Xegl lives!

Adam Jackson ajax at nwnk.net
Tue May 24 12:47:51 PDT 2005


On Tuesday 24 May 2005 15:14, Jon Smirl wrote:
> On 5/24/05, Lars Knoll <lars at trolltech.com> wrote:
> > Well, it's working just fine in a way that the implementation is correct.
> > If you ask the toolkit people (mainly cairo and Qt), you'd get a
> > different answer. For lots of things we would like to do Xorg is really
> > slow.
>
> The fastest Cairo platform by 10x is glitz. The core of Xgl is based
> on glitz. Seems to me like the fastest path to high performance Cairo
> would be to get Xgl going.

Do you have numbers to back up that claim?  I'd be particularly interested in 
comparing glitz performance to the normal Render core on Xati or on nVidia's 
driver.

> Even if you make modifications to the open source XAA drivers
> accelerating render there is no guarantee that NVidia/ATI will pick
> them up. Chasing that path to accelerating Cairo performance is
> ultimately a dead end.

I don't care if the vendors pick them up.  Their drivers are their problem, 
not mine (well, modulo work responsibilities).

It is vastly simpler to bring up a kdrive-esque driver with decent Render 
accel than it is to bring up a DRI driver.  There are at least 20 distinct 
chipsets where we have no DRI driver, with the degree of difficulty in adding 
support ranging from "poke registers and hope" up through "fully documented". 
[1]  You could spend two days adding Render accel in a classic DDX, or two 
months bringing up a DRI driver.

Render performance is increasingly relevant and XAA is flat out unacceptably 
bad at Render.  If you don't provide a solution, even a stopgap one, for 
older chipsets, you're telling those users to get bent.  I don't even care if 
that solution is nuking XAA and using shadowfb everywhere, it's still work 
that has to be done.

- ajax
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg/attachments/20050524/c61ca588/attachment-0001.pgp>


More information about the xorg mailing list