[Ffmpeg-devel-irc] ffmpeg-devel.log.20171029

burek burek021 at gmail.com
Mon Oct 30 03:05:03 EET 2017


[00:05:40 CEST] <ubitux> what's the state of our vorbis encoder?
[00:05:48 CEST] <ubitux> is it good compared to libvorbis?
[00:06:08 CEST] <ubitux> seems my shitty rpi3 isn't able to hold real time encoding, i wonder if it would help to use ffmpeg encoding
[00:07:48 CEST] <atomnuker> I haven't yet merged the last gsoc patches (I really should)
[00:09:01 CEST] <ubitux> atomnuker: btw, dunno if you were told, but mainline odroid is working
[00:09:09 CEST] <ubitux> mainline on odroid c2*
[00:09:15 CEST] <atomnuker> yep, I'm using it on my 2 c2 boards
[00:09:24 CEST] <atomnuker> libvorbis doesn't seem to have any simd
[00:09:31 CEST] <ubitux> yeah
[00:09:36 CEST] <atomnuker> we have mdct simd for most platform so it should be faster
[00:09:42 CEST] <ubitux> i'm assuming that's why i'm getting lags
[00:09:55 CEST] <ubitux> but it's curious because i didn't have any issue with an old armv7 board
[00:10:00 CEST] <atomnuker> the codec is good enough that even the worst encoder sounds decent, even at 54 kbps
[00:10:19 CEST] <ubitux> ok
[00:11:01 CEST] <ubitux> btw, funny that libvorbis has more autotools build files than source code
[00:11:27 CEST] <nevcairiel> autotools is just evil like that
[00:11:54 CEST] <ubitux> seriously, there is like at least 20 files for the build in the root dir
[00:12:17 CEST] <atomnuker> do we have the float_dsp equivalent for non-float stuff?
[00:12:49 CEST] <ubitux> fixed_dsp?
[00:13:26 CEST] <ubitux> libavutil/fixed_dsp.h
[00:13:42 CEST] <ubitux> seems it was introduced for mips
[00:14:25 CEST] <atomnuker> nah, I need it for video
[00:16:41 CEST] <nevcairiel> fixed_dsp is for fixed point numbers like the aac fixed point decoder, not sure how those compare to actual integer video data
[00:40:16 CEST] <durandal_1707> admit it: ffmpeg is dead
[00:45:33 CEST] <jamrial> ubitux: it can't even do real time?
[00:46:24 CEST] <ubitux> i get a lot of stuttering
[00:46:38 CEST] <jamrial> even a mid-low range x86 cpu can do 20x real time libvorbis with one thread
[00:46:42 CEST] <jamrial> how shitty can that thing be o_o
[00:46:51 CEST] <ubitux> :)
[00:49:49 CEST] <durandal_1707> i cant introduce new flag for mb types, so i will add private option to mpeg2video decoder
[00:51:02 CEST] <durandal_1707> why nobody calls a vote for hwaccel patch thing?
[00:53:53 CEST] <jamrial> durandal_1707: wm4 didn't want to do it
[00:54:22 CEST] <jamrial> durandal_1707: you mean new option to mpegvideo for the debug mv stuff?
[00:55:44 CEST] <durandal_1707> yes
[00:55:48 CEST] <ubitux> durandal_1707: private option in mpeg2 should be pretty fine for now, designing a clean api for mb types is going to take a lot of time to make everyone happy
[00:58:27 CEST] <jamrial> if that's enough to not restore the disabled public symbols in avcodec.h and options_table.h, it should be fine for now
[01:41:29 CEST] <TD-Linux> ubitux, my rpi2 encodes with libvorbis at 10x realtime...
[01:42:32 CEST] <TD-Linux> er 4.8x realtime
[01:44:02 CEST] <TD-Linux> (also libopus does 6x-12x realtime depending on complexity setting)
[01:46:46 CEST] <cone-514> ffmpeg 03Carl Eugen Hoyos 07master:1da8c4ec85c1: lavfi/palettegen: Allow setting the background colour.
[01:49:35 CEST] <TD-Linux> ubitux, one thing to keep in mind if you're trying to do really low latency stuff is that rpi3 core migrations are slow. try locking to one core with taskset
[02:36:25 CEST] <cone-514> ffmpeg 03Carl Eugen Hoyos 07master:8b43039fe19c: lavfi/palettegen: Fix compilation after 1da8c4e.
[03:38:38 CET] <BBB> durandal_1707: the private data for the avfilter is fine
[03:38:49 CET] <BBB> durandal_1707: just document it as unspecified data and codec-specific, so we can chagne it over time
[04:11:38 CET] <cone-514> ffmpeg 03James Almer 07master:ca2b7794234c: avformat: remove use of deprecated AVFMT_FLAG_KEEP_SIDE_DATA flag
[05:33:45 CET] <cone-514> ffmpeg 03Steven Liu 07master:f5208307618d: avformat/hlsenc: fix base_output_dirname is null when basename_size is 0 bug
[05:33:46 CET] <cone-514> ffmpeg 03Steven Liu 07master:c3e279e75227: avformat/hlsenc: fix missing first segment bug in fmp4 mode
[05:33:47 CET] <cone-514> ffmpeg 03Steven Liu 07master:073986c5aeb8: avformat/hlsenc: reindent hlsenc code
[09:35:36 CET] <cone-162> ffmpeg 03Martin Vignali 07master:7da254886fbf: libavcodec/magicyuv : remove unneed variable assignment
[09:40:17 CET] <ubitux> TD-Linux: a bunch of things may be the cause of the trouble; i'm relying on MPD to do the encode, and i'm setting a relatively high profile iirc
[09:40:39 CET] <ubitux> i haven't benched libvorbis in particular, i just know that's what being used
[09:41:28 CET] <ubitux> in wasn't an issue previously on an old A10
[09:42:02 CET] <ubitux> it maybe also be related to the approximate aarch64 mainline support on rpi
[12:16:30 CET] <cone-162> ffmpeg 03Paul B Mahol 07master:ca106d6a5c4c: avcodec/magicyuvenc: improve compression
[13:06:32 CET] <JEEB> atomnuker: as I noted I was outside the reach of internets and I don't have my related ssh keys on my phone anyways
[13:06:45 CET] <JEEB> I'd otherwise had pushed them
[13:20:34 CET] Action: kierank wonders how to remove mpeg4video from mpegvideo.c without breaking ER
[13:24:53 CET] <BBB> is ER still dependent on mpegvideo in itself?
[13:25:05 CET] <BBB> if so, I think the answer is to refactor ER to become itself mpegvideo-independent
[13:25:07 CET] <BBB> which is *hard*
[13:36:21 CET] <kierank> not dependent actually
[13:36:25 CET] <kierank> it does seem possible
[14:17:17 CET] <cone-162> ffmpeg 03James Almer 07master:7df37dd319f2: avutil/frame: deprecate getters and setters for AVFrame fields
[14:32:17 CET] <cone-162> ffmpeg 03James Almer 07master:307c24b32f30: avformat: move public AVStream fields up in the struct
[14:32:18 CET] <cone-162> ffmpeg 03James Almer 07master:b7785d10b00c: avformat: deprecate getters and setters for AVFormatContext and AVStream fields
[14:36:38 CET] <cone-162> ffmpeg 03James Almer 07master:c8e8092543dd: avformat/avformat.h: fix pairs typo
[14:55:43 CET] <cone-162> ffmpeg 03Carl Eugen Hoyos 07master:9b0510a8e359: lsws/yuv2rgb: Fix yuva2rgb32 on big endian hardware.
[15:05:34 CET] <jamrial> JEEB: if nicolas doesn't push his patchset by the time you're back then feel free to do it. git head has been broken long enough already and should be fixed asap
[15:09:37 CET] <JEEB> jamrial: sure. getting to a computer in about 3h
[15:09:49 CET] <jamrial> JEEB: thanks
[15:22:06 CET] <cone-162> ffmpeg 03Martin Vignali 07master:e9930883a26c: libavcodec/bswapdsp : add AVX2 func for bswap_buf (swap uint32_t)
[15:30:18 CET] <cone-162> ffmpeg 03Carl Eugen Hoyos 07master:75bd01090f5b: lavf/rtpenc: Add support for 24 bit pcm encoding as defined by RFC 3190.
[18:31:18 CET] <cone-501> ffmpeg 03Anton Schubert 07master:3dca4887aef4: dashenc: fix bitrate estimation with correct scaling
[18:31:18 CET] <cone-501> ffmpeg 03Peter Große 07master:44ede215b662: dashenc: set DASH related options for the subsequent matroska muxer when using webm
[18:31:18 CET] <cone-501> ffmpeg 03Peter Große 07master:1443859a8d0d: dashenc: don't write header data before the first packet arrives
[18:31:18 CET] <cone-501> ffmpeg 03Peter Große 07master:2911dabd51dc: dashenc: allow AdaptationSets to have different average frame rates
[18:31:18 CET] <cone-501> ffmpeg 03Anton Schubert 07master:fe334712ddc6: dashenc: Don't output frameRates at both AS and Representation-level. Only output maxFrameRate at AS-level.
[18:31:18 CET] <cone-501> ffmpeg 03Peter Große 07master:3c838e6442bb: dashenc: copy stream frame rate to output stream
[18:31:18 CET] <cone-501> ffmpeg 03Michael Niedermayer 07master:fe533628b960: avcodec/jpeglsdec: Check ilv for being a supported value
[18:31:19 CET] <cone-501> ffmpeg 03Michael Niedermayer 07master:f80224ed19a4: avcodec/jpeglsdec: Check for end of bitstream in ls_decode_line()
[18:42:16 CET] <cone-501> ffmpeg 03Nicolas George 07master:bfb1a946255f: examples/avio_reading: return AVERROR_EOF at EOF.
[18:42:17 CET] <cone-501> ffmpeg 03Nicolas George 07master:a606f27f4c61: lavf/avio: temporarily accept 0 as EOF.
[18:42:18 CET] <cone-501> ffmpeg 03Nicolas George 07master:158a79c31d55: lavf/aviobuf: return EINVAL when reading from a write-only context.
[18:42:25 CET] <JEEB> jamrial: pushed it
[18:43:07 CET] <JEEB> had to modify the lavf/version.h bit a bit since in the middle there was another define added in the middle, but that didn't really seem to be a problem other than making the automagic application of hte patch fail
[18:47:05 CET] <jamrial> JEEB: cool :D
[18:47:39 CET] <jamrial> JEEB: reply to the thread saying you pushed it in that case
[18:47:49 CET] <JEEB> already done
[18:50:51 CET] <jamrial> huh, i didn't get anything
[18:52:10 CET] <jamrial> there. it spent like five minutes stuck in the ml queue or something
[18:52:26 CET] <JEEB> funky
[19:42:54 CET] <cone-501> ffmpeg 03James Almer 07master:ddd0b19169dc: ffprobe: remove usage of deprecated getter functions
[19:42:55 CET] <cone-501> ffmpeg 03James Almer 07master:5cc5130c813f: ffmpeg: remove usage of deprecated getter functions
[19:42:56 CET] <cone-501> ffmpeg 03James Almer 07master:2245476e5c45: avdevice: remove usage of deprecated setter and getter functions
[21:24:30 CET] <cone-501> ffmpeg 03Carl Eugen Hoyos 07master:59ceee5896af: fate: Fix some of the fitsdec tests on big-endian.
[22:14:19 CET] <jkqxz> "Currently 709 or 601 is decided based on frame size."  Good old BT.((frame.height>600)*108+601), always the best colourspace.
[22:18:42 CET] <JEEB> :D
[23:10:58 CET] <cone-501> ffmpeg 03Carl Eugen Hoyos 07master:4ccc1ef2a3b1: fate: Add bitexact flag to the invalid-elst test.
[00:00:00 CET] --- Mon Oct 30 2017


More information about the Ffmpeg-devel-irc mailing list