The National Energy Research Scientific Computing Center (NERSC) will host on October 17th, 2019 a one-day training event on using Appentra’s Parallelware Trainer tool to learn how to use OpenMP and OpenACC directives to develop code for Graphical Processing Units (GPUs). The objective is to improve the learning experience of NERSC users to program the GPU.
This is the second event of a series of workshops, taught by Appentra’s CEO Manuel Arenaz in collaboration with NERSC staff.
The first event was in June 5th, 2019 and covered introductory contents to program the GPUs using OpenMP/OpenACC. This second event will cover more advanced topics for optimization of GPU code, particularly addressing how to minimize data transfers, how to exploit massive parallelism and how to optimize the usage of the GPU memory.
The lectures emphasize the key concept of parallel code pattern and how to use these patterns to learn best practices for parallel programming on GPUs. The workshop also included practical sessions where the participants worked on realistic codes as well as on their own codes using OpenMP and OpenACC offloading directives.
Further information and registration:
|8:15 – 8:45||Morning refreshment and coffee|
|8:45 – 9:00||Welcome and Introductions||Manuel Arenaz (Appentra)|
|9:00 – 9:30||Lecture 1: An introduction to OpenMP/OpenACC optimizations for GPUs|
|9:30 – 10:15||Lecture 2: Patterns to minimize data transfers, optimize memory usage and exploit massive parallelism|
|10:15 – 10:30||Break|
|10:30 – 11:00||Lecture 3: Minimizing data transfers|
|11:00 – 11:30||Lecture 4: Optimizing memory usage|
|11:30 – 12:00||Lecture 5: Exploiting massive parallelism|
|12:00 – 13:00||Working Lunch (hands-on activities)|
|13:00 – 14:00||Practical 5A: Parallelizing the calculation of Laplace2D|
|14:00 – 15:0||Hands-on time with your code||Manuel Arenaz & NERSC Staff|
This event will be presented both online using Zoom technology and in person at NERSC/LBNL (visitor info) in Berkeley, CA. The training will be held at Building 59, Room 4102 (CRT Building, aka Wang Hall).
Playlist of the previous workshop
If you missed the previous workshop on the Parallelware Trainer Tool at NERSC held in June 2019, you can now see the full session in this playlist.
About Parallelware Trainer:
Parallelware Trainer is an interactive, real-time code editor with features that facilitates learning, usage, and implementation of parallel programming by understanding how and why sections of code can be parallelized. The learning process is based on decomposing codes into parallel code patterns, learning what parallelization strategies are applicable to those patterns, and the actual generation of OpenMP/OpenACC-enabled parallel code according to best practices for parallel programming.