omp

OpenMP, short for Open Multi-Processing, is an API and ecosystem for shared-memory parallel programming in C, C++, and Fortran. It provides a portable, scalable model that enables developers to express parallelism through compiler directives, library routines, and environment variables, with the goal of improving performance on multi-core and multi-processor systems without altering core language syntax.

Core features of OpenMP include a range of parallel and work-sharing constructs. Parallel regions create multi-threaded execution, while work-sharing constructs such as for (or do in Fortran), sections, and single distribute work among threads. Synchronization is offered through barrier, critical, atomic, and ordered constructs. The data environment is controlled with clauses that specify how variables are shared or privatized among threads (shared, private, firstprivate, lastprivate, reduction). OpenMP also supports dynamic scheduling, nested parallelism, and, in newer versions, task-based parallelism with constructs like task and taskwait, as well as support for vectorization via SIMD directives. Since OpenMP 4.0, devices such as GPUs can be targeted through offload directives like target, teams, and distribute, with data movement controlled by map clauses.

Governance and ecosystem: the standard is developed and maintained by the OpenMP Architecture Review Board (ARB). OpenMP specifications have evolved through multiple versions, adding features for threads, tasks, atomics, SIMD, and accelerator offloading. Implementations are available in major compilers including GCC, Clang (libomp), Intel’s ICC/OneAPI, IBM, and others, often with additional vendor-specific extensions. OpenMP is widely used in scientific computing, engineering, and data processing to exploit shared-memory parallelism while remaining portable across platforms. Limitations include overhead from threading, dependence on shared memory, and varying performance characteristics across compilers and architectures.