Applicability of Program Comprehension to Sparse Matrix Computations

Space-efficient data structures for sparse matrices typically yield programs in which not all data dependencies can be determined at compile time. Automatic parallelization of such codes is usually done at run time, e.g. by applying the inspector-executor technique, incurring tremendous overhead. -- Program comprehension techniques have been shown to improve automatic parallelization of dense matrix computations. We investigate how this approach can be generalized to sparse matrix codes. We propose a speculative program comprehension and parallelization method. Placement of parallelized run-time tests is supported by a static data flow analysis framework