Lecture 3

Learning objectives

After this class, you should be able to:

1. Explain the following: (i) fork-join model used in OpenMP, (ii) shared variable, and (iii) private variable.
2. Explain the purpose of the following directives, clauses, and functions: (i) parallel, (ii) omp_get_thread_num, (iii) for, (iv) single, (v) nowait, (vi) sections, (vii) section, (viii) omp_get_wtime, (ix) omp_get_wtick, (x) parallel for, (xi) firstprivate, (xii) lastprivate, (xiii) private, (xiv) omp_get_num_procs, (xv) omp_set_num_threads, (xvi) omp_get_max_threads, (xvii) critical, (xviii) reduction, and (xix) schedule.
3. Given sequential code, use the above features to parallelize the code.
4. Given a code that is incorrectly parallelized using OpenMP, point out the errors.

Reading assignment

1. OpenMP tutorial at: http://www.llnl.gov/computing/tutorials/openMP.
2. Reference -- OpenMP 2 C/C++ specifications: http://www.openmp.org/mp-documents/cspec20.pdf.

Exercises and review questions

• Exercises and review questions on current lecture's material
  1. Parallelize the following loop using OpenMP.

     a = 0;
for (i=0; i < n; i++)
     

     a += b[i];
     

     Give an example to demonstrate the utility of the single pragma.

     Use OpenMP to parallelize matrix multiplication using the loop ordering that gave better performance sequentially.

     Compare the performance of the OpenMP example with inverted loop on slide 53 of lecture 3, with a code that parallelizes the inner loop of the original sequential code. Report your timings on the discussion board for for different values of m, n, and number of threads.

• Preparation for the next lecture
  1. None.

---

Last modified: 13 Jan 2010