On 1 Jan 2002, Owen Taylor wrote:
>
> Mark Vojkovich <[EMAIL PROTECTED]> writes:
>
> > On 30 Dec 2001, Owen Taylor wrote:
> >
> > > [2] Right now, to get decent performance, the source must be
> > > in system memory - we clearly need this for decent performance
> > > for the software fallback, and XAA also only supports this.
> > > For the local rendering case, we can enforce the source
> > > being in the right place by using a ShmPixmap. For the
> > > remote case, we have no way to enforce this, and given
> > > sufficient video ram, we'll get bad performance.
> >
> >
> > It wouldn't be too difficult to force sources out of video ram
> > when they are used in composite operations. I do that for other
> > cases such as offscreen pixmaps being used as tiles in GCs used
> > on system memory pixmaps.
> >
> > Yes, XAA assumes all data is in system memory and doesn't
> > optimize when both source and destination are in video ram, which
> > should be the fastest case of all. The drivers can hook this
> > high enough to do this themselves, but XAA doesn't have low-level
> > functions to do this. I can add this to NVIDIA's drivers easily
> > enough, but the problem is testing. I've hardly seen anything
> > resembling a test or performance suite for Render. I can't accelerate
> > anything I can't test, which is why XAA will only accelerate some
> > very simple cases.
>
> It's perhaps a bit of a chicken and egg problem ... it's hard to write
> accurate performance tests until there are apps, and no compelling
> reason to write apps using the more complicated bits of RENDER as long
> as its faster to do things on the client side.
If there is software support for these operations in RENDER, then
at least you can get working applications. Performance tuning can
be done on the working applications. If there aren't test apps,
I can't ensure that the result is even visually correct. It took
me a hell of alot of time to program up GeForce's register combiners
to do the right thing even for the simple cases we accelerate now.
Good tests are things like the "HelloX-XRender" app that was
floating around. Lightweight, simple tests that don't require
lots of support libraries. If it draws the whole scene everytime
you get an expose or a resize that's good enough for performance
testing. I just opaque resize and optimize until it's interactive.
As for booting the sources out of videoram, there should
probably be a check at the top of XAAComposite() that does
something like:
if((pSrc->pDrawable->type == DRAWABLE_PIXMAP) &&
IS_OFFSCREEN_PIXMAP(pSrc->pDrawable) &&
!OFFSCREEN_PIXMAP_LOCKED(pSrc->pDrawable))
{
XAAPixmapPtr pPriv = XAA_GET_PIXMAP_PRIVATE(pSrc->pDrawable);
FBAreaPtr area = pPriv->offscreenArea;
XAARemoveAreaCallback(area); /* clobbers pPriv->offscreenArea */
xf86FreeOffscreenArea(area);
}
Mark.
_______________________________________________
Render mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/render
