OpenMP

include/k-means.h

@@ -2,10 +2,10 @@
 #define __KMEANS__
 
 #include <vector>
-#include "Point.h"
+#include "point.h"
 #include "readWrite.h"
 #include <iostream> // debug
 
 void KMeans(std::vector<Point>* p,int epochs, int k);
 
-#endif
+#endif

include/readWrite.h

@@ -5,9 +5,9 @@
 #include <sstream>
 #include <vector>
 
-#include "Point.h"
+#include "point.h"
 
 std::vector<Point> readCsv(std::string path);
 void writeCsv(std::vector<Point>* p, std::string s);
 
-#endif
+#endif

src/k-means.cpp

@@ -10,20 +10,23 @@
 #include "../include/k-means.h"
 #include <omp.h>
 // #define DEBUG
-// #define OMP
+//#define OMP
 
 void KMeans(std::vector<Point>* p,int epochs, int k){
     std::vector<Point> centroids;
     std::srand(time(nullptr));
     std::vector<int> npts(p->size());
-    std::vector<double> sx(k), sy(k), sz(k);
+    std::vector<double> sx(k), sy(k), sz(k);    //cumulative x, y, x 
     #ifdef DEBUG
         std::cout << "=== DEBUG ===" << std::endl;
     #endif
     // init clusters with some random point
     #ifdef OMP
     // omp_set_num_threads(k);
     #pragma omp num_threads(k)
+	#ifdef DEBUG
+		std::cout << "N threads : [" << omp_get_num_threads() << std::cout << "]\n"; 
+	#endif
     #pragma omp parallel for 
     #endif
     for(size_t i = 0; i < k; i++){
@@ -39,18 +42,15 @@ void KMeans(std::vector<Point>* p,int epochs, int k){
         std::cout << "epoch - " << it << "/" << epochs << std::endl; 
         #endif
         // assign pts to a cluster
+		size_t j;
         #ifdef OMP
-	// fix different directive needed
-        // #pragma omp parallel for 
+        #pragma omp parallel for private(j) 
         #endif
         for(size_t i = 0; i < centroids.size(); ++i)
         {
             #ifdef DEBUG
             std::cout<<"centroid N: [" << i << "]" << std::endl; 
             #endif
-            #ifdef OMP
-            #pragma omp parallel for 
-            #endif
             for(size_t j = 0;j < p->size(); ++j)
             {
                 #ifdef DEBUG
@@ -77,6 +77,14 @@ void KMeans(std::vector<Point>* p,int epochs, int k){
             p->at(i).setMinDist(__DBL_MAX__); // reset
         }
         //compute new centroids
+		#ifdef OMP
+		omp_set_num_threads(centroids.size());
+		int cthreads = omp_get_num_threads();
+		#ifdef DEBUG
+		std::cout << "N cthreads : [" << cthreads << std::cout << "]\n";
+		#endif
+		#pragma omp parallel for 
+		#endif
         for(size_t i = 0;i < centroids.size(); ++i)
         {
             centroids.at(i).setX(sx[i] / npts[i]);

src/main.cpp

@@ -3,7 +3,7 @@
 #include <vector>
 #include <string>
 
-#include "../include/Point.h"
+#include "../include/point.h"
 #include "../include/k-means.h"
 #include "../include/readWrite.h"
 
@@ -26,12 +26,12 @@ int main(int argc, char** argv){
             std::cerr << e.what() << '\n';
             return EXIT_FAILURE;
         }
-        clock_t stop = clock();
-        clock_t elapsed = (((float)(stop - start)) / CLOCKS_PER_SEC) * 1000; // ms
-        std::cout<< "[ K-Means clustering algorithm ]" << std::endl;
-        std::cout<< "-> Details : data [ " << path << "] | " << epochs << " epochs | " 
+		clock_t stop = clock();
+        clock_t elapsed = (((float)(stop - start)) / CLOCKS_PER_SEC) * 1000; // s
+        std::cout<< "[ K-Means clustering algorithm ]\n" <<
+        			"-> Details : data [ " << path << "] | " << epochs << " epochs | " 
                 << clusters << " clusters;" << std::endl;
-        std::cout << "Sequential running time : "<< elapsed << " ms" << std::endl;
+        std::cout << "Run time : "<< elapsed << " s" << std::endl;
         writeCsv(&points, "data/out.csv");
     }
     return EXIT_SUCCESS;

src/readWrite.cpp

@@ -28,4 +28,4 @@ void writeCsv(std::vector<Point>* p, std::string s){
         file << it->getX() << ", " << it->getY() << ", " << it->getZ() <<", " << it->getK() << std::endl;
     }
     file.close();
-}
+}