From: Bryan Meredith <bryan@bear-mountain.demon.co.uk>
  To  : ggi-develop@eskimo.com
  Date: Fri, 05 Feb 1999 23:34:13 +0000

Re: KGI - mmap of accel registers to user space.

James Simmons wrote:
> 
> On Fri, 5 Feb 1999, Jos Hulzink wrote:
> 
> > On Thu, 4 Feb 1999, Bryan Meredith wrote:
> >
> > > Dear All,
> > >
> > >     this doesn't seem to be implemented at the moment.
> > >
> > >     Is anyone working on it?
> > >
> > >     Is it worth working on, given that James and Fabrice are cooking
> > > something up (along with many others ...).
> 
> Actually its me and Fabrice right now plus jeff from SGI has shown a
> interest in the patch. The patch is under alot of fire plus its meant for
> the 2.3.x kernel. Their is a good chance that the patch could get rejected
> or the accel part get removed. Also right now the way the patch works is
> it batches accel commands and then sends them in ioctl call. This reduces
> the overhead of context switching. It doesn't mmap accels. It was done
> that way to avoid the problems with poorly designed hardware.
> 
> > I'm testing this on the ViRGE driver (by increasing the framebuffer size
> > from 4 MB to 32 MB), but at the moment any attempt to get 2D accel working
> > fails (The accel engine actually does something, but it far from what it
> > should do).
> >
> Virge is a very touchy card. Its one of those poorly designed cards.

I am lucky enough to have a matrox mil2 - this is not a touchy card
AFAIK.

> 
> > I'm wondering if the mmap mechanism is the reason this fails. I'll keep
> > you informed.
> >
> Where is your code for this ?

Apart from requiring a call to be implemented by each driver in order to
get the address and length of the region, it should be straight forward
to add this to the kgi stuff that is currently in the archive. Is it
worth me producing a patch (for kgi and the matrox driver) or is this
not the right thing to do at the moment?

Bryan
-- 

Oh Bugger...