Join us for a webinar introducing how to use OpenMP for programming your GPU. You don’t have to be afraid of programming your GPU with the OpenMP API. The OpenMP features for heterogeneous programming have you covered!
In this talk, we will introduce what the OpenMP API considers a device for offloading computation as well as the underlying OpenMP device and execution model. Then, we will explore the OpenMP device features using a simple running example that you can use as the foundation for your own, more complex algorithms. We’ll walk through how to exploit parallelism on a GPU, how to optimize data transfers, and how to integrate offloaded computation into the OpenMP task world. The simple example code will guide us through the journey of enabling the example for massively parallel execution on a GPU.
At the end of this talk, you’ll be able to write your first OpenMP offload code and explore more sophisticated OpenMP offload features. And we will of course allow for plenty of time for your questions.
Suggested background: Ideally you will have basic understanding of the OpenMP directive syntax and how parallelism works in the OpenMP API, especially around working-sharing constructs and OpenMP tasks.
This presentation will be delivered by Michael Klemm, CEO of the OpenMP ARB and Principal Member of Technical staff in the HPC Center of Excellence at AMD. Michael’s day-to-day business is to bring applications to AMD GPUs and get the last bit of performance out of them.