[Spice-devel] OpenGL "passthrough"

Alon Levy alevy at redhat.com
Wed Nov 9 09:06:17 PST 2011


On Wed, Nov 09, 2011 at 04:57:23PM +0000, Mosebach  Kai wrote:
> Both would be very helpful :-) And important from a market/competitive
> point of view.
> 
> Please let me have a look at the patches if possible,
> 

I'm so far from that the mere mention of it here was just a result of
asking. It is a thought of a thought.

> Thanks Kai
> 
> On 11/9/11 5:54 PM, "Alon Levy" <alevy at redhat.com> wrote:
> 
> >On Wed, Nov 09, 2011 at 03:57:30PM +0000, Mosebach  Kai wrote:
> >> On 11/9/11 3:51 PM, "Alon Levy" <alevy at redhat.com> wrote:
> >> 
> >> 
> >> >On Wed, Nov 09, 2011 at 01:52:12PM +0000, Mosebach  Kai wrote:
> >> >> Hi guys,
> >> >> 
> >> >> Just a though :
> >> >> 
> >> >> In a scenario where a kvm-host system has a openGL capable display
> >> >> adapter, wouldn't it be possible (and relatively simple compared to a
> >> >>full
> >> >> VGA passthrough) to just
> >> >> pass the OpenGL interface/access into the VMs QXL Adapter making this
> >> >>one
> >> >> (HW accelerated) OpenGL aware?
> >> >
> >> >There are threads in qemu-devel about using pci passthrough for using
> >> 
> >> I tried for long time now to pass through PCIe Hardware into the guest
> >> with all kind of PCIe cards w/o too much off success. So that¹s not
> >>what I
> >> want ;-)
> >> 
> >> >the host GPU from the guest. If you actually want to view the results
> >> >then yes, it's relatively simple in theory. Maybe also in practice, but
> >> >I've been struggling with a slightly larger problem, doing OpenGL and
> >> >Direct3D passthrough, for almost a year now, with no results to show :/
> >> >
> >> >I do have patches on my harddrive somewhere by Izik Eidos that do a
> >> >proof of concept for GL passthrough from a windows guest to a
> >> >windows/linux client, I could ask him if it's ok to post them if you'd
> >> >like. (and if I can find them).
> >> >
> >> >Basically it adds qxl messages for all the GL commands, adds a surface
> >> >type, tracks all commands from the last SwapBuffers (I guess
> >> >wglSwapBuffers - the equivalent of glXSwapBuffers) per surface, and
> >> >passes all of those to the client. Rendering on the server, that is the
> >> >host, will be a little simpler.
> >> 
> >> So (if I got you right and you got me right) we talk about the QXL
> >>device
> >> on the KVM guest receiving OpenGL commands and passing them (thru
> >> qemu-kvm) downwards to the OpenGL capable hardware on the KVM Host
> >> (server). The client (spicec or similar) should not be involved at all.
> >>Is
> >> there any work in progress (for rendering on the server) - similar to
> >>the
> >> RemoteFX project of HyperV?
> >> 
> >
> >I know I'm working on enabling both server side and client side
> >rendering, not aware of anyone else.
> >
> >> 
> >> Cheers Kai
> >> 
> >> >
> >> >> 
> >> >> Thanks for enlightenment,
> >> >> 
> >> >> Kai
> >> >> 
> >> >> _______________________________________________
> >> >> Spice-devel mailing list
> >> >> Spice-devel at lists.freedesktop.org
> >> >> http://lists.freedesktop.org/mailman/listinfo/spice-devel
> >> 
> 


More information about the Spice-devel mailing list