🐛 Changed my Pi solution to only use #pragma omp for and not already the reduction sum

answers/02/pi_numerical_integration.cpp

@@ -1,6 +1,7 @@
 #include <iomanip>
 #include <iostream>
 #include <omp.h>
+#include <vector>
 
 using namespace std;
 
@@ -9,15 +10,28 @@ int main() {
   double width = 1.0 / double(num_steps); // width of a rectangle
   double sum = 0.0; // for summing up all heights of rectangles
 
+  int max_threads = omp_get_max_threads();
+  vector<double> thread_sums(max_threads, 0.0); // one sum per thread
+
+
   double start_time = omp_get_wtime(); // wall clock time in seconds
-  #pragma omp parallel for reduction(+:sum)
+  #pragma omp parallel
-  for (int i = 0; i < num_steps; i++) {
+  {
-    double x = (i + 0.5) * width; // midpoint
+    int thread = omp_get_thread_num();
-    sum = sum + (1.0 / (1.0 + x * x)); // add new height of a rectangle
+    
+    #pragma omp for
+    for (int i = 0; i < num_steps; i++) {
+      double x = (i + 0.5) * width; // midpoint
+      thread_sums[thread] += (1.0 / (1.0 + x * x)); // add new height of a rectangle
+    }
   }
+  for (int i = 0; i < max_threads; i++) {
+    sum+= thread_sums[i];
+  }
+
   double pi = sum * 4 * width; // compute pi
   double run_time = omp_get_wtime() - start_time;
 
   cout << "pi with " << num_steps << " steps is " << setprecision(17)
-       << pi << " in " << setprecision(6) << run_time << " seconds\n";
+  << pi << " in " << setprecision(6) << run_time << " seconds\n";
 }