EXA and MigrationHeuristic

Eric Anholt eric at anholt.net
Wed May 28 12:11:37 PDT 2008

On Wed, 2008-05-28 at 20:46 +0200, Juliusz Chroboczek wrote:
> Dear all,
> After my recent adventures with Radeon, XAA and EXA, I've come to the
> conclusion that setting MigrationHeuristic to greedy drastically helps
> performance of Firefox redraws on both Radeon and i965GM hardware.

The 965 render acceleration is still broken in that for small objects
(like text) it's slower than software.  Someone with the time needs to
fix that problem, by queuing up multiple primitives in a single vertex
buffer, so that we don't sync the chip every primitive.  There's a
commit attempting that in


on the 965-render-vb branch, but it's been done better in
intel-batchbuffer (emitting larger primitive lists).

Until someone fixes the driver problem, I would be opposed to use of the
965 as a metric for changing server defaults.

> By drastically, I mean it -- a redraw of a full-screen Firefox goes
> from 3 to 5 seconds (!) down to a fraction of a second.
> Now I haven't read the sources of Firefox (let him who has cast me the
> first stone), but I believe that Firefox performs client-side backing
> store using absolutely humonguous pixmaps.  While we may argue that
> what Firefox does is stupid, I think we should try to support this
> kind of usage.
> Isn't the default value of MigrationHeuristic incorrect ?  Should we
> switch to greedy as default ?  (Julien tells me that he believes that
> Ubuntu have already made the switch.)
>                                         Juliusz
> _______________________________________________
> xorg mailing list
> xorg at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/xorg
Eric Anholt                             anholt at FreeBSD.org
eric at anholt.net                         eric.anholt at intel.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: This is a digitally signed message part
URL: <http://lists.x.org/archives/xorg/attachments/20080528/1a17ea2e/attachment.pgp>

More information about the xorg mailing list