Research¶

Effective Performance Programming by Re-Connecting Developers and Compilers

My research uses strong theoretical reasoning to bring innovations to real-world compilation and programming language problems. My current objective is to rethink performance programming by re-connecting developers and compilers. Today, performance programming is no longer limited to the optimization of low-level code but often includes the use of domain-specific compilers, constraint programming libraries, complex performance models, and automatic (potentially learned) strategies to search for optimal code transformations.

My primary objectives are:

making compilation more modular, predictable, automatic, and trustworthy
bringing open-source compiler innovation to an increasingly broad set of targets from GPUs over FPGAs to custom hardware, and
breaking down the barriers between compilers and programmers by enabling their interaction via the programming language environment.

I dream of a future where performance programming is an intuitive play between the programmer and the compiler. A game that smoothly moves between manual as well as automatic techniques and works across software and hardware.

I am particularly interested in the following areas:

Compilers
Static & dynamic analysis
- Abstract interpretation based program analysis
- Performance and cache models
- Test case generation
Human-compiler interface
Domain-specific compilation
- Deep learning
- Climate science
High-performance computing
- Loop optimization & polyhedral compilation
- Vectorization
Compilation for accelerators: GPU, FPGA
Open-source hardware
- Compiler support for RISC-V
- Software/hardware co-design
SMT and constraint solving
- Mixed integer linear programming
- SAT solving
- Parametric counting using Barvinok’s algorithm
Automatic theorem proving for compiler verification and constraint solving
Effective compilation from functional languages to imperative code
Artificial intelligence for compilers and constraint solvers