Best practices for performance


OpenMP is an application programming interface (API) for shared-memory multiprocessor system for C, C++ and Fortran. It is supported by all major compilers on many different platforms and operating systems.

The idea behind OpenMP is simple: the developer writes the code in the way they are used to, but in addition they use special OpenMP compiler pragmas to provide the information needed to run the code on a multithreaded system.

The OpenMP runtime is in charge of distributing the problem onto several CPU cores, data movements and data synchronization. The model is lightweight and has found usage in many domains where multiprocessor programming is important.

Originally, OpenMP was created as an API for shared-memory multiprocessor systems. From OpenMP 4.0, it also supports offloading the computations to accelerators.