[Ffmpeg-devel-irc] ffmpeg.log.20160531

burek burek021 at gmail.com
Wed Jun 1 02:05:01 CEST 2016


[01:08:36 CEST] <Fox_> Hi I have a question about encoding using ProRes - is it possible to change to 8-bit and yuvj422p? My source files like that but a friend of mine cannot open my files in FCP so I encoded them using ProRes but am just wondering.
[01:10:30 CEST] <furq> Fox_: http://vpaste.net/A2Fci
[01:10:33 CEST] <furq> doesn't look like it
[01:11:04 CEST] <Fox_> Okay I guess that's fine it's not a big deal, thanks for the info.
[01:12:02 CEST] <Fox_> Thanks again
[01:51:28 CEST] <grrk-bzzt> The ffmpeg wiki mentions x264 has a lossless mode (by using -qp 0) but I remember someone explaining to me that it's not lossless
[01:51:35 CEST] <grrk-bzzt> Can someone confirm that?
[01:52:20 CEST] <CoJaBo> grrk-bzzt: Unless your input is already YUV, there's a colorspace conversion which is lossy
[01:52:25 CEST] <kepstin> grrk-bzzt: it is lossless if used correctly. In some cases, e.g. if you have rgb source and use the 'libx264' encoder (instead of 'libx264rgb'), the color conversion is lossy.
[01:52:54 CEST] <grrk-bzzt> Alright
[01:52:55 CEST] <CoJaBo> ..can it actually do rgb? Never seen that before
[01:53:06 CEST] <grrk-bzzt> So it's about the pixel format
[01:54:42 CEST] <kepstin> if you have an 8-bit video and use a 10-bit x264 library or vice-versa, i'd also expect some changes.
[04:43:53 CEST] <AndrewMock> I take it that the release heads are built into 'stable' binaries?
[04:48:03 CEST] <AndrewMock> looks like it
[08:39:18 CEST] <AndrewMock> Which library(ies) is used on Linux when doing the whole OpenCL-x264-lookahead thing?
[10:09:27 CEST] <nifwji2_> does anyone remember flipnote hatena?
[10:10:05 CEST] <nifwji2_> it was a great place to share animations made in the flipnote studio app.
[10:10:11 CEST] <nifwji2_> then they shut it down.
[10:10:37 CEST] <nifwji2_> but some guys managed to reverse engineer the code.
[10:10:50 CEST] <nifwji2_> so now there is a new flipnote hatena.
[10:10:55 CEST] <nifwji2_> anyway
[10:11:23 CEST] <nifwji2_> I was thinking about encoding all of the flipnotes I downloaded over the years into mp4 files or webms
[10:12:18 CEST] <nifwji2_> I have a program and it's source code on my computer that can decode the format.
[10:12:24 CEST] <nifwji2_> I just need to learn python.
[10:12:55 CEST] <nifwji2_> and then figure out how to encode them into any format I want.
[10:13:28 CEST] <nifwji2_> how difficult would it be to do this?
[10:13:41 CEST] <nifwji2_> would it be really simple?
[10:48:14 CEST] <nifwji2_> I was just trying to download a video off of twitter.
[10:48:42 CEST] <nifwji2_> I looked at the source found a link then found a link in that file then found a bunch of links in that file.
[10:49:03 CEST] <nifwji2_> that bunch of links was the video split up into 3 second parts.
[10:49:28 CEST] <nifwji2_> why not just store the video as one file?
[10:53:53 CEST] <wismas> Hi! I have 100 thousand audio files that I have to concatenate, with silence every two files, and without an easily specified order, so that I have to manually build a list to give to ffmpeg
[10:55:34 CEST] <wismas> each file is 30 characters long too, so while I don't know how ffmpeg commands work exactly, a single command would be at least 3 million characters
[10:56:00 CEST] <wismas> at an high level what kind of command(s) can I use to do the concatenation?
[11:14:22 CEST] <nifwji2_> where did you get 100000 audio files?
[11:33:51 CEST] <wismas> nifwji2_, a text to speech API.
[11:38:27 CEST] <IntelRNG> It looks like a work for a script rather than a single ffmpeg line.
[12:02:01 CEST] <dannyhook> hi folks
[12:02:37 CEST] <dannyhook> I need help about udp streaming. Who can help me?
[13:35:26 CEST] <jcdejong> hm, apparently I was disconnected, so not sure if your question was already answered @wismas. But can't you put the list of files in a concat.txt file and use -f concat -i concat.txt to do the concatenation?
[14:27:30 CEST] <dannyhook> who is online?
[14:30:33 CEST] <durandal_1707> nobody
[14:34:49 CEST] <dannyhook> I'm streaming rawvideo to udp. When I read flow, video has random split effect due to unrielability of protocol. Does anybody knows a workaround?
[14:52:48 CEST] <DHE> don't use UDP I'm afraid
[14:53:02 CEST] <DHE> it's especially bad if the connection travels between ports of varying levels of speed or congestion
[14:53:48 CEST] <Mavrik> Yeah, that configuration is pretty much a recipe for disaster
[14:54:03 CEST] <Mavrik> But I believe we talked about that here already :P
[15:15:23 CEST] <dannyhook> I have to send to streaming flows into the same machine. One has to be raw and another compressed. On demand I have to read these flows. Can you give me a hint?
[15:15:37 CEST] <dannyhook> *two streaming flows
[15:17:25 CEST] <DHE> having direct experience with high end switches (eg: Cisco & Junpier) I can tell you they'll cause problems in situations like that. So I can only imagine anything cheaper will just do worse
[15:18:26 CEST] <dannyhook> ok
[15:18:46 CEST] <dannyhook> Then I need a hint
[15:21:47 CEST] <P4Titan> Hello all. I have a question: I make a call to avcodec_encode_audio2 with 256 bytes of data, but the output is always 4 bytes of data long. Does anyone know why that occurs?
[15:24:26 CEST] <dannyhook> non ti aiuta nessuno qui'
[16:17:29 CEST] <DHE> P4Titan: that sounds a little small for most codecs. are you fulfilling the block size requirement?
[16:19:40 CEST] <P4Titan> What exactly do you mean?
[16:20:24 CEST] <DHE> codecs do have a minimum input size. I think AAC needs either 1536 or 2048 samples at a time
[16:20:41 CEST] <P4Titan> hmm, intersting
[16:21:22 CEST] <DHE> for using AAC as an example..
[16:21:32 CEST] <P4Titan> Should I look at the AVFrame->nb_samples?
[16:23:38 CEST] <DHE> See the transcode_aac.c example file. They use an av_audio_fifo to buffer audio because the input and output codecs can have mismatched frame size requirements
[16:24:07 CEST] <P4Titan> yes, I have implemented that fifo from looking at that file
[16:25:37 CEST] <DHE> "const int output_frame_size = output_codec_context->frame_size;"
[16:26:14 CEST] <DHE> you should be giving the codec frames with this many samples
[16:27:58 CEST] <P4Titan> if I pastie you code, could you please give it a look?
[16:34:48 CEST] <DHE> only if it's really simple. a bit busy here
[16:35:22 CEST] <P4Titan> ok
[16:35:32 CEST] <P4Titan> is this frame_size in terms of bits or bytes
[16:35:51 CEST] <P4Titan> because I set the frame->nb_samples to output_codec_context->frame_size
[16:36:10 CEST] <P4Titan> yet write to the frame: frame_size / sizeof(double)
[16:36:20 CEST] <P4Titan> is that bad?
[17:13:55 CEST] <P4Titan> DHE: When trying to encode pcm into ffmpeg, my first pass I get 21 bytes in the output
[17:14:22 CEST] <P4Titan> but with these errors: [trace] encoder.c :: ERROR :: [mp4 @ 0x4d4600]
[17:14:22 CEST] <P4Titan> [trace] encoder.c :: ERROR :: Timestamps are unset in a packet for stream 0. This is deprecated and will stop working in the future. Fix your code to set the timestamps properly
[17:14:22 CEST] <P4Titan> [trace] encoder.c :: ERROR :: [mp4 @ 0x4d4600]
[17:14:22 CEST] <P4Titan> [trace] encoder.c :: ERROR :: Encoder did not produce proper pts, making some up.
[17:14:42 CEST] <P4Titan> And then the output drops to 4 bytes. Could this potentially be the issue?
[18:02:33 CEST] <pfelt> greatest of mornings to you all
[18:03:01 CEST] <pfelt> between 2.8 and 3.0 we deprecated AVCodec.pkt  is that struct available somewhere else?
[18:03:27 CEST] <pfelt> specifically, i'm trying to set the pts on a frame
[18:04:47 CEST] <pfelt> er& misspoke.  that should be AVCodecContext
[18:05:18 CEST] <gmh> Somethings tell me you guys are into video :P Sony KD-65X9005A or Samsung UE65JU6075 ?
[19:09:38 CEST] <f00bar80> asking how I can analyse the relation bitrate , quality, resolution ? just i need a clarification to what can be identified as affecting factors
[19:11:53 CEST] <klaxa> what
[19:12:03 CEST] <klaxa> oh
[19:12:30 CEST] <klaxa> psnr, ssim? subjective quality scoring?
[19:12:54 CEST] <klaxa> i'm still not entirely sure what you want to know though
[19:21:03 CEST] <f00bar80> quality/bitrate/resolution/cpu usage analysis
[19:23:42 CEST] <DHE> you might want to grab a copy of x264 and read the help guide, or just use the presets for "fast", "slow", "superfast", etc
[19:26:38 CEST] <f00bar80> DHE: whats the use of profile:v high .. I couldn't find anything about that in the ffmmeg man
[19:27:58 CEST] <DHE> that's a profile. it sets the features the decoder will have. some decoders (like older cell phones) only support baseline, for example
[19:28:11 CEST] <DHE> though 'high' will usually raise CPU requirements since it allows more features
[19:32:02 CEST] <emitchell> thebombzen: hey sorry, i let your answer about my question a few days ago slip off my scrollback. I can re-ask if you want, sorry about that!
[19:32:17 CEST] <f00bar80> DHE: mroe features like what? as well suppose all clients are on 1M ADSL connection what can be the suitable bitrate/resolution?
[19:32:25 CEST] <kepstin> note that even if you say e.g. profile "high", x264 might use a lower profile, particularly in the faster encoding presets.
[19:32:49 CEST] <thebombzen> emitchell: what was your question?
[19:32:52 CEST] <DHE> DHE: for example, pixels can be encoded in smaller block sizes. b-frames can be allowed. different encoding algorithms and transformations can be used. I'm probably forgetting a few things...
[19:34:31 CEST] <f00bar80> DHE: suppose all clients are on 1M ADSL connection what can be the suitable bitrate/resolution?
[19:34:43 CEST] <emitchell> thebombzen: I have a sporting event I want to split into it's component matches based on start/end times from an API. My question was: is there any concern about splitting a live video stream into a file?
[19:34:49 CEST] <furq> f00bar80: that's something only you can answer
[19:35:16 CEST] <furq> it massively depends on your source and how powerful the encoding machine's cpu is
[19:35:25 CEST] <DHE> f00bar80: depends on the content and what your encoder can do. I would start at something like 480p and see how it goes, and then play with it from there.
[19:35:36 CEST] <kepstin> f00bar80: well, with a 1M ADSL, you obviously need a bit rate <1M (you'll probably want between 500-800kbit depending on overhead, etc.)
[19:35:48 CEST] <thebombzen> emitchell: depends on what you mean by split. do you mean fork the stream into two identical copies, or do you mean create a new file every 10 minutes?
[19:35:51 CEST] <DHE> and don't forget you have, what, 128k audio?
[19:35:54 CEST] <emitchell> I can totally play around with steaming it and trying to save to a file but i just want to make sure this is a) possible and b) wont result in useless resulting videos
[19:36:14 CEST] <emitchell> thebombzen: i would like to create a new file every $minutes
[19:36:27 CEST] <emitchell> depending on the start/end time from the API
[19:36:39 CEST] <kepstin> emitchell: what's the format (video codec, container, etc.) of the original video stream?
[19:36:40 CEST] <DHE> emitchell: there's a "segment" output format (-f segment) which can help with that, but you'll need to give it a lot of options. like segment size, output type, file list format, etc
[19:36:50 CEST] <f00bar80> DHE: based on a 4 cores  Intel(R) Xeon(R) CPU E3-1271 v3 @ 3.60GHz
[19:36:50 CEST] <emitchell> kepstin: it's an RTMP stream
[19:37:07 CEST] <emitchell> DHE: thank you
[19:37:14 CEST] <furq> we don't know how powerful every cpu is with regard to encoding 8 streams of unknown resolution
[19:37:15 CEST] <kepstin> emitchell: you might be best off just starting up a new ffmpeg for each segment and stopping it when the segment's done
[19:37:24 CEST] <furq> just try various settings and see if they work
[19:37:40 CEST] <f00bar80> DHE: what about the Bitrate ?
[19:37:57 CEST] <kepstin> emitchell: probably start each ffmpeg a little early to make sure you catch a keyframe before your desired actual start point
[19:37:59 CEST] <emitchell> kepstin: okay ill give that a shot
[19:38:46 CEST] <f00bar80> Is there a way to analyse the encoded stream regarding the current machine and connection I'm playing this stream on?
[19:39:08 CEST] <thebombzen> well that answer isn't the one I gave a while ago. I thought you meant fork a stream. I don't actually know beyond what kepstin just said.
[19:39:20 CEST] <emitchell> yeah my set up is: RTMP stream comes in, we record it on the server, we split video off the live stream as it is being recorded, and then we pipe the incoming stream with some overlays out to a video streaming service (twitch)
[19:40:55 CEST] <emitchell> thebombzen: thanks for your help. i'll play around with it and ill come back if i have any more questions. Thanks!
[19:41:43 CEST] <DHE> f00bar80: usually you'd go for audio from 64 to 128k, which leaves you around 800 kilobits (tops) for video
[19:42:01 CEST] <DHE> again, don't be afraid to experiment a little bit
[19:42:45 CEST] <f00bar80> DHE: yea but i want any analysing tool on the client end .. based on which i can see the experiments results
[19:42:51 CEST] <furq> f00bar80: you can use https://github.com/mariusae/trickle to simulate a slow connection
[19:43:14 CEST] <furq> trickle -d 1024 ffplay mystream.m3u8
[19:43:32 CEST] <f00bar80> Also can i use and depend on the Kush Gauge formula?
[19:43:47 CEST] <furq> the kush gauge pretty much seems like nonsense to me
[19:43:56 CEST] <furq> i guess you could use it as a starting point
[19:44:30 CEST] <furq> you are going to have to rely on trial and error if you want to optimise this though
[19:45:11 CEST] <DHE> video encoding is as much an art as a science
[19:45:26 CEST] <f00bar80> what's the effect of buffer size flag on the encoded output ?
[19:45:26 CEST] <furq> yeah there's no correct answer we can give you
[19:45:48 CEST] <furq> the buffer size is there solely to ensure you don't exceed the bandwidth limit
[19:46:00 CEST] <furq> it might reduce quality a bit but you either need it or you don't
[19:46:13 CEST] <DHE> a large buffer lets you spike above 1 Mbit when you need high quality in an action scene if it leads in with low bandwidth first (client builds a buffer)
[19:46:58 CEST] <DHE> depending on the client and streaming method, the buffer size might be pre-determined
[19:47:18 CEST] <DHE> for example, a DVD calls for ~1800 kilobit buffer
[19:47:26 CEST] <furq> that flag is mostly useful for hardware devices which have a known fixed size buffer
[19:47:47 CEST] <furq> i'm not sure how useful it really is for streaming
[19:47:51 CEST] <f00bar80> even -fflags +genpts can't be found in man ffmpeg
[19:48:57 CEST] <furq> f00bar80: https://www.ffmpeg.org/ffmpeg-formats.html#Format-Options
[19:49:15 CEST] <furq> or man ffmpeg-formats
[19:56:18 CEST] <kepstin> f00bar80: for streaming, the buffer size is more or less related to how long the player will spend "buffering" at the start before it starts playing. Bigger means longer wait, but bigger means that the video quality will be more consistent.
[19:57:36 CEST] <kepstin> with a bigger buffer size, which means more delay, it can make keyframes bigger so the video will overall look better.
[20:03:23 CEST] <f00bar80> when tried to play the encoded stream , i see no motion but sound is fine , this can be due to what?
[20:05:47 CEST] <f00bar80> this is the encoding i ran >
[20:05:53 CEST] <f00bar80> http://vpaste.net/X33HH
[20:10:39 CEST] <Frod> hello all
[20:11:01 CEST] <Frod> can some one tell me what is wrong with this line
[20:11:02 CEST] <Frod> ffmpeg -i KMg2wjOI5BV67w60f7zJ.mp4 -vcodec libx264 -vf "movie=/var/www/minilogo.png [watermark]; [in][watermark] overlay=10:10 [out], scale=min(640\,iw):trunc(ow/a/2)*2" -vb 800k -acodec aac -strict -2 -ab 96k -ar 44100 -f mp4 /enc/KMg2wjOI5BV67w60f7zJ_temp.mp4
[20:11:26 CEST] <c_14> that's not a vf, that's a filter_complex
[20:11:53 CEST] <c_14> and if you still need -strict experimental for the aac encoder, you should update your version of ffmpeg
[20:12:25 CEST] <Frod> the error i get is Simple filtergraphis does not have exactly one input and output
[20:12:48 CEST] <Frod> c_14: i will try to look for the filter_complex
[20:13:42 CEST] <c_14> just replace -vf with -filter_complex
[20:15:21 CEST] <Frod> my ffmpeg is version ffmpeg version 2.8.5
[20:15:58 CEST] <Frod> c_14: with exactly the same syntax ??
[20:16:37 CEST] <c_14> should be fine, ye
[20:16:42 CEST] <c_14> and you want to update to at least 3.0.1
[20:17:05 CEST] <c_14> eh, 3.0.2
[20:21:06 CEST] <f00bar80> somebody mentioned a client side analysis tool or low connection analyser .. plz resend
[20:25:37 CEST] <Frod> c_14: by just replacing the -vf i got this error
[20:25:38 CEST] <Frod> Stream specifier 'in' in filtergraph description movie=/var/www/minilogo.png [watermark]; [in][watermark] overlay=10:10 [out], scale=min(640\,iw):trunc(ow/a/2)*2 matches no streams.
[20:25:58 CEST] <c_14> replace in with 0:v
[20:26:15 CEST] <c_14> also, eh wait
[20:26:24 CEST] <c_14> get rid of [out]
[20:29:58 CEST] <f00bar80> ppl any comment
[20:33:38 CEST] <f00bar80> this is the encoding i ran >  http://vpaste.net/X33HH
[20:33:53 CEST] <f00bar80> when tried to play the encoded stream , i see no motion but sound is fine , this can be due to what?
[20:39:24 CEST] <CoJaBo> f00bar80: I had weirdness happen with superfast, does it still happen with veryfast?
[20:42:35 CEST] <DHE> f00bar80: why are you using -dn ?
[20:51:07 CEST] <f00bar80> DHE: I thought it allows me not to keep disk space
[20:51:27 CEST] <DHE> .... no
[20:53:41 CEST] <f00bar80> so what is its use then ?
[20:54:11 CEST] <f00bar80> DHE: and that's the reason why there's delay of/no motion
[20:54:57 CEST] <DHE> that's what I'm asking you...
[21:01:12 CEST] <f00bar80> DHE: as i said , i thought it doesn't copy data so it can help in not using extra disk space , so if it's not... then what is its usage ?
[21:05:19 CEST] <DHE> I'm not really sure. the description is vague, but it's in the video section so I'm calling it there
[21:08:43 CEST] <f00bar80> DHE: so what else can be the reason for the no motion issue , could it be the player ? I'm using VLC
[21:09:12 CEST] <P4Titan`> Question: if I wish to convert pcm to aac, would I need a resampler like the transcode_aac.c file
[21:10:10 CEST] <tlhiv_work> whenever i capture audio and video from v4l using ffmpeg and i specify -r (different than default), the audio and video are not in sync and one gets done playing before the other
[21:10:32 CEST] <tlhiv_work> is there "postprocessing" that i can do to fix that?
[21:11:57 CEST] <DHE> put the -r before the output file, not the input file
[21:12:19 CEST] <tlhiv_work> DHE: let me try ... thanks
[21:13:23 CEST] <P4Titan`> So, what would be the process of loading pcm data into a fifo so that it can be later encoded into an aac?
[21:13:35 CEST] <kepstin> tlhiv_work: '-r' as an input option causes the frame timestamps to be rewritten, it doesn't change the framerate that the capture is running at. Try using "-framerate" as an input option instead.
[21:13:53 CEST] <tlhiv_work> kepstin: thanks ... trying that
[21:15:02 CEST] <tlhiv_work> i'm trying to fix a lot of "Past duration (time) too large" ... not sure what's causing it ... i'm trying to capture raw video/audio and store to raw and postprocess/postcompress
[21:15:41 CEST] <kepstin> I have no idea what those 'past duration too large' messages even mean :/
[21:16:51 CEST] <tlhiv_work> me either ... the capturing or playback (i can't determine which) is choppy ... the audio seems fine, but the video is choppy
[21:17:29 CEST] <tlhiv_work> ffmpeg-2.8-custom -framerate 15 -f v4l2 -s 1920x1080 -i /dev/video0 -f alsa -i hw:1,0 -c:a pcm_s16le -c:v rawvideo -qp 0 -q:a 0 -q:v 0 -r 15 -y out.mkv
[21:20:23 CEST] <P4Titan`> Does anyone have any thoughts?
[21:26:04 CEST] <kepstin> tlhiv_work: if you're using raw video / uncompressed audio, the -qp and -q output options are meaningless.
[21:26:53 CEST] <kepstin> tlhiv_work: I'd suggest removing the -r 15 output option if you're using -framerate 15 on the input, the -r option might adjust the timing a little bit, possibly causing some judder/choppiness.
[21:29:22 CEST] <tlhiv_work> still having issues ... oh well it could be a computer speed problem
[21:30:14 CEST] <kepstin> with raw video like that, your speed limit might just be your hard drive write speed.
[21:30:32 CEST] <kepstin> using a fast lossless compresser like ffvhuff might help
[21:32:01 CEST] <debianuser> tlhiv_work: raw video probably takes a lot of space, you may hit hdd write speed limit :) Maybe try adding some fast compression, -c:v libx264 -preset realtime   for example?
[21:32:23 CEST] <tlhiv_work> debianuser: thanks ... let me try that
[21:32:43 CEST] <debianuser> heh, kepstin typed that while I was looking for the preset name :)
[21:32:54 CEST] <ChocolateArmpits> realtime is a preset ?
[21:33:06 CEST] <ChocolateArmpits> When did that get introduced ?
[21:33:26 CEST] <kepstin> debianuser: yeah, 'realtime' isn't a preset that x264 has in its list...
[21:34:52 CEST] <debianuser> yeah... it's not :) I was still searching... found it: it's `-preset ultrafast -tune zerolatency` :)
[21:36:39 CEST] Action: kepstin notes that libvpx does have some fun realtime (deadline) modes where it allows itself to spend up to Xms per frame, but x264 doesn't have any features like that.
[21:38:54 CEST] <ChocolateArmpits> kepstin: You can force the input to be read at it's native rate with -re input command
[21:39:15 CEST] <kepstin> ChocolateArmpits: that's completely unrelated.
[21:39:16 CEST] <ChocolateArmpits> so a 30fps video will be read and transcoded at 30fps provided there is enough performance potential
[21:39:26 CEST] <ChocolateArmpits> kepstin: really ?
[21:40:05 CEST] <furq> -re makes no difference with a live input like v4l
[21:40:40 CEST] <ChocolateArmpits> well that's to be expected
[21:40:42 CEST] <kepstin> ChocolateArmpits: the way the libvpx dealine mode works is that given a realtime source (e.g. webcam, live stream, or you're using -re or whatever) it will auto-adjust the encoder settings so that it's encoding at realtime speed.
[21:40:55 CEST] <kepstin> mostly used for e.g. webrtc video chat and applications like that
[21:41:25 CEST] <ChocolateArmpits> ok so it will adjust the quality to meet realtime encoding
[21:51:22 CEST] <f00bar80> ppl no coment on what can be the reason .. of no motion in encoded output , referring to the vpv i sent before
[23:54:48 CEST] <rainabba> Assuming that I know little to nothing about setting up a build environment, can anyone point me to a good resource on doing so? I've tried and couldn't get a build; I believe because libraries that I was trying to build with; were not where they should be (or what ever would hint the compiler to their location wasn't setup). I'm perfectly comfortable in bash and for years, I built kernels,
[23:54:48 CEST] <rainabba> but I say assume I know nothing because I'm clearly missing one or more steps :)
[23:55:24 CEST] <JEEB> first of all I would recommend not trying to build a binary with the kitchen sink
[23:55:25 CEST] <CoJaBo> ^ I, too, would like to know this
[23:55:39 CEST] <CoJaBo> I suck at compiling anything
[23:56:01 CEST] Action: CoJaBo might just wait for the next static build >_>
[23:56:02 CEST] <rainabba> JEEB: Right now my primary needs are x264, aac and opencl support. I'll be working with prores and applying some video filters too.
[23:56:05 CEST] <JEEB> you can first of all think that all decoders are already included in libavcodec itself so no external libraries are required for that
[23:56:09 CEST] <DHE> start by running 'configure' without options, see what it comes up with
[23:56:26 CEST] <JEEB> for LC-AAC the new internal encoder is good enough, for HE-AAC you might still need something
[23:56:39 CEST] <JEEB> opencl in x264 is pretty much useless so no, you don't need it
[23:56:44 CEST] <rainabba> So just start with clone from the official ffmpeg repo?
[23:56:52 CEST] <rainabba> JEEB: opencl for unsharp
[23:57:00 CEST] <JEEB> ooh-kay...
[23:57:04 CEST] <JEEB> not sure how optimal that is :P
[23:57:27 CEST] <rainabba> Me either, but if it's factors faster, I have to try. Current build only uses 6 of 36 CPU cores :/
[23:57:36 CEST] <JEEB> it most probably is ont
[23:57:38 CEST] <JEEB> *not
[23:57:49 CEST] <DHE> then make -j 20
[23:58:28 CEST] <rainabba> Okay, what it looks like from htop anyway.
[23:58:28 CEST] <DHE> I pulled 20 out of my ass because 36 somehow seems high
[23:58:30 CEST] Action: rainabba reasearching -j so no stupid questions are asked
[23:58:46 CEST] <rainabba> Oh, it's 36 :)  BIG ASS EC2 instances
[23:58:54 CEST] <CoJaBo> lol
[23:59:15 CEST] <DHE> pfft, I have bare metal with 32 cores sitting next to my desk
[23:59:15 CEST] <rainabba> Playing around with 32-core G2 (quad GPU) and 36-core
[23:59:20 CEST] <CoJaBo> I think the most I've ever had access to is 34
[23:59:44 CEST] <JEEB> anyways, even with just 6 cores used you should be getting quite the speeds and capability. but you should in any case be rather calculating where your bottleneck is
[23:59:48 CEST] <rainabba> g2.8xlarge and c4.8xlarge
[23:59:52 CEST] <JEEB> whether it's decoding, filtering or encoding
[00:00:00 CEST] --- Wed Jun  1 2016


More information about the Ffmpeg-devel-irc mailing list