[Ffmpeg-devel-irc] ffmpeg-devel.log.20130201

burek burek021 at gmail.com
Sat Feb 2 02:05:03 CET 2013


[00:14] <Compn> pross has a bayer swscale eh
[00:15] <Compn> now i forgot what codec it was that needed that
[00:17] <pigoz> is swig supposed to run ok on libavformat headers? seems like it doesn't like attribute_deprecated at all. Maybe I'm doing something terribly wrong. swig -cffi path/to/avformat.h
[00:18] <pigoz> oh well I'll figure out tomorrow :)
[00:56] <cone-469> ffmpeg.git 03Michael Niedermayer 07master:131fac1c12ed: vc1dec: fix block_off
[00:56] <cone-469> ffmpeg.git 03Sebastian Sandberg 07master:c6779c513117: vc1dec: fieldtx is only valid for interlaced frames
[00:59] <michaelni> nevcairiel, do you want to review the remaining patches from "Sebastian Sandb ( *K) [FFmpeg-devel] [PATCH */3] vc1dec: *" ?
[01:10] Action: Compn pings bayer rgb patch
[01:11] <Compn> michaelni : should i ask kostya about those patches ?
[01:14] <Compn> vc1 patches that is
[01:15] <michaelni> Compn, nevcairiel worked on interlaced vc1
[01:16] <Compn> ah
[01:16] <michaelni> i dunno if kostya did, but he is of course welcome to review/approve/appy the patches
[01:16] <michaelni> apply
[01:16] <Compn> kostya did not iirc
[02:09] <cone-469> ffmpeg.git 03Allan Kristensen 07master:4789c25d83c4: matroska: fixed missing S_DVBSUB subtitles
[03:20] <cone-469> ffmpeg.git 03Michael Bradshaw 07master:c58c67398d03: libopenjpegenc: add support for pix fmt gbrp (8-16 bit)
[03:20] <cone-469> ffmpeg.git 03Michael Bradshaw 07master:28f36dce3e5c: libopenjpegenc: simplify switch statement
[03:20] <cone-469> ffmpeg.git 03Michael Niedermayer 07master:ffd02d0c1974: Merge remote-tracking branch 'mjbshaw/master'
[03:39] <highgod> Hi,I use git to get the latest version, but I can't comile correctly.There seems some errors in file libm.h,./libavutil/libm.h:52:32: error: static declaration of 'cbrt' follows non-static declaration
[03:53] <michaelni> highgod, you did try make distclean ?
[03:58] <Compn> highgod : did you rm the old checkout and pull a new one ?
[04:02] <cone-469> ffmpeg.git 03Michael Niedermayer 07master:086566a557ee: dict: fix memleak
[04:50] <highgod> Hi,I create a new folder and get the code
[05:00] <Compn> does it work highgod ?
[05:05] <highgod> I try to reget again,Thanks Compn.
[05:05] <highgod> I want to add our openCL filter code in the latest code.we will add a OpenCL wrapper.Is it OK?
[05:15] <Compn> highgod : you should ask michaelni
[05:15] <Compn> someone on the list said that opencl was only good for newest intel cpu
[05:16] <Compn> i dont know enough about opencl to make an opinion 
[05:30] <highgod> OK, I think add a option to make user can choose whether to use CPU or GPU
[05:31] <highgod> I reget the code, but the same error
[05:38] <Compn> hmmm
[05:39] <Compn> highgod : what ./configure options are you using ?
[05:42] <highgod> yes,just ./configure
[05:42] <highgod> and my git set autocrlf = false
[05:44] <Compn> are you using ccache ?
[05:44] <Compn> what are you compiling on ?
[05:44] <Compn> what arch
[05:45] <Compn> highgod : we ask, because our automated ffmpeg system says everythings ok http://fate.ffmpeg.org/
[05:46] <highgod> no,I just compile on my computer using mingw
[05:47] <highgod> I'm confused that why alldevice.o error. 
[05:48] <Compn> paste make output to pastebin or similar site
[05:49] <wm4> Compn: tired of mplayer support?
[05:50] <Compn> wm4 : why, people asking for me to return /
[05:51] <highgod> sorry,Compn,what is pastebin?
[06:00] <Compn> highgod : pastebin.com or pastebin.ca
[06:01] <Compn> highgod : just a website you can paste text at for public share
[06:01] <Compn> it may be banned in certain countries....
[06:01] <highgod> Thanks
[06:01] <Compn> or just put the make output somewhere we can see it
[06:02] <Compn> but not in the channel , too much text
[06:03] <highgod> http://pastebin.ca/2309543
[06:05] <Compn> highgod : did you try rm libavutil/libm.h && git pull
[06:06] <highgod> Compn : No,but I create a new folder and get the code.Should I need to do this operation?
[06:07] <Compn> cant hurt to try :)
[06:07] <Compn> but i dont know whats wrong
[06:07] <highgod> OK,Thanks
[06:07] <highgod> I will try
[06:08] <Compn> your copy of libm.h looks broken , does it have the windows line endings again ? 
[06:08] <Compn> dos2unix app might fix it 
[06:15] <Daemon404> disable autocrlf in git.
[06:20] <highgod> Daemon404 : I have set the autocrcf = false
[06:20] <Daemon404> you need to re-clone after ofc
[14:24] <cone-686> ffmpeg.git 03Diego Biurrun 07master:cb30bf7f03ac: doc/developer: Add web links for all suggested licenses
[14:24] <cone-686> ffmpeg.git 03Diego Biurrun 07master:8b0fda180e99: doc/developer: Add ISC license to list of acceptable licenses
[14:24] <cone-686> ffmpeg.git 03Diego Biurrun 07master:8787847dc6b3: doc/developer: Drop obsolete MPlayer reference
[14:24] <cone-686> ffmpeg.git 03Anton Khirnov 07master:f2960097e42d: bink: fix a check for the first frame.
[14:24] <cone-686> ffmpeg.git 03Michael Niedermayer 07master:69a6eddd607a: Merge commit 'f2960097e42ddf9a356bab6547f87906f6999e0a'
[14:29] <cone-686> ffmpeg.git 03Anton Khirnov 07master:1730ca2eca42: bink demuxer: check malloc return value
[14:29] <cone-686> ffmpeg.git 03Michael Niedermayer 07master:1bc51a7a74fc: Merge commit '1730ca2eca42974af0c29f38a0770997cba2f0da'
[14:39] <cone-686> ffmpeg.git 03Anton Khirnov 07master:9ec8971060a0: bink demuxer: set framerate.
[14:39] <cone-686> ffmpeg.git 03Diego Biurrun 07master:4eef2ed70744: ppc: fmtconvert: Drop two unused variables.
[14:39] <cone-686> ffmpeg.git 03Diego Biurrun 07master:6c1a7d07eb41: Use proper "" quotes for local header #includes
[14:39] <cone-686> ffmpeg.git 03Michael Niedermayer 07master:911e27068840: Merge remote-tracking branch 'qatar/master'
[14:43] <divVerent> how are patches to be submitted BTW? have a question not answered on the ffmpeg.org/developer.html page
[14:43] <divVerent> just git-send-email? but how would I then put in extra explanation (screenshot links, etc.)
[14:44] <divVerent> or are attachments allowed on the list
[14:46] <iive> attachments are ok, afaik
[14:52] <Compn> divVerent : its best to put attachments and screenshots on the bug trac
[14:52] <Compn> i think
[14:53] <Compn> otherwise you'll be sending a screenshot to 1000+ subscribers (however big ffmpeg-devel is)
[14:55] <Compn> oh
[14:55] <Compn> i misunderstood
[14:55] <Compn> :D
[14:56] <Compn> divVerent : isnt there a intro message to get-send-email where you can add links and description and stuff ?
[14:56] <Compn> i've seen people send them, a 0/* mail
[14:56] <JEEB> yeah, cover thingy
[14:56] <JEEB> cover page or whatever it was
[14:56] <Compn> yeah that
[15:00] <divVerent> Compn: I'd have sent the screenshot as link
[15:00] <divVerent> 14:55:54         @Compn | divVerent : isnt there a intro message to get-send-email where you can add links and description and stuff ?
[15:00] <divVerent> not that I know of
[15:00] <divVerent> but well, can't use git-send-email directly anyway
[15:01] <divVerent> or rather, not from the email address I would subscribe
[17:48] <Marco-123> hi all. I have a board with an ARM processor (and a NEON hardware accelerator), which is running Ubuntu Linux v10.04. I am building ffmpeg and my goal is to make sure it is taking full advantage of the hardware on the board. It's a freescale IMX53 btw. After much googling, I set a few flags in the configure script, and the script tells me "NEON enabled: yes" but the encoding performance is quite slow (same as when it says "
[17:48] <Marco-123> can someone provide advice on how to build this project properly?
[17:48] <Marco-123> or if there is a conclusive way to tell if the hardware accelerators are being used? another point is that the configure also tells me "enabled hwaccels: " and nothing is listed there.
[17:51] <durandal_1707> the options in table for lut(component options), afade(curve types) are displayed wrong, who is doxygen expert?
[17:53] <divVerent> durandal_1707: speaking of the power-of-two division issue BTW which patcheck shows on my patch... just saying because I saw some other post about it
[17:53] <divVerent> I intentionally ignored it, as I want to highlight that 32768 is the max sample value of S16
[17:53] <divVerent> so >> 15 is not really an option :P
[17:54] <divVerent> oh, and I evaded the patcheck message this way :P
[17:54] <divVerent>     const double w = 1. / (sqrt(nb_freq) * 32768.);
[17:54] <durandal_1707> what patcheck reported?
[17:54] <divVerent> and it now says "sqrt(nb_freq) * 32768. is the max value of a fft bucket"
[17:54] <divVerent>     const double w = 1. / sqrt(nb_freq) / 32768.;
[17:54] <divVerent> I had this in a previous version
[17:54] <divVerent> then patcheck wanted me to replace the / 32768 by >> 15
[17:55] <divVerent> which doesn't even work for double :P
[17:55] <durandal_1707> ahh, patcheck have own bugs...
[17:55] <divVerent> just mentioning it, as some other mailing list post mentioned the same patcheck message and that it's useless now :P
[17:55] <divVerent> especially given that we can expect compilers to handle this too
[17:56] <divVerent> or rather, "mostly handle it"
[17:56] <divVerent> i.e. for unsigned types
[17:56] <durandal_1707> what other mailing list?
[17:56] <divVerent> ffmpeg-devel
[17:56] <cone-686> ffmpeg.git 03Carl Eugen Hoyos 07master:99eedfc40086: doc/muxers.texi: Fix mp3 picture attachment documentation.
[17:56] <divVerent> there was some other unrelated post
[17:56] <divVerent> right after I subscribed
[17:56] <divVerent> http://ffmpeg.org/pipermail/ffmpeg-devel/2013-February/138323.html
[17:56] <divVerent> this one, last two lines
[17:57] <divVerent> oh, oops... haha
[17:57] <divVerent> I thought this referred to patcheck, but it does not :P
[17:57] <cone-686> ffmpeg.git 03Carl Eugen Hoyos 07release/1.0:5c1a1b47f378: doc/muxers.texi: Fix mp3 picture attachment documentation. (cherry picked from commit 99eedfc40086972987aa27df8b1259c8bf15b20c)
[17:57] <cone-686> ffmpeg.git 03Carl Eugen Hoyos 07release/1.1:9bcb84810f53: doc/muxers.texi: Fix mp3 picture attachment documentation. (cherry picked from commit 99eedfc40086972987aa27df8b1259c8bf15b20c)
[18:00] <divVerent> durandal_1707: but anyway, I did notice more false positives with patcheck :P
[18:00] <divVerent> it really seems just a crude tool
[18:01] <divVerent> possibly never written:intensity_colors
[18:01] <divVerent> possibly constant     :intensity_colors
[18:01] <divVerent> e.g. it says that for the options
[18:02] <durandal_1707> patcheck is mostly used for common typos you could not spot and style issues
[18:02] <divVerent> yes
[18:02] <divVerent> it did find accidental tabs in my code that I then fixed
[18:05] <michaelni> Marco-123, you can add some av_log/printf into a neon function to be really sure if its used
[18:08] <Compn> grumble
[18:08] <Compn> where is mov_text in the documentation ?
[18:08] <Compn> hey cehoyos :)
[18:09] <Marco-123> michaelni: at the moment i am using a library that works with ffmpeg so I don't issue ffmpeg commands directly. However, for this purpose I guess I should! Would you elaborate more on what I can look for when issuing that command to tell if neon is being used? Im guessing that because its a neon function it wont even work if neon isnt being used.
[18:09] <cehoyos> Hi
[18:11] <cehoyos> Compn: [15:12] <Compn> durandal_1707 : yes, but is it fixed or is it a mis-duplicate tag ? :P  <-- I don't understand this sentence...
[18:15] <Marco-123> michaelni: thanks for your help, too. 
[18:18] <Compn> cehoyos : i just wanted someone to verify the bug was fixed
[18:19] <Compn> since it was marked duplicate, that sentence is asking if it was a valid duplicate bug
[18:19] <Compn> or a new bug
[18:20] <Compn> interesting
[18:21] <Compn> i couldnt get quicktime to open our ttext sample
[18:22] <cone-686> ffmpeg.git 03Stefano Sabatini 07master:b868275ddbec: doc/muxers: fix level for the segment options subsection
[18:22] <Compn> saste : where in the docs is the "3gpp timed text" subtitle option ?
[18:23] <saste> Compn, why do you ask me?
[18:23] <Compn> because you are active ? :)
[18:23] <Compn> and i see you just commit some docs :)
[18:23] <Compn> i'd like to fix it, if i can figure out if its missing. or it maybe that i am blind
[18:23] <saste> Compn, i don't even know what "3gpp timed text" is
[18:24] <Compn> mov text subs format
[18:24] <saste> ubitxu?
[18:24] <Compn> hes in japan
[18:24] <Compn> soaking up all that radiation
[18:24] <saste> i don't think it is documented at all
[18:25] <saste> yes japan is cheap since then
[18:25] <durandal_1707> saste: the lut rgb options and afade curve type options in doxy are displayed wrong what is correct syntax?
[18:25] <saste> don't miss the opportunity
[18:25] <Compn> so theres no list of subtitle formats and their options ?
[18:25] <Compn> it would be cool to visit japan again
[18:25] <saste> Compn, ffmpeg -h full may help
[18:26] <Compn> saste : it doesnt list either
[18:29] <saste> durandal_1707, item foo NL description
[18:29] <saste> i'm going to fix lut
[18:31] <Compn> http://ffmpeg.org/pipermail/ffmpeg-cvslog/2012-August/053256.html
[18:31] <Compn> theres the commit
[18:32] <cehoyos> Compn: Do you have any reason to assume it was not a duplicate?
[18:34] <cehoyos> Compn: Muxing mov_text in mov is not QT-compatible, see ticket #1845
[18:34] <Compn> cehoyos : no reason. just curiosity
[18:36] <Compn> ok thanks for bug 1845
[18:36] <Compn> we have mov_text samples
[18:36] <Compn> that work in quicktime
[18:36] <Compn> cehoyos : http://samples.ffmpeg.org/mov/subtitles-embedded/
[18:36] <Compn> at least http://samples.ffmpeg.org/mov/subtitles-embedded/BCDisassembly1RightSideTxt2.mov works in quicktime, did not test the rest
[18:38] <cehoyos> Compn: Which QT version did you test?
[18:38] <Compn> 7.7.3
[18:39] <cehoyos> I ask because this file is around 10years older than the QT version mentioned in ticket 1845
[18:40] <cehoyos> Could you try ffmpeg -i BCDisassembly1RightSideTxt2.mov -codec copy -map 0 out.mov and test if out.mov shows the subtitles in QT?
[18:40] <Compn> sure
[18:41] <Compn> yes that works
[18:41] <Compn> in quicktime
[18:41] <cehoyos> Also with -vcodec mpeg4 -qscale 2 -strict -2 -acodec aac ?
[18:42] <Compn> theres no -scodec listed 
[18:42] <Compn> do you want -scodec copy ?
[18:42] <cehoyos> Could you test both -scodec copy and scodec mov_text ?
[18:42] <Compn> sure
[18:43] <Compn>  -vcodec mpeg4 -qscale 2 -strict -2 -acodec aac -scodec copy works  in quicktime
[18:44] <Compn> -scodec mov_text shows weird things in qt
[18:44] <Compn> like the text is 100 pt font stuck in a 20pt box, and color dark red maybe ?
[18:44] <Compn> yes, if i fullscreen the video, i can see text
[18:45] <Compn> kind of hard to explain :)
[18:45] <cehoyos> You could try to enlarge the video radically: -s 1280x960
[18:47] <Compn> cehoyos : http://i.imgur.com/n7AeD4l.jpg
[18:47] <Compn> thats with -scodec mov_text
[18:48] <cehoyos> Thank you for the tests!
[18:50] <Compn> in my screenshot, you can see that the subs are at the top with ffmpeg's reencode
[18:50] <Compn> but in the original file, the subs are at the bottom...
[18:51] <Compn> probably the alignment tag does nt get copied
[18:51] <Compn> for some reason
[18:51] <Compn> cehoyos : thank you for getting bug reports in order :)
[18:52] <cehoyos> Regarding ticket #1840: Your suggestion for -c:s mov_text will not work in 50% of all cases...
[18:52] <cehoyos> Especially not for the sample from ticket #756
[18:54] <Compn> if (textsubs) then
[18:54] <Compn> i think dvdsub (vobsub?) in mp4/mov is rare
[18:54] <cone-686> ffmpeg.git 03Stefano Sabatini 07master:902fa77fd716: doc/filters: fix syntax and description for the lut* options
[18:54] <cone-686> ffmpeg.git 03Stefano Sabatini 07master:986540b65f6e: doc/filters: create a dedicated examples section for lut*, and itemize
[18:54] <cone-686> ffmpeg.git 03Stefano Sabatini 07master:423856c4ae85: doc/filters: fix syntax and typo for the afade curve option
[18:56] <Compn> oh that cpcdemo.mov is    Stream #0:2(eng): Subtitle: eia_608 (c608 / 0x38303663)
[18:57] <Compn> thats different than mov_text :\
[18:57] <cehoyos> (Apart from the fact that dvdsub, dvbsub, pgssub and maybe even xsub are not rare:) How is the rareness related to giving bad advice?
[18:58] <Compn> cehoyos : you are right, how about "please use -scodec (supported mov subtitle list)" then ?
[18:58] <cehoyos> Looks like ticket #846
[18:58] <cehoyos> I think the current error message is the best possible
[19:01] <Compn> ok, then feel free to close 1840 bug as duplicate of 1845 
[19:02] <cehoyos> But 1840 was never a duplicate of 1845 (you insisted yesterday in making it one)
[19:02] <Compn> comment #3 pretty much makes it his problem 
[19:02] <Compn> makes it a duplicate, i mean
[19:03] <Compn> ok mark it invalid then :)
[19:03] <Compn> no worries
[19:03] <Compn> cehoyos : new problem, where is mov_text listed ?
[19:03] <Compn> i think its missing in documentation
[19:03] <Compn> i cant find supported subtitle list either
[19:03] <cehoyos> No, comment three changed the original report to something (more important but) unrelated
[19:04] <cehoyos> I sometimes commit documentation fixes (saste: I hope you are ok with my changes an hour ago), but I don
[19:04] <cehoyos> ... I don't feel responsible for documenting features I don't know much about.
[19:05] <Compn> i dont mind documenting it myself
[19:05] <Compn> i mean
[19:05] <Compn> i want to fix it
[19:05] <Compn> i just dont know where it should go :)
[19:05] <cehoyos> Then send a patch
[19:06] <Compn> curious why this mov_text does not work, no video stream ? http://samples.ffmpeg.org/mov/subtitles-embedded/nova_eu_3012c01_caps.mov-captions-subtitles-in-quicktime.mov
[19:07] <Compn> it works in quicktime i mean
[19:07] <Compn> but not ffplay 
[19:12] <cehoyos> compn: I believe it would work with ffplay but ffplay does not support mov_text (or does it and I just don't know how?)
[19:13] <Compn> cehoyos : oh, i forgot ffplay doesnt support mov_text
[19:13] <Compn> then its my bad
[19:18] <Compn> cehoyos : are there any bugs i should look at ?
[19:18] <cehoyos> Compn: mov_text is documented afaict: http://ffmpeg.org/general.html#Subtitle-Formats
[19:19] <Compn> i'm not good with programming ..
[19:20] <Compn> cehoyos : yes, i found it , mov_text is listed with ffmpeg -codecs
[19:21] <Compn> i was trying ffmpeg -formats ... i'm dumb :(
[19:21] <Compn> is a subtitle a format or codec, who knows :D
[19:25] <durandal_1707> no more comments on showspectrum?
[19:26] <saste> durandal_1707, wait a day or so, so ubitux can comment
[19:26] <durandal_1707> ubitux: seriously?
[19:28] <saste> durandal_1707, he's the author of the filter
[19:28] <saste> i can't have a look at it now, i'll try to do it tonight
[19:32] <durandal_1707> saste: not listed in MAINTAINERS for that file ........................
[19:34] <nevcairiel> he is listed as author at the top of the file =p
[19:39] <durandal_1707> nevcairiel: so i would need to wait for every single persona persona at the top of the file?
[19:40] <nevcairiel> its only one person, and if that person is an active commiter to ffmpeg, yes, you should
[19:41] <nevcairiel> its just polite to ask the author to review a patch to his code
[19:41] <durandal_1707> well i will just add another persona at the top of the file which will simplify things a lot
[19:41] <nevcairiel> that doesnt change who the author of the file is
[19:45] <durandal_1707> even if code was so badly writen that it needs rewrite or adds more code that was already present ....
[19:57] <saste> durandal_1707, i'm asking you to wait a day, not till the end of the world
[19:58] <durandal_1707> saste: but end of world is near!!!!
[19:58] <saste> durandal_1707, no it's already passed
[19:59] <saste> can't say when it will be the next one
[20:03] Action: durandal_1707 will ask ubitux for special authorisation code
[20:04] <Compn> lol
[20:04] <Compn> durandal_1707 : work on something else in the meantime :)
[20:04] <Compn> i remember ffmpeg used to have much longer patch review times
[20:04] <Compn> 37 patch resubmissions :D
[20:05] <Compn> wonder what the highest one was
[20:05] <saste> durandal_1707, the-matrix-effect filter
[20:05] <Compn> try 42 ?
[20:06] <saste> durandal_1707, to say the truth, i have an unfinished port of it, based on jwz code
[20:06] <durandal_1707> of what?
[20:07] <durandal_1707> matrix effect? it still not in?
[20:07] Action: durandal_1707 sad
[20:08] <burek> is "Subtitle: dvb_teletext" the same as "Subtitle: dvb_subtitle" ?
[20:09] <Compn> no
[20:09] <Compn> well
[20:09] <Compn> i dont know :D
[20:09] <Compn> i dont think so
[20:09] <nevcairiel> no its not the same
[20:09] <nevcairiel> dvb has two kind of subtitles
[20:10] <burek> it seems that ffmpeg converts dvb_teletext to dvb_subtitle when -c copy is used
[20:10] <nevcairiel> sounds like a bug
[20:10] <burek> yeah, i guess i should report it
[20:10] <burek> generally, should i wait and ask for people to report a bug
[20:11] <burek> or if i notice in on the forum, should i just report it myself
[20:11] <burek> i mean, do i somehow take away their credits or something?
[20:11] <burek> because i'd not want to do that
[20:12] <burek> notice it*
[20:12] <nevcairiel> credit for reporting a bug?
[20:12] <burek> well i dunno, i just ask.. :)
[20:12] <nevcairiel> you could always mention the person and link to the forum
[20:12] <nevcairiel> if you care
[20:12] <burek> yeah, that could be ok i guess
[20:13] <burek> ok, thanks :) back to reporting it then :)
[20:39] <Marco-123> hi all. (my sys: ubuntu 10.04, arm proc, neon media processor). i am using a software library that uses FFMPEG. so i dont work with ffmpeg directly. however, i suspect that my hardware accelerators are not being used. so now I am to get into things. id like to write a minimal app, that will tell me (yes/no, conclusively) whether or not the hardware accelerators are being used by the ffmpeg libraries. could someone help me w
[20:42] <Marco-123> i was thinking of having a sample videos, maybe encoding/decoding them with various codecs (h263plus or h264), audio and video, and trying to see if the hwaccel is being used along the way (??). or maybe there is a command that will tell me info directly. 
[20:42] <burek> does ffmpeg support hwaccel right now?
[20:44] <cehoyos> Marco-123: Are you talking about hwaccel or asm / neon support?
[20:52] <michaelni> nevcairiel, maybe you want to comment/review on the 2 alterantive [PATCH 1/3] vc1dec: the second field is written wrong to the picture / vc1dec: Fix destination pointers in context for fields
[20:53] <michaelni> its better to have a second oppinon on which is better than me just pushing one in a few days
[20:53] <nevcairiel> all i can do is check the bitstream parsing against the spec, because thats really all i know about how vc1 works :d
[20:55] <michaelni> :(
[20:56] <michaelni> sadly no bitstream parsing changes in the ones i pointed to, but thanks anyway
[20:57] <nevcairiel> thats also the reason my patch was never finished, i wrote the bitstream parsing, and never had time or motivation to figure out how it fits into everything else
[20:59] <Marco-123> cehoyos: hmm. maybe my understanding is off. i have a neon chip (media processor?), and i thought that this was the hardware accelerator. maybe my terminology and understanding of the thing is wrong.
[20:59] <michaelni> nevcairiel, [PATCH 3/3] vc1dec: interlaced B-frames <----- contains some bitstream parsing in the added code
[21:00] <nevcairiel> neon is not what we usually call a "hardware accelerator", neon is just "SIMD", what in our desktop CPUs is MMX/SSE
[21:09] <Marco-123> nevcairiel: ah. ill read about hardware accelerators now. in the mean time, can an SIMD help my encoding/decoding performance?
[21:09] <nevcairiel> yes, significantly
[21:09] <nevcairiel> without SIMD, de/encoding is much slower
[21:10] <Marco-123> if my board has a neon SIMD, might it also have a GPU? or is there an overlap of these two functionalities and only one should be found?
[21:10] <nevcairiel> neon is part of the CPU, it has nothing to do with your GPU, so sure, it can have a GPU as well
[21:10] <nevcairiel> although in a SoC stuff gets mixed :p
[21:13] <Marco-123> alright i will look into my board's datasheet again for GPU details (it probably doesnt have one). so i guess my original question is: what is a good approach for conclusively knowing whether or not my ffmpeg encoding/decoding is acutally making use of the neon simd?
[21:15] <wm4> are there fate tests for dvdsub?
[21:16] <wm4> Marco-123: my guess is that if there's ARM specific asm in the source code, it uses neon
[21:17] <nevcairiel> his question seems to be how to know if that asm is actually used =p
[21:17] <Skyler_> see if neon: yes in configure?
[21:17] <nevcairiel> isnt there a runtime check now too
[21:17] <wm4> Marco-123: compile with asm disabled, and compare speed to normal configuration
[21:18] <wm4> or rather, neon disabled, if there's a configure option
[21:18] <Marco-123> nevcairiel: after much googling, im able to build ffmpeg so that the configure script reads "NEON enabled: yes"
[21:18] <Marco-123> --enable-neon haha
[21:18] <Marco-123> still took much googling!
[21:18] <nevcairiel> i would've thought configure enables such things on its own by probing
[21:18] <wm4> well then try with --disable-neon and compare speed, then you should get some conclusive results
[21:19] <nevcairiel> but what do i know about arm
[21:19] <Marco-123> wm4: yes this is a good plan. forgive my ignorance but I can't find on google what ASM in this context stands for
[21:19] <Marco-123> assembler?
[21:20] <wm4> yes
[21:20] <nevcairiel> yes, assembly code, hand-optimized code that uses neon
[21:20] <Marco-123> thank you
[21:21] Last message repeated 1 time(s).
[21:21] <Marco-123> so in your opinions: if the configure script reads "NEON enabled: yes", should I stop worrying about ffmpeg using it?
[21:21] <Marco-123> is it a good indicator?
[21:22] <wm4> sure, why not
[21:24] <Marco-123> ok. to answer why not: im still seeing not so great encoding performance. granted my set up is embedded and thus not as fast as a desktop computer, so this might be the cause
[21:26] <Marco-123> guys thanks for your help. you've cleared up a lot of fuzziness in my brain about this stuff
[21:26] <Marco-123> :D
[21:41] <Skyler_> If you want to be absolutely sure, disable NEON and see how much slower it is ;)
[23:05] <cone-774> ffmpeg.git 03Matthieu Bouron 07master:5a67e30b1c71: ffmpeg: fix broken channel_layout option
[23:10] <cone-774> ffmpeg.git 03Matthieu Bouron 07release/1.1:c8dace272829: ffmpeg: fix broken channel_layout option
[23:29] <cone-774> ffmpeg.git 03Carl Eugen Hoyos 07master:6a50e8a190bc: Do not change codec in flv streams if the user has forced a codec.
[23:42] <cone-774> ffmpeg.git 03Carl Eugen Hoyos 07release/1.0:299dc3baeaaf: Do not change codec in flv streams if the user has forced a codec.
[23:42] <cone-774> ffmpeg.git 03Carl Eugen Hoyos 07release/1.1:3445bec6fc1d: Do not change codec in flv streams if the user has forced a codec.
[23:58] <cone-774> ffmpeg.git 03Ronald S. Bultje 07master:8af915c21b03: vp56: remove clear_blocks call, and clear alpha plane U/V DC only.
[00:00] --- Sat Feb  2 2013


More information about the Ffmpeg-devel-irc mailing list