B2 - parallel regions and work sharing

parallel region

• the parallel directive defines a region of code executed on all threads parallelly

• the number of threads can be determined at runtime

• the default value is the number of processors (usually CPU cores)

• this can be overridden by:

  • setting the environment variable OMP_NUM_THREADS
  • calling omp_set_num_threads(P)
  • using the modifier num_threads(P) when declaring the parallel region

• the parallel directive creates necessary parallelism but does not share the work

• eg: in the following block of code, for P = 12, there will be 120 messages:

#pragma omp parallel private(i)
for (i = 0; i < 10; i++) {
	printf("Hello world %d\n", i);
}

• so, instead of sharing, the work is being replicated on each thread

work sharing

• the parallel for directive is a work-sharing construct
• it complements the parallel construct

#pragma omp parallel for
for (i = 0; i < 10; i++) {
	printf("Hello world %d\n", i);
}

• now, for P = 12, 10 messages will be printed

• different iterations execute concurrently on $P$ threads

• i is private by default in a for directive

• conditional compilation can be useful to compile statements only if OpenMP support is enabled:

#ifdef _OPENMP
... statememnts ...
#endif