[Intel-gfx] 3D hardware acceleration through the server?

[Bob Schulman]

Is there a code path in the X11R7 code base available from
intellinuxgraphics.org, where hardware acceleration for GL rendering would
occur in the X server rather than on the X client side? If so, what bread
crumbs can you toss my way so I figure out how to do a build for the
server-side so that GL requests which come over the wire get parsed and
dispatched, and then eventually bubble down to the DRI layer?

Or...is _all_ DRI magic strictly client-side? (If all DRI magic is
client-side, well, umm, why is that? Why preclude over-the-wire clients from
getting hw acceleration?)

Thanks.

bob

-----Original Message-----
From: Dave Airlie [mailto:airlied at gmail.com] 
Sent: Monday, October 12, 2009 12:14 AM
To: Bob Schulman
Cc: intel-gfx at lists.freedesktop.org
Subject: Re: [Intel-gfx] silliest of questions

On Mon, Oct 12, 2009 at 1:39 PM, Bob Schulman <bob at rexdog.net> wrote:
> For the Intel graphics source that's available through
> intellinuxgraphics.org, I cannot figure out if accelerated OpenGL is
> supported through your supplied source code (available through
> freedesktop.org) or not. I see that DRI is supported in the source code
> base, but I don't see where OpenGL API calls are transformed into pokes of
> 3D registers in any of your source. Am I missing something?
>
>
>
> From what I can tell, OpenGL is supported for this source base through
Mesa,
> but I just don't see where fastpaths through (or around) the Mesa code get
> set up so that accelerated rendering, using the 3D registers of whatever
> Intel part has been chosen, get set up.

All in Mesa, the GL api goes mostly via magic libGL stuff into the DRI 3D
driver
which has entrypoints in places in mesa src/mesa/main directory,
these call hooks in the driver function table, which in Intels case call
drivers
in src/mesa/main/drivers/dri/intel,i965,i915

Dave.