00:12karolherbst: pmoreau: we forgot about joinats/joins in the spirv thing
01:05imirkin: karolherbst: the super-smart thing would be to avoid super-smart things
01:08glisse: karolherbst: between did you had time to look into ACPI issue on your dell laptop ?
01:08annadane: so re: nouveau and plasma freezing issues, there's a thread on freedesktop.org about someone who disabled the network manager interface and who then had stable uptime; i did that by removing plasma-nm with no joy, and still got the issues
01:09annadane: i've now unchecked "allow applications to block compositing", we'll see how that goes
01:11karolherbst: glisse: not with enough depth
01:11karolherbst: imirkin: well yeah... I just didn't know that I could put any reg into the bra in input IR
01:11karolherbst: imirkin: but now I get the felling as the input nearly everything is possible
01:11imirkin: karolherbst: only predicate
01:12imirkin: karolherbst: just look at how it's presently done in from_tgsi
01:12imirkin: and assume that if from_tgsi doesn't do it, it's illegal
01:12imirkin: not 100% true, but true enough
01:12karolherbst: yeah, in the end I am usually looking up in there
01:13imirkin: i'd strongly recommend against modifiers/etc too
01:13imirkin: a lot of stuff is written with certain assumptions in place about what kinds of things come in at various stages
01:15annadane: once i get another computer and if i'm still having these issues i can ssh in
01:15annadane: i'm surprised this isn't more widely discussed or if it is no one has any idea of why this happens
01:16annadane: like i said, someone disabled their network thingy or whatever it is and they got it to work
01:16annadane: it's, pardon my french, annoying as shit
01:17karolherbst: imirkin: yeah, I try to keep it as plain as possible
01:20karolherbst: imirkin: the next thing I will implement is basic control flow stuff. I think I got the if/then/else edges and ops correctly, but now I have to deal with register values, because there are no phi nodes
01:20karolherbst: and after that I think most of the super basic stuff is already covered
01:20imirkin: and perhaps now my indexing comments make sense
01:21karolherbst: it made sense before as well, I just hoped I could deal with that a bit later
01:21karolherbst: but I also wasn't expecting so much progress so fast...
01:22karolherbst: I already got like 50 piglit tests passing matching glsl-1.10 and execution
01:23karolherbst: and most are just missing some ops and that basic control flow thing
01:29annadane: i observe that nouveau does everything better than the proprietary driver _except_ the freezing
01:29annadane: so i would love to have a magic bullet solution
01:29imirkin: annadane: can't have it all
01:29annadane: clearly not. :/
01:29imirkin: annadane: actually as far as actual gpu tasks are concerned, nouveau does everything almost exclusively worse
01:30imirkin: however as far as integration with the open-source ecosystem goes, generally better
01:30imirkin: although the places it works well are generally limited to the environments the developers use
01:30imirkin: since e.g. i wouldn't tolerate my desktop freezing every 5 minutes
01:32annadane: nvidia confuses me. there were a bunch of articles saying they would collaborate in open source on the one hand but then they've been predatory at other times
01:32annadane: obviously you folks know about all that more intimately
01:35imirkin: practically zero collaboration
01:35imirkin: they do help out with the bringup for their Tegra parts though
01:35imirkin: and they did contribute the whole "secure firmware" loading logic
01:45karolherbst: from a dev perspective this secure firmware thing is broken beyond repair anyway
01:51karolherbst: imirkin: anyway, do you which theory/algorithm the edge categories in codegen follow? because they don't make sense from a strict dfs point of view
01:52karolherbst: like if you have if/then/else edges, not both the then and else block can have a forward edge to the next common block
01:52karolherbst: in a strict dfs sense one would have to be tree and the other cross
01:53annadane: result: still freeze, but now mouse freezes too
02:17dzll: Hi guys, is GTX1050 supported in nouveau drivers?
02:18dzll: I'm using centos, which seems contains an old version of eouveau that does not support GTX1050....
02:25eb0t: hey my video card has started freezing after a major system upgrade
02:25eb0t: the dmesg says
02:26eb0t: not sure where to start
02:26eb0t: machine keeps freezing once i startx and i am using nvidia geforce 8500gt
02:28eb0t: i also updated mesa in the process...but prior to this everything was working perfectly
02:29FireBurn|Work: I still think you'd have been quicker downgrading mesa just to rule it out :P
03:13imirkin: eb0t: you have an older GPU with less memory on it. are you, by any chance, using a new DE?
03:14imirkin: dzll: mmm... if GTX 1050 is GP107, then it's mildly supported. if it's GP108, then it's ... mildly / 4 supported
03:17FireBurn|Work: imirkin: eb0t upgraded 120 packages including going from 17.0.6 to 17.1.10
03:18FireBurn|Work: Didn't updrage the kernel though
03:22imirkin: ok, well i strongly doubt a mesa change had anything to do with it
03:22imirkin: i'd look at packages starting with the letters 'q' and 'k'
03:23FireBurn|Work: I'm not sure if he's using Plasma or not
03:24FireBurn|Work: I'm assuming you're meaning Qt & kde?
03:25imirkin: subtle, right? :)
03:27FireBurn|Work: Is that the sRGB thing?
03:28FireBurn|Work: [02:12] <eb0t> i did an update yesterday and since then it has been a big problem..not sure what to do, as sometimes when it fails and i have to hard shut down..when i start again..the machine logo is also messed up a bit the boot screen ..which makes me think i may be looking at hardware failure
03:29imirkin: if it's the guy with the G86, then yeah, hw failure - those things were right in the timeperiod of those solder joint failures
03:31FireBurn|Work: I thought that was only laptops ( I asked if it was and it wasn't )
03:31FireBurn|Work: I had an old Sony laptop that was affected
03:32imirkin: mostly laptops yea
03:34FireBurn|Work: Kernel wasn't updates, possibly a hardware issue, thought it best to rule out mesa, though it'd have to be something pretty weird to persist between boots
03:34imirkin: yeah, i mean tesla doesn't get a ton of work done nowadays
03:34imirkin: and the stuff that *does* happen is mostly obscure bug fixes, not freeze-worthy things
03:35imirkin: 17.0.x is pretty recent for those
03:35imirkin: let's see what happened between 17.0 and 17.2..
03:35imirkin: er, 17.1
03:36imirkin: i guess some things did get changed which could have caused crashes
03:36imirkin: but ... like segfault crashes
03:36imirkin: not like gpu hang crashes
03:37FireBurn|Work: I figured it was the quickest way to rule it out
03:39imirkin: yeah, really nothign interesting for nv50 in 17.1 vs 17.0
05:54dzll: imirkin: Thanks. what's GP107 and GP108? I'm not familiar with these... How can I find out if it's 107 or 108?
06:11orbea: dzll: lspci would be able to tell you
06:13dzll: orbea: thanks! So it's GP107.
06:13dzll: so in which version of nouveau is the GTX1050 supported? My system has 1.0.13, but it does not work...
06:14dzll: I'm using centos 7.4
06:32eb0t: /buffer gentoo
10:46karolherbst: anybody aware of an issue with nouveau if somebody wants to resize a firefox window from the bottom?
10:47karolherbst: some user gets a PAGE_NOT_PRESENT error
10:47karolherbst: using XWayland
13:57karolherbst: imirkin: ... implement seq is kind of annoying. it is a pure float set on equal instruction, but nir also uses that for ints. now imagine it does seq 0xffffffff 0xffffffff
14:04imirkin: wtf is seq?
14:05imirkin: read the nir docs. they explain in some detail how the op should behave, and provide a C model for constant folding.
14:06imirkin: binop("seq", tfloat32, commutative, "(src0 == src1) ? 1.0f : 0.0f") # Set on Equal
14:07imirkin: ok, so it's float. if it's being used to check that 0xffffffff == 0xffffffff, it's a NIR bug.
14:07imirkin: (or a bug in the input to NIR)
14:25imirkin: if they decree otherwise, then treat it as an integer. but it can't be both ways -- for floats, one has to flush denorms, which clearly wouldn't be good for integers.
14:34karolherbst: imirkin: I can just enable native_ints in the nir opts and then I get feq and ieq instead...
14:34karolherbst: I guess I just stop caring about the s* variants then
14:35imirkin: you definitely have to enable native ints
14:35imirkin: without native ints you get pretty broken code.
14:35karolherbst: suddenly after enabling it, a lot of tests started to pass
14:35imirkin: no native int is for float-only hw, like pre-nv50
14:36karolherbst: yeah, I just thought it might be easier to start working with nir, apperantly it isn't
14:36imirkin: native ints simplify a lot of stuff :)
14:36imirkin: but when you don't have them, you don't have them
14:37imirkin: so you gotta live with it
16:16karolherbst: uhh wow, silly issues. something is odd with the getScratch() value numbering...
16:17karolherbst: numbers are getting reused, but the number is decreasing, weird
16:18karolherbst: ohh crap. I lookup in the wrong map
19:55karolherbst: fun "ld u32 $r1 c0[$r255+0x0]" in the end result
21:19pmoreau: Time to update the kernel I am running on my laptop (I’m on 4.13.4 currently :o).
21:21pmoreau: Time to add in some cutting edge patches as well, with reclocking + clockgating.
21:23pmoreau: karolherbst: stable_reclocking_kepler_v8 seems to be your latest version, is that right?
21:23karolherbst: uhm, no
21:23karolherbst: "clk_cleanup_for_real" is :D
21:24pmoreau: I know where some more cleanup should be done ;-p
21:25pmoreau: Awesome, and that one was rebased recently, so it has a higher chance to apply nicely on top of master
21:59eb0t: i get this error when my machine freezes any ideas
21:59eb0t: nouveau 0000:04:00.0: fb: trapped write at 00408c5700 on channel 2 [1fb1f000 systemd-logind] engine 00 [PGRAPH] client 0b [PROP] subclient 0c [DST2D] reason 00000002 [PAGE_NOT_PRESENT
22:02eb0t: i am using nvidia ge8600
22:03eb0t: running on gentoo
22:03eb0t: currently had to take the card out and use motherboard graphics and seems to be working
22:03eb0t: not sure if this means the card is burnt out or if it is a software issue
22:14karolherbst: eb0t: I am sure it doesn't happen with 4.13?
22:31eb0t: ah i am currently on 4.12
22:31eb0t: ill leave the card out until 4.13 shows up
23:43Exagone313: So it seems that I wasn't the only one with a problem with linux 4.14 :P I built last 4.13 version and I have no problem on it. Is the problem identified/fixed somewhere?
23:44Exagone313: is it fixed in 4.14.5 by any chance?
23:55karolherbst: Exagone313: doubtful