[FFmpeg-devel] [PATCH] Clip-Wrapped MXF support (attempt #4)

Reimar Döffinger Reimar.Doeffinger
Wed Jan 26 18:48:23 CET 2011


On Wed, Jan 26, 2011 at 11:11:29AM +0200, Maksym Veremeyenko wrote:
> Reimar D?ffinger ???????(??):
> >On Tue, Jan 25, 2011 at 01:38:10PM +0200, Maksym Veremeyenko wrote:
> >>@@ -524,13 +530,35 @@ static int mxf_read_source_package(void *arg, ByteIOContext *pb, int tag, int si
> >> static int mxf_read_index_table_segment(void *arg, ByteIOContext *pb, int tag, int size, UID uid)
> >> {
> >>+    MXFIndexTableSegment *segment = (MXFIndexTableSegment *)arg;
> >
> >Pointless cast.
> 
> will fix
> 
> >
> >>+                /* check small EditUnitByteCount for audio */
> >>+                if (!mxf->current_klv_bsize)
> >>+                    mxf->current_klv_bsize = 1024;
> >
> >IMO this "default" is a lot too small, one packet per 1 kB is a
> >massive overhead if you end up using it for e.g. a 100 MBit/s
> >MPEG-2 just because it does not have any such metadata setting.
> >1 kB sized packets means > 2% memory overhead in general and
> >for above example it would be about 12000 demuxer calls per second!
> >Apart from that I retract my earlier objections, I wasn't aware
> >the whole file would be read into memory currently.
> 
> does 1440000 bytes block is good enough - that magic count odd to
> DV25...50 frames...

Performance-wise it's suboptimal, it is not even a multiple a page
size, and for network file systems a multiple of 32 kB would be better.



More information about the ffmpeg-devel mailing list