[Ffmpeg-devel-irc] ffmpeg-devel.log.20160524

burek burek021 at gmail.com
Wed May 25 02:05:03 CEST 2016


[00:06:41 CEST] <cone-609> ffmpeg 03ZhouXiaoyong 07master:05a546181ff1: avcodec/mips/blockdsp_mmi: Version 2 of the optimizations for loongson mmi
[00:06:42 CEST] <cone-609> ffmpeg 03ZhouXiaoyong 07master:377e5db3db4a: avcodec/mips/pixblockdsp_mmi: Version 2 of the optimizations for loongson mmi
[00:06:43 CEST] <cone-609> ffmpeg 03ZhouXiaoyong 07master:8392794c9227: avcodec/mips/idctdsp_mmi: Version 2 of the optimizations for loongson mmi
[02:01:47 CEST] <philipl_> exit
[04:40:38 CEST] <cone-513> ffmpeg 03Michael Niedermayer 07master:4c82ccad2b3c: avformat/mov: Avoid "Unintended sign extension"
[13:21:52 CEST] <ln-> oh, ffmpeg has a tech support and customer service phone in canada
[13:26:53 CEST] <iive> pages deleted, ip blocked.
[13:28:22 CEST] <thardin> they're tenacious
[13:28:42 CEST] <BtbN> It might be time to entirely block public editing of the wiki.
[13:30:34 CEST] <iive> i think we might try some tweaking of the anti-spam engine
[13:31:08 CEST] <thardin> putting something simple like "what's the name of this project?" as a text input usually goes a long way
[13:31:08 CEST] <BtbN> I don't think it's too much of a problem to require manual activation of new accounts
[13:31:22 CEST] <thardin> that's what the mxflib forums do
[13:32:23 CEST] <thardin> spammers are only going to put so much effort into a custom antispam thing, if any
[13:49:43 CEST] <michaelni> if someone has a patch for trac to make registration require specifying the project name that would be welcome
[14:53:35 CEST] <Compn> finally someone gave me vqc1 sample :)
[14:54:09 CEST] <durandal_170> Compn: what it is?
[14:56:08 CEST] <Compn> old codec
[14:56:13 CEST] <Compn> doesnt look to be mjpeg 
[14:56:33 CEST] <Compn> https://samples.ffmpeg.org/drivers32/new/VQC1/
[14:57:17 CEST] <farfel_> not sure if this is the right place - I have a question about libavcodec API
[14:58:04 CEST] <Compn> farfel_ : are you working on ffmpeg or libavcodec? or are you working on your own project ?
[14:58:15 CEST] <farfel_> yes, own project
[14:58:20 CEST] <Compn> then probably ask in #ffmpeg
[14:58:25 CEST] <farfel_> ok, thanks
[14:58:28 CEST] <Compn> no problem
[15:23:02 CEST] <Compn> durandal_170 : want sample ?
[15:25:02 CEST] <BBB> I really dont think the trac spam is that much of an issue
[15:25:16 CEST] <BBB> I mean, it doesnt hurt us in any way, we just manually delete some pages
[15:25:57 CEST] <BtbN> As long as they don't start editing legitimate pages, yeah
[15:28:47 CEST] <Compn> says the person not on spam/trac patrol...
[15:28:56 CEST] <Compn> ask llogan or carl about it
[15:46:06 CEST] <omerjerk> Hi everyone 
[15:46:25 CEST] <omerjerk> Carl had previously suggested to use softfloat api instead of this - https://github.com/omerjerk/FFmpeg/blob/float/libavcodec/alsdec.c#L289
[15:46:55 CEST] <omerjerk> but softfloat api doesn't seem to have the sign bit data - https://www.ffmpeg.org/doxygen/2.2/structSoftFloat.html
[15:50:02 CEST] <omerjerk> and if I do av_float2int(float value) >> 31, that'd defeat the whole purpose I guess.
[15:54:45 CEST] <jkqxz> What is wrong with sign bit = mant < 0?  (Or value < 0.0 while it was a float.)
[16:07:30 CEST] <omerjerk> umm okay. that's also correct. 
[16:08:01 CEST] <omerjerk> jkqxz: So I'll go ahead with your solution. 
[16:09:00 CEST] <omerjerk> I'll send my patch to the mailing list by tomorrow, to get it reviewed further. 
[16:15:43 CEST] <andrey_turkin> I wonder what's happening with my nvenc patches. Nobody looked at them yet? Or nobody's going to because of some reason?
[16:29:23 CEST] <BBB> I dont think theres any specific reason
[16:36:57 CEST] <jkqxz> andrey_turkin:  I think BtbN was working on related things (CUDA/NVENC), and might be able to make some comment on it here.
[16:49:17 CEST] <andrey_turkin> mmkay... guess I'll keep waiting for some reaction then
[17:00:00 CEST] <BtbN> Did I miss a patch on the ml?
[17:01:03 CEST] <andrey_turkin> http://ffmpeg.org/pipermail/ffmpeg-devel/2016-May/194502.html and thread replies
[17:01:28 CEST] <BtbN> Anyway, I'm about to replace the nvenc encoder in ffmpeg with a merged libav/ffmpeg one. So patches aren't exactly easy at the moment.
[17:02:11 CEST] <andrey_turkin> ah shoot. So i guess I did that for nothing
[17:02:37 CEST] <andrey_turkin> if you are going to merge in libav features then these patches are useless
[17:02:57 CEST] <BtbN> https://git.khirnov.net/libav.git/log/?h=nvenc that's the current status. Patching is done so far, "just" need to review and test for regressions from the ffmpeg implementation.
[17:03:30 CEST] <BtbN> And I have to figure out how to do the actual merge. I guess just plainly replacing it would be the easiest.
[17:04:24 CEST] <andrey_turkin> hmm, maybe you still can you my first 3 patches to bring ffmpeg version a bit closer to libav
[17:05:04 CEST] <BtbN> I'm already short on time for just one review.
[17:06:25 CEST] <andrey_turkin> anyway, as long as CUDA frames support lands in ffmpeg I'd be happy
[17:06:37 CEST] <BtbN> Isn't that mostly useless at the moment?
[17:06:49 CEST] <andrey_turkin> not for me
[17:06:55 CEST] <nevcairiel> the libav version supports cuda frame input
[17:07:04 CEST] <BtbN> yes, but nothing else supports cuda frames.
[17:07:21 CEST] <andrey_turkin> I know but libav doesn't support things from ffmpeg like e.g. frame metadata
[17:08:56 CEST] <BtbN> I'll try to get the merge on its way this weekend.
[17:10:07 CEST] <BtbN> Not sure how to handle it though, can't realy send a merge to the ML.
[17:10:15 CEST] <andrey_turkin> why not?
[17:10:26 CEST] <BtbN> I could send a link to the merged branch
[17:13:14 CEST] <andrey_turkin> what's wrong with sending a patch to the ML? Of course it would look scary but still
[17:13:51 CEST] <nevcairiel> some people care to finish their work before sending it =p
[17:14:24 CEST] <BtbN> There is no simple patch
[17:14:35 CEST] <andrey_turkin> I don't see much difference between sending a patch to ML and pointing to a branch with the same commitdiff
[17:14:39 CEST] <BtbN> I'm pretty sure format-patch on the merge commit won't produce anything sensible
[17:22:46 CEST] <andrey_turkin> so let me reiterate; you got all relevant stuff from ffmpeg and merged it to libav (so libav now has best of both worlds when it comes to nvenc). Now you want to bring it all back to ffmpeg and the plan is to make a commit which basically replaces nvenc implementation with libav version. Sounds about right?
[17:22:46 CEST] <michaelni> ive added some simple custom anti spam check to ffmpeg trac, tests & comments welcome
[17:23:28 CEST] <pfelt> greatest of morning everyone!
[17:23:42 CEST] <BtbN> andrey_turkin, i want to merge the development branch we used.
[17:23:50 CEST] <BtbN> So a merge commit, two parents and stuff.
[17:25:46 CEST] <nevcairiel> any distinct changes should ideally be submitted as such
[17:26:36 CEST] <nevcairiel> having one huge merge change everything causes headaches for bug finding later
[17:27:37 CEST] <andrey_turkin> I didn't think libav and ffmpeg actually merged to each other (as in git merges)
[17:29:59 CEST] <BtbN> well, we took the libav nvenc encoder, and put all missing features and fixes from ffmpeg into it.
[17:30:07 CEST] <BtbN> So there aren't realy any distinct changes on the ffmpeg side.
[17:46:54 CEST] <andrey_turkin> BtbN: so this branch you pointed to - is this that's slated to get into ffmpeg? I mean, is it going to be a merge with --strategy=their so whole existing content will be thrown out and replaced with what's in this branch?
[17:51:23 CEST] <BtbN> not sure how to do that yet
[18:01:39 CEST] <michaelni> BtbN, IIUC you didnt work on the ffmpeg codebase ? if so theres no magic its just not in ffmpeg
[18:02:32 CEST] <BtbN> yes, it's a libav tree. Just copying over the nvenc files and adjusting the build system works though.
[18:02:59 CEST] <andrey_turkin> it works, but that commit would be hell to review
[18:03:32 CEST] <michaelni> BtbN, thats possible but isnt that a maintaince nightmare ?
[18:03:51 CEST] <BtbN> yes, that's why I'm not sure how to do the merge yet.
[18:04:33 CEST] <andrey_turkin> if you (anyone) wants, I can try to break out some of the difference between the two codebases into (semi)readable chunks, so that remaining diff is more approachable
[18:07:34 CEST] <michaelni> BtbN, without looking at the diff, i suspect that spliting the diff manually into self contained changes is the only clean option
[18:07:52 CEST] <andrey_turkin> as I said I'd be happy to do that
[18:08:10 CEST] <BtbN> that sounds close to impossible, as the implementations aren't common at all.
[18:08:55 CEST] <michaelni> i still dont understand why you didnt do the work on the ffmpeg codebase ?
[18:09:06 CEST] <michaelni> in the first place that is ...
[18:09:26 CEST] <BtbN> Less stuff to change on the libav side to reach a common set of features.
[18:10:16 CEST] <BtbN> We could also just merge your patches on the ML, and would have achived pretty much the same. Which might be the better thing to do now, as you already wrote them.
[18:10:19 CEST] <andrey_turkin> BtbN: I think it's not that hard to bring diff down a lot. Remaining churn will be harder to deal with but it would be at least redable
[18:12:46 CEST] <andrey_turkin> It would be great to have same (or at least minimally divergent) code at least in some parts of the two combined codebases
[18:16:15 CEST] <michaelni> BtbN, you are the maintainer of that code, you have to decide what you prefer / what is best. I dont object to replace the code if you prefer to maintain the resulting code and history
[18:22:24 CEST] <BtbN> andrey_turkin, your patches bring the ffmpeg encoder quite close to the libav one, I think the remaining differences(primarily the diffrent dts/pts calculation from what I can see) can be merged individually.
[18:27:30 CEST] <andrey_turkin> my thoughts exactly
[18:28:00 CEST] <andrey_turkin> and then it's just 2-3 more mechanical patches to rename some fields and we are done
[18:32:50 CEST] <andrey_turkin> well, if we are going this way then my initial question remains - please someone review and apply my patches )
[18:33:42 CEST] <BtbN> I hope next weekend is still early enough?
[18:37:54 CEST] <andrey_turkin> I don't really care about timeframes as long as someone eventually gets to them. I'll be doing some more of these in the mean time
[20:23:23 CEST] <BtbN> michaelni, the antispam redirects me to a page that trac has no handler for.
[20:24:04 CEST] <BtbN> oh, nevermind. Didn't notice that I was on the mplayer trac...
[20:37:01 CEST] <cone-557> ffmpeg 03Michael Niedermayer 07master:35b37591c034: avfilter/af_hdcd: Replace complexer expression by litteral constant
[20:37:01 CEST] <cone-557> ffmpeg 03dsmudhar 07master:04d35004c3d0: libavfilter/vf_hwdownload: show error when ff_formats_ref fails
[20:59:05 CEST] <cone-557> ffmpeg 03dsmudhar 07master:49640ae315ab: libavcodec/dv_tablegen: fixed wrong printf format
[21:39:00 CEST] <BBB> anyone know what type of algebra Im looking for? I have a list of N numbers and am trying to create M (where M <= N) averages of subgroups such that the sum of some kind of error over each number and its subgroup average is minimized
[21:39:14 CEST] <BBB> brute-force is somewaht easy but subideal
[21:42:09 CEST] <fritsch> BBB: can you write the optimization problem in tex?
[21:42:20 CEST] <fritsch> BBB: so that I am sure what you are looking for?
[21:42:21 CEST] <BBB> I dont know tex :-p
[21:42:38 CEST] <fritsch> M averages of subgroups
[21:42:52 CEST] <fritsch> example: {1,2,3,4,5} ->
[21:42:55 CEST] <fritsch> tell me :-)
[21:43:27 CEST] <fritsch> i want to get the definition of subgroups, is it elements of the power set?
[21:43:31 CEST] <fritsch> non disjunct?
[21:43:37 CEST] <fritsch> or exclusiv and non overlapping?
[21:44:04 CEST] <fritsch> or in other words: the way you formulate your problem does not ring a bell :-)
[21:47:49 CEST] <BBB> lets say you have {1,2,3,4,5,6,7,8,9,10} (N=10) and M=2
[21:48:54 CEST] <BBB> if I create two averages of 3 and 7, then my sum_error (lets assume my error metric is sse) is (3-1)^2 + (3-2)^2 etc. up until 6, where its (7-6)^2 up to 10
[21:49:25 CEST] <BBB> I Can just try two averages and set them iteratively between anything in between 1 and 10 and try everything
[21:49:27 CEST] <BBB> but thats slow
[21:49:54 CEST] <BtbN> andrey_turkin, do you happen to have a git branch somewhere i can pull the nvenc patches from?
[21:50:15 CEST] <andrey_turkin> I can upload them
[21:52:06 CEST] <fritsch> BBB: is the list sorted?
[21:53:26 CEST] <fritsch> what you are basically doing is a k-means with k=M
[21:53:48 CEST] <fritsch> so you need a clustering algorithm
[21:54:13 CEST] <fritsch> see: https://en.wikipedia.org/wiki/K-means_clustering
[21:54:43 CEST] <fritsch> it's np hard :-)
[21:54:55 CEST] <fritsch> so if you find a better solution tell me, hehe
[21:55:50 CEST] <andrey_turkin> BtbN: https://github.com/tea/FFmpeg/commits/nvenc-merge . There is one extra patch at the front of the queue not related to nvenc per se
[21:55:57 CEST] <BBB> ok so it is k-means
[21:56:09 CEST] <fritsch> from what you describe, yes
[21:56:10 CEST] <BBB> k-means kind of sucks
[21:56:15 CEST] <BBB> and yes the list can be sorted
[21:56:27 CEST] <fritsch> find M clusters of N elements
[21:57:07 CEST] <BBB> yes
[21:57:16 CEST] <fritsch> so that all elements in a cluster M' of M have a distance d' which is minimal compared to other clusters M''
[21:59:17 CEST] <fritsch> BBB: the results needs to be "fully correct"?
[22:02:00 CEST] <BBB> no
[22:02:07 CEST] <BBB> just something that is not entirely stupid
[22:02:24 CEST] <BBB> when does anything ever have to be correct, right? I mean, its not like Im shooting rockets at Pluto :-p
[22:04:00 CEST] <fritsch> fast approximate nearest neighbour
[22:14:35 CEST] <fritsch> BBB: if you get the array sorted it's much easier, if it's sorted and disjunct, e.g. no doublettes, even more and if it's sequential it's trivial :-)
[00:00:00 CEST] --- Wed May 25 2016


More information about the Ffmpeg-devel-irc mailing list