2016-054 – Graph Algorithms from Linear Algebra (GALA) Library

This is a design and implementation of a library of graph algorithms for heterogeneous high-performance compute (HHPC) architectures that contain GPU processing units in addition to general purpose CPU’s. While the highest performing implementations of these algorithms are complex and intimately tied to the underlying architecture, the objective of this library is to make high-performance graph analytics on new and emerging computer architectures more accessible to a broad base of users. We accomplish through a design that separates the concerns of expressing graph algorithms from the details of mapping to specific underlying computing architectures. This design includes the application programming interface (API), including data structures and algorithms that work for both CPU and GPU architectures. Raymond Taylor rtaylor@andrew.cmu.edu 412-268-2000