Added cheap parallelization.

2 files changed, 13 insertions, 5 deletions

diff --git a/Makefile b/Makefile
index 3c6169b..84be93f 100644
--- a/Makefile
+++ b/Makefile
@@ -4,7 +4,7 @@ BLOSSOM_DIRS := $(BLOSSOM_DIR) $(BLOSSOM_DIR)/MinCost $(BLOSSOM_DIR)/GEOM
 BLOSSOM_SOURCES := $(filter-out $(BLOSSOM_DIR)/example.cpp, $(foreach dir, $(BLOSSOM_DIRS), $(wildcard $(dir)/*.cpp)))

 BLOSSOM_OBJS := $(patsubst %.cpp, %.o, $(BLOSSOM_SOURCES))

 

-CFLAGS := -flto -O3 -Os -march=native -D_NDEBUG -DPERFECT_MATCHING_DOUBLE

+CFLAGS := -flto -fopenmp -O3 -Os -march=native -D_NDEBUG -DPERFECT_MATCHING_DOUBLE

 CXX = clang++

 LD = ld.lld

 LIBS := -lrt

diff --git a/order.cpp b/order.cpp
index 3ca57af..c2deb38 100644
--- a/order.cpp
+++ b/order.cpp
@@ -25,8 +25,14 @@ int main(int argc, char* argv[]) {
   Rng r;
   Graph G(n, r);
 
-  std::vector<Coordinate> data(G.vertices.size() / 2);
+#pragma omp declare reduction(vec_int_plus : std::vector<long int> : \
+                                  std::transform(omp_out.begin(), omp_out.end(), omp_in.begin(), omp_out.begin(), std::plus<long int>())) \
+                      initializer(omp_priv = decltype(omp_orig)(omp_orig.size()))
 
+  std::vector<long int> data_x(G.vertices.size() / 2);
+  std::vector<long int> data_y(G.vertices.size() / 2);
+
+#pragma omp parallel for reduction(vec_int_plus : data_x) reduction(vec_int_plus : data_y)
   for (unsigned i = 0; i < m; i++) {
     PerfectMatching pm(G.vertices.size(), G.edges.size());
 
@@ -40,15 +46,17 @@ int main(int argc, char* argv[]) {
     for (unsigned i = 0; i < G.vertices.size() / 2; i++) {
       unsigned j = pm.GetMatch(i);
 
-      data[i] += G.vertices[i].coordinate;
-      data[i] -= G.vertices[j].coordinate;
+      data_x[i] += G.vertices[i].coordinate[0];
+      data_y[i] += G.vertices[i].coordinate[1];
+      data_x[i] -= G.vertices[j].coordinate[0];
+      data_y[i] -= G.vertices[j].coordinate[1];
     }
   }
 
   std::cout << n << std::endl;
 
   for (unsigned i = 0; i < G.vertices.size() / 2; i++) {
-    std::cout << data[i][0] << " " << data[i][1] << std::endl;
+    std::cout << data_x[i] << " " << data_y[i] << std::endl;
   }
 
   return 0;