01:49 sooda: hi markz :)
01:50 markz: yeah, konsta.
10:58 karolherbst: imirkin: fyi, having my card activated since boot, bbswitch blacklisted, bumblebeed deactivated and module loaded before X11 start didn't help
11:01 xexaxo: karolherbst: I'm curious if the BB people will ever update this.
11:01 xexaxo: fix even.
11:02 xexaxo: I'd imagine that they can check if the module has the pmops and stop leave it be, rather than trying to switch off the device.
12:06 imirkin: karolherbst: doh, oh well
12:08 karolherbst: but it seems like even on a secod X server I can't do much, so I don't kow, maybe I should try out somethig without GLX
12:08 karolherbst: which I think should work anyway
12:31 karolherbst: imirkin: what is also strange, that if I load the nouveau driver after X got started, the first time X restarts or crashes itself for reasons I can't figure out :/ doesn't happen after the first time anymore
13:06 karolherbst: imirkin: but all in all, the driver is much more stable with the patch. No screen hangs or something crazy while module (un)loading
15:02 giaud: Hi, I have problem with shader generating UCMP instruction with negation
15:02 giaud: UCMP TEMP[4].x, TEMP[2].xxxx, -TEMP[3].xxxx, IMM[1].zzzz
15:02 imirkin: for which gpu?
15:03 imirkin: do you have a full sample shader that it miscompiles?
15:03 giaud: NV96
15:04 imirkin: the sample shader will let me look at what's going on
15:05 giaud: http://hastebin.com/iwaxunawok.avrasm
15:06 imirkin: giaud: i was hoping tgsi shader ;)
15:06 imirkin: coz then i can feed that to nouveau_compiler easily
15:06 imirkin: i assume you have it since you pasted that UCMP instruction
15:07 giaud: http://hastebin.com/vehawoceqe.avrasm
15:07 imirkin: excellent, thanks!
15:07 imirkin: and i guess the instruction 47: is where you think it's going wrong?
15:08 giaud: yes
15:08 imirkin: grrrrr... you have a european locale which breaks printing of the floats :(
15:08 imirkin: IMM[0] FLT32 { 1,0000, 2,0000, 1,5000, 0,0000}
15:08 imirkin: heh
15:09 imirkin: ok, got it building
15:10 giaud: nouveuau emits integer negation on float values
15:12 imirkin: EMIT: presin f32 $r1 $r1 (8)
15:12 imirkin: EMIT: sin f32 $r1 $r1 (8)
15:12 imirkin: EMIT: neg u32 $r3 $r1 (8)
15:12 imirkin: yeah
15:12 imirkin: that looks like BS
15:12 imirkin: but otoh
15:12 imirkin: 46: SIN TEMP[3].x, TEMP[3].xxxx
15:12 imirkin: 47: UCMP TEMP[4].x, TEMP[2].xxxx, -TEMP[3].xxxx, IMM[1].zzzz
15:12 imirkin: i dunno if that's right
15:12 imirkin: i need to double-check what UCMP does
15:20 imirkin: giaud: this is a very questionable fix: http://hastebin.com/goyahateye.hs
15:32 imirkin: er wait, not quite
15:32 giaud: problem with UCMP started in commit d8da6deceadf5e48201d848b7061dad17a5b7cac
15:34 imirkin: yeah, give me a min to come up with a real fix
15:43 imirkin: giaud: i think something like http://hastebin.com/ucudogitet.mel might work better
15:52 imirkin: giaud: is this coming from an application or a trace i could test btw?
15:57 giaud: yes, this version works better. I can create trace.
15:58 imirkin: just want to double-check it on nvc0 as well
15:58 imirkin: it uses slightly diff instructions
16:14 imirkin: giaud: let me know if you want a reported-by mention in the commit
16:49 giaud: imirkin: trace file https://www.dropbox.com/s/pwcjd4xtw3mdm3o/test2.6.trace?dl=0
16:50 imirkin: giaud: ok cool
16:51 imirkin: huh weird. it worked on nvc0 even before. i wonder why.
16:54 imirkin: wtf. it gets different tgsi than nv50 on input. that's extremely odd
17:07 imirkin: ah, looks like nv50 limits control flow depth to 4 (???)
17:19 giaud: if it help i can send trace with full shader
17:20 imirkin: giaud: nope, all good. did my fix work for you? it worked for me...
17:20 giaud: yes it work
17:20 imirkin: giaud: let me know if you want a Reported-by mention
17:21 imirkin: (i.e. i'd need your name/email). if not, i'll just push
17:21 giaud: ok just push
17:25 imirkin: ok, pushed. should be included in mesa 10.6.2. thanks for the report!
18:41 Karlton: bleh, libdrm-2.4.62 asks you to install valgrind unless you pass --disable-valgrind
18:43 imirkin: --enable-valgrind=auto
18:43 imirkin: unfortunate bug that slipped in. it's fixed at HEAD
18:47 Karlton: okay :)
18:49 d9500: my 2008-era nvidia gpu finally bit the dust recently...i haven't really had to follow the development thoroughly and learn which modern video cards nouveau supports best because it pretty much worked "out of the box" with my previous card.
18:49 d9500: any suggestions?
18:50 d9500: nothing Maxwell, at least as long as nvidia is requiring firmware the driver can't generate
18:50 d9500: but anything from previous generations is fine
18:56 imirkin: d9500: there's some reclocking support for keplers (GKxxx)
18:58 d9500: something like this?
18:58 d9500: wait, are we allowed to post a URL or is that a kick?
18:58 imirkin: urls are fine
18:59 imirkin: (wtf kind of channels don't allow urls??)
18:59 d9500: http://www.amazon.com/PNY-GeForce-Graphics-Cards-VCGGT7201D3LXPB/dp/B00MNGI8QW
18:59 imirkin: yeah... that one might be one of the GK208B's actually, not a lot of people have them, so it may have some issues
19:00 imirkin: although i think they should work mostly ok
19:00 imirkin: note that it's a really low-end one... only 8x pci-e among other things
19:00 d9500: yes, according to the wiki it's a 208B
19:00 d9500: and low-end is ctually what i'm looking for
19:00 imirkin: heh, well don't trust the wiki's mapping of model names to chips
19:01 imirkin: marketing people like to joke around a lot
19:01 d9500: the computer i'd be installing it in has something like a 300W power supply
19:01 imirkin: but it claims to support 4k, which means it's def a kepler
19:01 d9500: so i'd like a card that sips power
19:01 imirkin: there are some fanless GK208's...
19:01 imirkin: search for GT 630
19:03 imirkin: http://www.newegg.com/Product/Product.aspx?Item=N82E16814500367
19:03 imirkin: pretty sure that's a rebranded GT 630
19:04 d9500: and we have a winner!
19:04 imirkin: i assume you don't actually plan on doing anything 3d-related
19:04 d9500: imirkin, thanks. it's passively cooled too. the card i had, alas, finally quit spinning its fan
19:04 d9500: nah
19:05 d9500: i run lxde
19:05 d9500: don't game
19:05 d9500: don't use blender or do any modeling
19:05 imirkin: this will be *much* more than sufficient.
19:05 d9500: so likely not
19:05 imirkin: you could also just buy an older card on ebay
19:05 imirkin: can generally get them for $10 or so
19:05 d9500: like a 520?
19:05 imirkin: haha those are such crap
19:05 imirkin: GF119's
19:05 d9500: bottom of the bucket?
19:06 imirkin: almost as bad as the card i have, GF108
19:06 d9500: lousy hardware, or simply doesn't play nicely with nouveau?
19:07 imirkin: it's not exactly great hw to start with
19:07 imirkin: and fermi has no reclocking, so you're really stuck at very low frequencies
19:08 d9500: wait, kepler is newer than fermi, but it got reclocking support in kernel first?
19:10 imirkin: yes
19:11 imirkin: note that the GK208 isn't exactly a speed demon either
19:12 d9500: i'm assuming it would still beat the integrated geforce 6150se that's on the motherboard
19:12 imirkin: hehehe
19:13 imirkin: and you can only use the vga port with nouveau
19:13 imirkin: the stupid dvi port is on some external thing with those.
19:14 d9500: on the gk208's? or that pile of crap 6150 that's built-in which can't function for 10 minutes without a kernel panic?
19:14 imirkin: con the built-in 6150se
19:15 imirkin: it should largely work though...
19:15 imirkin: if you have a panic message from it, i could take a look
19:16 imirkin: i did just fix a bug that might have been panic'ing all pre-nv50 cards
19:16 d9500: alas, that computer is now lying on the op table awaiting surgery
19:16 d9500: aka on my floor
19:16 imirkin: hehe
19:17 d9500: interesting about the panics with cards prior to nv50 though
19:17 imirkin: http://cgit.freedesktop.org/~darktama/nouveau/commit/?id=d4df53da5aa302d156eebfbdfcae041d5b1a85d5
19:17 imirkin: it killed on my nv44 and i was trying to figure out why
19:17 d9500: i've been running a dinosaur of a dell laptop with an nv30 for a couple years now with very little trouble, besides of course the card being slow and outdated
19:18 d9500: i'll check that bug out
19:18 d9500: commit
19:18 imirkin: well, there's only a certain likelihood of hitting it
19:18 imirkin: basically the number of commands has to overflow the buffer size by 4 or fewer
19:18 imirkin: so it's far from a sure thing to get hit
19:19 d9500: it'll be in kernel 4.2?
19:20 imirkin: mmmaybe
19:21 imirkin: skeggsb missed the 4.2 feature pull, but perhaps he'll make a fixes push at some point
19:23 d9500: i can wait
19:23 d9500: i've no idea how long compiling a kernel from the nouveau tree would take on a pentium M, but i'm not about to attempt it
19:28 d9500: anyway, thanks again for the help...i'll look into a kepler like the one you linked or a similar card
19:28 d9500: out for now...take care and thank you again..