[FFmpeg-devel] [PATCH] MXF index table based seeking
Tomas Härdin
tomas.hardin at codemill.se
Mon May 16 10:27:59 CEST 2011
Georg Lippitsch skrev 2011-05-15 15:35:
> Here is my first try to add image table based seeking to the MXF demuxer.
>
> What the patch basically does is:
> 1. Parse through the whole file in mxf_read_header (instead of breaking
> at the first essence container). If an index table is found, it is
> parsed and stored in an MXFIndexTableSegment.
This seems OK, but it'll be very slow for large frame wrapped files.
What you could do to speed it up is to make it partition aware.
Unfortunately the partitions only contain FooterPartition and
PreviousPartition offsets, which means one would have to parse the file
partly in reverse (you can't rely on there being a Random Index Pack).
I've been toying around with doing something like that. The logic is
rather tricky though. I parse forward until I run into essence, then I
seek to the last partition. After that I parse forward and seek to the
previous partition if I run into essence or another partition. This
repeats until I get back to where I was before I did the seek to the footer.
Rebased patches attached. The first one simply adds some code that
parses MXF partitions as it runs across them. The second one does the
parsing I explained above. The commit message is due to the exact same
problem your patch has - my code didn't handle 11 GiB frame wrapped
files very well :o
> 2. In mxf_parse_structural_metadata, the MXFIndexTableSegments are
> parsed and an AVIndexEntry array is created from them
> Seeking is then done based on the AVIndexEntry array.
Sounds OK.
> What I didn't really understand is how to link between MXF tracks and
> the slices in the index table. Sometimes, there is a track for every
> slice in the index table, sometimes there is not. So this part of the
> code is probably wrong, but help on this is very appreciated!
The MXF book attempts to explain slices, but I still haven't managed to
figure out how they work. What I can say is that they're for handling
mixing CBR and VBR media, which is a fairly typical use case (VBR video,
CBR audio).
I can't comment too much on the index parsing code since I don't quite
understand how MXF's indexes work.
> @@ -907,6 +1044,7 @@ static int mxf_read_header(AVFormatContext *s, AVFormatParameters *ap)
> {
> MXFContext *mxf = s->priv_data;
> KLVPacket klv;
> + int64_t file_size = avio_size(s->pb);
>
> if (!mxf_read_sync(s->pb, mxf_header_partition_pack_key, 14)) {
> av_log(s, AV_LOG_ERROR, "could not find header partition pack key\n");
> @@ -914,18 +1052,24 @@ static int mxf_read_header(AVFormatContext *s, AVFormatParameters *ap)
> }
> avio_seek(s->pb, -14, SEEK_CUR);
> mxf->fc = s;
> - while (!url_feof(s->pb)) {
> + while (avio_tell(s->pb) < file_size) {
!s->pb->eof_reached?
> const MXFMetadataReadTableEntry *metadata;
>
> if (klv_read_packet(&klv, s->pb) < 0)
> return -1;
> PRINT_KEY(s, "read header", klv.key);
> av_dlog(s, "size %lld offset %#llx\n", klv.length, klv.offset);
> +
> if (IS_KLV_KEY(klv.key, mxf_encrypted_triplet_key) ||
> - IS_KLV_KEY(klv.key, mxf_essence_element_key)) {
> - /* FIXME avoid seek */
> - avio_seek(s->pb, klv.offset, SEEK_SET);
> - break;
> + IS_KLV_KEY(klv.key, mxf_essence_element_key) ||
> + IS_KLV_KEY(klv.key, mxf_system_item_key)) {
> + if (!mxf->essence_offset) {
> + mxf->essence_offset = klv.offset;
> + if (IS_KLV_KEY(klv.key, mxf_system_item_key))
> + mxf->system_item = 1;
> + }
> + avio_skip(s->pb, klv.length);
> + continue;
This is the part that'll be quite slow.
> }
>
> for (metadata = mxf_metadata_read_table; metadata->read; metadata++) {
> @@ -945,13 +1089,15 @@ static int mxf_read_header(AVFormatContext *s, AVFormatParameters *ap)
> if (!metadata->read)
> avio_skip(s->pb, klv.length);
> }
> + /* FIXME avoid seek */
> + avio_seek(s->pb, mxf->essence_offset, SEEK_SET);
You might want to account for not finding any essence.
Anyway, hopefully the attached patches prove useful. The first one could
probably be applied as-is. Baptiste?
/Tomas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Parse-MXF-partitions.patch
Type: text/x-patch
Size: 6682 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20110516/06e3d32b/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-Speed-up-MXF-metadata-and-index-parsing.patch
Type: text/x-patch
Size: 7089 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20110516/06e3d32b/attachment-0001.bin>
More information about the ffmpeg-devel
mailing list