The Power of Performant Code Abstraction in Real World HPC Applications


The Power of Performant Code Abstraction in Real World HPC Applications

Bussmann, M.; Eckert, C.; Huebl, A.; Jung, F.; Widera, R.; Zacharias, M.; Zenker, E.

Many-core architectures are going to stay for a while. Thus, scaling real world applications on HPC systems with thousands of nodes and hybrid architectures will be important to application developers. We present two real world applications, plasma simulations and high repetition rate data analysis, that strongly benefit from speed ups seen with accelerator hardware. From our viewpoint as users with real-world problems we discuss our path towards scalable, hardware-independent, performant application codes. Our focus is not to reinvent the wheel, but to make full use of the power of existing C++ codes by fully abstracting all parts of a code that require optimization, separating concepts used in the code from their optimized implementation. We do this by providing light-weight, single-purpose, zero-overhead interfaces to central code components. In the talk I will discuss our experiences with this approach, the reusable interfaces that have come out of this and the vision behind putting everything together.

Keywords: many core; picongpu; simulation; hpc

Related publications

  • Invited lecture (Conferences)
    SPPEXA Annual Plenary Meeting, 25.-27.01.2016, Garching b. München, Deutschland

Permalink: https://www.hzdr.de/publications/Publ-23153
Publ.-Id: 23153