As the use of GPUs in HPC is spreading fast, several models are now available to programmers. During the tutorial we will introduce the directive-based approach to GPU programming. While being quite simple to implement, it allows to reach performance similar to lower level paradigms.
We will present the basic concepts of OpenACC and OpenMP-target models through several examples and hands-on sessions. The main pitfalls and good practice will be presented.
We will also give some insights of the tools available to assess the performance.
Learning Outcomes :
- Ability to execute a piece of code on a GPU using OpenACC or OpenMP-target
- Assessment of the performance of the code
Basic knowledge of Fortran and/or C and/or C++
No prior knowledge of OpenACC or OpenMP is required