Graph Transformation and Specialized Code Generation For Sparse Triangular Solve (SpTRSV)

Sparse Triangular Solve (SpTRSV) is an important computational kernel used in the solution of sparse linear algebra systems in many scientific and engineering applications. It is diffcult to parallelize SpTRSV in today's architectures. The limited parallelism due to the dependencies between calculations and the irregular nature of the computations require an effective load balancing and synchronization mechanism approach. In this work, we present a novel graph transformation method where the equation representing a row is rewritten to break the dependencies. Using this approach, we propose a dependency graph transformation and code generation framework that increases the parallelism of the parts of a sparse matrix where it is scarce, reducing the need for synchronization points. In addition, the proposed framework generates specialized code for the transformed dependency graph on CPUs using domain-specific optimizations.