00:16imirkin: RSpliet: it's common to have a self-jump at the end of every shader. i've seen the blob do that for a looong time
00:17imirkin: pendingchaos: glad you liked my little hacky idea :) i like cheating when possible.
00:18imirkin: kwizart: a297 == 3d class. method 0d78 == VERTEX_BUFFER_COUNT. 4 is ... apparently an illegal buffer count? that seems odd.
00:20imirkin: wait, that's basically the number of vertices in a draw
00:20imirkin: i'm not sure how that can ever be illegal
00:21HdkR: imirkin: Doesn't everyone love a little infinite loop at the end of your shader?
00:22airlied: we can't recover if you run off the end, but we can recover from infintei loops :-P
00:23imirkin: kwizart: well, there's a slight difference in how we do instanced indirect draws, but i can't imagine it'd matter
00:24imirkin: kwizart: my guess is some kind of concurrency messing everything up
00:34RSpliet: imirkin: I know, I've observed that many times as well... But I haven't seen it in a slot where there's supposed to be sched codes before
00:34imirkin: yeah, that shouldn't happen :)
00:35imirkin: right. 0x300 should definitely have a sched code there...
00:35imirkin: since it's every 0x40 or 0x80 bytes or whatever
00:36RSpliet: Every 0x40 I think, one in 8, 64-bit opcodes
00:37RSpliet: Anyway, it's a useless observation mainly (on a 2yo driver as well), just funny to sometimes see traces of what I only assume is ad-hoc engineering
00:37imirkin: every 4th 64-bit opcode
00:37RSpliet: That's Maxwell
00:37imirkin: or more like, one sched in every group of 4
00:37imirkin: kepler is right, one in 8
00:38imirkin: kepler was weird though ... you didn't HAVE to have the sched's
00:38imirkin: although i'm not sure you could mix a program where you had them vs didn't
00:39RSpliet: Presumably that would have the same effect as having all sched codes zero - so dropping a lot of potential for higher performance on the floor...
00:40imirkin: but yeah - definitely better to have them
00:42RSpliet: I think the rationale was "why do stuff in HW on every run of every warp if we can make the compiler sort out the insn scheduling problem for us instead"
00:43RSpliet: So HW can do fancy stuff like dual issue and minimal issue distance, but leaves the dependency tracking up to the compiler... makes sense to me tbh
00:43imirkin: just ... better docs ;)
00:43imirkin: and also ... that's what they said for alpha
00:44imirkin: mattst88: were you ever able to find a baseline for that iMac G4?
00:44RSpliet: I don't think I was a conscious human being when people said things about alpha :-P
00:44mattst88: imirkin: not yet, and I suspect I won't have time until after XDC
00:44imirkin: ah ok
00:45imirkin:has a weakness for useless old hardware
00:45mattst88: my CD build failed last night for some reason. that one is supposed to just have OF FB enabled without nouveau or nvidia_fb
00:45mattst88: imirkin: me too :)
00:46imirkin: thankfully practical space considerations prevent me from going too crazy with that stuff
00:46imirkin: i have a drawerful of nvidia boards, that's about it
00:46imirkin: (well, and the G5)
00:46mattst88: you're in NYC?
00:46HdkR: The G5 helps prevent you from going crazy?
00:46mattst88: The G5 keeps him warm in the winter :)
00:47imirkin: drowns out the street noise too
00:47gnarface: it's like sleeping with 5 coffee machines :)
00:47RSpliet: Altavista was created by DEC... mind blown
00:48mattst88: and Amazon ran on a bunch of DEC Alphas :)
00:59imirkin: the good ol' days
03:42rhyskidd: oh hello: https://pastebin.com/2hUNTuBc
03:42rhyskidd: PCI device: 0x10de:0x1e87, class 0x030000
03:45imirkin: is that GP104 or GV104?
03:45imirkin: looks like GV104
03:47rhyskidd: has GDDR6
03:47imirkin: ah cool
05:45`ani`: <andreas> [ 371.592322] nouveau 0030:01:00.0: disp: ERROR 1  02  chid 0 mthd 0000 data 00000400
05:45`ani`: <andreas> [ 371.592362] nouveau 0030:01:00.0: disp: ERROR 1  02  chid 1 mthd 0000 data 00000400
05:45`ani`: <andreas> https://pastebin.com/7jGj0VJ8
05:46`ani`: I don't know what any of these messages mean.
12:56karolherbst: `ani`: something with enabling the display went wrong
12:56karolherbst: or well, while starting up the display engine
12:57karolherbst: `ani`: do you have any displays connected when nouveau gets loaded?
12:57karolherbst: and maybe a full dmesg would help
12:57karolherbst: ohh wait
12:57karolherbst: different person affected
13:03`ani`: karolherbst: yes I have a display connected.
13:03`ani`: I'm the person.
13:04`ani`: do you want the full log or?
13:08karolherbst: I am sure skeggsb might have some ideas maybe
13:08`ani`: I have issues with booting first that I need to sort out before giving the log.
19:16bkeys: Is there a table or wiki where I can see which models of GPU work well with Nouveau
19:20bkeys: Is it safe to assume that the cards listed here will work good?
19:29kernel-3xp: only thing i know, fermi reclocking doesnt work
20:10Lyude: karolherbst: might have someone else with some GP108M PMU related issues poke their head in here
20:14karolherbst: Lyude: what do you mean?
20:14Lyude: karolherbst: their kernel is hanging on the GPU init, and the backtrace seems to indicate it's related to the PMU bringup, about to post it in just a moment
20:14karolherbst: ohh, I see
20:18Lyude: karolherbst: https://lyude.net/~lyudess/tmp/int-miner-backtrace.txt
20:18Lyude: this is an older kernel so it may or may not already be fixed, but I figured it was worth poking you
20:19Lyude: (apparently they run gentoo normally, so getting feedback on whether or not a new kernel fixes this will take a little while)
21:08Lyude: karolherbst: looks like it still happens on 4.18.8
21:09karolherbst: Lyude: I will be able to take a closer look in roughle 30 minutes
21:09Lyude: karolherbst: sgtm
21:52karolherbst: Lyude: https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/commit/?id=85c5d90fc155d78531efa5d2b02e92aaef7e4b88
21:52karolherbst: ohh wait
21:52karolherbst: forget it
21:52karolherbst: it was for !gp108
21:53karolherbst: mhh, nvkm_pmu_reset fails
21:53karolherbst: skeggsb: ^^
21:53karolherbst: know anything about fails inside nvkm_pmu_reset?
23:08Lyude: someone mind letting in the email I just sent to nouveau's ML? got put on hold because of all of the recepients again :s