02:22wutno: thank imirkin, that was it :) got llvm10 and it just werks
06:31pmoreau: karolherbst: one ack could be good even if it’s not enforced by the system. It will most likely be unneeded in most cases, but still good to have a second pair of eyes to look at it and see if there are obvious mistakes like me forgetting to update one of the rows of the table.
15:15karolherbst: imirkin: btw, somebody already fixed the NIR regression in nouveau so with that merged we can enable the shader-db runs on volta+ :) https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8585
15:15imirkin: go for it.
15:17karolherbst: I left a comment, maybe the MR author picks it up :D
15:24imirkin: fk. looks like the "fd import" logic in libdrm_nouveau didn't account for the possibility of a process importing a fd into itself
15:26karolherbst: uhh.. annoying
15:28imirkin: i don't immediately see how though
15:28imirkin: looks fine
15:28imirkin: but the results definitely aren't :)
15:29emersion: yeah, wlroots likes to import DMA-BUFs into itself
15:29emersion: … and GNOME was looking into doing something similare
15:29imirkin: it's a reasonable thing to do
15:29emersion: it sounds strange, but it makes our life easier
15:29imirkin: and does appear like it should work though
15:29imirkin: just need to debug why
15:30imirkin: (why it doesn't, that is)
15:30imirkin: or perhaps it works fine and i just can't read - another popular option
17:06MrFlyback: I'm looking for a graphics card for my computer that is supported by nouveau
17:06MrFlyback: Can anyone suggest an affordable card with good support?
17:07imirkin: define 'good support'
17:08MrFlyback: Good support as in working
17:08imirkin: define 'working'
17:08imirkin: i'm not just being a jerk - it means diff things to diff people
17:08MrFlyback: One that can show me my youtube videos and episodes of the golden girls :P
17:08MrFlyback: It can be new, it can be old
17:08imirkin: gotcha. so acceleration isn't a concern
17:09imirkin: all you really want is something that will put up a picture on your monitor(s)
17:09MrFlyback: well, if it can do some games, that is good too
17:09imirkin: speaking of ... what's your (desired) monitor setup?
17:09MrFlyback: Single monitor, FHD
17:09imirkin: FHD = 1920x1080 right?
17:10imirkin: ok. well i think you'd be reasonably well-served by almost any cheap nvidia GPU. they all have approximately equally poor stability when acceleration is in play, but you can disable acceleration.
17:10imirkin: a recent cheapo gpu would be a GT1030
17:11imirkin: but there's plenty to be found, anything that's like GT6xx or higher number.
17:11imirkin: if you're looking to play video games, you're going to be better served by the GT(X) 6xx/7xx series GPUs
17:11MrFlyback: I don't play any demanding games
17:13imirkin: all that said, i'd really recomend against getting an nvidia gpu at all
17:14imirkin: nouveau is fun and all, but in terms of stability for end users, esp in modern desktop environments, it ain't great
17:14MrFlyback: So i'm better off with AMD?
17:14imirkin: i think so. you'll find people who've had problems with AMD of course.
17:15imirkin: that's not to say we don't try to make it a good experience, but there's a lot more effort going into the AMD driver than nouveau
17:16HdkR: AMD RX 580 on the AMD side works pretty well and is relatively cheap :P
17:17MrFlyback: I found a HD 4850
17:17MrFlyback: I'm looking for silent cards
17:18HdkR: silent or passive?
17:18MrFlyback: Passive, yes
17:18HdkR: Ah, nothing in AMD class that is 100% passive. Only ones that turn fans off under low/no load
17:19MrFlyback: Guess i'll stick to my laptop then
17:19MrFlyback: HdkR, the one i found has 0% fans though
17:19imirkin: some nvidia boards have passive cooling
17:20MrFlyback: there are lots of passive cards
17:21ccr: I wonder why this question about passively cooled (nvidia) cards seems to pop up rather often
17:21imirkin: they're the only ones that make passively cooled GPUs
17:21imirkin: or rather, only their chips are used in passively cooled boards
17:22imirkin: except for some truly ancient AMD ones
17:22imirkin: like HD3xxx/4xxx era
17:22HdkR: Hopefully with RDNA2 having better perf/w, low end passive will come back to them
17:22MrFlyback: I'm looking at a passively cooled HIS Radeon HD 6670 now
17:22imirkin: yeah, exactly.
17:22MrFlyback: I'm fine with older cards
17:22RSpliet: Well, actually
17:22imirkin: iirc that's terascale
17:23imirkin: HD 6670 is TURKS
17:23imirkin: which is NI
17:24HdkR: hm, the videocardz link through on that one show it going in to a server chassis. Hopefully that card is actually passive rather than needing chassis airflow
17:24imirkin: support for it is basically abandonned. it still works ok though.
17:24imirkin: (NI = northern islands)
17:24HdkR: It's hard to recommend something with Terascale these days though :D
17:24MrFlyback: So i'm not gonna have a good time with a passively cooled card, ok
17:24RSpliet: HdkR: yes, that's the main target for passive cards. Had quite an elaborate description of why that makes sense, with the server providing plenty of airflow already and now having fewer moving parts in the server to breal
17:25imirkin: MrFlyback: all in one's definition of a good time ;)
17:25HdkR: RSpliet: Yes, just wouldn't make sense to a consumer that can't hit those CFM levels in a silent build :)
17:25imirkin: MrFlyback: otoh, those boards are pretty cheap, so buy one and see if it meets your needs
17:26RSpliet: But just wanted to be "that guy" pointing out AMD does do stuff :-P
17:26RSpliet: Or well, not AMD but... you catch my drift
17:26imirkin: RSpliet: you mean "sell GPUs which need fans without the fans"?
17:26RSpliet: imirkin: Exactly!
17:27HdkR: It's the perfect market, Nvidia gets in on that with Tesla as well :D
17:27HdkR: Tesla branding, not uarch
17:27imirkin: yeah, and in a server chassis, that makes perfect sense
17:27MrFlyback: I'm just gonna follow what h-node says
17:27imirkin: those fans are just ones that didn't make it past QA on the jet assembly line
17:29RSpliet: Sooo, the ones they used to stick on the GTX290
17:29imirkin: yes. but like 5 of them :)
17:32ccr: passively cooled cards to be used only in siberian winter
17:33imirkin: ccr: yeah, the fans freeze there, so not much cohice
17:38imirkin: [obv fans don't freeze, but cars do -- gets too cold for the oil, so it turns to sludge]
17:39ccr:lives in northern-ish finland, and is aware of these things :P
17:40imirkin: is there a southern part of finland? i thought it was all north... :p
17:40ccr: depends on one's point of view, I suppose .. :P
18:12karolherbst: imirkin: I think we have to silence the unknown cap warning :/ now it pollutes the CI log :D
18:14imirkin: i added stuff
18:14imirkin: feel free to add more stuff
18:14imirkin: i don't want to get rid of the warning
18:15imirkin: it tells us what "new" things we need to look at
18:24karolherbst: yeah.. I know
18:24karolherbst: maybe we should add a test and add it to CI to assert on anything new? :D
18:24imirkin: then just remove the default
18:24imirkin: and make it a compile error
18:25imirkin: the point though was that people were sick of having to modify drivers when they added new caps
18:25imirkin: although tbh i didn't see what the big deal was
18:25imirkin: and i still wanted to evaluate the new caps that were coming in
18:25imirkin: so i kept a warning
18:25imirkin: i mean a debug warning
18:58karolherbst: imirkin: with that nir fix on Turing: [17159/17159] Pass: 11006 Fail: 36 Skip: 6094 ExpectedFail: 0 UnexpectedPass: 0 Crash: 23 Timeout: 0 Missing: 0 Flake: 0 Duration: 23:06 Remaining: 0 :)
18:58karolherbst: CTS + GTF
18:59karolherbst: ohh all issues inside the CTS, nice
18:59imirkin: got a list of CTS fail?
18:59imirkin: (+ crash)
19:00imirkin: i only ran the GL 4.5 cts, but there were no failures on GP108
19:00karolherbst: I know that most is nir related
19:00imirkin: KHR-GL46.shader_storage_buffer_object.advanced-unsizedArrayLength -- ok, that's a lot of them, i guess you screwed something up in the nir
19:00karolherbst: need to rework some bits
19:00karolherbst: layout stuff
19:01imirkin: well, you probably just don't get the length right
19:01imirkin: or the length means something else in nir
19:01karolherbst: something more annoying. I know what this is about
19:01imirkin: yeah, that _mostly_ looks like nir fail
19:01karolherbst: I just never bothered with fixing as it really doens't matter in real world scenarios
19:02karolherbst: fixing the other easy to fix things first :D
19:02imirkin: KHR-GL46.texture_cube_map_array.sampling -- that one might have nothing to do with nir. or you're forgetting *6 or something
19:03karolherbst: what *6 by the way?
19:03imirkin: cube arrays are 6 faces
19:03karolherbst: sure, but TGSI also doesn't have a *6 around
19:03imirkin: i think you have to *6 stuff somewhere
19:04imirkin: oh, maybe not?
19:04imirkin: maybe just for images
19:04karolherbst: might be.. anyway, I will figure it out
19:04imirkin: or could be in glsl_to_tgsi
19:05karolherbst: the shared mem stuff: SHARED_CONFIG_TOO_SMALL :)
19:05karolherbst: might be
19:05karolherbst: I actually thought I fixed the shared memory size stuff...
19:07karolherbst: maybe it's something turing specific
19:07imirkin: hm, i don't see the *6. i could be wrong.
19:08imirkin: it's been ages since i added support for it.
19:08karolherbst: "SHARED_MEMORY_SIZE : 0x10000"
19:12karolherbst: ehhh.. please no
19:13imirkin: yes. this.
19:19karolherbst: mhh " GL_MAX_COMPUTE_SHARED_MEMORY_SIZE = 49152" okay, at least we don't expose too much