Teleconference 2017-01-18

Attendees: Ken Moreland (SNL), James Kress (ORNL), Matt Larsen (LLNL), Alberto Villarreal (Intel), Dave Pugmire (ORNL), Tom Fogal (NVIDIA), Rob Maynard (Kitware), Dave Lonie (Kitware), Ollie Lo (LANL), Tom Otahal (SNL), Chris Sewell (LANL), Mark Kim (ORNL)

ECP reporting. Need to report monthly on progress. Ken needs to gather progress on milestones on a monthly basis. The question is should we do this at the regular VTK-m meeting or have a separate PI meeting.

Ken is introduces a reduce by key worklet.

James is working through checking in tests to his builder classes.

Matt is working on using raycasting to implement radiography and adding new features as necessary.

Matt is looking into doing ZFP compression in VTK-m.

Alberto is from Intel's software development group. He is our point of contact for Intel architecture.

Dave P. is getting prepared for the git LFS move to allow us to check in larger data files into VTK-m (generally for testing). From there we will be looking at using rendering to do image comparisons for testing.

Looking into streamlines testing.

Rob has been looking into the VTK-m/ParaView integration. Is teasing out some bugs and working out issues.

Rob also working on reducing vtables and other exports.

David L. is working on an alternative to virtual methods for CUDA that get around a problem of having to malloc on the device. It basically binds function pointers.

Ollie is fixing a bug in floating point atomics.

Tom Otahal working on a filter wrapper for the magnitude worklet. Noticed that the magnitude worklet only took Vec3, so expanding it to include other Vec types. Notices that when passing in a vec of integers, got integers. Resolved to convert integer types to vtkm::FloatDefault. Magnitude of integers makes little sense. This should probably change in VectorAnalysis.h.

The contour tree feature is now officially merged into VTK-m.

Unified memory code is available in the repository as a merge request but not merged. There are probably several changes that should be made first. Ollie will be taking over this work.

Chris is working on highlight slides for contour trees and streaming.

We have all agreed to move to git LFS to store data files and baselines. We expect to have all data using LFS to be in a single directory, probably named "data". Rob will set that up.