00:24khronosschoty: hi I noticed that after upgrading beyond kernel 4.4.38 that when I resume from suspend to ram the screen does not come back on. If I press the power button however once dropped back into tty1 the screen comes back on. I was wondering if this might be a nouveau related issue and how I might go about finding out?
00:26imirkin_: what GPU?
00:28imirkin_: unfortunately that's a 1+ year old kernel, so it can be difficult to remember what all has changed since then
00:28imirkin_: nearly 2y
00:28khronosschoty: nvidia nvs 4200m
00:29karolherbst: khronosschoty: what was the latest kernel you tried?
00:29imirkin_: that's GF108 right?
00:29karolherbst: and the earliest one with the issue
00:29khronosschoty: series is latest i tried
00:29imirkin_: pastebin dmesg after this happens
00:30imirkin_: (4200M seems to be GF119)
00:30imirkin_: how is the screen connected?
00:40khronosschoty: its a laptop
00:41imirkin_: LVDS or eDP?
00:41imirkin_: nothing in dmesg from nouveau
00:42imirkin_: i wonder if it's just like backlight not being restored
00:42khronosschoty: that is it
00:42imirkin_: all that stuff changed in some linux kernel
00:42khronosschoty: I think
00:42imirkin_: not necessarily related to nouveau
00:42imirkin_: do you know which bl controller works for you?
00:42imirkin_: is it the acpi one, or the nouveau one?
00:42khronosschoty: no idea
00:42imirkin_: check in /sys
00:43imirkin_: like /sys/class/backlight maybe?
00:43imirkin_: (i don't remember)
00:43imirkin_: [ 916.320921] video LNXVIDEO:00: Restoring backlight state
00:43imirkin_: that indicates that acpi is winning
00:44imirkin_: try booting with acpi_backlight=vendor
00:44karolherbst: imirkin_: I am wondering if there is no way to fix this without such parameters....
00:45imirkin_: karolherbst: well, it's been fixed back and forth many times
00:45imirkin_: basically people ship broken ACPI things
00:45karolherbst: well right.... I guess on windows there are simply blacklist or vendors ship crappy drivers "fixing" this?
00:45imirkin_: yeah no clue.
00:45imirkin_: hansg spent a bunch of time fixing it
00:46imirkin_: i believe after he was done the situation was much improved
00:49karolherbst: I bet we will never know for sure if the ACPI stuff is broken or not
00:49imirkin_: you should ask him
00:49imirkin_: (he's at RH)
00:50karolherbst: yeah, I know
00:50imirkin_: (and apparently hangs out in the RH irc chans)
00:50karolherbst: we already talked about other stuff
00:58karolherbst: ohh wow, 10 fps on min settings with in xcom2 with nouveau, that's...bad
01:05karolherbst: 30 fps on nvidia
01:10karolherbst: imirkin_: I highly doubt that this is all an issue within the shaders, but maybe it is
01:17karolherbst: mhh, spilling, meh
01:21karolherbst: imirkin_: why are locals used here? https://gist.github.com/karolherbst/4e5beaca58b7d2ba272fed4f17d94d7f
01:23karolherbst: those offset loads?
01:25imirkin_: 56: ld u64 $r0d l[$r4+0x0] (8)
01:25imirkin_: what's the original shader?
01:25karolherbst: some array stuff
01:26karolherbst: vec4 x0;
01:26karolherbst: two of those
01:26imirkin_: vec3 presumably
01:26karolherbst: no, vec4
01:26imirkin_: so ...
01:27imirkin_: there *is* something we could do
01:27imirkin_: which is to recompile the program with the value that's in c1[0x60]
01:27karolherbst: well, they don't touch .w though
01:28imirkin_: yea wtvr
01:28imirkin_: of course that's a UBO
01:28imirkin_: so any time that changes, we have to recompile
01:28karolherbst: doesn't matter anyhow
01:28karolherbst: sounds terrible
01:28imirkin_: but maybe it doesn't change often
01:29imirkin_: presumably it can only be a handful of values
01:29imirkin_: i.e. 0..5
01:29karolherbst: well on 07 I had 3 instead of 10 fps ;)
01:29karolherbst: so I don't think the recompiles are the bottleneck here
01:30karolherbst: well there are a few shaders were we spill for real though, I think I want to look at those first
01:30imirkin_: i just mean -
01:30imirkin_: we COULD recompile
01:30karolherbst: and they also use some compute shaders
01:30karolherbst: imirkin_: yeah, we could
01:31karolherbst: how much effort? if it isn't too much I could actually work on that
01:31imirkin_: a bunch... dunno
01:31karolherbst: I would basically think this is already handled by gallium in some way
01:31imirkin_: it's not.
01:31karolherbst: well if it would be designed the smart way I think
01:31imirkin_: gallium doesn't do variants.
01:32karolherbst: I see
01:32imirkin_: (it can, but only for dumb things)
01:32imirkin_: and nouveau has things set so that st/mesa doesn't do any variants
01:32karolherbst: why not?
01:32imirkin_: part 1 of this would be to add proper variants to nouveau
01:32imirkin_: because none of the dumb things are needed?
01:32karolherbst: well most likely
01:33karolherbst: but then there is the example where it is actually, but we will never know
01:33imirkin_: this allows us to get the program at glLinkProgram() time rather than at glDraw time
01:33imirkin_: i don't think you understand this stuff very clearly
01:33karolherbst: anyhow... I would have assumed that something like this kind of belongs into gallium and that the driver should just manage the compiles, uploads and using the right variant on the hw
01:33imirkin_: which is ok - it's all rather subtle
01:34imirkin_: perhaps. but that's totally not how it works.
01:35karolherbst: imirkin_: okay, so currently nouveau compiles at linking time, because we don't to variants, which is fine for now, because that won't add silly lags on draw time
01:36karolherbst: I think what radeonsi currently does is to trigger an async variant compile and swap it out on the fly later on, right?
01:37karolherbst: shaders like those are fun: https://gist.githubusercontent.com/karolherbst/eb8f1b2653cd27b05df9bb7d20afbad5/raw/14d58f3fb2c199d7c8b7008395570780487e520a/gistfile1.txt
01:38karolherbst: imirkin_: does this make any sense? bar sync u32 # $r63 $r63
01:38karolherbst: I would say no, but most likely it means something else
01:39imirkin_: yeah. that's aka barrier();
01:39imirkin_: (i think)
01:42imirkin_: that's a pretty monster shader =/
01:43karolherbst: I will see if I get frameretracer to work on nouveau
01:43karolherbst: would be fun to have it working
01:47imirkin_: this looks an aweful lot like something i've seen... popCount maybe?
01:49karolherbst: guess again
01:50karolherbst: thing about the most terrible reason why those instruction come up and then thing of something even more terrible
01:55imirkin_: (popCount == bitCount btw)
01:58karolherbst: do you really want to know what the wrote into their shaders?
01:59karolherbst: it is basically what you see in the shader output
01:59imirkin_: but it's inside a function right?
01:59karolherbst: rephrase: it is literally what you see in that shader
02:00karolherbst: "r4.zw = uintBitsToFloat(floatBitsToUint(r4.zw) & uvec2(0x33333333, 0xcccccccc));" ;)
02:00imirkin_: but is that inside a function, or all inside main()?
02:00imirkin_: i don't need to :p
02:00imirkin_: i know how those converters work.
02:00imirkin_: sometimes they define fallback functions though
02:00karolherbst: yeah well, not here
02:00imirkin_: there's about 100% chance this is bitCount or bitReverse though
02:02imirkin_: my money's on bitReverse
02:02karolherbst: okay, so frameretracer basically just works with intel
02:03karolherbst: or it can't handle prime setups
02:04imirkin_: iirc it hardcodes stuff
02:04karolherbst: yeah, seems that way
02:04imirkin_: i was asking janesma stuff about it in #dri-devel shortly after XDC
02:04imirkin_: check the logs
02:16karolherbst: well, that tool doesn't even work on intel for me
02:23imirkin_: i never got it going either
02:23imirkin_: but i gave up pretty early on
05:32imirkin: anyone have a pre-GF117/9 fermi plugged in running nouveau?
10:33Tomin: imirkin: I have GF104, is that what you are looking for? I'm currently using Nvidia drivers though, because any graphics is so slow with nouveau. Yesterday after upgrading to F27 I was using nouveau for a little bit before deciding that FX and Cura are not working fast enough with it
11:02chillfan: Any fix for missing textures in a wine directx game?
13:00tarragon: curious about something. I got over a crash on a game by deleting games own chache mechanism AND mesa's .cache/mesa* temp files.
13:01tarragon: which raisins some further questions. Should the .cache/mesa* files be deleted after mesa/libdrm/kernel upgrade??
15:55imirkin: tarragon: nah - the mesa version is encoded in there. the key changes.
16:38Tom^: so still no one that wants to take care of this 780ti i have in a box ready to be shipped? , karolherbst ?
16:39Tom^: its using valuable shelf place. :D
16:39Tom^: imirkin: or you?
17:00karolherbst: Tom^: well... I was kind if panning to get such an external thunderbolt GPU case at some point
17:00karolherbst: Tom^: but you can ship it to me nontheless
17:00Tom^: karolherbst: sure, just need some adress
17:00karolherbst: Tom^: I don't have direct access to a kepler2 GPU anyway, so it might come in handy
17:01Tom^: and il post it on monday
17:09imirkin: Tom^: i'm further, and i wouldn't have anywhere to plug it in =/
17:09imirkin: that thing needs more power than i think i have
17:09imirkin: this comp is from 2011... let's see if i can track down what PSU i got for it
17:09imirkin: i think it was a nice one
17:10imirkin: er, 2010
17:10imirkin: oh wow. that's more than i thought.
17:15Tom^: 1000w? holy smokes
17:32karolherbst: imirkin: :D for newest desktop you usually choose 600W or something :D
17:32karolherbst: even with a highend GPU
17:32karolherbst: imirkin: well, if you want you can have it
17:56imirkin: well, doesn't have to be Tom^'s - but does it have enough plugs?
17:56imirkin: i know high-end gpu's want various extra connectors
17:56Tom^: iirc you need two of those uh pcie 8x connectors
17:58imirkin: 1 x 8-Pin EPS 12V :(
17:59Tom^: hm https://www.geforce.com/hardware/desktop-gpus/geforce-gtx-780-ti/specifications . "One 8-pin and one 6-pin"
17:59imirkin: oh wait, that's the wrong thing
17:59imirkin: 2 x 6-Pin, 2 x 6+2-Pin
17:59imirkin: presumably 8-pin = "6+2 pin"
17:59Tom^: you are good to power two of these beasts at the same time then
18:23imirkin: Tom^: as long as i supply my own nuclear power plant :)
18:29imirkin: anyways, i suspect karolherbst would be the better recipient for such hw - he has a lot more time to play with this stuff than i do
18:30Tom^: sure, karol's pmed me so unless you dont want it il probably send it to him in a few days
18:31Tom^: make things a little bit easier, just saw i had to provide some kind of decleration of goods for sending to the states.
18:31Tom^: not that it isnt doable but damn you americans! :D
18:31imirkin: yeah, annoying
18:31imirkin: it's easier to ship stuff within EU i suspect
18:32Tom^: yeah, i just smack country, postal code, or well pretty much adress. go to the local post office. done
18:32Tom^: on the box.
18:33imirkin: inter-economic zone stuff is annoying
18:35Tom^: speaking of nothing, i went full blown geek. https://i.imgur.com/SmPwR3h.jpg :D no need for fan control here
18:36Tom^: kind of nice having two very silent fans keeping both cpu and gpu at ~50C on full load
18:47karolherbst: imirkin: well I don't own a desktop yet...
18:48imirkin: so we both suck is what you're saying ;)
18:48karolherbst: and I am sure I can't bring _that_ card to work
18:49Tom^: if neither of you want it now, it just means my valuable shelf space is gonna be used until you do. so hehe
18:49karolherbst: Tom^: wait until Monday, I have to discuss a few things before...
18:49RSpliet: I think I already have access to one in the computer lab here...
18:50Tom^: karolherbst: i still have pci soundcards and couple ddr2 sticks and a 60gb hdd on the shelf too. so its not like its the first thing to go if it ever gets overcrowded. :D. so yeah take the time you need
18:52RSpliet: Oh no, it's a regular GTX 780. Either way don't have a cause for it at the moment, and don't think I should turn my priorities around to find one
18:52imirkin: it'd just be useful to do proper performnace testing
18:52imirkin: since doing it on e.g. a GK208 with DDR3 is a little ... well ... not representative :)
18:53RSpliet: yeah... and otherwise for regression testing
18:53RSpliet: which would have been fun if I had time for ins sched or other perf optimisations, but... I don't really (boo)