Attendees: Ken Moreland (SNL), James Kress (OU), Chris Sewell (LANL), Pat Fasel (LANL), Berk Geveci (Kitware)
Pat is gaining some steam on implementing statistics in VTK-m. This work will require some additions to the device adapter (particularly a reduction operation). The reduce will probably look similar to the current scan interface, which will probably be generally OK.
There is also some thought about including imaging algorithms in VTK-m.
- Is this appropriate? There is, after all a clear distinction between VTK and ITK. There might be good reason for that.
- Berk: ITK has lots solvers, which are probably not a good fit for VTK-m.
- There are also lots of stencil-based algorithms like convolution that apply an operation to a local neighborhood.
- These types of algorithms would probably fit in VTK-m.
- Several algorithms like this in VTK imaging pipeline. Still commonly used.
- That said, these algorithms are already pretty fast. There might not be a huge speed advantage to run on the GPU.
- Splatting might be a good use case.