[Ffmpeg-devel-irc] ffmpeg-devel.log.20171218

burek burek021 at gmail.com
Tue Dec 19 03:05:02 EET 2017


[00:03:41 CET] <ubitux> "deprived of kids" sounds weird
[00:04:05 CET] <Chloe> ubitux: i think durandal_1707 was asking 'were the kids or you deprived of chocolate'
[00:04:54 CET] <ubitux> i know, it just sounded creepy so i wanted to point it out :) 
[00:05:17 CET] <ubitux> (because that's funny. i think. maybe.)
[00:09:56 CET] <TimothyGu> how do you all get to travel so much
[01:35:45 CET] <SortaCore> TimothyGu: they're evading Interpol
[01:36:24 CET] <SortaCore> anyone know why frame rate/duration is butchered in RTSP H264 copied to MP4 H264?
[01:36:47 CET] <SortaCore> plays fine in vlc, but codec info is all wrong
[02:07:15 CET] <cone-644> ffmpeg 03Aurelien Jacobs 07master:3f8874406773: aptx: add codec cap SMALL_LAST_FRAME and INIT_THREADSAFE as appropriate
[02:23:08 CET] <SortaCore> *raises hands in praise to BBB*
[02:23:18 CET] <BBB> Ive seen that crap in several channels now :-/
[02:23:33 CET] <Fenrirthviti> Had to leave 6 channels I was in
[02:23:37 CET] <Fenrirthviti> due to inactive ops. Very annoying.
[02:23:43 CET] <rcombs> set mode +q $~a
[02:23:49 CET] <Fenrirthviti> Seen spam for that guy for months, actually.
[02:23:51 CET] <DHE> I'm not a fan of it, but there are modes you can set on the channel that only allows registered users to join the channel or to talk (two different modes)000000
[02:23:56 CET] <iive> freenode stuff is flooder with complaints too
[02:24:03 CET] <rcombs> DHE: the latter is +q $~a
[02:24:09 CET] <Fenrirthviti> He seems to have pissed off a bunch of people and now they're trying to slander him by making it seem like this is him spamming 
[02:24:11 CET] <rcombs> I just removed BEL support from my client
[02:25:10 CET] <SortaCore> a hop into Inspect Elements fixed that mess
[02:25:17 CET] <SortaCore> anyone have a tip for my framerate/duration problem?
[02:25:35 CET] <BBB> no, sorry :/
[02:26:23 CET] <Shiz> almost all freenode chans are being spammed
[02:26:27 CET] <jamrial> BBB: nice, that was quick
[02:26:35 CET] <BBB> luck
[02:26:37 CET] <BBB> :-p
[02:26:51 CET] <BBB> I happened to be around
[02:28:08 CET] <jamrial> hehe
[09:53:52 CET] <ubitux> $
[11:12:35 CET] <atomnuker> my main laptop's not booting after leaving it off for a week because of some debian/gdm issue
[11:14:09 CET] <atomnuker> tempted to install gentoo but I require nothing but total systemd integration for everything and gentoo doesn't because ignorant bastards want openrc to work
[11:14:31 CET] <SortaCore> windows 2000 best OS
[11:18:41 CET] <durandal_1707> atomnuker: you nuked laptop, awesome
[11:18:58 CET] <atomnuker> I didn't
[11:19:11 CET] <atomnuker> it just stopped booting by itself
[11:21:29 CET] <SortaCore> NSA kill switch activated
[11:22:21 CET] <atomnuker> systemd haters are the equivalent of conservative bastards
[11:22:23 CET] <atomnuker> "I don't have anything against systemd. What I dislike is how systemd is being forcibly implemented. If systemd was that beneficial to me, I would be voluntarily embracing it as I have done so many other changes, but it is not that beneficial and so I will not embrace it, especially not when it is forced on me."
[11:22:35 CET] <durandal_1707> atomnuker: do you have all your work on cloud?
[11:22:59 CET] <atomnuker> most, I have a 6tb hdd and an 8tb one
[11:24:09 CET] <atomnuker> look at that quote, its absolutely disgusting and wrong
[11:24:44 CET] <durandal_1707> you cited whom ?
[11:25:31 CET] <atomnuker> https://github.com/dantrell/gentoo-project-gnome-without-systemd/blob/master/POLITICS.md#rationale
[11:26:03 CET] <iive> well, gentoo is one of the few distributions that actually let you choose if you want to use systemd
[11:29:56 CET] <atomnuker> by default its openrc so everything's meant to work the old way
[11:31:38 CET] <atomnuker> "For me, I have a mask file that includes every major piece of code poettering has touched. I will NEVER allow systemd, pulse or avahi on any of my Gentoo systems. gconf and dconf have also been masked as they create a lot of "systemd compatible" crap, polluting my system for no good reason."
[11:31:52 CET] <atomnuker> absolutely delicious
[11:36:31 CET] <iive> well, unfortunately, systemd is still incomplete. It lacks a text editor.
[11:38:43 CET] <Chloe> atomnuker: doesn't gentoo support systemd if you want it
[11:40:14 CET] <atomnuker> it does, but its bound to be badly integrated in a non-standard way with the rest of the distributions
[11:42:33 CET] <iive> atomnuker: what?!
[11:42:39 CET] <Chloe> atomnuker: aren't overlays pretty standard and integrated, especially official ones?
[11:43:05 CET] <Chloe> though I'm not experienced with gentoo so I don't really know
[11:43:36 CET] <michaelni> atomnuker, about "laptop's not booting", probably its a different issue but last time i saw one/mine not booting it was because ubuntu put UEFI files in a different directory than what the notebook bios looked into resulting in the thing acting as if there was no OS installed
[11:44:25 CET] <iive> i've also heard that systemd tries to mount everything from fstab and panics if it cannot.
[11:44:51 CET] <atomnuker> no, the issue here is that I get "failed to start user UID=111" on the boot log and trying to use a different VT just makes the screen flash back to the login screen if I put my login info
[11:45:01 CET] <atomnuker> ssh just immediately exits without running anything
[11:45:38 CET] <atomnuker> thought its because of gdm but using a rescue shell to remove the executable doesn't fit anything
[11:45:39 CET] <iive> What?!!
[11:45:52 CET] <atomnuker> yeah, weird
[11:46:08 CET] <bencoh> what if you disable ssh pty allocation and try running a command there?
[11:46:50 CET] <iive> gdm is for X, can't you boot to text mode or something?
[11:47:40 CET] <iive> gdm - gnome display manager.
[11:49:51 CET] <Chloe> nice, having a comment after an extern breaks configure
[11:50:32 CET] <Chloe> in the lists anyway
[12:00:58 CET] <atomnuker> iive: it never goes to gdm, its stuck in text mote
[12:01:00 CET] <atomnuker> *mode
[12:01:19 CET] <atomnuker> and also like I said if you try to login from another VT you don't get a shell
[12:02:42 CET] <jkqxz> init=/bin/bash
[12:04:13 CET] <jkqxz> (Sounds like a systemd problem...)
[12:04:35 CET] Action: iive hides
[12:15:11 CET] <atomnuker> init=/bin/bash gets stuck before the kernel even boots up
[12:15:47 CET] <iive> you add it to the kernel line, if that wasn't clear
[12:16:06 CET] <iive> also, check if you have bash executable at that place
[12:18:44 CET] <jkqxz> If you're inside the initrd rather than the initial root being a real disk then it might not have bash.
[12:19:09 CET] <iive> it might look at initrd filesystem...  initrd systems prefer bisybox that emulates all commands, including sh and bash
[12:19:16 CET] <iive> busybox
[12:20:15 CET] <iive> the initrd is specified on the kernel line, so you can remove it when you add the init.
[12:20:31 CET] <iive> i assume grub that allows editing.
[12:21:26 CET] <atomnuker> jkqxz: yeah, I should try /bin/sh
[12:23:27 CET] <stevenliu> fdisk -l lookup the disk partitions
[12:56:11 CET] <iive> atomnuker: how is it going?
[13:26:20 CET] <atomnuker> dunno, I wanna reinstall anyway because debian's kde hasn't been updated in months and because I want to use f2fs
[13:26:46 CET] <BtbN> Why would you want to use f2fs? Are you installing on an SD Card?
[13:26:50 CET] <Chloe> atomnuker: why do you want to use f2fs?
[13:27:29 CET] <atomnuker> its faster than btrfs for nvmes, its not only for sd cards, its for anything flash based
[13:27:54 CET] <BtbN> The primary point of f2fs is to look like fat32, so the crappy wear-leveling of cheap flash can handle it
[13:28:02 CET] <BtbN> On propper SSDs ext4 is preferable
[13:31:12 CET] <atomnuker> ext4 doesn't even have any flash optimizations, its stale, "stable" and old by now
[13:31:29 CET] <atomnuker> https://www.phoronix.com/scan.php?page=article&item=linux-nvme-raid&num=3 does show f2fs is faster than btrfs
[13:31:40 CET] <BtbN> it's not hard to be faster than btrfs
[13:32:20 CET] <atomnuker> its faster than ext4 too
[13:32:42 CET] <BtbN> I see ext4 on top there.
[13:33:09 CET] <BtbN> or so slightly behind that it's probably just noise
[13:34:42 CET] <atomnuker> on one benchmark, yeah
[13:35:01 CET] <atomnuker> all others have f2fs being faster
[13:35:59 CET] <Chloe> 'stable' sounds like a better quality than 'fast but maybe not as stable' for a filesystem
[13:36:44 CET] <Chloe> Guess you just have to weigh up speed vs reliability
[13:38:21 CET] <BtbN> I remember there being some fuzzing test that completely wrecked f2fs with almost every attempt, and ext4 was not really affected at all
[13:40:16 CET] <iive> atomnuker: if you reinstall, before you find and fix the problem, I'm going to use this case as example how crappy systemd is
[13:40:32 CET] <iive> because it makes impossible to debug and fix your system :P
[13:40:47 CET] <iive> if anything goes wrong with it.
[13:41:25 CET] <ubitux> the ffmpeg community hasn't enough of its own drama it needs to brings the systemd one into it as well
[13:44:05 CET] <iive> well, if you know how to fix atomnuker's issue, go ahead.
[13:45:02 CET] <ubitux> it doesn't sound related to multimedia in a meaningful way
[13:45:22 CET] <iive> well, if you cannot log into your development machine...
[13:45:53 CET] <ubitux> yeah well, i don't expect help from ffmpeg-devel if i loose my home keys
[13:47:06 CET] <atomnuker> iive: you're just bringing in systemd because its init related when I don't think it has any connection at all!
[13:47:14 CET] <atomnuker> you're a systemd hater too, aren't you!
[13:50:49 CET] <nevcairiel> every sane person should be
[13:56:40 CET] <atomnuker> right, nevcairiel's correct, I should be using windows because it has no systemd
[13:57:38 CET] <BBB> every knows mac > windows
[13:57:42 CET] <BBB> <3
[13:59:21 CET] <Chloe> atomnuker: you could even try run kde on linux on windows
[14:00:15 CET] <cone-347> ffmpeg 03Ronald S. Bultje 07master:df3222d4bb18: libvmaf: exit gracefully if the library fails.
[14:11:24 CET] <iive> atomnuker: well, what is the problem then?
[14:17:09 CET] <iive> while the problem might not be in systemd itself, it kind of makes it from hard to impossible to locate the real problem.
[14:17:27 CET] <iive> and you are above avarage linux user, admin and programmer.
[14:23:31 CET] <BtbN> imo it got way easier to locate and debug problems with systemd and journald. You just have to learn it once. But lets not discuss that here please.
[14:27:24 CET] <iive> BtbN: give a hand to atomnuker then. :D
[14:52:19 CET] <cone-347> ffmpeg 03Mateusz 07master:6260ab60a80f: avfilter/vf_overlay: fix packed_rgb case
[15:25:50 CET] <cone-347> ffmpeg 03Paul B Mahol 07master:5f2c4d044f76: doc/filters: update biquad filters with commands section
[16:51:27 CET] <thealexbarney> I released the source to an ATRAC9 decoder last week, so I figured I'd drop in and see how adding that into FFmpeg is going. From what I gather, atomnuker will be handling that?
[16:51:59 CET] <thealexbarney> I posted a more standalone decoder if that's of any use: https://github.com/Thealexbarney/LibAtrac9
[17:14:19 CET] <atomnuker> thealexbarney: will do it by the re'd spec you released + whatever I don't understand from the decoder code you released
[17:35:31 CET] <ubitux> why decode_{audio,video} set decode_failed to 1 on EOF?
[17:35:38 CET] <ubitux> is this on purpose? (fftools/ffmpeg.c)
[17:37:19 CET] <wm4> who knows
[17:38:14 CET] <wm4> doesn't seem to matter
[17:38:34 CET] <ubitux> doesn't end up printing an EOF "error"?
[17:38:56 CET] <ubitux> ah, no right there is a check before that
[17:39:04 CET] <jdarnley> Has Agner Fog done his instruction timing on an AVX-512 Skylake/Cannonlake or not?
[17:39:15 CET] <jamrial> on verbose/debug level it does print a "end of file reached" message or so, i recall
[17:39:37 CET] <jdarnley> I was looking at his tables but it only seems to list Knights Landing
[17:39:57 CET] <jdarnley> Gramner: do you know anything about  that ^ ?
[17:44:08 CET] <thealexbarney> atomnuker, there are a few things I haven't gotten around to adding to that spec. I'm also available to clarify anything, although I probably won't be on IRC too much. Most of my knowledge of audio coding techniques comes from RE'ing the HCA and ATRAC9 codecs over the past ~half-year with no outside help, so I wouldn't be surprised if there were some oddities in the source because of that.
[17:53:48 CET] <ubitux> filter-overlay_rgb(a)_rgba seems to be failing randomly
[17:53:50 CET] <ubitux> is this known?
[17:54:31 CET] <ubitux> or maybe that was fixed recently
[17:54:39 CET] <ubitux> (looking at fate)
[18:03:20 CET] <ubitux> k it's fixed
[18:28:33 CET] <jdarnley> Gramner: on a related note do you know why nasm complains about "pcmpgtq m0, m1" under "INIT_YMM avx512"?
[18:29:49 CET] <Gramner> because that instruction doesn't have an evex version
[18:29:57 CET] <jdarnley> oh
[18:30:16 CET] Action: jdarnley opens the manual
[18:31:36 CET] <jdarnley> oh right, it can only put the results in a k register
[18:34:36 CET] <jdarnley> Sorry to bug you with that, I wanted to replace this emulation with the proper instruction anyway.
[18:42:32 CET] <Gramner> https://software.intel.com/sites/default/files/managed/ad/dc/Intel-Xeon-Scalable-Processor-throughput-latency.pdf for SKL-X instruction latencies
[18:43:30 CET] <Gramner> and yes, some instructions are slow as hell. should improve with newer µarchs
[18:43:46 CET] <jdarnley> Thank you very much
[18:45:28 CET] <jdarnley> :( no index for it
[18:47:34 CET] <jdarnley> Wow, what a lot of instructions that begin with v! /s
[18:51:35 CET] <Gramner> vpconflictd zmm is one of the bestr ones at a blazing 74/17.5
[18:51:54 CET] <jdarnley> omfg
[18:52:18 CET] <Gramner> and I still made one DSP function in x264 faster using it
[18:52:21 CET] <jdarnley> that must be basically a scalar operation under the hood
[18:52:49 CET] <Gramner> its microcoded. 3/1 on xeon phi though
[18:53:07 CET] <jdarnley> VPCONFLICTD/QDetect Conflicts Within a Vector of Packed Dword/Qword Values into Dense Memory/ Register
[18:53:48 CET] <jdarnley> Test each dword/qword element of the source operand (the second operand) for equality with all other elements in the source operand closer to the least significant element. Each elements comparison results form a bit vector, which is then zero extended and written to the destination according to the writemask.
[18:56:56 CET] <jdarnley> A question: does "vpmaxsq ymm1{k1}, ymm2, [rdi]" hold one of the values for "max(a,b)" in the gp register?
[18:57:21 CET] <Gramner> no, it's a memory arg
[18:57:32 CET] <jdarnley> okay
[18:58:23 CET] <jdarnley> What does "m64bcst" mean in the instruction reference for pmaxsq?
[18:59:53 CET] <atomnuker> Gramner: what's vpconflictd useful for?
[19:00:08 CET] <Gramner> finding duplicate values
[19:00:43 CET] <jdarnley> "An operand that can be a ZMM register, a 512-bit memory location or a 512-bit vector loaded from a 64-bit memory location."  Is that a broadcasted value then?
[19:01:35 CET] <Gramner> probably
[19:04:19 CET] <atomnuker> Gramner: where'd you want to find duplicate values? quantization?
[19:05:54 CET] <Gramner> I use it in mbtree cost propagation to avoid race conditions by detecting duplicate mb references
[19:06:18 CET] <Gramner> or well, race condition is probably not the correct term here, but similar
[19:53:52 CET] <thebombzen> this error should not happen: [null @ 0x55dd1c69c440] Application provided invalid, non monotonically increasing dts
[19:54:06 CET] <thebombzen> the NULL muxer should not care if the application provided invalid DTS or PTS or whatever
[19:54:10 CET] <thebombzen> seeing as it discards all output
[19:54:21 CET] <thebombzen> should I open an issue on TRAC about this or is it known?
[19:58:27 CET] <cone-347> ffmpeg 03Michael Niedermayer 07master:74f408cc8e9a: avformat/mov: Fix the the typo
[20:06:04 CET] <jkqxz> thebombzen:  That message comes from the lavf core, not from the null muxer.  It's always invalid, regardless of whether you are throwing away the output or not, so seems fair enough.
[20:11:37 CET] <durandal_1707> what you guys think about xvc codec?
[20:14:04 CET] <thebombzen> I guess what I care more about is that it's repeated over and over again and ffmpeg's automatic error message deduplication  doesn't catch it
[20:14:08 CET] <thebombzen> because the timestamps are different
[20:14:46 CET] <thebombzen> perhaps the automatic error message deduplication should be made a bit more robust so DTS warnings and errors are not as spammy. 
[22:11:41 CET] <atomnuker> jamrial: is there an aarch64 equivalent of something like http://www.felixcloutier.com/x86/ ?
[22:12:21 CET] <atomnuker> (or was it ubitux who did aarch64 simd)
[22:12:26 CET] <jamrial> it was him
[22:12:47 CET] <durandal_1707> nooo, write filter, leave arch optimizations
[22:13:11 CET] <atomnuker> im just curious to see what it has in terms of instructions
[22:13:48 CET] <durandal_1707> nothing useful, its sloow, i bet
[22:15:46 CET] <atomnuker> I already know its slow and worse in every way than x86 for manual asm
[22:16:30 CET] <atomnuker> (I'd rather have tens of bytes for a single instruction than lose x86 addressing)
[22:25:47 CET] <durandal_1707> atomnuker: is there big book like one for x86?
[22:27:05 CET] <atomnuker> that's what I'd like to know, I wanna write some aarch64 simd to prove its worse and not fun
[22:29:28 CET] <iive> ubitux was doing some aarch64
[22:31:10 CET] <atomnuker> https://en.wikipedia.org/wiki/Replay_system
[22:31:22 CET] <atomnuker> oh wow, " When this happens, the replay system signals the scheduler to stop, then repeatedly executes the failed string of dependent operations until they have completed successfully."
[22:31:50 CET] <atomnuker> never knew pentium 4 had such a huge hack
[22:37:25 CET] <wm4> I thought such mechanisms were common in modern CPUs
[22:46:05 CET] <TD-Linux> durandal_1707, it's a nice idea, though I see no reason to use it over AV1 (but I'm super biased)
[22:56:06 CET] <thardin> 31 stages? that's nothing compared to modern GPUs
[22:58:07 CET] Action: kierank wonders what michaelni is talking about on the ml
[22:58:27 CET] <kierank> on what planet do you need to stream ffmpeg structs over a network
[22:58:50 CET] <thardin> isn't that what nut is?
[22:59:02 CET] <BtbN> it's what ffm is iirc
[22:59:05 CET] <nevcairiel> no, thats what ffm iis, basically
[22:59:17 CET] <kierank> point still stands
[22:59:54 CET] <thardin> I guess it's useful when building pipelines in some cases
[23:00:25 CET] <thardin> can't for the life of me think of a case however
[23:01:38 CET] <kierank> conflates the idea of software structures and network structures
[23:01:40 CET] <philipl> The only thing better than C structs on the wire are internal structs on the wire.
[23:01:45 CET] <thardin> insert objection to libavfilter here
[23:02:01 CET] <thardin> it being graph based that is
[23:02:42 CET] <BtbN> wm4, adding a reserved field? Isn't adding more fields always possible?
[23:02:59 CET] <jamrial> BtbN: sizeof avpacket is part of the abi
[23:03:07 CET] <BtbN> oh wow
[23:03:17 CET] <BtbN> yeah, add something now then
[23:03:35 CET] <BtbN> Maybe even a bit more than just one field
[23:03:44 CET] <wm4> it could end up unused, or so
[23:03:45 CET] <philipl> Round up to the nearest gigabyte, just to be safe.
[23:04:11 CET] <jamrial> wm4: worst case scenario, it's removed during the next bump
[23:04:26 CET] <wm4> (I'm no sure how we'd achieve compatibility anyway with the old API)
[23:04:52 CET] <Chloe> jamrial: I think you mean the bump after the next next bump at least
[23:05:57 CET] <jamrial> no. a reserved field that never ends up being used can be removed without issue at the first chance
[23:06:49 CET] <wm4> anyway, there's the idea floating around that sizeof(AVPacket) should not be part of the ABI anymore, but I think it's too early for that
[00:00:00 CET] --- Tue Dec 19 2017


More information about the Ffmpeg-devel-irc mailing list