EXA patches

[Michel Dänzer]

On Wed, 2009-08-05 at 01:39 -0400, Alex Deucher wrote:
> 2009/8/4 Michel Dänzer <michel at daenzer.net>:
> > On Tue, 2009-08-04 at 11:54 +1000, Dave Airlie wrote:
> >> On Tue, 2009-08-04 at 09:49 +0800, Joel Feiner wrote:
> >> > 2009/8/4 Michel Dänzer <michel at daenzer.net>
> >> >         <snip>
> >> >
> >> >         I wonder if maybe the slowdown I'm seeing is because the
> >> >         radeon driver
> >> >         is temporarily lacking UploadToScreen and DownloadFromScreen
> >> >         hooks with
> >> >         KMS, though I'm not sure what those would be used for with
> >> >         text
> >> >         rendering.
> >> >
> >> > Off-topic, but if I may enquire out of intellectual curiosity: why
> >> > doesn't the KMS version of Radeon have UTS and DFS acceleration?
> >>
> >> The code is in the branch, but its really ugly, I was hoping to actual
> >> do it cleaner which might involve a new kernel interface to ask for the
> >> current placement of a buffer object so we can decide whether to just
> >> memcpy or we need to blit for DFS. I'm not sure I saw a good reason for
> >> UTS, doing Host data blits isn't really useful with BOs at least with
> >> the current code.
> >
> > One potential benefit of UTS is pipelining. But I agree that DFS is
> > probably more important.
> 
> FWIW... untested.
> http://www.botchco.com/alex/xorg/radeon_kms_uts.diff

You want to avoid any explicit flushes or even waits for destination BO
idle, those would defeat any potential pipelining benefits. Otherwise
looks like it could work though. :)


-- 
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer