Teleconference 2017-11-29

Jump to navigation Jump to search

Attendees: Ken Moreland (SNL), Matt Letter (SNL), Ollie Lo (LANL), Rob Maynard (Kitware), Sujin Philip (Kitware), Dave Pugmire (ORNL), Tom Otahal (SNL), Aaron Knoll (NVIDIA), Mark Kim (ORNL), Cyrus Harris (LLNL), Matt Larsen (LLNL), James Kress (ORNL)

Ken will send Ollie an email about updates for ECP.

Matt Letter is working on the CMake refactor. The refactor currently requires CMake 3.9 if CUDA is enabled. One of the CUDA dashboards is failing because CMake is 3.6. There are also some dashboards failing because it defaults to OpenGL rendering backend. Instead, it should default to None.

Is VTK-m's integration with ParaView and VTK undergoing substantial changes? There is an external ECP milestone to integrate VTK-m through Catalyst through Nalu. Rob reports that the big change is that there will be an explicit VTK-m dataset in VTK. However, this should not significantly impact code that uses the Catalyst interface.

Recent ParaView master is now using VTK-m for LOD all the time.

Rob has been reworking the dispatch and is starting to remove FunctionInterface from within the dispatcher code (but not fully removed).

Sujin is working on converting the coordinate system classes to use virtual methods. We are starting with this (rather than more general ArrayHandle/fields) because there is a fixed set of types that we support. There are some complications with the NVIDIA hardware with virtual method implementations not being available on the device.

Dave P. is working on multiblock documentation.

Yesterday ORNL had converstion with NWChem application to duplicate for them what was done for fusion in the showcase.

Matt Larsen continues to refactor raycasting. Also making pretty images for Ken and is resurrecting path tracer.

Matt Larsen also has an MR to wrap device adapter algorithms so that you don't have specify a device adapter (use TryExecute instead).