- This event has passed.
HPC Workshop, Fall 2019
November 2, 2019 @ 9:00 AM – 1:00 PM
This HPC workshop will be instructed by Stuart Slattery of Oak Ridge National Laboratory. This course will overview the Kokkos programming model for performance portability (https://github.com/kokkos/kokkos). Kokkos is actively being developed by national laboratory staff within the Department of Energy (DOE) to allow developers to compose complex C++ applications through template metaprogramming concepts capable of running on the variety of supercomputers DOE is in the process of procuring in the transition to exascale computing. This includes multithreaded CPU architectures, NVIDIA GPUs, AMD GPUs, Intel GPUs, as well as ARM platforms. In this class we will cover the basics of Kokkos in a tutorial format with interactive code exercises. This includes:
- Portable memory allocation and code execution
- Managing data access patterns for performance portability
- Simple 1D data parallel computational patterns
- Thread safety, thread scalability, and atomic operations
- Hierarchical patterns for maximizing parallelism
- Multidimensional parallelism
When: Saturday, November 2nd from 9:00am – 1:00pm. Lunch will be provided.
Where: The computer lab located in Moore 100A. You do not need to bring your own machine, computers that run Linux will be provided.
Student prerequisites
- Want to use all HPC node architectures in a single code base; including GPUs
- Are familiar with C++ including templates
- Are familiar with data parallelism
- A little familiar with OpenMP
- A little familiar with NVIDIA GPU architecture
- Want GPU programming to be easier
- Would like portability as long as it doesn’t hurt performance
- You must be a student at the University of Pennsylvania
To register for the course, you must click here and fill out the RSVP. Due to the limited number of machines available, we will need to cap this class at 22 people.