AccelDFS issues with which AGP bridges?

[Alan Swanson]

On Wed, 2006-12-20 at 08:22 +0100, Michel Dänzer wrote:
> On Tue, 2006-12-19 at 21:39 +0000, Alan Swanson wrote:
> > On Tue, 2006-12-19 at 21:05 +0000, Alan Swanson wrote:
> > > Regarding the AccelDFS option defaulting to off for AGP due to issues
> > > with some AGP bridges, I'd like to know which AGP bridges those are?
> 
> If we knew exactly which ones are problematic, this option wouldn't be
> necessary...

Half of my query was really so that anybody wondering the same had an
answer in the future.

> > > I've enabled this with my two systems without observing any issues 
> 
> Your AGP bridges obviously aren't affected then. Most of them probably
> aren't, still we can't enable this by default without a complete
> blacklist of broken bridges - or a whitelist of known working bridges
> maybe...
> 
> > > with yesterdays GIT (except EXA still being horrible when switching
> > > workspaces with Gnome but perhaps slightly less so).
> 
> Try the xserver exa-damagetrack branch if you haven't.

Been giving it a whirl with yesterdays changes after figuring out what
code is different and using the following as a patch to the 7.2 branch.
I've added it to the bug 6929 to help any other testers.

$ git diff server-1.2-branch..exa-damagetrack exa miext/damage

Still takes a second or two when switching workspaces with Gnome;
Metacity non-compositing with Nautilus background and running various
terminals, Galeon, Evolution and other little apps. It feels like it's
chucked out all the pixmaps for a workspace on switching out and then
chunders away reloading them when switching back.

> > Desktop: AthlonXP 2500 - Nvidia Nforce2 - 8500
> > 
> >   GetImage10x10   9120.0/sec 14200.0/sec 56%
> >   GetImage100x100  154.0/sec   660.0/sec 329%
> >   GetImage500x500    6.1/sec    24.5/sec 302%
> > 
> > Laptop: Pentium-M 1.6 - Intel 855PM - Mobility 9000
> > 
> >   GetImage10x10   10900.0/sec 17700.0/sec 62%
> >   GetImage100x100   177.0/sec   801.0/sec 353%
> >   GetImage500x500     7.6/sec    31.3/sec 312%
> 
> Not bad, is it? :)

A distinct improvement!

-- 
Alan.

"One must never be purposelessnessnesslessness."