[Ffmpeg-devel-irc] ffmpeg-devel.log.20150705

burek burek021 at gmail.com
Mon Jul 6 02:05:02 CEST 2015

[01:56:45 CEST] <michaelni> anyone alive ? Compnn maybe ?
[01:57:07 CEST] <michaelni> ive just switched samples.ffmpeg.org to the new server if peolpe want to test
[01:57:31 CEST] <BtbN> oh, there already is a new server?
[02:03:48 CEST] <BtbN> Seems to work and also seems decently fast
[02:04:48 CEST] <Compnn> yes?
[02:05:30 CEST] <Compnn> michaelni : Tracing route to samples.mplayerhq.hu []
[02:05:43 CEST] <Compn> samples.mphq resolving to old ip
[02:05:52 CEST] <michaelni> ye i just see
[02:05:53 CEST] <Compn> theres lots of domains to switch around :P
[02:06:13 CEST] <Compn> thanks for working on it michaelni 
[02:06:35 CEST] <michaelni> samples should be the new IP in a few min
[02:08:05 CEST] <michaelni> git.ffmpeg.org switched to new IP too
[02:08:41 CEST] <BtbN> shows the ffmpeg homepage now. Or is that what it allways did?
[02:09:00 CEST] <michaelni> probably
[02:09:29 CEST] <BtbN> Wasn't it a redirect to the videolan git?
[02:09:56 CEST] <michaelni> source.ffmpeg.org was and is
[02:10:01 CEST] <BtbN> ah
[02:14:10 CEST] <Compn> are the git.mplayerhq.hu server up ?
[02:14:17 CEST] <Compn> mirror of mplayer svn 
[02:14:22 CEST] <Compn> not important just curious
[02:15:56 CEST] <michaelni> mplayer svn & git are still on the old box someone still needs to look in to them
[02:16:16 CEST] <Compn> ask reimar/beastd i guess
[02:17:38 CEST] <michaelni> yes, if noone else does i might look too but its very long i touced  svn
[02:24:33 CEST] <michaelni> www subdomain switched too
[02:39:39 CEST] <jamrial> where was everything hosted until now and why the sudden move?
[02:39:45 CEST] <jamrial> it's not really explained in the ml thread
[02:48:55 CEST] <llogan> management changes in the IT department at the university in Hungary. or something like that.
[02:49:23 CEST] <llogan> IIRC, internal hosting is being moved to "cloud" service (I hate that term).
[02:50:43 CEST] <jamrial> ah, i see
[02:54:21 CEST] <jamrial> pretty crappy it ultimately happened with a one day warning
[02:56:02 CEST] <llogan> we had a notification with no known deadline on July 1
[02:56:16 CEST] <llogan> i was hoping we would have ~2 weeks
[03:15:31 CEST] <Compn> cloud is so shitty 
[03:15:42 CEST] <Compn> just constant downtime.
[03:18:49 CEST] <rcombs> does that mean it rained?
[03:34:19 CEST] <Compn> llogan : tell arpi to check the managment cv ,  usually people with these "great ideas" are fibbing about their degrees...
[03:34:20 CEST] <Compn> ;p
[04:39:40 CEST] <philipl> wm4: It's not in the latest qsv patches? ;-)
[13:40:38 CEST] <lookatmeyou> help
[13:40:43 CEST] <lookatmeyou> how to convert a raw nv21 video to bgr24 uncompressed avi?
[13:40:54 CEST] <lookatmeyou> ffmpeg -s 480x640 -pix_fmt nv21 -i video.nv21 -vcodec rawvideo -pix_fmt bgr24 video.avi
[13:41:02 CEST] <lookatmeyou> I use this command, but it prints "video.nv21: Invalid data found when processing input"
[13:41:09 CEST] <lookatmeyou> Is there anything wrong?
[13:44:09 CEST] <wm4> #ffmpeg
[13:45:19 CEST] <lookatmeyou> Thanks
[13:54:08 CEST] <BtbN> philipl, lossless behaves fine on Windows/MSVC on my 760
[13:54:43 CEST] <BtbN> Some error messages might need a higher level, so it doesn't fail without any indication what went wrong.
[13:54:54 CEST] <BtbN> Testing on my linux box now
[14:01:08 CEST] <cone-828> ffmpeg 03Philip Langdale 07master:671bdd4b0977: avcodec/nvenc: Add support for H.264 High 444 Predictive encoding
[14:01:32 CEST] <BtbN> philipl, works as expected.
[14:02:18 CEST] <wm4> philipl: what do you mean by that?
[14:02:27 CEST] <wm4> <philipl> wm4: It's not in the latest qsv patches? ;-)
[14:04:20 CEST] <BtbN> And i finaly fixed my cygwin msvc build env.
[14:04:30 CEST] <BtbN> Needed more quotes
[14:10:13 CEST] <nevcairiel> cygwin is just a terrible way to do that :-(
[14:11:40 CEST] <BtbN> Best way i found
[14:11:46 CEST] <BtbN> better than a windows command prompt
[14:13:12 CEST] <wm4> is powershell useable these days?
[14:13:36 CEST] <BtbN> If you invest enough time to learn all the strange commands it has
[14:13:44 CEST] <BtbN> I prefer a unix-like shell
[14:14:32 CEST] <nevcairiel> yeah if you know how to then powershell is fine
[16:52:12 CEST] <cone-828> ffmpeg 03Rostislav Pehlivanov 07master:e06578e392e3: aacenc: use the new function for setting special band scalefactor indices
[16:52:13 CEST] <cone-828> ffmpeg 03Rostislav Pehlivanov 07master:117b15f4a8d9: aaccoder: remove previous PNS implementation from twoloop
[16:52:14 CEST] <cone-828> ffmpeg 03Rostislav Pehlivanov 07master:38fd4c2e664a: aaccoder: add a new perceptual noise substitution implementation
[17:11:30 CEST] <cone-828> ffmpeg 03Rostislav Pehlivanov 07master:0b233900fa36: aacenc: add support for coding of IS spectral coefficients
[17:11:31 CEST] <cone-828> ffmpeg 03Rostislav Pehlivanov 07master:e8576dc8dfc9: aacenc: implement Intensity Stereo encoding support
[17:41:57 CEST] <philipl> wm4: It was a joke about the fact that qsv presumably wraps the hevc decoder
[17:42:04 CEST] <philipl> and we're getting lots of random qsv patches
[17:42:09 CEST] <wm4> ok
[17:42:17 CEST] <wm4> I wonder why qsv is so popular
[17:42:30 CEST] <philipl> because it's the only option on windwos for encoding?
[17:42:49 CEST] <wm4> yeah, but the latest patches were for linux too?
[17:43:02 CEST] <BtbN> Yeah, because it works there, too
[17:43:07 CEST] <BtbN> And there is no libva encoder
[17:43:28 CEST] <BtbN> So of course people want to use it, even though it's just another stupid wrapper
[17:44:50 CEST] <philipl> Well, none of us is volunteering to write the native encoder, for understandable reasons. So either its nothing or a wrapper.
[17:45:20 CEST] <BtbN> I already did write a libva based encoder, for vlc though. And i couldn't figure out how to do B-Frames, and rate-control. So i gave up.
[17:45:28 CEST] <philipl> and qsv is a better choice of wrapper than yami.
[17:45:45 CEST] <wm4> what rules libyami out for encoding?
[17:46:03 CEST] <philipl> We scared the intel guy away :-)
[17:46:03 CEST] <BtbN> It's just a wrapper around libva
[17:46:20 CEST] <BtbN> with questionable future
[17:46:36 CEST] <BtbN> specialy as QSV exists for free on linux now.
[17:46:38 CEST] <wm4> why not just use gstreamer
[17:46:38 CEST] <philipl> and given that qsv is already a windows dependency for a good reason, it makes much more sense to enable it on linux than use a completely different wrapper
[17:47:13 CEST] <BtbN> you mean a generic gstreamer wrapper in lavc? ;)
[17:47:26 CEST] <philipl> lavc -> gstreamer -> lavc sounds pretty awesome to me.
[17:47:28 CEST] <wm4> something like this *shudder*
[17:48:05 CEST] <philipl> Does qsv expose the vp8 encode/decode?
[17:48:31 CEST] <BtbN> The primary problem with QSV is, that it most likely will never be packaged by distributions
[17:48:49 CEST] <wm4> why not? legal shit?
[17:49:06 CEST] <BtbN> Basicaly, yes
[17:50:24 CEST] <BtbN> no idea if it's still not possible to download the SDK without an intel account, but last time i checked, you still needed one
[19:05:28 CEST] <klaxa> BtbN: i downloaded it yesterday, you still need to create an account
[19:44:12 CEST] <ubitux> so the aac patch is merged? :o
[19:45:21 CEST] <nevcairiel> no, different patch
[19:45:31 CEST] <ubitux> oh, ok
[19:45:35 CEST] <BtbN> there were a lot of aac patches recently
[19:45:58 CEST] <ubitux> 'made some (slow) progress on the selective color for those interested
[19:47:08 CEST] <ubitux> remaining is black param, relative adjustement, and the usual round trip of fate/doc/test
[19:47:25 CEST] <ubitux> then it should be good for a first draft
[19:47:37 CEST] <BtbN> Would be nice if the ffmpeg encoder could get a bit closer to libfdk
[19:50:53 CEST] <nevcairiel> psy optimizations are closely kept secrets for aac apparently
[19:52:30 CEST] <BtbN> You can keep those secret even if the source of the encoder is 100% open?
[19:52:46 CEST] <wm4> open binary!
[19:56:41 CEST] <wm4> ( http://www.vapoursynth.com/2012/10/open-binary-introducing-a-practical-alternative-to-open-source/ )
[19:57:11 CEST] <nevcairiel> BtbN: you can read the source, understand the algorithms, and re-implement it in fresh code, but of course you cant just take the code =)
[20:12:16 CEST] <cone-828> ffmpeg 03Michael Niedermayer 07master:47496eb97cff: avcodec/rawenc: Use ff_alloc_packet() instead of ff_alloc_packet2()
[20:31:06 CEST] <rcombs> where do you even get the linux qsv lib
[20:31:26 CEST] <rcombs> and can you actually redistribute binaries linking to it without angering the GPL lords?
[20:34:23 CEST] <durandal_1707> is there specification that dictates various volume metering?
[20:41:05 CEST] <ubitux> durandal_1707: ebu r128?
[20:41:48 CEST] <durandal_1707> there are others too
[20:42:37 CEST] <durandal_1707> I want to add them to avf_showvolume
[20:42:55 CEST] <J_Darnley> replaygain, maybe?
[20:43:37 CEST] <durandal_1707> no, that is something else
[21:43:10 CEST] <wm4> always nice to see when FFmpeg adds new public API without any documentation: dc92464aee01a78c9963f86f24600ddf9c825258
[21:43:39 CEST] <wm4> you can tell just by looking at the code, and running git blame always confirms it
[21:43:56 CEST] <wm4> Libav still has a good reason to exist after all
[21:45:15 CEST] <kierank> rcombs: feel free to pay for medialooks if you don't like the gpl lords
[21:45:27 CEST] <kierank> I'm sure those binary blobs will work well
[21:45:39 CEST] <kierank> or you could pay $1000 for their ffmpeg wrapper
[21:46:00 CEST] <rcombs> kierank: quality
[21:47:40 CEST] <wm4> what $1000 ffmpeg wrapper?
[21:48:25 CEST] <kierank> wm4: http://blog.medialooks.com/0LyGa6/
[21:48:54 CEST] <wm4> ah, that one
[21:49:03 CEST] <kierank> aka a lot of broadcast software
[21:49:07 CEST] <kierank> ffmpeg with a flashy gui
[22:55:49 CEST] <philipl> I just realised cuda has vdpau interop. So it's theoretically possible to do vdpau -> nvenc without readback.
[22:55:56 CEST] <philipl> I wonder how much you gain by doing it.
[23:00:22 CEST] <nevcairiel> less cpu usage, probably no speed up
[23:00:27 CEST] <nevcairiel> nvidia is quite efficient with that
[23:05:54 CEST] <Compn> isnt that what the whole quicksync intel qsv is supposed to do ?
[23:06:04 CEST] <Compn> once its up and committed...
[23:08:33 CEST] <cone-828> ffmpeg 03Michael Niedermayer 07master:0e7bbdbeaf7f: avcodec/avuienc: Initialize output data
[23:08:34 CEST] <cone-828> ffmpeg 03Michael Niedermayer 07master:c0db6320a639: avcodec/avuienc: Use ff_alloc_packet()
[23:08:59 CEST] <cone-828> ffmpeg 03Michael Niedermayer 07release/2.7:a7fa1c9b2b18: ffmpeg_opt: Fix forcing fourccs
[23:24:07 CEST] <philipl> I went and built the latest nvenc sdk samples and we're way slower than the example encoder.
[23:27:31 CEST] <philipl> I can't gte the encode frame rates that nvidia document (1111fps) but I can get 854 from the example and we're ~390
[23:31:02 CEST] <Compn> wasnt there an nvenc from nvidia ? and someone rewrote their own implementation ?
[23:53:08 CEST] <philipl> Compn: I mean, the example app in the nvenc sdk is much faster than using nvenc through ffmpeg
[23:53:23 CEST] <philipl> BtbN and I have discussed this before. There's no obvious reason.
[00:00:00 CEST] --- Mon Jul  6 2015

More information about the Ffmpeg-devel-irc mailing list