Hi,
On 3. May 2011, at 22:10, erij89 [via Software] wrote:
> I'm new here and I have a few general questions in order to figure out if
> Equalizer is applicable for what we want to use it for.
Welcome! The goal of Equalizer is to solve the *generic* problems of writing a
parallel and potentially distributed OpenGL application. If it is not
applicable, we should figure out why and change it. ;)
> We are trying to build an out-of-core rendering server (stateless!) that
> should be able to generate different GBuffers depending on the request (Query
> String). Those GBuffers are e.g. color, depth, objectID images of a
> particular view. The rendered image should be postprocessed - if needed.
> We are currently using osg for that and a pp-library that works directly on
> the imagedata on the gpu.
Just to make sure: The render server delivers the GBuffers to the client, not
simply a rendered image - correct?
> So if we would split the image, so that a quarter of it would be rendered on
> a different gpu/pc we would not have the data of the pixels right next to the
> splitting area. Is there an easy way to enlargen the area each gpu renderes
> (so that there are enough pixels beyond the corners) and after that, put the
> image together?
>From my current understanding I would first assemble the partial images on the
>destination, and then run the pp on it. My assumption here is that the pp does
>not reduce the amount of data. If it does, it might be worthwhile to do the pp
>on the source channels.
> Is there an easy way to use our existing osg code with shaders that genereate
> the pbuffers for every Node?
Yes, just do it. :)
Of course there might be details to work out, but in the draw task method you
have native access to a GL context. Have a look at osgScaleViewer for OSG
integration.
> The most important thing is - if splitting an image - if it would be possible
> to put it back together, even if there are 3-4 representations of the same
> image (color, depth, normals, objectid) ?
You would have to write your own implementation of Channel::frameReadback and
Channel::frameAssemble to read and assemble the different representations using
the correct source/destination, but this is hardly rocket science.
> We are using mutliple GB of Model files and Texture data. Can all
> Equalizer-Nodes that that data, so that it would not have to be loaded
> multiple times (if using one pc)?
One one PC you would only configure one node (process) with n pipe (threads),
one for each GPU. The pipe threads naturally have access to shared data.
HTH,
Stefan.
--
View this message in context:
http://software.1713.n2.nabble.com/IPP-and-GBuffer-generation-and-usage-tp6328643p6329824.html
Sent from the Equalizer - Parallel Rendering mailing list archive at Nabble.com.
_______________________________________________
eq-dev mailing list
[email protected]
http://www.equalizergraphics.com/cgi-bin/mailman/listinfo/eq-dev
http://www.equalizergraphics.com
