Moderator: Jason D. Sewall, Intel Corporation
Panelists: Barbara Chapman, Stony Brook University; Olivier Giroux, Nvidia Corporation; Gihan Mudalige, University of Warwick; Sunita Chandrasekaran, University of Delaware; Christian Trott, Sandia National Laboratories; Andrew Richards, Codeplay Software Ltd
Time: Tuesday, November 13th, 1:30pm – 3pm
Hardware is evolving in complexity and diversity, enabling us to reach ever-higher peaks of performance. However, this comes at a cost to software development: some hardware requires special consideration that can hinder portability, and even portable code that runs well on one device may not run well on another. Given the great expense of software development and validation, we continually hunt for new tools—and augmentations to old ones—that mitigate these costs. Tools for developing single-source, cross-platform codes—like Kokkos, RAJA, OpenACC and OpenMP—are popular, while libraries like LIBXSMM use new code-generation techniques to improve performance.
Can we build a do-it-all compiler? Are prescriptive or descriptive approaches better? How do you prepare for the Next Great Thing? This panel discussion features experts in popular libraries and languages for portable, high performance computing, each with a different philosophy, a distinct approach to these problems, and a willingness to stand up and defend their perspective.