00:30imirkin: gnurou: have you had any stability issues with your TK1? esp as far as ethernet is concerned?
00:36gnurou: imirkin: not that I remember, and I am using network boot + NFS root
00:37gnurou: imirkin: which kernel are you using?
00:37imirkin: gnurou: 4.3... any time i try to tab-complete in bash, the thing dies
00:37imirkin: with not-a-whisper over netconsole or any other network thing
00:37gnurou: wow - but I fail to see how this could be related to ethernet?
00:38imirkin: gnurou: i'll try 4.4 when i get around to doing it again...
00:38imirkin: gnurou: well, network activity seems to trigger the problem, and i don't get anything over the network when the issue happens. seems connected :)
00:38gnurou: maybe try the staging/work branch on https://github.com/Gnurou/linux
00:39gnurou: it's a bit outdated, but I am using it without any issue that I can tell
00:39imirkin: ok, i'll keep that in mind
00:39gnurou: if your root is on eMMC or SD card, maybe the problem comes from here? Especially if bash-tab can trigger this
00:39imirkin: nfsroot
00:39imirkin: everything over network
00:39gnurou: ah, so same setup as me
00:39imirkin: no screen connected
00:40imirkin: and i don't have enough screens/connectors to boot it with a screen plugged in
00:40imirkin: (specifically i only have 1 screen with a single digital input)
00:40imirkin: (which i'm using)
00:41imirkin: i was working on the ASTC stuff but it became a bit frustrating after the 50th reset
00:42imirkin: i _think_ i got it work, but need to retest
00:42imirkin: on the off chance you feel like trying it out: https://github.com/imirkin/mesa/commit/1ba596c899332673f0cad644eca6ce41747e2bed
00:43imirkin: [unfortunately the version i tested was totally broken and didn't notice... by the time i fixed it, i was already sick of the resets]
00:45imirkin: and i guess i should also throw ETC1 into the list, i've been told it's actually a subset of ETC2 and thus should be supported by an ETC2 decoder
02:47user_6521: allah is doing
02:47user_6521: sun is not doing allah is doing
02:47user_6521: moon is not doing allah is doing
02:47user_6521: stars are not doing allah is doing
02:47user_6521: planets are not doing allah is doing
02:48user_6521: galaxies are not doing allah is doing
02:48user_6521: oceans are not doing allah is doing
02:49user_6521: mountains are not doing allah is doing
02:49user_6521: trees are not doing allah is doing
02:49user_6521: mom is not doing allah is doing
02:49user_6521: dad is not doing allah is doing
02:49user_6521: boss is not doing allah is doing
02:50user_6521: job is not doing allah is doing
02:51user_6521: dollar is not doing allah is doing
02:51user_6521: degree is not doing allah is doing
02:51user_6521: medicine is not doing allah is doing
02:51user_6521: customers are not doing allah is doing
02:52user_6521: you can not get a job without the permission of allah
02:52user_6521: you can not get married without the permission of allah
02:52gryffus: lol :DDD
02:53user_6521: nobody can get angry at you without the permission of allah
02:53user_6521: light is not doing allah is doing
02:53user_6521: fan is not doing allah is doing
02:53user_6521: businessess are not doing allah is doing
02:53user_6521: america is not doing allah is doing
02:54user_6521: fire can not burn without the permission of allah
02:54user_6521: knife can not cut without the permission of allah
02:54user_6521: rulers are not doing allah is doing
02:55user_6521: governments are not doing allah is doing
02:55user_6521: sleep is not doing allah is doing
02:55user_6521: hunger is not doing allah is doing
02:55user_6521: food does not take away the hunger allah takes away the hunger
02:56user_6521: water does not take away the thirst allah takes away the thirst
02:56user_6521: seeing is not doing allah is doing
02:57user_6521: hearing is not doing allah is doing
02:57user_6521: seasons are not doing allah is doing
02:57user_6521: weather is not doing allah is doing
02:57user_6521: humans are not doing allah is doing
02:58user_6521: animals are not doing allah is doing
02:58user_6521: the best amongst you are those who learn and teach quran
02:59niko: user_6521: could you stop that ?
02:59user_6521: one letter read from book of allah amounts to one good deed and allah multiplies one good deed ten times
03:00niko: user_6521: that's a no ?
03:00kelsoo: user_6521 is not doing allah is doing
03:01kelsoo: I suspect
03:01user_6521: hearts get rusted as does iron with water to remove rust from heart recitation of quran and rememberance of death
03:01user_6521: heart is likened to a mirror
03:02user_6521: when a person commits one sin a black dot sustains the heart
03:04user_6521: to accept islam say that i bear witness that there is no deity worthy of worship except allah and muhammad peace be upon him is his slave and messenger
03:05user_6521: read book fazaileamaal.com
03:05user_6521: read book muntakhabahadith.com
03:06user_6521: need spiritual teacher visit alhaadi.org.za
03:06niko: user_6521: if you don't stop your proselytizing you will end quieted
03:08niko: pq: ^
03:13pq: bleh
03:14user_6521: allah created the sky without any pillars
03:14pq: oh cool
03:14gryffus: thank guys
03:14user_6521: allah makes the sun rise from the east and sets it in the west
03:15pq: I wasn't sure I had powers here :-)
03:15karolherbst: :D
03:16vedranm: pq: did it work?
03:16vedranm: conspiracy theory: this is actually someone from NVIDIA
03:16karolherbst: well he is still in the user list
03:46karolherbst: that should never happen right? https://gist.github.com/karolherbst/4bab6c6cded381d82186 :D
04:11RSpliet: karolherbst: never say never
04:17karolherbst: RSpliet: do you think it would be a good idea to just code a pmu script into the kernel from the blob (I mean port the reg writes and waits to nouveau) and then try to get everything else right first?
04:17karolherbst: and then to take care of the pmu scripting?
04:18RSpliet: what do you mean?
04:18RSpliet: do you mean taking NVIDIA's scripts and sort-of-unconditionally copy-paste them in nouveau as a starting point?
04:19karolherbst: something like that, just to check if everything else is in order
04:19karolherbst: not that I play around with the pmu parts and something else is messed up already
04:19RSpliet: I personally get the feeling that this approach is a bit limiting in the sense that the scripts don't contain writes to registers that already contain the correct value
04:20RSpliet: (unless they are triggers rather than values)
04:20karolherbst: RSpliet: k, but when I take the script from boot values to the next pstate?
04:20karolherbst: this should somehow work then, or not?
04:21RSpliet: that doesn't solve the problem, since boot also initialises the values using the initscripts
04:21RSpliet: my approach has been more VBIOS driven. I do take the script as my reference to see if I ticked all the boxes
04:21RSpliet: but per VBIOS bit I see what it changes in a script
04:22RSpliet: (yes, I actually recently started using meld for that)
04:22karolherbst: meld is nice though
04:22karolherbst: I also use it for visual diffing
04:23RSpliet: that way you don't narrow your view towards the one register you're trying to sort out :-)
08:50pmoreau: Interesting: never thought my MCP79 could drive a 1440p screen with glxspheres at 60fps, on top of the 900p laptop screen.
10:12duelle: After having enabled reverse PRIME and added two external screens I have some problems mostly regarding some GUI tools: keyboard shortcuts not working anymore, misplaced or completely invisible context menus, very small fonts. Do you know of such issues and possibly how to get rid of them?
10:13duelle: One interesting thing is for the example of Firefox: If I try to open the menu on the main screen, it displays just fine. If I move the window to an external one it is not visible at all.
10:31karolherbst: pmoreau: :D
10:33karolherbst: duelle: at first this sounds a bit like a window manager bug or something else strange.
10:33karolherbst: duelle: what desktop are you using?
10:33duelle: karolherbst: xfce/xfwm
10:33karolherbst: duelle: you could try out if disabling compositon changes anything or you could like for example install kwin or metactiy or anything else
10:34pmoreau: karolherbst: I just need to figure out how to get Nouveau to that same level of awesomeness. :-)
10:34karolherbst: like after installing kwin you can simply replace xfwm by doing kwin --replace (or kwin_x11 for the plasma 5 version)
10:34karolherbst: pmoreau: ohh that was with binary driver?
10:34pmoreau: karolherbst: Sure!
10:34karolherbst: pmoreau: so does nouveau fail performing at 60 fps or at the display setup? :D
10:35duelle: karolherbst: One thing I just saw in my logs is from xsettingsd: "trap divide error" and a stack trace.
10:35pmoreau: karolherbst: With Nouveaum the external screen doesn't even light up and at some point, the whole screen freezes.
10:35karolherbst: pmoreau: mhhh
10:35karolherbst: pmoreau: try out a lower resolution for it ;)
10:35karolherbst: pmoreau: but I am sure that old gpus usually are able to run 1440p
10:35duelle: karolherbst: Disabling composition does not help.
10:36pmoreau: karolherbst: Rather, the whole laptop freezes. Same issue on 1080p, depending on the adaptor.
10:36karolherbst: pmoreau: and with 900p?
10:37pmoreau: karolherbst: With one of the adaptor on 1080p, I got the external screen running, but starting an OpenGL program would make it flicker until reboot.
10:37karolherbst: mhh
10:37pmoreau: I haven't tested 900p
10:37karolherbst: pmoreau: I am pretty sure that some of these can also do 2560 x 1600
10:38pmoreau: But now I know the chipset can manage 1440p ;-)
10:38karolherbst: pmoreau: is it a macbook?
10:38pmoreau: NBP mid 2009
10:38karolherbst: because then you can bet they all can do 2560 x 1600
10:39pmoreau: How so?
10:39karolherbst: because why should apple sell a 30" 2560 x 1600 display when some modesl aren't able to drive it :p
10:39mwk: imirkin: my tests passed without modification on G80 and GT216, seems the mul.sat thing is the only things varying with chipset in fp32 & sfu insns
10:39mwk: unless hmm... unless they take different operands
10:39mwk: I'll have to throw in tests for s[] and c[] inputs
10:39mwk: but that's probably best left to the next gen of tests
10:39pmoreau: karolherbst: When did they release their first 1600p monitor?
10:40karolherbst: no idea, but I am sure everything after 2008 can drive these
10:40karolherbst: maybe there are some exceptions to this
10:40karolherbst: pmoreau: https://support.apple.com/en-us/HT201927
10:41mwk: oh, and pre-G200 GPUs ignore the "64-bit" flag in cvt instructions
10:41mwk: making them alias the 32-bit counterparts
10:42pmoreau: karolherbst: Seems like it should :-) But well, I'll be happy if Nouveau can get 1080p and run OpenGL program on it.
10:42karolherbst: pmoreau: and I think it came around 2004 or later
10:42duelle: karolherbst: I just tried using "metacity --replace" and the problem persists with metacity.
10:43karolherbst: pmoreau: from wikipedia "All Power Mac G5, PowerBook G4, and Mac Pro models that were introduced after the 30-inch model was released are capable of supporting it without the use of any adapters."
10:43karolherbst: duelle: k, then maybe this is indeed a nouveau problem, but it really shouldn't :/
10:43karolherbst: it's just the intel gpu rendering stuff and sends it to the gpu
10:43karolherbst: as image data
10:43karolherbst: basically
10:44duelle: karolherbst: Might this xfsettingsd thing be an issue related to this?
10:44karolherbst: could be
10:44karolherbst: duelle: could you try to just start plain X?
10:44karolherbst: without any desktop running on it?
10:44karolherbst: you can start xfwm or anything on the xterm if you want though
10:45karolherbst: but just not the desktop environment itself
10:45pmoreau: Anyhow, I made a nice 150MB MMT trace of plugging/setting up/unsetting/unplugging the external screen, will see if I can find anything.
10:45karolherbst: :D
10:46duelle: karolherbst: Do I have to disable lightdm in some way. Or is it enough to go into tty and start x from there?
10:46karolherbst: pmoreau: we had a discussion about something related here recently, with a g3x or g4x gpu or something
10:46karolherbst: duelle: also disable lightdm
10:46karolherbst: pmoreau: seems like there is a way to enable some special memory regions to do something special to speed up the transfer rate to the display or something
10:47pmoreau: Interesting!
10:47duelle: karolherbst: ok thanks, I will test it and come back then.
10:47karolherbst: pmoreau: and it was you :D
10:48pmoreau: Oh, that NISO_POLLER thingy?
10:48karolherbst: yeah
10:48pmoreau: Well, without enabling it, the laptop lockuped
10:48karolherbst: :D
10:48karolherbst: k
10:48karolherbst: so more is needed
10:49pmoreau: And by lockup, I mean lockup at boot while Nouveau is initialising everything.
10:50pmoreau: Turns out it is usually set up by the VBIOS, but those MBPs VBIOS didn't do it.
10:51karolherbst: funny
10:52pmoreau: Not that much I would say but, … 4.4 is the first kernel that supports that MBP without locking up on boot.
10:53pmoreau: Some 6 years after being released…
10:53karolherbst: :D
10:54karolherbst: those apple laptops are special anyway and not many care about linux anyway on these
10:56duelle: karolherbst: Do I have to provide any further paramters to startx to keep the X server running? It seems to start up and terminate (gracefully) afterwards.
10:57karolherbst: duelle: xinit should work usually
10:58karolherbst: ohh mhh
11:00imirkin: mwk: are you saying that GT21x does f64 cvts?
11:00mwk: imirkin: I'm saying it recognizes the opcode
11:00imirkin: mwk: i thought all f64 things were G200-only
11:00mwk: ... and gives you an illegal opcode error
11:00imirkin: hehehe
11:01imirkin: fair enough :)
11:01mwk: as opposed to pre-G200, which aliased the opcode to f32
11:02karolherbst: oh lol, I got some kickstarter rewards laying around since mid last year :D
11:03mwk: imirkin: FWIW I don't have the f64 ops test-covered yet
11:03mwk: but then they can't differ between chipsets at all, can they now...
11:04imirkin: mwk: let's say it's unlikely :)
11:04mwk: well
11:04mwk: if they changed the ISA between G200.A1 and G200.B1, I'm going to search for an axe
11:05imirkin: aka "unlikely"
11:06mwk: SM 1.2999995
11:06duelle: karolherbst: xinit did the trick. But after having started xfwm/metacity and firefox, I can still reproduce the menu issue.
11:06mwk: aka pentium edition
11:06imirkin: duelle: nouveau just draws whatever picture the intel ddx hands it
11:11duelle: imirkin: I understand. So this would rather be a topic for intel-gfx? Or what would you suggest?
11:12imirkin: duelle: perhaps yeah.
11:13duelle: Thanks.
13:56RSpliet: mwk: did you confirm ex2 sat on pre-G200 as well?
13:59imirkin: yeah, i think ex2 sat is everywhere
13:59imirkin: iirc i even knew about it
13:59imirkin: but... forgot.
14:00RSpliet: I was hoping for a night of a bit of fun, but apparently the TGSI compiler already converts 1/exp2(x) to neg preex2, ex2.
14:01imirkin: RSpliet: :)
14:02imirkin: RSpliet: i'm going to throw sat support in
14:02RSpliet: I already have a patch
14:03imirkin: oh
14:03imirkin: send it :)
14:03RSpliet: not before I test it on 9600GT
14:05imirkin: RSpliet: this is my untested diff: http://hastebin.com/mopexuluwo.coffee
14:06RSpliet: ah heh, you even went for the short form
14:06imirkin: there's no short form version
14:07mwk: RSpliet: it's everywhere
14:08mwk: the only difference in modifiers on f32 instructions that I found is fmul.sat
14:16RSpliet: imirkin: yes, after blinking again I figured you just added an assertion there
14:16imirkin: RSpliet: right. just in case :)
14:16imirkin: RSpliet: i've had my share of bugs with this stuff
14:16imirkin: as have you, iirc :p
14:17RSpliet:whistles and looks away
14:17RSpliet: .. yes
14:17imirkin: i think i've finally fixed all of those various issues... many of which weren't your fault
14:17imirkin: but the whole thing with immediates and short registers, etc
14:18imirkin: now the RA knows what's up too
14:33imirkin: gnurou: actually looks like i was testing with 4.4-rc5. i'm going to try it out with 4.4-final now.
14:44pmoreau: There is no remainder op in nv50_ir, or did I missed it?
14:46imirkin: pmoreau: like modulo? float or integer?
14:47imirkin: pmoreau: for integer there's OP_MOD
14:48pmoreau: That's true
14:51RSpliet: imirkin: consider your untested diff now a tested diff
14:52imirkin: RSpliet: ah, i assumed you'd want to send you rown
14:52RSpliet: the only alteration I'd make is line 18, write the assertion as assert(i->op == OP_EX2); - but that's only a minor readability thing
14:52RSpliet: nah, yours is more complete already, and closer to the person who pushes :-P
14:55imirkin: RSpliet: heh ok. yeah, i thought about checking i->op instead.....
14:58RSpliet: imirkin: fwiw btw, test -> http://ur1.ca/of408
15:04imirkin: thanks
22:47karolherbst: aouz