Compiling Imperfectly-nested Sparse Matrix Codes with Dependences

Abstract

We present compiler technology for generating sparse matrix code from (i) dense matrix code and (ii) a description of the indexing structure of the sparse matrices. This technology embeds statement instances into a Cartesian product of statement iteration and data spaces, and produces efficient sparse code by identifying common enumerations for multiple references to sparse matrices. This approach works for imperfectly-nested codes with dependences, and produces sparse code competitive with hand-written library code.