Research Interests

Very generally speaking, I am interested in scientific computing. More specifically, my work can be divided into three categories:

Parallel Algorithms. Specifically I have worked on parallel algorithms for clusters, GPUs, multiprocessors, and vector processors (SIMD).
Productivity and Portability through SaC. Single-Assignment C (SaC) is a minimalistic language for scientific computing that does not expose memory or parallelism. This makes it easy to use. The compiler can generate parallel code for CPU, GPU, and clusters from a single specification. If the parallelisation is not too sophisticated, the performance is comparable to code written in C or Fortran.
Numerical Stability. This is a new interest of mine, about getting a grip on the accuracy for numerical algorithms. I am currently working on a problem in computational geometry.

Publications

Minimizing Communication in the Multidimensional FFT. A new parallel algorithm for computing the discrete Fourier transform on supercomputers. bib
Rank-Polymorphism for Shape-Guided Blocking. On showing how locality optimisations can be elegantly expressed in a language with support for arrays of arbitrary dimension. bib
Shray: An Owner-Compute Distributed Shared-Memory System. A software system for easier programming of supercomputers. bib
Modulo in high-performance code: strength reduction for modulo-based array indexing in loops. An optimization for programming languages that translates pretty ways of programming a certain class of computations (stencils) to efficient ways. bib

Multi-GPU Code Generation for Out-Of-Core Problems. A compiler backend for the programming language SaC that can leverage multiple GPUs, also on data structures that do not fit on any single GPU. To appear in TFP'25 proceedings.
Partitioning In-Place on Massively Parallel Systems. An algorithm for splitting an array in two on a GPU, without needing a significant amount of additional memory. To appear in EuroPar'25 proceedings.

Comparing Functional Array Languages. Large collaborative paper with Futhark, Accelerate, APL, DaCe, and SaC. Submitted to JPDC.

The Tensor Product of Bulk Synchronous Parallel Algorithms. This derives a multidimensional parallel FFT algorithm from a one-dimensional parallel one using category theory. This is my master thesis, and the paper Minimizing Communication in the Multidimensional FFT is essentially this thesis with the category theory stripped out.

My colleagues at Radboud's software science group: