Understanding the Performances of SMVP on Multiprocessor Platform

Sparse Matrix Vector Product (SMVP) is an important kernel in many scientific applications. In this paper we study the performances of this kernel on multiprocessor platform using four different compression format (CSR, CSC, ELL and COO). Our aim is to extract runtime environment parameters, matrix characteristics and algorithm parameters that impact performances. This work is in the context of implementing an auto-tuner system for Optimal sparse Compression Format (OCF) selection.