[Intel-gfx] Too many memory managers

[Eric Anholt]

On Wed, 2008-08-06 at 13:44 +0200, Lukas Hejtmanek wrote:
> On Mon, Aug 04, 2008 at 12:49:43AM -0700, Keith Packard wrote:
> > I'm switching the 2D driver over so that it uses GEM objects for
> > pixmaps. I've ripped exa apart and created 'uxa' which doesn't deal with
> > pixmap allocation at all, and just focuses on acceleration. That's
> > working fine, and now the intel driver can create GEM objects for every
> > pixmap.
> 
> wouldn't the pixmap creation be too expensive compared to traditional
> allocation?
> 
> And another question related to your blog:
> > fast writes and really slow reads, but no penalty for switching rapidly
> > between GPU and CPU
> 
> is the read penalty still noticable if you read doublequadwords using sse 
> and also explicitly prefetching data?

Past use of mmx, sse, and friends for fb reads has seen on the order of
10% improvements according to comments I've read.  You can't prefetch
since they're uncached reads.

-- 
Eric Anholt
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: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20080806/794d6495/attachment.sig>