[Ffmpeg-devel-irc] ffmpeg-devel.log.20180227

burek burek021 at gmail.com
Wed Feb 28 03:05:04 EET 2018


[01:01:39 CET] <gagandeep> kierank: the course on coursera that i was looking into is beyond the simple codec functioning (like image enhancement and other stuff)
[01:02:20 CET] <gagandeep> kierank: i also encountered a term called information theory, should i look into it for basic understanding regarding compression and decompression
[01:02:28 CET] <kierank> not really
[01:03:42 CET] <gagandeep> kierank: for the last two days i was struggling with installing the ffmpeg using source and then later figuring out something called pkg-config through which one can finally use libraries
[01:04:01 CET] <kierank> you don't need pkg-config to build ffmpeg
[01:04:12 CET] <gagandeep> i was trying to follow a tutorial for using libav
[01:05:07 CET] <kierank> https://trac.ffmpeg.org/wiki/CompilationGuide/Generic
[01:05:09 CET] <kierank> just follow that
[01:06:04 CET] <gagandeep> i followed that and had ffmpeg already, it was to use the libav that i was struggling as i had never used external libraries
[01:07:29 CET] <gagandeep> basically i was confused that how avcodec.h is including avutil.h as it was not in the same directory
[01:07:45 CET] <gagandeep> when i was looking into the source code
[01:08:06 CET] <kierank> you shouldn't need to use any of the libraries for this
[01:08:14 CET] <kierank> this entire task is doable with ffmpeg.c
[01:08:20 CET] <kierank> and you don't need to read that
[01:08:25 CET] <kierank> just do ./ffmpeg -i blah
[01:09:38 CET] <gagandeep> what do you mean by this line
[01:12:04 CET] <gagandeep> kierank:what do you mean by ./ffmpeg -i
[01:12:12 CET] <kierank> it's the ffmpeg application
[01:12:23 CET] <kierank> that it builds by default
[01:13:10 CET] <gagandeep> oh!! so if one runs ffmpeg, all the files will be built automatically
[01:13:55 CET] <kierank> ?
[01:14:04 CET] <kierank> no, you build the git tree and then you get the ffmpeg application
[01:39:10 CET] <BBB> greentea: in case you didnt get an answer yet, just ask :)
[02:18:09 CET] <cone-288> ffmpeg 03James Almer 07master:f235359b2b11: avformat/utils: don't overwrite the return value of read_packet()
[05:23:32 CET] <cone-288> ffmpeg 03Serhii Marchuk 07master:6d763ab58811: avformat/dashenc: add option to change HTTP method
[05:23:33 CET] <cone-288> ffmpeg 03Karthick Jeyapal 07master:e3851f6a3739: avformat/dashenc: Removed usage of deprecated 'filename' variable
[08:36:41 CET] <cone-697> ffmpeg 03Tobias Rapp 07master:fa0c9d69d3d7: fftools/ffmpeg: replace call to av_strerror with av_err2str
[11:39:55 CET] <bogdanc> where is the definition of AVFrame ?
[11:42:21 CET] <nevcairiel> libavutil/frame.h
[11:42:37 CET] <wm4> a quick grep would tell you too
[11:45:52 CET] <bogdanc> couldnt really tell from all the "AVFrame" in all the headers
[11:46:16 CET] <nevcairiel> "struct AVFrame" would probably find it right away i reckon
[11:47:04 CET] <bogdanc> didn't cross my mind to also put typedef in there
[11:47:24 CET] <bogdanc> or struct..
[12:23:27 CET] <wm4> so would the nvidia headers change actually require real hw to test it? (that means whether it really manages to dlopen the driver etc.)
[12:28:31 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:bbed942dfd64: avcodec/rscc: Skip empty frames (nb_tiles == 0)
[12:28:31 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:7414d0bda776: avcodec/utvideodec: Check subsample factors
[12:50:21 CET] <nevcairiel> wm4: probably not, its just a header afterall, it just needs to find it
[13:33:15 CET] <BtbN> wm4, no, it's all runtime-loaded.
[13:33:34 CET] <BtbN> And I suspect it might actually work on MSVC, if you convince configure and pkg-config hard enough. Will try again right now.
[13:37:49 CET] <BtbN> Visual Studio wants to update itself before doing anything. Great.
[13:41:26 CET] <wm4> lol
[13:41:43 CET] <BtbN> I don't overly care about MSVC, but I don't want to break a configuration that works right now. If I manage to make it build, I'll just push the patch.
[13:42:07 CET] <BtbN> It estimates 2 hours for the update. Yeah, I'll go grab food I guess.
[14:00:49 CET] <wm4> blocking forced autoupdate are just MS' thing, huh
[14:40:02 CET] <nevcairiel> dont think i have ever seen that with msvc
[14:40:11 CET] <nevcairiel> it tells you there is an update, but forced?
[14:40:47 CET] <wm4> dunno, apparently it stopped BtbN from doing his thing
[14:42:14 CET] <BtbN> nevcairiel, it looks like it was started previously, but canceled
[14:42:38 CET] <BtbN> so the whole thing was in a half-updated state. Even cl on the commandline just opened the visual studio installed and started the update
[14:43:23 CET] <BtbN> but it's done now, so I can actually do stuff now
[15:33:33 CET] <BtbN> Does msys really have cygpath?
[15:33:42 CET] <JEEB> yes
[15:34:43 CET] <BtbN> ok, so I will just add a parameter to the nv-codec-headers repo Makefile for it to write native Windows paths into the pkg-config file and call it a day
[15:35:37 CET] <nevcairiel> mingw can understand native windows paths, so maybe it should always do that on windows then? could probably do that without parameters
[15:36:13 CET] <BtbN> Cygwin gcc should as well
[15:36:28 CET] <BtbN> But how to reliably detect Windows in a Makefile?
[15:37:40 CET] <nevcairiel> ifeq ($(OS),Windows_NT)
[15:37:54 CET] <nevcairiel> unless cygwin is evil and overrides that
[15:38:01 CET] <BtbN> it doesn't seem to
[15:44:25 CET] <BtbN> This whole thing is made annoying by the 15 minute breaks to wait for configure to run
[15:47:29 CET] <BtbN> But it takes way longer now than it did before. So I guess this was a success
[15:47:41 CET] <nevcairiel> heh
[15:47:51 CET] <nevcairiel> can always grep the config.log to check
[15:48:15 CET] <BtbN> I want to build-test it anyway, so might as well wait
[15:56:02 CET] <jdarnley> BtbN: If you're actually running in cygwin then use dash.  It cuts the run time to a mere 2/3rds.
[15:56:25 CET] <jdarnley> I can't say whether msys or other shells have it.
[15:58:45 CET] <nevcairiel> dash comes with  msys as well
[16:05:38 CET] <BtbN> so explicitly invoke configure with dash you mean? I'll try that
[16:06:15 CET] <nevcairiel> one of these days someone should figure out how to make the slow fork emulation fast
[16:08:37 CET] <jdarnley> Yes I do mean runf: dash ./configure --whatever
[16:14:20 CET] <BtbN> The only way to make it fast it to make the Kernel do it
[16:16:11 CET] <BtbN> still takes ages with dash
[16:23:03 CET] <cone-549> ffmpeg 03Timo Rothenpieler 07master:27cbbbb33f25: compat: remove in-tree NVidia headers
[16:26:00 CET] <JEEB> yayifications
[16:29:23 CET] <wm4> nice
[16:30:45 CET] <wm4> who can add nv-codec-headers.git to the github mirror?
[16:32:20 CET] <durandal_1707> oh noo, you ignored Nicolas and pushed patch
[16:33:28 CET] <wm4> no he didn't
[16:33:44 CET] <wm4> the discussion was over when he confirmed it worked on MSVC (or actually pushed a fix to the headers repo)
[16:35:42 CET] <durandal_1707> you missed last nicolas mail then
[16:38:50 CET] <jamrial> BtbN: run configure with --disable-everything
[16:39:02 CET] <jamrial> it doesn't disable autodetect stuff
[16:39:44 CET] <jamrial> and it's much faster since it's not doing a thousand recursive dep checks
[18:02:13 CET] <BtbN> durandal_1707, you mean the one to wait a week?
[18:02:35 CET] <BtbN> That was if nobody comes up to fix MSVC. Which happened, as I finally got around to fixing it.
[18:56:38 CET] <BtbN> I guess it's michaelni for the github mirror? But I'm not sure who managed the actual mirroring
[18:59:14 CET] <jamrial> BtbN: nicolas sent his email at the exact same minute you sent your "fixed and pushed" email
[18:59:21 CET] <jamrial> hence the confusion
[19:03:53 CET] <BtbN> I think his mail was also about "Push if nobody fixes it"
[19:04:00 CET] <BtbN> It's fixed, so I don't see an issue
[19:38:27 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:090c0abff9c8: avcodec/bintext: sanity check dimensions
[19:38:28 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:49062a90174b: avcodec/exr: fix invalid shift in unpack_14()
[19:38:29 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:3981fb8d2a03: avcodec/g2meet: Check tile dimensions with av_image_check_size2()
[19:38:30 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:8b5c29b6c2ab: avcodec/truemotion2rt: Check input buffer size
[19:38:31 CET] <cone-549> ffmpeg 03Michael Niedermayer 07master:f9cb17f988cc: avcodec/msmpeg4dec: Check for input end in msmpeg4v34_decode_mb()
[21:27:38 CET] <chance83> is there a way to stop ffprobe decode errors by having it wait to decode until it gets sps/pps?
[21:31:15 CET] <atomnuker> FUCK YEAH IT WORKS
[21:31:17 CET] <BtbN> if a file does not provide sps/pps immediately it's broken. Also, not a question for -devel
[21:31:18 CET] <atomnuker> https://pars.ee/temp/sharpen_vulkan.webm
[21:31:35 CET] <atomnuker> vaapi decode, mapping to vulkan, unsharp_vulkan and a hwdownload
[21:31:51 CET] <atomnuker> speed is _an_order_of_magnitude_ better than opencl
[21:32:07 CET] <durandal_1707> show the numbers
[21:32:19 CET] <atomnuker> 6 frames per second vs 60
[21:32:24 CET] <chance83> BtbN think of udp stream
[21:32:45 CET] <durandal_1707> atomnuker: something is not right!
[21:32:45 CET] <atomnuker> yep, 6 frames per second for opencl sharpening with vaapi interop
[21:32:47 CET] <BtbN> what's with a UDP stream?
[21:33:12 CET] <atomnuker> its 25 for nvidia with opencl and software decoding and manual uploading
[21:33:12 CET] <BtbN> you better use some container that repeats the extradata frequently for a lossy transport
[21:33:31 CET] <chance83> when you attach you arent gauranteed to have it at the start
[21:33:47 CET] <BtbN> well, can't decode until you get it
[21:33:54 CET] <BtbN> also, still not a topic for -devel
[21:34:41 CET] <atomnuker> of course it looks ruined by 20 fucking minutes of libvpx-vp9 encoding but oh well
[21:37:58 CET] <atomnuker> keep in mind this is for a 4k video so its quite intensive
[21:38:46 CET] <atomnuker> I could improve things a lot more for vulkan too, I do work on 16x16 tiles but should probably tune this to whatever the device has
[21:39:08 CET] <durandal_1707> atomnuker: is that convolution of 3x3 kernel?
[21:39:29 CET] <atomnuker> yep, standard sharpen/blur kernel
[21:39:54 CET] <BtbN> now write a CUDA one and compare :P
[21:40:30 CET] <atomnuker> can't, none of the cuda filters work for me, despite hardware decoding working
[21:40:39 CET] <atomnuker> because configure can't find cuda.h
[21:40:40 CET] <BtbN> uh, what? What breaks?
[21:40:51 CET] <BtbN> you need the CUDA SDK to build them
[21:41:09 CET] <atomnuker> from the nvidia website?
[21:41:10 CET] <BtbN> most linux distros have it in their (non-free) repos
[21:41:21 CET] <BtbN> at least Ubuntu and Gentoo do
[21:41:31 CET] <atomnuker> I think I have the sdk, its like a gig on archlinux and provides cuda
[21:41:52 CET] <BtbN> Yeah, that's all you need. Probably need to set proper include paths, as it usually goes into /opt
[21:43:22 CET] <atomnuker> right, that bit me when I tried to install vulkan's validation layer which went into /usr/local/etc/ but here they should've been in /usr/local/share
[21:51:48 CET] <cone-549> ffmpeg 03James Almer 07master:40102a213740: avcodec/Makefile: skip nvdec.h header when nvdec is not enabled
[22:00:51 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:876ecfccfb27: avcodec/dxtory: Remove code that corrupts dimensions
[22:00:52 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:b4135fb335f0: avcodec/diracdec: Use int64 in global mv to prevent overflow
[22:00:53 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:afc85dacba4b: avcodec/jpeg2000dwt: Fix integer overflows in sr_1d53()
[22:00:54 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:6648d3fef6b0: avcodec/aacdec_templat: Fix integer overflow in apply_ltp()
[22:00:55 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:6822bd50c1ea: avcodec/diracdec: Fix integer overflow in mv computation
[22:00:56 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:24a3c45da511: avcodec/cavsdec: Check alpha/beta offset
[22:00:57 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:789a12b140ba: avcodec/smc: Check input packet size
[22:00:58 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:fbf690d79a61: avcodec/utvideodec: Check subsample factors
[22:00:59 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:add3c2468e96: avcodec/bintext: sanity check dimensions
[22:01:00 CET] <cone-549> ffmpeg 03Michael Niedermayer 07release/3.0:b910b3492665: Changelog: update
[23:40:19 CET] <cone-549> ffmpeg 03James Almer 07n3.0.11:HEAD: avcodec/Makefile: skip nvdec.h header when nvdec is not enabled
[00:00:00 CET] --- Wed Feb 28 2018


More information about the Ffmpeg-devel-irc mailing list