[Ffmpeg-devel-irc] ffmpeg-devel.log.20151204

burek burek021 at gmail.com
Sat Dec 5 02:05:02 CET 2015


[00:14:13 CET] <cone-020> ffmpeg 03Rostislav Pehlivanov 07master:ed08cbd7b172: aacenc_ltp: fix out of bounds memory access
[00:36:08 CET] <atomnuker> aac-pns-encode fails on kfreebds, last know commit is listed as N-76962-g736e2e2
[00:36:37 CET] <atomnuker> seems it started failing some time ago, but I can't find any commit with a hash containing g736e2e2
[00:39:34 CET] <nevcairiel> The g is not part of the hash
[00:40:44 CET] <nevcairiel> (It's hex after all)
[00:40:52 CET] <llogan> that g confuses many people. including me a few times
[00:47:01 CET] <atomnuker> one of ganesh's commits broke it, probably the tablegen utils and  maybe changed bit allocation on that particular system
[00:52:59 CET] <rcombs> what's the g mean anyway
[00:55:26 CET] <kierank> git
[01:00:23 CET] <rcombs> ah
[02:05:20 CET] <cone-020> ffmpeg 03Michael Niedermayer 07master:cafb19560401: avformat/utils: Fix "libavformat/utils.c:927:35: runtime error: signed integer overflow: -2450238577049583619 - 9223090561878065151 cannot be represented in type long long"
[02:05:21 CET] <cone-020> ffmpeg 03Michael Niedermayer 07master:77a644e6fa4a: avcodec/h264_direct: Check CO-POCs before use
[02:24:06 CET] <BBB> nevcairiel: whenever I see the words nal, slice etc used in vp9 context, I get the creeps :D
[02:38:07 CET] <cone-020> ffmpeg 03Sebastian Dröge 07master:eb3628d87f67: mpegtsenc: Fix off-by-one in indexing into Opus channel mapping arrays
[02:38:08 CET] <cone-020> ffmpeg 03Vicente Olivert Riera 07master:a27401a05ba3: mips: rename mipsdspr1 to mipsdsp
[03:19:30 CET] <cone-020> ffmpeg 03Michael Niedermayer 07master:9696a01f1262: avformat/mpjpegdec: Fix "libavformat/mpjpegdec.c:269:9: warning: passing argument 3 of av_stristart from incompatible pointer type"
[03:52:48 CET] <cone-020> ffmpeg 03Yann Coupin 07master:d89c4846e0d6: avformat/movenc: HE-AAC correct FourCC in ISML
[05:15:46 CET] <Mista_D>  HWAccelIntro is missing any info about QSV encoders/decoderes. Is there other place that shows it? 
[06:29:26 CET] <Timothy_Gu> Mista_D: there used to be a post on ffmpeg-devel about how to use it. cant comment on the usefulness though, and I gave up trying to install Intel's media sdk on linux a long time ago
[06:30:06 CET] <Timothy_Gu> grep for a thread originated by Stefano Sabatini, I believe Ivan Uskov was the one who answered it
[06:30:14 CET] <Timothy_Gu> might be a few months ago
[09:48:19 CET] <BtbN> QSV is not a classic ffmpeg hwaccel.
[10:28:07 CET] <ubitux> why are these ppl hijacking my bool thread
[10:28:09 CET] <ubitux> :(
[13:26:10 CET] <durandal_170> Compn: drones are using equirectangular
[13:32:34 CET] <Compn> my single greatest fear!
[13:32:35 CET] <Compn> ehe
[13:38:32 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:18268f761bff: avcodec/vp3: Fix "runtime error: left shift of negative value"
[13:38:33 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:9d44e236187c: ffmpeg: Check pkt_dts for AV_NOPTS_VALUE
[14:16:46 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:2ff61c3c1a0a: avcodec/dnxhddec: Fix runtime error: left shift of negative value
[14:16:47 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:0c1d62ab9d75: avformat/oggparsetheora: Check duration to be not AV_NOPTS_VALUE (and positive) before use
[14:55:35 CET] <cone-435> ffmpeg 03Hendrik Leppkes 07master:d6b988b505d6: libavutil/tablegen: add missing math.h include
[14:55:36 CET] <cone-435> ffmpeg 03Hendrik Leppkes 07master:42a9aead854f: avcodec: include tablegen compat header before the tablegen functionality
[14:55:37 CET] <cone-435> ffmpeg 03Hendrik Leppkes 07master:3c4e49187f83: libavcodec/aacsbr_tablegen: fix tablegen on windows
[15:01:33 CET] <kierank> BBB: is there a standard way of unpacking coefficients with multiple bit depths or is it a case of if(foo){AV_WN16}else{AV_WN32}?
[15:01:50 CET] <kierank> ah seems macroed in h264
[15:01:58 CET] <BBB> I do that, with macros and inline funcs
[15:02:09 CET] <BBB> so the check is essentially once per coef block readoing, not once per coef
[15:02:40 CET] <kierank> yeah
[15:02:45 CET] <kierank> atomnuker: ^
[15:03:07 CET] <BBB> youre probably going to say dude this is pretty hacky at some point, looking at how codecs implement multi-bitdepth
[15:03:14 CET] <BBB> and I cant entirely deny it
[15:03:17 CET] <kierank> can't any other way
[15:03:19 CET] <kierank> see*
[15:03:22 CET] <BBB> but it works and the performance impact is low, so I think its ok
[15:03:25 CET] <BBB> yeah exactly
[15:03:33 CET] <atomnuker> dude this is pretty hacky
[15:04:34 CET] <wm4> just write it in C++ and use templates
[15:07:37 CET] <BBB> c++ is unsafe, you should use python
[15:08:32 CET] <Daemon404> only pure functional languages are safe
[15:08:55 CET] <Daemon404> furthermore you should formally prove your code and algo is correct
[15:09:02 CET] <Daemon404> and publish 6 papers
[15:13:30 CET] <BBB> dont forget to get tenure
[15:14:49 CET] <Daemon404> that doesnt exist anymore
[15:19:03 CET] <durandal_170> python is soo slow
[15:20:55 CET] <cone-435> ffmpeg 03Andreas Cadhalpun 07master:488e9a06f014: ffmdec: only seek back at EOF if a server is attached
[15:20:56 CET] <cone-435> ffmpeg 03Andreas Cadhalpun 07master:a611375db532: ffmdec: reject zero-sized chunks
[15:21:03 CET] <BBB> durandal_170: thats only because your datacenter isnt big enough
[15:21:13 CET] <BBB> havent you heard of high performance computing?
[15:21:14 CET] <BBB> its very simple
[15:21:18 CET] <BBB> if your software is too slow
[15:21:21 CET] <BBB> just buy more datacenters
[15:21:28 CET] <BBB> and now your software is fast again!!!!!
[15:21:41 CET] <BBB> \o/
[15:22:54 CET] <Daemon404> BBB, preferably using: FORTRAN, SciPy, OpenMP, and 900 nested loops
[15:23:05 CET] <BBB> yes, openmp
[15:23:12 CET] <BBB> and gccs vectorization
[15:23:21 CET] <BBB> I hear its really good at vectorizing memset(0) calls
[15:23:24 CET] <BBB> and thats about all it does
[15:23:26 CET] <kierank> double precision everywhere too
[15:24:45 CET] <nevcairiel> speaking of gcc, 5.3 was released
[15:25:31 CET] <nevcairiel> the only noteworthy highlight is skylake/avx-512 support, if you are into intrinsics
[15:25:36 CET] <BBB> speaking of papers, did you guys know I recently published a new paper?
[15:25:48 CET] <BBB> I think Im at 6 papers now, actually, so there you go Daemon404 
[15:26:54 CET] <Daemon404> ;)
[15:28:03 CET] <BBB> yeah I just counted, if I count only real published and peer-reviewed ones, its 6 papers
[15:28:09 CET] <BBB> if I include conference papers, its 8
[15:28:23 CET] <Daemon404> eg the best papers ive read usually come from cons
[15:28:31 CET] <BBB> write papers, or write simd
[15:28:32 CET] <BBB> hm...
[15:28:33 CET] <BBB> difficult
[15:28:46 CET] <Daemon404> (siggraph)
[15:52:31 CET] <cone-435> ffmpeg 03Clément BSsch 07master:fb99ef0bd39a: avcodec: use AV_OPT_TYPE_BOOL in a bunch of places
[15:52:32 CET] <cone-435> ffmpeg 03Clément BSsch 07master:0e62b5d1ef13: avfilter: use AV_OPT_TYPE_BOOL in a bunch of places
[15:52:33 CET] <cone-435> ffmpeg 03Clément BSsch 07master:43ecec0f0386: avformat: use AV_OPT_TYPE_BOOL in a bunch of places
[15:52:34 CET] <cone-435> ffmpeg 03Clément BSsch 07master:c1f114a8c4ff: swresample: use AV_OPT_TYPE_BOOL for linear_interp and cheby options
[15:52:35 CET] <cone-435> ffmpeg 03Clément BSsch 07master:010caed261ac: swscale: use AV_OPT_TYPE_BOOL for {src,dst}_range options
[15:52:36 CET] <cone-435> ffmpeg 03Clément BSsch 07master:60532348d254: avcodec/mpegvideo: use constants for rc_strategy
[15:52:37 CET] <cone-435> ffmpeg 03Clément BSsch 07master:7234e04e358b: ffmpeg: enable echoing with command and debug modes
[15:56:49 CET] <nevcairiel> hm gcc 5.3 tarballs not up yet
[15:56:52 CET] <nevcairiel> lazy gnus
[15:59:05 CET] <Daemon404> are you sure its released and not gnu-released
[15:59:14 CET] <Daemon404> iirc gcc team had some weird defintiion of released
[15:59:54 CET] <nevcairiel> they posted it on their main website
[16:00:01 CET] <nevcairiel> GCC 5.3 released [2015-12-04]
[16:01:39 CET] <nevcairiel> there is a tag and  everything
[16:01:41 CET] <nevcairiel> just no tarballs
[16:02:30 CET] <ubitux> note: my fate instances should be re-up in the next 24 hours if everything goes well; sorry about the downtime
[16:06:54 CET] <ubitux> "UndefinedBehaviorSanitizer gained a few new sanitization options"
[16:06:56 CET] <ubitux> cool.
[16:07:28 CET] <nevcairiel> note that this is the 5.0 changelog
[16:07:35 CET] <nevcairiel> 5.3 changes are far on the bottom
[16:08:02 CET] <ubitux> oh.
[16:08:28 CET] <ubitux> :(
[16:09:05 CET] <ubitux> michaelni: what's the usual way of testing sws?
[16:09:21 CET] <ubitux> using vf_scale?
[16:09:31 CET] <kierank> fate
[16:09:45 CET] <ubitux> yes well i mean within fate
[16:10:03 CET] <ubitux> context: testing the nv/rgb convert
[16:10:40 CET] <nevcairiel> not sure sws is that well tested in fate
[16:10:59 CET] <BBB> it has tons of tests
[16:11:04 CET] <BBB> all lavfi-fmt tests use sws
[16:11:15 CET] <nevcairiel> but you could probably just use a simple ffmpeg command line and request a specific outout format and size
[16:11:24 CET] <BBB> but they take forever :-p
[16:11:45 CET] <BBB> have I ever mentioned I hate race conditions
[16:11:53 CET] <ubitux> +1
[16:11:53 CET] <BBB> I just had another one, SOOOOOOOOO frustrating to debug these things
[16:12:23 CET] <ubitux> BBB: i spent a day writing a threadmessage fuzzer in order to make sure it wasn't broken
[16:12:30 CET] <ubitux> (and it was, in a way)
[16:12:37 CET] <BBB> poor you
[16:12:44 CET] <ubitux> and here the complexity is very very low
[16:13:17 CET] <ubitux> btw, if you want to review that patchset& :)
[16:13:32 CET] <nevcairiel> if the code is simple enough its usually easier to stare at it for a day than to try to reproduce it
[16:17:25 CET] <wm4> I don't know, I added a lot of threads to a very old single threaded codebase (mplayer), and the only kind I've run into was some missed wakeups
[16:19:30 CET] <ubitux> BBB: so where is the race this time?
[16:19:44 CET] <BBB> not in ffmpeg :-p
[16:19:46 CET] <BBB> in my own code
[16:19:48 CET] <ubitux> ahah
[16:19:50 CET] <ubitux> ok
[16:20:21 CET] <durandal_170> closed source?
[16:22:03 CET] <BBB> not open, for now, sorry
[16:22:19 CET] <BBB> its that stream analyzer I showed at VDD
[16:22:54 CET] <nevcairiel> its written using crazy mac tools anyway
[16:22:56 CET] <nevcairiel> keep it
[16:22:56 CET] <nevcairiel> :D
[16:23:15 CET] <ubitux> and hide it
[16:23:21 CET] <BBB> :(
[16:23:23 CET] <Daemon404> everything for writing guis is pain
[16:23:27 CET] <Daemon404> he just chose what is least pain to him
[16:24:19 CET] <ubitux> which reminds me i need to resurect the qp move to lavfi
[16:26:49 CET] <BBB> yes!
[16:28:40 CET] <wm4> isn't swift open source now
[16:29:28 CET] <J_Darnley> (Its not GPL so no!)
[16:30:18 CET] <Daemon404> wm4, cocoa is not
[16:30:32 CET] <Daemon404> J_Darnley, surely the stdlib isnt
[16:30:37 CET] <Daemon404> or at least it must have an exceptions
[16:30:38 CET] <Daemon404> -s
[16:31:20 CET] <J_Darnley> I just remember a link to a ballsy request on github to change the license to gpl
[16:31:28 CET] <Daemon404> that wasnt ballsy
[16:31:32 CET] <Daemon404> it was a lame troll
[16:31:47 CET] <J_Darnley> If no asks it won't happen
[16:32:07 CET] <Daemon404> it was not a serious PR
[16:54:49 CET] Action: Daemon404 wonders why mips is so interested in lavc
[17:02:51 CET] <atomnuker> they're too slow to run stuff there realtime?
[17:03:30 CET] <atomnuker> and they probably don't want to make their CPUs more performant
[17:07:15 CET] <ubitux> BBB: done
[17:11:53 CET] <nevcairiel> oh gcc tarballs showed up
[17:14:59 CET] <Daemon404> time to asess the damage
[17:15:00 CET] <Daemon404> ?
[17:15:14 CET] <nevcairiel> its a bugfix release, unlikely to break much
[17:15:36 CET] <nevcairiel> after i finally managed to figure out why my strip was randomly failing in mingw
[17:16:34 CET] <Daemon404> o ok
[17:17:39 CET] <Daemon404> hmm
[17:17:53 CET] Action: Daemon404 is sure this options thread will in no way be controversial or massive
[17:21:08 CET] <wm4> "This is a rather long explanation on ideas " yeah, no
[17:23:12 CET] <wm4> just going over it without reading everything, there are probably good ideas in it, but still, changing Everything rarely works out too well
[17:25:50 CET] <Daemon404> wm4, a skim makes it seemed a little to aimed at lavfi
[17:25:54 CET] <Daemon404> too*
[17:29:32 CET] <wm4> he wants to make the parsers recursive, instead of embedding quoted strings
[17:36:09 CET] <BBB> I think his complaint about avoptions being designed primarily for ffmpeg.exe is correct
[17:36:41 CET] <Daemon404> i dont have any complains about the api myself
[17:36:43 CET] <ubitux> our tools also have some local options btw
[17:36:44 CET] <Daemon404> i use it fairly extensively
[17:36:53 CET] <ubitux> while we could maybe use AVOptions :p
[17:37:18 CET] <ubitux> what i miss the most about AVOptions are lists of items
[17:37:23 CET] <ubitux> :(
[17:38:03 CET] <nevcairiel> using AVOptions from API is usually fine
[17:38:29 CET] <BBB> I think he doesnt mind the api
[17:38:35 CET] <BBB> I think he minds the string handling in the api
[17:38:56 CET] <BBB> the overabundance of strings, used for lists of devices when dealing with capturing video on mac/windows
[17:39:06 CET] <BBB> or used for all kind of other things
[17:39:17 CET] <BBB> a lot of that could be more clever
[17:39:35 CET] <nevcairiel> clever is the death of simple
[17:39:54 CET] <nevcairiel> and complex is the death of understandable
[17:41:22 CET] <fritsch> word for sunday ..
[17:41:34 CET] <wm4> the places where the API wants lists are disgusting
[17:41:48 CET] <wm4> (though I doubt it would improve with these avoption changes)
[17:59:14 CET] <cone-435> ffmpeg 03Paul B Mahol 07master:5d2cc00dd019: avfilter: add audio emphasis filter
[18:01:37 CET] <durandal_170> watch fate for me
[18:09:27 CET] <durandal_170> 30 band audio equalizer?
[19:09:43 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:a85b02dcf70f: avcodec/jpeg2000dec: Check bpno in decode_cblk()
[19:09:44 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:689a8674131c: avformat/msf: Check channels when reading the header
[19:09:45 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:ab7ff38052f1: avformat/flvdec: Fix left shift of 137 by 24 places cannot be represented in type int
[21:27:38 CET] <cone-435> ffmpeg 03Vicente Olivert Riera 07master:c8419035f48f: mips: remove --mips32r5 configure option
[21:27:39 CET] <cone-435> ffmpeg 03Vicente Olivert Riera 07master:1a538a149063: mips: add p5600 core optimizations
[22:30:24 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:4f2419888ba4: avcodec/pgssubdec: Fix left shift of 255 by 24 places cannot be represented in type int
[22:30:25 CET] <cone-435> ffmpeg 03Michael Niedermayer 07master:325b59368dae: swscale/utils: Fix for runtime error: left shift of negative value -1
[00:00:00 CET] --- Sat Dec  5 2015


More information about the Ffmpeg-devel-irc mailing list