Utilizing Recursive Storage in Sparse Matrix-Vector Multiplication - Preliminary Considerations

Computations with sparse matrices on "multicore cache based" computers are affected by the irregularity of the problem at hand, and performance degrades easily. In this note we propose a recursive storage format for sparse matrices, and evaluate its usage for the Sparse Matrix-Vector (SpMV) operation on two multicore and one multiprocessor machines. We report benchmark results showing high performance and scalability comparable to current state of the art implementations.

• Type: Conference
• Year: 2010
• Where: CATA
• Authors: Michele Martone, Salvatore Filippone, Salvatore Tucci, Marcin Paprzycki, Maria Ganzha
• Keywords: CATA 2010, Computer Science, Multicore Cache, Sparse Matrices, Sparse Matrix-Vector