speed dfference using clipping with EXA acceleration

[Olivier Fourdan]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi all,

I have a bug that reports that the xfwm4 (xfce window manager) embedded
compositor is slower in 4.4 beta2 than in 4.4 beta1.

The user is using the ATI open source driver with EXA.

One of the changes that occurred between beta1 and beta2 is the use of
clipping in beta2 for rendering windows frame with different opacity
using techniques such as :

- - XFixesIntersectRegion()
- - XFixesSetPictureClipRegion()
- - XRenderComposite() using the region computed just above as clipping
- - XFixesSubtractRegion()

instead of "cutting" the window border areas to render in separate calls
to XRenderComposite() with different mask/src/dst coordinates that
describe just the window decorations area.

Reverting that change makes things fast again for the user.

Is that expected? In one case, using 4 calls to XRenderComposite() with
smaller areas is fast, while in the other case, there is just one call
to XRenderComposite() on a large area but with a clipping that leaves
just the decorations area (ie most of the area will not be rendered) and
it's slow...

What's the best (I mean technically speaking, because I clearly prefer
using clipping as I find that more "elegant", but unfortunately it's
slower)?

Thanks in advance,
Cheers,
Olivier.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFE5NqxFHBtvh5LRUARAst2AJ0U8x50zSxmzW0TNJoQaevhENvAiQCfZTDe
rZCZgIrtqvb7jUPx0gEs4WA=
=trqn
-----END PGP SIGNATURE-----