[Ffmpeg-devel-irc] ffmpeg-devel.log.20180602

burek burek021 at gmail.com
Sun Jun 3 03:05:04 EEST 2018

[02:10:06 CEST] <cone-026> ffmpeg 03Jacob Trimble 07master:9827bb88e7dc: libavformat/mov: Fix heap buffer overflow.
[13:11:21 CEST] <cone-226> ffmpeg 03Paul B Mahol 07master:e28b1fa6e99c: avfilter: add adeclick and adeclip audio filters
[13:46:39 CEST] <durandal_1707> atomnuker: i get this one /video/out/wayland_common.c:901:14: error: XDG_TOPLEVEL_STATE_TILED_RIGHT undeclared (first use in this function) --> is some other wayland now needed?
[13:49:18 CEST] <JEEB> yes that needs a newer version of wayland-protocols, when merging his changes I bumped that in configure
[13:49:28 CEST] <JEEB> and the wayland people told me "just vendor the XML"
[13:49:30 CEST] <JEEB> :V
[13:50:50 CEST] <JEEB> https://github.com/mpv-player/mpv/commit/d8288877923728dad3c5ce6c735a9b4cde4af705
[14:27:32 CEST] <durandal_1707> JEEB: i have latest protocols installed already...
[14:29:32 CEST] <durandal_1707> looks like i need do do full clean
[14:29:37 CEST] <JEEB> not sure at which point the headers get generated but most likely you missed some point or something by not running configure (or the wayland-protocols 1.14 is too new, which is why #wayland was OK with just vendoring that crap)
[14:47:44 CEST] <gagandeep> kierank: that 2nd frame is working now, i didn't need to do any av_frame stuff as the buffers which held the data of coefficients were not being freed so i could use them directly
[14:48:06 CEST] <gagandeep> just working out the offset values and 'prescales' to get bitexact output
[14:48:21 CEST] <kierank> you should start doing bitexact testing on all the samples
[14:48:42 CEST] <durandal_1707> gagandeep: how many past frames cfhd can use?
[14:49:12 CEST] <gagandeep> right now 3d transform is only using 2 frames as group
[14:49:24 CEST] <gagandeep> which are passed through temporal transform
[14:50:01 CEST] <gagandeep> i don't think it is working like mpeg where past frames are used for construction of new frames
[14:50:45 CEST] <gagandeep> it just used lowpass bands of two frames(by doing spatial transform) through a temporal transform
[14:52:56 CEST] <gagandeep> a total of 17 subbands are used to reconstruct 2 frames and cycle repeats
[14:53:17 CEST] <kierank> gagandeep: sure, but if you are doing threading you need to make sure the first frame has finished decoding (well not decoding per se, but coefficients are done)
[14:53:26 CEST] <kierank> gagandeep: did you remove my in place optimisations then?
[14:53:49 CEST] <gagandeep> which?, i only removed .capabilities
[14:54:31 CEST] <gagandeep> other than that changed how some error handling was done as mountain sample was not having some tags in right place
[14:54:33 CEST] <kierank> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/cfhd.c#L305
[14:54:50 CEST] <kierank> i'm surprised you don't need to remove the inplace optimisations
[14:55:29 CEST] <gagandeep> i have read the code and i didn't see any problem with any of the transform stuff
[14:55:38 CEST] <kierank> interesting
[14:56:43 CEST] <gagandeep> though i am assuming the part of code you are asking about is after all subbands are read, right?
[14:57:09 CEST] <kierank> during allocation
[14:57:11 CEST] <kierank> so
[14:57:12 CEST] <kierank> 8:32:33 PM <durandal_1707> kierank: ff_thread_report_progress() ff_thread_await_progress() this are calls for synchronization between different threads, (A)PNG is i think simplest decoder that use them
[14:57:23 CEST] <kierank> after bitexact testing I think you need to add threading
[14:57:40 CEST] <gagandeep> yeah, will begin to read on threading
[14:58:02 CEST] <gagandeep> will look in png
[14:58:56 CEST] <kierank> you probably need to copy pointers between threads when frame 1 has finished unpack
[15:01:11 CEST] <durandal_1707> gagandeep: you also call ff_thread_release_buffer() after you do not need thread frame any more...
[15:01:48 CEST] <gagandeep> ok, guys, will ask when i start looking into it on probably monday
[15:01:57 CEST] <gagandeep> then will be asking a lot
[15:02:05 CEST] <gagandeep> like a fool
[15:02:08 CEST] <gagandeep> :)
[15:02:49 CEST] <kierank> that ami_stuff guy needs to come and provide samples
[15:03:11 CEST] <gagandeep> also might need to explore how cineformsdk handles threads
[15:03:14 CEST] <kierank> gagandeep: you should make new temporal_inverse_filter function
[15:03:17 CEST] <kierank> gagandeep: i would not do that
[15:03:21 CEST] <gagandeep> k
[15:03:22 CEST] <kierank> just look at how ffmpeg does it
[15:03:28 CEST] <gagandeep> ok
[15:03:35 CEST] <kierank> when you do simd, you can't do if (!temporal_for_highpass) {
[15:03:44 CEST] <kierank> well you can but it's inefficient
[15:03:56 CEST] <gagandeep> k, will keep in mind
[15:04:38 CEST] <kierank> it's good there are simple simd functions
[16:27:49 CEST] <rindolf> Hi all! tmm1 : some days ago i sent a patch to http://ffmpeg.org/pipermail/ffmpeg-devel/ but it isn;t there yet
[16:29:37 CEST] <JEEB> probably  in the moderation queue then
[16:29:52 CEST] <JEEB> never 100% understood who was maintaining that
[16:47:21 CEST] <cone-226> ffmpeg 03Gyan Doshi 07master:e50a5c9c4eb7: doc/ffmpeg: rewrite Stream Selection chapter
[18:20:01 CEST] <JEEB> would be nice if someone could ack https://patchwork.ffmpeg.org/patch/9228/
[18:20:22 CEST] <JEEB> (or tell how this should be "properly" fixed if that's not the way to fix it)
[18:21:38 CEST] <JEEB> also the committer is listed as pedro arthur... http://git.videolan.org/?p=ffmpeg.git;a=commit;h=bdf1bbdbb4ebb342c0267d0f77cd06e717197e65 I wonder if he's on IRC
[18:25:01 CEST] <jdarnley> JEEB: from the description you gave it sounds like an okay solution for now
[18:25:36 CEST] <jdarnley> If something else is added with depends on the dnn thing then I gues it will need it own "module" in configure
[18:25:36 CEST] <JEEB> yea someone on #ffmpeg just had most filters in libavfilter disabled
[18:25:41 CEST] <JEEB> yea
[18:25:45 CEST] <JEEB> that's the way I felt
[18:37:33 CEST] <durandal_1707> atomnuker: actual useful vulkan filter would be lut3d
[18:46:30 CEST] <jamrial> JEEB: patch ok
[18:48:17 CEST] <JEEB> ok
[18:49:29 CEST] <cone-226> ffmpeg 03Jan Ekström 07master:76daf4529bf2: lavfi/Makefile: move dnn_*.o under vf_srcnn
[18:49:32 CEST] <JEEB> pushed
[19:19:13 CEST] <cone-226> ffmpeg 03Paul B Mahol 07master:af31084399b8: avfilter/vf_lut3d: unbreak haldclut with planar rgb formats
[21:48:32 CEST] <cone-226> ffmpeg 03Michael Niedermayer 07master:5ee203076fa1: avcodec/vp3: Fix end of bitstream check in unpack_superblocks()
[23:48:01 CEST] <atomnuker> TD-Linux: actually forcing gpu rendering on linux makes things worse
[23:48:28 CEST] <atomnuker> firefox with gpu rendering always damages the whole screen, if you leave it disabled it'll damage only areas that need to be damaged
[00:00:00 CEST] --- Sun Jun  3 2018

More information about the Ffmpeg-devel-irc mailing list