[Intel-gfx] What's the story with Gallium and Intel GPUs?

[Eric Anholt]

On Tue, 26 Jan 2010 19:42:12 +1000, Dave Airlie <airlied at gmail.com> wrote:
> On Tue, Jan 26, 2010 at 7:06 PM, Olivier Galibert <galibert at pobox.com> wrote:
> > Or that could be "Is there a story in the first place?" :-)
> >
> > I'm interested in intel gfx hardware (it's what I have in my laptop),
> > and I'd like to see if I can help things up to the point where I can
> > play with geometry shaders and stuff.  Maybe I can, maybe I can't,
> > time will tell.
> >
> > One of the things I'm trying to figure out is the place of Gallium in
> > all that.  Having it in the separate directory with separate drivers
> > in the mesa git is a little strange, and google only says that the
> > intel gfx devs are not interested by gallium at that point.
> 
> Thats not that strange, gallium drivers are separate drivers they
> don't sit in the same place in the tree as they are different.
> 
> >
> > One the other hand, it looks like it may be a requirement for GLSL
> > 1.5/OpenGL 3.2 (in the future, obviously).
> 
> Not really. Gallium is a driver architecture, it doesn't mean its
> impossible to do things in other ways.
> 
> >
> > So what is it?  A future requirement?  An alternative way of doing
> > things?  Is the interest in gallium for intel gpus only deferred until
> > kms/uxa is considered stable or there is an active rejection going on?
> > And if so, where should I look for the technical reasons, if any?
> 
> Gallium is a new driver architecture, which means the drivers have to
> be written from close to scratch to use it. This means introducing a lot
> of regressions and expending a lot of time to get the gallium  based mesa
> driver back to the same level as the current Intel Mesa driver. So
> far Intel haven't decided that the end outweighs the transisition.
> The Intel drivers have undergone a pretty major rewrite already and really
> only just started coming back out the other side about 6 months ago,
> doing this again is probably not the easiest path to happiness.
> 
> So I don't think the Intel devs are willing to undergo this process again
> so soon, there may be some point that the advantages of the G3D
> architecture outweighs the pain and suffering in moving a driver to it.
> 
> There are community maintained i915g/i965g drivers (mostly VMware
> devs) and there is plenty of work on those to be done.

This.  Converting to gallium is a *ton* of work (see various starts and
abandonments of doing gallium intel drivers already), and the gain at
the end is... well, there's no clear gain.  If somebody steps up and
produces a gallium driver that's faster/more conformant, then we'd be
interested, but I've seen no indication that that would happen.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20100126/4ad43680/attachment.sig>