04:01sneil: I've noticed the piglit test arb_compute_shader-indirect-compute "freezes" my desktop on a GM107 (mouse moves but nothing else works)
05:15dboyan_: sneil: Which version of mesa are you using?
09:30imvr: Hello, after installing nouveau, chrome doesnt get any hardware acceleration
09:31imvr: I am on ubuntu 16.04
09:32imvr: anything wrong with my setup ?
09:32karolherbst: imvr: depends. What gpu and what kernel?
09:33karolherbst: chances are you have a maxwell card and a way too old kernel
09:35imvr: alright thanks will try to get updates
14:24sneil_: dboyan_: git version as of last night
14:32dboyan_: sneil_: Can you try reverting commit d03bfb078 and see?
14:37sneil_: will do when I'm back home
14:37sneil_: How long ago was that commit?
14:39dboyan_: Just a few days
14:40sneil_: ah, found it
14:41dboyan_: I suspect it most because it touches indirect compute handling.
14:41dboyan_: If it is problematic, we'll find out why
14:43dboyan_: imirkin: ^ any ideas?
15:35imirkin: sneil_: did this happen before too?
15:36imirkin: sneil_: i wasn't aware of any issues with indirect compute on gm107
15:40imirkin: sneil_: can you try running with NV50_PROG_SCHED=0 and see if that helps?
15:40imirkin: dboyan_: you should be able to test on your GK208 as well - same impl of indirect there
15:45sneil_: imirkin: it happened a couple of days ago too, but I don't know if my checkout had d03bfb078 then
15:45sneil_: I'll give that a try too when I can
15:46imirkin: ok, well, good luck guys. i'll be largely unavailable for the remainder of the week i think.
15:48hakzsam: I'm not aware of any compute failures on gm107 as well
15:49karolherbst: well depends on how you define "failures"
15:50karolherbst: hakzsam: with hitman Nouveau generates tons of OOB errors which are compute related
15:51karolherbst: not sure if on gm107 though
15:51karolherbst: but I am sure that's totally unrelated to that
15:53hakzsam: failures == piglit/deqp failures :)
15:58hakzsam: imirkin: btw, one might want to implement something similar to R600_DEBUG=vm,check_vm for nouveau because it's... VERY useful
15:58hakzsam: especially, when used with GALLIUM_DDEBUG
15:58hakzsam: basically, it waits N ms after every draw call, and check dmesg logs in order to detect VM faults and other things
15:59hakzsam: and it extracts current shaders, object states, etc
17:20television: nouveau 0000:01:00.0: imem: PRAMIN exhausted
17:36imirkin: hakzsam: yes... i've considered that. but i was lazy.
17:36imirkin: hakzsam: or you mean the literal functionality? hm.
17:36imirkin: hakzsam: anyways, feel free to implement
17:37imirkin: television: are you doing something weird and hitting weird issues? or regular usage?
17:41imirkin: not sure what would cause that. skeggsb is the one you're looking for though.
17:41imirkin: sounds like an object leak somewhere or something
17:42television: imirkin: also the bug mutated. in 2016, with older kern version, no wayland, same gnome though, it used to happen without any use.
17:42television: like just sitting there idle on my desk
17:42imirkin: oh, those errors are ... expected
17:42television: it'd just hang the kernel oo the point where sysrq reisub won't help anything
17:42imirkin: the gr_tlb_flush timeout thing happens. no idea why. the card wedges, we can't unwedge it.
17:43television: I think the old bug and the new one I'm having now are related
17:44television: also nvidia's propiatary drivers suddenly stopped working after most recent arch update I did last week, like login screen would show GDM just fine
17:44imirkin: yeah can't help you with blob drivers :)
17:44television: but after login, just black screen with a shitload of nvrm errors
17:44television: I know, just thought I'd mention
17:45television: imirkin: I posted here about a year ago about the other nouveau issue I've been having on the older softwate
17:45television: someone mentioned it was an "off by one" bug? and they wanted me to firescope it
17:45television: I finally got the stuff to firescope it
17:46television: I'm thinking of downgrading my system to a year ago and seeing if I can capture it
17:47television: it'd hang regularly, like just 15mins of either browsing or just sitting idle doing nothing but with screen on, it'd hang to where SysRq wouldn't be able to flush anything to disk
17:47television: no logs
17:47television: however I noticed some tells letting me know its about to crash while using it, giving me time to SysRq emergency sync at least
17:48television: (mouse cursor skips and jerks as I'm moving it, time to sysrq)
17:49television: if I hesitate more than 3 secs, it freezes to where you don't get any response from sysrq
17:52television: imirkin: I notice random blue splotches in WebGL stuff
17:52television: any idea what that is?
18:06karolherbst: skeggsb: the only problem I have with parralel runs in piglits are test which use the textureGather binary
18:06karolherbst: no idea if I will be able to look into that at some point, but maybe you know what's up?
18:08imirkin: television: yeah, long loops terminate randomly, thus generating random colors
18:09imirkin: television: https://bugs.freedesktop.org/show_bug.cgi?id=78161
18:10imirkin: and those webgl demos are often full of long loops
18:10imirkin: all the raytracing stuff
18:11television: imirkin: all of a sudden some of my webgl demos started turning into triangles with lines going everywhere
18:11television: I forgot to screenshot it
18:11imirkin: my latest thinking on that issue is that we mess something up with context switches, but ... unfortunately there's no additional thinking after that
18:11television: anyway this was the thing that showed up with weird colors
18:11imirkin: random triangles/lines happens when VBO data gets messed up
18:11imirkin: i've seen it occasionally, no clue why
18:11television: ah thanks
18:11imirkin: there are a lot of undiagnosed bugs on nv50 unfortunately
18:12imirkin: that i've spent a bunch of time trying to diagnose and thus far have been unsuccessful
18:12imirkin: unfortunately getting at the relevant states is hard, and there's no docs
18:12television: I'm considering desoldering my GPU and soldering in a better one :^3
18:13television: maybe something amd :^}
18:13imirkin: that might void your warranty :)
18:13television: LOL what warranty :D
18:13Lyude: television: if you do the foss community will thank you :)
18:13imirkin: AMD is definitely the way to go for a well-supported OSS experience
18:13imirkin: or Intel
18:13imirkin: definitely not NVIDIA
18:14television: though I've heard nV started trying to put an effort into opensourcing some junk?
18:14television: recently? like for their newer cards? idk
18:14Lyude: television: not really, no
18:14television: ah ok
18:15Lyude: they do upstream work for tegra, but that is about it
18:16Lyude: and that's really not very useful for what we need from them atm, e.g. firmware
18:16television: on a totally different subject, hashcat worked fine with my old AMD HD7970 with propiatary drivers, but a new version of hashcat came out that only supports the AMDGPU drivers, I waited and waited, and eventually AMDGPU supported my 7970, so I switched to it, but now hashcat keeps crashing on its benchmark when it never used to crash before
18:16television: running out of memory when it should have lots of ram
18:16Lyude: reported a bug?
18:17television: well I told hashcat on irc about it, they were like: eh throw out your gpu, get a new one lol
18:17television: im gonna report it later though
18:18Lyude: television: it could also be an amdgpu bug potentially
18:18imirkin: not sure if there's official SI support in amdgpu
18:18television: I'll post it on hashcat's github first though and see what they say
18:18television: they might send me to amdgpu
18:19imirkin: they'll say "driver bug"
18:19imirkin: since it clearly is
18:19imirkin: [almost clearly]
18:20endrift: hey so I asked a question a few days ago but I think I did it while everyone was asleep so I never noticed an answer: is there magic I need to do to get a GeForce 6800 GT working on my PowerMac G5?
18:20television: I like my SI gpus though :3 hope they get better support.
18:20television: amdgpu supports them enough to do non-hashcat things just fine :D
18:20endrift: I found some threads about it being broken from a few months ago but they were never resolved
18:21endrift: if that magic is "buy a Radeon" I'm ok with that, I just want to know if there's anything I can do short of that
18:22endrift: I did the 4K pages thing and the disabling MSI interrupts thing, but neither helped
18:22endrift: I'm running Linux 4.10.x (I forget the patch release) and mesa head
18:23television: endrift: I'm runing 4.10x too and amdgpu-pro and the GPU works great for desktop stuff and browsing stuff
18:23endrift: television: on PPC?
18:23television: on x86
18:23endrift: yeah I think PPC is part of the problem here
18:23endrift: or it might be. dunno
18:24television: anyway I'm going afk have fun y'all
18:24endrift: I was basically told "yeah imirkin is probably the only one who would know" so...here I am
18:25endrift: I'm willing to get my hands dirty doing kernel debugging if that's needed, too
18:25imirkin: endrift: AGP or PCIe?
18:25endrift: on another topic, I saw that nouveau isn't supported on FreeBSD and I was wondering if anyone here might know why and/or if I can do some work to port it
18:26endrift: can ask #freebsd if needed
18:26endrift: imirkin: AGP
18:26imirkin: can't say anything about freebsd or anything else. netbsd did port the code. freebsd can too if they like.
18:26imirkin: ok, so... interesting. can you try nouveau.config=NvPCIE=0 ?
18:27endrift: sure, let me boot the computer right now
18:27imirkin: i recently had an issue with a NV4A with a PCI interface, which normally has an AGP interface
18:27imirkin: however i think your case is a little different
18:27imirkin: oh, also it's a nv40, not nv41+. so it won't matter. hrm.
18:28endrift: yeah this is nv40
18:28imirkin: worth checking anyways
18:28endrift: ok, the computer is booting; I'll edit the yaboot config and reboot as soon as it comes up
18:28imirkin: i netboot mine ;)
18:28imirkin: same as i do my arm boards. heh.
18:28imirkin: it's about as powerful as one
18:29imirkin: although eats a TON more power
18:29endrift: I want to do netbooting stuff on some of my machines but I can't do PXI without majorly screwing with my network config
18:30imirkin: it took quite a bit of configuring to get all of that to work
18:30imirkin: happy to share my findings though
18:30endrift: well my router is an Airport Extreme so I'd need to replace the DHCP server entirely to do PXI :/
18:30imirkin: *obviously* :)
18:31endrift: so I'm looking at the dmesg and apparently it says "nouveau [pci-id]: pci: failed to acquire agp"
18:31endrift: that might be related....
18:32endrift: anyway same error
18:33endrift: checkerboard screen (with more static than usual) and gr: intr 00100000 [ERROR] nsource 00000010 [LIMIT_COLOR] nstatus 04000000 [PROTECTION_FAULT] ch 1 [0008e000 X] subc 3 class 004a mthd 0404 data 0a000400
18:33endrift: in dmesg
18:33imirkin: right, so ... something in the mappings is messed up
18:33imirkin: also, there is no agp, since agp is disabled on ppc
18:34imirkin: because the gart logic on ppc is ... non-functional
18:34endrift: does that mean I'm SOL unless I get a different card?
18:34imirkin: anyways, ftr, i have a PowerMac7,3 with a FX5200 (nv34) AGP and it works ok
18:34endrift: FX is a bit old for what I'm looking for
18:34imirkin: i assume you have a PowerMac7,3 as well since you have AGP?
18:34endrift: I think so yeah
18:35endrift: a 6200 would be fine if that would work
18:35endrift: I just want hardware OGL2
18:36imirkin: right, but for it to work well on mac, it needs to have an apple OF blob
18:36imirkin: so you can't just run out to best buy and buy any ol' 6200...
18:36endrift: ohhh do none of the 6 cards have that?
18:36endrift: oh yeah I was looking at ebay for "powermac g5 6200"
18:36endrift: it's like $30
18:36imirkin: that said, it's all just a regular computer
18:36endrift: I don't think this model has PCI-e, does it?
18:37imirkin: so if you're willing to lose the OF fb, you can use any board you like
18:37endrift: ah I'm looking for dual booting
18:37endrift: and PPC is specifically required for what I'm doing
18:37endrift: namely testing software on PPC with a real debugger as opposed to on the Wii :P
18:38imirkin: well if you're looking to be testing dolphin on nouveau's nv30 driver, you're going to be sorely disappointed.
18:38endrift: ha no the software isn't Dolphin
18:38endrift: although sometimes I test the software INSIDE Dolphin
18:38imirkin: also the 3d driver is pretty buggy on BE
18:38endrift: maybe I can work on that
18:39imirkin: i have a lot of confusion about what the various "BE mode" bits do
18:39imirkin: i think index buffers get over-swapped
18:39endrift: Lyude was telling me yesterday I should get my hands dirty with nouveau or some other kernel drivers
18:39endrift: I'm looking for kernel-level jobs and I have very little experience (but non-zero) in the kernel
18:39endrift: yes that was me
18:39Lyude: oooooh. hi!
18:39endrift: you kinda ruined my day fwiw <_<
18:40Lyude: I'm uh, sorry? :(
18:40endrift: anyway getting my hands dirty on nouveau would probably be good for me anyway
18:40imirkin: either way, i'm not sure what's going wrong with your setup
18:40imirkin: sounds like it might be mmu related, but i'm not sure
18:40endrift: I'll try getting a different card I guess
18:40imirkin: you can poke around in the xf86-video-nv driver to see if it does something special for your board
18:41endrift: at some point
18:41endrift: do I need a drm driver for that?
18:41endrift: or just to not modeset into nouveau?
18:41imirkin: for reading code?
18:41endrift: I thought you meant testing if nv works with the card
18:41imirkin: testing is nice too, i suppose
18:41imirkin: for that, yeah, just don't load nouveau drm
18:41endrift: I think I built that into my kernel heh
18:41endrift: but I can always rebuild
18:41endrift: it just...takes a while on this machine
18:42endrift: I need to send a patch to Linux because their g5_defconfig doesn't include support for the powermac g5, only the imac
18:42endrift: it leaves out the wifi card and PMU drivers
18:43imirkin: hmmm... i think PMU was included on mine
18:44endrift: eh I just needed to tick off the windfarm_lm72 or whatever driver and it worked fine
18:44endrift: and I needed to tick off b43 for the wifi card
18:53karolherbst: *sigh* that liveOnly thing won't be easy. Not a single piglit regression :(
18:53karolherbst: but maybe those piglit things are all too trivial for it anyway
21:57pmoreau: Alea jacta est
21:57pmoreau: Let’s see which comments I’ll get
22:05mupuf: pmoreau: yeah!
22:06pmoreau: mupuf: It was about time, after sitting for quite a bit of time on a branch, and being re-written/polished a bit once in a while
22:06karolherbst: pmoreau: it's not tight to nouveau at all, is it?
22:06pmoreau: This one, absolutely not.
22:07karolherbst: how much of OpenCL does work with this?
22:07karolherbst: under nouveau?
22:07karolherbst: or are there follow up patches required?
22:08pmoreau: If NIR gets what is missing to express OpenCL kernels, you could easily add a conversion step from SPIR-V to NIR for example.
22:08pmoreau: Follow up patches are required.
22:08pmoreau: This is only to have clover accept SPIR-V binaries, and forward them to drivers that accept SPIR-V as input. So far, no drivers does that.
22:09pmoreau: The follow-up series will have Nouveau advertise SPIR-V as a valid input IR, as well as add a `nv50_ir_from_spirv.cpp` file that does stuff.
22:11karolherbst: ahh, I see
22:39james9999: hey is this a good guide to learn to program nouveau? https://nouveau.freedesktop.org/wiki/IntroductoryCourse/
22:39imirkin: not really
22:39imirkin: best to just read the code and ask questions
22:41james9999: what tools besides vim and gcc?
22:41imirkin: have a look at envytools
22:43james9999: Tools for people envious of nvidia's blob driver.
22:44imirkin: or the phonetic version of NV tools
22:45james9999: well now i feel dumb
22:45james9999: didn't get that on first try lol
23:43karolherbst: ohhh, I got a good idea for the PMU counter implementation, nice. I guess I will finish that soon then