Using semantical information to enhance the parallel sparse performance

This work presents a novel strategy for the parallelization of applications containing sparse matrix references using the data-paxallel paradigm. Our approach is a first stop to converge to the automatic parallelization by re ducing the number of directives on code. We have used the semantical relati onship of vectors composing a high-level data structure to enhance the perf ormance of the parallel code, applying a sparse privatization and a multi-l oop analysis. We also study the building/updating of a sparse matrix at run -time, solving the problem of using pointers and some levels of indirection s on the left hand side. A detailed analysis about several temporary buffer s useful for sparse communications is described in this paper. The evaluati on of our strategy has been performed on a Cray T3E with sparse matrix tran sposition algorithm as a case of study.