This course gives basic to intermediate level knowledge on HIP programming. It starts with a basic introduction to writing GPU applications using HIP and their corresponding CUDA counterparts. Then, building on this, more advanced topics are introduced. These enable one to write efficient HIP applications, including asynchronous execution with streams, utilizing performance tools and how to write efficient kernels. Finally multi-GPU programming will be discussed.

Learning outcome

After the course, the participants will be able to write simple GPU programs, know how to analyse the performance of their GPU programs, and understand how to combine GPUs with MPI.

Prerequisites and content level

No prior knowledge on GPU programming or HIP/CUDA is required, but the participants are assumed to have a working knowledge of the C++ (or C) programming language and to be familiar with MPI.

The content level of the course is broken down as: beginner’s – 0%, intermediate – 60%, advanced – 40%, community-targeted content – 0%.

The event is organised at the CSC Training Facilities located in the premises of CSC at Keilaranta 14, Espoo, Finland.

Agenda (tentative)

Day 1, Thursday 27.3.2025, 9:00-16:00

Introduction to GPUs and GPU programming
Kernels, synchronisation, streams
Memory allocation/access, unified memory

Day 2, Friday 28.3.2025, 9:00-16:00

Fortran and HIP
Kernel optimisation, profiling, and libraries
Multi-GPU programming

Lecturers: Cristian-Vasile Achim (CSC), Juhani Kataja (CSC), Juhana Lankinen (CSC)

Language: English