NHR PerfLab Seminar: The Linear Algebra Mapping Problem and how programming languages solve it (September 12, online)
Title: The Linear Algebra Mapping Problem and how programming languages solve it
Speaker: Prof. Paolo Bientinesi, Ph.D., Umeå University, Sweden
Date and time: Tuesday, September 12, 2023, 2:00-3:00 p.m. CEST
Abstract:
The computational workflows of countless applications across science and engineering involve operations on matrices. To support such a broad demand, the numerical linear algebra community has delivered over the course of the years an assortment of high-performance libraries, offering application-independent kernels that serve as universal building blocks. However, despite their quality, such libraries are not utilized as often as one would imagine. Indeed, often times their usability is limited by an interface and a level of abstraction that are inconvenient for many users. Furthermore, typically users face operations that are more complex than those offered by the library kernels, and the decomposition of a target operation in terms of a set of available kernels is far from being a trivial task. We refer to this task as the “Linear Algebra Mapping Problem” (LAMP). This problem is solved internally by all those programming languages that offer high-level matrix syntax, such as Matlab, Python, R, and C++ in combination with libraries such as Armadillo or Eigen. These languages tremendously help users’ productivity. However, we show that in terms of performance, they are still immature as their solutions to LAMPs are severely suboptimal.
Short Bio:
Paolo Bientinesi is Professor in High-Performance Computing at Umeå University, and director of the High-Performance Computing Center North (HPC2N, Sweden). He completed his Laurea degree in computer science at the University of Pisa (Italy, 1998), and received his Ph.D. from The University of Texas at Austin (US, 2006). Before moving to Sweden, he was professor at RWTH Aachen University (Germany, 2008). His research interests include matrix and tensor operations, automatic algorithm & code generation, performance modeling, and computer music. Paolo Bientinesi leads the research group High-Performance and Automatic Computing (HPAC, hpac.cs.umu.se, https://github.com/HPAC).
For a list of past and upcoming NHR PerfLab seminar events, see: https://hpc.fau.de/research/nhr-perflab-seminar-series/