[Ffmpeg-devel-irc] ffmpeg-devel.log.20131213

burek burek021 at gmail.com
Sat Dec 14 02:05:02 CET 2013

[00:15] <cone-204> ffmpeg.git 03Michael Niedermayer 07master:180a0b1bcb52: avcodec/hevc_sei: fix use of uninitialized variables in decode_nal_sei_frame_packing_arrangement()
[00:36] <clever> Daemon404: and its fixed
[00:39] <clever> VO: [x11] 1920x1072 => 1920x1072 Planar YV12
[00:39] <clever> [swscaler @ 0xc3f0d0]using unscaled yuv420p -> rgb565le special converter
[00:40] <clever> [swscaler @ 0xc3f0d0]No accelerated colorspace conversion found from yuv420p to rgb565le.
[00:40] <clever> ok, thats going to really suck!
[00:41] <Daemon404> what exactly are you converting to that colorspace for
[00:41] <clever> mplayer didnt auto-detect a decent VO, so its converting the 1080p frames to rgb565le
[00:42] <clever> and sending them to xorg, over shm i think
[00:42] <wm4> oh god vo_x11
[00:42] <clever> wm4: exactly what i was thinking
[00:46] <clever> -vo sdl does work, and avoids the colorspace conversion
[00:47] <wm4> depends entirely what sdl is doing internally
[00:48] <clever> lets see, 30 frames with sdl takes 11 seconds
[00:48] <clever> wait no, 50
[00:49] <clever> 50 frames with x11 takes 11 seconds
[00:49] <clever> ok, strange, x11 and sdl have the same performance...
[00:50] <clever> [gl] no GLX support present
[00:50] <clever> wm4: would this be an xorg problem, or an mplayer problem?
[00:50] <wm4> on that platform it probably supports egl and gles only
[00:51] <wm4> mplayer seems to have some support for that, but AFAIK it's extremely crappy
[00:51] <wm4> (works on some other ARM board only)
[00:51] <clever> and i didnt enable it at compile time
[00:51] <wm4> so try that
[00:51] <clever> i suppose i could just make a vo_rpi that uses video_render
[00:55] <clever> hmmm, maybe fbdev might give more performance
[00:56] <clever> 10 seconds to render 50 frames, slight visual corruption, and it had a segfault inside uninit_vcodec
[00:57] <clever> and fbdev2 wants bgra!
[00:57] <clever> ah, fbdev wants rgb565le, ok, neither is good right now
[01:02] <clever> wm4: yeah, looks like the current design just wont work, -vo null gets 11 fps
[01:02] <clever> its just too much data to be shuffling arround
[01:02] <clever> doesnt matter how fast the vo is, it wont beat null
[01:09] <cone-204> ffmpeg.git 03Michael Niedermayer 07master:db8a53d86a71: avformat/sdp: fix memleak on error
[02:12] <clever> the root problem right now seems to be that just getting a frame out of the gpu takes too long (0.032982), so i'm trying to get 2 at once, and let the async tasks do its job
[02:12] <clever> the only side-effect, is that it will return the 'wrong' frame, one frame old
[02:12] <clever> but it may already be doing that, i dont know how much omx is buffering on me
[02:16] <cone-204> ffmpeg.git 03Michael Niedermayer 07master:c5f15f40b9b2: avformat/rtpdec_h264: fix null pointer dereferences
[02:16] <cone-204> ffmpeg.git 03Michael Niedermayer 07master:5d8e4f6da03c: avformat/rtpenc: check av_packet_get_side_data() return, fix null ptr dereference
[04:24] <cone-204> ffmpeg.git 03Lou Logan 07master:d546a59a8139: docs/outdevs: add a few simple ALSA examples
[09:46] <cone-169> ffmpeg.git 03Stefano Sabatini 07master:bed4eeab3304: MAINTAINERS: remove ffserver maintainership status notice
[09:46] <cone-169> ffmpeg.git 03Stefano Sabatini 07master:93f08827253c: ffserver: improve error feedback in case of av_write_frame() error
[09:46] <cone-169> ffmpeg.git 03Stefano Sabatini 07master:65a3429e7ac1: ffprobe: add cosmetical empty line
[09:47] <cone-169> ffmpeg.git 03Even Wiik Thomassen 07master:43f116d1621a: examples/demuxing_decoding: return error when no codec found
[10:37] <cone-169> ffmpeg.git 03Clément BSsch 07master:b02d87936b3b: examples/demuxing_decoding: print the decoding error when it happens.
[11:56] <saste> do we have some native tool to shift timestamps, with no transcoding?
[11:59] <ubitux> -itsoffset
[11:59] <ubitux> ?
[12:02] <saste> uhm seems so...
[13:38] <BBB> ubitux: ok current vp9-simd branch works - it's not very clean, I'll likely do a semi-rewrite in terms of register numbering, but at least it's something
[13:38] <BBB> ubitux: performance-wise this is pretty much going to be it for the full 16x16, I'll write subforms over the next few days, maybe this weekend
[13:40] <BBB> I'm not totally unhappy about it :)
[13:53] <ubitux> BBB: yeaah :D
[13:54] <ubitux> on my side, i'm almost done with the "fm" if :(
[13:55] <BBB> \o/
[13:56] <BBB> loopflter takes very long, don't worry about it feeling slow
[13:56] <BBB> vp8's version took me forever also, and it's trivial compared to the vp9 one
[13:59] <BBB> as for your /2 question, if the number is always positive, just use >>1
[13:59] <BBB> if the number can be negative, use (x+(x<0))>>1
[13:59] <BBB> +x<0 is the same as psubb(mx, pcmpgtw(pxor(my, my), mx))
[14:00] <BBB> er, gtb
[14:00] <BBB> not gtw
[14:01] <BBB> and then for the rightshift, do something to maintain the sign, e.g. mx>>1 = por(pand(mx, 0x8080808080808080), psrlw(mx, 1))
[14:01] <BBB> (since there's no psrab)
[14:03] <ubitux> i used michael's trick for the /2
[14:03] <ubitux> (the +1 and average with 0)
[14:04] <BBB> just make sure you understand that the rounding of / and pavgb is different
[14:04] <ubitux> yes that's why i did a +1
[14:04] <BBB> pavgb is >>, so -> -inf, / is -> 0
[14:04] <BBB> you might need to do a +x<0
[14:04] <ubitux> yes i think my current code assume unsigned all the time
[14:04] <BBB> pavgb(mx, pcmpgtb(0, mx))
[14:04] <ubitux> i need to check that&
[14:05] <BBB> ok, if that assumption is correct, then that's ok, yes
[14:06] <BBB> I'm currently toying with reordering register operations so we always use tX==mX
[14:07] <BBB> that would basically mean introducing a few macro wrappers that do more swaps, e.g. SUMSUB_AB in addition to SUMSUB_BA (where AB is just BA followed by SWAP A,B)
[14:07] <BBB> and likewise a macro wrapper for UNPACK_MULSUB_2W_4X
[14:08] <BBB> but maybe it's not worth it, I don't know
[15:20] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:8f9569cfacb9: avfilter/vf_pullup: Factor free_field_queue() out
[15:20] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:47eb15b989a0: avfilter/vf_pullup: fix memleak on error
[15:59] <cone-382> ffmpeg.git 03Carl Eugen Hoyos 07master:fb5518cfb76a: Force one stream in caf muxer.
[15:59] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:1d0995e3a59e: Merge remote-tracking branch 'cehoyos/master'
[16:51] Action: ubitux just ordered a rpi to test clever work
[16:51] <Compn> lol
[17:00] <cone-382> ffmpeg.git 03Pierre Yves MORDRET 07master:0f02ea0afe96: libavcodec/dvbsub: fix for Display Definition Segment
[17:01] <ubitux> can we have a sample for this ^ ?
[20:30] <kierank> ubitux: he said maybe
[20:30] <ubitux> ok :)
[20:31] <kierank> I do wonder what the sample is about
[20:31] <kierank> he said it wasn't a test suite
[20:35] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:7439475e69f3: avformat/network: check ff_socket_nonblock() return and print a debug message
[20:35] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:7d7a70136286: avformat/ipmovie: Fix use of uninitialized memory in OPCODE_INIT_VIDEO_BUFFERS
[20:35] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:0c23ee131257: avcodec/mjpegdec: use mallocz for progressive blocks
[20:36] <cone-382> ffmpeg.git 03James Almer 07master:782d48c3c126: configure: remove redundant definition of __USE_MINGW_ANSI_STDIO
[20:50] <cone-382> ffmpeg.git 03Diego Biurrun 07master:a1d7d5524799: build: Do not add -g to HOSTCFLAGS
[20:50] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:7b2a19cb9e66: Merge commit 'a1d7d5524799bd90bc5d12e8b3063238cc0c02dd'
[21:24] <clever> Compn: ive got an idea that may fix all of this extradata problems
[21:25] <clever> Compn: OMX_IndexParamNalStreamFormatSelect
[21:25] <clever> Sets the format of NAL units that will be presented to the decoder (OMX_NaluFormatStartCodes or OMX_NaluFormatOneNaluPerBuffer). By default the decoder will try to guess the type based on the codec configuration data it has been given. This is parameter is currently only relevant for AVC streams.
[21:27] <cone-382> ffmpeg.git 03Diego Biurrun 07master:bdc111a16209: configure: Detect newer (>=4.1) versions of MinGW32
[21:27] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:2a459f5d87f8: Merge commit 'bdc111a162094c14660d1e88839d103a4d79e42a'
[21:32] <cone-382> ffmpeg.git 03Diego Biurrun 07master:7c3388711b72: configure: Add -D__USE_MINGW_ANSI_STDIO=1 to CPPFLAGS on MinGW32
[21:32] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:1e3718e2a1e7: Merge commit '7c3388711b7243d059db39aa04b754baf2156934'
[21:42] <cone-382> ffmpeg.git 03Diego Biurrun 07master:79d9884e9590: build: Merge pthreads/w32threads OBJS declarations
[21:42] <cone-382> ffmpeg.git 03Michael Niedermayer 07master:0f242e62b4b4: Merge remote-tracking branch 'qatar/master'
[00:00] --- Sat Dec 14 2013

More information about the Ffmpeg-devel-irc mailing list