Advancing Compiler Techniques for Heterogeneous Computing

At the Chair for Compiler Construction in Cfaed (Center for Advancing Electronics Dresden), we’re dedicated to pushing the boundaries of knowledge in programming methods and compiler optimizations. Just as an artist chooses brushes and colors, programmers select tools and methods. Our group delves deep into these methods, exploring the strengths and potentials of dataflow and reactive programming paradigms. Once the code is written, we specialize in understanding compiler optimizations and developing new techniques to enhance compiler functionality. By optimizing code for CPUs, GPUs, FPGAs, or other emerging hardware architectures, we want to ensure that the software performs at its peak.
Our research focus lies mainly in these three areas:

1. Parallel Programming of Heterogeneous Platforms:
– We explore dataflow programming as a way to manage parallel architectures, optimize the scheduling of operations and different mapping models for offloading applications to hardware, optimizing resource usage and execution time.
2. Domain-Specific Languages (DSLs), Optimizations, and Tools:
– We raise the abstraction levels for particular application domains aiming to enable more powerful optimizations for large parallel systems.
– Collaborating with life science scientists, we define DSL abstractions and explore optimization potentials for diverse scientific computations, such as computational fluid dynamics and particle simulations.
3. Post-CMOS Systems:
– We explore programming flows for future heterogeneous systems, optimizing technologies like Racetrack Memories (RTMs) through compilation and architectural simulation tools.
– Investigate in and near-memory computing architectures, we explore the potential of memristor crossbars, content-addressable memories, and upmem platform for modern data-intensive workloads.