Teleconference 2018-10-31

From VTKM
Jump to: navigation, search

Attendees: Ken Moreland (SNL), Rob Maynard (Kitware), Mark Kim (ORNL), Matt Larsen (LLNL), Abhishek Yenpure (UO), Dave Pugmire (ORNL), Berk Geveci (Kitware), Sujin Philip (Kitware), Hauchang Liu (Kitware), Ollie Lo (LANL), Matt Letter (SNL)

Color table and separable compilation are now up with an MR. There is a Windows issue. NVIDIA is aware and will fix with whatever comes after CUDA 10. Will have to either ship a custom version of thrust or use an execution policy that brings over a chunk of thrust with it.

Renar is coming back on line later today with newer hardware. Still will have CUDA 9.

Hauchang has an MR to allow changing the runtime device information at runtime. Needs Ken to review.

Sujin has been working on a VTK data set that has a VTK-m data set internally.

Abhishek and Hank have been working on an EGPGV paper on point merging.

Dave has been working on finishing a paper. Almost have ghost cells for rectilinear structures done.

Ken asked Dave to put together a poster for ECP meeting.

Matt Larsen is working on a movie for SC and is going through review and release. Also going to have a demo station at the DOE booth.

Matt: several simulations use their own memory allocator where they grab a large pool of memory. That pool is still allocated when calling visualization routines, which leaves little memory left. Can we change our allocators to use this pool of memory. Rob: yes. There is probably some tricks we have to do with Thrust, but this should be doable.

Ollie has been working on a bubble detection algorithm with VTK-m. Simple filter to convert particle data into a grid and find regions with low density.

Ollie brought in Marvin Petersen from Kauserslaughtern. His adviser is Christoph Garth. Working on his masters thesis by implementing hyper grid tree in VTK-m.

Matt Letter has updated all the examples in the user's guide for recent changes in filters. He is now working on converting all examples of device adapter algorithm to algorithm.

Some thoughts on submitting an R&D100. It might be helpful submitting through labs. Although there are more levels of scrutiny, there is also a large amount of experience in submitting.