Hi, On Thu, Feb 25, 2016 at 8:26 PM, Wan-Teh Chang <wtc-at-google....@ffmpeg.org > wrote:
> On Thu, Feb 25, 2016 at 12:00 PM, Ronald S. Bultje <rsbul...@gmail.com> > wrote: > If anything, you can use atomic ints instead of regular ints if we don't > > already [1]. > > > [...] > > [1] https://ffmpeg.org/doxygen/trunk/atomic_8h_source.html > > Thanks for the link. avpriv_atomic_int_get and avpriv_atomic_int_set > perform the load and store with sequential consistency, which requires > a full memory barrier and is slower than simply relying on the > existing per-thread mutex. The drawback of my patch is that it uses > more memory. I think that's the right trade-off, but I would be happy > to use an atomic int. Please let me know what you prefer. Is probably OK since it doesn't introduce new lock/unlock pairs, yes. We typically don't care about a few bytes of memory. Ronald _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
