Attendees: Ken Moreland (SNL), Ollie Lo (LANL), David Pugmire (ORNL), Tom Otahal (SNL), David Camp (LBL), Matt Larsen (LLNL), Manish Mathai (UO), Alberto Villarreal (Intel), James Kress (ORNL), Rob Maynard (Kitware), Dave Lonie (Kitware), Sujin Philip (Kitware), Abhishek Yenpure (UO)
Ollie has been implementing scan by key. He has an open merge request and it is being reviewed and revised.
David P has been working on getting PIC particle advection working on CUDA. It is now working on CUDA. The one issue with CUDA is that if they are keeping track of particle histories they are allocating a very large array for all particles and largest possible stream. Possible ways around: dynamically allocate or do in rounds.
Working on isosurfacing pipeline in strawman. Trying to run TAU on this.
James working on getting VTK-m integrated into Xlotl. Part of the PSI SciDAC.
Tom send out results on tiling study for KNL architecture. In short recap, tiling with a 512 or 1024 buffer seems about optimal. This was for applying the elevation filter.
Alberto has been using performance tools to analyze vectorization on the code they are benchmarking.
David suggests for particle advection you want to do it in parts and send it in clusters and send it to the core.
Matt has been doing some dedicated HPC time this week. It involved strawman and VTK-m. Part of the work includes an isosurface in the rendering library. Matt and Dave P. are doing similar things and should coordinate and compare.
UO students have been getting familiar with VTK-m.
Rob has been working on Marching Cubes. He has been separating normals generation from the MC pass. This saves on memory. This in turn gives a nice performance bump. Also added some code to allow the code to follow fast paths for permutations.
Rob has also been cleaning up the ForEach classes.
Rob has been doing some GTC prep.
Sujin is cleaning up code for the virtual implicit functions.
Dave L. is finishing a patch on cleaning up virtual tables in several control classes.