:24 ncm: Hi -- the support matrix seems to say that NV110-series, including NV117, are supported, but my 1.0.12 driver on debian just says "NV117 - unknown"
19:25 karolherbst: ncm: that's why you have to use the modesetting X driver
19:25 imirkin_: ncm: xf86-video-nouveau does not support NV117
19:26 imirkin_: ncm: the X server should fall back to modesetting and move merrily on
19:26 ncm: Should it be easy for me to find out how to persuade X to use modesetting?
19:27 ncm: Usually I see modesetting in terms of how to install a kernel module.
19:29 imirkin_: there's an X ddx called "modesetting"
19:30 imirkin_: which is built on top of KMS for modesetting and GLAMOR for acceleration
19:37 GreatEmerald: What could be the reason that I get a "modeset(0): eglInitialize() failed" error when using the mosesetting DDX?
19:37 imirkin_: your Xorg is of a particular vintage
19:37 imirkin_: which doesn't play nice with mesa
19:38 imirkin_: something changed in the egl init
19:38 imirkin_: not sure if it was reverted, or what
19:38 GreatEmerald: Hm, I have "X.Org X Server 1.18.4"
19:38 imirkin_: yeah, iirc it was "every X server older than something"
19:38 imirkin_: which i'm sure is not released
19:39 GreatEmerald: Oh, so it will get fixed in some future release? All right then
19:39 imirkin_: i think there's some solution afoot
19:40 imirkin_: i'm not really plugged into these things
19:40 imirkin_: oh, another reason is that your GPU does not support acceleration with nouveau
19:40 imirkin_: but i figured that wasn't the issue you were running into
19:40 GreatEmerald: That'd be nice, given that on openSUSE Tumbleweed, right now llvmpipe is the only usable driver...
19:40 imirkin_: wait, what gpu do you have?
19:40 GreatEmerald: GTX 660
19:40 imirkin_: and why are you using modesetting?
19:41 GreatEmerald: Because I'm trying to debug the GPU hang issue
19:41 GreatEmerald: If I use nouveau, it always hangs randomly and mysteriously
19:41 ncm: how does one "use the modesetting ddx"? Google reveals nothing.
19:41 GreatEmerald: ncm: xorg.conf, driver "modesetting"
19:42 imirkin_: GreatEmerald: yeah, but right now you're not using acceleration at all, so nothing's using nouveau, so no hangs
19:42 GreatEmerald: Yeap, exactly
19:42 GreatEmerald: I know it's a long shot
19:42 imirkin_: you could achieve the same effect by loading nouveau with nouveau.noaccel=1
19:43 GreatEmerald: Hm, I could try that
19:43 ncm: meaning, don't mention nouveau in xorg.conf, just "modesetting"?
19:43 GreatEmerald: ncm: Yeap
19:43 karolherbst: ncm: remove yxour xorg.conf
19:43 imirkin_: ncm: or just don't have an xorg.conf at all - things Just Work (tm) nowadays
19:43 ajax: well. if you don't have the nouveau ddx installed, removing xorg.conf will do that.
19:43 imirkin_: even if you do have the nouveau ddx installed
19:44 imirkin_: it'll fail to load and X should fall back to the next one
19:44 GreatEmerald: Well, "nouveau" is higher on the priority list than "modesetting"
19:44 imirkin_: er, fail to probe
19:44 ajax: oh, it fails to bind now?
19:44 ajax: that's awful nice of it
19:44 ncm: I am on a Lenovo P70 with "optimus". Things are very not-just-working.
19:44 karolherbst: the only valid reason for having a xorg.conf is if you have a broken display or need to hackaround xorg prime sillyness :p
19:44 imirkin_: ajax: pretty sure...
19:44 ncm: I don't even care about acceleration, I just want it to drive external monitors.
19:45 GreatEmerald: So if you have both, you need to tell it to specifically use "modesetting" or else it will use nouveau
19:45 ncm: Great, trying it. Thanks!
19:45 karolherbst: GreatEmerald: no
19:45 karolherbst: GreatEmerald: except, the nouveau ddx supports your gpu
19:45 karolherbst: if not, it falls back to modesetting
19:46 GreatEmerald: Yea, but most people have a GPU that nouveau supports
19:46 karolherbst: maxwell ain't one
19:46 imirkin_: ajax: hm crap. for GM10x it'll still try to load
19:46 karolherbst: and for everything not bein maxwell, you should use the nouveau ddx
19:47 ajax: https://cgit.freedesktop.org/nouveau/xf86-video-nouveau/commit/src/nv_driver.c?id=3e2e0faa2ee1cce9c1bb5c7ad80d0592460f3edc
19:47 ncm: wait, so "modesetting" won't work? It also has the intel thing, but that can't see the external monitor ports.
19:47 ajax: imirkin_: presumably 0x110 is just gm110?
19:47 imirkin_: oh right. and HasKMS is checked in PciProbe
19:47 imirkin_: so it should b egood
19:47 imirkin_: 0x110 = gm10x
19:48 imirkin_: [and by "good" i mean "not try to bind to the device"]
19:48 GreatEmerald: karolherbst: Well, modesetting has glamor acceleration which is potentially interesting
19:48 GreatEmerald: Also https://www.phoronix.com/scan.php?page=news_item&px=Nouveau-Vs-Modesetting
19:48 imirkin_: GreatEmerald: glamor = using acceleration. if acceleration = hangs on your gpu, that won't help.
19:49 imirkin_: in fact, chances are that the nouveau ddx's accel is just fine for you
19:49 imirkin_: and you should use the nouveau ddx, and just remove nouveau_dri.so
19:49 imirkin_: which will avoid having an opengl impl
19:49 imirkin_: so normal things will be fast
19:49 GreatEmerald: imirkin_: Yea, I figure as much, I'm just arguing the point that using modesetting with nouveau is overall a bad idea
19:50 imirkin_: GreatEmerald: i've made that argument before :)
19:51 ncm: So, rename nouveau_dri.so?
19:51 imirkin_: ncm: that comment was meant for GreatEmerald
19:51 ncm: Ok.
19:51 karolherbst: ncm: just remove your xorg.conf and everything should turn out fine
19:52 GreatEmerald: imirkin_: What do you mean by "normal things"? Wouldn't that just cause the driver to fail to load?
19:52 imirkin_: GreatEmerald: i think i've sufficiently confused you ...... nevermind.
19:52 GreatEmerald: Hah
19:53 GreatEmerald: But yea, right now on Tumbleweed, the packaged NVIDIA proprietary drivers don't support kernel 4.7 (surprise surprise), nouveau hangs my GPU, and the only alternative is llvmpipe...
19:53 ncm: I started with no xorg.conf and it wasn't fine. It only worked at all with BIOS setting graphics = discrete, and then used the frame buffer driver.
19:54 karolherbst: GreatEmerald: do you change the pstate?
19:54 karolherbst: ncm: odd, what kernel are you using?
19:54 ncm: switching to "hybrid" and setting the "intel" driver, I got some acceleration, but lost all access to the second screen.
19:54 GreatEmerald: karolherbst: What do you mean by "change"?
19:54 ncm: kernel is 4.6.4 stock debian
19:55 ncm: X is 1.18.4
19:55 karolherbst: ncm: k, you may want to use reverse prime then, but...
19:56 karolherbst: ncm: well, if you care about power consumption you want the intel to be your main one
19:56 ncm: <googles reverse prime>
19:56 karolherbst: also that gives you a higher stability, in theory
19:56 karolherbst: ncm: https://nouveau.freedesktop.org/wiki/Optimus/
19:57 karolherbst: imirkin_: by the way, is _anybody_ working on a proper solution for the reverse prime situation, so that it works also for wayland and dri3?
19:57 GreatEmerald: karolherbst: And do you mean the CPU pstate, or the power states of the GPU?
19:58 GreatEmerald: Either way, I have them al left on defaults
19:58 GreatEmerald: all*
20:11 GreatEmerald: Oh, so you meant by nouveau.pstate. I could try changing it and see what happens, but probably nothing good
20:12 karolherbst: not really with stock nouveau
20:12 imirkin_: karolherbst: not me!
20:12 karolherbst: with my patches yes, but unlikely that they fix those "random" hangs
20:12 karolherbst: imirkin_: I know .D
20:13 imirkin_: and i think airlied has been too enamored in vulkan to work on anything like that
20:13 karolherbst: ahh crap
20:13 imirkin_: but i'm not sure that reverse prime is related to ... dri
20:13 GreatEmerald: Yea, according to the power management table, NVE0 has WIP reclocking anyway, so I guess that means it wouldn't really do much to begin with
20:13 karolherbst: maybe we can convince some of the amd devs, doesn't amd actually need this?
20:14 imirkin_: i have a very poor idea how it's all hooked up though
20:14 karolherbst: imirkin_: my understanding is: it doesn't work for wayland anyway
20:14 karolherbst: and nobody really seems to care
20:14 imirkin_: they'll care once wayland gets users
20:14 imirkin_: right now it has none afaik
20:15 GreatEmerald: Oh, and apparently there's no more pstate since 4.5
20:15 karolherbst: imirkin_: yeah, but my point is, we most likely need some in kernel mechanism, just like prime offloading
20:16 imirkin_: it's tricky... may involve a copy
20:16 imirkin_: since the remote gpu may not be able to scan out from gart at all
20:16 karolherbst: right
20:16 karolherbst: of course it needs to be flexible enough to provide multiple links
20:16 imirkin_: it's in the long list of Things I Don't Worry About
20:17 karolherbst: it should be also possible to use SLI for that, like for prime offloading as well
20:19 karolherbst: ohh right
20:19 karolherbst: new trello card :)
20:22 GreatEmerald: Hm, there seems to be some interesting debug logging options available
20:23 GreatEmerald: This kernel has CONFIG_NOUVEAU_DEBUG=5, so would "nouveau.debug=debug" enable that extra logging?
20:23 hakzsam: a bunch of new features in trello :)
20:24 JodaZ: how lang has power management been the major roadblock for nouveau now?
20:24 karolherbst: hakzsam: add more if you can think of anything which isn't there :)
20:24 JodaZ: *how long
20:25 hakzsam: karolherbst, don't want to think :)
20:25 karolherbst: JodaZ: well it is pretty much done for kepler+ now though, or at least the biggest thinks are already either prototyped or in review queue
20:25 hakzsam: my todolist is already full of things for the next motns
20:25 hakzsam: *months
20:25 karolherbst: hakzsam: :D add them all!
20:25 JodaZ: karolherbst, does kepler+ include maxwell?
20:26 hakzsam: karolherbst, most of them are already in trello actually
20:26 hakzsam: karolherbst, and others are not really interesting (ie. bugfixes like F1)
20:26 karolherbst: JodaZ: yes
20:26 karolherbst: JodaZ: we can fully reclocked maxwell2 gpus, too
20:26 karolherbst: JodaZ: but, we can't control the fans due to missing signed firmware
20:27 karolherbst: hakzsam: I see
20:27 karolherbst: hakzsam: bugs belong into bugzilla anyway
20:27 hakzsam: yeah, but not all bugs
20:27 JodaZ: controlling fw needs signed fw?
20:28 JodaZ: you can't just take the signed FW from the official driver?
20:28 JodaZ: seems like fair use to me
20:29 karolherbst: JodaZ: first we have to actually get everything we need to upöload them
20:30 karolherbst: and that is a bunch of stuff
20:30 karolherbst: second: we have to find it first
20:31 karolherbst: JodaZ: if it would be that easy, it would be already done (provided we have enough time actually)
20:31 JodaZ: i dunno, i assumed legal shenanigans
20:31 karolherbst: well
20:31 karolherbst: we can't distribute them either
20:33 JodaZ: so at the end of this, you will license your whitebox results to some chinese company to make knockoff nvidia hardware, right?
20:33 karolherbst: why should we?
20:34 karolherbst: we don't do this for money you know, but actually for fun
20:34 JodaZ: why shouldn't you tho, seeing that nvidia never contributed
20:34 karolherbst: they actually did
20:35 JodaZ: with what
20:35 karolherbst: tegra support
20:35 JodaZ: rofl
20:36 karolherbst: well, those are kepler and maxwell based
20:36 karolherbst: and actually quite alike to the desktop counterparts (with exceptions)
20:36 karolherbst: also
20:36 karolherbst: the secure boot stuff was actually implemented by an nvidia dev
20:36 JodaZ: its obvious why they are doing it for those tho, its because there they have no particular advantage over the competition
20:37 JodaZ: oh so very very tactical
20:38 karolherbst: what do you mean?
20:38 imirkin_: JodaZ: knowing which bits in the hardware do what is like 0.01% of the effort to making hardware that does those things when those bits are twiddled.
20:38 JodaZ: imirkin_, free drivers tho
20:39 imirkin_: hehe
20:54 idl0r: hm, something between mesa 11.2.2 and 12.0.1 causes the OSD menu of my VDR to be very slow
20:54 idl0r: using nouveau/vdpau
20:55 imirkin_: idl0r: probably dri3
20:55 imirkin_: unfortunately there's no way to turn it off =/
20:55 idl0r: i'll try to figure out whether 12.0.0 is affected as well
20:55 idl0r: hm
20:55 imirkin_: 12.0.1 = 12.0.0 + build fix
20:55 idl0r: ah, ok
20:56 imirkin_: https://github.com/imirkin/mesa/commit/778501f3775f587a18817be4c7514cdc1ca980d3
20:56 idl0r: video itself is ok it seems but as soon as i open the OSD it will be slow as hell and also effect playback of the video
20:56 imirkin_: try that patch, and enable the env var
20:56 imirkin_: i noticed mplayer was slow for me too
20:56 imirkin_: i use vdpau pretty rarely though
21:02 idl0r: imirkin_: will try that. thanks