vout 4 performance issues
benny at via.ecp.fr
Fri Jun 8 08:01:04 CEST 2001
Got all the tools you may want on Solaris, but unfortunately there are
still a few bugs on Sparc that prevent me to get any image. If you guys
can try to run the vlc on the CTI's machines in export display, I then
can send you whatever you need (well, obviously without the
overlay/MMX/whatever optimizations, but at least you'll know if there is
a CPU hog in the vlc).
Samuel Hocevar wrote:
> I'm kinda stuck with my vout 4 development. It was supposed to
> be faster due to the video being decoded directly into the overlay
> surfaces, and I had benchmarked such improvements, but now that the
> performance problems have been fixed in current CVS, vout 4 is slower
> than current CVS (even after applying the same fix, of course).
> Here are the figures on my computer (average values on 5 tests,
> Pentium III 650, 256KB cache, Dolby_Broadway stream):
> released vlc 0.2.80, -A dsp, -V sdl : 62% cpu (X: 20% cpu)
> CVS vlc 0.2.80, -A dsp, -V sdl : 51% cpu (X: 20% cpu)
> vout4 vlc 0.2.80, -A dsp, -V sdl : 54% cpu (X: 1% cpu)
> CVS vlc 0.2.80, -A dsp, -V dummy : 38% cpu
> vout4 vlc 0.2.80, -A dsp, -V dummy : 43% cpu
> As you can see, vout 4 vlc is really slower, even in dummy mode,
> which I fail to explain. The speed loss in SDL overlay mode is smaller
> but that's probably because X already eats a lot less CPU with vout 4.
> Actually I'm beginning to think I left a huge bug somewhere in
> the code. Any ideas on how to track this problem? Is there anyone
> with efficient profiling tools on his platform? AFAIK, Linux lacks a
> multithreaded profiler.
> Tarballs can be found here for comparison (cvs == today's CVS):
More information about the vlc-devel