On 08/25/2017 05:35 PM, wm4 wrote:
+static inline void set_pts(V4L2Buffer *out, int64_t pts)
+{
+ if (pts == AV_NOPTS_VALUE) {
+ /* invalid timestamp: not sure how to handle this case */
+ out->timestamp.tv_sec = 0;
+ out->timestamp.tv_usec = 0;
+ } else {
+ AVRational v4l2_timebase = { 1, 1000000 };
+ int64_t v4l2_pts = av_rescale_q(pts, out->context->time_base,
v4l2_timebase);
+ out->timestamp.tv_sec = v4l2_pts / INT64_C(1000000);
+ out->timestamp.tv_usec = v4l2_pts % INT64_C(1000000);
+ }
+}
Why does it require a fixed timebase? A decoder shouldn't even look at
the timestamps, it should only pass them though. Also, not using DTS
will make it a nightmare to support containers like avi.
I suspect the decoder tries to "fix" timestamps, or maybe even does
something particularly bad like reordering frames by timestamps. This
is NOT something that should be in a kernel API.
(FFmpeg native decoders_and_ hwaccels pass through both PTS and DTS,
and don't touch their values.)
ok I will just pass through the dts/pts and return them unmodified for
decoding and retest.
what about for encoding? should I just read the pts from the driver,
rescale it back to time_base and set both pts&&dts on the returned
packet to the this same value?
after this I think I am ready to post v7 with all the review comments
addressed so thanks a lot for all the responses!
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
