[Ffmpeg-devel-irc] ffmpeg-devel.log.20140904

burek burek021 at gmail.com
Fri Sep 5 02:05:02 CEST 2014


[00:00] <Diogo> Llogan: i already write in ffmpeg user  https://ffmpeg.org/pipermail/ffmpeg-user/2014-August/023066.html and this thread  https://ffmpeg.org/pipermail/ffmpeg-user/2014-August/023133.html
[00:01] <Diogo> llogan: But i already tried all solutions and notthing...
[00:32] <wm4> ubitux: what do you think about reimar's idea of detecting utf16?
[00:58] <J_Darnley> Huh?  If x264asm has no movh for YMM regs, what are you supposed to emulate with?
[01:02] <J_Darnley> oh, movdqa does 32 bytes
[01:02] <J_Darnley> (so much for a double quad word)
[01:23] <kierank> it does?
[01:47] <jamrial> yaeh, movdqa with ymm does 32 bytes, with xmm 16
[01:47] <jamrial> if you want "movh" with INIT_YMM you just use mova xm*
[01:57] <J_Darnley> Now I'm wondering why there is a difference between in semantics between x264 and ffmpeg on xmN and ymN
[01:58] <Timothy__Gu> J_a
[01:58] <Timothy__Gu> J_Darnley: what?
[01:59] <J_Darnley> Or perhaps my version if just old, hang on.
[02:01] <Timothy__Gu> michaelni: I don't have my regular laptop right now, so I won't be able to read your GPG mail until this weekend.
[02:03] <michaelni> Timothy__Gu, the mail just says that i applied your patch and the new texi thing works
[02:05] <J_Darnley> well my x264 was old but the difference remains
[02:05] <J_Darnley> I means these differences: http://pastebin.com/WNVaFnTK
[02:19] <Timothy__Gu> michaelni: \o/
[02:26] <llogan> Timothy__Gu: does that address #3232?
[02:45] <Timothy__Gu> Llogan: no
[02:45] <Timothy__Gu> The patch on ML does
[02:58] <cone-704> ffmpeg.git 03Timothy Gu 07master:17ad5fbb8af3: Use makeinfo to generate html doc for the new website
[03:27] <llogan> Timothy_Gu: yes, that's what i was referring to...apparently i can't read because it mentions 3232 in the patch
[04:01] <cone-704> ffmpeg.git 03Michael Niedermayer 07master:bb8b752553f8: postproc: use FFMIN/FFMAX where appropriate
[04:25] <Zeranoe> ubitux: The modplug issue was on their end, the old configure was not doing proper checks and when it did the lib detection failed.
[05:04] <jamrial> is someone going to fix the cross-site font loading problem in fate.ffmpeg.org?
[07:18] <ubitux> Zeranoe: ok great, thank you
[07:20] <ubitux> wm4: yes that could be better, assuming you are willing to do it :)
[10:13] <durandal_1707> is sup demuxer & muxer going to be applied soon?
[10:19] <nevcairiel> there is still a few timestamp questions open for both
[10:56] <tmm1> i'm wondering if its possible to change encoding parameters like resolution/bitrate/crf "on the fly" during transcoding
[11:02] <BtbN> For h264/x264 and at least mpeg-ts it definitly is.
[11:03] <BtbN> But i don't think there's support in ffmpeg for it
[11:42] <nevcairiel> if you use avcodec programmatically you can change some properties
[11:42] <nevcairiel> with libx264 anyway
[11:43] <nevcairiel> supported are: AR, VBV, bitrate, crf, qp, interlaced
[11:43] <nevcairiel> resolution probably needs to flush the encoder, and re-create it
[11:44] <nevcairiel> all those above you can just change in AVCodecContext/avoptions and they will apply
[11:46] <tmm1> cool that's encouraging
[11:48] <tmm1> i'm thinking of adding a control domain socket option to ffmpeg so there's a way to send commands to a running instance
[11:49] <rcombs> tmm1: for the record, I would probably come up with a way to use that
[11:49] <tmm1> i find -progress lacking a lot of times so a way to read out more state would be nice too
[13:51] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:4f63d6d3b5c0: avdevice/x11grab: fix error handling in pixel formats
[14:04] <cone-611> ffmpeg.git 03Dmitry Volyntsev 07master:fe8f4c71e305: libavdevice/v4l2: fix descriptors leak on error paths
[14:19] <ubitux> the rdft/fft api really is convoluted
[14:20] <ubitux> it looks like it was designed to avoid too large code generation
[14:20] <ubitux> like having check on the mode in the callback itself (inverse, sign, ...) instead of having multiple callbacks
[15:12] <wm4> shit, russian .smi file spotted in the wild
[15:12] <nevcairiel> let me guess, ansi encoded in some weird code page? :D
[15:13] <wm4> utf-8 fortunately
[15:17] <cone-611> ffmpeg.git 03Steven Liu 07master:03efd7308289: avformat/flvdec: read the correct bits into the tag type
[15:37] <ubitux> michaelni: are you really going to add this wget?
[15:37] <ubitux> michaelni: please add a comment that it's for debian package
[15:38] <ubitux> also, i thought Andreas said it wasn't going to solve the issue?
[15:46] <cone-611> ffmpeg.git 03Andreas Cadhalpun 07master:c4abee734dcb: Remove non-free tests/lena.pnm and adapt FATE tests to depend on lena.pnm in the SAMPLES directory
[15:48] <michaelni> ubitux, no, i just meant the lena removial and use of it through fatesamples, its the minimum we need to resolve this, improvments on top of it are welcome if someone has ideas
[16:09] <ubitux> ok
[16:11] <J_Darnley> I wonder how much it would cost to buy the rights to the image.
[16:11] <ubitux> what's the point of having libpostproc externally?
[16:11] <nevcairiel> that wouldnt make debian any happier J_Darnley :P
[16:12] <J_Darnley> Do they dislike the content too?
[16:13] <ubitux> if we pay for it, it means we can distribute it
[16:13] <ubitux> not debian
[16:14] <J_Darnley> I mean buy the the entire copyright to the image meaning the person could place it under CC0 (if they wished)
[16:14] <nevcairiel> the image is a scan from a playboy from the 70s, good luck with t hat endavour
[16:14] <ubitux> too soon©
[16:15] <J_Darnley> It might be easier to lobby for a reduction in copyright length.
[16:16] <ubitux> maybe we should just come up with OC
[16:16] <ubitux> who's gonna pose for a sexy photo?
[16:17] <ubitux> http://lucy.pkh.me/lena.pnm and http://lucy.pkh.me/lena-debian.pnm weren't that a bad idea after all
[16:18] Action: ubitux hides
[16:21] <kierank> Get stallman to call Hugh hefner
[16:22] <ubitux> lena is no more :(
[17:25] <ubitux> wm4: i don't think pp belongs outside
[17:25] <ubitux> eventually it should be merged into libavfilter, but that's all
[17:25] <wm4> your thoughts are morally and technically wrong
[17:26] <wm4> I'm sure Daemon404 can confirm
[17:26] <ubitux> there is no reason to have it outside
[17:26] <ubitux> except haters who think that not using it means it should be dropped
[17:27] <ubitux> we have a pp filter using that library; while it might not belong at the root of the project, it could be merged into the pp filter
[17:27] <ubitux> postprocessing is kind of a very video specific thing
[17:28] <ubitux> now i can understand you don't like the implementation itself, but that's not a reason to drop it
[17:41] <ubitux> wm4: so, utf-16 is ready or you want to do some more stuff on it?
[17:43] <wm4> I'll send a new patch in some hours
[19:17] <cone-611> ffmpeg.git 03Diego Biurrun 07master:2b8b0da09cf0: fdctdsp: cosmetics: Drop one unnecessary if-block level
[19:17] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:a52e74db17e5: Merge commit '2b8b0da09cf0dc2535de42f95110eb633f36d8c8'
[19:45] <cone-611> ffmpeg.git 03Diego Biurrun 07master:803f8992f11d: build: cosmetics: Group hwaccel OBJS declarations together
[19:45] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:91a8b933fb6d: Merge commit '803f8992f11d1eb301672cd13c685a7b1d04c467'
[19:57] <cone-611> ffmpeg.git 03Diego Biurrun 07master:f20518568a77: build: Split WMA frequencies into a separate object file
[19:57] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:5a20f1885d3b: Merge commit 'f20518568a77a6138fc74021ce56013ab72907ba'
[20:15] <cone-611> ffmpeg.git 03Diego Biurrun 07master:4d55e9de2789: vc1: Split bits used in libavformat into a separate header
[20:15] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:ddc6adaa280a: Merge commit '4d55e9de27894ddfb337b28cd7aa6d14a12666d2'
[20:46] <cone-611> ffmpeg.git 03Anton Khirnov 07master:12f0388f9cb3: Add release notes for 11.
[20:46] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:69a68d301345: Merge commit '12f0388f9cb32016ac0dacaeca631b088b29bb96'
[20:55] <cone-611> ffmpeg.git 03Diego Biurrun 07master:d9792b773516: Mark 11 release in the changelog
[20:55] <cone-611> ffmpeg.git 03Michael Niedermayer 07master:53b089200557: Merge commit 'd9792b773516a560ecb99694b8ee745a50027fac'
[20:56] <michaelni> Anyone wants to write / check / complete the release_notes ? Timothy_Gu ? ubitux ?
[20:56] <michaelni> for 2.4
[21:56] <wm4> with git send-email, how do I adjust the subject line? I want to add a "v2" without having to do that for each patch
[21:58] <kepstin-laptop> wm4: git send-email takes most of the options from git format-patch; see e.g. the '-v' option there.
[21:59] <wm4> that's exactly what I wanted, thanks
[22:13] <michaelni> ubitux, is "SSE2 version of vf_idet'sfilter_line" ok to apply or you want to do another round of review ?
[22:14] <ubitux> michaelni: not motivated for release notes; i feel bad for wm4 because of utf-16 though :p
[22:14] <ubitux> michaelni: fine with me with idet; no more comment
[22:15] <wm4> ubitux: what, did I miss the release
[22:16] <ubitux> libav released 11 or something, or is soon going to do it, and michael wants to sync with it
[22:24] <cone-611> ffmpeg.git 03skal 07master:406a9ccffea8: avfilter/vf_idet: MMX/MMXEXT/SSE2 implementation of idet's filter_line()
[22:26] <j-b> some on
[22:26] <j-b> "skal"
[22:27] <J_Darnley> I don't think it is the first nick we've used
[22:28] <j-b> It's just ridiculous.
[22:28] <Rodeo> ???
[22:28] <llogan> i would have preferred "FireWind2000"
[22:29] <Rodeo> oh OK
[22:29] <Rodeo> there's an email in the commit message fortunately
[22:33] <llogan> it's nit level annoying though that his real name was used in previous commits
[22:33] <llogan> but not here
[22:36] <J_Darnley> Oh... That patch has a problem.  It is using paddq in mmx
[22:37] <ubitux> J_Darnley: sorry i probably didn't spot every problem in the patch
[22:37] <nevcairiel> paddq is not mmx?
[22:37] <nevcairiel> it seems to be
[22:40] <J_Darnley> ubitux: not your fault.  I missed it too
[22:41] <J_Darnley> I should lookup how 64-bit math is done ox x86 and fix it
[22:41] <J_Darnley> *on
[22:41] <jamrial> it doesn't even need to use paddq. paddd is enough
[22:41] <wm4> ubitux: patches on ML
[22:41] <ubitux> wm4: great
[22:42] <ubitux> still no fate tests?
[22:42] <J_Darnley> jamrial: then it is probably just a typo
[22:43] <wm4> ubitux: no
[22:45] <nevcairiel> whats wrong with paddq anyway, except that its not really "packed" in mmx, and just operates on the entire register?
[22:45] <wm4> ubitux: the "speed up probing" patch should probably have you as author
[22:45] <ubitux> wm4: lol no it's fine :)
[22:47] <jamrial> so fate-vsynth needs the fate samples to be rsynced now?
[22:47] <ubitux> oh patch 6 is cool
[22:49] <J_Darnley> nevcairiel: it is sse2
[22:49] <J_Darnley> it will cause an illegal instruction if you don't have sse2
[22:50] <ubitux> wm4: thanks for the patchset :) i'll probably apply 1-5 by tomorrow after some tests, and leave some more time for the last one
[22:50] <wm4> would be nice if this could be in 2.4
[22:50] Action: wm4 stares at michaelni 
[22:50] <ubitux> michaelni: can you wait ~24 hours?
[22:50] <ubitux> wm4: "On invalid UTF-18"; UTF-16, right?
[22:51] <nevcairiel> J_Darnley: an instruction on mmx registers, but it was added later?
[22:51] <ubitux> (patch 1)
[22:52] <wm4> ubitux: yes
[22:52] <nevcairiel> oh well, if its not even needed at all
[23:06] <ubitux> wm4: mmh, INVALIDDATA for a \0? isn't it what will be returned at EOF as well?
[23:09] <wm4> ubitux: hm, guess you're right
[23:09] <wm4> for srt it's still fine, but it makes the function not generally useful
[23:13] <ubitux> wm4: one way could be to have an optional unexpected_chr counter pointer that will be incremented in such case
[23:13] <ubitux> that will still read everything, but you can stop after one run
[23:13] <ubitux> gtg, hf, 'night
[23:14] <wm4> ubitux: I could also just move this code to srtdec
[23:31] <jamrial> the vf_idet fate test doesn't seem to use use the asm code
[23:47] <J_Darnley> WTF git?!  git: 'send-email' is not a git command. See 'git --help'.
[23:49] <J_Darnley> Did someone forget to build it when they rolled out 2.0?
[23:55] <kepstin-laptop> it's an optional install on some linux distributions, split into a separate package
[23:59] <J_Darnley> I just reverted to 1.7.9
[23:59] <J_Darnley> I'll look at what's going on later
[00:00] --- Fri Sep  5 2014


More information about the Ffmpeg-devel-irc mailing list