Hate to jump in before I have all the facts (usually means a major rash of
foot in mouth disease), but modular/XAA is definitely the way to go. It's
lightweight and fast and far removed from the tight PCI & x86isms of the
past. XFree86 4.0 will not be shipping any of the old servers (such as the
old S3, S3V or Mach16 or similar).

The new modular architecture that is coming to fruition (I worked on the
earliest builds 3.9 when MetroLink's stuff was merged in) uses a new version
of XAA that is not as dependant or fussy about architecture. It does
pre-suppose PCI or AGP right now, but the bus stuff is now modularlised, and
therefore ripe for additions. THe old monolithic code was/is a nightmare in
this context. Therefore NetBSD could make a driver for a specific driver &
architecture (such as alpha or 68k) and it would work on all binary
compatible ports. It also means that the users would have access to the
latest XFree86 builds, whilst still using a NetBSD approved driver.

The new modular code is not only faster (we saw about a 15% immediate boost
on my MGA Mill 1 from the already _VERY_ fast 3.3.x server) but very
lightweight in terms of memory usage because it only loads those things that
are in use at the time, rather than the entire lot.

The other stuff is we definitely want to pursue DRI and the other 3D stuff.
Alphas kick ass when it comes to floating point, let's get the platform one
of choice for engineering work. If we stick to the old servers, you will not
get the free rides that are coming in XFree86 4.0

Andrew

-----Original Message-----
From: port-alpha-owner@netbsd.org [mailto:port-alpha-owner@netbsd.org]On
Behalf Of Jason Thorpe
Sent: Friday, 22 October 1999 6:57 AM
To: R. C. Dowdeswell
Cc: Bernd Ernesti; port-alpha@netbsd.org
Subject: Re: XFree86 on NetBSD/alpha


On Thu, 21 Oct 1999 13:43:35 -0700
 "R. C. Dowdeswell" <elric@mabelode.imrryr.org> wrote:

 > Yeah, although I did have a look at said code -- it doesn't really
 > look like there is a lot of acceleration in there.  (Only solid
 > fills).  I've been planning to put my accelerations in XF86 when
 > we move over to it.

For TGA, we should not use the current XF86 code.  We should use our
Sane Framebuffer Model when possible, and probably change XalphaNetBSD
to be XdecNetBSD, because it could work w/ e.g. SFB/SFB+, too.

        -- Jason R. Thorpe <thorpej@nas.nasa.gov>