Where the theory and practice of software meets hardware
The Science of High-Performance Computing group focuses on the computer science that underlies high-performance linear algebra libraries. Formal derivation techniques are employed to derive families of algorithms from which the best for a specific architecture or situation can be chosen. By carefully layering software, the resulting algorithms are implemented as highly portable, high-performance software libraries. This has yielded open source software that is widely used in academia, by the national labs, and by industry. The techniques are shared through outreach activities that include a number of Massive Open Online Courses.
Website
Directors
Faculty and Research Staff
Staff
The talk "The BLIS Framework: Experiments in Portability," which received the SIAM Activity Group on Supercomputing Best Paper Prize 2020, is a nice introduction to the group’s activities: https://www.youtube.com/watch?v=1biep1Rh_08.
Projects
Collaborative Research: Frameworks: Beyond the BLAS: A framework for accelerating computational and data science.
National Science Foundation Award CSSI-2003921.
PI: Robert van de Geijn and Devin Matthews (SMU).
Co-PIs: Margaret Myers, Field Van Zee, Devangi Parikh.