Introduction to OpenMP

• A parallel programming model, an API
• Comprised of
  • Compiler directives
  • Runtime Library Routines
  • Environment Variables
• Programming Model

• All OpenMP programs begin as a single process: the master thread. The master thread executes sequentially until the first parallel region construct is encountered.
• FORK: the master thread then creates a team of parallel threads.

• JOIN: When the team threads complete the statements in the parallel region construct, they synchronize and terminate, leaving only the master thread.

• Compiler directives statements (start with #pragma omp) tell compiler that here follows a parallel region

  • a parallel region is bounded with { }
• Runtime Library Routines can give you some informations about your environment
  • omp_get_thread_num(). Tells the thread id of the current thread.
  • omp_get_num_threads(). Tells the total thread number. If you don't tell compiler how many threads you want to have in your program, it will be the number of your cores by default.
  • ... you can find more on internet about this.
• Environment Variables tell compiler that some variables are public and some variables are private.
  • As OpenMP is a shared memory model, variables are shared by threads by default.
  • You can define private variables
    • by private() statement.
    • variables defined in parallel region are private variables.
• More details later.