Skip to content

Propagate id3 to shared indices

Robert Maynard requested to merge robertmaynard/vtk-m:propagate_id3_to_shared_indices into master

Dramatically improves performance when performing Cell based algorithm on Uniform Grids.

Note the following performance numbers include both Ken's work to share fetched indices and this work to not transition to flat index space when scheduling with a vtkm::Id3

On Tesla K40 the cuda vtkm marching cube benchmark sees runtime go from 0.89s to 0.61s. Likewise on a 2.6GHZ Intel machine the TBB marching cube benchmark sees a runtime of 8.2627s go to 5.83962s.

Merge request reports