This repository contains two mini-apps that were created to improve the MOAB remapping workflow for Climate problems.
-
Kd-tree mini-app implementation: This example loads a 2d/3d unstructured mesh file and performs several queries to locate the points belonging to elements in the mesh.
Typical usage:
./kdtreeApp -i MESH_FILE -d 3 -n 100000 -r 1There are parameters that can be specified for the application. a)
-i <path>- Path to the mesh file to load and query. b)-d <int>- Leading dimension of the mesh file. Default is 3. c)-n <int>- Number of queries to perform on the tree datastructure. Default is 1000000. c)-r <int>- Number of in-memory refinements to perform to have a much bigger mesh to query. Default is 0. -
SpMV(n) mini-app implementation: This example loads a remapping weight matrix from file, and forms a SparseMatrix datastructure in memory. Then it applies the matrix onto a random vector multiple times to compute average time for SpMV.
Typical usage follows:
./spmvApp OPERATOR_PATH -t -v 1 -n 100There are parameters that can be varied here. a)
OPERATOR_PATH- This is a required argument specifying the path to the NetCDF-based matrix weights file. a)-t- Enable both Ax and A^Tx operations in timing profile. Default is False. b)-v <int>- This argument controls the number of vectors onto which the operator is applied. This is useful to mimic transfer of multiple real world fields between components. Default is 1. c)-n <int>- This number controls the total iterations to perform to get an average SpMV timing. Default is 100.