[Ffmpeg-devel-irc] ffmpeg-devel.log.20190109

[burek]

[00:08:26 CET] <diggidydave> Hello all, I'm new to development using ffmpeg libs (and, generally, to av stuff), and I'm trying to get my bearings, so to speak...
[00:08:52 CET] <diggidydave> I have a lot of cameras, each of which will sometimes be sending h264 encoded video via RTP to a dedicated endpoint.  I was thinking I will have a broker that sets up the endpoint and ultimately causes a service to listen for that RTP data, and then it will do some other things with that stream.  Right now, I'm trying to figure out the "entry point" into ffmpeg for that RTP data.
[00:09:49 CET] <diggidydave> I have raw RTP packets coming in, and I can get the raw payload out of those of course.  Can anyone point me towards any good examples of how to ingest RTP from UDP packets?
[00:16:24 CET] <Compn> are you developing a patch for ffmpeg
[00:16:26 CET] <Compn> or for yourself ?
[00:17:07 CET] <diggidydave> this is for myself, am I in the wrong channel?
[00:17:44 CET] <diggidydave> oh, man I didn't see the header up there. I saw the #ffmpeg message which seemed to suggest this channel, but I didn't see this one pointing my back there. :-) my aopologies
[00:29:55 CET] <Compn> tno worries
[16:07:25 CET] <durandal_1707> how to do horizontall add with avx/ymm?
[16:17:51 CET] <atomnuker> of floats? 4 shuffles and 4 adds or 2 haddps
[16:18:10 CET] <jamrial> durandal_1707: HADDD macro
[16:19:19 CET] <jamrial> haddps and phaddd are very slow, so only use them when you want to add two regs horizontally the exact way the instructions handle it
[16:20:57 CET] <atomnuker> isn't HADDD integer only?
[16:27:50 CET] <jamrial> yeah
[16:30:58 CET] <jamrial> if he wants to add horizontally all 8 floats in an ymm reg, it's vextractf128, addps, movhlps, addps, shufps, addss, using a tmp reg
[16:56:55 CET] <jamrial> durandal_1707: see scalarproduct_float in libavutil/x86/float_dsp.asm
[16:57:19 CET] <jamrial> after the loop is the horizontal add, including some x86_32 specific bits
[18:14:38 CET] <durandal_1707> anybody see where i failed to rewrite this: https://github.com/richardpl/FFmpeg/commit/a7a594ec951f8a592359ab68043cde087335338f ?
[18:16:54 CET] <BtbN> durandal_1707, first thing that springs to my eye is that v is reset between the two calls you have now, while previously it keps growing over the two for-loops
[18:18:30 CET] <durandal_1707> BtbN: i use cache + S
[18:28:12 CET] <durandal_1707> i found it
[18:29:20 CET] <BtbN> What is it?
[18:30:02 CET] <durandal_1707> v was not set, it should be 2 * S later in code...
[21:20:02 CET] <cone-299> ffmpeg 03Paul B Mahol 07master:ed3b64402ef7: avfilter/af_anlmdn: ignore too small values
[21:39:13 CET] <llogan> michaelni: got a rough idea when you're going to release 4.1.1?
[23:13:15 CET] <thebombzen> I sent a patch a while ago to the mailing list but it had a problem, so I fixed it and sent v2 but never got a response. is it okay to bump it?
[23:15:43 CET] <durandal_1707> yes
[23:29:39 CET] <thebombzen> I ended up rerunning make fate and reformatting the patch and resending with git send-email but the patch contents should be the same
[00:00:00 CET] --- Thu Jan 10 2019