Intel Tool Aims To Help Developers Move From OpenACC To OpenMP

A new open-source tool made public by Intel engineers last month aims to help migrate codebases from using OpenACC to OpenMP. In turn the OpenMP-based offloading is preferred for Intel's XPU offloading strategy.

Thank god for terminal command line completion as the intel-application-migration-tool-for-openacc-to-openmp tool as it's called is a mouthful. The intel-application-migration-tool-for-openacc-to-openmp command aims to do as implied by its name of helping to automatically convert OpenACC constructs to the semantically-equivalent OpenMP. Automatic conversion of OpenACC constructs to OpenMP has been attempted by various parties/projects in the past to mixed success.

The intel-application-migration-tool-for-openacc-to-openmp is written in Python (3.6+) and aims to get OpenACC codebases converted to OpenMP with minimal time/input required of the user. However, the generated OpenMP code isn't necessarily the most efficient, so it will still require some developer intervention and profiling to ensure the most optimal code is created. The intel-application-migration-tool-for-openacc-to-openmp currently supports C/C++ and Fortran code.

While there has been OpenACC work for both GCC and LLVM/Clang, OpenMP is more of a common focus for Intel both for CPUs and offloading to GPUs and other XPUs -- and generally better supported by other vendors too, aside from NVIDIA. OpenMP offloading for Intel accelerators is supported using their oneAPI DPC++/C++ and oneAPI Fortran compilers.

Downloads for this 3-clause BSD licensed tool and more information on the tool's capabilities can be found via Intel's GitHub.