Teleconference 2017-02-15

Attendees: Ken Moreland (SNL), Matt Larsen (LLNL), Tom Otahal (SNL), James Kress (ORNL), Rob Maynard (Kitware), Berk Geveci (Kitware), Dave Lonie (Kitware), Tom Fogal (NVIDIA), Ollie Lo (LANL), Mark Kim (ORNL), Hank Childs (UO), David Rogers (LANL)

Hank reports that the first full week of April 4-6 should work for a code sprint at UO @ Eugene. Hank will take the lead on logistics.

Tom Otahal is looking at a benchmark for image processing at look at how will it is working on KNL.

Tom reports there is a Kokkos-enabled version of SPARTA that runs on Trinity. That could be a good motivation to get VTK-m working with Kokkos.

Matt brought up ALPINE version to latest version of VTK-m.

Matt is running into some issues with the standard C++ library. There appears to be some mis-match between standard library being used and the compiler version.

Tom Peterka will be working on using their feature tracking methods in CODAR. Plan to incorporate that into VTK-m.

Dave L. pulling things into the vtkm_cont library. He ran into a dependency issue with the device adapters.

Rob working on issues with VTK-m/VTK Accelerator module. Last issue was with the renaming of exceptions. Latest RC of ParaView offers a plugin that makes VTK-m filters available. Currently built with TBB backend.

Tom F. has some merge requests that he needs folks to look at. Tom also highly encourages us to move to asynchronous scheduling. During his latest timings he found that the majority of time was spent while synchronizing the device.

Ollie has been benchmarking unified memory. Directly allocating CUDA memory has no performance difference from using a thrust vector. Ollie has been having issues with nvprof on the benchmark programs. Oddly, our benchmark code seems to crash with nvprof.