OpenMP

What is it?

In Distributed Computing, OpenMP is a C++, C and Fortran programming interface which implements multi-threading parallel programming. It basically consists of a set of compiler directives, routines and environment variables which will influence the run-time execution to spawn multiple threads to process a certain region of the code.

For example, for a code which will calculate an integral version of $\pi$, expressed by: $\pi =4{\int }_0^1\frac{1}{1+x^2}dx$Can be implemented for each thread to calculate each set of rectangles separately, improving performance:

#include<stdio.h>
#include<stdlib.h>
#include<omp.h>
#include<cmath>
 
constexpr float totalHeight = 4.0f;
constexpr float totalX = 1.0f;
constexpr int rectangles = 1000000;
 
float calculateX(int rectangleNumber) { // Rectangle center
    float rectangleWidth = totalX / rectangles;
    return (rectangleNumber * rectangleWidth) + (rectangleWidth / 2.0f);
}
 
float calculateY(float x) {
    return totalHeight / (1 + pow(x, 2.0));
}
 
float calculateArea(float y)
{
    float rectangleWidth = totalX / rectangles;
    return y * rectangleWidth;
}
 
int main() {
    double start = omp_get_wtime();
    float heights[rectangles];
    float totalHeight = 0.0f;
 
    // Calculates height of each rectangle of the integral.
    // Then sums all heights into a single variable for total area calculation.
    #pragma omp parallel for reduction(+:totalHeight)
    for (int i = 0; i < rectangles; i++) {
        double x = calculateX(i);
        double y = calculateY(x);
        totalHeight += y;  // Thread-safe because of the reduction clause
    }
 
    float totalArea = calculateArea(totalHeight);
    double totalTime = omp_get_wtime() - start;
 
    printf("Area: %f\n", totalArea);
    printf("Duration: %f\n", totalTime);
}