[Ffmpeg-devel-irc] ffmpeg.log.20160811

[burek]

[00:43:43 CEST] <fa0> I'm trying to compile ffmpeg against libdc1394-2.2.4, but any version starting at 2.1.3 complains it can't find libdc1394-2.pc in pkgconfig, does anyone know about this, a fix?
[00:54:47 CEST] <fa0> At the terminal all it says is; ERROR: libdc1394-2 not found using pkg-config
[00:55:49 CEST] <fa0> But when I look at /usr/lib64/pkgconfig I see; libdc1394-2.pc
[01:54:17 CEST] <Spring> is it correct that frames per second have nothing to do with the filesize, but only time?
[01:54:38 CEST] <Spring> time x bitrate
[02:16:46 CEST] <DHE> Spring: most [common] compression schemes have flexible quality<->size tradeoffs. so you can have a 1080p image at 1 megabit if you're okay with it looking like a colour smear
[02:17:23 CEST] <DHE> framerate doesn't enter into it except as information to the codec so it can calculate how to allocate bytes to each frame
[02:54:01 CEST] <DHE> Spring: understand?
[03:47:13 CEST] <starstuff> Hi, I have a video that is in a quicktime container and I want to put it into a different container that is not quicktime.
[03:47:23 CEST] <starstuff> what's the best open source container format right now?
[03:49:57 CEST] <starstuff> what's the difference between webM and MPEG H.264?
[03:55:36 CEST] <starstuff> I want to use whatever is the open-source video format that is the most widely accepted on the web today. Can anyone help please?>
[04:02:40 CEST] <Spring> DHE, kind of. With more frames would that help with encoding or just mean that the 'extra' frames become more 'smeared' for the same bitrate? Just wondering if more fps would be better with higher quality settings vs lower fps.
[04:06:34 CEST] <Spring> in my case I doubt it matters much since there is little movement but just wanted to be sure so I can explain to others
[04:07:49 CEST] <kepstin> given the same bitrate, having a higher framerate will always result in lower quality video. The difference depends on the amount of motion and various other factors.
[04:08:11 CEST] <kepstin> Even if the frames are identical, it still has to signal that "there is a frame here that's the same as the last one", which takes some bits :)
[04:13:07 CEST] <starstuff> Hi, how do I convert a video I have?  It's in the quicktime container form and I don't want that.  What's the best open source format right now that is most widely used and accepted?
[04:14:18 CEST] <Spring> kepstin, so what I read about framerate having no effect isn't correct then. Good to know.
[04:14:44 CEST] <kepstin> Spring: it'll still result in the same file size of course
[04:14:51 CEST] <kepstin> Spring: since file size is just bitrate * time
[04:15:04 CEST] <Spring> right
[04:15:47 CEST] <kepstin> starstuff: not sure what you mean by "open source"; there's open-source software that encode and decode quicktime files and most codecs in them just fine
[04:15:52 CEST] <kepstin> e.g. ffmpeg ;)
[04:17:15 CEST] <starstuff> well, I mean that I just want to know what is the video/audio formats that have legal problems or whatever.  Isn't there an open source video/audio format now I heard?
[04:18:18 CEST] <kepstin> oh, you mean formats that (as far as anyone knows) don't have patent issues?
[04:18:25 CEST] <starstuff> sure, yes.
[04:18:29 CEST] <Spring> VPx is open source and decent, and the WebM container supports open source audio formats like opus and ogg
[04:18:35 CEST] <kepstin> then probably webm container with either vp8 or vp9 video and vorbis or opus audio
[04:18:49 CEST] <kepstin> vorbis/vp8 is more compatible, opus/vp9 is newer and better
[04:18:50 CEST] <starstuff> okay I heard about webm container with vp8 video
[04:19:11 CEST] <starstuff> so how do I convert this quicktime file to webm with vp8 video?
[04:19:21 CEST] <kepstin> ffmpeg -i file.mov file.webm
[04:19:36 CEST] <kepstin> (but that'll use some default settings with medium-poor quality, probably)
[04:19:40 CEST] <starstuff> I can't seem to find ffmpeg in my distribution's repository.
[04:19:57 CEST] <starstuff> I have Debian Jessie.
[04:20:06 CEST] <kepstin> debian jessie has ffmpeg...
[04:20:07 CEST] <starstuff> I could have sworn I used to be able to install ffmpeg but it's not in here anymore
[04:20:34 CEST] <starstuff> # aptitude install ffmpeg No candidate version found for ffmpeg     No candidate version found for ffmpeg No packages will be installed, upgraded, or removed. 0 packages upgraded, 0 newly installed, 0 to remove and 0 not upgraded. Need to get 0 B of archives. After unpacking 0 B will be used.
[04:20:36 CEST] <kepstin> oh, it's in backports
[04:20:49 CEST] <kepstin> i think
[04:21:18 CEST] <kepstin> should be `apt-get -t jessie-backports install ffmpeg`
[04:21:40 CEST] <starstuff> E: The value 'jessie-backports' is invalid for APT::Default-Release as such a release is not available in the sources
[04:21:56 CEST] <kepstin> ok, follow the instructions on https://backports.debian.org/Instructions/ :)
[04:23:27 CEST] <starstuff> kepstin: so with that command you gave, how do I modify it so it uses highest quality possible?  I don't want any added compression.  I'm looking for the best conversion quality.
[04:24:01 CEST] <kepstin> well, "best" is hard to say. Any time you do a lossy transcode, there will be some quality loss
[04:24:24 CEST] <kepstin> if you make the file bigger, less quality will be lost. it's a tradeoff you have to decide
[04:25:01 CEST] <starstuff> I don't mind if it's big.  Right now the file says it's in a quicktime container and the video codec is listed as H.264.  I don't know what that is.
[04:25:15 CEST] <starstuff> but I'm pretty sure I want webm/vp8
[04:25:31 CEST] <starstuff> for some reason when I right-click the file in gnome it shows there's nothing for the audio codec.
[04:25:38 CEST] <kepstin> but, assuming you want to go with vp9 and opus, a good start would be "ffmpeg -i file.mov -c:a opus -c:v libvpx-vp9 -crf 30 -b:a 128k output.webm", then adjust the crf value smaller for higher quality, bigger for larger file.
[04:26:06 CEST] <starstuff> is vp8 more compatible right now? I want other people on the web to be able to see it
[04:26:07 CEST] <Spring> with vp9 you need to add -b:v 0 to get actual CRF mode
[04:26:22 CEST] <kepstin> hmm, it doesn't do that by default?
[04:26:33 CEST] <Spring> I'm just going by the docs
[04:26:50 CEST] <kepstin> and with vp8, crf mode is more complicated - you have to set a bitrate. Most people just use -b:v "some really big value", e.g. 50M or whatever
[04:26:57 CEST] <Spring> https://trac.ffmpeg.org/wiki/Encode/VP9
[04:27:04 CEST] <starstuff> it's weird that it says its a quicktime container but the file extension is .mp4.  Is that messed up?
[04:27:15 CEST] <JEEB> i think the bit rate is only needed for vp8
[04:27:26 CEST] <JEEB> vp9 has crf only crf
[04:27:57 CEST] <kepstin> JEEB: the question is, do you have to explicitly state '-b:v 0' to get vp9 crf mode to not use a max bitrate :)
[04:28:07 CEST] <Spring> "To trigger this mode, you must use a combination of -crf and -b:v 0. -b:v MUST be 0."
[04:28:41 CEST] <kepstin> starstuff: the iso mp4 container is based on the quicktime container, ffmpeg has shared code for them
[04:28:44 CEST] <starstuff> so, back to my question, is vp8 more compatible across the web? Will less people be able to view the video if it is in vp9?
[04:29:00 CEST] <Spring> whether this has changed and the docs haven't been updated I'm not sure
[04:29:17 CEST] <kepstin> starstuff: assuming people don't disable auto-update on their browser, it's probably about the same now. Latest firefox and chrome should both do vp9 fine i think
[04:29:42 CEST] <Spring> vp9 has been fine for years in both Firefox and Chrome for me
[04:29:53 CEST] <starstuff> okay then what would I do to convert this quicktime container video with H.264 codec to a webm container file with vp9 codec?
[04:30:08 CEST] <starstuff> with minimal quality loss!
[04:30:18 CEST] <kepstin> starstuff: I gave you an example command, just add the '-b:v 0' option as well
[04:30:41 CEST] <kepstin> and you will have to tweak the -crf value yourself to balance between the amount of quality loss and filesize you want
[04:31:10 CEST] <starstuff> ffmpeg -i file.mov -c:a opus -c:v libvpx-vp9 -crf 1000000000 -b:v 0 -b:a 128k output.webm
[04:31:12 CEST] <starstuff> would that work?
[04:31:28 CEST] <kepstin> no, -crf value is (iirc) a range between around 2 and 63 for vp9
[04:31:37 CEST] <kepstin> where smaller numbers are higher quality = larger files
[04:31:38 CEST] <starstuff> is 63 the best?
[04:31:41 CEST] <starstuff> oh ok
[04:31:48 CEST] <Spring> no, lower is better
[04:31:54 CEST] <starstuff> ffmpeg -i file.mov -c:a opus -c:v libvpx-vp9 -crf 2 -b:v 0 -b:a 128k output.webm
[04:31:58 CEST] <starstuff> like that?  ^
[04:32:12 CEST] <kepstin> 2 is almost certainly going to give you a file that'll look pretty, but be "way too big" :)
[04:32:15 CEST] <Spring> you'd have a crazy high filesize but you're getting the idea
[04:32:21 CEST] <kepstin> start somewhere between 15 and 30
[04:33:16 CEST] <starstuff> okay I'm doing -crf 2 just because I want to see the resultant file size :)
[04:33:43 CEST] <htoetl> is there a way to overlay and loop the same text over video that would fade in / out all the time? i used to do it in many steps - first creating a video with a single loop of fading in/out, then make a long looped video and then using it as a source for overlay
[04:34:05 CEST] <kepstin> starstuff: heh, in the spirit of the scientific method ;)
[04:34:14 CEST] <htoetl> is there a way to add this without intermediate steps and separate video?
[04:35:21 CEST] <starstuff> mmm this might take some time I think.
[04:36:03 CEST] <kepstin> yeah. vp9 is not fast
[04:36:52 CEST] <Spring> a 20 second 1440p clip takes me 5 minutes to encode
[04:36:59 CEST] <Spring> at crf 30-ish
[04:37:24 CEST] <Spring> JEEB, you have some experience with vp9, from the fb2k logs I remember you said it doesn't have great rate control, right?
[04:37:58 CEST] <JEEB> yeah but qith crf you're not specifying a specific one anyways (avg bit rate)
[04:38:38 CEST] <JEEB> crf is usually variable quantizer that tries to be closer to "cobstant quality"
[04:39:06 CEST] <JEEB> and yes writing with one hand at a station inäs not optimal :)
[04:40:48 CEST] <JEEB> if you are encoding for the web or otherwise rate limited stuff you limit stuff with -maxrate and -bufsize
[04:41:05 CEST] <JEEB> think of it as maxrate over bufsize
[04:41:21 CEST] <JEEB> (and hope libvpx doesn't screw up)
[04:52:55 CEST] <Alca_> Hi Guys
[04:53:11 CEST] <Alca_> is there anyone alive here :D
[04:53:28 CEST] <Alca_> i need some help with FFmpeg
[04:56:55 CEST] <Alca_> Hello
[05:11:22 CEST] <Spring> Alca_, I'd just post the question, then idle to see if anyone can help
[05:13:57 CEST] <deweydb> If you were going to buy a video card specifically for FFMPEG hardware acceleration in linux. what card would you pick?
[05:14:15 CEST] <deweydb> this card's one and only task will be this. nothing else.
[05:32:47 CEST] <kepstin> deweydb: in most cases, I wouldn't bother, and would spend more money on a cpu instead :)
[05:33:27 CEST] <deweydb> hmm
[05:35:07 CEST] <kepstin> unless you're doing realtime video encoding while you're cpu's busy with something else, a software encoder will probably give better results.
[05:36:45 CEST] <deweydb> i'm not doing much, but it seems to take quite a long time.  I'm taking a video clip, stripping the audio, adding water mark, adding an intro graphic, adding some text intro/outro segments, then layering a song ontop.
[05:37:03 CEST] <deweydb> the videos are about 2 mins long in total.
[05:37:14 CEST] <deweydb> it takes my i7 2600k about 15-20 mins per video
[05:37:30 CEST] <deweydb> granted thats a pretty old CPU
[05:43:38 CEST] <Spring> does ffmpeg have some option to display the filesize of an audio/video stream without demuxing them?
[05:43:52 CEST] <Spring> mediainfo doesn't display it
[05:59:58 CEST] <kepstin> Spring: that's not really information stored in a header anywhere, you have to demux the file and count the sizes of the streams separately to get the info
[06:00:19 CEST] <kepstin> i dunno, ffprobe might be able to do it with the '-count_frames' option? have to check
[06:06:05 CEST] <Spring> that's fine, I'll probably just demux it then
[06:17:30 CEST] <htoetl> how to create a video with a transparent background?
[06:49:42 CEST] <antiPoP> Hi I'm reading http://ksloan.net/watermarking-videos-from-the-command-line-using-ffmpeg-filters/ , and I was wondering how to make an animated watermark, but only for a time fraction (eg from 00:50 to 01:00)
[06:58:18 CEST] <Mysoft> does anyone know where i could find win32 static libraries (libavcodec.a , libavformat.a, libavutil.a, libswscale.a) ?
[06:58:37 CEST] <Ashish> i wants to control cpu utilization when run ffmpeg commands. how can i do it ? we have centos server with 8 CORE cpu. each one has Intel(R) Xeon(R) CPU E3-1265L V2 @ 2.50GHz processer.
[06:58:47 CEST] <Mysoft> so that i can create a possible static .exe with smaller size (as it will only use the required parts)
[06:59:44 CEST] <Mysoft> i wanted to avoid compiling myself for now, based on how bad luck i have compiling C stuff
[07:05:54 CEST] <Ashish> can anyone help us here?
[07:06:41 CEST] <tdr> Ashish, what kind of things are you trying to do, centos supports cgroups easy enough
[07:08:13 CEST] <Ashish> do you mean with ffmpeg ?
[07:08:23 CEST] <tdr> Ashish, to contorl
[07:08:25 CEST] <tdr> control
[07:09:11 CEST] <Ashish> we are doing nothing right now. we are just executing commands.
[07:10:46 CEST] <tdr> Ashish, i meany you can use the os to control the use too, depening on what youre trying to tune, its easier to do it with the OS vs ffmpeg cli
[07:15:25 CEST] <Ashish> how can we do this with os ?
[07:15:55 CEST] <Ashish> we are creating mp4 video from property photos which pan and zoom effect.
[07:16:47 CEST] <Ashish> we are doing image concation, doing text overlay, image ovelay, text fade in, converting it to flv.
[07:18:46 CEST] <tdr> ok you said you wanted to control cpu utilization
[07:19:04 CEST] <Ashish> Yes
[07:19:07 CEST] <tdr> are you needing realtime scheduling or want to limit cores or cpu frequencey or ?
[07:20:09 CEST] <Ashish> currently when we are developing video, our cpu usage is getting higher. server load is going to above 10
[07:20:50 CEST] <Ashish> if we can limit cpu core and cpu frequency.
[07:33:56 CEST] <Ashish> @tdr, are you there ?
[07:34:06 CEST] <Ashish> have you find something useful ?
[08:31:26 CEST] <bencc> is the CPU cache important to ffmpeg encoding performance?
[08:31:37 CEST] <bencc> will I see a big difference between 8MB and 20MB?
[08:56:47 CEST] <Mysoft> heh i found this... that is what i wanted
[08:56:47 CEST] <Mysoft> https://sourceforge.net/projects/ffdec/?source=navbar
[08:56:51 CEST] <Mysoft> but this one is old
[09:50:36 CEST] <durandal_170> bencc: yes its impact is huge
[13:13:32 CEST] <wallbroken> why everyday are there new releases?
[13:17:16 CEST] <mrelcee_> every day?  it was over  a month between 3.11 and 3.12  - if you're talking about nightlies, well that's the nature of nightlies..
[13:22:09 CEST] <Mysoft> heh you could be on north pole
[13:22:17 CEST] <Mysoft> and only have nighly builds every 2 months
[13:22:17 CEST] <Mysoft> :P
[13:23:28 CEST] <mrelcee_> heheheh
[13:24:02 CEST] <mrelcee_> i just spotted 3.1.2 is out.   i'm running a git snapshot from about 3 weeks ago..     i think I'll wait a bit, it's doing fine..
[13:24:19 CEST] <Mysoft> heh
[13:24:21 CEST] <mrelcee_> 3.1.1 was crashing for me a lot..
[13:24:27 CEST] <Mysoft> while i'm here i should request a bug ticket
[13:24:31 CEST] <Mysoft> to fix the win32 builds :P
[13:24:53 CEST] <mrelcee_> people still run win32?  :)
[13:25:07 CEST] <Mysoft> i don't know if people do
[13:25:09 CEST] <Mysoft> but i do :P
[13:26:07 CEST] <Mysoft> as libavcodec is using _wfopen_s and swscanf_s
[13:26:14 CEST] <Mysoft> from msvcrt.dll
[13:26:17 CEST] <Mysoft> but those are vista+ only
[13:26:48 CEST] <Mysoft> they should either use that dynamically or load it from the msvcr100.dll (or newer)
[13:27:07 CEST] <mrelcee_> i've been in the  mac and unix world for a long time..   I JUST put a win7 vm up on my server to run as an itunes server for home sharing my huge music library...    first windows install I've had in a long time
[13:27:16 CEST] <Mysoft> heh
[13:27:34 CEST] <Mysoft> my computer will never see anything "newer" than windows XP
[13:27:36 CEST] <mrelcee_> i'm killing my OSX Server service by service
[13:27:42 CEST] <Mysoft> except ofcourse by ReactOS :)
[13:27:46 CEST] <Mysoft> but i wont dowgrnade to vista+
[13:27:50 CEST] <Mysoft> *downgrade
[13:28:36 CEST] <Mysoft> (anyway i already fixed the problem with my patcher) that implements several stuff that is vista+
[13:28:51 CEST] <Mysoft> but it's pointless to not run on XP just because of that :)
[13:28:54 CEST] <mrelcee_> vista got okay after SP3.   Win7 is pretty tolerable on good hardware.  8/8.1 can blow me..    except i'm afraid of it's snaggletooth..      win10 my kids and wife are running and I don't hear many complaints..
[13:29:53 CEST] <Mysoft> nah vista still uses 2x ram than XP
[13:30:01 CEST] <Mysoft> and requires 2x more hdd space
[13:30:02 CEST] <bencoh> are you actually replacing an osx server with ... win7? oO
[13:30:09 CEST] <mrelcee_> bencoh: no
[13:30:18 CEST] <Mysoft> and has all the "annoying" unsecurity crap like trusted installer and UAC
[13:30:19 CEST] <mrelcee_> replacing osx server with freebsd.
[13:30:29 CEST] <Mysoft> i wont abdicate of that freedom :P
[13:30:30 CEST] <bencoh> ah :)
[13:30:43 CEST] <mrelcee_> i needed itunes though so I stuck a virtualbox VM of win7 on it
[13:30:46 CEST] <Mysoft> (whihc is the same reason i wont switch to linux) :P
[13:31:10 CEST] <mrelcee_> family has been using it and noticed nothing...   success
[13:33:37 CEST] <mrelcee_> right now everything the osx server was doing is moved over to the freebsd server - EXCEPT file sharing..      I'm NFS mountiung everything over gigether to provide services from the freebsd box..    i've been hatcheting old crap i have 1% chance of needing in the next 10 years that I have saved because I could..     so I can avoid buying all new drives and transfer the data over to another drive on the freebsd box...  then make a zfs pool..
[13:34:25 CEST] <mrelcee_> my drives are in a raid5 firewire 800 box formatted HFS+     not so goood for just plugging in to the BSD box..
[13:35:05 CEST] <mrelcee_> besides i don't have FW800 in that machine...
[13:53:20 CEST] <wallbroken> https://ffmpeg.zeranoe.com/builds/win64/static/
[13:53:24 CEST] <wallbroken> as you can see
[13:53:33 CEST] <wallbroken> there is a build per day
[13:58:15 CEST] <mrelcee_> is there something the version you have isnt' doing right or are you just interersted in being on the bleeding edge?
[13:58:50 CEST] <wallbroken> the question was: [13:13] <wallbroken> why everyday are there new releases?
[13:59:43 CEST] <wallbroken> you said: there aren't every day builds, and i demonstrate you thate there are. now the question is still the original
[13:59:45 CEST] <mrelcee_> because development is always happening and they do quasi-nightly releases just like you can check out the source via git and compile your own...
[13:59:57 CEST] <mrelcee_> there's skipped days in there   :)
[14:00:27 CEST] <mrelcee_> anyhow ive given you an answer there...
[14:00:44 CEST] <mrelcee_> ny question is why do you care
[14:03:14 CEST] <mrelcee_> you're also mistaking a quasi-nightly snapshot build for a "release"     A release gets a version number not a build number.
[17:33:41 CEST] <jack_thompson>     Stream #0:0: Video: vp8, yuv420p, 480x360, SAR 1:1 DAR 4:3, 1k fps, 29.97 tbr, 1k tbn, 1k tbc (default)
[17:33:46 CEST] <jack_thompson> 1k fps? :o
[17:34:18 CEST] <jack_thompson> ffmpeg version N-78307-gf89bc66 Copyright (c) 2000-2016 the FFmpeg developers
[17:35:30 CEST] <DHE> some file formats have variable FPS
[17:35:44 CEST] <DHE> mkv is a big one there
[17:35:55 CEST] <kepstin> with vp8, that's probably webm (mkv), which normally uses a 1ms (1/1000s) timebase
[17:36:07 CEST] <kepstin> so if ffmpeg can't guess a constant fps, that seems like what it would show
[17:37:56 CEST] <jack_thompson> alright.. thanks :)
[17:40:04 CEST] <viric> how do people encode to variable fps?
[17:40:18 CEST] <viric> is it from a camera with variable fps?
[17:40:59 CEST] <DHE> depends on the source. sometimes people use a frame dropper that eliminates identical frames (eg periods of black). that qualifies since the framerate drops
[17:44:01 CEST] <viric> as there is always some noise... the definition of 'identical frames' may involve some threshold
[17:44:25 CEST] <kepstin> many webcams, particularly in low light, reduce their framerate
[17:44:32 CEST] <viric> ah yes
[17:45:35 CEST] <kepstin> some 60i video content containing a mix of interlaced and telecined stuff will often be turned into a vfr file containing sections at 30fps (deinterlaced) and 24fps (detelecined)
[17:46:41 CEST] <Mysoft> whoa
[17:46:55 CEST] <kepstin> that's pretty rare nowadays, but used to be more common :/
[17:47:15 CEST] <Mysoft> my software video player with libav+directx is faster than mpc-hc with nv12+overlay O_o
[17:48:29 CEST] <Mysoft> maybe because i'm using 16bpp output instead of 32
[17:49:06 CEST] Action: kepstin doesn't know of any video card newer than the voodoo 3 where 16bpp was actually faster than 32bpp
[17:49:14 CEST] <iive> nv12 is 12 bit
[17:49:56 CEST] <Mysoft> iive yes but the ddraw version of mpc-hc would be 32bpp (or desktop)
[17:50:03 CEST] <Mysoft> either way i doesnt know how it got faster hehe
[17:50:12 CEST] <Mysoft> but that's great for that 2004 notebook :)
[17:50:17 CEST] <kepstin> well, nv12 is 12bpp, but that's because of subsampling. it's still 8 bits per sample
[17:50:25 CEST] <Mysoft> now i can have 1280x1024 at 60fps (with no deblock filter at least)
[17:50:37 CEST] <Mysoft> *1280x720
[17:50:43 CEST] <iive> the overlay means that the colorspace conversion is done at runtime by the hardware
[17:51:11 CEST] <iive> so, probably it doesn't use nv12 at all and fallbacks to something else
[17:51:19 CEST] <kepstin> unless you're on a modern design video card which doesn't really have "overlays" any more, but emulates them with shaders :/
[17:51:30 CEST] <Mysoft> 2004 notebook... :P
[17:51:41 CEST] <iive> then it uses the code word "blitter"
[17:51:43 CEST] <Mysoft> it has OpenGL 2.0 :]
[17:52:12 CEST] <Mysoft> couldnt get it really doing hw acc with dxva there yet (for h264)
[17:52:23 CEST] <iive> older cards also prefered yuy2 type of colorspace.
[17:52:40 CEST] <Mysoft> i think i confused the nv12 with my pc
[17:52:54 CEST] <Mysoft> but my point is that there's no colorspace conversion for the overlay
[17:54:47 CEST] <kepstin> there might be, depending on the formats the overlay supports, but it'll usually just be a shuffle (e.g. between planar and packed)
[17:55:18 CEST] <Mysoft> but it's great since i implemented deblock ->  frameskip properly
[17:55:39 CEST] <Mysoft> so heopfully my player will handle 1920x1080p (15 effective fps)
[17:55:59 CEST] <Mysoft> on the notebook there... i will benchmark it will be faster
[17:56:13 CEST] <Mysoft> to upload the 1920x1080 with directx
[17:56:28 CEST] <Mysoft> or scale to 1280x720 prior upload
[17:56:50 CEST] <kepstin> hmm. 2004... you're probably talking about either a late model core 2 or nehalem-era core. Shouldn't have any issue with 1080p decoding, i'd expect.
[17:56:58 CEST] <kepstin> of course laptop chips tend to be slower
[17:57:28 CEST] <Mysoft> it's a sempron 3200+ (single core) 1.6ghz :P
[17:58:08 CEST] <Mysoft> but i made the player just as a "preview" option i by no means expected it to be that fast
[17:58:17 CEST] <Mysoft> i dunno why but it's 2x faster than ffplay
[17:59:28 CEST] <kepstin> ouch, a 2004 amd mobile chip? that was a bad era for amd mobile chips :)
[17:59:42 CEST] <Mysoft> well i think this notebook is really good :)
[18:00:11 CEST] <Mysoft> it is slight slower than my 3200+ desktop (but the desktop was 2.0ghz)
[18:00:29 CEST] <Mysoft> the ATI card it has... isnt the best regarding OpenGL support (DX works much better)
[18:00:51 CEST] <Mysoft> but i managed to upgrade the drivers to get OpenGL 2.0 (with 1.1 shaders! which is awesome :P)
[18:01:00 CEST] <Mysoft> *1.1 GLSL
[18:01:01 CEST] <furq> nehalem was 2009
[18:01:23 CEST] <furq> the core 2 wasn't introduced until 2006
[18:02:40 CEST] <Mysoft> i will later try to use my own overlay and re-check the performance
[18:02:45 CEST] <furq> i would have charitably assumed you meant netburst if you'd not mentioned core 2
[18:04:10 CEST] <Mysoft> and then do some tests... to see how far i can go into using shaders (or even the fixed pipeline) to do colorspace conversions on GL1.4-2.0 since i'm having lot of fun with those :D
[18:07:38 CEST] <Mysoft> altough it wont be necessary since divx and those formats that support "av_codec_set_lowres" are much faster already than h264
[18:07:53 CEST] <Mysoft> but i wonder what would look better... frameskip or lowres
[18:08:58 CEST] <Mysoft> i guess it depends on the video playing... like a superfast game playing 60fps (like quake 3) would look better at lowres than frameskip
[19:04:52 CEST] <besc> Anybody know how to reduce 60fps MP4 video with audio to 30fps? Want to keep the same speed, so drop every second frame?
[19:05:07 CEST] <furq> -vf fps=30
[19:08:06 CEST] <besc> furq: neat, thanks!
[19:56:04 CEST] <Cryp71c> I was in here a few days ago regarding this same issue but I realized I had messed something up when talking to you guys. The issue was a 10 minute, 250MB 1920x1080 30fps video taking ~50 minutes to transcode using libx264. The suggestions I got were to primarily look at the CRF value and the preset used.
[19:56:30 CEST] <Cryp71c> What I had messed up was the command I was using w/ ffmpeg to do the transcoding, I was wrong in what I provided.
[19:56:56 CEST] <Cryp71c> http://paste2.org/mWO7kMvB Is the actual command, and I wanted to make sure there wasn't anything else glaringly wrong with the various options we're using that could be affecting performance significantly
[19:56:57 CEST] <Cryp71c> http://paste2.org/mWO7kMvB
[19:58:55 CEST] <Mysoft> yay 1080p 30fps... with 5-8 frameskip (per second)... nice lol
[20:01:43 CEST] <furq> 50 minutes to encode 480*270 at preset medium?
[20:02:26 CEST] <furq> that's appalling
[20:02:44 CEST] <furq> my old core 2 duo at stock clocks is much faster than that
[20:03:28 CEST] <furq> Cryp71c: did you ever find out what cpu that vm is running
[20:03:42 CEST] <furq> if it's not an atom then something is badly wrong
[20:03:44 CEST] <TD-Linux> 15fps?
[20:04:02 CEST] <TD-Linux> oh are you targeting a psp?
[20:04:26 CEST] <furq> i guess it's for adaptive bitrate streaming
[20:12:16 CEST] <Cryp71c> furq, this is on a beta system so its not as strong as our prod boxes. Its definitely running on a single core and that could be shared, honestly I'm not 100% sure.
[20:12:38 CEST] <furq> 6fps for that output is really bad
[20:14:52 CEST] <Cryp71c> Anything suspicious with the command we're using for transcoding?
[20:15:08 CEST] <Cryp71c> Could disk performance be a significant factor?
[20:17:27 CEST] <Plorkyeran> if the input is uncompressed video then it's very easy to be i/o bound
[20:17:38 CEST] <Plorkyeran> if not then something has to be very wrong for i/o performance to matter
[20:17:48 CEST] <furq> the command looks fine
[20:17:56 CEST] <furq> i get 60+fps running that on a core 2 duo
[20:18:51 CEST] <furq> run top while it's transcoding, you should be able to see if you're io bound
[20:19:37 CEST] <furq> actually if it's 0.2x at 15fps then that's 3fps isn't it
[20:19:44 CEST] <furq> so 20x slower than a dualcore cpu from 2007
[20:19:49 CEST] <furq> that's not great
[20:20:31 CEST] <furq> i take it you're using ffmpeg from a distro repo, so it should be built with yasm etc
[20:20:33 CEST] <viric> ffmpeg in the mobile phone runs faster
[20:20:50 CEST] <viric> (termux)
[20:21:04 CEST] <Cryp71c> furq, what should I be looking for for determining i/o boundedness?
[20:21:28 CEST] <furq> the "%Cpu(s)" line
[20:21:43 CEST] <kepstin> should also check if the instance type is something like the amazon 'tiny' instances, where you have limited compute credits, and the instance turns hella slow when you run out.
[20:21:47 CEST] <furq> if you have high numbers for id (idle) or wa (iowait) then you're io bound
[20:21:54 CEST] <furq> although idk how well that'll work in a vm
[20:22:40 CEST] <kepstin> if the main cpu usage - us (user) - is significantly <100%, then you're limited by something other than cpu
[20:23:20 CEST] <kepstin> (or you're limited by cpu, because the cpu is being used for something else - e.g. stolen by vm host)
[20:24:00 CEST] <furq> also your ffmpeg process should be at the top of the process list (unless you changed the sort order)
[20:24:20 CEST] <furq> and be showing something close to (100 * core count) %CPU
[20:28:53 CEST] <Cryp71c> Here is another command used for the 1080p conversion:  -i /var/lib/qualboard/temp/vid-KiG6rW -ar 44100 -b:a 128k -b:v 1200k -f mp4 -r 30 -pix_fmt yuv420p -vf "scale=if(gt(a\,16/9)\,1280\,trunc(720*a/2)*2):if(lt(a\,16/9)\,720\,trunc(1280/a/2)*2),pad=1280:720:if(gt(a\,16/9)\,0\,(1280-iw)/2):if(lt(a\,16/9)\,0\,(720-ih)/2)" -metadata:s:v:0 rotate=0 -c:v libx264 -preset faster -crf 25 -c:a aac -strict experimental -y
[20:29:03 CEST] <Cryp71c> (The earlier one was for the "Low quality" which finishes much faster
[20:29:10 CEST] <Cryp71c> That one is for 1280, I believe.
[20:29:57 CEST] <kepstin> that one's using preset 'faster' rather than preset 'medium' (the default if unset), so i'd expect it to be faster, yes
[20:30:03 CEST] <Cryp71c> I just changed that one
[20:30:07 CEST] <kepstin> with slightly lower encoding efficiency, of course
[20:30:12 CEST] <Cryp71c> the crf and preset weren't there until a moment ago
[20:30:59 CEST] <furq> if it's the 720p conversion which runs at 3fps then that's not completely hopeless
[20:31:02 CEST] <furq> it's not good though
[21:12:59 CEST] <pgorley> what's the difference between d3d11va and dxva2 on windows? is one better than the other?
[00:00:00 CEST] --- Fri Aug 12 2016