About Me
I am a researcher at the University of Münster in Germany. My work focuses on compiler technologies – code generation and optimization – as well as programming language design for parallel architectures, including GPUs and CPUs, based on formal methods. My overall research goal is to provide Performance, Portability, and Productivity for data-parallel computations with particular focus on computations relevant for deep learning (e.g., linear algebra routines and stencil computations).
To achieve my goals, I am one of the main designers of a holistic code Generation & Optimization & Execution approach, consisting of three major sub-projects:
-
Multi-Dimensional Homomorphisms (MDH) – a novel algebraic formalism for expressing and reasoning formally about data-parallel computations. In particular, this project includes the design and specification of a Domain-Specific Language (DSL) for expressing such data-parallel computations, as well as the design and implementation of a compiler for this DSL – our compiler enables generating code (e.g., in CUDA, OpenMP, or OpenCL) that can be automatically optimized (auto-tuned) for state-of-the-art parallel architectures;
-
Auto-Tuning Framework (ATF) – a general-purpose auto-tuning approach that automatically optimizes parallel programs, based on numerical search techniques and optimized processes of generating, storing, and exploring the optimization spaces of state-of-the-art parallel implementations;
-
Host Code Abstraction (HCA) – a high-level programming abstraction that simplifies implementing and optimizing so-called host code which is required in modern parallel programming approaches (e.g., CUDA and OpenCL) to execute code on the devices of distributed, heterogeneous systems.
You can find my CV here.