Teleconference 2018-05-16

From VTKM
Jump to navigation Jump to search

Attendees: Ken Moreland (SNL), Abhishek Yenpure (UO), Manish Mathai (UO), Matt Letter (SNL), Matt Larsen (LLNL), Rob Maynard (Kitware), Allie Vecanti (Kitware), Haocheng Liu (Kitware), Sujin Philip (Kitware), Berk Geveci (Kitware), Ollie Lo (LANL)

Matt Letter has a MR for a fix for gradient vector output.

Rob as added a new multiblock example that uses multiple threads that distributes the work. Also shows other threads producing new work.

Wavelet compressor tests has been failing for mysterious reasons. An update to CUDA 8 seems to fix the problem.

Working on getting virtuals to work on Pascal hardware.

Sujin has been updating VTK-m in VTK. Now working implementing diy serialization routines so you can send data sets across MPI processes.

Allie is wrapping up some long-term branches.

Haocheng started working on VTK-m yesterday. Working on adding a float version of vtkm::Pi.

We talked a bit about CellLocator and PointLocator design (https://gitlab.kitware.com/vtk/vtk-m/issues/224). One of the biggest changes suggested was to use dynamic classes with virtual methods. There was an extended discussion on managing the lifetime of dynamic objects in the control environment. The consensus was in the short term to just build the locator hierarchy and leave the management to the users. If necessary, we can later look into creating a management class (e.g. DynamicCellLocator).