02:45chatter29: hey guys
02:45chatter29: allah is doings
02:45chatter29: sun is not doing allah is doing
02:45chatter29: to accept Islam say that i bear witness that there is no deity worthy of worship except Allah and Muhammad peace be upon him is his slave and messenger
05:02nyef: Ah, damn. I missed a conversation with someone who has a 3D Vision setup. /-:
07:19leberus: mupuf: Hi! Regarding temperature-related writes: You told me to export _max, _max_hyst, _crit and _crit_hyst. Should I export _emergency and _emergency_hyst too?
11:11karolherbst: do falcon really not have any max/min instructions? .... :/
11:12karolherbst: ohh maybe it is fine, I just should use cmpu rather than cmp
11:42karolherbst: something on the PMU doesn't save registers properly :(
14:35MikeB_: Hello everyone. Quick question related to hardware accelerated decoding here. (I hope you don't mind)
14:35MikeB_: I'm working on academic project under which I'm supposed to analyze how hardware acc. decoding of h264 videos works. I have to modify the source code of nouveau driver to f.e. replace few frames of video with "dummy" picture.
14:35MikeB_: My question is: Is it really possible to access/modify the process of GPU's decoding or is it a "black box" ? I'd be most thankful for any advices.
14:40karolherbst: MikeB_: more or less a black box. We use firmware from the nvidia driver on the video acceleration engines.
14:40karolherbst: one could figure out how those work and write firmware from scratch though
14:41imirkin: MikeB_: it's reasonably well understood actually
14:42imirkin: depending on the specific GPU generation and what exactly you're trying to do
14:42imirkin: although replacing frames of video with a dummy picture is pretty straightforward, and has nothing to do with the decoding process...
14:45MikeB_: I'm working with VP5. Regarding the dummy picture: What I'm trying to do is to "prove" that I have an understanding of the workflow of the whole process in any significant way.
14:49MikeB_: To be honest, I got lost when looking through the source code and documentation of this project and I'm not even able to find the part of the source code where h264 decoding is implemented/called.
14:51imirkin: look for nouveau_video_vp3* in src/gallium/drivers/nouveau
14:52imirkin: bsp = bitstream processing, vp = video processing
14:52imirkin: should probably have been called vld instead, but ... o well.
14:53imirkin: those are just the helpers, the actual commands are dispatched in, e.g., https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c#n133
14:57MikeB_: So I guess what I was missing was looking at Mesa's code.
14:57MikeB_: Thank you imirkin and karolherbst
15:30mupuf: leberus: yes, please do.
15:30mupuf: and I am sorry that I did not see that karolherbst told you not to export them anymore. I could have told him why it was here to begin with
15:31karolherbst: mupuf: I am wondering if it's still an issue though? I thought we always use those constant values everywhere now
15:31mupuf: on new HW, yes
15:31karolherbst: ohhh okay
15:32mupuf: on old HW, some vbios are a little fucked up
15:32mupuf: you had no way to know
15:32karolherbst: wondering what the nvidia driver does there
15:32karolherbst: mupuf: I've kind of remeber that we talked about this over a year ago though
15:34karolherbst: mupuf: by the way, I know a place where we could do a "nouveau dev con" without having to pay anything, which rooms which can be rent for sleeping in the same building
15:35karolherbst: if somebody is interested in having something like this :D
15:37orbea: could this mplayer crash be at all nouveau related? I ask only because I can only reproduce it with hwdec. It occurs sometimes when changing tracks in a playlist and started after upgrading ffmpeg-3.2 to 3.3. https://pastebin.com/bFYN0DPn
15:37pmoreau: A few days for drinking beers and hacking on Nouveau together sounds like a lot of fun! :-)
15:37karolherbst: pmoreau: yes, exactly my thought
15:38karolherbst: well alternative is to sleep at my place, so one would only have to pay for traveling
15:38karolherbst: there is also a fridge at the location
15:39karolherbst: and enough technical tools for stuff
15:41mupuf: karolherbst: a nouveau dev gathering could be fun and productive. But we should it be topical?
15:42karolherbst: what do you mean by your last question?
15:43imirkin: MikeB_: there's also an extremely primitive program to operate the VP2 engine (which is somewhat different from VP3+): https://github.com/imirkin/re-vp2/blob/master/decode_frame.c
15:44imirkin: orbea: anything's possible, but it doesn't seem like nouveau is involved. that said, iirc mplayer does NASTY stuff with creating/destroying vdpau contexts that caused me some heartache for a while. i thought it was all sorted though.
15:45imirkin: orbea: that said, it's failing in malloc, which suggests that there's some corruption
15:45imirkin: orbea: might want to try running it with valgrind
15:45orbea: ah, yea, I didn't think it was nouveau related, but when I discovered it was only with hwdec I thought it'd be good to be sure, but its kind of hard to test that with other hardware here
15:45orbea: alright, I can do that in a bit
17:27orbea: imirkin: i realized a few problems, the backtrace is different each time and it doesn't seem to crash in valgrind. gdb 1 - https://pastebin.com/Kb75BJJ7 2 - https://pastebin.com/VyYN5JPr valgrind - https://pastebin.com/ExYmAEgK
17:28imirkin: ==15631== Invalid write of size 1
17:28imirkin: that's not great.
17:36television: imirkin: hey remember me from a few days ago with the FX1800?
17:38television: anyway I found new info
17:39television: linux 4.10.13-1 kills my kernel consistantly with gnome3+wayland when in gparted doing ntfs copy (idk why? maybe something with drawing the progress bar?)
17:40television: but in linux 4.4.44, everything seems very stable
17:41television: (4.4.44 on some old version of system rescue cd)
17:41television: so far that thing seems to be the most stable I've ever seen nouveau handle my hardware
17:41television: wonder if theres a regression
17:45orbea: imirkin: i added those to the mplayer issue report, but they haven't shown much interest yet. https://trac.mplayerhq.hu/ticket/2324
17:46imirkin: orbea: you mean in the past 3 minutes? :p
17:47orbea: no, I opened the issue 3 weeks ago
17:48orbea: kind of forgot about it until now when slackware current updated to ffmpeg 3.3 too and I thought someone else may run into it
17:48imirkin: do they take interest in any issues on that bugtracker?
17:48orbea: im not sure, i guess I can try the mailing list too
17:48orbea: i dont know an alternative
17:50orbea: it happens with the llvmpipe too btw...
20:40pmoreau: imirkin: Thanks for pushing the series :-)
21:35imirkin: pmoreau: yw
21:44RSpliet: karolherbst: make sure to touch bases with Hans de Goede about a nouveau village. If he's not camping with revspace he might be interested