Teleconference 2015-06-03

Attendees: Ken Moreland (SNL), Rob Maynard (Kitware), Dave Pugmire (ORNL), Jeremy Meredith (ORNL), Berk Geveci (Kitware), Sujin Philip (Kitware)

There was a recent email thread about changing the behavior of ArrayHandle to allow writing to user arrays (created with, for example, make_ArrayHandle) although not necessarily be able to resize. No one had any major objection to this although there may be some complication with the storage becoming invalidated data is pushed to the execution environment.

There was quite a bit of talk about being able to pass an ArrayHandle to a worklet for random access. This can already be done indirectly by creating an ExecObject, but it would be convenient to do it directly. There was no argument for doing so for read only, but there was much discussion about being able to write to such an array since that provides a very easy way to shoot yourself in the foot.

The main use case for random access writes is to implement a scatter operation. One alternate suggestion was to implement a formal scatter operation in the device adapter. Another was to use the permutation array to implicitly scatter values given an index map. Rob will look into both ideas.

Gitlab is ready to go; it just needs a switch flipped. That will happen... soon.

The data model branch is pretty much ready to be merged into master. That will happen momentarily. Users should feel free to begin using the data model, but be prepared to bleed a bit as the API may need to be changed as issues arise.

There is now a TBB device adapter merged into master.

Kewei, Chris' student, is working on streamline and streamsurfaces in VTK-m. There may need to be some organization from the Oregon folks, although they are not currently working on this.