Parallel Computing for Data Science

5.3. OPENMP LOOP SCHEDULING OPTIONS 125

and

#pragma omp for schedule(dynamic,50)

The keywords static and dynamic correspond to the scheduling strategies

presented in Section 3.1, with the optional second argument being chunk

size as discussed in that section. The static version assigns chunks before

the loop is executed, parceled out in Round Robin manner.

The third scheduling option is guided. It uses a large chunk size in early

iterations, but tapers down the chunk size as the execution of the loop pro-

gresses. This strategy, also discussed in Section 3.1, is designed to minimize

overhead in the early rounds, but minimize load imbalance later on.