Multicore DSP

12Introduction to OpenMP

CHAPTER MENU

1. 12.1 Introduction to OpenMP
2. 12.2 Directive formats
3. 12.3 Forking region
   1. 12.3.1 omp parallel – parallel region construct
      1. 12.3.1.1 Clause descriptions
4. 12.4 Work‐sharing constructs
   1. 12.4.1 omp for
      1. 12.4.1.1 OpenMP loop scheduling
   2. 12.4.2 omp sections
   3. 12.4.3 omp single
   4. 12.4.4 omp master
   5. 12.4.5 omp task
5. 12.5 Environment variables and library functions
6. 12.6 Synchronisation constructs
   1. 12.6.1 atomic
      1. 12.6.1.1 Clauses
   2. 12.6.2 barrier
   3. 12.6.3 critical
7. 12.7 OpenMP accelerator model
   1. 12.7.1 Supported OpenMP device constructs
      1. 12.7.1.1 #pragma omp target
      2. 12.7.1.2 #pragma omp target data
      3. 12.7.1.3 #pragma omp target update
      4. 12.7.1.4 #pragma omp declare target
8. 12.8 Laboratory experiments
   1. 12.8.1 Laboratory experiment 1
   2. 12.8.2 Laboratory experiment 2
   3. 12.8.3 Laboratory experiment 3
   4. 12.8.4 Laboratory experiment 4
   5. 12.8.5 Laboratory experiment 5
9. 12.9 Conclusion
10. References

Chapter 1 showed that continually increasing the clock frequency of a processor in order to increase the performance is not an option anymore, and the way forward is to increase the number of cores. However, this introduces other problems; for instance, was the application written for parallel processing? Can this application be parallelised, and has its performance increased as a consequence of this parallelism? How much effort has to be put for parallelising it? And so on.

To solve this, software parallelisation is not an easy task for a compiler alone to accomplish. Software parallelisation ...