[Paraview] Coprocessing with accelerators

[Biddiscombe, John A.]

>
But yes Mark you are right. As it stands there will be a round trip in between the simulation and vis stages. We will have to extend Catalyst with the ability to create a vtkPistonDataObject that refers to the data already on the GPU in order to avoid it.
>

Since piston is a very thin wrapper around a dataobject and is easy to customize for your own specific requirements, passing a device ptr handle from your simulation code into a piston data object to be further processed on the GPU will be trivial. Where you may have trouble is doing the render on the GPU in the absence of a Xserver (if you're using cray XE/XK nodes for example).

JB
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20130820/871137a6/attachment.htm>