On Thu, 1 Aug 2002, Steve Kirkendall wrote:
> > From: Mark Vojkovich <[EMAIL PROTECTED]>
> > Date: Wed, 31 Jul 2002 12:42:10 -0700 (PDT)
> >
> > On Wed, 31 Jul 2002, Steve Kirkendall wrote:
> >
> > > The testxv.c program has an XFlush(display) call, but it's commented
> > > out. Because of this, the XShmPutImage(...) requests are all queued
> > > until Xlib's request queue fills up, at which time they are all sent
> > > in one big chunk. This could make the "frames per second" value
> > > fluctuate wildly, giving meaningless results. It wouldn't look very
> > > good on the screen, either. But try replacing the commented-out
> > > XFlush(...) call with the above code, first.
> >
> > XGetGeometry will flush.
> >
> > >
> > > Also, testxv.c calls XGetGeometry() inside the loop. I believe
> > > this is a round-trip request (can somebody confirm that?) which
> > > would slow down the display loop. You'd be better off watching
> > > for ConfigureNotify events.
> >
> > Actually XGetGeometry would work as an XSync(). This would
> > have the same effect as flushing and getting the notify event, plus
> > it gets geometry changes. It's probably more efficient overall than
> > flushing and waiting for Completion or ConfigureNotify events.
> >
>
> Let me try paraphrasing that, to see if I've got it right.
>
> The XGetGeometry() call will serve multiple roles: It will flush
> the request queue; it will wait for completion of the XvShmPutImage()
> request (because the X server always completes *ALL* pending requests
> before sending a response to a round trip request?); it will get the
> window's current geometry.
Yes.
>
> Consequently, the only change needed in testxv.c is, the last
> parameter of the XvShmPutImage() call should be changed from True
> to False so the X server won't send completion events. That will
> prevent the client's event queue from growing endlessly, which is
> what the original poster was concerned about.
>
> Is that right?
Yes. Using completion event doesn't really help unless you
want to get some work done between the time that you send the
XvShmPutImage request and the time it is completed. That way
you don't have to wait for the server in order to continue. On
an SMP system this can be a big win even if you only have a single
threaded application. But it's only of benefit if you are
trying to get work done (like rendering the next buffer) between when
you flush the PutImage and when you want to reuse the buffer.
Mark.
_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert
