Today at SC14, the OpenMP Architecture Review Board (ARB) announced a Code Conversion Challenge. The Challenge is to users to bring in shared memory or accelerator code (C++ AMP, CUDA, OpenACC®, OpenCL®, …) from real applications. The ARB will then demonstrate how to write them in modern OpenMP.
The applications can be written in any parallel programming model, either shared-memory or accelerator based. The OpenMP ARB will convert the applications to OpenMP, with a focus on source to source conversion, especially of the parallel regions of interest.
OpenMP, one of the most-capable high-level parallel languages
The ARB aims to show that reasonable programs written in another parallel programming model (e.g., OpenCL, Co-Array-Fortran, OpenACC, CUDA, C++, C++AMP or Cilk) can be rewritten in OpenMP and thus that modern OpenMP is among the most-capable high-level parallel languages.
Submission of applications
Applicants are welcome to submit their applications on the site https://github.com/OpenMP/Challenge. If the submitter agrees, the application will be listed on the OpenMP web site.
“We have designed modern OpenMP to be the best parallel language for the three general purpose languages C, C++ and Fortran”, said Michael Wong, OpenMP CEO. “It is a language that enables you to access all capabilities of your machine without dropping to another language. This Challenge will convince you of that or show us where we need to improve. Either way, the industry benefits.”