DMX and 3D Compositing Questions

[Christian Pirchheim]

Hi,

Creating large tiled displays using multiple distributed X-servers can
be done using the DMX-server [1]. DMX integrates with Xinerama extension
to pretend a single large display to client applications which are
actually rendered on multiple displays. Each of these X-displays is
provided by one backend X-server, DMX opens a fullscreen window on each
of the displays and manages I/O. DMX is often combined with Chromium
which allows for rendering fullscreen 3D applications.

Now, here comes my question: What does it take to make DMX compatible
with 3D compositing window managers? Background of my question is, that
we would like to create large seamless displays combining multiple
overlapping projectors. Within the composite window manager, the desktop
geometry and window textures are available, allowing to compensate for
geometric and photometric distortions.

The basic steps in 3D compositing are (1) redirection of window pixmaps
to off-screen buffers (Composite), (2) access those pixmaps as GL
textures (EXT_texture_from_pixmap) and (3) render the composite desktop
via OpenGL calls (GLX/OpenGL) indirectly or directly.

First, obviously, each backend X-server for itself must be
3D-compositing capable, including convenient graphics hardware and
drivers, providing the required X extensions (Composite, Damage, RandR,
XSync) and GL extensions (GL_EXT_texture_from_pixmap,
GL_ARB_texture_non_power_of_two, GL_EXT_framebuffer_object) and direct
rendering support.

But then? For (1), the Composite extension is required, which is
currently not supported by neither, Xinerama nor DMX. I assume,
Composite needs to be integrated in both. According to [1] DMX
implements GLX and supports both indirect and direct rendering on its
multiple backend servers. That fact sounds great, but what does it mean
concerning (2) and (3). Does AIGLX (indirect accelerated rendering) have
any relevance in this context?

I would be glad about any comments concerning these issues which will
help to understand the problem and also the correlations between the
various technologies.

Regards,
  Christian


[1] Martin, K. E.; Dawes, D. H. & Faith, R. E. Distributed Multihead X
Design VA Linux Systems, Inc., Red Hat, Inc., 2004