Offer #2018.31

Platform for Heterogenous/Parallel Computing

Linköping University, Sweden

The problem/challenge

Heterogeneous systems (e.g. those that use both GPUs and CPUs) provide significant performance increase. Yet, existing platforms for heterogeneous programming (OpenCL, CUDA, etc.) are difficult to use and the code is hard to maintain and optimize.

The invention/product

SkePU is a framework for heterogenous programming, i.e. where one can write heterogenous software and applications. The framework is:

  • Easy to use (High level of abstraction, no harder than writing C++)
  • Portable and platform-independent
  • Performance-portable (Automated optimization of execution flow)
  • Comes with pre-defined constructs (skeletons) such as scan, map, reduce, stencil etc.
Companies/industry we are looking for

We are looking for programmers using, or thinking of switching to, heterogeneous platforms. Specific industries include:

  1. Researchers and Research Centers working with super computers, machine learning and image processing
  2. Companies dealing with image processing, such as radar companies and medical imaging companies.
Next steps/Activities

We are seeking collaborations to:

  • Use SkePU (SkePU is licensed under the GNU General Public License)
  • Extend SkePU with further skeletons and data containers for specific domains or target architectures.

The research group for compiler technology and parallel computing at PELAB, Linköping University, Sweden, led by Prof. Christoph Kessler, has long-term experience in the design and implementation of high level parallel programming frameworks for heterogeneous parallel computer systems, based on software abstractions such as skeletons/patterns, multi-variant components, and data containers. The group also develops support techniques for heterogeneous computing, including automated performance and energy modeling as well as static and dynamic optimizations for mapping and scheduling, resource allocation, variant selection, memory management and data transfers.

Make an enquiry

Date published07/26/2018StatusLooking for collaborationTechnology areasCompiler TechnologyHeterogeneous ComputingMulticore SystemsOptimization TechnologiesParallel Programming