Open-Source Internship opportunity by OpenGenus for programmers. Apply now.
Reading time: 15 minutes
OpenMP is an open source library which is used for multi-threaded parallel processing and shared-memory multi-processor (core) computers. In OpenMP, part of program is a single thread and part is multi-threaded.
OpenMP has 3 components:
- directives that can be put into C/C++ or Fortran programs
- runtime library for setting and querying parallel parameters (number of threads)
- environment variables that define runtime parallel parameters (number of threads)
Approach
Two main approaches of OpenMP:
- loop-level
- parallel regions
Loop-Level Parallelism / fine-grained parallelism
- individual loops parallelized
- each thread assigned a unique range of the loop index
- execution starts on a single serial thread
- multiple threads are spawned inside a parallel loop
- after parallel loop execution is serial
- relatively easy to implement
Parallel Regions Parallelism / coarse-grained parallelism
- any sections of codes can be parallelized (not just loops)
- using the thread identifier to distribute the work
- execution starts on a single serial thread
- multiple threads are started for parallel regions (not necessarily at a loop)
- ends on a single serial thread
Advantages of OpenMP
Advantages of OpenMP are:
- Prevalence of multi-core computers
- Requires less code modification than using MPI
- OpenMP directives can be treated as comments if OpenMP is not available
- Directives can be added incrementally
Disadvantages of OpenMP
Disadvantages of OpenMP are:
- OpenMP codes cannot be run on distributed memory computers (exception is Intel's OpenMP)
- Requires a compiler that supports OpenMP (most do)
- limited by the number of processors available on a single computer
- often have lower parallel efficiency
- rely more on parallelizable loops
- tend to have a higher % of serial code
- Amdahl's Law - if 50% of code is serial will only half wall clock time no matter how may processors
Application
OpenMP is used in:
- Matlab
- Mathematica
- Libraries
- Intel Math kernel Library (MKL)
- GNU Scientific Library (GSL)
