IRTG Modern Inverse Problems (MIP)
Lade Veranstaltungen

« Alle Veranstaltungen

  • Diese Veranstaltung hat bereits stattgefunden.

EU Regional School with Prof. Paolo Bientinesi, Ph.D.

Mai 19 @ 13:30 - 17:00

Prof. Paolo Bientinesi, Ph.D. – High-Performance Matrix Computations: It’s not all about libraries

Director of the High-Performance Computing Center North (HPC2N)
Umeå University, Sweden



Matrix computations arise in virtually all computational disciplines and are often the bottleneck in countless workflows. In response to such a widespread demand, the numerical linear algebra community produces excellent libraries of high-performance kernels, to be used as universal building blocks. However, we observe that despite the quality of such libraries, end users are increasingly less likely to use them directly.
In fact, end users commonly deal with matrix computations that are significantly more complex than the functionality offered by said kernels, and expressing („decomposing“, „mapping“) a workflow in terms of the available kernels is a challenging task. End users typically circumvent the problem by adopting high-level languages such as Matlab, Python, R, or C++ with Eigen, which offer a convenient syntax for linear algebra, thus boosting productivity. However, these languages are still very immature when it comes to matrix computations, and in terms of performance, they make vastly suboptimal choices.
We refer to the problem of expressing a target matrix computation in terms of available building blocks as the „Linear Algebra Mapping Problem“ (LAMP). In this lecture we define the LAMP, and contrast the expertise and knowledge necessary to achieve almost-optimal solutions with the solvers implemented in the most popular programming languages.

In a nutshell: This lecture explains why high-performance libraries are a necessary, but not sufficient element towards high-performance matrix computations.



Mai 19
13:30 - 17:00




Rogowski Building, Room 115
Schinkelstraße 2
Aachen, 52062 Deutschland
Google Karte anzeigen