On Nov 10, 2009, at 1:39 PM, Bruce Wheaton wrote:
On Oct 29, 2009, at 4:10 PM, Jean-Daniel Dupas wrote:
Le 29 oct. 2009 à 23:35, Bruce Wheaton a écrit :
What is reordered_opaque for? I presume it helps with reordered
packets, and I see anecdotal evidence that if I want to override
get_buffer, I should set the frame's reordered_opaque to the
current codec pts, but I'm not sure if it's essential.
Currently, in my video decode tasks, I get the pts of the first
video packet, and use that as the pts of the video frame I obtain
(I stay in the decode job until I get a complete video frame). Is
that no longer a valid approach?
It will not work with codec that use frame reordering (H.264 for
example). To reorder the frame, the decoder will bufferize decoded
frames, and so when you call decode_frame with such decoder, the
decoded frame returned may not be the one corresponding to the
packet you pass as argument.
In such cases, packet pts will not match frame pts.
Sorry, looks like I have another question. In this frame re-
ordering, when I pass in an AVFrame, will I always get that same
AVFrame out, or will the decode sometimes want to give me a new frame?
I'm trying to set the opaque pointer of the frame I get in, but when
my decode routine finishes, I seem to get back another frame. Should
I pass in NULL then, if I'm just going to get a new frame back?
Huh, it's the same frame, but my opaque pointer was blanked, so my
get_buffer callback didn't work properly (it has a pointer to the
parent object that will make the actual storage).
Literally, between decode and get_buffer being called, the opaque gets
zero'ed.
I tried setting the AVFrame type to user earlier, in the hopes that
would persuade the codec to leave the data alone. I tried overriding
reget_buffer.
So now I'm wondering, am I trying to do something evil, or possibly
it's ffmpeg-mt that behaves differently? I think I'm using the opaque
data exactly in the way intended, but is it not expected to persist
across a get_buffer cycle?
Bruce
is->video_st->codec->reordered_opaque= pkt->pts;
len1 = avcodec_decode_video2(is->video_st->codec, frame,
&got_picture, pkt);
if ((decoder_reorder_pts || pkt->dts == AV_NOPTS_VALUE) &&
frame->reordered_opaque != AV_NOPTS_VALUE)
pts= frame->reordered_opaque;
else if (pkt->dts != AV_NOPTS_VALUE)
pts= pkt->dts;
else
pts= 0;
pts *= av_q2d(is->video_st->time_base);
I just started to use the ffmpeg-mt branch, if that matters, and I
did add:
#ifdef AVCODEC_HAS_REORDERED_OPAQUE
// take over pts for this frame to have it reordered
pic->reordered_opaque = c->reordered_opaque;
#endif
To my buffer callback.
Regards,
Bruce Wheaton
_______________________________________________
libav-user mailing list
[email protected]
https://lists.mplayerhq.hu/mailman/listinfo/libav-user
-- Jean-Daniel
_______________________________________________
libav-user mailing list
[email protected]
https://lists.mplayerhq.hu/mailman/listinfo/libav-user
_______________________________________________
libav-user mailing list
[email protected]
https://lists.mplayerhq.hu/mailman/listinfo/libav-user
_______________________________________________
libav-user mailing list
[email protected]
https://lists.mplayerhq.hu/mailman/listinfo/libav-user
