burek021 at gmail.com
Sun Nov 11 03:05:04 EET 2018
[02:02:31 CET] <philipl> BtbN: https://github.com/philipl/nv-codec-headers/commit/7b88dd0250e3cc8055e9e63922d1abfcd2990c11
[02:02:56 CET] <BtbN> So it was the syncornization after all?
[02:03:27 CET] <philipl> No
[02:03:38 CET] <philipl> But I got synchronization working and that will be necessary
[02:04:09 CET] <philipl> It's not really possible to do a pool strategy with images like I did with the buffer implementation
[02:04:26 CET] <philipl> so I need synchronization to make sure we don't copy the next frame until the previous one has been read out of the image.
[02:04:46 CET] <BtbN> I'll have a look tomorrow, way too late right now
[02:04:52 CET] <philipl> Sure. I'm surprised you're up
[02:05:51 CET] <BtbN> Well, there was a certain chess match.
[02:06:04 CET] <BtbN> That ended up taking almost 8 hours.
[02:47:52 CET] <cone-221> ffmpeg 03James Zern 07master:6d3890ee34fd: libvpxenc,vp9: add enable-tpl option
[02:59:13 CET] <cone-221> ffmpeg 03James Zern 07master:55e021f39b02: libvpxenc: extend auto-alt-ref range
[10:53:12 CET] <BtbN> philipl, lgtm. I'll wait a bit until doing another round of releases, in case something else comes up, which has happened surprisingly frequently lately.
[16:46:39 CET] <philipl> BtbN: thanks. Would you take the change with the kernel launch API stuff? It's not actually very much.
[16:47:46 CET] <philipl> https://github.com/philipl/nv-codec-headers/commit/692ef24e3242286a0b6bfa4db8469a4aab4e0bb7
[16:47:49 CET] <philipl> for reference
[16:49:25 CET] <BtbN> Well, it won't make it non non-free
[16:49:47 CET] <philipl> I know.
[16:50:04 CET] <philipl> But it at least scopes the problem to the ptx compilation
[16:53:36 CET] <BtbN> hm, I guess it also makes the binary at least runable without the nvidia driver installed, which might be a plus to some internal deployments.
[16:58:13 CET] <philipl> Yeah
[17:07:56 CET] <BtbN> Go ahead then
[17:26:17 CET] <philipl> Ok. I'll push these but I won't tag for release. Yeah?
[17:55:33 CET] <philipl> BtbN: https://opensource.stackexchange.com/questions/4304/redistributing-lgpl-2-1-code-compiled-by-a-proprietary-compiler
[17:55:43 CET] <philipl> not much clarity
[17:56:54 CET] <BtbN> Reads pretty clear to me
[17:57:09 CET] <BtbN> Unless we get it to compile and work with clang, it's non-free.
[17:57:18 CET] <philipl> I investigated that.
[17:58:17 CET] <BtbN> I wonder how things would look if the ptx would be commited. But I doubt that'd be accepted.
[17:58:40 CET] <philipl> There are missing language features, and, I think, the thing that sinks the whole ship, is that you end up linking in a copy of cudart to provide the cuda api implementations.
[17:58:51 CET] <philipl> It's not like there's a libc equivalent handing out on the GPU side.
[17:59:37 CET] <philipl> The other thing I was wondering about was what if the ptx was loaded as a separate file off the filesystem. That means you can rebuild the lgpl part any time you want however you want. and the ptx runs on the card (so separate address space, etc)
[18:00:00 CET] <philipl> Arguably, the ptx being inside the library is aggregation and not linking.
[18:00:10 CET] <philipl> but it makese the replacement requirement much harder
[18:01:10 CET] <philipl> So you could say "Here is ffmpeg". "If you download this ptx and place it in the right place, the filter works"
[18:02:35 CET] <BtbN> I don't think there is much arguing here. It's just non-free
[18:02:46 CET] <philipl> But there's an important point there.
[18:03:01 CET] <philipl> If you do what I just said what about the ffmpeg build is non-free if the ptx isn't in there?
[18:03:22 CET] <philipl> The argument then is that the ptx is non-free and how do you get that
[18:03:29 CET] <BtbN> It seems like cheap trickery to me
[18:03:51 CET] <BtbN> But you'll have to discuss that with an fsf lawyer
[18:04:18 CET] <philipl> It's not cheap. If you say the ptx is non-free and non-redistributable, the functionality is just as unavailable as it was before.
[18:05:10 CET] <philipl> I don't have have any commerical motivtions behind any of this; I'm just interested in maximising availability to end users.
[18:05:33 CET] <BtbN> I'm not against that at all, but that solution seems as anti-end-user as it gets.
[18:05:56 CET] <BtbN> (And would also be an API break)
[18:06:26 CET] <philipl> Sure.
[18:06:31 CET] <philipl> I suspect this is all futile anyway.
[18:06:52 CET] <philipl> It's really the proprietary compiler question. Everything else is rearranging deck chairs on the titanic.
[18:16:07 CET] <cone-517> ffmpeg 03bnnm 07master:27e114b4511b: avcodec/wmaprodec: improve XMA missing samples
[18:18:19 CET] <uau> hevc frameskip still seems broken (https://github.com/mpv-player/mpv/issues/6218); there's also a patch from Mark Wu on the mailing list without replies (from 2018-10-23)
[18:18:37 CET] <uau> could someone fix/apply that?
[18:20:14 CET] <philipl> BtbN: And now the pix fmt question. The easy/lazy thing is to punt on pix fmts for now and just do 444p16. Otherwise it's organising a vote?!?
[18:20:33 CET] <BtbN> Seems like it.
[18:21:56 CET] <philipl> Even though I'm in favour of the new pix fmts, I'd feel better if I could find some discussion of the industry position on this (like there is for P010). But I can't find any.
[18:23:24 CET] <philipl> We know the nvdec docs won't help - they only talk about P016 today and so I'm sure they'll just say 444p16 when they are updated.
[18:29:56 CET] <jamrial> uau: you don't have push access to the repo?
[18:32:06 CET] <uau> no
[18:32:50 CET] <jamrial> ok, will push in a bit
[18:38:12 CET] <cone-517> ffmpeg 03Mark Wu 07master:10bc4c3a7df7: avcodec/hevcdec: fix non-ref frame judgement
[19:05:48 CET] <j-b> 'morning
[19:14:19 CET] <durandal_1707> do you type in every channel that?
[19:33:00 CET] <atomnuker> its 18:00, its not even morning
[19:33:28 CET] <durandal_1707> he sleep too much
[19:33:37 CET] <atomnuker> worse, imagine if it was morning, then you'd be reminded its morning
[19:36:05 CET] <BradleyS> good morning :)
[19:36:33 CET] <cone-517> ffmpeg 03Mark Wu 07release/4.0:263b3ad40749: avcodec/hevcdec: fix non-ref frame judgement
[19:36:35 CET] <cone-517> ffmpeg 03Mark Wu 07release/4.1:11dff170ef79: avcodec/hevcdec: fix non-ref frame judgement
[19:47:22 CET] <durandal_1707> michaelni: when you will resign as leader for real?
[19:54:24 CET] <jamrial> durandal_1707: do something else if you're bored instead of stirring shit, please
[19:55:42 CET] <kierank> durandal_1707: why you troll michael
[20:18:06 CET] <durandal_1707> kierank: because he is still leader
[20:18:18 CET] <kierank> i don't disagree but why raise it now?
[20:21:09 CET] <durandal_1707> he is blocking CoC
[20:22:22 CET] <kierank> not the only one
[20:22:32 CET] <kierank> you got to follow vote process or whatever "rules" exist
[20:24:01 CET] <durandal_1707> kierank: no, we must use force now!
[20:24:15 CET] <kierank> should have said that at vdd
[20:26:28 CET] <BBB> yeah, we had this discussion at VDD
[20:26:33 CET] <BBB> I told all you guys to be there
[20:26:37 CET] <BBB> you werent there
[20:26:38 CET] <BBB> sorry
[20:26:50 CET] <BBB> I believe thilo said he would move it forward
[20:27:07 CET] <durandal_1707> jamrial: i got impression, that you are not very fond of me?
[20:27:34 CET] <BBB> I think jamrial wishes you could be more positive
[20:27:40 CET] <BBB> but I wont put words in his mouth
[20:27:51 CET] <jamrial> durandal_1707: i am fond of you, but you really seem to like to troll people when things are quiet
[20:41:27 CET] <durandal_1707> kierank: ok, i gonna do Carl statue then
[20:43:44 CET] <kierank> durandal_1707: you can buy bricks from elon musk boring company
[20:51:33 CET] <durandal_1707> kierank: i thought to make it from space frozen trash
[22:16:42 CET] <cone-517> ffmpeg 03Paul B Mahol 07master:0bc2326b90a7: avfilter/vf_lut3d: add cosine 1D interpolation
[22:16:43 CET] <cone-517> ffmpeg 03Paul B Mahol 07master:c47b48c0a629: avfilter/vf_lut3d: add spline 1D interpolation
[22:22:46 CET] <durandal_1707> 9 more commits
[00:00:00 CET] --- Sun Nov 11 2018
More information about the Ffmpeg-devel-irc