Design and Implementation of the GraphBLAS Template Library (GBTL)
July 2016 • Presentation
Scott McMillan, Samantha Misurda, Marcin Zalewski (Indiana University), Peter Zhang (Indiana University), Andrew Lumsdaine (Indiana University)
The design of the GraphBLAS Template Library separates graph algorithm development from performance tuning for heterogeneous high-performance computing architectures.
Software Engineering Institute
This presentation was given at the 2016 Annual Meeting of the Society for Industrial and Applied Mathematics in Boston, Massachusetts.
An important design goal of the GraphBLAS Template Library (GBTL) is to achieve a separation of concerns between graph algorithm development and performance tuning required for various heterogeneous high-performance computing (HHPC) architectures. In GBTL, the GraphBLAS API specifies the primitive operations and data structures that allow this separation of concerns. We implement this API for both generic CPU systems and GPU systems, and present a number of important graph algorithms written in terms of this API.