00:17 imirkin: skeggsb: any ideas about this? http://paste.debian.net/hidden/f7eb0900/ -- got it when running vdpauinfo with DRI_PRIME
02:24 imirkin: has anyone used tesla vdpau on semi-recent kernels?
02:35 imirkin: skeggsb: doubly odd that it's trying to write 3f ... from what i can see, the could would unconditionally end up writing like 0x400 to *cc8
17:58 karolherbst: imirkin: btw.. was there something you wanted to test regarding astc
17:58 karolherbst: I've got the jetson nano running now
17:59 karolherbst: (I also saw the option to run the system in big endian mode... and different page sizes, but I like how my life is going now, so I skip going through this kind of pain for now :p)
18:00 karolherbst: although I don't think that tegra soc even supports big endian
18:02 imirkin: yeah, just if it works...
18:04 karolherbst: okay
18:04 karolherbst: I guess we have piglit tests for that?
18:04 karolherbst: or CTS?
18:17 imirkin: welll ... it's actually gotten tougher now
18:17 imirkin: because st/mesa has gained astc emulation
18:17 imirkin: so you have to confirm that it's actually using the hw astc format
18:17 imirkin: but deqp has *tons* of astc tests
18:17 imirkin: the ones in piglit are kinda meh
18:47 karolherbst: ahh, okay, then I'll try deqp
19:09 karolherbst: mhhh, but I keep getting SIGPIPEs :/ I think something is still broken with my arm setup
19:09 karolherbst: uhm
19:09 karolherbst: SIGBUS
21:56 karolherbst: it works \o/ https://gist.githubusercontent.com/karolherbst/5effb6a495ac9cdd810c5e6ec7168200/raw/d9773b887cec9848f7b93360841e72d3bc2b0c9a/gistfile1.txt
21:56 imirkin: OpenGL vendor string: NVIDIA ?
21:57 karolherbst: tegra
21:57 imirkin: ah
21:57 karolherbst: uff
21:57 karolherbst: did I say it works?
21:57 imirkin: you said glxinfo works
21:57 karolherbst: https://gist.githubusercontent.com/karolherbst/b4e6538b3005cf3717a9f650ab4b44c9/raw/b0f02e0b79b5c73f71eefc7c10cb872aa5c66ab1/gistfile1.txt
21:57 karolherbst: ahh
21:57 karolherbst: yeah.. glxgears, doesn't
21:57 imirkin: close.
21:57 imirkin: i'm no expert, but i'd say "something's messed up" :)
21:58 karolherbst: yeah.. probably :D
21:58 karolherbst: although tagr said it should work
21:58 imirkin: i suspect the first set of errors is more instructive
21:58 karolherbst: although we also have this weirdo modesetting situation
21:58 imirkin: forget that shit
21:58 karolherbst: so modesetting ddx, rendering thorugh the nouveau mesa bits
21:58 imirkin: just use kmscube
21:58 karolherbst: yeah...
21:58 karolherbst: that was my original goal
21:58 karolherbst: but my sdl2 app jsut doesn't work
21:58 imirkin: kmscube.
21:58 karolherbst: runs perfectly fine on the tty on my laptop
21:59 imirkin: anyways, those errors indicate that the FIFO got messed up
21:59 karolherbst: yeah.. probably
21:59 imirkin: erm, actually ...
21:59 imirkin: this is a maxwell+ situation, right?
21:59 karolherbst: yes
22:00 imirkin: 0d78 = VERTEX_BUFFER_COUNT
22:00 karolherbst: uff.. fedora doesn't have kmscube.. anoying
22:01 karolherbst: ehh, and my serial console is always messing up
22:01 imirkin: most often you see such errors if the "BEGIN" command failed
22:01 imirkin: which jives with the INVALID_OPERATION for 19d0
22:01 imirkin: er actually, it doesn't
22:01 imirkin: 19d0 = CLEAR_BUFFERS
22:02 karolherbst: I am sure something fishy is going on
22:02 imirkin: i'm gonna go back to my earlier assessment...
22:02 imirkin: "something is messed up"
22:02 karolherbst: but ... I kind of don't want to build mesa on that machine :D
22:02 imirkin: cross-compile on desktop
22:02 imirkin: that's what i always did
22:02 imirkin: used to be easy with autoconf + gentoo
22:02 karolherbst: yeah....
22:02 karolherbst: well
22:02 karolherbst: I cross compiled the kernel
22:02 imirkin: yeah, coz it uses a reasonable "just give me the prefix to your suite" approach
22:02 karolherbst: yeah
22:02 imirkin: instead of meson's "make a file with who-knows-what in it"
22:03 karolherbst: maybe we should just add support for that prefix thing for meson as well
22:03 karolherbst: should be fairly trivial
22:03 karolherbst: "failed to initialize legacy DRM" uff
22:03 imirkin: and more generally the reliance on pkg-config (even in autoconf) causes more pain than is strictly necessary
22:03 karolherbst: kmscube
22:04 imirkin: oh hm. kmscube may need help. check with robclark.
22:04 karolherbst:checks his kernel config
22:04 imirkin: (on a separate display / render situation)
22:04 imirkin: you also need to run it as root
22:04 karolherbst: yeah.. I do
22:04 imirkin: anyways, i've given up getting anyone to do things better ... meson, mesa, etc
22:04 imirkin: just causes a lot of heartache for me
22:05 imirkin: much easier to unsubscribe and move on
22:05 karolherbst: ohh
22:05 karolherbst: maybe "no connected connector!"
22:05 karolherbst: maybe I have to run it on a tty
22:05 karolherbst: not a ttyS :p
22:05 imirkin: it's not about that
22:05 imirkin: kmscube uses the first connected connector on the gpu
22:05 imirkin: to set a mode on
22:06 karolherbst: ohh
22:06 karolherbst: well
22:06 karolherbst: there is a HDMI display connected
22:06 imirkin: yeah, but like i said, kmscube may need help with the split situation
22:06 imirkin: not sure.
22:06 imirkin: are you giving it the tegra device node?
22:07 karolherbst: I give it nothing
22:07 imirkin: ok
22:07 karolherbst: ahhh
22:07 karolherbst: ./kmscube -D /dev/dri/card1
22:07 karolherbst: that works
22:08 karolherbst: even displays it on the display
22:08 karolherbst: ohhh
22:08 karolherbst: card1 is the tegra chip :D
22:09 karolherbst: that explains everything
22:09 karolherbst: imirkin: the PRIME stuff is all handled inside the tegra gallium driver btw
22:09 karolherbst: but... I think here it just selects the wrong device by default
22:10 karolherbst: mhh, but X used card1
22:11 karolherbst: "no primary bus or device found" that's... troublesome though
22:11 imirkin: yeah
22:11 imirkin: ok, so kmscube works ok?
22:11 karolherbst: yes
22:12 imirkin: that's a start
22:12 imirkin: that means something works sometimes :)
22:12 karolherbst: do we have an eglgears? :D
22:13 karolherbst: ohh, we do
22:13 imirkin: dunno if it works with gbm
22:13 karolherbst: one way to find out
22:14 karolherbst: that device is a bit funky though
22:15 karolherbst: it didn't came with any power supply and it only draws 2A from USB
22:15 karolherbst: but
22:15 karolherbst: the tegra itself is a 10W device
22:15 karolherbst: and that doesn't account for the dev board
22:15 karolherbst: so sometimes... the tegra just shuts of because it didn't get enough power
22:15 imirkin: 2A * 5V = 10W?
22:15 karolherbst: yeah
22:15 karolherbst: but.. you know, usb keyboard...
22:15 karolherbst: usb mice
22:16 imirkin: not to mention the usb rats...
22:16 karolherbst: 10W is just for the SoC
22:16 karolherbst: and then there is the ethernet port
22:16 karolherbst: but.. the tegra can do PoE
22:16 karolherbst: so I maybe want to use that instead
22:16 RSpliet: Surely nothing could wrong as long as you unplug your USB kettle and USB toaster
22:16 karolherbst: it also has a power jack for 4A though
22:17 karolherbst: but.. it's disabled by default
22:17 karolherbst: and I don't have anything fitting into it anyway
22:17 karolherbst: RSpliet: well.. my laptop totally fails to power it sadly
22:17 karolherbst: so I need to use my phone charger
22:17 karolherbst: which has... 2A
22:17 karolherbst: and my 4A charger doesn't work :D
22:17 karolherbst: well
22:17 karolherbst: with the jetson
22:18 karolherbst: inside l4t they even added a power mode daemon
22:18 karolherbst: and it essentially kills of 2 CPU cores if you go to the 5W mode :p
22:20 RSpliet: Well Tegras have a history of not running with all their cores
22:21 karolherbst: mine does
22:21 karolherbst: or what do you mean?
22:22 RSpliet: That fancy +1 core on the K1 of which I don't think anyone used it... the four A53's on the Tegra X1...
22:22 karolherbst: well
22:22 karolherbst: I only have 4 A57 cores, that's all
22:23 karolherbst: ohh, the other tegras indeed have more cores
22:23 karolherbst: oh well
22:23 RSpliet: Yep. The A53s are on the chip though. Somewhere... taking up valuable chip real-estate :-D
22:23 RSpliet: Oh the Nano... yes
22:24 karolherbst: ehhh...
22:24 karolherbst: why are serial consoles that broken :D
22:24 RSpliet: I imagine that's what they did for the "next gen Switch" to make it more power efficient. Strip out the A53!
22:24 karolherbst: I just want to use my htop.. and gdb... :D
22:24 karolherbst: mhh.. maybe it's just screen
22:24 karolherbst: but minicom was even worse
22:25 karolherbst: or maybe I should just use ssh...
22:25 HdkR: A Next gen switch sounds like some IoT controlled light switch
22:25 karolherbst: :D
22:26 karolherbst: it does
22:26 karolherbst: or well
22:26 karolherbst: just a network switch with fancy features :p
22:26 RSpliet: HdkR: yeah Nintendo is branching out to new markets...
22:26 karolherbst: for being just a 10W SoC that arm cores are indeed not that slow
22:27 HdkR: Supports QoS and has full unblocked throughput on every port. Next gen :P
22:27 karolherbst: wait, what? :D
22:27 karolherbst: isn't that the case for like... all switches already?
22:27 HdkR: Consumer things don't commonly D:
22:27 karolherbst: huh?
22:28 karolherbst: they should
22:28 karolherbst: only cisco hardware sucks :p
22:28 karolherbst: well
22:28 karolherbst: you can't achive max throughput anyway.. because various reasons
22:28 HdkR: If you look at the lowest end stuff that ends up in Walmarts or Bestbuy stores then they still hve throughput limits
22:29 HdkR: Even some of Netgears things at that level...
22:29 karolherbst: ohh, that's surprising
22:29 karolherbst: I mean.. if they only sell 100mbit witches... sure
22:29 karolherbst: but... who would buy that crap
22:29 karolherbst: *switches
22:29 HdkR: Probably someone that wants to only spend $15-$25 on a switch
22:30 karolherbst: you can get 1gbit switches for less
22:30 HdkR: I think Netgears cheapest 8 port that is full throughput is like $35
22:31 HdkR: Consumer hardware is such a wacky place of cutting corners
22:31 karolherbst: HdkR: Netgear GS908 Desktop Gigabit Switch, 8x RJ-45 (GS908-100)
22:32 karolherbst: that's like $20
22:32 HdkR: Whoa, $70 on Amazon, wtf
22:32 karolherbst: :D
22:32 karolherbst: 15€ here
22:32 HdkR: Oh no, different one. Amazon search failing
22:32 HdkR: $18
22:33 karolherbst: so... :p
22:33 HdkR: Huh, Guess I'm wrong :D
22:33 karolherbst: :D
22:34 karolherbst: imirkin: yeah... seems like that mesa-demos stuff is mostly X only
22:34 karolherbst: sad
22:36 karolherbst: HdkR: the Gbit times are long gone anyway.. we are now getting the first Tbit hardware :p
22:37 karolherbst: but seriously
22:37 karolherbst: 1gbit is way to slow
22:38 karolherbst: 5 years ago.. heck.. 2 years ago disc I/O was always the bottleneck... now literally everything else is
22:38 karolherbst: even nvme is a bottleneck today
22:40 HdkR: karolherbst: I would love to upgrade my entire network to 10gbit
22:40 karolherbst: yeah..
22:40 karolherbst: I've heard they want to make the next nvme standard based on pcie 4.0 and 8 lanes...
22:41 karolherbst: then all that becomes seriously fun
22:41 HdkR: whoa
22:41 karolherbst: yeah
22:41 karolherbst: we already have SSDs hitting the 3.0x4 limit
22:41 HdkR: Yea, we've had that for a while now
22:42 karolherbst: I think thats why intel came up witht hat optane stuff.. because the RAM bus is the only thing fast enough for insane speeds
22:42 karolherbst: but...
22:42 karolherbst: I think they messed it up on the last miles
22:43 HdkR: The new NVMe 4.0 drives are only hitting 5GB/s or so, but Samsung or someone is going to come out with a drive to quickly saturate that link
22:43 karolherbst: we are quite close to compression over nvme I think :p
22:43 karolherbst: that's really the only thing you can do today
22:44 karolherbst: but that means ssds have to be able to decompress
22:44 karolherbst: and usually people things it's smarter to do that on an fs level
22:44 HdkR: Just make all drives natively understand NTFS, HFS, Ext2/3/4, etc etc :P
22:45 karolherbst: sure :p
22:46 karolherbst: well.. anyway, something is messe up with X apps on the tegra
22:48 karolherbst: ehm
22:48 karolherbst: what's the difference between 1x MSAA and 0x MSAA?
22:48 karolherbst: ohh 1x we default to 2x
22:48 karolherbst: k
22:50 karolherbst: mhh
22:50 karolherbst: I think setting the clocks doesn't work
22:50 karolherbst: at least between the lowest and the highest pstate I see no perf difference in kmscube
22:51 karolherbst: maybe it's only slow because of memory..
22:51 karolherbst: "GPU core: -0.02 V (min = +4294967.28 V, max = +4294967.28 V)" :D
22:51 karolherbst: okay
22:53 karolherbst: okay.. something is indeed funky
22:54 karolherbst: starting sddm also shows me those errors
22:54 karolherbst: I bet it's something triggered from within glamor
22:54 karolherbst: 19d0 and 17e0 again
23:03 imirkin: my bet is that fifo updates didn't process somehow, memory coherency whatever
23:03 imirkin: dunno
23:03 imirkin: so it's processing a new fifo with old commands
23:14 karolherbst: maybe
23:15 karolherbst: anyway.. that's for tomorrow