[Ffmpeg-devel-irc] ffmpeg-devel.log.20191127

[burek]

[00:04:47 CET] <cone-357> ffmpeg 03Carl Eugen Hoyos 07master:a2fbdc6898fc: lavc/qsvenc: Fix compilation for some build environments.
[11:37:58 CET] <Dmitri_Ovch> jamrial: "the avctx->refs check should probably be > 0 instead", i checked, in docs  "AMF_VIDEO_ENCODER_MAX_NUM_REFRAMES 0..16 Maximum number of reference frames", 0 is also valid value. Default value for encoder is 1. in encoder defaults: { "refs",       "-1"  }. Looks like "avctx->refs != -1" is correct check.
[11:39:30 CET] <mkver> jamrial isn't online right now.
[11:48:16 CET] <Dmitri_Ovch> mkver:Thank you, not observed. Then I'll send that message again later.
[12:05:32 CET] <cone-432> ffmpeg 03Steven Liu 07master:afee801276d5: avformat/hls: correct grammatical errors of m3u8_hold_counters option
[19:33:19 CET] <j-b> If people needed an extra datapoint about intrinsics vs .S/.asm files: the patch on h_scale gets +34% in .S, while it was 15% in intrinsics :)
[19:42:32 CET] <ubitux> j-b: and it's not even its finale form :))
[19:43:04 CET] <j-b> ubitux: true :)
[19:44:42 CET] <durandal_1707> intrinsics rules, asm sucks
[19:47:49 CET] <j-b> stop trolling
[19:59:33 CET] <durandal_1707> j-b: or?
[20:04:35 CET] <j-b> durandal_1707: or I'll be pissed.
[20:18:40 CET] <durandal_1707> ubitux: gonna apply patches
[21:02:41 CET] <ubitux> durandal_1707: which ones?
[21:14:58 CET] <durandal_1707> ubitux: EVERYTHING
[21:15:26 CET] <durandal_1707> nlmeans
[21:15:43 CET] <durandal_1707> >8 bit
[21:16:28 CET] <ubitux> durandal_1707: do you mind waiting a few days?
[21:16:36 CET] <ubitux> i wasn't aware of these patches
[21:16:51 CET] <ubitux> do not hesitate to ping me on irc, i'm not following the ml these days
[21:30:11 CET] <durandal_1707> if intrinsic is so bad that why everybody still use it?
[21:30:25 CET] <BtbN> Because it's less of a hassle
[21:32:18 CET] <Lynne> everyone uses them because everyone else uses them
[21:32:59 CET] <BtbN> You don't have to deal with an extra tool in your build, you can use more familiar constructs to layout the code, ...
[21:33:45 CET] <JEEB> "you can just stick asm in the middle of C"
[21:33:47 CET] <JEEB> is the primary reason
[21:33:56 CET] <JEEB> s/asm/SIMD/
[21:38:00 CET] <nevcairiel> mostly "because its easier"
[21:56:13 CET] <ubitux> pure asm also usually involves a hassle wrt build, runtime integration, and takes some maturity on the available header helper to not mess up too much
[21:56:38 CET] <ubitux> but all of this have been adressed in ffmpeg for quite some time so there isn't much of a reason to stick with intrinsics
[21:57:20 CET] <ubitux> (also: "everyone does c++ and javascript" is not a good argument)
[21:58:29 CET] <ubitux> JEEB: yeah this is actually the only remaining drawback
[21:58:38 CET] <yzhu132> Hi, I'm looking into contributing in helping with FFmpeg. Preferably with optimization and/or code clean up with the arm/aarch64 build.
[22:00:05 CET] <yzhu132> Could anyone point me in a direction that needs some help?
[22:00:23 CET] <cehoyos> There are both arm and aarch64 asm optimizations, feel free to add more
[22:00:32 CET] <cehoyos> (Don't use intrinsics)
[22:04:23 CET] <yzhu132> Got it, because the compiler already optimizes for that
[22:11:25 CET] <jdarnley> huh
[22:11:59 CET] <jdarnley> we say don't use intrinsics because we prefer assembly
[22:18:54 CET] <durandal_1707> anybody know how to trace executed functions via debugger ? maybe with radare2 ?
[22:31:13 CET] <durandal_1707> braw apparently uses tiles, that's all what i know so far :(
[23:49:12 CET] <durandal_1707> i gonna push 0day code exploits!
[00:00:00 CET] --- Thu Nov 28 2019