Teleconference 2017-01-04

Revision as of 18:36, 4 January 2017 by Kmorel (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Attendees: Ken Moreland (SNL), Matt Larsen (LLNL), Rob Maynard (Kitware), Berk Geveci (Kitware), James Kress (ORNL), Dongliang Chu (ORNL), Chris Sewell (LANL), Oli Lo (LANL)

Matt is doing a major refactor of ray tracing to support other work. Separating some of the pieces. For example, templating everything on ray precision (32 vs. 64-bit). Removing dependence on mapper so that multiple mappers can use same template compile.

Rob has been getting VTK branch with VTK-m accelerator get all the tests pass and get acceptance.

Current gradient filter is failing and not getting the same results as the VTK. Needs Ken's help in resolving this.

Work on VTK-m cont library. Have several array handles baked into it already. On OSX symbol visibility is different than other OS's. Rob has a merge request that is a step in the right direction.

The latest version of TBB was reporting errors with exclusive scan. Other errors were found as well. Rob has an MR open for increasing the size of arrays used during device adapter tests so that multiple threads are used as well as several MRs to fix problems.

James is working on new tests for data set add. Moving on to particle stuff. Initially data set representation.

Dongliang comparing Marching Cubes between VTK and VTK-m with Oxbow.

Chris comparing unified memory with streaming arrays on new Pascal architecture. The unified memory is actually faster on the new hardware. However, this is not yet in VTK-m. To use unified memory, you have to do a cuda malloc managed.