OpenACC users, members converse and collaborate

Every year, the OpenACC annual meeting brings together representatives from national laboratories, universities, and other research institutions to exchange information and expand the programming model’s uses in various science domains.

A directive-based and performance-portable parallel programming model designed to program many types of accelerators, OpenACC is compatible with the C, C++, and Fortran programming languages. OpenACC simplifies the process of porting codes from host devices to various high-performance computing platforms and architectures, significantly reducing the time and effort scientists and engineers spend programming. This resource is one of the major programming models supported on the Cray XK7 Titan and the IBM AC922 Summit, both supercomputers at the Oak Ridge Leadership Computing Facility (OLCF), a US Department of Energy (DOE) Office of Science User Facility located at DOE’s Oak Ridge National Laboratory (ORNL).

The 2018 meeting, which took place at ORNL from July 31 to August 3, included a user feedback session with presentations on diverse subjects such as using OpenACC for machine learning, astrophysics, and earth system modeling applications. The remainder of the event consisted of a series of discussions among members of the OpenACC organization, which was founded to provide education and support to researchers while helping them do more science and less programming. Through these activities, ORNL staff obtained key insights to help the OLCF optimize OpenACC applications on the recently launched Summit system.

“The OpenACC face-to-face meeting at ORNL gave us an opportunity to meet with users at the lab and from around the world,” said OLCF Computer Science Researcher Oscar Hernandez. “They helped us understand how they are using OpenACC and identified challenges we still need to solve, and this feedback was extremely useful as it will help us address those needs in future specifications.”

Members of the OpenACC organization from academia, industry, and government congregated to share knowledge, tackle challenges, and brainstorm improvements throughout a series of technical and outreach-focused discussions.

Based on this feedback, members prioritized technical topics such as improving OpenACC memory models, offloading data to both CPUs and GPUs, and clarifying the model’s existing functionality. Additionally, they strategized how to grow the user community and empower users, make OpenACC training widely accessible and effective, and bring more global members into the organization. Over 40 individuals attended these sessions, either in person or remotely.

Then, they prepared for OpenACC 2.7, the next version of the programming model scheduled for release in November at the International Conference for High Performance Computing, Networking, Storage, and Analysis. They also discussed the preliminary implementation of deep copy (data movement of complex data structures) and reviewed memory allocation capabilities, as well as a new clause that allows users to run OpenACC on the host device without assistance.

“It’s been great to see how the OpenACC specification has developed over the past 7 years and how many applications have been and are being ported to parallel and accelerated programming with OpenACC,” said chair of the OpenACC language committee Michael Wolfe, who works on OpenACC implementation in the PGI compiler suite at NVIDIA. “The language committee continues to be driven by feedback from our users to focus on features they need for performance on current and upcoming hardware.”

OpenACC member organizations represented at the meeting were NVIDIA, ORNL, Helmholtz-Zentrum Dresden-Rossendorf (HZDR), the University of Delaware, Indiana University, the Swiss National Supercomputing Centre (CSCS), Mentor Graphics, Stony Brook University, Advanced Micro Devices, and Cray.

Presenters at the user feedback session were Arghya Chatterjee, Bronson Messer, Piet Jones, and Matt Norman of ORNL; Xuechao Li of Concordia University; Guido Juckeland of HZDR; Will Sawyer of CSCS; Enrico Calore of the University of Ferrara; Ron Caplan of Predictive Science; and Brian Dobbins of the National Center for Atmospheric Research.

“I have to express my gratitude to the ORNL staff for their continued support, through participation in committees and hosting and supporting events like this meeting,” Wolfe said.

ORNL is managed by UT-Battelle for the Department of Energy’s Office of Science, the single largest supporter of basic research for the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit