[FFmpeg-devel-irc] IRC log for 2010-09-05

irc at mansr.com irc at mansr.com
Mon Sep 6 02:00:22 CEST 2010


[04:37:37] <ramiro> Dark_Shikari: ping
[04:38:38] <ramiro> does it make sense that interleaveBytes (in libswscale/rgb2rgb_template.c) is faster in mmx-only code? in mmx2 code it's slower (the only difference is an extra prefetch), and sse2 is just as slow as mmx2.
[04:39:15] <ramiro> oh, no. there's also movntq instead of movq for mmx2
[04:42:05] <Dark_Shikari> be VERY careful how you bench code that uses movntq.
[04:42:40] <Dark_Shikari> It will be very dependent on cache size + resolution + exactly what you're benching + whether or not you're benching what reads the image data after it's converted
[04:42:46] <Dark_Shikari> + whether it's read immediately after, etc
[04:44:48] <ramiro> hmm. looping a few thousand times over the same conversion using START/STOP_TIMER
[04:46:26] <ramiro> if I convert only once it's still slower.
[04:47:47] <Dark_Shikari> might be more interesting if you used a different memory area each time
[04:49:01] <ramiro> as src or dst?
[04:49:43] <Dark_Shikari> both
[04:50:35] <Dark_Shikari> anyways it depends entirely on the things I said
[04:50:37] <Dark_Shikari> it's a high-level decision
[07:25:33] <KotH> hoi zäme
[07:32:04] <_av500_> moin
[07:32:07] <_av500_> janneg: ping
[09:35:29] <mru> morning
[09:36:57] <pJok> mornings mru
[09:37:15] * mru looks around, sees only one morning
[09:37:24] <pJok> its sunday
[09:37:34] <mru> ah, I looked around spatially, should have looked temporally
[09:37:36] <pJok> they are all hungover
[09:37:37] <mru> now I see many mornings
[09:38:24] <pJok> watching danish news about danish rocket amateurs launching a test flight
[09:45:03] <CIA-11> libswscale: reimar * r32050 /trunk/libswscale/mlib/yuv2rgb_mlib.c:
[09:45:03] <CIA-11> libswscale: Fix mlib compilation: add mising include for SwsContext.
[09:45:03] <CIA-11> libswscale: Patch by Aleksey Cheusov [vle gmx net].
[09:46:05] <mru> lol mlib
[09:52:42] <ohsix> mlib?
[09:53:15] <mru> exactly
[10:11:10] <CIA-11> ffmpeg: reimar * r25042 /trunk/libavcodec/x86/ (vp3dsp.asm vp8dsp.asm vc1dsp_yasm.asm h264_weight.asm):
[10:11:10] <CIA-11> ffmpeg: Use "d" suffix for general-purpose registers used with movd.
[10:11:10] <CIA-11> ffmpeg: This increases compatibilty with nasm and is also more consistent,
[10:11:10] <CIA-11> ffmpeg: e.g. with h264_intrapred.asm and h264_chromamc.asm that already
[10:11:10] <CIA-11> ffmpeg: do it that way.
[13:27:05] <pentanol> wbs hi, I heard talk about rtp testing tool, I do similar things right now, but I'm a bit confused should I use priv_data or not after I did rtp_parse_packet and move RTPDemuxContext to AVFormatContext and then put it to the file?
[13:31:01] <pentanol> wbs, also it loos like we can't stream few streams in one time, such as: ffmpeg -i tst.avi -f rtp -nv rtp://127.0.0.1:10001 -f rtp -na rtp://127.0.0.1:10002 , but on a receiver side should be scaler for joining both stream to one file.
[14:44:06] <lu_zero> pentanol: what are you doing?
[14:44:21] <lu_zero> pentanol: you can
[14:44:29] <lu_zero> the line is lightly different
[14:45:11] <lu_zero> http://cgit.lscube.org/cgit.cgi/flux/tree/scripts/record.sh
[15:40:45] <KotH> wth is frei0r?
[15:40:45] <KotH> a gstreamer clone?
[15:40:52] <mru> it's a silly name
[15:42:25] <kshishkov> why? It's SupahHaxx0r name
[15:42:51] <kshishkov> (which tells quite a lot)
[15:43:59] * kshishkov thinks one should not bother about such stuff until they come to us with some offer
[15:45:29] * KotH thinks one should not bother about such stuff until they come to us with some offer for a lot of swiss chocolate
[15:48:51] * kshishkov thinks it happens only in KotH's dreams
[15:51:47] <KotH> well.. there was a time when people came to me and asked whether i want this or that anime. why shouldnt it happen that people ask me whether i want chocolate?
[15:52:19] <funman> KotH: do you want swiss chocolate?
[15:52:51] <kshishkov> KotH: because you live in a country with hyped chocolate
[15:52:56] <mru> funman: KotH is made of swiss chocolate
[15:53:17] <KotH> funman: ofcourse!
[15:53:37] <kshishkov> KotH: then go buy it, you lazy lout!
[15:53:40] <funman> well me too, if i find a free supply i'll tell you!
[15:53:45] <KotH> kshishkov: i know it's not swedish cocolate, but IMHO it's better than any other (at least it's definitly better than any other i tried)
[15:54:20] <kshishkov> KotH: some French chocolate is definitely better than any Swiss chocolate I tried - 100% cocoa
[15:54:47] * KotH is not fan of sandstorm (aka 100% cocoa)
[15:55:12] <kshishkov> why such nickname?
[15:55:31] <KotH> because, 100% cocoa is a powder
[15:55:42] <funman> i am a fan of milka chocolate
[15:56:05] <kierank> so am i
[15:56:15] <kshishkov> KotH: ever heard of cocoa butter?
[15:56:42] <KotH> funman: milka was swiss... then a german compnay bought it, which got bought by an abomination
[15:56:46] <kshishkov> funman: and I'm suspicious about violet cows, being Ukrainian and such...
[15:57:00] <kshishkov> Nestle?
[15:57:07] <KotH> kraft
[15:57:35] <KotH> nestle doesnt buy chocolatiers, they buy springs
[15:57:45] <funman> http://5f.img.v4.skyrock.net/5f9/loulavache/pics/441881633_small.jpg
[15:58:03] <kshishkov> KotH: still, if you mix coca powder with cocoa butter you'll get real tasty 100% cocoa chocolate
[16:00:10] * funman remembers KotH offering swiss chocolate at linuxtag 2008
[16:00:43] <funman> i also remember 2 swiss guys offering weed
[16:00:53] <KotH> lol
[16:01:02] <funman> swiss are nice people!
[16:01:10] <lu_zero> pfff
[16:03:56] <lu_zero> KotH: you should come around and I'll bring you to two traditional chocolatier
[16:04:17] <funman> lu_zero: where are you?
[16:04:29] <lu_zero> Torino
[16:05:02] <KotH> lu_zero: i never had some real torino chocolate...only that pseudo stuff they sell here, right next to the german chocolate
[16:05:26] <lu_zero> http://www.pastiglieleone.it/fabbrica-cioccolato.php
[16:05:37] <lu_zero> you mean mass produced ferrero one?
[16:05:51] * KotH doesnt consider ferrero as chocolate
[16:05:55] <lu_zero> It's more Alessandria
[16:05:57] <lu_zero> it isn't
[16:06:20] <lu_zero> is more hazelnut and other addictive substances
[16:06:28] <lu_zero> Ferrero -> Nutella
[16:06:36] <KotH> juup :)
[16:07:21] <mru> hazelnut addictive?
[16:07:38] <KotH> mru: shortly before it becomes an alergic reaction
[16:08:13] * kshishkov is addicted to inlagd sill
[16:09:30] <lu_zero> mru: also cocoa
[16:09:42] <lu_zero> apparently hazelnut+cocoa is a deadly mix
[16:09:59] <kshishkov> mandels are better
[16:10:02] <mru> tastes ok
[16:10:10] <lu_zero> mandels are poisonous
[16:10:15] <lu_zero> different ^^
[16:10:27] <mru> those are called almonds
[16:10:55] <lu_zero> uhm
[16:11:22] <lu_zero> we have some traditional recipes for bitter almonds as well...
[16:11:42] <mru> they are toxic in large quantities
[16:12:00] <kshishkov> yet cianide tastes great!
[16:12:21] <lu_zero> in small quantities =P
[16:13:38] <kshishkov> indeed
[16:13:52] <kshishkov> in large ones it's a taste for lifetime
[16:16:22] <lu_zero> theheh
[21:05:14] <CIA-11> ffmpeg: stefano * r25043 /trunk/libavcodec/avcodec.h:
[21:05:14] <CIA-11> ffmpeg: Add #ifndef around FF_API_MM_FLAGS.
[21:05:14] <CIA-11> ffmpeg: The #ifndef is required to allow for example some automated regression
[21:05:14] <CIA-11> ffmpeg: tests by simply configuring with: --extra-cflags="-DFF_API_FOO=0".
[21:39:06] <CIA-11> ffmpeg: aurel * r25044 /trunk/libavformat/matroskadec.c: matroskadec: allow uint and float elements with length = 0
[22:16:37] <felipec> all right, the tidsp hwaccel seems to be working fine... patches sent ;)
[22:26:37] <CIA-11> ffmpeg: aurel * r25045 /trunk/libavformat/utils.c: move stream info arrays into a struct to ease future dynamic allocation
[22:51:34] <pentanol> lu_zero thanks for points, I was wonders how that did vlc, I guess they use PT(payload types) in rtp tod spliting video and audio and use just one pair ports
[23:11:26] <lu_zero> pentanol: for rtp?
[23:26:48] <pentanol> yep *for
[23:29:21] <pentanol> also I've crazy idea for split video and audio between few network channels, I'm not quite shure, but it should be done in own mux\demuxer. But how does split it and don't lose stream data if I lose one of network channels.


More information about the FFmpeg-devel-irc mailing list