[Ffmpeg-devel-irc] ffmpeg-devel.log.20190418

[burek]

[09:41:24 CEST] <cone-591> ffmpeg 03Sam John 07master:88325f4b34dd: Updated documentation for libaom encoder options.
[13:41:18 CEST] <cone-915> ffmpeg 03Carl Eugen Hoyos 07master:09f8b21deb16: doc/encoders: Fix libvpx option name arnr-maxframes.
[14:45:43 CEST] <cehoyos> Hi! vaapi works fine here, can anybody explain to me what I have to do to use qsv? I did install libmfx, what else is needed? Where do I get the necessary shared library?
[14:47:00 CEST] <BtbN> iirc you need a special kernel and the Intel Media SDK for Linux from their Developer Resources
[14:47:15 CEST] <BtbN> There isn't really any reason to use QSV on Linux
[14:47:27 CEST] <cehoyos> So on Linux qsv never works without a patched kernel? So I don't want to use it, correct?
[14:47:48 CEST] <BtbN> libva just works and offers almost the exact same features
[14:47:59 CEST] <cehoyos> On Windows, qsv is the only hardware acceleration (with Intel) for encoding, right?
[14:48:04 CEST] <BtbN> yes
[14:48:12 CEST] <BtbN> Well, and d3d11va for decoding
[14:48:22 CEST] <cehoyos> (And dxva2)
[14:48:30 CEST] <BtbN> Same thing but kinda dead
[14:48:35 CEST] <cehoyos> And do I need anything apart from libmfx on Windows?
[14:48:43 CEST] <cehoyos> It worked when I tested it yesterday.
[14:48:53 CEST] <cehoyos> Patching the kernel maybe?
[14:48:56 CEST] <BtbN> Everything comes with the driver. If you use libmfx_dispatch, that's all you need.
[14:49:05 CEST] <cehoyos> Ok, will test, thank you!
[14:51:08 CEST] <BtbN> https://github.com/Intel-Media-SDK/MediaSDK also seems to be an official open source dispatcher library for both OS
[14:52:54 CEST] <cehoyos> But several docs suggest to use libmfx
[14:53:27 CEST] <cehoyos> Once you find out that the CMake.txt file is just a trap, it is not difficult to install (at least on Linux, I wonder how to autobuild on Windows)
[14:58:21 CEST] <nevcairiel> i dont think the k ernel patches are needed anylonger on recent drivers
[14:59:32 CEST] <BtbN> You crosscompile it from Linux or use MinGW I guess
[14:59:59 CEST] <cehoyos> I use the subsystem now, works really great imo.
[15:00:08 CEST] <nevcairiel> the official mediasdk from intel has a visual studio solution, you can also use that if you are not into mingw
[15:00:13 CEST] <cehoyos> nevcairiel: So what is needed?
[15:00:22 CEST] <cehoyos> On Linux, I mean.
[15:00:37 CEST] <cehoyos> I don't have installed VS yet on this system
[15:00:45 CEST] <cehoyos> Want to avoid it a little longer;-)
[15:04:44 CEST] <nevcairiel> apparently qsv on linux is still as undocumented as ever
[15:04:56 CEST] <nevcairiel> maybe jkqxz has some hints
[15:05:39 CEST] <nevcairiel> a more portable encoding solution on linux is using vaapi
[15:14:30 CEST] <JEEB> ya
[15:14:43 CEST] <JEEB> thankfully intel added support for it as well, instead of trying to keep to qsv only
[15:14:50 CEST] <JEEB> windows is the thing where you get sad
[15:15:34 CEST] <BtbN> QSV works just fine on Windows though, as it comes with the driver you install anyway
[15:15:43 CEST] <JEEB> yup
[15:15:45 CEST] <BtbN> So the thin dispatcher is all you need
[15:19:05 CEST] <cehoyos> Ok, will test
[15:19:24 CEST] <cehoyos> Since there are many qsv reports that nobody ever tested, I would have liked to also install it on Linux though.
[15:21:06 CEST] <BtbN> That chain of github stuff is what you need to figure out then
[15:21:15 CEST] <cehoyos> I am building it right now.
[15:21:24 CEST] <cehoyos> I wonder how to install: Won't it overwrite libmfx
[15:21:26 CEST] <cehoyos> ?
[15:21:41 CEST] <BtbN> I'd just install it somewhere in your homedir
[15:21:50 CEST] <BtbN> never even giving it the change to install anything as root
[15:24:11 CEST] <cehoyos> Ok, it loads the library now.
[15:24:30 CEST] <cehoyos> It seems to be implemented so smart that it only works with additional parameters;-)
[15:24:48 CEST] <wbs> nevcairiel: re patchwork, you're right with how it associates usernames; it stores a email->realname mapping, with the first sender name for each email address that it has gotten. and then, as you guessed, rewrites patches with a realname based on only the email address in the mail
[15:31:38 CEST] <nevcairiel> yeah it seemed logical to do that
[15:31:48 CEST] <nevcairiel> multiple people dont usually share mail addresses
[15:32:08 CEST] <BtbN> So if you Typo your name the first time you send a patch, you're stuck with it forever?
[15:32:57 CEST] <nevcairiel> presumably it updates the name
[15:33:05 CEST] <nevcairiel> but there is still only one global record for every mail address
[15:33:54 CEST] <nevcairiel> ultimately that rewrite script should really  learn to insert From: lines into the pathc c ontent itself to fix it
[15:34:18 CEST] <nevcairiel> .. and my spacebar is going crazy again
[21:10:09 CEST] <wbs> BtbN: true. if you typo your name in the first patch, you need to ask the patchwork admin to reset it for you
[22:04:25 CEST] <cone-741> ffmpeg 03Lou Logan 07master:d7e8e62797d8: doc/mailing-list-faq: add DMARC munging
[23:37:04 CEST] <mkver> I just found out that I am affected by the mailing list name mangling.
[23:37:45 CEST] <mkver> In fact, I am the author of the majority of the commits by "Oliver Collyer via ffmpeg-devel" on patchwork (https://patchwork.ffmpeg.org/project/ffmpeg/list/?submitter=776).
[23:38:35 CEST] <mkver> But googlemail is just another name for gmail and plenty of people use gmail without issue. Does anyone have a clue why I could be different?
[23:42:38 CEST] <durandal_1707> dunno, if pushed patch was wrongly attributed raise it on mailing list
[23:49:57 CEST] <JEEB> we right now seem to have four mistakenly committed patches in the repo find'able with `git log |grep "via ffmpeg-devel"`
[23:50:29 CEST] <durandal_1707> only four? what about Elvis ones?
[23:50:42 CEST] <JEEB> this is specific to the "via ffmpeg-devel" stuff :P
[23:56:30 CEST] <JEEB> mkver: it seems like those who committed ffmpeg-devel at ffmpeg.org e-mail addressed people at least downloaded the patches from the mailing list, and not patchwork. since the author name seems to be correct
[23:57:03 CEST] <JEEB> so while patchwork doing not-so-nice things is bad, it doesn't seem to have been affected wrong name being put onto commits in the repo
[23:59:13 CEST] <JEEB> or at least that's how it looks because there's no Oliver Collyer there, and just one occurence of each name
[23:59:42 CEST] <mkver> Is the reason that e.g. Jamrial is not affected that he commits his stuff himself?
[00:00:00 CEST] --- Fri Apr 19 2019