Attendees: Ken Moreland (SNL), Abhishek Yenpure (UO), James Kress (ORNL), Matt Larsen (LLNL), Rob Maynard (Kitware), Allie Vacanti (Kitware), Tom Otahal (SNL), Dave Pugmire (ORNL), Mark Kim (ORNL), Manish Mathai (UO/LLNL)
Last week James and Dave P. went to ADIOS hackathon at Kitware and got some hacked together work with VTK-m. Were able to get rendering out of ADIOS memory buffer. Dave P. is also working on a standalone library to read ADIOS files into VTK-m. Not sure where that will go.
With recent additions of log scale plots, James' integration to Xlotl is pretty much complete.
Matt added capability in connectivity tracer to integrate emission into X-Ray rendering.
There was a discussion about moving the "raycast" canvas as a concrete base class to all canvases.
Tom O. is going to look at implementing radix sort in the TBB device adapter.
Abhishek and Dave P. have been working on streamlines and maintaining history of particles. Trying rounds based on number of particles and rounds based on number of steps. Have been doing a study of these approaches with unified memory. Been dealing with how to keep track with how many particles are in a worklet and how many have stopped.
Allie is working on making composite arrays writable. This is on hold because MSVC 2013 does not support the appropriate C++ features.
Allie has merged the Results classes. Instead of IsValid there IsDataSetValid and IsFieldValid. That is a bit confusing because there is no more IsValid. The running idea is that there should still be an IsValid.
Allie is working on interop. You no longer need to specify the device when doing interop. It now figures out what device the data are already on.
Sujin has been working on cell set explicit. The build capability is compiled in.
Sujin is now using CUDA streams instead of the default streams.
Rob also refactor the CopyInto of the ArrayHandle (i.e. got rid of them).
More refactoring of using over typedef.
Working on the point gradient/imaging/neighborhood worklet algorithms.
Sujin has done some benchmarking on comparing using real virtual methods vs our faux virtual methods. On the newest CUDA architectures the difference in runtime is quite small. Sujin is in the process of converting to the real virtual methods.