<span class="var-sub_title">OpenACC API User Experience, Vendor Reaction, Relevance, and Roadmap</span> SC18 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

OpenACC API User Experience, Vendor Reaction, Relevance, and Roadmap


Authors: Duncan Poole (Nvidia Corporation), Sunita Chandrasekaran (University of Delaware), Robert Henschel (Indiana University)

Abstract: OpenACC, a directive-based high-level parallel programming model, has gained rapid momentum among scientific application users - the key drivers of the specification. The user-friendly programming model has facilitated acceleration of over 130 applications including CAM, ANSYS Fluent, Gaussian, VASP, Synopsys on multiple platforms and is also seen as an entry-level programming model for the top supercomputers (Top500 list) such as Summit, Sunway Taihulight, and Piz Daint. As in previous years, this BoF invites scientists, programmers, and researchers to discuss their experiences in adopting OpenACC for scientific applications, learn about the roadmaps from implementers and the latest developments in the specification.

Long Description: OpenACC is a widely popular model used by application developers to accelerate and parallelize applications on multicore and massively parallel processors while preserving a single code base and enabling cross-platform portability. With a ‘more science, less programming’ motto, OpenACC fosters a cross-platform API complementary to and interoperable with OpenMP, MPI, CUDA, and OpenCL. A number of applications including Gaussian, ANSYS, and VASP have been ported to GPUs using OpenACC.

The model targets X86 processors (such as Intel Skylake, AMD EPYC), OpenPOWER processors, NVIDIA’s GPUs, and the PEZY-SC manycore processor, FPGAs, AMD GPUs and Sunway’s Shenwei proceessor. OpenACC non-commerical source-to-source compilers include OpenARC and Omni compilers. Current discussions center on technical challenges to overcome to support the next generation of HPC platforms such as ARM. The standard values such discussions, that are often taking place in research groups and academic institutions, and to further encourage broad participation in the standard, voting rights have been amended accordingly; details will be shared with BoF attendees.

This interactive BoF will bring together the user and developer communities to brainstorm new feature requests, clarifications of definitions and implementations of existing features, and ideas for future direction. The direction of the standard is heavily driven by OpenACC users/application developers. They provide feedback via usergroup meetings held 3-4 times a year. Feedback includes new feature requests or clarifications to existing ones. An updated version of the specification with revised definitions for some of the current features and how OpenACC plans to address different memory models, will be released during the SC time frame. We will share updates on two topics of great importance to attendees deep copy (user requested feature) and support for OpenACC in GCC.

Representatives from SPEC HPG will present results of the OpenACC dataset of the SPEC ACCEL benchmark, comparing OpenACC compilers’ performances on different hardware and software architectures. This will be used as a jumping off point to discuss user experience with different OpenACC compilers.

The topic on Hackathons has always kindled a vibrant conversation among the attendees who are either seeking help to accelerate their codes or are interested to either hosting similar hackathons or sending teams to such events. Since SC14, over 125 kernels, miniapps and applications spanning astrophysics, climate modeling, nuclear physics and quantum chemistry have been ported to large scale machines using OpenACC via hackathons held around the globe.

We will share success stories from two major applications at the BoF. (1) Using OpenACC to port FLASH (ORNL CAAR code), a publicly available, massively parallel AMR code; (2) Using OpenACC to accelerate the Synopsys TCAD Sentaurus Device EMW simulator to speed up optical simulations of image sensors.

Training and Education: Attendees will be updated on the lecture slides, Jupyter notebooks and availability of example codes via docker/containers that they can use.

New OpenACC members will share their enthusiasm and motivation behind joining the organization.

All of the above topics have steered excellent discussions among the attendees in the past years. We will ensure to keep up the momentum.





Back to Birds of a Feather Archive Listing