On Sat, 25 Jun 2016, Martin Storsjö wrote:
On Sat, 25 Jun 2016, Anton Khirnov wrote:
Quoting Martin Storsjö (2016-06-24 22:44:56)
If the calculated offset from write_packet would be cached somewhere,
that can be used instead of recalculating the offset for the stream.
---
libavformat/internal.h | 3 ++-
libavformat/movenc.c | 3 ++-
libavformat/mux.c | 18 +++++++++++++++---
3 files changed, 19 insertions(+), 5 deletions(-)
diff --git a/libavformat/internal.h b/libavformat/internal.h
index bbdfd2f..1d970dc 100644
--- a/libavformat/internal.h
+++ b/libavformat/internal.h
@@ -449,6 +449,7 @@ void ff_format_io_close(AVFormatContext *s,
AVIOContext **pb);
* @return a pointer to the next packet, or NULL if no packet is queued
* for this stream.
*/
-const AVPacket *ff_interleaved_peek(AVFormatContext *s, int stream);
+const AVPacket *ff_interleaved_peek(AVFormatContext *s, int stream,
+ AVPacket *buf, int add_offset);
#endif /* AVFORMAT_INTERNAL_H */
diff --git a/libavformat/movenc.c b/libavformat/movenc.c
index aadfa06..866a402 100644
--- a/libavformat/movenc.c
+++ b/libavformat/movenc.c
@@ -3262,7 +3262,8 @@ static int mov_flush_fragment(AVFormatContext *s,
int force)
for (i = 0; i < s->nb_streams; i++) {
MOVTrack *track = &mov->tracks[i];
if (!track->end_reliable) {
- const AVPacket *next = ff_interleaved_peek(s, i);
+ AVPacket buf;
+ const AVPacket *next = ff_interleaved_peek(s, i, &buf, 1);
if (next) {
track->track_duration = next->dts - track->start_dts;
if (next->pts != AV_NOPTS_VALUE)
diff --git a/libavformat/mux.c b/libavformat/mux.c
index c41c477..6ed46ae 100644
--- a/libavformat/mux.c
+++ b/libavformat/mux.c
@@ -614,12 +614,24 @@ int ff_interleave_packet_per_dts(AVFormatContext *s,
AVPacket *out,
}
}
-const AVPacket *ff_interleaved_peek(AVFormatContext *s, int stream)
+const AVPacket *ff_interleaved_peek(AVFormatContext *s, int stream,
+ AVPacket *buf, int add_offset)
{
AVPacketList *pktl = s->internal->packet_buffer;
while (pktl) {
- if (pktl->pkt.stream_index == stream)
- return &pktl->pkt;
+ if (pktl->pkt.stream_index == stream) {
+ *buf = pktl->pkt;
+ if (add_offset && s->internal->offset != AV_NOPTS_VALUE) {
+ int64_t offset = av_rescale_q(s->internal->offset,
+
s->internal->offset_timebase,
+
s->streams[buf->stream_index]->time_base);
+ if (buf->dts != AV_NOPTS_VALUE)
+ buf->dts += offset;
+ if (buf->pts != AV_NOPTS_VALUE)
+ buf->pts += offset;
+ }
+ return buf;
+ }
pktl = pktl->next;
}
return NULL;
--
2.7.4 (Apple Git-66)
This is starting to look a bit messy IMO. The return value and the buf
parameter are now redundant, so I think the return value can now be an
int.
Sure
And I would feel better if you stripped the data/size/buf/side_data
fields from the returned packet, since it's now neither a plain pointer
into the intereleaving queue nor a proper reference.
Hmm, I'm a little undecided about that. Being able to peek at the data
could also be seen as useful in some way; not in the current way it's
used, but could prove useful otherwise.
Alternatively, we could also solve it this way instead:
http://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=46a60fe184cfd39cc9517c860c7404d60977456d
Then we can still return a plain reference to the packet within the queue,
letting the user choose whether to apply the offset or not.
// Martin
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
