[FFmpeg-devel] [PATCH 2/3] avfilter/avfilter: avfilter_register() that works in O(1) time

Ivan Kalvachev ikalvachev at gmail.com
Tue Dec 17 21:03:28 CET 2013


On 12/17/13, Stefano Sabatini <stefasab at gmail.com> wrote:
> On date Tuesday 2013-12-17 01:05:21 +0100, Michael Niedermayer encoded:
>> On Mon, Dec 16, 2013 at 05:16:37PM +0100, Stefano Sabatini wrote:
>> > On date Sunday 2013-12-08 03:55:22 +0100, Michael Niedermayer encoded:
>> > > This also flips the order of the list
>> > > avfilters should be unique and thus the order should not matter
>> > >
>> > > Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
>> > > ---
>> > >  libavfilter/avfilter.c |    8 +++-----
>> > >  1 file changed, 3 insertions(+), 5 deletions(-)
>> > >
>> > > diff --git a/libavfilter/avfilter.c b/libavfilter/avfilter.c
>> > > index 2567ce9..c564b45 100644
>> > > --- a/libavfilter/avfilter.c
>> > > +++ b/libavfilter/avfilter.c
>> > > @@ -476,7 +476,6 @@ AVFilter *avfilter_get_by_name(const char *name)
>> > >
>> > >  int avfilter_register(AVFilter *filter)
>> > >  {
>> > > -    AVFilter **f = &first_filter;
>> > >      int i;
>> > >
>> > >      /* the filter must select generic or internal exclusively */
>> > > @@ -488,10 +487,9 @@ int avfilter_register(AVFilter *filter)
>> > >                      || (!input->start_frame && !input->end_frame));
>> > >      }
>> > >
>> > > -    filter->next = NULL;
>> > > -
>> > > -    while(*f || avpriv_atomic_ptr_cas((void * volatile *)f, NULL,
>> > > filter))
>> > > -        f = &(*f)->next;
>> >
>> > > +    do {
>> > > +        filter->next = first_filter;
>> > > +    } while (filter->next != avpriv_atomic_ptr_cas((void * volatile
>> > > *)&first_filter, filter->next, filter));
>> >
>> > why do you need a loop for this?
>>
>
>> consider 2 libs or an app and a lib try to use avfilter_register() at
>> the same time
>> we need to insert our filter atomically but its 2 operations so if
>> something else inserted something after we did the first step but
>> before the second then we have to retry
>
> Isn't a mutex a cleaner solution?

I think the cleanest solution is to have a static constant table.
Do we really need to register everything?


More information about the ffmpeg-devel mailing list