[00:07:13 CET] <cone-185> ffmpeg 03Andreas Cadhalpun 07master:ff8816f7172b: aacsbr: ensure strictly monotone time borders [00:07:14 CET] <cone-185> ffmpeg 03Andreas Cadhalpun 07master:a9c20e922cee: sbrdsp_fixed: assert that input values are in the valid range [00:07:15 CET] <cone-185> ffmpeg 03Andreas Cadhalpun 07master:fdc94db37e89: sbr_qmf_analysis: sanitize input for 32-bit imdct [00:15:31 CET] <cone-185> ffmpeg 03Paul B Mahol 07master:df2ce130a65c: avfilter/af_aemphasis: more declarations above [00:45:14 CET] <philipl> What a hilarious conversation this is becoming. [00:49:20 CET] Action: J_Darnley wonders which one [01:09:57 CET] <philipl> J_Darnley: the gpl nvenc one [01:10:11 CET] <J_Darnley> ah [01:31:39 CET] <ethe> Hi, I registered for the trac a few hours ago, but I havent received my confirmation email? I've checked the email, it is the right email; checked my spam as well. Any ideas? [01:44:41 CET] <J_Darnley> ethe: stick around and I'm sure someone can help you. [02:09:15 CET] <ethe> J_Darnley It's really late, so I'll be back tomorrow. My email is: joshh@protonmail.ch [02:09:32 CET] <J_Darnley> fine [02:09:34 CET] <J_Darnley> ... [02:10:11 CET] <J_Darnley> I'm sure someone can find that account. [02:19:32 CET] <J_Darnley> :O ... :( [02:40:36 CET] <kierank> michaelni: in build_vlc what does symbols_wrap mean? [02:49:46 CET] <kierank> and what does TEX_VLC_BITS mean? [02:51:04 CET] <kierank> seems to just be a number plucked from thin air [02:55:08 CET] <michaelni> *_VLC_BITS is the table size, that with the "depth" of the tables is restricted by how long the longest VLC is, beyond that tuning the value allows adjusting between resolving a VLC with a single lookup or requireing smaller tables and CPU data cache [02:56:15 CET] <kierank> how does that relate to symbols_wrap? [02:56:25 CET] <michaelni> the parameters to build vlc (if i dont mix things up) specifiy the vlc table layout like data type, space beteen elements [02:59:42 CET] Action: michaelni afk (will awnser any remaining questions tomorrow) [the table layout was meant as the input to build vlc] [03:00:56 CET] <beastd> J_Darnley: Trac account verification is pending for ethe . Maybe mail delivery took its time. I will look if it is still pending tomorrow. [03:01:44 CET] <J_Darnley> Okay. If I see him when I wakeup I will pass on the message. [03:02:12 CET] <beastd> ok. thanks. [09:28:32 CET] <khali> hello [11:52:24 CET] <durandal_1707> anybody tried sofalizer? [12:22:48 CET] <rcombs> hey who do I complain to about dumb mov things [12:22:57 CET] <rcombs> MOV chapters are dumb [12:22:59 CET] <BtbN> Apple [12:23:27 CET] <rcombs> "it's a subtitle stream with a 'frame' at the start point of each chapter containing the title" [12:23:43 CET] <rcombs> that's silly but what's sillier is lavf exposing it as a normal subtitle stream [12:23:55 CET] <rcombs> and not providing any way of identifying it as such (as far as I can tell) [12:24:20 CET] <rcombs> not sure if we should avoid exporting the stream at all, or just add an AV_DISPOSITION or some shit [12:24:27 CET] <nevcairiel> i think its possible to identify such streams, if they are muxed properly [12:24:47 CET] <nevcairiel> and if identifiable, it should just make chapters out of it, and not expose the stream [12:24:53 CET] <nevcairiel> i thought we could detect them [12:24:58 CET] <rcombs> we do detect them [12:25:01 CET] <rcombs> and make chapters out of them [12:25:07 CET] <rcombs> but it's also exported as a regular AVStream [12:25:37 CET] <rcombs> I think it's because the bit that tells us it's a chapter stream comes after the actual track definition atom [12:26:00 CET] <rcombs> so there'd be some awkward juggling required to remove a stream at that point [12:26:23 CET] <rcombs> maybe just call it AVMEDIA_TYPE_DATA [12:26:55 CET] <rcombs> I just want a way to detect and ignore it [12:28:38 CET] <nevcairiel> a disposition might be the easiest non-breaking way [12:29:25 CET] <rcombs> yeah, seems easy enough, if a bit silly from an API perspective [12:30:05 CET] Action: rcombs hands nevcairiel the Hard Problem of naming it [12:44:15 CET] <khali> does ffmpeg rely on a library to write still pictures (-f image2)? [12:47:54 CET] <nevcairiel> khali: no [12:50:49 CET] <khali> nevcairiel: not even for jpeg? I see it links with libopenjpeg [12:51:26 CET] <nevcairiel> it uses that optionally for decoding or encoding, not for writing files [12:51:35 CET] <nevcairiel> but its not a required dependency for anything [12:52:03 CET] <khali> ok [12:52:27 CET] <khali> ahhh [12:52:47 CET] <khali> jpeg (and tiff) use YUV colorspace [12:52:59 CET] <khali> while png and bmp use RGB colorspace [12:53:08 CET] <khali> that would explain the difference I'm seeing [12:55:27 CET] <khali> conversion from yuv410p to (at least) RGB24 seems to be broken [12:57:28 CET] <khali> I guess I should create a ticket with my findings [13:24:58 CET] <khali> done [14:12:24 CET] <J_Darnley> ethe: did you get the message left for you last night? [14:17:38 CET] <kartikey0303> hello i am new here. Can someone guide me how can i contribute. [14:19:29 CET] <J_Darnley> What can you do and what do you want to do? [14:20:52 CET] <J_Darnley> Succesfully building ffmpeg is usually the first thing to do if you want to contribute code. [15:00:22 CET] <cone-551> ffmpeg 03Michael Niedermayer 07master:5e8b9e746fd5: avutil/softfloat: Assert that the exponent did not overflow the legal range in av_normalize1_sf() [15:00:23 CET] <cone-551> ffmpeg 03Tom Marecek 07master:46cbb4c2312c: ffmpeg: change command line option -dump to work without -loglevel debug [15:16:59 CET] <ethe> J_Darnley no [15:16:59 CET] <ethe> however, I did get my confirmation email (eventually) [15:31:34 CET] <atomnuker> how in the hell is ffmpeg misidentifying a Dirac file for an AAC one [15:33:31 CET] <atomnuker> there's a BBCD at the start of every header, the extension is .drc and yet it's considered an AAC file [15:34:15 CET] <nevcairiel> is there even a raw dirac demuxer [15:34:39 CET] <atomnuker> yeah, there is [15:43:13 CET] <Compn> do we do filename extension probe ? [15:43:18 CET] <Compn> i know mplayer does... ;P [15:43:21 CET] Action: Compn runs [15:44:15 CET] <nevcairiel> there eare extension probes as well, but magic-number probes are usually favorable [15:44:20 CET] <nevcairiel> *are [15:47:21 CET] <durandal_1707> nobody tried sofalizer filter :( [16:06:28 CET] <atomnuker> that's odd, if I force the raw dirac demuxer to return max score, diracdec gets init'd [16:07:01 CET] <atomnuker> but since avctx->width, height and pix_fmt are unset it won't ever get to the decode frame function [16:07:18 CET] <nevcairiel> it cant parse this from the bitstream? [16:08:35 CET] <atomnuker> it's a raw demuxer, it checks if the first four bytes are 'BBCD' [16:08:43 CET] <atomnuker> that's all it does [16:08:59 CET] <nevcairiel> .. and they are not? [16:09:22 CET] <atomnuker> they are [16:09:42 CET] <atomnuker> but unless I explicitly force max detection score it gets misdetected as AAC [16:10:07 CET] <nevcairiel> it seems like it does use max score if all the tests pass [16:10:11 CET] <nevcairiel> so which one doesnt? [16:10:57 CET] <atomnuker> a file which I encoded myself [16:23:38 CET] <atomnuker> hm, read_frame_internal() seems to fail and prevents the decoder from probing for size [16:31:43 CET] <RiCON> seems like the same issue that was happening with vtt/srt [16:37:10 CET] <cone-551> ffmpeg 03Ganesh Ajjanagadde 07master:93afb338a405: lavfi/formats: fix segfault when allocation fails [16:37:11 CET] <cone-551> ffmpeg 03Ganesh Ajjanagadde 07master:a0ddebfedf46: lavc/nellymoserdec: replace pow by exp2 [16:37:12 CET] <cone-551> ffmpeg 03Ganesh Ajjanagadde 07master:df679f12649d: lavc/dcaenc: avoid wasteful cos calls [16:37:13 CET] <cone-551> ffmpeg 03Ganesh Ajjanagadde 07master:b8e5b1d78682: lavc/mdct_template: use lrint instead of floor hack [18:39:06 CET] <durandal_1707> BBB: one of variables is increased by 1 every time and than & operation is done how this can be done in simd? [18:44:00 CET] <BBB> sayagain? [18:44:43 CET] <BBB> can you write the c code? [18:44:47 CET] <BBB> durandal_1707: ^ [18:46:30 CET] <durandal_1707> its in sofalizer on ml, see convolute function [18:47:17 CET] <durandal_1707> its IR with 512 coefficients for each channel and sample [18:48:37 CET] <BBB> wr = (wr + 1) & modulo; /* update ringbuffer write position */ ? [18:48:44 CET] <Daemon404> g 42 [18:49:40 CET] <BBB> I would just use padd{w,d} followed by pand [18:49:45 CET] <BBB> but you dont seem to like that? [18:50:24 CET] <BBB> Im not sure I understand what the algorithm does [18:53:04 CET] <durandal_1707> BBB: no, the deepest for loop [18:53:18 CET] <BBB> *dst += *(buffer[l] + ((read++) & modulo)) * *(temp_ir++); ? [18:54:05 CET] <BBB> I dont think I understand the algorithm, I need to udnerstand the ago to give useful simd suggestions [18:54:33 CET] <BBB> is this for channel repositioning? [19:08:57 CET] <durandal_1707> BBB: its 5.1 to stereo for example [19:09:15 CET] <BBB> so why not use swresample for that? [19:09:26 CET] <BBB> I thought that was the whole point of swr [19:09:50 CET] <durandal_1707> This is not simple mixing [19:10:12 CET] <durandal_1707> uses delayed samples [19:10:56 CET] <BBB> so lets say I have float fivepointone[n][6] [19:11:06 CET] <BBB> how does this algorithm convert it to float twopointzero[n][2]? [19:11:56 CET] <BBB> because this code is like gibberish to me [19:12:57 CET] <durandal_1707> it works [19:13:26 CET] <BBB> Ive written code when I was drunk, and it works [19:13:26 CET] <durandal_1707> makes surround experience with headphones [19:14:16 CET] <durandal_1707> its delayed samples are stored in ringbuffer [19:14:39 CET] <BBB> delayed, you mean for (int n = 0; n < size; n++) { read sample n-2, n-1, n? }? [19:14:48 CET] <BBB> (per channel) [19:14:51 CET] <BBB> or something else? [19:15:32 CET] <durandal_1707> its just prev encountered samples [19:15:34 CET] <khali> is there any way to force non-direct mode of a filter? [19:16:08 CET] <durandal_1707> khali: who are you? [19:17:12 CET] <durandal_1707> what you mean by direct mode? [19:17:33 CET] <khali> I am the maintainer of the vf_delogo filter, is that answer your question [19:17:47 CET] <khali> or I am a 37 yo man living in France [19:18:11 CET] <khali> or am I a kernel developer working on ffmpeg on my scarce spare time [19:19:02 CET] <BBB> so it seems to me that you have src[][6], you reorder than into buffer[ring_size][6] [19:19:03 CET] <BBB> right? [19:19:38 CET] <khali> (and my real name is Jean Delvare, if it matters) [19:20:03 CET] <durandal_1707> khali: sorry, first time encounter [19:20:14 CET] <durandal_1707> on irc [19:20:17 CET] <BBB> and then you use the dotproduct of ir * buffer for the output [19:20:24 CET] <BBB> (per channel) [19:20:25 CET] <BBB> is that right? [19:20:46 CET] <durandal_1707> yes, afaik [19:22:15 CET] <khali> durandal_1707: if av_frame_is_writable(in), the filter will just write to the buffer, this is direct mode [19:22:25 CET] <BBB> durandal_1707: ok& so how do you want to simd this? [19:22:30 CET] <durandal_1707> khali: There is aperm filter, if you talking about that [19:23:21 CET] <durandal_1707> BBB: only this bastard of dot product [19:23:34 CET] <khali> durandal_1707: looks promising, thanks [19:23:39 CET] <BBB> that may be harder than it looks ... [19:23:46 CET] <khali> I suspect there's a bug in non-direct case, will try to reproduce it [19:23:46 CET] <BBB> why not simd by doing N samples at once? [19:24:40 CET] <BBB> Im also not sure I understand why the reordering is necessary [19:24:49 CET] <BBB> if its just to have one instead of n_channels ir tables, I dont think its worth it [19:25:02 CET] <BBB> reordering in simd will be a bloody pain, whereas integrating it into ir order is free [19:26:52 CET] <durandal_1707> it because ffmpeg channel layouts is not used in reality [19:27:01 CET] <durandal_1707> are* [19:27:25 CET] <durandal_1707> lfe channel is not last one [19:28:49 CET] <BBB> but cant you work around that by doing lfe separately from the dsp code here? [19:28:56 CET] <BBB> and using 0 as a coeff for the lfe channel for the rest? [21:38:23 CET] <cone-281> ffmpeg 03Lou Logan 07master:5f55be700568: doc/issue_tracker.txt: various updates [21:54:14 CET] <BBB> does anyone here have experience with amazon EC2 instances? [totally off-topic] [21:54:38 CET] <JEEB> I had a free one for a while [21:54:40 CET] <JEEB> never did the APIs [21:57:48 CET] <BBB> JEEB: so, can the free one run ffmpeg commands? [21:58:05 CET] <BBB> like does it have stuff like disk space, does it work for stuff taking 100% of one cpu, etc? [21:58:52 CET] <JEEB> it should. it's limited as hell but it should [22:05:19 CET] <Plorkyeran> there's a non-persistent local disk [22:05:44 CET] <JEEB> oh, they switched to such? [22:05:51 CET] <JEEB> I remember at one point it was persistent [22:06:08 CET] <JEEB> since people ran irssi instances with logging on them [22:06:15 CET] <Plorkyeran> and you can use as much cpu as you can as they throttle you automatically rather than sending angry emails [22:06:25 CET] <JEEB> yeah [22:06:25 CET] <Plorkyeran> no, it never has been persistent [22:06:29 CET] <JEEB> oh [22:06:35 CET] <Plorkyeran> but that doesn't stop people from treating them as if they are [22:06:46 CET] <JEEB> true that [22:06:51 CET] <Plorkyeran> you can attach EBS volumes that look like a normal filesystem though [22:07:07 CET] <Plorkyeran> I just dont' think the free plan includes that [22:07:28 CET] <Plorkyeran> it'd be cents per month for just some irc logs and config files [22:15:04 CET] <BBB> EBS is not included? [22:15:14 CET] <BBB> or local non-persistent disk is not included? [22:15:27 CET] <BBB> I guess Im hoping someone has done this already [22:15:37 CET] <BBB> I want some 20-or-so instances running ffmpeg during work hours [22:15:47 CET] <BBB> just various invocations that I need (encodes) [22:16:30 CET] <BBB> if I do that fulltime, 24hr/day*0.067$/hr*30days/month*20 =~ $1000 [22:16:33 CET] <BBB> thats a little much [22:16:40 CET] <BBB> so Im wondering if I can use the t2.micro instances [22:16:44 CET] <BBB> theyre like 1 ct/hr [22:16:54 CET] <BBB> but I need disk, for obvious reasons [22:19:33 CET] <Plorkyeran> EBS is something like 10 cents per GB-month [22:19:48 CET] <Plorkyeran> if you need persistent storage you'd have to pay for 100% uptime of that [22:20:00 CET] <Plorkyeran> but you only need to pay for the CPU for the instances when you're actually using them [22:20:20 CET] <Plorkyeran> (if you spin them up and shut them down rather than jsut leaving them running...) [22:20:52 CET] <Plorkyeran> the transient storage on the instances sticks around as long as the instance does and should be fine for scratch space [22:21:52 CET] <BBB> this is for t2.micro? or for m3? [22:22:36 CET] <Daemon404> theis is for all instances [22:22:38 CET] <Daemon404> this* [22:23:16 CET] <BBB> I guess I need to read a guide on things like how to programmatically start/stop instances? [22:23:33 CET] <BBB> I guess what I want is a script that starts 20 instances and then runs some stuff on each of them and then shuts them down [22:25:22 CET] <Daemon404> they have libraries for most languages [22:25:38 CET] <BBB> https://aws.amazon.com/sdk-for-python/ \o/ [22:47:12 CET] <TD-Linux> BBB, I run a bunch of video codec tests on EC2 at https://arewecompressedyet.com/ [22:47:34 CET] <BBB> you own that domain? [22:47:38 CET] <TD-Linux> yes [22:48:11 CET] <TD-Linux> I use a bunch of c4.8xlarge instances [22:48:18 CET] <BBB> cool [22:48:27 CET] <BBB> CanIHaveYourScriptsPlease? [22:48:30 CET] <BBB> (are they public?) [22:49:06 CET] <TD-Linux> yes, this is the website https://github.com/tdaede/awcy [22:49:20 CET] <TD-Linux> and this is the "backend" https://github.com/tdaede/rd_tool [22:50:13 CET] <TD-Linux> it's not very pretty btw, it's still under continuous development, and web things in general are not my specialty :) [22:51:06 CET] <BBB> its probably 1000x better than what I can hack up in some short amount of time :-p [22:52:39 CET] <durandal_1707> michaelni: do you know a way how to reduce say 128 tap filter to 32 one? [22:53:07 CET] <TD-Linux> see awsremote.py in rd_tool for the part that talks to AWS. [22:54:53 CET] <cone-281> ffmpeg 03Michael Niedermayer 07master:676a93f2d86d: avformat: Add av_program_add_stream_index() [22:54:54 CET] <cone-281> ffmpeg 03Michael Niedermayer 07master:8f948b6244e4: ffmpeg: Add basic support to mux multiple programs [22:54:55 CET] <cone-281> ffmpeg 03Michael Niedermayer 07master:edfc835a56a5: avformat/mpegtsenc: Add basic multi program support [22:57:34 CET] <michaelni> durandal_1707, hmm, some filters might be factorizable in several filter steps, also normal "convolutional" filters can be done in frequency domain by simple complex multiplication, so for sufficiently long filters a FFT + IFFT may be faster [22:58:04 CET] <michaelni> s/may/is/ [22:58:39 CET] <BBB> TD-Linux: thanks!!! [23:00:31 CET] <BBB> TD-Linux: so & maybe stupid question, but what happens to running instances that dont do anything? [23:01:17 CET] <TD-Linux> BBB, you get billed for uptime on EC2 regardless of CPU usage [23:01:36 CET] <TD-Linux> it's in per hour increments. so you have to manually shut off idle instances [23:01:59 CET] <BBB> so if I run 20 machines for 10 minutes each, it costs me 20*1hr? [23:02:00 CET] <BtbN> Why not just shut them down automaticaly once they're done? [23:02:35 CET] <TD-Linux> BtbN, well I mean it is automatic on AWCY. I meant your scripts have to do it somewhere [23:02:40 CET] <BtbN> "Each time you transition an instance from stopped to started, we charge a full instance hour, even if transitions happen multiple times within a single hour." [23:02:41 CET] <TD-Linux> BBB, yup. [23:02:51 CET] <BBB> hum& sucky [23:03:09 CET] <TD-Linux> the startup time is also kind of long. they are full VMs [23:03:24 CET] <BBB> I see, thats why you keep them running& ok [23:03:34 CET] <BBB> I probably need to design this a bit so it wont bankrupt me immediately [23:03:43 CET] <TD-Linux> well right now if there are no jobs queued they turn off after a 5 minute delay [23:03:48 CET] <BBB> anyway Ill start with your script, that looks pretty good [23:04:18 CET] <TD-Linux> google compute engine has 1 minute billing increments which looks interesting, but I've never tried it [23:05:02 CET] <durandal_1707> michaelni: there is frequency domain mode but its marked as medium quality [23:05:22 CET] <durandal_1707> how much faster it can be? [23:09:52 CET] <michaelni> something like O(n*m) vs. O(n*logn + m*logm) [23:51:21 CET] <kierank> michaelni: do you know what max_depth does in GET_RL_VLC [00:00:00 CET] --- Sat Dec 12 2015