Invited talk

Speaker

Samuel Thibault, University of Bordeaux, Laboratoire Bordelais de Recherche en Informatique (LaBRI), France

Samuel Thibault is Professor at the University of Bordeaux, and part of the Inria STORM team. His research revolves around task and data transfer scheduling in parallel and distributed runtime systems. He is currently focused on the design of the StarPU runtime, and more particularly its scheduling heuristics for heterogeneous architectures and for distributed systems.

Recursive tasks for fun and profit

Recursive tasks have been proposed in various task-based runtime systems. While the details vary, a lot of commonalities arise, and similar performance benefits are observed by different research teams. On the application side, expressing computation recursively is also a pattern that arises commonly for expressing e.g algorithms on compressed data such as H-matrices. We will thus consider the current proposals and results, and discuss the kinds of benefits that we can expect in the long run. This can include making it easier to express complex algorithms with complex data structures, but also improve the efficiency of the execution in various ways, by using different granularities to drive the available parallelism and balance tasks granularities between heterogeneous resources, as well as parallelizing the tasks submissions and decoupling tasks from communication.