diff --git a/amgprec/impl/aggregator/MatchBoxPC.h b/amgprec/impl/aggregator/MatchBoxPC.h
index 530933e5..88e205ba 100644
--- a/amgprec/impl/aggregator/MatchBoxPC.h
+++ b/amgprec/impl/aggregator/MatchBoxPC.h
@@ -64,6 +64,7 @@
 #include "dataStrStaticQueue.h"
 
 using namespace std;
+#define NUM_THREAD 4
 
 #ifdef __cplusplus
 extern "C"
@@ -203,6 +204,13 @@ extern "C"
                            staticQueue &privateQMsgType,
                            staticQueue &privateQOwner);
 
+    inline void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
+                    MilanLongInt *verLocPtr,
+                    MilanLongInt *verLocInd,
+                    MilanInt myRank, 
+                    MilanReal *edgeLocWeight,
+                    MilanLongInt *candidateMate);
+
     void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
         MilanLongInt NLVer, MilanLongInt NLEdge,
         MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanReal *edgeLocWeight,
diff --git a/amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp b/amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp
index d470b1ab..980824aa 100644
--- a/amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp
+++ b/amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp
@@ -5,6 +5,7 @@
 #include "findOwnerOfGhost.cpp"
 #include "computeCandidateMate.cpp"
 #include "initialize.cpp"
+#include "parallelComputeCandidateMateB.cpp"
 
 // ***********************************************************************
 //
@@ -258,26 +259,22 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
     /////////////////////////////////////////////////////////////////////////////////////////
     // Compute the Initial Matching Set:
 
+    /*
+     * OMP PARALLEL_COMPUTE_CANDIDATE_MATE_B has been splitted from
+     * PARALLEL_PROCESS_EXPOSED_VERTEX_B in order to better parallelize
+     * the two.
+     * PARALLEL_COMPUTE_CANDIDATE_MATE_B is now totally parallel.
+     */
+
+    PARALLEL_COMPUTE_CANDIDATE_MATE_B(NLVer,
+                                      verLocPtr,
+                                      verLocInd,
+                                      myRank,
+                                      edgeLocWeight,
+                                      candidateMate);
+
 #pragma omp parallel private(k, u, w, v, k1, adj1, adj2, adj11, adj12, heaviestEdgeWt, ghostOwner, privateMyCard, isEmpty) firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner) default(shared) num_threads(4)
     {
-        /*
-         * OMP PARALLEL_COMPUTE_CANDIDATE_MATE_B has been splitted from
-         * PARALLEL_PROCESS_EXPOSED_VERTEX_B in order to better parallelize
-         * the two.
-         * In particular PARALLEL_COMPUTE_CANDIDATE_MATE_B is now totally parallel.
-         */
-
-#pragma omp for schedule(static)
-        for (v = 0; v < NLVer; v++)
-        {
-#ifdef PRINT_DEBUG_INFO_
-            cout << "\n(" << myRank << ")Processing: " << v + StartIndex << endl;
-            fflush(stdout);
-#endif
-            // Start: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
-            candidateMate[v] = firstComputeCandidateMate(verLocPtr[v], verLocPtr[v + 1], verLocInd, edgeLocWeight);
-            // End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
-        }
 
         /*
          * PARALLEL_PROCESS_EXPOSED_VERTEX_B
@@ -476,7 +473,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
             }
         }
 
-#pragma omp master
+#pragma omp single
         {
             tempCounter.clear(); // Do not need this any more
         }
@@ -715,11 +712,11 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
                                     candidateMate[NLVer + Ghost2LocalMap[v]] = -1;
                                 if (v != Mate[u - StartIndex])
                                 { // u is local
-                                    // Build the Message Packet:
-                                    // Message[0] = u; //LOCAL
-                                    // Message[1] = v; //GHOST
-                                    // Message[2] = SUCCESS;  //TYPE
-                                    // Send a Request (Asynchronous)
+                                  // Build the Message Packet:
+                                  // Message[0] = u; //LOCAL
+                                  // Message[1] = v; //GHOST
+                                  // Message[2] = SUCCESS;  //TYPE
+                                  // Send a Request (Asynchronous)
 
 #ifdef PRINT_DEBUG_INFO_
                                     cout << "\n(" << myRank << ")Sending a success message: ";
diff --git a/amgprec/impl/aggregator/parallelComputeCandidateMateB.cpp b/amgprec/impl/aggregator/parallelComputeCandidateMateB.cpp
new file mode 100644
index 00000000..ced93456
--- /dev/null
+++ b/amgprec/impl/aggregator/parallelComputeCandidateMateB.cpp
@@ -0,0 +1,36 @@
+#include "MatchBoxPC.h"
+#include <stdio.h>
+#include <iostream>
+#include <assert.h>
+#include <map>
+#include <vector>
+#include "primitiveDataTypeDefinitions.h"
+#include "dataStrStaticQueue.h"
+#include "omp.h"
+
+inline void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
+                                              MilanLongInt *verLocPtr,
+                                              MilanLongInt *verLocInd,
+                                              MilanInt myRank,
+                                              MilanReal *edgeLocWeight,
+                                              MilanLongInt *candidateMate)
+{
+
+    MilanLongInt v = -1;
+
+#pragma omp parallel private(v) default(shared) num_threads(4)
+    {
+
+#pragma omp for schedule(static)
+        for (v = 0; v < NLVer; v++)
+        {
+#ifdef PRINT_DEBUG_INFO_
+            cout << "\n(" << myRank << ")Processing: " << v + StartIndex << endl;
+            fflush(stdout);
+#endif
+            // Start: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
+            candidateMate[v] = firstComputeCandidateMate(verLocPtr[v], verLocPtr[v + 1], verLocInd, edgeLocWeight);
+            // End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
+        }
+    }
+}