[Ffmpeg-devel-irc] ffmpeg-devel.log.20150715

[burek]

[00:00:17 CEST] <BtbN> I don't think dumping the stream would help. Or does mjpeg have timestamps and stuff?
[00:10:01 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:522256b374fe: avfilter/internal: Improve docs about ff_request_frame()
[00:10:02 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:bfd17046c138: avcodec/mpeg12dec: Fix decoding of faulty interlaced mpeg2
[01:02:20 CEST] <durandal_1707> ubitux: it doesn't work for bunch of filters, swscaler reports zero vector in scaling
[01:22:47 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:c9c7263e5820: avformat/mov: Fix opening relative references
[02:00:35 CEST] <jamrial> J_Darnley: you want the removegrain asm to be GPL then?
[02:18:34 CEST] <J_Darnley> jamrial: yes
[03:06:27 CEST] <Compn> has perl's logo always looked like an onion vagina ?
[03:07:37 CEST] <Compn> https://www.fastly.com/img/open_source/perl.png
[03:08:12 CEST] <J_Darnley> I see the TOR onion there
[04:10:08 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:901922e0475c: avformat/utils: adjust short skip threshold in ff_configure_buffers_for_index() to avoid seeking if a packet is skiped
[04:10:09 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:90dd6ad22cee: avformat/utils: inform the user if the protocol cannot be determined
[05:16:37 CEST] <cone-562> ffmpeg 03Michael Niedermayer 07master:dffae122d0f4: avcodec/mjpegdec: Fix DC overflow in decode_block()
[06:55:53 CEST] <charismatic> Hi! Is ffmpeg still in need of a host?
[09:22:37 CEST] <bove> Can someone help me understand this line in a demuxer? pkt->pts = pkt->dts = pkt->pos / packet_size;
[09:33:08 CEST] <rcombs> bove: looks like it's from a raw-type demuxer?
[09:33:30 CEST] <rcombs> bove: that's setting the packet's timestamp to its in-file position, divided by the size of each packet
[09:33:39 CEST] <rcombs> so the timestamp is effectively its packet number
[09:35:23 CEST] <bove> rcombs: That's right. Thanks
[09:36:54 CEST] <bove> rcombs: I'm writing a demuxer for SGO Mistika .js format, which is a 4k block aligned rawvideo format with header
[09:37:53 CEST] <rcombs> OK?
[09:44:26 CEST] <bove> I'm now able to decode all the frames, but the rawvideo decoder throws an error when trying to decode the last 4096 bytes
[09:52:36 CEST] <cone-392> ffmpeg 03James Darnley 07master:bff724260840: avfilter/vf_removegrain: add x86 and x86_64 SSE2 functions
[09:52:37 CEST] <cone-392> ffmpeg 03Paul B Mahol 07master:eea08efc0df3: avfilter/x86/vf_psnr.asm: split one line of license text into two
[10:41:07 CEST] <durandal_170> anybody tried deband filter?
[10:48:05 CEST] <bove> nope. problems?
[11:40:12 CEST] <bove> I'm trying to decode a raw picture and pix_fmt bgr24 almost works, but the picture is upside down. Is there a way to scan lines in reverse?
[11:56:32 CEST] <durandal_170> bove: there is filter for that
[13:00:47 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:8f2c045a0e21: avformat/mov: Implement a same origin policy for references instead of only allowing a subset of relative pathes in references
[13:18:53 CEST] <bove> durandal_170: Is there a way to set that from the demuxer?
[13:20:22 CEST] <durandal_170> no, why you would need that anyway, its not demuxer job
[13:21:49 CEST] <bove> It's just the way the streams are stored in that format and there's no matching pix_fmt
[13:23:02 CEST] <cone-392> ffmpeg 03Arttu Ylä-Outinen 07master:ae3f4e479a7b: MAINTAINERS: add myself as libkvazaar maintainer
[13:23:03 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:9d8070afe11a: avfilter/internal: add more words to sentance
[13:35:40 CEST] <cone-392> ffmpeg 03hSÇ 07master:2c34389551b6: configure: loongson disable expensive optimizations in gcc O3 optimization
[13:54:42 CEST] <durandal_170> bove: there is no pixel format for inverse scanlines
[13:58:34 CEST] <wm4> bove: you output raw video?
[13:58:40 CEST] <wm4> bove: I mean, the demuxer
[14:00:39 CEST] <bove> wm4: Yes. st->codec->codec_id               = AV_CODEC_ID_RAWVIDEO;
[14:01:45 CEST] <wm4> then there's no way
[14:02:14 CEST] <wm4> it's trivial to flip the decoded image, though
[14:02:52 CEST] <wm4> you could also use the displaymatrix stuff - it's normally used to signal rotation from container to decoder, but could also be used for flipping the image
[14:03:12 CEST] <nevcairiel> i think a rgb flip can even be done with negative line size
[14:04:59 CEST] <wm4> works with yuv too
[14:09:12 CEST] <bove> nevcairiel: Where would I but that? Just negative codec->height?
[14:10:49 CEST] <wm4> you can't in the demuxer
[14:11:36 CEST] <nevcairiel> if its rgb, you might be able to produce a special extradata token to make the rawvideo decoder flip it
[14:11:44 CEST] <nevcairiel> its magic from AVI raw video support
[14:12:24 CEST] <nevcairiel> ie. write the string "BottomUp" into extradata
[14:12:26 CEST] <nevcairiel> and voila!
[14:13:33 CEST] <BtbN> Won't that confuse a lot of other filters/muxers?
[14:13:57 CEST] <nevcairiel> negative linesize? yeah it doesnt work with everything
[14:14:11 CEST] <BtbN> No, stuff like that in extradata
[14:14:34 CEST] <nevcairiel> well, its not like its the only format that would have it, the logic in rawdec is there for a reason
[14:15:31 CEST] <nevcairiel> demux rawvideo from avi, and it might have this token as well
[14:15:38 CEST] <nevcairiel> (if its flipped)
[14:16:46 CEST] <bove> grep -I BottomUp libavformat/* shows multiple hits
[14:19:45 CEST] <wm4> michaelni: can you please put up all your changes for review
[14:20:32 CEST] <wm4> michaelni: commit 8f2c045a0e21559c0cdec44805bdc26457658f9c is ridiculous and doesn't make much sense from the API user perspective
[14:22:34 CEST] <durandal_170> D404|Ghetto: if you need help with vf_reverse I sent vf_random to the ml
[14:23:47 CEST] <D404|Ghetto> durandal_170: ok.
[14:23:56 CEST] <D404|Ghetto> i am still internetless at night, until Friday
[14:24:01 CEST] <D404|Ghetto> so i will see if i have time to look.
[14:24:08 CEST] <D404|Ghetto> (my mail client server is down...)
[14:29:42 CEST] <michaelni> wm4, if you want to review 8f2c045a0e21559c0cdec44805bdc26457658f9c, please do
[14:30:12 CEST] <wm4> it's a bit too late now, isn't it
[14:30:37 CEST] <D404|Ghetto> some people think review means reading ffmpeg-cvs
[14:38:48 CEST] <michaelni> Can someone review the AAC_fixed_decoder patches on ffmpeg-devel ?
[14:39:42 CEST] <michaelni> also "[FFmpeg-devel] [PATCH 1/2] apng: Support inter-frame compression" needs a review
[14:40:18 CEST] <michaelni> anyone has a opinion on "[FFmpeg-devel] [PATCH] lavc/srtdec: handle leading whitespace before tag names" ?
[14:45:28 CEST] <wm4> the aac fixed decoder thing has been in limbo for weeks
[14:45:33 CEST] <wm4> isn't apng a gsoc?
[14:47:19 CEST] <kierank> hmm where's ludmila
[14:50:54 CEST] <michaelni> apng is a gsoc project yes, was just wondering if anyone wants to/will review it or if i need to find someone to review / should do it myself
[15:54:38 CEST] <durandal_170> michaelni: I tested, and it works. But I didn't made deep inspection
[16:01:16 CEST] <cone-392> ffmpeg 03hSÇ 07master:a0ad6c85aca9: avcodec: loongson optimized h264pred with mmi
[16:01:17 CEST] <cone-392> ffmpeg 03ZhouXiaoyong 07master:95418634b1e8: avcodec: loongson optimized mpegvideo dct unquantize with mmi
[16:03:22 CEST] <adevelop> Hi
[16:34:16 CEST] <AstralStorm> uh, avcodec_open2 is supposedly not thread safe
[16:35:05 CEST] <AstralStorm> I have no idea why, since it uses ff_lock_codec and even initializes threads inside
[16:36:23 CEST] <AstralStorm> looking at the code, the lock is not taken in case of FF_CODEC_CAP_INIT_THREADSAFE
[16:36:28 CEST] <AstralStorm> I'll have to check this
[16:45:44 CEST] <AstralStorm> indeed, most codecs are not INIT_THREADSAFE
[16:46:30 CEST] <AstralStorm> this is terrible, I now have to ensure nobody attempts to use ffmpeg in the same process directly
[16:55:00 CEST] <kierank> there are static tables that need to be initialised
[16:55:12 CEST] <kierank> you need to set it up at the beginning of your process
[16:56:21 CEST] <AstralStorm> except I can't know when the codec will be used
[16:56:33 CEST] <AstralStorm> or which of those
[16:57:25 CEST] <AstralStorm> I can't even have the encoding context ready early, because the muxer might be different
[17:03:10 CEST] <ubitux> http://b.pkh.me/4k.gif ffmpeg ftw
[17:03:37 CEST] <ubitux> [] gif 4k 
[17:06:43 CEST] <durandal_170> 4k gif, how big is?
[17:08:43 CEST] <ubitux> actually not exactly 4k but well
[17:08:52 CEST] <ubitux> this one is ~5M
[17:09:04 CEST] <phh> i want moar pixels
[17:09:21 CEST] <michaelni> durandal_170, if noone does "deep inspection" then it should be ok
[17:10:59 CEST] <michaelni> after fixing the found issues that is
[17:26:27 CEST] <nevcairiel> ubitux: re videotoolbox, the hardware not being fast enough for 240fps is no valid answer? =p
[17:27:39 CEST] <ubitux> it is
[17:28:04 CEST] <ubitux> using avfoundation frameworks handle it ok
[17:28:21 CEST] <nevcairiel> arent those slowmo videos, recorded at 240, supposed to playback at 60
[17:28:33 CEST] <ubitux> yeah
[17:29:03 CEST] <ubitux> are you saying there are skipping packets?
[17:29:41 CEST] <nevcairiel> just saying that if you play them at 60, you can take 16ms per frame :)
[17:30:10 CEST] <ubitux> assuming i'm not decoding everything
[17:30:32 CEST] <ubitux> i wonder if that's sane to do
[17:31:22 CEST] <nevcairiel> you decode everything, just slower .. that way you get the slow-motion effect
[17:31:25 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:4a6b8289e646: avcodec/mips/mpegvideo_mmi: Use av_assert2()
[17:31:42 CEST] <ubitux> i can't decode everything
[17:31:50 CEST] <ubitux> at least with ffmpeg
[17:31:57 CEST] <nevcairiel> why not
[17:32:16 CEST] <ubitux> ok i see what you mean
[17:32:33 CEST] <ubitux> but yeah no i need non-slowmo playback
[17:32:54 CEST] <nevcairiel> no idea if thats really how it works, but 240 fps at 1080p or what these are is challenging for many hardware decoders
[17:33:10 CEST] <ubitux> 720p in this case, but sure
[17:33:43 CEST] <philipl> re-encode while skipping frames?
[17:35:31 CEST] <ubitux> it's just a decoding here
[17:35:50 CEST] <ubitux> i'm just wondering how the avfoundation framework (and other apps) manage to playback without slowmo 240 fps footages
[17:36:10 CEST] <philipl> Is there special metadata in the stream regarding framerate?
[17:36:11 CEST] <ubitux> while the decode calls take ~5ms with ffmpeg (which is too much)
[17:36:27 CEST] <ubitux> so assuming this is indeed the best the hw can do
[17:36:36 CEST] <ubitux> it would suggest it's just not decoding everything
[17:36:40 CEST] <philipl> Right
[17:36:48 CEST] <philipl> If the encoding is done correctly, it can skip
[17:36:51 CEST] <nevcairiel> its not like the display could show them all
[17:36:52 CEST] <ubitux> but how does it manage to do this with a 30 frame gop?
[17:37:09 CEST] <philipl> Maybe the slow-mo part is different?
[17:37:10 CEST] <nevcairiel> drop non-ref frame
[17:37:13 CEST] <nevcairiel> +s
[17:37:23 CEST] <nevcairiel> ffmpeg has discard modes for that, try one? :)
[17:37:37 CEST] <philipl> Maybe it changes the GOP mid-stream for the slow mo section
[17:37:51 CEST] <ubitux> alright, well let me try that
[18:01:48 CEST] <ubitux> nevcairiel: seems it works damn well, thx
[18:02:08 CEST] <ubitux> now i need to figure out if that's indeed what it does internally with the frameworks
[18:06:45 CEST] <nevcairiel> which discard did you use? non-ref, non-key?
[18:09:23 CEST] <ubitux> nonref
[18:39:52 CEST] <durandal_170> J_Darnley: have time to write more asm?
[18:40:19 CEST] <J_Darnley> I am an unemployed layabout, so yes.
[18:46:40 CEST] <J_Darnley> ...  What did you want me to do?  More filter work?
[18:47:57 CEST] Action: J_Darnley can never remember how to use `ln`
[18:57:12 CEST] <JEEB> if you are interested in some newer scaling/depth/colorspace conversion things, you could look into https://github.com/EleonoreMizo/fmtconv or https://github.com/sekrit-twc/zimg and see if those could be used in the context of ffmpeg
[19:14:12 CEST] <rcombs> michaelni: 'looking at it, i suspect the unknown = 1 case might "break" or rather not work correctly with this' <-- re: my srtdec patch; I don't understand the question here?
[19:14:39 CEST] <rcombs> OH, further down
[19:14:47 CEST] <rcombs> never mind, I've got it
[19:20:57 CEST] <durandal_170> J_Darnley: you can look in deband
[19:25:28 CEST] <J_Darnley> sure, i'll look
[20:07:54 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:a5d44d5c220e: swscale/utils: Clear pix buffers
[21:13:27 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:269a71b68ae7: avcodec/mpeg4videodec: Clarify the mpeg4_unpack_bframes reference.
[21:29:03 CEST] <cone-392> ffmpeg 03Luca Barbato 07master:c88c5eef53ff: hevc: Split the struct setup from the pps parsing
[21:29:04 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:4e160ecdbc05: Merge commit 'c88c5eef53ff1619724ba47b722da64ec0593dab'
[22:43:40 CEST] <cone-392> ffmpeg 03Luca Barbato 07master:4512ee78e19f: mpegts: Mark the muxer as supporting variable fps
[22:43:41 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:1680f9c58850: Merge commit '4512ee78e19fdb011bdec1b3a8dc0b315c82a81e'
[22:51:18 CEST] <cone-392> ffmpeg 03Henrik Gramner 07master:2cb34f82b92c: checkasm: Add unit tests for h264qpel
[22:51:18 CEST] <cone-392> ffmpeg 03Michael Niedermayer 07master:cbd4a1dbdeda: Merge commit '2cb34f82b92c15b811f5c03dc7f61a4baf6e02e3'
[22:59:01 CEST] <philipl> fate-suite.ffmpeg.org doesn't provide access to the samples anymore.
[22:59:10 CEST] <philipl> I assume this is related to the host move.
[22:59:28 CEST] <philipl> But for example, this used to be the url to get a specific sample: http://fate-suite.ffmpeg.org/hevc-conformance/LTRPSPS_A_Qualcomm_1.bit
[23:03:51 CEST] <nevcairiel> i assume rsync still works?
[23:06:22 CEST] <philipl> I think it works.
[23:07:42 CEST] <philipl> Yeah, it does.
[23:15:02 CEST] <cone-392> ffmpeg 03Donny Yang 07master:99b0cadd1388: apng: Fix wrong default final frame delay in muxer
[23:45:33 CEST] <michaelni> philipl, fate-suite http should be fixed
[23:53:28 CEST] <philipl> michaelni: yay. Thanks.
[00:00:00 CEST] --- Thu Jul 16 2015