14:00RSpliet: http://ssabook.gforge.inria.fr/latest/book.pdf - is a pretty darn good resource on SSA compilers
16:22kuerv0: Ive installed xorg, but when I type startx X launches but only shows the test pattern, no desktop manager, any idea ?
16:23Sarayan: what's in your .xstart?
16:23Sarayan: sorry, .xinitrc
16:24kuerv0: exec dbus-launch --exith-with-session startfluxbox
16:25Sarayan: I guess that's what's having a problem
16:25kuerv0: but the same if I type xinit /usr/bin/xterm
16:25Sarayan: you mean the xterm doesn't show up?
16:25kuerv0: only test pattern, no xterm, neither keyboard or mouse
16:25kuerv0: yep, neither xterm neeither nothing, only test pattern
16:25Sarayan: anything interesting is Xorg.0.log?
16:26kuerv0: let me pastebin, I cant see anything strange
16:29kuerv0: when was just installed it worked, it entered desktop manager, but with too low resolution, but after exit X and enter again failed and from since...
16:30Sarayan: yeah, I see nothing special
16:31Sarayan: try addign something akin to echo "I was here" > /tmp/hello.txt in you .xinitrc to see whether it goes there?
16:31kuerv0: ok let me see
16:38kuerv0: if I quicly switch between ttys I can see the taskbar of fluxbox, but suddenly again the pattern
16:39Sarayan: nothing should mean it never went over the echo
16:39Sarayan: so fluxbox should not be started at all
16:42kuerv0: but it doesnt work for example xinit /usr/bin/X neither xinit /usr/bin/xterm
16:42kuerv0: always show the patter
16:43Sarayan: why xinit?
16:44kuerv0: only for testing
16:44kuerv0: neither startx /usr/bin/xterm
16:47kuerv0: yep, so much :S
16:48imirkin: kuerv0: X isn't supposed to start anything
16:48imirkin: so ... working as intended
16:49imirkin: if you run "xinit" (no args) it should start an xterm.
16:50kuerv0: nope :( only test pattern
16:50kuerv0: tried to start X as root, but doestn work
16:51imirkin: silly question... do you have xterm installed?
16:51kuerv0: yes, I have it
16:51imirkin: perhaps something changed in xorg 1.20...
16:53imirkin: also nowadays it doesn't display the test pattern
16:53imirkin: it should just be a black background
16:54kuerv0: pfff so weird then
16:54kuerv0: perhaps something is broken but dont know what
17:20karolherbst: kuerv0: can you do a "xinit xterm"?
17:21karolherbst: but mhh, you tried that, didn't you?
17:22kuerv0: yes Ive tried, and doesnt work
17:22karolherbst: I guess that test pattern thing just gues in the foreground
17:22karolherbst: and everything works, just that client is silly
17:23kuerv0: is possible,
17:23kuerv0: if I quicly switch between ttys I can see the taskbar of fluxbox, but suddenly again the pattern
17:23kuerv0: so maybe is workin but cant remove the pattern
17:24karolherbst: check for applications inside top
17:24karolherbst: and just kill it
17:24karolherbst: if you find it
17:24karolherbst: if this fixes it, you can figure out why it always starts
17:24karolherbst: maybe some local files
17:25karolherbst: like .xinitrc .xsession or .xsessionrc ...
17:25kuerv0: mmm, gonna see
17:25karolherbst: I think xinitrc would be the one in the end
17:25karolherbst: but there are also the system wide session files :/
17:26karolherbst: anyway, if you find that the executeable name you could search for files containing it. Maybe that helps to figure it out
17:26kuerv0: good idea
17:34kuerv0: cant see anything strange
17:35karolherbst: mhh, maybe something weirdo inside the X config? In doubt you could remove all X config files not owned by packages
17:39kuerv0: tried with no luck
18:10joepublic: Today I learned that fake chinese nvidia cards are not just slow, but rather largely nonfunctional past initially putting up a display.
18:11imirkin: on the bright side, they're cheap!
18:11joepublic: My hypothesis, "functional but dirt dog slow", was disproven experimentally.
18:13joepublic: I dunno, I looked them up, and it seems people are selling them for $US40 - $60 which does not sound all that cheap for a nonfunctional device.
18:13imirkin: if you're buying on ebay, you have to be a little careful
18:13imirkin: once you know what you're looking for, they're quite obvious
18:13joepublic: I was also alarmed to see on ebay that I didn't have to look very hard to find them, there are zillions of them.
18:14imirkin: they tend to be the top results :)
18:14imirkin: since ... they're cheap
18:14joepublic: To buy on ebay, I would look for a brand name card whose pictures match the pictures on that maker's website. The chinese fakes all seem to look pretty generic.
18:15imirkin: or are for brands that don't exist
18:17joepublic: Now I wonder if someone gave me this card because it does not work, but most of the chinese fakes do work, or if all those zillions of chinese fakes stop working after several minutes (or upon starting a benchmark or game) like this one I have here does.
18:18joepublic: Because if they are all useless like this one, ebay/ebay buyers have a serious and alarming problem in the form of an endless supply of these things.
18:21kuerv0: ups, sorry, wrong chat
18:21HdkR: When Maxwell switched over to signed firmwares it sort of destroyed that business model. Will take a long time for everyone to get hardware being Maxwell+ only though
18:23joepublic: Their business model does not seem to have been affected (lots of newish model numbers advertised on ebay that all show the pictures of the generic one DVI, one VGA, one HDMI fake card)
18:24HdkR: Sure, but the hardware they are rebranding will become more and more apparently different and their stock will....eventually die out
18:24HdkR: Will take 5-10 years to stomp out
18:24karolherbst: HdkR: they have nvflash
18:25karolherbst: or other tools to just flash whatever vbios
18:25joepublic: One must wonder why ebay is so complicit (other than getting a financial cut of each scam transaction)
18:25HdkR: karolherbst: Can you still rebrand a GTX 1060 to somethign random?
18:25karolherbst: HdkR: you can rebrand a fermi to a GTX 1070 :p
18:25karolherbst: but it also works for maxwell+
18:26HdkR: Hm. I thought the signing knocked out the vbios trickery as well
18:26joepublic: when you say "works", do these cards by and large actually work, just slowly, or are they mostly work-a-little-then-die like the one I have here?
18:26karolherbst: HdkR: what about those high end overclockers? If you look hard enough, you actually find signed vbios you can just flash on your GPU with overclocked settings as well
18:27karolherbst: HdkR: also, how do OEMs set their own OEM clocks ;)
18:27karolherbst: joepublic: normally you sell older crappy cards with newer/high-end model names
18:28orbea:wonders when nvidia will be cooler like amd
18:28karolherbst: HdkR: the trust model depends on nvidia trusting all employees of all OEMs, guess how well that works out ;)
18:28HdkR: Yea, but those are already signed and people are just stripping them from different vendors?
18:28joepublic: The one I have here has a sticker that says "GTX660" but clearly isn't one, but in testing, it dies slowly one black region of the screen at the time, or dies quickly if you start something demanding (glmark2, for example)
18:28karolherbst: HdkR: OEMs can more or less sign their own vbios, no?
18:28karolherbst: how would that work out otherwise?
18:29karolherbst: they have to send every single vbios to nvidia and get that signed there?
18:29karolherbst: I highly doubt that
18:29HdkR: I'm not sure exactly how it works with the vbios
18:29karolherbst: I am more or less sure, that with the right tool (which I am 100% sure exists) you can flash whatever vbios on the GPU
18:29karolherbst: I think the hardware one isn't even verified?
18:29karolherbst: only the software vbios is checked?
18:30HdkR: gross then
18:30karolherbst: I think mupuf told me that on chip vbios doesn't have to be signed
18:30karolherbst: only software loaded ones
18:30HdkR: That's disgusting and should be fixed :/
18:30orbea: s/fixed/tossed/ :P
18:31karolherbst: sadly one year ago maxwell gpus were still expensive
18:31HdkR: Guess I'll find out in a few generations when GT 1030s are rebranded as RTX 5060 cards or something
18:31karolherbst: I just don't know if there is a legal way to get that flashing tool though
18:32HdkR: Eh, it's china, they don't give a flip
18:33karolherbst: I meant for us
18:33karolherbst: I was thinking about buying really cheap maxwell GPUs and do reverse engineering by flashing the vbios
18:33karolherbst: there has to be a legal way to do that
18:34karolherbst: maybe Nvidia would be nice and provide us with such tools, but then there will be an NDA attached for sure :/
18:35joepublic: attract a throwaway developer, talk him or her through flashing some cards, mail you the cards, then he or she goes about his or her merry way, never having touched nouveau?
18:36karolherbst: yeah, I am sure that will work out splendidly
18:37joepublic: I would say that that conclusion is probably a little beyond what's warranted.
18:38HdkR: Attracted any developers to Nouveau is difficult with interest dying due to signed blobs never getting released
18:39joepublic: Yeah I heard those were coming RSN
19:48mupuf: karolherbst: not entirely sure about this though
21:36MichaelP: Video card Nvidia GeForce GT 730 Arch Linux Plasma 5.14.2... nouveau is laggy... oogle chrome is choppy on scrolling content youtube and vlc kinda choppy...
21:39imirkin: MichaelP: nouveau + plasma = fail, i think
21:40imirkin: but pastebin dmesg + xorg logs, perhaps there's something more obvious going on
21:44MichaelP: dmesg https://bpaste.net/show/87421d488d9f
21:44imirkin: do you have nothing plugged into it?
21:45imirkin: oh nm, that's i915
21:45MichaelP: xorg.logs https://bpaste.net/show/0aae9a55bbec
21:46MichaelP: Is built in that i don't use.
21:46imirkin: appears that your monitor is being reprobed with some regularity
21:47imirkin: i assume you're not constantly plugging and unplugging the HDMI cable?
21:47MichaelP: monitor is my 32 inch tv going thru hdmi
21:47MichaelP: nope pluged in 24/7
21:48imirkin: seems like every 5-10 seconds the EDID is being re-read
21:48imirkin: which is an indication of something worse going on
21:48imirkin: when you say "choppy" do you just mean laggy scrolling, or like entire screen blinks?
21:49imirkin: well, one thing you could try is clocking your gpu higher, and see if that fixes anything
21:49imirkin: you can adjust the pstate by futzing with /sys/kernel/debug/dri/1/pstate
21:49MichaelP: If i use nvidia driver.. everythin is great
21:49imirkin: (cat it to see the pstates, then you can flip between them by echo'ing the pstate id into the file)
21:49imirkin: yeah, nvidia reclocks automatically
21:51MichaelP: sounds like that be a lot of work
21:54MichaelP: /sys/kernel/debug/dri/1/pstate .... 07: core 405 MHz memory 810 MHz ... 0f: core 653-901 MHz memory 1600 MHz ....DC: core 405 MHz memory 810 MHz
21:58imirkin: so try "echo 0f > /sys/kernel/.../pstate"
22:01MichaelP: 2nd line like ... echo 0f: core 653-901 MHz memory 1600 MHz
22:05MichaelP: i see here https://www.phoronix.com/scan.php?page=news_item&px=linux-4.5-nouveu-pstate-howto ... on that line at the end adding AC DC * to re-clock card
22:12imirkin: MichaelP: echo 0f
22:12imirkin: as i said
22:15MichaelP: imirkin: so putting echo in font of 0f
22:15imirkin: <imirkin> so try "echo 0f > /sys/kernel/.../pstate"
22:18MichaelP: that adds the AC DC * to the end of that line
22:19MichaelP: guess reboot see if it works any better
22:34karolherbst: imirkin: ntw, did you plan to take another look at my texs patches? I updated the print one some weeks ago
22:34imirkin: i'm so overloaded with other stuff
22:34imirkin: i can't keep it all in my head.
22:35imirkin: need to get ear plugs or something, it all seems to leak out otherwise
22:36karolherbst: wondering if I should run piglit on a maxwell1/2 GPU though, it could be they added something in pascal, allthough I highly doubt that. I guess I will do the pre push run on my maxwell2 though
22:39karolherbst: the whole series is here, but I didn't update any of the other patches and I think you were fine with the other patches: https://github.com/karolherbst/mesa/commits/scalar_tex
22:39imirkin: karolherbst: ok, well, the last patch is hard to review. but it all seems fine in principle
22:39imirkin: so if the series passes the piglit tex-miplevel-selection and other texture tests
22:39imirkin: then the series is R-b me
22:40HdkR: oh? more TEXS stuff?
22:40karolherbst: yeah... the actual opcode stuff is quite annoying, because it looks like "we just put the most common things used in random order" :/
22:40karolherbst: imirkin: k thanks. Will run piglit before pushing anyway
22:40karolherbst: HdkR: you know it's true :p
22:41HdkR: I do
22:41HdkR: TEXS is a mess
22:43karolherbst: HdkR: just look at this patch: https://github.com/karolherbst/mesa/commit/21ce49a402b2d43cf3c39190cb1038094e6dc6be :/
22:45HdkR: yuzu currently completely messes up texs as well, is fun :)
22:45HdkR: because the operanda are so random
22:47HdkR: Does the texs support cause most non-scalar trxture fetches to vanish?
22:49karolherbst: most of the texs thingies are still not perfectly scalar
22:49karolherbst: but most of them should be gone, yes
22:49HdkR: sure, they just max out at vec2 instead of vec4
22:49karolherbst: our RA is just not that perfect and adds movs because it can't place registers better
22:50HdkR: I see
22:50karolherbst: and those usually go away if you not use quad regs
22:50karolherbst: that's where the -1% instruction count comes from
22:52HdkR: I wonder why the encoding space is even wasted having the vector texture fetches
22:52HdkR: seems redundant
22:53karolherbst: HdkR: how so?
22:53karolherbst: you still need those anyway
22:53karolherbst: it's not like you would safe any space if you remove them
22:54karolherbst: or... for what would you use the space instead
22:54HdkR: why? just convert to the scalar versions :D
22:54karolherbst: HdkR: you can't have TEXS with a 0x5 or 0x6 mask
22:54karolherbst: check the patch ;)
22:54karolherbst: you will see what I mean
22:55karolherbst: HdkR: "RegAlloc::InsertConstraintsPass::isScalarTexGM107"
22:55karolherbst: ohh and indirect textures/samplers won't work as well
22:55karolherbst: and... other things