Sparse matrix problems are di cult to parallelize e ciently on message-passing machines, since they access data through multiple levels of indirection. Inspector executor strategies, which are typically used to parallelize such problems impose signi cant preprocessing overheads. This paper describes the runtime support required by new compilation techniques for sparse matrices and evaluates their performance, highlighting optimizations and improvements over previous techniques.
Manuel Ujaldon, Shamik D. Sharma, Joel H. Saltz, E