ICSA Faculty Talk - 29/09/2020
Talk Title: Effective mixed-domain compilation using multi-level rewriting
Abstract: Today's increasingly common mixed-domain compilation problems require us to overcome the classical single-domain compiler designs and the costly compartmentalization they bring along. Software-hardware co-design, the compilation for heterogeneous hardware architectures, or programming models that translate imperative programs directly into hardware designs (e.g., HLS) all these workflows mix highly orthogonal compilation problems. While an all-unifying compiler remains unattainable, we make the case that multi-level rewriting exposes surprising similarities in compilers for three vastly different domains, (a) mathematical reasoning about Presburger arithmetic, (b) a climate compiler solving finite difference computations, and (c) hardware design via System Verilog. Each compiler for itself can in our opinion be foundational for its very domain. We close this talk posing three questions: (1) which additional domains can we conquer, (2) do hybrid-compilers enable new workflows, and (3) how can we make the design of mixed-domain compilers fundamentally easier?
Bio: Tobias Grosser is a Reader (associate professor) in the School of Informatics at the University of Edinburgh. Combining compilation, programming languages, and run-time systems he works on making performance programming an effective and intuitive game. His use of strong mathematical reasoning often enables optimal, precise, and trustworthy solutions. As an expert in polyhedral compilation and open-source enthusiast, Tobias Grosser has led the integration of advanced loop optimization frameworks into production compilers. His current research covers the full software-hardware stack from high-performance computing for climate modelling down to the effective compilation of hardware design languages, including the necessary foundational math libraries. Tobias holds a master from the University of Passau, received as Google Fellow a PhD from Pierre and Marie Curie University in Paris, and was awarded an SNSF Ambizione Fellowship for his work at ETH Zurich.