On Thu, Jun 03, 1999 at 12:04:02AM -0400, Andrew Gillham wrote:
> ./quake2: can't resolve symbol '__register_frame_info'

I haven't tried q3test..

I just want hardware 3D to improve 'xlock -modelist allgl' performance :-)

> I'll be working on building glx under NetBSD/i386 later tonight, or 
> tomorrow.  I don't want to try to reinvent the wheel if someone is already
> working on this.
> 
> So some questions:
> 	1.  Is it likely the diffs to the XFree86 nvidia driver will be
> 	    put in the tree, or will we wait for XFree86 to incorporate
> 	    them, and import their code?

The NVidia source code hasn't been released yet as far as I know, neither 
the 2D improvements or the 3D code.

Once it is, I assume the 2D changes will be put in xsrc if sufficiently
significant and taken from the next XFree86 release if less important,
as with other drivers.

> 	2.  What about the diffs to Mesa-3.0 provided by nvidia?

Don't know, but I assume the Mesa (and in turn, XFree86) people will
be integrating them.

> 	3.  Is there someone more Xpert (pun intended) who is planning on
> 	    porting glx?  Particularly the nvidia glx code.

The impression I have is that most of the work/code is in the GLX
hardware driver.

I seem to recall (not quite sure) that NVidia's GLX code is based on
Precision Insight's 'direct rendering' code which is also not freely
available yet (but will be quite soon, I believe).

Anyway, XFree86 (including the version that is in xsrc) can use
a GLX module (such as the G200 one mentioned below) by just using
'Load "glx.so" in 'Section "Module"'. Since this stuff is also
currently very much in flux, I think there is little point in
integrating it or even making packages from it.

> 	4.  Does anyone else care about having hardware 3D under NetBSD?

I play around with the Matrox G200 GLX module
( http://reality.sgi.com/ripperda_engr/glx/ ) and it works reasonably
well and is improving very quickly.

I do have it working natively with the xsrc XF86_SVGA. There are/were
a few minor Linux-isms in the build process. It also seemed to surface
a i386 binutils bug that I haven't looked into yet, but with a little
build hand-holding it does work.

[BTW, (possibly flamebait, sorry) these people are very eager to
use PCI bus-master DMA. From userspace that is. Any thoughts on this
from a NetBSD perspective?]


-- 
Soren