GPU programming using directives training

When: 13 – 14 March 2025 09:00 – 17:00 GMT

Where: Room 4.20 Charteris Land, Edinburgh

The ExCALIBUR PAX-HPC project, in collaboration with ARCHER2, are running a training course on GPU programming using directives. Given that future supercomputers will all contain some sort of accelerators, most likely GPUs, it is crucial that the software development community are able to ensure that their codes fully exploit these technologies.

The OpenMP offload and OpenACC APIs support portable GPU programming using directives in Fortran, C and C++. This course will provide an introduction to GPU programming with directives, aimed at scientific application programmers wishing to develop their own software. The course will give a background on GPU architecture and programming API, and then focus on using OpenMP offload directives. The course will cover offloading code and data movement, as well as more advanced topics such as interoperability with HIP and performance tuning. We will also cover the similarities and differences between OpenMP and OpenACC.

Note that although the course will use AMD GPUs for practical exercises, it is also appropriate for those wishing to use NVIDIA or Intel GPUs.

For more details and to sign up see the training course website.