Authors: Daniel Howard (NSF National Center for Atmospheric Research (NCAR)), Fernanda Foertter (University of Alabama), Jack Wells (NVIDIA Corporation), Thomas Schwinge (BayLibre)
Abstract: Through its mission to reduce time to science by nurturing parallel computing skills, the Open Accelerated Computing Consortium (OpenACC) invites you to explore the future of directive-based programming. In the age of agentic coding assistants, this BoF examines the evolving roles of OpenACC, OpenMP, standard language parallelism, and abstraction libraries like Julia for Accelerators or Kokkos for GPU-resident runtimes. Panels including compiler developers and community leaders will summarize the current state of programming for accelerators, followed by open discussion comparing practices and imagining new futures. Developers, researchers, compiler vendors, and the broader HPC community are welcome to shape the conversation.
Long Description: The Open Accelerated Computing Consortium (OpenACC), which stewards the OpenACC directive-based programming standard for parallel compute hardware, represents a vibrant community of academic, national laboratory, and industry institutions committed to accelerated high-performance computing. Developed via continuous feedback since 2011 from a diverse user base with usability and portable performance in mind, OpenACC provides an easy on-ramp to parallelism on many devices, including x86, Arm, and OpenPOWER CPUs as well as NVIDIA and AMD GPUs, PEZY-SC manycore processors, and FPGAs.
This year's BoF is organized around the theme of Programming Accelerators with Directives in the Age of AI. As large language models and agentic coding assistants reshape how developers write, port, and optimize accelerated software, the community faces new questions about the role directives will play alongside these emerging tools. This includes new opportunities to lower the barrier to entry for developers without deep GPU expertise. Rather than a status update on implementations of OpenACC alone, this session invites attendees to compare current approaches and imagine divergent futures for accelerating code across a variety of accelerator architectures and vendor platforms.
Opening panels will survey the current landscape of directive-based programming spanning vendors building compilers to HPC users running the code. Implementers from open-source and vendor-supported projects will report progress on OpenACC, with explicit attention to portability across NVIDIA and AMD GPUs, x86, Arm, and other architectures. Discussions will situate OpenACC alongside OpenMP, ISO standard language parallelism (C++ and Fortran), and abstraction libraries such as Kokkos and Julia for Accelerators (JACC), examining where each excels and how they interoperate for GPU-resident runtimes. This comparative, portability-first framing directly addresses the concerns application developers raise about performance on heterogeneous platforms.
Following brief presentations, a fishbowl discussion and open Q&A will surface successes, challenges, feature requests, and clarifications from users and application developers. We will also probe how AI-assisted tooling might change that feedback loop itself. The bulk of the session will be deliberately interactive because the direction of the OpenACC specification is driven by this community. Your participation and feedback gathered here feeds directly into the technical committee's roadmap such that attendees will see their input shape subsequent revisions and future priorities of the OpenACC community.
Over its 15 year history, OpenACC's impact is well established, including over 200 applications such as Quantum ESPRESSO, NWChem, ICON, VASP, Gaussian, and OVERFLOW run in production on systems such as Frontier, Perlmutter, LUMI, JUWELS, Alps, Miyabi, and LineShine. The model remains complementary to and interoperable with OpenMP, MPI, CUDA, and accelerated numerical libraries, while the organization increasingly emphasizes performance and portability through ISO standard languages kept current with the OpenACC specification. Nonetheless, the future of directive based programming must adapt alongside other competing standards in order to collaborate effectively with the whole HPC community and achieve the primary goal of reducing time to science.
We welcome software developers, researchers, educators, compiler vendors, and the broader HPC community, especially first-time attendees and users of non-NVIDIA platforms, to attend and shape this conversation.
https://www.openacc.org
https://www.openhackathons.org
Website: https://www.openacc.org/events/openacc-birds-feather-bof-sc26