PVPiston
– GPU accelerated filters for ParaView
This
plugin integrates LANLÕs
Piston (http://viz.lanl.gov/projects/PISTON.html), GPGPU accelerated filters
into ParaView.
To
use it you must compile ParaView with the
PARAVIEW_USE_PISTON cmake configuration option. Once compiled
and loaded new filters appear in the Piston submenu of the Filters menu. Note
that currently the plugin is only compatible with
standalone, serial ParaView runs.
To
use Piston, first apply the SendToPiston filter. This
filter moves data from the CPU onto the GPU where the rest of the filters do
their work. Currently, this filter is only applicable to uniform grid data that
has a single floating point scalar array. Polygonal
data is also accepted, but there are no Piston filters yet that
operate on this type of data.
Next
apply the Piston sort, threshold, slice or contour filters to process the data
on the GPU. Finally, apply the GetFromPiston filter
to bring the data back to the CPU, where it can be further processed with
standard VTK filters.
Data
produced by the piston filters is rendered via the vtkPistonMapper.
If you start ParaView with the:
Ò--use-cuda-interopÓ argument, the mapper will do its work directly in GPU space without first
transferring the data back to the CPU for rendering.