The Data_N_Comp_Reorder package includes various sparse matrix reordering algorithms that aim to improve the data locality and therefore the performance of computations that iterate over the sparse matrix data structure. The package include a benchmark that visits edges in an irregular mesh, a molecular dynamics benchmark, and drivers for reordering triangle and tetrahedral input files.

The drivers are capable of accepting many command-line parameters to configure the data and computation reordering strategy. They are also instrumented with timers and can be used with PAPI to record hardware counter data such as number of L1 cache misses.


More information about some of the reordering algorithms included, how to find or generate larger datasets, and experimental results can be found in the following paper:

Metrics and Models for Reordering Transformations (pdf)
Michelle Mills Strout and Paul D. Hovland.
In Proceedings of the The Second ACM SIGPLAN Workshop on Memory System Performance (MSP), pages 23-34, June 8, 2004.

Last updated .... June 30, 2005