On Wed, Dec 21, 2011 at 06:46:35AM -0800, Ronald S. Bultje wrote:
> On Wed, Dec 21, 2011 at 5:09 AM, Janne Grunau <[email protected]>wrote:
> > On 2011-12-21 01:55:15 +0000, Måns Rullgård wrote:
> > > Janne Grunau <[email protected]> writes:
> > >
> > > > On 2011-12-20 17:25:42 -0800, Ronald S. Bultje wrote:
> > > >>
> > > >> Missing #else fallback? (Can simply default to 1.)
> > > >
> > > > I had that first but it will break passing 0 to x264, although it may
> > > > be a safe assumption that if our autodetection fails, x264 will also
> > > > default to 1 thread.
> > >
> > > The best value for x264 is not the number of CPUs, and it knows better
> > > what works best for the settings in use.
> >
> > noted and solved
> >
> > > The number CPUs is also not always the best number of threads for us.
> > > Especially frame-based threading benefits from having a few more threads
> > > than CPUs.  On 2 CPUs, 4 threads are often needed to fully saturate the
> > > system.
> >
> > I'm using sched_getaffinity() now and will do some tests how threading
> > scales up to number of cores * 2. I can test on dualcore with and
> > without hyper threading and quadcore.
> 
> If you have a script, I can help, I have 12- and 16-core machines here.

I can test on a 24-core buildhost if necessary.

Diego
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to