OpenACC

OpenACC is a parallel programming standard designed to simplify writing code that executes on accelerators such as GPUs, while maintaining portability across architectures. It uses compiler directives (pragmas) that annotate C, C++, and Fortran code to indicate which regions can be offloaded to an accelerator and how data should be moved between host and device. The goal is to enable incremental adoption: existing scalar code can be gradually annotated to exploit parallel hardware with minimal restructuring.

OpenACC provides data directives (copy, copyin, copyout, create, present) to manage memory and compute directives (kernels, parallel) to specify regions of code to execute on the device. It also supports loop-level directives (independent, collapse) and a hierarchical execution model (gang, worker, vector) to map work to accelerators. Offloading is designed to be portable: the same OpenACC-annotated code can run on different accelerator architectures supported by a compliant compiler.

The standard is governed by the OpenACC standards committee and is hosted at openacc.org, with evolution through multiple versions since its inception in 2011. Compiler support exists from several vendors, with tools such as the NVIDIA HPC SDK and Cray compilers providing OpenACC support; other compilers have partial or evolving support.

OpenACC is commonly used in high-performance computing for scientific simulations, data analysis, and workloads that benefit from parallel accelerators, offering portability and a higher-level programming model compared with device-specific APIs. See also OpenMP, CUDA, OpenCL.