Teleconference 2019-01-09

Attendees: Ken Moreland (SNL), Robert Maynard (Kitware), Dave Pugmire (ORNL), Abhishek Yenpure (UO), Berk Geveci (Kitware), Matt Larsen (LLNL), HaoCheng Liu (Kitware)

ECP notes and announcements

• December milestones were completed and submitted on time. Thanks to everyone.
• Annual meeting is next week
  • Jim Ahrens will give feedback from our October review Wednesday @ 11:30 (location TBA?).
    • PIs should be there. Others working on the project are welcome to attend.
  • We will have an organization meeting to discuss FY20-FY22 work sometime Wednesday.
    • PIs should be there. Others working on the project are welcome to attend.
    • If you haven't, vote for time at doodle poll: https://doodle.com/poll/qttxkkz6wuqn9aa7
    • Notes on confluence at: https://confluence.exascaleproject.org/display/STDA05/FY20-FY22+Planning

ECP deliverable updates:

• [MS-19/04] Ghost Cells
  • Dave added a filter that removes ghost cells
  • Just made a filter that adds a ghost layer array to structured grids (to mark external areas as ghost).
  • It was suggested that for completion of the milestone, the histogram filter should be modified to optionally accept a ghost cell array to ignore inputs from ghost cells.
• [MS-19/05] Merge Points
  • Ken is on the hook for merging the Merge Points worklet in.
  • The functionality will actually be part of clean grid.
• [MS-19/06] Connected Components
  • No one from LANL is here.
• [MS-19/07] Advect Time Varying
  • Not much to report right now.
  • Some preliminary work has been done.
  • Most of what we need is already there.
  • Dave P. does not see any problem with getting this done.

There has been some discussion on the mailing list about supporting a mask on the output of a worklet. In this case the worklet would only be executed on a select set of entries of the output. This is different than a scatter operation, which compresses the output to only those generated.

Virtuals were merged in on December 28. Some tweaks on library size and such. Will continue to work on virtuals. Need to improve fast paths. Need virtual cell sets.

Haocheng has an MR to combine all rendering tests into one executable. Next, Haocheng will work on improving the timer/synchronization.

Dave has been looking at adding explicit cell sets for particle advection. Need a cell locator.

Abhishek has an MR for cell locators for structured grids. It should be merged in soon.