B1 - OpenMP

• it includes a set of compiler directives to describe parallelism in the source code

• also includes a subroutine library

• uses a set of environment variables

• parallelism based on threads

• the program begins as a single process, the master thread

• it executes sequentially until the first fork, where a team of parallel threads is created

• upon completion, the threads synchronise and terminate or join, bringing control back to a single master thread

• in C, directives take the form of pragmas

#pragma omp parallel default(shared) private(x,y)

• for a sequential compilation, it will be interpreted as comments
• each C directive applies only to the next statement or block, hence a structured block {...} is always needed

#include <omp.h>
#include <stdio.h>

int main()
{
    int nthreads, tid;

    #pragma omp parallel private(tid)
    {
        tid = omp_get_thread_num();
        printf("Hello world from thread = %d\n", tid);

        if (tid == 0)
        {
            nthreads = omp_get_num_threads();
            printf("Number of threads = %d\n", nthreads);
        }
    }
}

• can combine parallel directive with for directive:

#pragma omp parallel for
	for (i = 0; i < i_max; i++)
	{
		z[i] = a * x[i] + y;
	}

• there are no data dependencies, and each iteration must be independent of the others
• the loop iterations get split across threads