TIES3440 COM2: Portable GPU Programming (JSS35) (1 op)

This course provides a practical, hands‑on introduction to portable GPU programming. Participants will learn how to develop hardware‑agnostic, high-performance applications for diverse accelerated computing environments using OpenMP offload and Kokkos as example frameworks. The course combines lectures with hands-on exercises on the LUMI and Roihu supercomputers, enabling participants to work with both AMD and NVIDIA GPUs to explore cross‑platform portability.

After completing the course, participants should be able to

· Explain the key architectural features of modern GPUs and their implications for performance

· Develop hardware-agnostic accelerated applications using OpenMP offload and/or Kokkos to express parallelism

· Implement effective memory management strategies across host and accelerator environments

· Compare and critically assess different GPU programming models in terms of portability, performance, and ease of use

· Basic skills to operate in a Linux command line environment.

· Basic working knowledge in programming with C or C++. Necessary C++ constructs will be introduced to those familiar with C only. The OpenMP section may also be completed using Fortran instead of C or C++.

· Prior exposure to scientific computing concepts is helpful but not required.

· Participants should bring their own laptop for accessing the supercomputers where hands‑on exercises are conducted. Any laptop capable of SSH access and web browsing is sufficient.