[Ffmpeg-devel-irc] ffmpeg-devel.log.20170103

[burek]

[00:05:57 CET] <durandal_1707> in my fft decoder subframes do not line up closely sometimes. what's cause?
[00:18:42 CET] <BBB> are you missing an overlap?
[00:27:41 CET] <atomnuker> bofh_: QMF? but that's only used in like mp3 and atrac, no?
[01:00:52 CET] <cone-382> ffmpeg 03Andreas Cadhalpun 07master:91e6a64d2ea6: wmavoice: truncate spillover_nbits if too large
[01:00:53 CET] <cone-382> ffmpeg 03Andreas Cadhalpun 07master:e8651f51aa46: wmavoice: validate block alignment
[01:35:00 CET] <bofh_> atomnuker: ...and SBR
[01:35:25 CET] <bofh_> complex, mercifully, not cosine
[02:08:25 CET] <kierank> BBB: was a idct scalefactor nug
[02:08:26 CET] <kierank> bug
[02:11:41 CET] <BBB> cool, great job figuring that out
[02:11:54 CET] <BBB> (given that you dont have a source reference that is in any way useful ;) )
[02:13:48 CET] <kierank> going to be hard to get it to fit nicely in mpegvideo.c
[02:48:47 CET] <BBB> kierank: welcome to mpegvideo.c coding ;)
[03:26:00 CET] <cone-382> ffmpeg 03Michael Niedermayer 07master:4221c68edbc8: avformat/flvenc: Check for extradata allocation failure
[03:26:01 CET] <kierank> fixed 444 (turns out file was rgb) as well
[03:53:59 CET] <BBB> kierank: very good!
[03:54:10 CET] <kierank> can't get mpegvideo.c to like rgb though
[03:54:13 CET] <BBB> now to merge your changes back upstream (see you in 6 months)
[03:54:15 CET] <kierank> yeah
[03:54:28 CET] <BBB> why wouldnt mpegvideo like rgb?
[03:54:35 CET] <kierank> some pixfmt change thing
[03:54:40 CET] <BBB> oooo
[03:54:45 CET] <kierank> seems to have some assumption for yuv somewhere
[03:54:47 CET] <BBB> right, you need to change pixfmt as well as colorspace
[03:54:49 CET] <BBB> right
[03:54:52 CET] <BBB> all over the place
[03:55:01 CET] <kierank> if I put rgb at the top of the pix_fmts list in mpeg4videodec it works
[03:55:09 CET] <kierank> but if I put it below yuv422p10 it doesn't...
[11:17:41 CET] <JEEB> hmm, was there a way in FATE to show results for a specific hash?
[11:19:45 CET] <JEEB> for example if I'd  like to see all results for 4221c68
[15:02:00 CET] <wm4> the troubles of limited RGB and such: http://kodi.wiki/view/Video_levels_and_color_space
[15:02:03 CET] <wm4> (fuck HDMI)
[15:04:33 CET] <JEEB> yes, limited RGB is something funky
[15:14:38 CET] <jkqxz> That page is written as if the settings are likely to actually work.
[15:19:43 CET] <cone-540> ffmpeg 03Tobias Rapp 07master:5796048f6a1e: avformat/avidec: clean-up identifier names
[15:21:00 CET] <wm4> I think the idea is that the settings sometimes works, and then you can be happy
[15:21:07 CET] <wm4> if not, add more hacks
[15:33:16 CET] <durandal_170> BBB_: i got overlap but now I get tones which should not be there
[15:33:54 CET] <BBB_> hum ...
[15:34:07 CET] <BBB_> why is my nick always taken
[15:35:18 CET] <RiCON> BBB: /msg nickserv help ghost
[15:35:24 CET] <BBB> RiCON: yeah I did that
[15:35:34 CET] <BBB> RiCON: but it takes a msg ;) I wish it was automatic
[15:35:53 CET] <durandal_170> simple sine wave gets corrupt after some frames
[15:35:58 CET] <RiCON> it's automatic in certain server, not sure if on freenode
[15:36:20 CET] <Gramner> some clients have auto-ghost features as well
[15:36:24 CET] <durandal_170> this codec have big internal cache
[15:38:04 CET] <BBB> how did you do the overlap?
[15:42:02 CET] <durandal_170> this is qdmc codec
[15:42:20 CET] <durandal_170> it have frame of 32 subframes
[15:43:15 CET] <durandal_170> I decode tones from packet which are grouped in 5
[15:43:48 CET] <durandal_170> each tone have offset into one of subframes
[15:44:35 CET] <fritsch> bofh_: did you decide on padding vs prime + butterfly yet?
[15:44:56 CET] <durandal_170> meaning if single bit is wrong in first subpacket  next frame is bad
[15:45:13 CET] <durandal_170> *subframe
[15:53:46 CET] <BBB> durandal_170: wait, dont we have a qdmc decoder already?
[15:53:52 CET] <BBB> durandal_170: or youre trying to fix a bug in it?
[15:54:17 CET] <durandal_170> BBB: qdm2 we have
[15:54:29 CET] <BBB> I always thought we had qdmc as well as qdm2
[15:55:00 CET] <durandal_170> older is supported via audiotoolbox in lavc
[15:55:42 CET] <BBB> I see, no source decoder
[15:55:45 CET] <BBB> how irritating
[15:55:55 CET] <BBB> qdm2 gives no inspiration for how it might work?
[15:56:11 CET] <BBB> (Im expecting the two to be similar, like how various vpN codecs are related also)
[15:56:15 CET] <durandal_170> nope this one is much simpler
[15:56:51 CET] <durandal_170> I think I got some tone scales wrong
[15:57:26 CET] <BBB> simple = nice
[15:57:35 CET] <BBB> thats cool that youre working on it
[16:00:10 CET] <cone-540> ffmpeg 03Bela Bodecs 07master:557c0df9a854: avformat/hlsenc: size and duration in segment filenames
[16:01:05 CET] <durandal_170> BBB: see last change in github.com/richardpl/FFmpeg/tree/qdmc
[16:04:48 CET] <BBB> oh right an inverted table
[16:04:52 CET] <BBB> that would be hard to debug
[16:04:55 CET] <BBB> gj
[16:57:07 CET] <bofh_> fritsch: by which you mean gotten around to running perf analyses on both? not yet, sec.
[16:58:18 CET] <bofh_> durandal_170: cool. out of curiosity what the heck even used qdmc? even qdm2 seems rare and that iirc very rapidly deprecated qdmc. I've never found a sample in the wild of it.
[16:59:39 CET] <durandal_170> bofh_ :used prior to mp3 era
[17:09:35 CET] <BBB> Ive seen a qdmc sample once
[17:48:25 CET] <durandal_1707> BBB: so what I should look for?
[17:48:55 CET] <BBB> whats the current problem? still tones that shouldnt be there?
[17:49:40 CET] <BBB> how about you email benjamin larson (he wrote qdm2, right?)
[17:49:43 CET] <BBB> maybe he would have ideas
[17:49:46 CET] <BBB> Im more a video guy :-p
[18:07:03 CET] <durandal_1707> BBB: wave diverge from reference decoder output
[18:07:20 CET] <BBB> that sounds like an accumulation of error
[18:07:29 CET] <BBB> so something is wrong earlier one causing it to go slightly off
[18:07:33 CET] <BBB> and then from there it goes haywire
[18:07:43 CET] <BBB> but its hard to say what originally causes wrong given the little information I have ;)
[22:59:58 CET] <durandal_1707> BBB: apparently decoder overreads input in those cases that gives bad output
[23:00:17 CET] <BBB> aha! a bug
[23:04:48 CET] <Compn> durandal_1707 : what pixfmt output is pixlet? yuv420p10 ?
[23:05:02 CET] <Compn> or 12 or 16 ? 
[23:05:09 CET] <durandal_1707> 16
[23:05:27 CET] <Compn> 16 only then ? i have to ask...
[23:05:50 CET] <durandal_1707> yes, only
[23:05:53 CET] <Compn> thanks
[23:05:57 CET] <Compn> makes things easier for me
[23:06:05 CET] Action: Compn bbl
[00:00:00 CET] --- Wed Jan  4 2017