[Ffmpeg-devel-irc] ffmpeg-devel.log.20170117

[burek]

[01:09:44 CET] <philipl> haasn: haha.
[01:09:52 CET] <philipl> I'm not aware of any cuda vulkan interop.
[01:10:04 CET] <philipl> Sounds like something that would show up in SDK 9.0
[01:10:19 CET] <nevcairiel> nvidia said they want to do that, not sure if it was released yet. not that vulkan is particularly useful for video
[01:11:55 CET] <philipl> It's overkill to be sure, but you'll see vulkan renderers for media players. Someone's working on one for mpv, IIRC.
[01:12:21 CET] <nevcairiel> its not like low-level hardware access gets you anything
[01:12:32 CET] <nevcairiel> you dont have draw calls with thousands of vertices or anything
[01:12:41 CET] <nevcairiel> just simple uploads and shaders
[01:12:42 CET] <philipl> It doesn't. But it gets you off a legacy API onto the new hotness.
[01:13:04 CET] <philipl> That's enough to motivate some people.
[01:13:07 CET] <nevcairiel> the high-level APIs like OpenGL or DirectX < 12 are not going away
[01:13:27 CET] <philipl> I'm not the one working on it :-)
[03:44:16 CET] <michaelni> ubitux, what variable in ffv1 is supposed to be undefined ? line 58 is "for (x = 0; x < w; x++)" is it w ? but there are loops using that before that dont claim its undefined
[03:46:10 CET] <michaelni> did you try to disable optimizations in gcc for the valgrind test ? (i dont have the same versions here so i cant easily repro)
[07:45:50 CET] <wm4> nevcairiel: actually vulkan might be semi-useful because better control over resource management and timing
[07:47:38 CET] <nevcairiel> that reminds me that I cursed at OpenGL for a week because its handling of vsync is so terrible
[07:47:38 CET] <nevcairiel> but DX doesnt have that issue at least
[07:50:02 CET] <nevcairiel> (also the "implicit" exclusive mode that you cant control at all was  super annoying, not sure if that happens anywhere but windows though)
[07:56:29 CET] <wm4> implicit exclusive fullscreen is a windows-only and driver-specific thing
[07:56:38 CET] <wm4> what's better with DX at vsync?
[07:56:43 CET] <wm4> they seem equally bad to me
[07:58:11 CET] <nevcairiel> opengl seems to like to waste cpu cycles while waiting for vsync, or at least all common drivers do that
[07:58:15 CET] <nevcairiel> with DX I never saw that
[07:58:40 CET] <nevcairiel> its rather annoying for simple user interfaces or video playback when people go like, wtf why is it eating my cpu
[08:00:37 CET] <wm4> I heard madvr actually polls vsync status in a loop
[08:01:24 CET] <nevcairiel> outside of exclusive mode probably, but it doesnt result in high cpu usage somehow
[08:01:44 CET] <nevcairiel> in exclusive mode you can just let the driver present frames in the vsync interval
[08:02:09 CET] <wm4> and that somehow doesn't work in non-exclusive mode?
[08:02:55 CET] <nevcairiel> not with desktop composition sitting in between, i guess
[08:03:15 CET] <wm4> sure is a mess
[10:09:33 CET] <cone-871> ffmpeg 03Matthieu Bouron 07master:adf5dc90a951: avutil/tests: add aes_ctr, audio_fifo and imgutils to .gitignore
[10:59:02 CET] <ubitux> michaelni: yeah it looks like a false positive
[11:16:59 CET] <wm4> jkqxz: could you push libav 7081620aca36e616ea96f71fd71d2703e3abae09 to ffmpeg too?
[11:17:15 CET] <wm4> that's a pretty important bug fix if you actually want to use those vdpau functions
[11:51:15 CET] <mateo`> michaelni: hi, we updated the merge commit (https://github.com/mbouron/FFmpeg/tree/merge-libav), if you can give it a try again, that would be very helpful.
[11:52:55 CET] <mateo`> we now also use h->missing_fields > 1 to trigger the call to h264_select_output_frame
[12:39:08 CET] <michaelni> ubitux, can you add a entry to tests/fate-valgrind.supp to suppress that ? otherwise spoting real failurs becomes harder
[12:39:44 CET] <ubitux> there is the noundef instance
[12:39:51 CET] <ubitux> but we should probably open a bug
[13:40:12 CET] <michaelni> mateo`, cant find more issues
[14:39:17 CET] <jkqxz> wm4:  Sure, I can do it later today.  (That's a sufficiently trivial merge that I should just do it, right?)
[14:40:44 CET] <wm4> yeah
[14:40:48 CET] <wm4> and an obvious bug fix
[14:42:41 CET] <cone-133> ffmpeg 03Anton Khirnov 07master:f450cc7bc595: h264: eliminate decode_postinit()
[14:42:41 CET] <cone-133> ffmpeg 03Matthieu Bouron 07master:bdbbb8f11edb: Merge commit 'f450cc7bc595155bacdb9f5d2414a076ccf81b4a'
[14:42:49 CET] <ubitux> yey, next one
[14:44:09 CET] <nevcairiel> I already uploaded the samples for the new fate tests in the following commits, btw, before I noticed how involved this patch was and ran out of time with holidays and vacation
[14:44:38 CET] <ubitux> the next one is also a nightmare yeah
[14:44:49 CET] <ubitux> but it's going to cleanup important stuff contrary to the one we just merged
[14:44:59 CET] <ubitux> thx for the upload
[14:45:16 CET] <ubitux> we already merged it actually, we "just" need to fix the bugs
[14:46:07 CET] <nevcairiel> I can probably help again in February when I'm back home from vacation
[14:47:12 CET] <ubitux> nevcairiel: let's hope we merged that commit since then ;)
[14:47:29 CET] <nevcairiel> Plenty more after that one ;)
[14:47:35 CET] <ubitux> yeah, 800+
[14:47:44 CET] <ubitux> hopefully not as nightmerish
[14:47:50 CET] <mateo`> hopefully there is a lot of noop
[14:47:51 CET] <ubitux> we spent about a week on this one
[14:48:28 CET] <nevcairiel> Between Christmas and going to Australia for a month I didn't get as far as I hoped last year
[14:48:50 CET] <nevcairiel> And these blockers
[14:49:39 CET] <nevcairiel> I tried to do this postinit one for a couple hours but then noticed how incompatible it turned out and basically had to start over
[14:50:06 CET] <nevcairiel> And never got around to that
[14:52:40 CET] <mateo`> michaelni: btw, thanks a lot for your help
[15:06:16 CET] <michaelni> np
[17:02:10 CET] <beastwick> Hi, I want to develop an app that can use x11grab functionality provided by ffmpeg's libraries. I see libavdevice/x11grab.c but no header I can include in my project. How do I proceed? Can I link with an object file of this file?
[17:02:28 CET] <beastwick> Should I copy the device code into my own project?
[17:02:59 CET] <beastwick> I also don't see any examples that use the code in x11grab.c anywhere else in the ffmpeg project, or at least, my results thus far haven't turned anything up.
[17:05:15 CET] <ubitux> you need to use the format/device API, but this is the wrong channel, see the topic
[17:08:20 CET] <beastwick> @ubitux, thanks, yeah I just figured that I have to use something like that, moving back to ffmpeg.
[17:56:04 CET] <cone-133> ffmpeg 03Kacper MichajBow 07master:2064a3b8dfd8: configure: Don't disable SSA Optimizer on MSVC v19.00.24218+.
[20:40:00 CET] <cone-133> ffmpeg 03Michael Niedermayer 07master:b05d8e7184f2: libavformat/mpegtsenc: support hevc with missing in stream headers like h.264
[00:00:00 CET] --- Wed Jan 18 2017