Open PhD Student Position in System Software

Multi-Domain DSL Design and Compilation for Heterogeneous Parallel Systems

Linköping University invites applications for a fully funded PhD student position in Computer Science at PELAB, Research group for compiler technology and parallel computing:

Modern computer systems are increasingly parallel and heterogeneous, and new domain-specific accelerator types e.g. for AI/ML computations are introduced at a rapid pace, each with their own platform-specific programming model and API. This poses a challenge for writing efficient software for such systems yet keeping it portable and adaptive to future system configurations. High-level, domain-specific programming frameworks based on patterns/skeletons, such as SkePU, are a promising approach towards this goal.

We investigate the design and implementation of high-level, pattern-based multi-domain (X+AI) programming models for modern heterogeneous parallel computer systems with AI accelerators, and the development of compiler and runtime optimizations for the efficient execution of multi-domain programs on AI accelerator hardware, such as tensor cores in recent GPUs, systolic arrays as in TPUs, and reconfigurable architectures.
Beyond conceptual work e.g. on programming model design and solutions to key technical problems in program analysis, parallelization, optimization, adaptivity support, and low-level code generation, the project also involves prototype development and evaluation, including e.g. the extension of a modern C++ based high-level programming framework for heterogeneous parallel systems, such as SkePU, and adaptation of an open-source DSL compiler framework such as TVM or LLVM/MLIR.

Qualification:

• Specific requirements for this project:
  • We expect a solid background in mathematical foundations, algorithms, computer architecture, and combinatorial optimization; preferably also in compiler technology and in parallel and/or GPU programming;
  • very good programming skills in modern C++ (esp., template metaprogramming) and familiarity with Linux.
  • Some background in machine learning / deep learning techniques is desirable.
  • We also welcome practical experience with the internals of some modern open-source compiler framework.
  • An interest in teaching is desirable.
• For the general requirements (CUGS), such as general eligibility criteria, expected English language skills and other general requirements, see the main announcement text.

For more information about the project, contact: Prof. Christoph Kessler

Page responsible: Webmaster
Last updated: 2021-07-02