[Ffmpeg-devel-irc] ffmpeg-devel.log.20180918

[burek]

[00:21:16 CEST] <cone-015> ffmpeg 03hwren 07master:6f23c327c5a0: lavc/libdavs2: remove unused header files
[00:21:16 CEST] <cone-015> ffmpeg 03hwren 07master:c0e9824c0f25: lavc/libdavs2: fix error return codes
[00:21:16 CEST] <cone-015> ffmpeg 03hwren 07master:bc1097a2bf4f: lavc/libxavs2: remove nonexistent parameters
[00:56:30 CEST] <cone-015> ffmpeg 03Sergey Lavrushkin 07master:bd10c1e9a86c: libavfilter: Removes stored DNN models. Adds support for native backend model file format in tf backend.
[09:57:59 CEST] <liyou> what is swscale function?  it is so hard for me to understand its logic.
[11:05:09 CEST] <J_Darnley> I think I've seen a function or a group of functions somewhere in ffmpeg that do aligned/unaligned pointer detection for assembly.
[11:06:27 CEST] <J_Darnley> I can't remember where it was.  Maybe in swresample or avresample.
[11:07:17 CEST] <J_Darnley> Does anyone know what I'm talking about?
[11:08:39 CEST] <durandal_1707> J_Darnley: i only see it in zimg
[11:09:02 CEST] <durandal_1707> for other cases i do not remmember
[11:10:34 CEST] <nevcairiel> J_Darnley: swresample rematrix.asm has a and u variants of functions, if thats what you want
[11:11:44 CEST] <nevcairiel> and a quick test at the top of the a function for alignment,  that switches to the u if its not fullfilled
[11:29:58 CEST] <J_Darnley> nevcairiel yes I think that's what I want
[11:42:15 CEST] <Gramner> just use unaligned moves
[11:43:09 CEST] <Gramner> if you're thinking about just having mova instead of movu that is, since there's no penalty for using unaligned loads compared to aligned ones as long as the data is aligned
[11:43:45 CEST] <Gramner> as far as non-ancient machines at least, dunno about various historical artifacts
[11:44:30 CEST] <atomnuker> I thought there were no unaligned/aligned penalties at all for modern machines
[11:44:47 CEST] <Gramner> there is, it's just a lot smaller than it used to be
[11:45:15 CEST] <Gramner> iirc amd does worse than intel on unaligned also
[11:45:55 CEST] <atomnuker> so with aligned data movu == mova?
[11:45:59 CEST] <Gramner> yes
[11:46:03 CEST] <atomnuker> nice
[11:47:15 CEST] <Gramner> which klind of makes sense when you consider that avx instructions doesn't have alignment restrictions on memory operands
[11:48:16 CEST] <Gramner> so all instructions with memory operands are essentially unaligned loads
[13:02:52 CEST] <Yagiza> Hello!
[13:03:38 CEST] <Yagiza> Do we have a way to specify protocol for muxing when using custom I/O?
[13:08:35 CEST] <JEEB> no, I think at that point it's 100% up to you
[13:08:40 CEST] <JEEB> you handle the protocol
[13:14:31 CEST] <Yagiza> So, I have no way to use SRTP endryption procedures from FFMpeg libraries when using custom I/O?
[13:14:41 CEST] <Yagiza> encryption
[13:35:02 CEST] <Yagiza> And what about decryption?
[14:01:15 CEST] <Yagiza> When I give SDP demuxer an SDP which contains SRTP-related commands, but will specify "sdp_flags=custom_io" it will ignore those commands and will use unencrypted RTP?
[14:02:43 CEST] <Yagiza> It seems we need to fix this issue.
[15:30:21 CEST] <J_Darnley> Gramner: that's good to know.
[15:30:24 CEST] <J_Darnley> I guess I won't bother with oodles of functions, especially ones that'll probably just mov{a,u} the data.
[15:32:32 CEST] <J_Darnley> I do recall 1 in this bunch that manages to do a multiply (I think) straight from the memory.
[15:33:05 CEST] <J_Darnley> But even if I need to be careful with that one I can probably just throw out half the func already.
[16:43:39 CEST] <durandal_1707> i gonna write state of art audio denoiser
[17:28:10 CEST] <BradleyS> durandal_1707: curious, based on... ?
[17:28:44 CEST] <BradleyS> i have a fair amount of experience with commercial ones and would be glad to evaluate samples
[17:44:35 CEST] <durandal_1707> BradleyS: some java code i decompiled
[17:45:11 CEST] <TheAMM> My lord, is that legal?
[18:29:49 CEST] <cone-364> ffmpeg 03Maxym Dmytrychenko 07master:a2041a652264: qsvenc: AV_PIX_FMT_QSV path should release frame
[18:29:50 CEST] <cone-364> ffmpeg 03James Almer 07master:f4cf6ba8c964: Merge commit 'a2041a6522642859ce64af1c618d6fb90a50d4af'
[18:33:46 CEST] <akravchenko188> hello guys, I saw patches with vulcan hwcontext support, but they still are not in repo
[18:33:59 CEST] <akravchenko188> is it going to be published?
[18:38:58 CEST] <atomnuker> yes, I wanted to wait for the drm modifier patchset but I should probably resubmit the patches to address some comments jkqxz had
[18:39:59 CEST] <atomnuker> also I wanted to wait for intel to fix their crap where using proper synchronization or anything it didn't like caused a reset in the gpu
[18:40:30 CEST] <atomnuker> but I think I'll shame them by getting the patches in and letting other people too know their drivers are broken
[18:41:35 CEST] <atomnuker> akravchenko188: are you planning on adding an amd interop with amf(?)
[18:42:41 CEST] <akravchenko188> yes
[18:43:35 CEST] <akravchenko188> I also sent patch with amf colorspace converter/scaler, still on review
[18:45:38 CEST] <akravchenko188> atomnuker: to be clear, you mean dx11 -> opencl interop?
[18:48:17 CEST] <atomnuker> no, I thought you meant a vulkan <-> amf, hence that's why you asked
[18:48:45 CEST] <atomnuker> though there's a vaapi interop already which works well
[19:50:02 CEST] <akravchenko188> ok I see, I will, and I wanted to add amf decoder and Vulcan support to amf converter/scaler and amf encoder
[19:58:53 CEST] <JEEB> was there a way to separate the AVOptions for the protocol and the demuxer?
[19:59:01 CEST] <JEEB> IIRC both protocol and demuxer are in the lavf context
[20:00:42 CEST] <atomnuker> akravchenko188: there's already an amf decoder, isn't there?
[20:20:06 CEST] <akravchenko188> atomnuker: not  yet, since ffmpeg has DXVA decoder, it was enough to have hw pipeline for windows. 
[20:22:11 CEST] <akravchenko188> Now AMD published linux driver with amf library which works with vulkan. now we could implement full hw pipeline on linux using hwcontext_vulkan in FFmpeg
[20:40:05 CEST] <kurosu> I'm disappoint
[20:40:21 CEST] <kurosu> dav1d subtitle should be dav1d is dav1d av1 decoder
[20:41:34 CEST] <kurosu> (or stands for, etc)
[20:48:12 CEST] <durandal_1707> kurosu: where are slides?
[20:48:26 CEST] <kurosu> I'm just referring to videolan.org program
[20:48:47 CEST] <durandal_1707> i want slides and code
[20:49:56 CEST] <kurosu> don't we all
[20:50:40 CEST] <kurosu> but I guess one has to shake BBB's hand first in person during VDD
[20:50:54 CEST] <atomnuker> no choice but to finish the slides at the venue, because you don't know what aspect ratio the screen will be in
[20:51:09 CEST] <BBB> <3
[20:51:50 CEST] <kurosu> you need dav1d to output its presentation to display first
[20:55:21 CEST] <gnafu> kurosu: I actually love that idea.  You could encode each slide as an AV1 frame, then rig up a slide program that calls the dav1d library to decode each slide.
[20:55:37 CEST] <gnafu> slide1.ivf, slide2.ivf, etc.
[20:55:56 CEST] <gnafu> Or of course rig it up to step through frames of a single video.
[20:56:09 CEST] <gnafu> But then you have to rerender the whole slide deck if something changes ;-).
[21:06:40 CEST] <kurosu> also depends if screen content stuff, such as ibc and palette, was implemented
[21:28:00 CEST] <durandal_1707> on ibc there were played AV1 demos with what software?
[21:29:54 CEST] <gnafu> durandal_1707: I know there were some AV1-in-Firefox-using-libaom playback demos.
[21:30:17 CEST] <gnafu> And some live encoding with rav1e (640x480, 15-30 fps).
[21:30:33 CEST] <gnafu> I don't know if dav1d was used or not.
[23:40:12 CEST] <atomnuker> lol, servo has regressed to using gstreamer
[23:40:25 CEST] <atomnuker> what is happening with firefox multimedia people?
[23:41:05 CEST] <atomnuker> it took them many years to realize 709 != 601, they used libstagefright to do demuxing (!) which traumatised them so much they do ALL their own demuxing now
[00:00:00 CEST] --- Wed Sep 19 2018