On Fri, Jun 10, 2016 at 1:23 PM, Anton Khirnov <an...@khirnov.net> wrote:
> Quoting Vittorio Giovara (2016-06-10 16:52:39)
>> On Fri, Jun 10, 2016 at 4:56 AM, Anton Khirnov <an...@khirnov.net> wrote:
>> > Quoting Vittorio Giovara (2016-06-10 00:55:18)
>> >> Store data from each stsd in a separate extradata buffer, keep track of
>> >> the stsc index for read and seek operations, switch buffers when the
>> >> index differs. Decoder is notified with an AV_PKT_DATA_NEW_EXTRADATA
>> >> packet side data.
>> >>
>> >> Since H264 supports this notification, and can be reset midstream, enable
>> >> this feature only for multiple avcC's. All other stsd types (such as
>> >> hvc1 and hev1) need decoder-side changes, so they are left disabled for
>> >> now.
>> >>
>> >> Signed-off-by: Vittorio Giovara <vittorio.giov...@gmail.com>
>> >> ---
>> >> Only modified to avoid caching a zero-sized extradata.
>> >> Vittorio
>> >>
>> >> libavformat/isom.h | 8 ++++
>> >> libavformat/mov.c | 123
>> >> ++++++++++++++++++++++++++++++++++++++++++++++++++---
>> >> 2 files changed, 125 insertions(+), 6 deletions(-)
>> >>
>> >> diff --git a/libavformat/isom.h b/libavformat/isom.h
>> >> index aec623b..75aa70b 100644
>> >> --- a/libavformat/isom.h
>> >> +++ b/libavformat/isom.h
>> >> @@ -105,6 +105,8 @@ typedef struct MOVStreamContext {
>> >> MOVStts *ctts_data;
>> >> unsigned int stsc_count;
>> >> MOVStsc *stsc_data;
>> >> + int stsc_index;
>> >> + int stsc_sample;
>> >> unsigned int stps_count;
>> >> unsigned *stps_data; ///< partial sync sample for mpeg-2 open gop
>> >> int ctts_index;
>> >> @@ -137,6 +139,12 @@ typedef struct MOVStreamContext {
>> >> unsigned int rap_group_count;
>> >> MOVSbgp *rap_group;
>> >>
>> >> + /** extradata array (and size) for multiple stsd */
>> >> + uint8_t **extradata;
>> >> + int *extradata_size;
>> >> + int last_stsd_index;
>> >> + int stsd_count;
>> >
>> > Nit: nb_stsd would be more consistent
>> >
>> >> +
>> >> int32_t *display_matrix;
>> >> } MOVStreamContext;
>> >>
>> >> diff --git a/libavformat/mov.c b/libavformat/mov.c
>> >> index 125919f..9e2d0e2 100644
>> >> --- a/libavformat/mov.c
>> >> +++ b/libavformat/mov.c
>> >> @@ -1771,8 +1771,7 @@ static int mov_skip_multiple_stsd(MOVContext *c,
>> >> AVIOContext *pb,
>> >> int video_codec_id = ff_codec_get_id(ff_codec_movvideo_tags, format);
>> >>
>> >> if (codec_tag &&
>> >> - (codec_tag == AV_RL32("avc1") ||
>> >> - codec_tag == AV_RL32("hvc1") ||
>> >> + (codec_tag == AV_RL32("hvc1") ||
>> >> codec_tag == AV_RL32("hev1") ||
>> >> (codec_tag != format &&
>> >> (c->fc->video_codec_id ? video_codec_id !=
>> >> c->fc->video_codec_id
>> >> @@ -1857,6 +1856,20 @@ int ff_mov_read_stsd_entries(MOVContext *c,
>> >> AVIOContext *pb, int entries)
>> >> return ret;
>> >> } else if (a.size > 0)
>> >> avio_skip(pb, a.size);
>> >> +
>> >> + if (sc->stsd_count > 1) {
>> >
>> > Are all those checks really necessary? Naively I'd expect that
>> > stsd_count==1 is not really any different from stsd_count=n for any
>> > larger n.
>>
>> I just wanted to avoid needlessly allocating the extra layers and
>> copying data around.
>
> What data would be copied around? There might be some tiny additional
> overhead in read_header, but nothing otherwise.
There also is some overhead in packet_read and packet_seek, but if
dropping the condition does make the code a little simpler.
I'll resend with this change then.
--
Vittorio
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
