Re: [Dri-devel] Re: sched_yield()

[Denis O. K. <do...@di...>]

Quoting Denis Oliver Kropp (do...@di...):
> Quoting Keith Whitwell (ke...@tu...):
> > Michel Dänzer wrote:
> > >The bottom line is that with a 2.5 kernel, sched_yield() will discard
> > >the time slice of the process, so misuse of it will cause bad
> > >performance.
> > 
> > I guess it's possible that this could be a problem.  The yield is there to 
> > allow two graphics apps to somewhat smoothly share the card & cpu - 
> > otherwise you get very non-smooth interleaving or even starvation of some 
> > of the graphics apps on 2.4.
> 
> I've also noticed non-smooth interleaving after adding multi application support
> to DirectFBGL (about to commit in the next hours). I added sched_yield() at the
> end of IDirectFBGL::Unlock() and it got very smooth, but I'm not that happy
> about the solution. I think I will add the necessary pieces of sched_yield() to
> the fusion call (ioctl) that releases the long-time lock of the graphics card
> in DirectFB.

Anyone interested in multiple blended hardware OpenGL windows (in multiple
processes) should have a look at our screenshots page ;)

It shows quake2 and tuxracer on SDLGL using DirectFBGL on DRI ;)

-- 
Best regards,
  Denis Oliver Kropp

.------------------------------------------.
| DirectFB - Hardware accelerated graphics |
| http://www.directfb.org/                 |
"------------------------------------------"

                            Convergence GmbH