[Ffmpeg-devel-irc] ffmpeg-devel.log.20190717

[burek]

[00:15:03 CEST] <juandl> Hello everyone, I'm trying to implement a filter following this guide mostly: https://fossies.org/linux/ffmpeg/doc/writing_filters.txt but when I try to run it using "-lavfi myfilter" I get "No such filter: myfilter" am I missing something?
[03:33:12 CEST] <Lynne> why is pwc, one of the big four auditors, the last strongholds of modern day slavery, interested in categorizing random codecs we've supported?
[03:34:14 CEST] <Lynne> fun thing to do: go to urbandictionary and search for definitions for each company's name
[09:14:27 CEST] <kurosu> Gah, I don't know what this UlfZibis tries/thinks he is achieving...
[10:14:12 CEST] <nevcairiel> feels like it'll get worse before it gets better with this guy
[10:27:32 CEST] <kurosu> nevcairiel: I'm bikeshedding, but I'd have closed this one with invalid - I wouldn't even want to acknowledge that kind of "bug" reports
[10:27:37 CEST] <kurosu> thanks for doing it anyway
[10:28:28 CEST] <nevcairiel> he send his FFSIGN change to the ML as well, in case you didnt see that
[10:42:00 CEST] <durandal_1707> don't be so negative, he tries to code something useful
[11:18:03 CEST] <kurosu> "Because anyone I ask including mathematicians, is the opinion that the sign of 0 is positive." <- that could be read as coming from a high horse
[11:18:10 CEST] <kurosu> hopefully durandal_1707 is right
[11:22:04 CEST] <thardin> I get the feeling sgn(0) == 0
[11:22:53 CEST] <jdarnley> But what if you're running on a sign-magnitude machine?  sign(-0) should be negative?
[11:23:09 CEST] <rcombs> I dunno what mathematicians he asked but conceptually zero is neither positive nor negative
[11:23:22 CEST] <thardin> wikipedia thinks sgn(0) == 0 at least
[11:23:55 CEST] <rcombs> (floating-point negative zero is a curiosity of the way IEEE floats are implemented, and while it can be meaningful for certain applications, it's not reflective of how real numbers work in a mathematical sense)
[11:24:41 CEST] <jdarnley> But what matters is what the hardware does, at least following from his MAX/MIN comments
[11:26:00 CEST] <rcombs> which comments?
[11:26:01 CEST] <j-b> How do you define the sign of 0?
[11:26:21 CEST] <jdarnley> by what the sign bit says lol
[11:26:47 CEST] <thardin> rcombs: you can certainly define a number system which has a negative zero
[11:26:55 CEST] <rcombs> thardin: sure but it's not reals
[11:27:00 CEST] <thardin> of course it wouldn't be \mathbb{R}
[11:27:01 CEST] <rcombs> nor integers
[11:27:04 CEST] <thardin> aye
[11:27:05 CEST] <jdarnley> this one https://trac.ffmpeg.org/ticket/8022
[11:27:12 CEST] <thardin> reminds me of how super-dense time works
[11:27:29 CEST] <thardin> which is a union of a real and an integer
[11:28:00 CEST] <rcombs> jdarnley: oof
[11:28:41 CEST] <rcombs> that kinda hurts to look at
[11:30:27 CEST] <rcombs> I'm kinda curious what the guy's trying to do
[11:31:23 CEST] <rcombs> seems like he's probably (relatively) new at software and trying to work his way around the codebase for some sort of project, and has an unfortunate tendency to report things he doesn't understand as bugs instead of asking why they are the way they are
[11:31:45 CEST] <rcombs> part of why that hurts is that it reminds me of me like 5 years ago
[11:31:56 CEST] <kurosu> rcombs: make ffmpeg palatable to mathematicians?
[11:32:09 CEST] <rcombs> (and still occasionally)
[11:32:32 CEST] <rcombs> kurosu: make C palatable to mathematicians
[11:32:57 CEST] <kurosu> Google tells me he might be/have been a Java programmer
[11:33:29 CEST] <rcombs> if you wanted a rigorously correct version of the sign function for reals and integers, you'd have something like ((x) > 0) ? 1 : ((x) < 0) ? -1 : 0
[11:33:43 CEST] <rcombs> (not sure if that's the right operator precedence, nested ternaries etc)
[11:34:11 CEST] <rcombs> but of course the point isn't to be rigorously correct
[11:34:45 CEST] <rcombs> even if that didn't add any bugs, you wouldn't want it, because it'd add an extra comparison that might survive the optimizer in some cases
[11:36:48 CEST] <thardin> you can just say "FFSIGN is FFSIGN"
[11:36:57 CEST] <thardin> as I see someone already did
[11:37:07 CEST] <kurosu> the psign* x86 instruction does vector*vector(FFSIGN(x)), so some people share our rationale
[11:39:13 CEST] <rcombs> this reminds me, are there any good time profiler navigation UI tools for linux
[11:39:52 CEST] <rcombs> I'm really used to Instruments.app on macOS and haven't found anything like that, perf seems a lot more difficult to interpret
[11:42:32 CEST] <rcombs> usually if I need to test perf for something on linux I just end up writing a unit test in checkasm, or something similar in whatever project it's in
[11:42:53 CEST] <rcombs> which is useful, and works well for a single tight loop, but not as well for telling you where all your cycles are going in a whole program
[11:43:54 CEST] <kurosu> rcombs: no idea, maybe https://developer.amd.com/amd-uprof/ (possibly proprietary, previous version codexl had at least a github repo)
[11:44:21 CEST] <rcombs> if on Intel I guess there's VTune
[11:44:32 CEST] <rcombs> I should try that out sometime
[11:45:41 CEST] <kurosu> VTune has a commercial license iirc, dunno how it might intersect with open source development
[11:46:50 CEST] <JEEB> https://software.intel.com/en-us/system-studio/choose-download
[11:46:58 CEST] <JEEB> > free commercial license
[11:47:18 CEST] <rcombs> afaik it's "free but it'll nag you about renewing the free license periodically"
[11:47:21 CEST] <JEEB> yea
[11:47:42 CEST] <rcombs> doesn't help much if you wanna do ARM stuff though
[11:47:56 CEST] <JEEB> yea
[11:49:13 CEST] <rcombs> VTune apparently provides ridiculous amounts of info on where your bottleneck in the CPU's pipeline is, so it's useful even when trying to optimize a single tight loop
[11:54:44 CEST] <jdarnley> rcombs: see if the perf website has any featured links for a front-end
[11:54:53 CEST] <jdarnley> they probably don't
[11:55:30 CEST] <durandal_1707> ncurses is best front-end
[12:07:07 CEST] <kierank> rcombs: afaiu it's just perf
[12:07:11 CEST] <kierank> with a gui
[12:07:16 CEST] <kierank> but might be useful
[13:34:36 CEST] <_bluez> Hey! Can anyone tell me if (in mov demuxer) I can get the location of mdat before it is actually parsed?
[13:35:48 CEST] <_bluez> I need to read some bytes from the beginning of mdat
[13:37:24 CEST] <JEEB> sounds like you need to export that somehow :P
[13:38:40 CEST] <_bluez> like while probing?
[13:42:13 CEST] <_bluez> export how?
[13:50:56 CEST] <JEEB> you might want to detail what exactly you want and why you cannot do that while parsing mdat?
[13:58:49 CEST] <_bluez> In heif files, the information about grid of tiles e.g. how many rows and cols is present at the top (for apple files at least) of mdat
[13:59:43 CEST] <_bluez> but now that I think of it maybe I wont need that information until later ':D
[14:00:52 CEST] <_bluez> I'll try to do that and ask again if not possible
[14:01:21 CEST] <_bluez> Sorry ':D
[14:07:19 CEST] <JEEB> _bluez: sounds like it's a specific box that you just have to parse from mdat or so?
[14:08:29 CEST] <JEEB> and you could export it either track-specific or global depending on if it's global metadata or track-specific metadata
[14:10:14 CEST] <_bluez> its not exactly a box but yeah kinda like that... its an item entry just like a picture or tile, except it has those information instead of image data
[14:12:47 CEST] <_bluez> but how do i export and from which part in the code?
[14:43:53 CEST] <durandal_1707> kurosu: it works with all samples i have
[14:44:20 CEST] <kurosu> durandal_1707: I suspect (for malformed bitstreams that some transcoding services will have to handle) you may need to flush one codec or the other
[14:44:55 CEST] <kurosu> or there's a special property of the coding structure that allows it for non-devious files, but is not generally guaranteed
[14:45:59 CEST] <kurosu> I'd probably alloc the decoder on demand, and ask_for_sample if it changes within a file/decoder instance/...
[14:47:27 CEST] <kurosu> that or we wait for Google's fuzzing reports to come in
[17:26:56 CEST] <mkver> durandal_1707: Did you test compilation with my matroska patch with MSVC?
[17:28:31 CEST] <durandal_1707> mkver: nope, let FATE handle it :)
[17:28:53 CEST] <durandal_1707> wait for some MSVC guru to appear
[21:48:19 CEST] <durandal_1707> jamrial: have you seen that MSVC fails on fate? do you have MSVC to test patch on ml?
[21:48:56 CEST] <mkver> I have already talked to James. He has no MSVC.
[21:49:09 CEST] <durandal_1707> crap
[22:22:06 CEST] <durandal_1707> ^ nevcairiel ?
[22:26:10 CEST] <nevcairiel> i can run it a bit later
[22:26:46 CEST] <mkver> Thanks, nevcairiel!
[22:57:01 CEST] <cone-588> ffmpeg 03James Almer 07master:056865a1d679: avcodec: clarify the allocation requirements for intra_matrix and inter_matrix fields
[23:20:05 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:a7e02cf3ad6f: avcodec/ivi: Ask for samples with odd tiles
[23:20:06 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:fff2bdc8b835: avformat/takdec: Free buffer on error pathes
[23:20:07 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:bb326795d135: avformat/takdec: Check for multiple streaminfo
[23:20:08 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:15008db0fac6: avcodec/parser: Check next index validity in ff_combine_frame()
[23:20:09 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:1707dbdf49b2: avcodec/dnxhd_parser: remove unneeded code
[23:20:10 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:827faa18cfd0: avcodec/dnxhd_parser: Optimize insufficient buf size case
[23:20:11 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:f6d1b18b3d58: avformat/rawdec: Make the raw packet size configurable
[23:20:12 CEST] <cone-588> ffmpeg 03Michael Niedermayer 07master:2d900d8fe0aa: avcodec/dnxhd_parser: Fix parser when input does not have nicely sized packets
[00:00:00 CEST] --- Thu Jul 18 2019