Try improve OpenMP version of matchbox

amgprec/amg_d_matchboxp_mod.f90

@@ -724,6 +724,7 @@ contains
            & vnl, mate, iam, np,ictxt,&
            & msgis,msgas,msgprc,ph0t,ph1t,ph2t,ph1crd,ph2crd,info,display_inp)
       if (do_timings) call psb_toc(idx_cmboxp)
+      if (iam==0) write(0,*) iam,' buildmatching from PMatchBox:', info,ph0t,ph1t,ph2t
       if (debug) write(0,*) iam,' buildmatching from PMatchBox:', info
       if (debug_sync) then
         call psb_max(ictxt,info)

amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp

@@ -489,7 +489,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
         ///////////////////////////////////////////////////////////////////////////////////
         /////////////////////////// PROCESS MESSAGES //////////////////////////////////////
         ///////////////////////////////////////////////////////////////////////////////////
-
+	//startTime = MPI_Wtime();
         processMessagesD(NLVer,
                         Mate,
                         candidateMate,

amgprec/impl/aggregator/processExposedVertex.cpp

@@ -33,12 +33,13 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_BD(MilanLongInt NLVer,
   MilanLongInt v = -1, k = -1, w = -1, adj11 = 0, adj12 = 0, k1 = 0;
     MilanInt ghostOwner = 0, option, igw;
 
-    //#pragma omp parallel private(option, k, w, v, k1, adj11, adj12, ghostOwner) \
+#pragma omp parallel private(option, k, w, v, k1, adj11, adj12, ghostOwner) \
-    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner) \
+  firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx,	\
+	       privateQGhostVtx, privateQMsgType, privateQOwner)	\
   default(shared)   num_threads(NUM_THREAD)
 
     {
-      //#pragma omp for reduction(+			\
+      #pragma omp for reduction(+			\
                           : PCounter[:numProcs], myCard \
                           [:1], msgInd                  \
                           [:1], NumMessagesBundled      \
@@ -216,12 +217,13 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_BS(MilanLongInt NLVer,
     MilanLongInt v = -1, k = -1, w = -1, adj11 = 0, adj12 = 0, k1 = 0;
     MilanInt ghostOwner = 0, option, igw;
 
-    //#pragma omp parallel private(option, k, w, v, k1, adj11, adj12, ghostOwner) \
+#pragma omp parallel private(option, k, w, v, k1, adj11, adj12, ghostOwner) \
-    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner) \
+    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, \
+		 privateQGhostVtx, privateQMsgType, privateQOwner) \
      default(shared)   num_threads(NUM_THREAD)
 
     {
-      //#pragma omp for reduction(+			\
+#pragma omp for reduction(+				\
                           : PCounter[:numProcs], myCard \
                           [:1], msgInd                  \
                           [:1], NumMessagesBundled      \

amgprec/impl/aggregator/processMatchedVerticesAndSendMessages.cpp

@@ -51,8 +51,9 @@ void processMatchedVerticesAndSendMessagesD(
     MilanLongInt localVertices = 0;
 #endif
     //#pragma omp parallel private(k, w, v, k1, adj1, adj2, adj11, adj12, ghostOwner, option) \
-    firstprivate(Message, privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx,\
+  firstprivate(Message, privateU, StartIndex, EndIndex, privateQLocalVtx, \
-		 privateQMsgType, privateQOwner, UChunkBeingProcessed) default(shared) \
+	       privateQGhostVtx, privateQMsgType, privateQOwner, UChunkBeingProcessed) \
+  default(shared)							\
       num_threads(NUM_THREAD)						\
   reduction(+								\
 	    : msgInd[:1], PCounter					\