[lkml]   [2019]   [Nov]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [linux-sunxi] Re: [PATCH 1/3] media: cedrus: Properly signal size in mode register
Hi Stefan,

On Thu 07 Nov 19, 09:24, Stefan Monnier wrote:
> > Do you know if we have a way to report some estimation of the maximum supported
> > fps to userspace? It would be useful to let userspace decide whether it's a
> > better fit than software decoding.
> Even if the fps ends up too low for the player's taste, I can't imagine
> why software decoding would be preferable: it seems it could be only
> even (substantially) slower. Or are there speed-up options in software
> decoding not available in hardware decoding (such as playing only every
> N'th frame, maybe?).

This may be true for the Allwinner case as we know it today but not true in
general. It could happen that the CPU is perfectly able to decode as fast as or
faster than the hardware implementation, but userspace would still try to use
hardware video decoding when it can provide good enough performance so that the
CPU can do other things in the meantime.

Having a good idea of the expected performance is important for userspace to
make this kind of policy decision.

This is kind of a common misconception that hardware offloading always implies
a performance improvment. In our cases where the CPU is a bottleneck, it is
more often true than not, but this is by far not true in general.



Paul Kocialkowski, Bootlin
Embedded Linux and kernel engineering
[unhandled content-type:application/pgp-signature]
 \ /
  Last update: 2019-11-09 13:59    [W:0.047 / U:34.148 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site