Timothy Mattson provides Workshops on CPU and GPU Parallel Programming

This is a full day workshop that introduces parallel programming. It is a hands-on workshop meaning most of the learning happens through exercises. We use OpenMP for writing parallel code for the cores inside a CPU. Even if your goal is scaling across a full cluster, OpenMP is the right place to start. Why? The amount you need to learn before writing code is minimal. That means less time spent listening to someone lecture about parallelism and more time actually writing parallel code. While the emphasis is on programming the cores on a CPU, we will cover many of the core concepts you need for programming clusters and GPUs.!

The future is built around clusters with heterogenous nodes composed of CPUs and GPUs. The smart programmer wants to use all the hardware on a node. OpenMP is one of the few (if not only) ways to program both CPUs and GPUs from within a single programing model. In this hands-on workshop, we will cover GPU programming with OpenMP. I assume you have experience programming CPUs with OpenMP (such as by taking my prior workshop that introduces OpenMP) but no experience with GPU programming. Running from morning to mid-afternoon (three quarters of a day), we’ll cover everything you need to write well optimized code for a GPU. We will focus particularly on understanding GPU programming models in general (i.e., what you need to make sense of CUDA, OpenACC or OpenMP) and how they compare to CPU programming.

This is a 90-minute workshop that covers the key topics EVERY scientist MUST know about numbers on computers. In particular, we cover the fundamental issues raised by the IEEE-754 standard and what you need to know to avoid writing programs that cause disasters. We then talk about how to safely use random numbers on a computer with a particular focus on how random numbers fall apart on parallel computers.