[FFmpeg-devel] MPEG-PS demuxer index memory usage

[Paul Kelly]

On Mon, 7 Jan 2008, Michael Niedermayer wrote:
> On Mon, Jan 07, 2008 at 01:36:39AM +0100, Baptiste Coudurier wrote:
>> Paul Kelly wrote:
>>> [...]
>>>
>>> +    max_entries= s->max_index_size / sizeof(AVIndexEntry);
>>> +    if(max_entries == 0)
>>> +        return -1;
>>> +    if((unsigned)st->nb_index_entries >= max_entries){
>>> +        int in, out= 0;
>>> +        /* Halve the size of the index by removing every second entry */
>>> +        for(in=0; in<st->nb_index_entries; in+= 2)
>>> +            st->index_entries[out++]= st->index_entries[in];
>>> +        st->nb_index_entries= out;
>>> +    }
>>> +
>>>      entries = av_fast_realloc(st->index_entries,
>>
>> Just a note to remember that mov demuxer is building index containing
>> all samples, and absolutely need all index entries to seek and demux, so
>> you cannot remove any index entry with the current implementation.
>
> yes, right, i should have realized that earlier ...
> so i guess the best would be an explicitly called ff_reduce_index() which
> does the above. This also safes us from changing all av_add_index_entry()

Do you mean to call ff_reduce_index() before av_add_index_entry() in 
individual demuxers, if it's appropriate to how they use the index? The 
attached patch does that, but only for AVFMT_GENERIC_INDEX and the mpegps 
demuxer (to solve my specific problem). If it's the right approach I could 
try adding it to the other demuxers too, but it might be better if someone 
more familiar with how they work did that.

Paul
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ff_reduce_index.patch
Type: text/x-diff
Size: 4231 bytes
Desc: 
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080107/0e3bed4e/attachment.patch>