Merge branch 'PolySmooth' into development

amgprec/impl/Makefile

@@ -22,22 +22,22 @@ MPFOBJS=$(SMPFOBJS) $(DMPFOBJS) $(CMPFOBJS) $(ZMPFOBJS)
 MPCOBJS=amg_dslud_interface.o amg_zslud_interface.o
 
 
-DINNEROBJS= amg_dmlprec_bld.o amg_dfile_prec_descr.o \
+DINNEROBJS= amg_dmlprec_bld.o amg_dfile_prec_descr.o amg_dfile_prec_memory_use.o \
 	amg_d_smoothers_bld.o amg_d_hierarchy_bld.o amg_d_hierarchy_rebld.o \
 	amg_dmlprec_aply.o   \
 	$(DMPFOBJS) amg_d_extprol_bld.o
 
-SINNEROBJS=   amg_smlprec_bld.o amg_sfile_prec_descr.o \
+SINNEROBJS=   amg_smlprec_bld.o amg_sfile_prec_descr.o amg_sfile_prec_memory_use.o \
 	amg_s_smoothers_bld.o amg_s_hierarchy_bld.o amg_s_hierarchy_rebld.o \
 	amg_smlprec_aply.o   \
 	$(SMPFOBJS) amg_s_extprol_bld.o 
 
-ZINNEROBJS=   amg_zmlprec_bld.o amg_zfile_prec_descr.o \
+ZINNEROBJS=   amg_zmlprec_bld.o amg_zfile_prec_descr.o amg_zfile_prec_memory_use.o \
 	amg_z_smoothers_bld.o amg_z_hierarchy_bld.o amg_z_hierarchy_rebld.o \
 	amg_zmlprec_aply.o   \
 	$(ZMPFOBJS)  amg_z_extprol_bld.o
 
-CINNEROBJS=   amg_cmlprec_bld.o amg_cfile_prec_descr.o \
+CINNEROBJS=   amg_cmlprec_bld.o amg_cfile_prec_descr.o amg_cfile_prec_memory_use.o \
 	amg_c_smoothers_bld.o amg_c_hierarchy_bld.o amg_c_hierarchy_rebld.o \
 	amg_cmlprec_aply.o   \
 	$(CMPFOBJS)  amg_c_extprol_bld.o

amgprec/impl/aggregator/MatchBoxPC.cpp

@@ -67,12 +67,13 @@ void dMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
 #endif
 
 
-#define TIME_TRACKER
-    #ifdef TIME_TRACKER
-        double tmr = MPI_Wtime();
-    #endif
+#undef TIME_TRACKER
+#ifdef TIME_TRACKER
+  double tmr = MPI_Wtime();
+#endif
 
-#ifdef OPENMP
+#if defined(OPENMP)
+	//fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
         dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(NLVer, NLEdge,
 							   verLocPtr, verLocInd, edgeLocWeight,
 							   verDistance,  Mate,
@@ -91,11 +92,11 @@ void dMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
 #endif
 
 
-  #ifdef TIME_TRACKER
-    tmr = MPI_Wtime() - tmr;
-    fprintf(stderr, "Elaboration time: %f for %ld nodes\n", tmr, NLVer);
-  #endif
-
+#ifdef TIME_TRACKER
+	tmr = MPI_Wtime() - tmr;
+	fprintf(stderr, "Elaboration time: %f for %ld nodes\n", tmr, NLVer);
+#endif
+	
 #endif
 }
 
@@ -113,13 +114,24 @@ void sMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
   fprintf(stderr,"MatchBoxPC: rank %d nlver %ld nledge %ld [ %ld %ld ]\n",
 	  myRank,NLVer, NLEdge,verDistance[0],verDistance[1]);
 #endif
+#if defined(OPENMP)
+	//fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
+  salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(NLVer, NLEdge,
+							      verLocPtr, verLocInd, edgeLocWeight,
+							      verDistance,  Mate,
+							      myRank, numProcs, C_comm,
+							      msgIndSent, msgActualSent, msgPercent,
+							      ph0_time, ph1_time, ph2_time,
+							      ph1_card, ph2_card );
+#else
   salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(NLVer, NLEdge,
-							   verLocPtr, verLocInd, edgeLocWeight,
-							   verDistance,  Mate,
-							   myRank, numProcs, C_comm,
-							   msgIndSent, msgActualSent, msgPercent,
-							   ph0_time, ph1_time, ph2_time,
-							   ph1_card, ph2_card );
+							    verLocPtr, verLocInd, edgeLocWeight,
+							    verDistance,  Mate,
+							    myRank, numProcs, C_comm,
+							    msgIndSent, msgActualSent, msgPercent,
+							    ph0_time, ph1_time, ph2_time,
+							    ph1_card, ph2_card );
+#endif
 #endif
 }
 

amgprec/impl/aggregator/MatchBoxPC.h

@@ -59,7 +59,11 @@
 #include <assert.h>
 #include <map>
 #include <vector>
+#ifdef OPENMP
+// OpenMP is included and used if and only if the OpenMP version of the matching
+// is required
 #include "omp.h"
+#endif
 #include "primitiveDataTypeDefinitions.h"
 #include "dataStrStaticQueue.h"
 
@@ -177,252 +181,416 @@ extern "C"
 #define MilanRealMin MINUS_INFINITY
 #endif
 
-    // Function of find the owner of a ghost vertex using binary search:
-    MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
-                              MilanInt myRank, MilanInt numProcs);
+/* These functions are only used in the experimental OMP implementation, if that
+is disabled there is no reason to actually compile or reference them. */
 
-    MilanLongInt firstComputeCandidateMate(MilanLongInt adj1,
-                                           MilanLongInt adj2,
-                                           MilanLongInt *verLocInd,
-                                           MilanReal *edgeLocWeight);
+  // Function of find the owner of a ghost vertex using binary search:
+  MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
+			    MilanInt myRank, MilanInt numProcs);
+  
+  MilanLongInt firstComputeCandidateMateD(MilanLongInt adj1,
+					 MilanLongInt adj2,
+					 MilanLongInt *verLocInd,
+					 MilanReal *edgeLocWeight);
 
-    void queuesTransfer(vector<MilanLongInt> &U,
-                        vector<MilanLongInt> &privateU,
-                        vector<MilanLongInt> &QLocalVtx,
-                        vector<MilanLongInt> &QGhostVtx,
-                        vector<MilanLongInt> &QMsgType,
-                        vector<MilanInt> &QOwner,
-                        vector<MilanLongInt> &privateQLocalVtx,
-                        vector<MilanLongInt> &privateQGhostVtx,
-                        vector<MilanLongInt> &privateQMsgType,
-                        vector<MilanInt> &privateQOwner);
+  
+  void queuesTransfer(vector<MilanLongInt> &U,
+		      vector<MilanLongInt> &privateU,
+		      vector<MilanLongInt> &QLocalVtx,
+		      vector<MilanLongInt> &QGhostVtx,
+		      vector<MilanLongInt> &QMsgType,
+		      vector<MilanInt> &QOwner,
+		      vector<MilanLongInt> &privateQLocalVtx,
+		      vector<MilanLongInt> &privateQGhostVtx,
+		      vector<MilanLongInt> &privateQMsgType,
+		      vector<MilanInt> &privateQOwner);
+  
+  bool isAlreadyMatched(MilanLongInt node,
+			MilanLongInt StartIndex,
+			MilanLongInt EndIndex,
+			vector<MilanLongInt> &GMate,
+			MilanLongInt *Mate,
+			map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
+  
+  MilanLongInt computeCandidateMateD(MilanLongInt adj1,
+				    MilanLongInt adj2,
+				    MilanReal *edgeLocWeight,
+				    MilanLongInt k,
+				    MilanLongInt *verLocInd,
+				    MilanLongInt StartIndex,
+				    MilanLongInt EndIndex,
+				    vector<MilanLongInt> &GMate,
+				    MilanLongInt *Mate,
+				    map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
+  
+  void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
+		  MilanLongInt StartIndex, MilanLongInt EndIndex,
+		  MilanLongInt *numGhostEdgesPtr,
+		  MilanLongInt *numGhostVerticesPtr,
+		  MilanLongInt *S,
+		  MilanLongInt *verLocInd,
+		  MilanLongInt *verLocPtr,
+		  map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+		  vector<MilanLongInt> &Counter,
+		  vector<MilanLongInt> &verGhostPtr,
+		  vector<MilanLongInt> &verGhostInd,
+		  vector<MilanLongInt> &tempCounter,
+		  vector<MilanLongInt> &GMate,
+		  vector<MilanLongInt> &Message,
+		  vector<MilanLongInt> &QLocalVtx,
+		  vector<MilanLongInt> &QGhostVtx,
+		  vector<MilanLongInt> &QMsgType,
+		  vector<MilanInt> &QOwner,
+		  MilanLongInt *&candidateMate,
+		  vector<MilanLongInt> &U,
+		  vector<MilanLongInt> &privateU,
+		  vector<MilanLongInt> &privateQLocalVtx,
+		  vector<MilanLongInt> &privateQGhostVtx,
+		  vector<MilanLongInt> &privateQMsgType,
+		  vector<MilanInt> &privateQOwner);
+  
+  void clean(MilanLongInt NLVer,
+	     MilanInt myRank,
+	     MilanLongInt MessageIndex,
+	     vector<MPI_Request> &SRequest,
+	     vector<MPI_Status> &SStatus,
+	     MilanInt BufferSize,
+	     MilanLongInt *Buffer,
+	     MilanLongInt msgActual,
+	     MilanLongInt *msgActualSent,
+	     MilanLongInt msgInd,
+	     MilanLongInt *msgIndSent,
+	     MilanLongInt NumMessagesBundled,
+	     MilanReal *msgPercent);
+  
+  void PARALLEL_COMPUTE_CANDIDATE_MATE_BD(MilanLongInt NLVer,
+					 MilanLongInt *verLocPtr,
+					 MilanLongInt *verLocInd,
+					 MilanInt myRank,
+					 MilanReal *edgeLocWeight,
+					 MilanLongInt *candidateMate);
+  
+  void PARALLEL_PROCESS_EXPOSED_VERTEX_BD(MilanLongInt NLVer,
+					 MilanLongInt *candidateMate,
+					 MilanLongInt *verLocInd,
+					 MilanLongInt *verLocPtr,
+					 MilanLongInt StartIndex,
+					 MilanLongInt EndIndex,
+					 MilanLongInt *Mate,
+					 vector<MilanLongInt> &GMate,
+					 map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+					 MilanReal *edgeLocWeight,
+					 MilanLongInt *myCardPtr,
+					 MilanLongInt *msgIndPtr,
+					 MilanLongInt *NumMessagesBundledPtr,
+					 MilanLongInt *SPtr,
+					 MilanLongInt *verDistance,
+					 MilanLongInt *PCounter,
+					 vector<MilanLongInt> &Counter,
+					 MilanInt myRank,
+					 MilanInt numProcs,
+					 vector<MilanLongInt> &U,
+					 vector<MilanLongInt> &privateU,
+					 vector<MilanLongInt> &QLocalVtx,
+					 vector<MilanLongInt> &QGhostVtx,
+					 vector<MilanLongInt> &QMsgType,
+					 vector<MilanInt> &QOwner,
+					 vector<MilanLongInt> &privateQLocalVtx,
+					 vector<MilanLongInt> &privateQGhostVtx,
+					 vector<MilanLongInt> &privateQMsgType,
+					 vector<MilanInt> &privateQOwner);
+  
+  void PROCESS_CROSS_EDGE(MilanLongInt *edge,
+			  MilanLongInt *SPtr);
+  
+  void processMatchedVerticesD(
+			      MilanLongInt NLVer,
+			      vector<MilanLongInt> &UChunkBeingProcessed,
+			      vector<MilanLongInt> &U,
+			      vector<MilanLongInt> &privateU,
+			      MilanLongInt StartIndex,
+			      MilanLongInt EndIndex,
+			      MilanLongInt *myCardPtr,
+			      MilanLongInt *msgIndPtr,
+			      MilanLongInt *NumMessagesBundledPtr,
+			      MilanLongInt *SPtr,
+			      MilanLongInt *verLocPtr,
+			      MilanLongInt *verLocInd,
+			      MilanLongInt *verDistance,
+			      MilanLongInt *PCounter,
+			      vector<MilanLongInt> &Counter,
+			      MilanInt myRank,
+			      MilanInt numProcs,
+			      MilanLongInt *candidateMate,
+			      vector<MilanLongInt> &GMate,
+			      MilanLongInt *Mate,
+			      map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+			      MilanReal *edgeLocWeight,
+			      vector<MilanLongInt> &QLocalVtx,
+			      vector<MilanLongInt> &QGhostVtx,
+			      vector<MilanLongInt> &QMsgType,
+			      vector<MilanInt> &QOwner,
+			      vector<MilanLongInt> &privateQLocalVtx,
+			      vector<MilanLongInt> &privateQGhostVtx,
+			      vector<MilanLongInt> &privateQMsgType,
+			      vector<MilanInt> &privateQOwner);
+  
+  void processMatchedVerticesAndSendMessagesD(
+					     MilanLongInt NLVer,
+					     vector<MilanLongInt> &UChunkBeingProcessed,
+					     vector<MilanLongInt> &U,
+					     vector<MilanLongInt> &privateU,
+					     MilanLongInt StartIndex,
+					     MilanLongInt EndIndex,
+					     MilanLongInt *myCardPtr,
+					     MilanLongInt *msgIndPtr,
+					     MilanLongInt *NumMessagesBundledPtr,
+					     MilanLongInt *SPtr,
+					     MilanLongInt *verLocPtr,
+					     MilanLongInt *verLocInd,
+					     MilanLongInt *verDistance,
+					     MilanLongInt *PCounter,
+					     vector<MilanLongInt> &Counter,
+					     MilanInt myRank,
+					     MilanInt numProcs,
+					     MilanLongInt *candidateMate,
+					     vector<MilanLongInt> &GMate,
+					     MilanLongInt *Mate,
+					     map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+					     MilanReal *edgeLocWeight,
+					     vector<MilanLongInt> &QLocalVtx,
+					     vector<MilanLongInt> &QGhostVtx,
+					     vector<MilanLongInt> &QMsgType,
+					     vector<MilanInt> &QOwner,
+					     vector<MilanLongInt> &privateQLocalVtx,
+					     vector<MilanLongInt> &privateQGhostVtx,
+					     vector<MilanLongInt> &privateQMsgType,
+					     vector<MilanInt> &privateQOwner,
+					     MPI_Comm comm,
+					     MilanLongInt *msgActual,
+					     vector<MilanLongInt> &Message);
+  
+  void sendBundledMessages(MilanLongInt *numGhostEdgesPtr,
+			   MilanInt *BufferSizePtr,
+			   MilanLongInt *Buffer,
+			   vector<MilanLongInt> &PCumulative,
+			   vector<MilanLongInt> &PMessageBundle,
+			   vector<MilanLongInt> &PSizeInfoMessages,
+			   MilanLongInt *PCounter,
+			   MilanLongInt NumMessagesBundled,
+			   MilanLongInt *msgActualPtr,
+			   MilanLongInt *MessageIndexPtr,
+			   MilanInt numProcs,
+			   MilanInt myRank,
+			   MPI_Comm comm,
+			   vector<MilanLongInt> &QLocalVtx,
+			   vector<MilanLongInt> &QGhostVtx,
+			   vector<MilanLongInt> &QMsgType,
+			   vector<MilanInt> &QOwner,
+			   vector<MPI_Request> &SRequest,
+			   vector<MPI_Status> &SStatus);
+  
+  void processMessagesD(
+		       MilanLongInt NLVer,
+		       MilanLongInt *Mate,
+		       MilanLongInt *candidateMate,
+		       map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+		       vector<MilanLongInt> &GMate,
+		       vector<MilanLongInt> &Counter,
+		       MilanLongInt StartIndex,
+		       MilanLongInt EndIndex,
+		       MilanLongInt *myCardPtr,
+		       MilanLongInt *msgIndPtr,
+		       MilanLongInt *msgActualPtr,
+		       MilanReal *edgeLocWeight,
+		       MilanLongInt *verDistance,
+		       MilanLongInt *verLocPtr,
+		       MilanLongInt k,
+		       MilanLongInt *verLocInd,
+		       MilanInt numProcs,
+		       MilanInt myRank,
+		       MPI_Comm comm,
+		       vector<MilanLongInt> &Message,
+		       MilanLongInt numGhostEdges,
+		       MilanLongInt u,
+		       MilanLongInt v,
+		       MilanLongInt *SPtr,
+		       vector<MilanLongInt> &U);
+  
+  void extractUChunk(
+		     vector<MilanLongInt> &UChunkBeingProcessed,
+		     vector<MilanLongInt> &U,
+		     vector<MilanLongInt> &privateU);
 
-    bool isAlreadyMatched(MilanLongInt node,
-                          MilanLongInt StartIndex,
-                          MilanLongInt EndIndex,
-                          vector<MilanLongInt> &GMate,
-                          MilanLongInt *Mate,
-                          map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
+  MilanLongInt firstComputeCandidateMateS(MilanLongInt adj1,
+					 MilanLongInt adj2,
+					 MilanLongInt *verLocInd,
+					  MilanFloat *edgeLocWeight);
 
-    MilanLongInt computeCandidateMate(MilanLongInt adj1,
-                                      MilanLongInt adj2,
-                                      MilanReal *edgeLocWeight,
-                                      MilanLongInt k,
-                                      MilanLongInt *verLocInd,
-                                      MilanLongInt StartIndex,
-                                      MilanLongInt EndIndex,
-                                      vector<MilanLongInt> &GMate,
-                                      MilanLongInt *Mate,
-                                      map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
+  MilanLongInt computeCandidateMateS(MilanLongInt adj1,
+				    MilanLongInt adj2,
+				    MilanFloat *edgeLocWeight,
+				    MilanLongInt k,
+				    MilanLongInt *verLocInd,
+				    MilanLongInt StartIndex,
+				    MilanLongInt EndIndex,
+				    vector<MilanLongInt> &GMate,
+				    MilanLongInt *Mate,
+				    map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
+  
+  void PARALLEL_COMPUTE_CANDIDATE_MATE_BS(MilanLongInt NLVer,
+					 MilanLongInt *verLocPtr,
+					 MilanLongInt *verLocInd,
+					 MilanInt myRank,
+					 MilanFloat *edgeLocWeight,
+					 MilanLongInt *candidateMate);
+  
+  void PARALLEL_PROCESS_EXPOSED_VERTEX_BS(MilanLongInt NLVer,
+					 MilanLongInt *candidateMate,
+					 MilanLongInt *verLocInd,
+					 MilanLongInt *verLocPtr,
+					 MilanLongInt StartIndex,
+					 MilanLongInt EndIndex,
+					 MilanLongInt *Mate,
+					 vector<MilanLongInt> &GMate,
+					 map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+					 MilanFloat *edgeLocWeight,
+					 MilanLongInt *myCardPtr,
+					 MilanLongInt *msgIndPtr,
+					 MilanLongInt *NumMessagesBundledPtr,
+					 MilanLongInt *SPtr,
+					 MilanLongInt *verDistance,
+					 MilanLongInt *PCounter,
+					 vector<MilanLongInt> &Counter,
+					 MilanInt myRank,
+					 MilanInt numProcs,
+					 vector<MilanLongInt> &U,
+					 vector<MilanLongInt> &privateU,
+					 vector<MilanLongInt> &QLocalVtx,
+					 vector<MilanLongInt> &QGhostVtx,
+					 vector<MilanLongInt> &QMsgType,
+					 vector<MilanInt> &QOwner,
+					 vector<MilanLongInt> &privateQLocalVtx,
+					 vector<MilanLongInt> &privateQGhostVtx,
+					 vector<MilanLongInt> &privateQMsgType,
+					 vector<MilanInt> &privateQOwner);
+  void processMatchedVerticesS(
+			      MilanLongInt NLVer,
+			      vector<MilanLongInt> &UChunkBeingProcessed,
+			      vector<MilanLongInt> &U,
+			      vector<MilanLongInt> &privateU,
+			      MilanLongInt StartIndex,
+			      MilanLongInt EndIndex,
+			      MilanLongInt *myCardPtr,
+			      MilanLongInt *msgIndPtr,
+			      MilanLongInt *NumMessagesBundledPtr,
+			      MilanLongInt *SPtr,
+			      MilanLongInt *verLocPtr,
+			      MilanLongInt *verLocInd,
+			      MilanLongInt *verDistance,
+			      MilanLongInt *PCounter,
+			      vector<MilanLongInt> &Counter,
+			      MilanInt myRank,
+			      MilanInt numProcs,
+			      MilanLongInt *candidateMate,
+			      vector<MilanLongInt> &GMate,
+			      MilanLongInt *Mate,
+			      map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+			      MilanFloat *edgeLocWeight,
+			      vector<MilanLongInt> &QLocalVtx,
+			      vector<MilanLongInt> &QGhostVtx,
+			      vector<MilanLongInt> &QMsgType,
+			      vector<MilanInt> &QOwner,
+			      vector<MilanLongInt> &privateQLocalVtx,
+			      vector<MilanLongInt> &privateQGhostVtx,
+			      vector<MilanLongInt> &privateQMsgType,
+			      vector<MilanInt> &privateQOwner);
+  
+  void processMatchedVerticesAndSendMessagesS(
+					     MilanLongInt NLVer,
+					     vector<MilanLongInt> &UChunkBeingProcessed,
+					     vector<MilanLongInt> &U,
+					     vector<MilanLongInt> &privateU,
+					     MilanLongInt StartIndex,
+					     MilanLongInt EndIndex,
+					     MilanLongInt *myCardPtr,
+					     MilanLongInt *msgIndPtr,
+					     MilanLongInt *NumMessagesBundledPtr,
+					     MilanLongInt *SPtr,
+					     MilanLongInt *verLocPtr,
+					     MilanLongInt *verLocInd,
+					     MilanLongInt *verDistance,
+					     MilanLongInt *PCounter,
+					     vector<MilanLongInt> &Counter,
+					     MilanInt myRank,
+					     MilanInt numProcs,
+					     MilanLongInt *candidateMate,
+					     vector<MilanLongInt> &GMate,
+					     MilanLongInt *Mate,
+					     map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+					     MilanFloat *edgeLocWeight,
+					     vector<MilanLongInt> &QLocalVtx,
+					     vector<MilanLongInt> &QGhostVtx,
+					     vector<MilanLongInt> &QMsgType,
+					     vector<MilanInt> &QOwner,
+					     vector<MilanLongInt> &privateQLocalVtx,
+					     vector<MilanLongInt> &privateQGhostVtx,
+					     vector<MilanLongInt> &privateQMsgType,
+					     vector<MilanInt> &privateQOwner,
+					     MPI_Comm comm,
+					     MilanLongInt *msgActual,
+					     vector<MilanLongInt> &Message);
+  
+  void processMessagesS(
+		       MilanLongInt NLVer,
+		       MilanLongInt *Mate,
+		       MilanLongInt *candidateMate,
+		       map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+		       vector<MilanLongInt> &GMate,
+		       vector<MilanLongInt> &Counter,
+		       MilanLongInt StartIndex,
+		       MilanLongInt EndIndex,
+		       MilanLongInt *myCardPtr,
+		       MilanLongInt *msgIndPtr,
+		       MilanLongInt *msgActualPtr,
+		       MilanFloat *edgeLocWeight,
+		       MilanLongInt *verDistance,
+		       MilanLongInt *verLocPtr,
+		       MilanLongInt k,
+		       MilanLongInt *verLocInd,
+		       MilanInt numProcs,
+		       MilanInt myRank,
+		       MPI_Comm comm,
+		       vector<MilanLongInt> &Message,
+		       MilanLongInt numGhostEdges,
+		       MilanLongInt u,
+		       MilanLongInt v,
+		       MilanLongInt *SPtr,
+		       vector<MilanLongInt> &U);
+  
+  
+  void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
+			 MilanLongInt NLVer, MilanLongInt NLEdge,
+			 MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanReal *edgeLocWeight,
+			 MilanLongInt *verDistance,
+			 MilanLongInt *Mate,
+			 MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
+			 MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
+			 MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
+			 MilanLongInt *ph1_card, MilanLongInt *ph2_card);
 
-    void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
-                    MilanLongInt StartIndex, MilanLongInt EndIndex,
-                    MilanLongInt *numGhostEdgesPtr,
-                    MilanLongInt *numGhostVerticesPtr,
-                    MilanLongInt *S,
-                    MilanLongInt *verLocInd,
-                    MilanLongInt *verLocPtr,
-                    map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
-                    vector<MilanLongInt> &Counter,
-                    vector<MilanLongInt> &verGhostPtr,
-                    vector<MilanLongInt> &verGhostInd,
-                    vector<MilanLongInt> &tempCounter,
-                    vector<MilanLongInt> &GMate,
-                    vector<MilanLongInt> &Message,
-                    vector<MilanLongInt> &QLocalVtx,
-                    vector<MilanLongInt> &QGhostVtx,
-                    vector<MilanLongInt> &QMsgType,
-                    vector<MilanInt> &QOwner,
-                    MilanLongInt *&candidateMate,
-                    vector<MilanLongInt> &U,
-                    vector<MilanLongInt> &privateU,
-                    vector<MilanLongInt> &privateQLocalVtx,
-                    vector<MilanLongInt> &privateQGhostVtx,
-                    vector<MilanLongInt> &privateQMsgType,
-                    vector<MilanInt> &privateQOwner);
+  
+  void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
+			 MilanLongInt NLVer, MilanLongInt NLEdge,
+			 MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanFloat *edgeLocWeight,
+			 MilanLongInt *verDistance,
+			 MilanLongInt *Mate,
+			 MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
+			 MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
+			 MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
+			 MilanLongInt *ph1_card, MilanLongInt *ph2_card);
 
-    void clean(MilanLongInt NLVer,
-               MilanInt myRank,
-               MilanLongInt MessageIndex,
-               vector<MPI_Request> &SRequest,
-               vector<MPI_Status> &SStatus,
-               MilanInt BufferSize,
-               MilanLongInt *Buffer,
-               MilanLongInt msgActual,
-               MilanLongInt *msgActualSent,
-               MilanLongInt msgInd,
-               MilanLongInt *msgIndSent,
-               MilanLongInt NumMessagesBundled,
-               MilanReal *msgPercent);
 
-    void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
-                                           MilanLongInt *verLocPtr,
-                                           MilanLongInt *verLocInd,
-                                           MilanInt myRank,
-                                           MilanReal *edgeLocWeight,
-                                           MilanLongInt *candidateMate);
-
-    void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
-                                           MilanLongInt *candidateMate,
-                                           MilanLongInt *verLocInd,
-                                           MilanLongInt *verLocPtr,
-                                           MilanLongInt StartIndex,
-                                           MilanLongInt EndIndex,
-                                           MilanLongInt *Mate,
-                                           vector<MilanLongInt> &GMate,
-                                           map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
-                                           MilanReal *edgeLocWeight,
-                                           MilanLongInt *myCardPtr,
-                                           MilanLongInt *msgIndPtr,
-                                           MilanLongInt *NumMessagesBundledPtr,
-                                           MilanLongInt *SPtr,
-                                           MilanLongInt *verDistance,
-                                           MilanLongInt *PCounter,
-                                           vector<MilanLongInt> &Counter,
-                                           MilanInt myRank,
-                                           MilanInt numProcs,
-                                           vector<MilanLongInt> &U,
-                                           vector<MilanLongInt> &privateU,
-                                           vector<MilanLongInt> &QLocalVtx,
-                                           vector<MilanLongInt> &QGhostVtx,
-                                           vector<MilanLongInt> &QMsgType,
-                                           vector<MilanInt> &QOwner,
-                                           vector<MilanLongInt> &privateQLocalVtx,
-                                           vector<MilanLongInt> &privateQGhostVtx,
-                                           vector<MilanLongInt> &privateQMsgType,
-                                           vector<MilanInt> &privateQOwner);
-
-    void PROCESS_CROSS_EDGE(MilanLongInt *edge,
-                            MilanLongInt *SPtr);
-
-    void processMatchedVertices(
-        MilanLongInt NLVer,
-        vector<MilanLongInt> &UChunkBeingProcessed,
-        vector<MilanLongInt> &U,
-        vector<MilanLongInt> &privateU,
-        MilanLongInt StartIndex,
-        MilanLongInt EndIndex,
-        MilanLongInt *myCardPtr,
-        MilanLongInt *msgIndPtr,
-        MilanLongInt *NumMessagesBundledPtr,
-        MilanLongInt *SPtr,
-        MilanLongInt *verLocPtr,
-        MilanLongInt *verLocInd,
-        MilanLongInt *verDistance,
-        MilanLongInt *PCounter,
-        vector<MilanLongInt> &Counter,
-        MilanInt myRank,
-        MilanInt numProcs,
-        MilanLongInt *candidateMate,
-        vector<MilanLongInt> &GMate,
-        MilanLongInt *Mate,
-        map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
-        MilanReal *edgeLocWeight,
-        vector<MilanLongInt> &QLocalVtx,
-        vector<MilanLongInt> &QGhostVtx,
-        vector<MilanLongInt> &QMsgType,
-        vector<MilanInt> &QOwner,
-        vector<MilanLongInt> &privateQLocalVtx,
-        vector<MilanLongInt> &privateQGhostVtx,
-        vector<MilanLongInt> &privateQMsgType,
-        vector<MilanInt> &privateQOwner);
-
-    void processMatchedVerticesAndSendMessages(
-        MilanLongInt NLVer,
-        vector<MilanLongInt> &UChunkBeingProcessed,
-        vector<MilanLongInt> &U,
-        vector<MilanLongInt> &privateU,
-        MilanLongInt StartIndex,
-        MilanLongInt EndIndex,
-        MilanLongInt *myCardPtr,
-        MilanLongInt *msgIndPtr,
-        MilanLongInt *NumMessagesBundledPtr,
-        MilanLongInt *SPtr,
-        MilanLongInt *verLocPtr,
-        MilanLongInt *verLocInd,
-        MilanLongInt *verDistance,
-        MilanLongInt *PCounter,
-        vector<MilanLongInt> &Counter,
-        MilanInt myRank,
-        MilanInt numProcs,
-        MilanLongInt *candidateMate,
-        vector<MilanLongInt> &GMate,
-        MilanLongInt *Mate,
-        map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
-        MilanReal *edgeLocWeight,
-        vector<MilanLongInt> &QLocalVtx,
-        vector<MilanLongInt> &QGhostVtx,
-        vector<MilanLongInt> &QMsgType,
-        vector<MilanInt> &QOwner,
-        vector<MilanLongInt> &privateQLocalVtx,
-        vector<MilanLongInt> &privateQGhostVtx,
-        vector<MilanLongInt> &privateQMsgType,
-        vector<MilanInt> &privateQOwner,
-        MPI_Comm comm,
-        MilanLongInt *msgActual,
-        vector<MilanLongInt> &Message);
-
-    void sendBundledMessages(MilanLongInt *numGhostEdgesPtr,
-                             MilanInt *BufferSizePtr,
-                             MilanLongInt *Buffer,
-                             vector<MilanLongInt> &PCumulative,
-                             vector<MilanLongInt> &PMessageBundle,
-                             vector<MilanLongInt> &PSizeInfoMessages,
-                             MilanLongInt *PCounter,
-                             MilanLongInt NumMessagesBundled,
-                             MilanLongInt *msgActualPtr,
-                             MilanLongInt *MessageIndexPtr,
-                             MilanInt numProcs,
-                             MilanInt myRank,
-                             MPI_Comm comm,
-                             vector<MilanLongInt> &QLocalVtx,
-                             vector<MilanLongInt> &QGhostVtx,
-                             vector<MilanLongInt> &QMsgType,
-                             vector<MilanInt> &QOwner,
-                             vector<MPI_Request> &SRequest,
-                             vector<MPI_Status> &SStatus);
-
-    void processMessages(
-        MilanLongInt NLVer,
-        MilanLongInt *Mate,
-        MilanLongInt *candidateMate,
-        map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
-        vector<MilanLongInt> &GMate,
-        vector<MilanLongInt> &Counter,
-        MilanLongInt StartIndex,
-        MilanLongInt EndIndex,
-        MilanLongInt *myCardPtr,
-        MilanLongInt *msgIndPtr,
-        MilanLongInt *msgActualPtr,
-        MilanReal *edgeLocWeight,
-        MilanLongInt *verDistance,
-        MilanLongInt *verLocPtr,
-        MilanLongInt k,
-        MilanLongInt *verLocInd,
-        MilanInt numProcs,
-        MilanInt myRank,
-        MPI_Comm comm,
-        vector<MilanLongInt> &Message,
-        MilanLongInt numGhostEdges,
-        MilanLongInt u,
-        MilanLongInt v,
-        MilanLongInt *SPtr,
-        vector<MilanLongInt> &U);
-
-    void extractUChunk(
-        vector<MilanLongInt> &UChunkBeingProcessed,
-        vector<MilanLongInt> &U,
-        vector<MilanLongInt> &privateU);
-
-    void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
-        MilanLongInt NLVer, MilanLongInt NLEdge,
-        MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanReal *edgeLocWeight,
-        MilanLongInt *verDistance,
-        MilanLongInt *Mate,
-        MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
-        MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
-        MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
-        MilanLongInt *ph1_card, MilanLongInt *ph2_card);
 
     void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
         MilanLongInt NLVer, MilanLongInt NLEdge,

amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC.cpp

@@ -1303,16 +1303,16 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
 // SINGLE PRECISION VERSION
 
 void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
-                                                              MilanLongInt NLVer, MilanLongInt NLEdge,
-                                                              MilanLongInt* verLocPtr, MilanLongInt* verLocInd,
-							      MilanFloat* edgeLocWeight,
-                                                              MilanLongInt* verDistance,
-                                                              MilanLongInt* Mate,
-                                                              MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
-                                                              MilanLongInt* msgIndSent, MilanLongInt* msgActualSent,
-							      MilanReal* msgPercent,
-                                                              MilanReal* ph0_time, MilanReal* ph1_time, MilanReal* ph2_time,
-                                                              MilanLongInt* ph1_card, MilanLongInt* ph2_card ) {
+      MilanLongInt NLVer, MilanLongInt NLEdge,
+      MilanLongInt* verLocPtr, MilanLongInt* verLocInd,
+      MilanFloat* edgeLocWeight,
+      MilanLongInt* verDistance,
+      MilanLongInt* Mate,
+      MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
+      MilanLongInt* msgIndSent, MilanLongInt* msgActualSent,
+      MilanReal* msgPercent,
+      MilanReal* ph0_time, MilanReal* ph1_time, MilanReal* ph2_time,
+      MilanLongInt* ph1_card, MilanLongInt* ph2_card ) {
 #if !defined(SERIAL_MPI)
 #ifdef PRINT_DEBUG_INFO_
     cout<<"\n("<<myRank<<")Within algoEdgeApproxDominatingEdgesLinearSearchMessageBundling()"; fflush(stdout);

amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-#ifdef OPENMP
 // ***********************************************************************
 //
 //        MatchboxP: A C++ library for approximate weighted matching
@@ -244,7 +243,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
      * PARALLEL_COMPUTE_CANDIDATE_MATE_B is now totally parallel.
      */
 
-    PARALLEL_COMPUTE_CANDIDATE_MATE_B(NLVer,
+    PARALLEL_COMPUTE_CANDIDATE_MATE_BD(NLVer,
                                       verLocPtr,
                                       verLocInd,
                                       myRank,
@@ -269,7 +268,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
      * TODO: Test when it's actually more efficient to execute this code
      *       in parallel.
      */
-    PARALLEL_PROCESS_EXPOSED_VERTEX_B(NLVer,
+    PARALLEL_PROCESS_EXPOSED_VERTEX_BD(NLVer,
                                       candidateMate,
                                       verLocInd,
                                       verLocPtr,
@@ -321,7 +320,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
     vector<MilanLongInt> UChunkBeingProcessed;
     UChunkBeingProcessed.reserve(UCHUNK);
 
-    processMatchedVertices(NLVer,
+    processMatchedVerticesD(NLVer,
                            UChunkBeingProcessed,
                            U,
                            privateU,
@@ -430,7 +429,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
         /////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
         ///////////////////////////////////////////////////////////////////////////////////
 
-      processMatchedVerticesAndSendMessages(NLVer,
+      processMatchedVerticesAndSendMessagesD(NLVer,
 					    UChunkBeingProcessed,
 					    U,
 					    privateU,
@@ -490,8 +489,8 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
         ///////////////////////////////////////////////////////////////////////////////////
         /////////////////////////// PROCESS MESSAGES //////////////////////////////////////
         ///////////////////////////////////////////////////////////////////////////////////
-
-        processMessages(NLVer,
+	//startTime = MPI_Wtime();
+        processMessagesD(NLVer,
                         Mate,
                         candidateMate,
                         Ghost2LocalMap,
@@ -556,6 +555,488 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
     *ph2_card = myCard;                 // Cardinality at the end of Phase-2
 }
 // End of algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMate
+
+void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
+    MilanLongInt NLVer, MilanLongInt NLEdge,
+    MilanLongInt *verLocPtr, MilanLongInt *verLocInd,
+    MilanFloat *edgeLocWeight,
+    MilanLongInt *verDistance,
+    MilanLongInt *Mate,
+    MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
+    MilanLongInt *msgIndSent, MilanLongInt *msgActualSent,
+    MilanReal *msgPercent,
+    MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
+    MilanLongInt *ph1_card, MilanLongInt *ph2_card)
+{
+
+    /*
+     * verDistance: it's a vector long as the number of processors.
+     *              verDistance[i] contains the first node index of the i-th processor
+     *              verDistance[i + 1] contains the last node index of the i-th processor
+     * NLVer: number of elements in the LocPtr
+     * NLEdge: number of edges assigned to the current processor
+     *
+     * Contains the portion of matrix assigned to the processor in
+     * Yale notation
+     * verLocInd: contains the positions on row of the matrix
+     * verLocPtr: i-th value is the position of the first element on the i-th row and
+     *            i+1-th value is the position of the first element on the i+1-th row
+     */
+
+#if !defined(SERIAL_MPI)
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")Within algoEdgeApproxDominatingEdgesLinearSearchMessageBundling()";
+    fflush(stdout);
+#endif
+
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ") verDistance [" ;
+    for (int i = 0; i < numProcs; i++)
+      cout << verDistance[i] << "," << verDistance[i+1];
+    cout  << "]\n";
+    fflush(stdout);
+#endif
+#ifdef DEBUG_HANG_
+    if (myRank == 0) {
+      cout << "\n(" << myRank << ") verDistance [" ;
+      for (int i = 0; i < numProcs; i++)
+	cout << verDistance[i] << "," ;
+      cout  << verDistance[numProcs]<< "]\n";
+    }
+    fflush(stdout);
+#endif
+
+    // The starting vertex owned by the current rank
+    MilanLongInt StartIndex = verDistance[myRank];
+    // The ending vertex owned by the current rank
+    MilanLongInt EndIndex   = verDistance[myRank + 1] - 1;
+
+    MPI_Status computeStatus;
+
+    MilanLongInt msgActual = 0, msgInd = 0;
+    MilanFloat heaviestEdgeWt = 0.0f; // Assumes positive weight
+    MilanReal startTime, finishTime;
+
+    startTime = MPI_Wtime();
+
+    // Data structures for sending and receiving messages:
+    vector<MilanLongInt> Message; // [ u, v, message_type ]
+    Message.resize(3, -1);
+    // Data structures for Message Bundling:
+    // Although up to two messages can be sent along any cross edge,
+    // only one message will be sent in the initialization phase -
+    // one of: REQUEST/FAILURE/SUCCESS
+    vector<MilanLongInt> QLocalVtx, QGhostVtx, QMsgType;
+    // Changed by Fabio to be an integer, addresses needs to be integers!
+    vector<MilanInt> QOwner; 
+
+    MilanLongInt *PCounter = new MilanLongInt[numProcs];
+    for (int i = 0; i < numProcs; i++)
+        PCounter[i] = 0;
+
+    MilanLongInt NumMessagesBundled = 0;
+    // TODO when the last computational section will be refactored this could be eliminated
+    // Changed by Fabio to be an integer, addresses needs to be integers!
+    MilanInt ghostOwner = 0; 
+    MilanLongInt *candidateMate = nullptr;
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")NV: " << NLVer << "  Edges: " << NLEdge;
+    fflush(stdout);
+    cout << "\n(" << myRank << ")StartIndex: " << StartIndex << "  EndIndex: " << EndIndex;
+    fflush(stdout);
+#endif
+    // Other Variables:
+    MilanLongInt u = -1, v = -1, w = -1, i = 0;
+    MilanLongInt k = -1, adj1 = -1, adj2 = -1;
+    MilanLongInt k1 = -1, adj11 = -1, adj12 = -1;
+    MilanLongInt myCard = 0;
+
+    // Build the Ghost Vertex Set: Vg
+    // Map each ghost vertex to a local vertex
+    map<MilanLongInt, MilanLongInt> Ghost2LocalMap;
+    // Store the edge count for each ghost vertex
+    vector<MilanLongInt> Counter;
+    // Number of Ghost vertices
+    MilanLongInt numGhostVertices = 0, numGhostEdges = 0; 
+
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")About to compute Ghost Vertices...";
+    fflush(stdout);
+#endif
+#ifdef DEBUG_HANG_
+    if (myRank == 0)
+        cout << "\n(" << myRank << ")About to compute Ghost Vertices...";
+    fflush(stdout);
+#endif
+
+    // Define Adjacency Lists for Ghost Vertices:
+    // cout<<"Building Ghost data structures ... \n\n";
+    vector<MilanLongInt> verGhostPtr, verGhostInd, tempCounter;
+    // Mate array for ghost vertices:
+    vector<MilanLongInt> GMate; // Proportional to the number of ghost vertices
+    MilanLongInt S;
+    MilanLongInt privateMyCard = 0;
+    vector<MilanLongInt> PCumulative, PMessageBundle, PSizeInfoMessages;
+    vector<MPI_Request> SRequest;  // Requests that are used for each send message
+    vector<MPI_Status> SStatus;    // Status of sent messages, used in MPI_Wait
+    MilanLongInt MessageIndex = 0; // Pointer for current message
+    MilanInt BufferSize;
+    MilanLongInt *Buffer;
+
+    vector<MilanLongInt> privateQLocalVtx, privateQGhostVtx, privateQMsgType;
+    vector<MilanInt> privateQOwner;
+    vector<MilanLongInt> U, privateU;
+
+    
+    initialize(NLVer, NLEdge, StartIndex,
+               EndIndex, &numGhostEdges,
+               &numGhostVertices, &S,
+               verLocInd, verLocPtr,
+               Ghost2LocalMap, Counter,
+               verGhostPtr, verGhostInd,
+               tempCounter, GMate,
+               Message, QLocalVtx,
+               QGhostVtx, QMsgType, QOwner,
+               candidateMate, U,
+               privateU,
+               privateQLocalVtx,
+               privateQGhostVtx,
+               privateQMsgType,
+               privateQOwner);
+
+    finishTime = MPI_Wtime();
+    *ph0_time = finishTime - startTime; // Time taken for Phase-0: Initialization
+#ifdef DEBUG_HANG_
+    cout << myRank << " Finished initialization" << endl;
+    fflush(stdout);
+#endif
+
+    startTime = MPI_Wtime();
+
+    /////////////////////////////////////////////////////////////////////////////////////////
+    //////////////////////////////////// INITIALIZATION /////////////////////////////////////
+    /////////////////////////////////////////////////////////////////////////////////////////
+    // 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_BS(NLVer,
+                                      verLocPtr,
+                                      verLocInd,
+                                      myRank,
+                                      edgeLocWeight,
+                                      candidateMate);
+
+#ifdef DEBUG_HANG_
+    cout << myRank << " Finished Exposed Vertex" << endl;
+    fflush(stdout);
+#if 0
+    cout << myRank << " candidateMate after parallelCompute " <<endl;
+    for (int i=0; i<NLVer; i++) {
+      cout << candidateMate[i] << " " ;
+    }
+    cout << endl;
+#endif
+#endif
+    /*
+     * PARALLEL_PROCESS_EXPOSED_VERTEX_B
+     * TODO: write comment
+     *
+     * TODO: Test when it's actually more efficient to execute this code
+     *       in parallel.
+     */
+    PARALLEL_PROCESS_EXPOSED_VERTEX_BS(NLVer,
+                                      candidateMate,
+                                      verLocInd,
+                                      verLocPtr,
+                                      StartIndex,
+                                      EndIndex,
+                                      Mate,
+                                      GMate,
+                                      Ghost2LocalMap,
+                                      edgeLocWeight,
+                                      &myCard,
+                                      &msgInd,
+                                      &NumMessagesBundled,
+                                      &S,
+                                      verDistance,
+                                      PCounter,
+                                      Counter,
+                                      myRank,
+                                      numProcs,
+                                      U,
+                                      privateU,
+                                      QLocalVtx,
+                                      QGhostVtx,
+                                      QMsgType,
+                                      QOwner,
+                                      privateQLocalVtx,
+                                      privateQGhostVtx,
+                                      privateQMsgType,
+                                      privateQOwner);
+
+    tempCounter.clear(); // Do not need this any more
+
+#ifdef DEBUG_HANG_
+    cout << myRank << " Finished Exposed Vertex" << endl;
+    fflush(stdout);
+#if 0
+    cout << myRank << " Mate after Exposed Vertices " <<endl;
+    for (int i=0; i<NLVer; i++) {
+      cout << Mate[i] << " " ;
+    }
+    cout << endl;
+#endif
+#endif
+
+    ///////////////////////////////////////////////////////////////////////////////////
+    /////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
+    ///////////////////////////////////////////////////////////////////////////////////
+
+    // TODO what would be the optimal UCHUNK
+    vector<MilanLongInt> UChunkBeingProcessed;
+    UChunkBeingProcessed.reserve(UCHUNK);
+
+    processMatchedVerticesS(NLVer,
+                           UChunkBeingProcessed,
+                           U,
+                           privateU,
+                           StartIndex,
+                           EndIndex,
+                           &myCard,
+                           &msgInd,
+                           &NumMessagesBundled,
+                           &S,
+                           verLocPtr,
+                           verLocInd,
+                           verDistance,
+                           PCounter,
+                           Counter,
+                           myRank,
+                           numProcs,
+                           candidateMate,
+                           GMate,
+                           Mate,
+                           Ghost2LocalMap,
+                           edgeLocWeight,
+                           QLocalVtx,
+                           QGhostVtx,
+                           QMsgType,
+                           QOwner,
+                           privateQLocalVtx,
+                           privateQGhostVtx,
+                           privateQMsgType,
+                           privateQOwner);
+
+
+#ifdef DEBUG_HANG_
+    cout << myRank << " Finished Process Vertices" << endl;
+    fflush(stdout);
+#if 0
+    cout << myRank << " Mate after Matched Vertices " <<endl;
+    for (int i=0; i<NLVer; i++) {
+      cout << Mate[i] << " " ;
+    }
+    cout << endl;
+#endif
+#endif
+
+    /////////////////////////////////////////////////////////////////////////////////////////
+    ///////////////////////////// SEND BUNDLED MESSAGES /////////////////////////////////////
+    /////////////////////////////////////////////////////////////////////////////////////////
+
+    sendBundledMessages(&numGhostEdges,
+                        &BufferSize,
+                        Buffer,
+                        PCumulative,
+                        PMessageBundle,
+                        PSizeInfoMessages,
+                        PCounter,
+                        NumMessagesBundled,
+                        &msgActual,
+                        &MessageIndex,
+                        numProcs,
+                        myRank,
+                        comm,
+                        QLocalVtx,
+                        QGhostVtx,
+                        QMsgType,
+                        QOwner,
+                        SRequest,
+                        SStatus);
+
+    ///////////////////////// END OF SEND BUNDLED MESSAGES //////////////////////////////////
+
+    finishTime = MPI_Wtime();
+    *ph1_time = finishTime - startTime; // Time taken for Phase-1
+
+#ifdef DEBUG_HANG_
+    cout << myRank << " Finished sendBundles" << endl;
+    fflush(stdout);
+#endif
+
+    *ph1_card = myCard;                 // Cardinality at the end of Phase-1
+    startTime = MPI_Wtime();
+    /////////////////////////////////////////////////////////////////////////////////////////
+    //////////////////////////////////////// MAIN LOOP //////////////////////////////////////
+    /////////////////////////////////////////////////////////////////////////////////////////
+    // Main While Loop:
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")Entering While(true) loop..";
+    fflush(stdout);
+#endif
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+
+    while (true)  {
+#ifdef DEBUG_HANG_
+      //if (myRank == 0)
+      cout << "\n(" << myRank << ") Main loop" << endl;
+      fflush(stdout);
+#endif
+        ///////////////////////////////////////////////////////////////////////////////////
+        /////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
+        ///////////////////////////////////////////////////////////////////////////////////
+
+      processMatchedVerticesAndSendMessagesS(NLVer,
+					    UChunkBeingProcessed,
+					    U,
+					    privateU,
+					    StartIndex,
+					    EndIndex,
+					    &myCard,
+					    &msgInd,
+					    &NumMessagesBundled,
+					    &S,
+					    verLocPtr,
+					    verLocInd,
+					    verDistance,
+					    PCounter,
+					    Counter,
+					    myRank,
+					    numProcs,
+					    candidateMate,
+					    GMate,
+					    Mate,
+					    Ghost2LocalMap,
+					    edgeLocWeight,
+					    QLocalVtx,
+					    QGhostVtx,
+					    QMsgType,
+					    QOwner,
+					    privateQLocalVtx,
+					    privateQGhostVtx,
+					    privateQMsgType,
+					    privateQOwner,
+					    comm,
+					    &msgActual,
+					    Message);
+
+      ///////////////////////// END OF PROCESS MATCHED VERTICES /////////////////////////
+
+        //// BREAK IF NO MESSAGES EXPECTED /////////
+#ifdef DEBUG_HANG_
+#if 0
+      cout << myRank << " Mate after ProcessMatchedAndSend phase "<<S <<endl;
+      for (int i=0; i<NLVer; i++) {
+	cout << Mate[i] << " " ;
+      }
+      cout << endl;
+#endif
+#endif
+#ifdef PRINT_DEBUG_INFO_
+        cout << "\n(" << myRank << ")Deciding whether to break: S= " << S << endl;
+#endif
+
+        if (S == 0) {
+#ifdef DEBUG_HANG_
+	  cout << "\n(" << myRank << ") Breaking out" << endl;
+	  fflush(stdout);
+#endif
+	  break;
+        }
+        ///////////////////////////////////////////////////////////////////////////////////
+        /////////////////////////// PROCESS MESSAGES //////////////////////////////////////
+        ///////////////////////////////////////////////////////////////////////////////////
+
+        processMessagesS(NLVer,
+                        Mate,
+                        candidateMate,
+                        Ghost2LocalMap,
+                        GMate,
+                        Counter,
+                        StartIndex,
+                        EndIndex,
+                        &myCard,
+                        &msgInd,
+                        &msgActual,
+                        edgeLocWeight,
+                        verDistance,
+                        verLocPtr,
+                        k,
+                        verLocInd,
+                        numProcs,
+                        myRank,
+                        comm,
+                        Message,
+                        numGhostEdges,
+                        u,
+                        v,
+                        &S,
+                        U);
+
+        ///////////////////////// END OF PROCESS MESSAGES /////////////////////////////////
+#ifdef DEBUG_HANG_
+#if 0
+      cout << myRank << " Mate after ProcessMessages phase "<<S <<endl;
+      for (int i=0; i<NLVer; i++) {
+	cout << Mate[i] << " " ;
+      }
+      cout << endl;
+#endif
+#endif
+#ifdef PRINT_DEBUG_INFO_
+        cout << "\n(" << myRank << ")Finished Message processing phase: S= " << S;
+        fflush(stdout);
+        cout << "\n(" << myRank << ")** SENT     : ACTUAL= " << msgActual;
+        fflush(stdout);
+        cout << "\n(" << myRank << ")** SENT     : INDIVIDUAL= " << msgInd << endl;
+        fflush(stdout);
+#endif
+    } // End of while (true)
+
+    clean(NLVer,
+          myRank,
+          MessageIndex,
+          SRequest,
+          SStatus,
+          BufferSize,
+          Buffer,
+          msgActual,
+          msgActualSent,
+          msgInd,
+          msgIndSent,
+          NumMessagesBundled,
+          msgPercent);
+
+    finishTime = MPI_Wtime();
+    *ph2_time = finishTime - startTime; // Time taken for Phase-2
+    *ph2_card = myCard;                 // Cardinality at the end of Phase-2
+}
+
 #endif
 
 #endif

amgprec/impl/aggregator/amg_c_soc1_map_bld.F90

@@ -275,7 +275,8 @@ subroutine amg_c_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             disjoint = all(ilaggr(icol(1:ip)) == -(nr+1)).or.(ip==0)
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_c_soc2_map_bld.F90

@@ -309,7 +309,8 @@ subroutine amg_c_soc2_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             !
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_d_soc1_map_bld.F90

@@ -275,7 +275,8 @@ subroutine amg_d_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             disjoint = all(ilaggr(icol(1:ip)) == -(nr+1)).or.(ip==0)
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_d_soc2_map_bld.F90

@@ -309,7 +309,8 @@ subroutine amg_d_soc2_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             !
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_s_soc1_map_bld.F90

@@ -275,7 +275,8 @@ subroutine amg_s_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             disjoint = all(ilaggr(icol(1:ip)) == -(nr+1)).or.(ip==0)
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_s_soc2_map_bld.F90

@@ -309,7 +309,8 @@ subroutine amg_s_soc2_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             !
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_z_soc1_map_bld.F90

@@ -275,7 +275,8 @@ subroutine amg_z_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             disjoint = all(ilaggr(icol(1:ip)) == -(nr+1)).or.(ip==0)
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/amg_z_soc2_map_bld.F90

@@ -309,7 +309,8 @@ subroutine amg_z_soc2_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
             !
             if (disjoint) then
               locnaggr(kk)     = locnaggr(kk) + 1
-              itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
+              itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
+              itmp = itmp*nths+kk
               if (itmp < (bnds(kk)-1+locnaggr(kk))) then
                 !$omp atomic update
                 info = max(12345678,info)

amgprec/impl/aggregator/clean.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 // TODO comment
 #if !defined(SERIAL_MPI)
 

amgprec/impl/aggregator/computeCandidateMate.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 /**
  * Execute the research fr the Candidate Mate without controlling if the vertices are already matched.
  * Returns the vertices with the highest weight
@@ -9,9 +8,8 @@
  * @param edgeLocWeight
  * @return
  */
-#if !defined(SERIAL_MPI)
 
-MilanLongInt firstComputeCandidateMate(MilanLongInt adj1,
+MilanLongInt firstComputeCandidateMateD(MilanLongInt adj1,
                                        MilanLongInt adj2,
                                        MilanLongInt *verLocInd,
                                        MilanReal *edgeLocWeight)
@@ -44,7 +42,7 @@ MilanLongInt firstComputeCandidateMate(MilanLongInt adj1,
  * @param Ghost2LocalMap
  * @return
  */
-MilanLongInt computeCandidateMate(MilanLongInt adj1,
+MilanLongInt computeCandidateMateD(MilanLongInt adj1,
                                   MilanLongInt adj2,
                                   MilanReal *edgeLocWeight,
                                   MilanLongInt k,
@@ -62,7 +60,7 @@ MilanLongInt computeCandidateMate(MilanLongInt adj1,
     for (k = adj1; k < adj2; k++)   {
       if (isAlreadyMatched(verLocInd[k], StartIndex, EndIndex, GMate, Mate, Ghost2LocalMap))
 	continue;
-      
+
       if ((edgeLocWeight[k] > heaviestEdgeWt) ||
 	  ((edgeLocWeight[k] == heaviestEdgeWt) && (w < verLocInd[k])))     {
 	heaviestEdgeWt = edgeLocWeight[k];
@@ -70,7 +68,71 @@ MilanLongInt computeCandidateMate(MilanLongInt adj1,
       }
     } // End of for loop
       //  End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
-    
+
+    return w;
+}
+
+
+MilanLongInt firstComputeCandidateMateS(MilanLongInt adj1,
+                                       MilanLongInt adj2,
+                                       MilanLongInt *verLocInd,
+                                       MilanFloat *edgeLocWeight)
+{
+    MilanInt w = -1;
+    MilanFloat heaviestEdgeWt = 0.0f; // Assign the smallest 
+    int finalK;
+    for (int k = adj1; k < adj2; k++) {
+      if ((edgeLocWeight[k] > heaviestEdgeWt) ||
+	  ((edgeLocWeight[k] == heaviestEdgeWt) && (w < verLocInd[k])))     {
+	heaviestEdgeWt = edgeLocWeight[k];
+	w = verLocInd[k];
+	finalK = k;
+      }
+    } // End of for loop
+    return finalK;
+}
+
+/**
+ * //TODO documentation
+ * @param adj1
+ * @param adj2
+ * @param edgeLocWeight
+ * @param k
+ * @param verLocInd
+ * @param StartIndex
+ * @param EndIndex
+ * @param GMate
+ * @param Mate
+ * @param Ghost2LocalMap
+ * @return
+ */
+MilanLongInt computeCandidateMateS(MilanLongInt adj1,
+                                  MilanLongInt adj2,
+				  MilanFloat *edgeLocWeight,
+                                  MilanLongInt k,
+                                  MilanLongInt *verLocInd,
+                                  MilanLongInt StartIndex,
+                                  MilanLongInt EndIndex,
+                                  vector<MilanLongInt> &GMate,
+                                  MilanLongInt *Mate,
+                                  map<MilanLongInt, MilanLongInt> &Ghost2LocalMap)
+{
+    // Start: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
+
+    MilanInt w = -1;
+    MilanFloat heaviestEdgeWt = 0.0f; // Assign the smallest Value
+    for (k = adj1; k < adj2; k++)   {
+      if (isAlreadyMatched(verLocInd[k], StartIndex, EndIndex, GMate, Mate, Ghost2LocalMap))
+	continue;
+
+      if ((edgeLocWeight[k] > heaviestEdgeWt) ||
+	  ((edgeLocWeight[k] == heaviestEdgeWt) && (w < verLocInd[k])))     {
+	heaviestEdgeWt = edgeLocWeight[k];
+	w = verLocInd[k];
+      }
+    } // End of for loop
+      //  End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
+
     return w;
 }
 #endif

amgprec/impl/aggregator/extractUChunk.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 void extractUChunk(
     vector<MilanLongInt> &UChunkBeingProcessed,
     vector<MilanLongInt> &U,
@@ -28,4 +27,4 @@ void extractUChunk(
         }
 
     } // End of critical U // End of critical U
-}
+}

amgprec/impl/aggregator/findOwnerOfGhost.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 /// Find the owner of a ghost node:
 MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
                           MilanInt myRank, MilanInt numProcs)

amgprec/impl/aggregator/initialize.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
                 MilanLongInt StartIndex, MilanLongInt EndIndex,
                 MilanLongInt *numGhostEdges,
@@ -291,7 +290,7 @@ void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
                 //new (&U) staticQueue(NLVer + (*numGhostVertices));
                 U.reserve(NLVer + (*numGhostVertices));
 
-                // Initialize the private vectors                
+                // Initialize the private vectors
                 privateQLocalVtx.reserve(*numGhostVertices);
                 privateQGhostVtx.reserve(*numGhostVertices);
                 privateQMsgType.reserve(*numGhostVertices);

amgprec/impl/aggregator/isAlreadyMatched.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 /**
  * //TODO documentation
  * @param k
@@ -32,7 +31,7 @@ bool isAlreadyMatched(MilanLongInt node,
     */
     MilanLongInt val;
     if ((node < StartIndex) || (node > EndIndex)) // if ghost vertex
-    { 
+    {
 #pragma omp atomic read
         val = GMate[Ghost2LocalMap[node]];
         return val >= 0; // Already matched
@@ -43,4 +42,4 @@ bool isAlreadyMatched(MilanLongInt node,
     val = Mate[node - StartIndex];
 
     return val >= 0; // Already matched
-}
+}

amgprec/impl/aggregator/parallelComputeCandidateMateB.cpp

@@ -1,7 +1,6 @@
 #include "MatchBoxPC.h"
-#if !defined(SERIAL_MPI)
 
-void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
+void PARALLEL_COMPUTE_CANDIDATE_MATE_BD(MilanLongInt NLVer,
                                               MilanLongInt *verLocPtr,
                                               MilanLongInt *verLocInd,
                                               MilanInt myRank,
@@ -21,7 +20,35 @@ void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
             fflush(stdout);
 #endif
             // Start: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
-            candidateMate[v] = firstComputeCandidateMate(verLocPtr[v], verLocPtr[v + 1], verLocInd, edgeLocWeight);
+            candidateMate[v] = firstComputeCandidateMateD(verLocPtr[v], verLocPtr[v + 1],
+							 verLocInd, edgeLocWeight);
+            // End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
+        }
+    }
+}
+
+void PARALLEL_COMPUTE_CANDIDATE_MATE_BS(MilanLongInt NLVer,
+                                              MilanLongInt *verLocPtr,
+                                              MilanLongInt *verLocInd,
+                                              MilanInt myRank,
+                                              MilanFloat *edgeLocWeight,
+                                              MilanLongInt *candidateMate)
+{
+
+    MilanLongInt v = -1;
+
+#pragma omp parallel private(v) default(shared) num_threads(NUM_THREAD)
+    {
+
+#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] = firstComputeCandidateMateS(verLocPtr[v], verLocPtr[v + 1],
+							 verLocInd, edgeLocWeight);
             // End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
         }
     }

amgprec/impl/aggregator/processCrossEdge.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-
 void PROCESS_CROSS_EDGE(MilanLongInt *edge,
                         MilanLongInt *S)
 {
@@ -21,4 +20,4 @@ void PROCESS_CROSS_EDGE(MilanLongInt *edge,
 #endif
 
     // End: PARALLEL_PROCESS_CROSS_EDGE_B
-}
+}

amgprec/impl/aggregator/processExposedVertex.cpp

@@ -1,6 +1,5 @@
-#include "MatchBoxPC.h"
-#ifdef OPENMP
-void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
+#include "MatchBoxPC.h" 
+void PARALLEL_PROCESS_EXPOSED_VERTEX_BD(MilanLongInt NLVer,
                                        MilanLongInt *candidateMate,
                                        MilanLongInt *verLocInd,
                                        MilanLongInt *verLocPtr,
@@ -31,15 +30,16 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
                                        vector<MilanInt> &privateQOwner)
 {
 
-    MilanLongInt v = -1, k = -1, w = -1, adj11 = 0, adj12 = 0, k1 = 0;
+  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)    \
-    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner) \
-     default(shared)   num_threads(NUM_THREAD)
+    //#pragma omp parallel private(option, k, w, v, k1, adj11, adj12, ghostOwner) \
+  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      \
@@ -66,7 +66,7 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
 #pragma omp critical(Matching)
 	      {
 		if (isAlreadyMatched(verLocInd[k], StartIndex, EndIndex, GMate, Mate, Ghost2LocalMap)) {
-		  w = computeCandidateMate(verLocPtr[v], verLocPtr[v + 1], edgeLocWeight, 0,
+		  w = computeCandidateMateD(verLocPtr[v], verLocPtr[v + 1], edgeLocWeight, 0,
 					   verLocInd, StartIndex, EndIndex,
 					   GMate,  Mate,  Ghost2LocalMap);
 		  candidateMate[v] = w;
@@ -181,4 +181,189 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
 	
     } // End of parallel region
 }
+
+
+void PARALLEL_PROCESS_EXPOSED_VERTEX_BS(MilanLongInt NLVer,
+                                       MilanLongInt *candidateMate,
+                                       MilanLongInt *verLocInd,
+                                       MilanLongInt *verLocPtr,
+                                       MilanLongInt StartIndex,
+                                       MilanLongInt EndIndex,
+                                       MilanLongInt *Mate,
+                                       vector<MilanLongInt> &GMate,
+                                       map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+                                       MilanFloat *edgeLocWeight,
+                                       MilanLongInt *myCard,
+                                       MilanLongInt *msgInd,
+                                       MilanLongInt *NumMessagesBundled,
+                                       MilanLongInt *S,
+                                       MilanLongInt *verDistance,
+                                       MilanLongInt *PCounter,
+                                       vector<MilanLongInt> &Counter,
+                                       MilanInt myRank,
+                                       MilanInt numProcs,
+                                       vector<MilanLongInt> &U,
+                                       vector<MilanLongInt> &privateU,
+                                       vector<MilanLongInt> &QLocalVtx,
+                                       vector<MilanLongInt> &QGhostVtx,
+                                       vector<MilanLongInt> &QMsgType,
+                                       vector<MilanInt> &QOwner,
+                                       vector<MilanLongInt> &privateQLocalVtx,
+                                       vector<MilanLongInt> &privateQGhostVtx,
+                                       vector<MilanLongInt> &privateQMsgType,
+                                       vector<MilanInt> &privateQOwner)
+{
+
+    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) \
+    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, \
+		 privateQGhostVtx, privateQMsgType, privateQOwner) \
+     default(shared)   num_threads(NUM_THREAD)
+
+    {
+      //#pragma omp for reduction(+			\
+                          : PCounter[:numProcs], myCard \
+                          [:1], msgInd                  \
+                          [:1], NumMessagesBundled      \
+                          [:1])				\
+  schedule(static)
+        for (v = 0; v < NLVer; v++) {
+            option = -1;
+            // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+            k = candidateMate[v];
+            candidateMate[v] = verLocInd[k];
+            w = candidateMate[v];
+
+#ifdef PRINT_DEBUG_INFO_
+            cout << "\n(" << myRank << ")Processing: " << v + StartIndex << endl;
+            fflush(stdout);
+#endif
+
+#ifdef PRINT_DEBUG_INFO_
+            cout << "\n(" << myRank << ")" << v + StartIndex << " Points to: " << w;
+            fflush(stdout);
+#endif
+            // If found a dominating edge:
+            if (w >= 0) {
+#pragma omp critical(Matching)
+	      {
+		if (isAlreadyMatched(verLocInd[k], StartIndex, EndIndex, GMate, Mate, Ghost2LocalMap)) {
+		  w = computeCandidateMateS(verLocPtr[v], verLocPtr[v + 1], edgeLocWeight, 0,
+					    verLocInd, StartIndex, EndIndex,
+					    GMate,  Mate,  Ghost2LocalMap);
+		  candidateMate[v] = w;
+		}
+		if (w >= 0) {
+		  (*myCard)++;
+		  if ((w < StartIndex) || (w > EndIndex)) { // w is a ghost vertex
+		    option = 2;
+		    if (candidateMate[NLVer + Ghost2LocalMap[w]] == v + StartIndex) {
+		      option = 1;
+		      Mate[v] = w;
+		      GMate[Ghost2LocalMap[w]] = v + StartIndex; // w is a Ghost
+		    } // End of if CandidateMate[w] = v		  
+		  } // End of if a Ghost Vertex
+		  else  { // w is a local vertex
+		    if (candidateMate[w - StartIndex] == (v + StartIndex)) {
+		      option = 3;
+		      Mate[v] = w;                           // v is local
+		      Mate[w - StartIndex] = v + StartIndex; // w is local
+#ifdef PRINT_DEBUG_INFO_
+		      cout << "\n(" << myRank << ")MATCH: (" << v + StartIndex << "," << w << ") ";
+		      fflush(stdout);
+#endif
+		    } // End of if ( candidateMate[w-StartIndex] == (v+StartIndex) )
+		  }     // End of Else
+		} // End of second if
+	      }
+	      
+            } // End of if(w >=0)
+            else  {
+	      //#pragma omp critical(adjuse)
+	      {
+		// This piece of code is executed a really small number of times
+		adj11 = verLocPtr[v];
+		adj12 = verLocPtr[v + 1];
+		for (k1 = adj11; k1 < adj12; k1++) {
+		  w = verLocInd[k1];
+		  if ((w < StartIndex) || (w > EndIndex)) { // A ghost
+		    
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")Sending a failure message: ";
+		    cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+		    fflush(stdout);
+#endif
+		    (*msgInd)++;
+		    (*NumMessagesBundled)++;
+		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+		    // assert(ghostOwner != -1);
+		    // assert(ghostOwner != myRank);
+#pragma omp atomic
+		    PCounter[ghostOwner]++;
+		    
+		    privateQLocalVtx.push_back(v + StartIndex);
+		    privateQGhostVtx.push_back(w);
+		    privateQMsgType.push_back(FAILURE);
+		    privateQOwner.push_back(ghostOwner);
+		    
+		  } // End of if(GHOST)
+		}     // End of for loop
+	      }
+            }
+            // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+	    
+            switch (option)
+	      {
+	      case -1:
+                break;
+	      case 1:
+                privateU.push_back(v + StartIndex);
+                privateU.push_back(w);
+		
+#ifdef PRINT_DEBUG_INFO_
+                cout << "\n(" << myRank << ")MATCH: (" << v + StartIndex << "," << w << ")";
+                fflush(stdout);
+#endif
+		
+                //  Decrement the counter:
+                PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], S);
+	      case 2:
+#ifdef PRINT_DEBUG_INFO_
+                cout << "\n(" << myRank << ")Sending a request message (291):";
+                cout << "\n(" << myRank << ")Local is: " << v + StartIndex << " Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs) << endl;
+                fflush(stdout);
+#endif
+                (*msgInd)++;
+                (*NumMessagesBundled)++;
+                ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+                // assert(ghostOwner != -1);
+                // assert(ghostOwner != myRank);
+#pragma omp atomic
+                PCounter[ghostOwner]++;
+		
+                privateQLocalVtx.push_back(v + StartIndex);
+                privateQGhostVtx.push_back(w);
+                privateQMsgType.push_back(REQUEST);
+                privateQOwner.push_back(ghostOwner);
+                break;
+	      case 3:
+	      default:
+                privateU.push_back(v + StartIndex);
+                privateU.push_back(w);
+                break;
+	      }
+	    
+        } // End of for ( v=0; v < NLVer; v++ )
+	
+        queuesTransfer(U, privateU, QLocalVtx,
+                       QGhostVtx,
+                       QMsgType, QOwner, privateQLocalVtx,
+                       privateQGhostVtx,
+                       privateQMsgType,
+                       privateQOwner);
+	
+    } // End of parallel region
+}
 #endif

amgprec/impl/aggregator/processMatchedVertices.cpp

@@ -1,7 +1,5 @@
 #include "MatchBoxPC.h"
-
-#if !defined(SERIAL_MPI)
-void processMatchedVertices(
+void processMatchedVerticesD(
     MilanLongInt NLVer,
     vector<MilanLongInt> &UChunkBeingProcessed,
     vector<MilanLongInt> &U,
@@ -59,29 +57,29 @@ void processMatchedVertices(
     {
 
         while (!U.empty()) {
-	  
+
 	  extractUChunk(UChunkBeingProcessed, U, privateU);
-	  
+
 	  for (MilanLongInt u : UChunkBeingProcessed)   {
 #ifdef PRINT_DEBUG_INFO_
 	    cout << "\n(" << myRank << ")u: " << u;
 	    fflush(stdout);
 #endif
 	    if ((u >= StartIndex) && (u <= EndIndex))     { // Process Only the Local Vertices
-	      
+
 #ifdef COUNT_LOCAL_VERTEX
 	      localVertices++;
 #endif
-	      
+
 	      // Get the Adjacency list for u
 	      adj1 = verLocPtr[u - StartIndex]; // Pointer
 	      adj2 = verLocPtr[u - StartIndex + 1];
 	      for (k = adj1; k < adj2; k++) {
 		option = -1;
 		v = verLocInd[k];
-		
+
 		if ((v >= StartIndex) && (v <= EndIndex)) { // If Local Vertex:
-		  
+
 #ifdef PRINT_DEBUG_INFO_
 		  cout << "\n(" << myRank << ")v: " << v << " c(v)= " << candidateMate[v - StartIndex] << " Mate[v]: " << Mate[v];
 		  fflush(stdout);
@@ -92,62 +90,62 @@ void processMatchedVertices(
 		  if (mateVal < 0)  {
 #pragma omp critical
 		    {
-		      if (candidateMate[v - StartIndex] == u) {
+#pragma omp atomic read
+		      mateVal = Mate[v - StartIndex];
+		      // If the current vertex is pointing to a matched vertex and is not matched
+		      if (mateVal < 0)  {
+			
+			if (candidateMate[v - StartIndex] == u) {
 			  // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
-			w = computeCandidateMate(verLocPtr[v - StartIndex],
-						 verLocPtr[v - StartIndex + 1],
-						 edgeLocWeight, 0,
-						 verLocInd,
-						 StartIndex,
-						 EndIndex,
-						 GMate,
-						 Mate,
-						 Ghost2LocalMap);
-			
-			candidateMate[v - StartIndex] = w;
-			
+			  w = computeCandidateMateD(verLocPtr[v - StartIndex],
+						   verLocPtr[v - StartIndex + 1],
+						   edgeLocWeight, 0,
+						   verLocInd, StartIndex, EndIndex,
+						   GMate, Mate, Ghost2LocalMap);
+			  candidateMate[v - StartIndex] = w;
 #ifdef PRINT_DEBUG_INFO_
-			cout << "\n(" << myRank << ")" << v << " Points to: " << w;
-			fflush(stdout);
+			  cout << "\n(" << myRank << ")" << v << " Points to: " << w;
+			  fflush(stdout);
 #endif
-			// If found a dominating edge:
-			if (w >= 0)   {			  
-			  if ((w < StartIndex) || (w > EndIndex))  { // A ghost
+			  // If found a dominating edge:
+			  if (w >= 0)   {
+			    if ((w < StartIndex) || (w > EndIndex))  { // A ghost
 #ifdef PRINT_DEBUG_INFO_
-			    cout << "\n(" << myRank << ")Sending a request message:";
-			    cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+			      cout << "\n(" << myRank << ")Sending a request message:";
+			      cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
 #endif
-			    option = 2;
-			    
-			    if (candidateMate[NLVer + Ghost2LocalMap[w]] == v)  {
-			      option = 1;
-			      Mate[v - StartIndex] = w;     // v is a local vertex
-			      GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
-			      
-			    } // End of if CandidateMate[w] = v
-			  }     // End of if a Ghost Vertex
-			  else    { // w is a local vertex
-			    if (candidateMate[w - StartIndex] == v)  {
-			      option = 3;
-			      Mate[v - StartIndex] = w; // v is a local vertex
-			      Mate[w - StartIndex] = v; // w is a local vertex
+			      option = 2;
 			      
+			      if (candidateMate[NLVer + Ghost2LocalMap[w]] == v)  {
+				option = 1;
+				Mate[v - StartIndex] = w;     // v is a local vertex
+				GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
+				
+			      } // End of if CandidateMate[w] = v
+			    }     // End of if a Ghost Vertex
+			    else    { // w is a local vertex
+			      if (candidateMate[w - StartIndex] == v)  {
+				option = 3;
+				Mate[v - StartIndex] = w; // v is a local vertex
+				Mate[w - StartIndex] = v; // w is a local vertex
+				
 #ifdef PRINT_DEBUG_INFO_
-			      cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
-			      fflush(stdout);
+				cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+				fflush(stdout);
 #endif
-			    } // End of if(CandidateMate(w) = v
-			  }     // End of Else
-			}         // End of if(w >=0)
-			else
-			  option = 4; // End of Else: w == -1
-			// End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
-		      } // End of If (candidateMate[v-StartIndex] == u
+			      } // End of if(CandidateMate(w) = v
+			    }     // End of Else
+			  }         // End of if(w >=0)
+			  else
+			    option = 4; // End of Else: w == -1
+			  // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			} // End of If (candidateMate[v-StartIndex] == u
+		      }
 		    }     // End of task
 		  }         // mateval < 0
 		}             // End of if ( (v >= StartIndex) && (v <= EndIndex) ) //If Local Vertex:
 		else  { // Neighbor is a ghost vertex
-		  
+
 #pragma omp critical
 		  {
 		    if (candidateMate[NLVer + Ghost2LocalMap[v]] == u)
@@ -156,7 +154,7 @@ void processMatchedVertices(
 		      option = 5; // u is local
 		  }                   // End of critical
 		}                       // End of Else //A Ghost Vertex
-		
+
 		switch (option)
 		  {
 		  case -1:
@@ -166,7 +164,7 @@ void processMatchedVertices(
 		    // Found a dominating edge, it is a ghost and candidateMate[NLVer + Ghost2LocalMap[w]] == v
 		    privateU.push_back(v);
 		    privateU.push_back(w);
-		    
+
 		    (*myCard)++;
 #ifdef PRINT_DEBUG_INFO_
 		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
@@ -175,15 +173,16 @@ void processMatchedVertices(
 		    // Decrement the counter:
 		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], SPtr);
 		  case 2:
-		    
+
 		    // Found a dominating edge, it is a ghost
 		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
 		    // assert(ghostOwner != -1);
 		    // assert(ghostOwner != myRank);
+#pragma omp atomic		    
 		    PCounter[ghostOwner]++;
 		    (*NumMessagesBundled)++;
 		    (*msgInd)++;
-		    
+
 		    privateQLocalVtx.push_back(v);
 		    privateQGhostVtx.push_back(w);
 		    privateQMsgType.push_back(REQUEST);
@@ -192,7 +191,7 @@ void processMatchedVertices(
 		  case 3:
 		    privateU.push_back(v);
 		    privateU.push_back(w);
-		    
+
 		    (*myCard)++;
 		    break;
 		  case 4:
@@ -202,94 +201,384 @@ void processMatchedVertices(
 		    for (k1 = adj11; k1 < adj12; k1++) {
 		      w = verLocInd[k1];
 		      if ((w < StartIndex) || (w > EndIndex)) { // A ghost
-			
+
 #ifdef PRINT_DEBUG_INFO_
 			cout << "\n(" << myRank << ")Sending a failure message: ";
 			cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
 			fflush(stdout);
 #endif
-			
+
 			ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
 			// assert(ghostOwner != -1);
 			// assert(ghostOwner != myRank);
-			
+#pragma omp atomic		    
 			PCounter[ghostOwner]++;
 			(*NumMessagesBundled)++;
 			(*msgInd)++;
-			
+
 			privateQLocalVtx.push_back(v);
 			privateQGhostVtx.push_back(w);
 			privateQMsgType.push_back(FAILURE);
 			privateQOwner.push_back(ghostOwner);
-			
+
 		      } // End of if(GHOST)
 		    }     // End of for loop
 		    break;
 		  case 5:
 		  default:
-		    
+
 #ifdef PRINT_DEBUG_INFO_
 		    cout << "\n(" << myRank << ")Sending a success message: ";
 		    cout << "\n(" << myRank << ")Ghost is " << v << " Owner is: " << findOwnerOfGhost(v, verDistance, myRank, numProcs) << "\n";
 		    fflush(stdout);
 #endif
-		    
+
 		    ghostOwner = findOwnerOfGhost(v, verDistance, myRank, numProcs);
 		    // assert(ghostOwner != -1);
 		    // assert(ghostOwner != myRank);
-		    
+
 		    (*NumMessagesBundled)++;
 		    PCounter[ghostOwner]++;
 		    (*msgInd)++;
-		    
+
 		    privateQLocalVtx.push_back(u);
 		    privateQGhostVtx.push_back(v);
 		    privateQMsgType.push_back(SUCCESS);
 		    privateQOwner.push_back(ghostOwner);
-		    
+
 		    break;
 		  } // End of switch
-		
 	      } // End of inner for
 	    }
 	  } // End of outer for
-	  
+
 	  queuesTransfer(U, privateU, QLocalVtx,
 			 QGhostVtx,
 			 QMsgType, QOwner, privateQLocalVtx,
 			 privateQGhostVtx,
 			 privateQMsgType,
 			 privateQOwner);
-	  
+
 #pragma omp critical(U)
 	  {
 	    U.insert(U.end(), privateU.begin(), privateU.end());
 	  }
-	  
-	  privateU.clear();
-	  
+
 #pragma omp critical(sendMessageTransfer)
 	  {
-	    
 	    QLocalVtx.insert(QLocalVtx.end(), privateQLocalVtx.begin(), privateQLocalVtx.end());
 	    QGhostVtx.insert(QGhostVtx.end(), privateQGhostVtx.begin(), privateQGhostVtx.end());
 	    QMsgType.insert(QMsgType.end(), privateQMsgType.begin(), privateQMsgType.end());
 	    QOwner.insert(QOwner.end(), privateQOwner.begin(), privateQOwner.end());
 	  }
-	  
+
+	  privateU.clear();
 	  privateQLocalVtx.clear();
 	  privateQGhostVtx.clear();
 	  privateQMsgType.clear();
 	  privateQOwner.clear();
-	  
+
         } // End of while ( !U.empty() )
-	
+
 #ifdef COUNT_LOCAL_VERTEX
         printf("Count local vertexes: %ld for thread %d of processor %d\n",
                localVertices,
                omp_get_thread_num(),
                myRank);
-	
+
+#endif
+    } // End of parallel region
+}
+
+
+
+void processMatchedVerticesS(
+    MilanLongInt NLVer,
+    vector<MilanLongInt> &UChunkBeingProcessed,
+    vector<MilanLongInt> &U,
+    vector<MilanLongInt> &privateU,
+    MilanLongInt StartIndex,
+    MilanLongInt EndIndex,
+    MilanLongInt *myCard,
+    MilanLongInt *msgInd,
+    MilanLongInt *NumMessagesBundled,
+    MilanLongInt *SPtr,
+    MilanLongInt *verLocPtr,
+    MilanLongInt *verLocInd,
+    MilanLongInt *verDistance,
+    MilanLongInt *PCounter,
+    vector<MilanLongInt> &Counter,
+    MilanInt myRank,
+    MilanInt numProcs,
+    MilanLongInt *candidateMate,
+    vector<MilanLongInt> &GMate,
+    MilanLongInt *Mate,
+    map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+    MilanFloat *edgeLocWeight,
+    vector<MilanLongInt> &QLocalVtx,
+    vector<MilanLongInt> &QGhostVtx,
+    vector<MilanLongInt> &QMsgType,
+    vector<MilanInt> &QOwner,
+    vector<MilanLongInt> &privateQLocalVtx,
+    vector<MilanLongInt> &privateQGhostVtx,
+    vector<MilanLongInt> &privateQMsgType,
+    vector<MilanInt> &privateQOwner)
+{
+
+    MilanLongInt adj1, adj2, adj11, adj12, k, k1, v = -1, w = -1, ghostOwner;
+    int option;
+    MilanLongInt mateVal;
+
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+
+#ifdef COUNT_LOCAL_VERTEX
+    MilanLongInt localVertices = 0;
+#endif
+    //#pragma omp parallel private(k, w, v, k1, adj1, adj2, adj11, adj12, ghostOwner, option) \
+    firstprivate(privateU, StartIndex, EndIndex, privateQLocalVtx, privateQGhostVtx, \
+       privateQMsgType, privateQOwner, UChunkBeingProcessed) \
+       default(shared)  num_threads(NUM_THREAD)   \
+            reduction(+                           \
+                      : msgInd[:1], PCounter      \
+                      [:numProcs], myCard         \
+                      [:1], NumMessagesBundled    \
+                      [:1])
+    {
+
+        while (!U.empty()) {
+
+	  extractUChunk(UChunkBeingProcessed, U, privateU);
+
+	  for (MilanLongInt u : UChunkBeingProcessed)   {
+#ifdef PRINT_DEBUG_INFO_
+	    cout << "\n(" << myRank << ")u: " << u;
+	    fflush(stdout);
+#endif
+	    if ((u >= StartIndex) && (u <= EndIndex))     { // Process Only the Local Vertices
+
+#ifdef COUNT_LOCAL_VERTEX
+	      localVertices++;
+#endif
+
+	      // Get the Adjacency list for u
+	      adj1 = verLocPtr[u - StartIndex]; // Pointer
+	      adj2 = verLocPtr[u - StartIndex + 1];
+	      for (k = adj1; k < adj2; k++) {
+		option = -1;
+		v = verLocInd[k];
+
+		if ((v >= StartIndex) && (v <= EndIndex)) { // If Local Vertex:
+
+#ifdef PRINT_DEBUG_INFO_
+		  cout << "\n(" << myRank << ")v: " << v << " c(v)= " << candidateMate[v - StartIndex] << " Mate[v]: " << Mate[v];
+		  fflush(stdout);
+#endif
+#pragma omp atomic read
+		  mateVal = Mate[v - StartIndex];
+		  // If the current vertex is pointing to a matched vertex and is not matched
+		  if (mateVal < 0)  {
+#pragma omp critical
+		    {
+#pragma omp atomic read
+		      mateVal = Mate[v - StartIndex];
+		      // If the current vertex is pointing to a matched vertex and is not matched
+		      if (mateVal < 0)  {
+			
+			if (candidateMate[v - StartIndex] == u) {
+			  // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			  w = computeCandidateMateS(verLocPtr[v - StartIndex],
+						   verLocPtr[v - StartIndex + 1],
+						   edgeLocWeight, 0,
+						   verLocInd, StartIndex, EndIndex,
+						   GMate, Mate, Ghost2LocalMap);
+			  candidateMate[v - StartIndex] = w;
+#ifdef PRINT_DEBUG_INFO_
+			  cout << "\n(" << myRank << ")" << v << " Points to: " << w;
+			  fflush(stdout);
+#endif
+			  // If found a dominating edge:
+			  if (w >= 0)   {
+			    if ((w < StartIndex) || (w > EndIndex))  { // A ghost
+#ifdef PRINT_DEBUG_INFO_
+			      cout << "\n(" << myRank << ")Sending a request message:";
+			      cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+#endif
+			      option = 2;
+			      
+			      if (candidateMate[NLVer + Ghost2LocalMap[w]] == v)  {
+				option = 1;
+				Mate[v - StartIndex] = w;     // v is a local vertex
+				GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
+				
+			      } // End of if CandidateMate[w] = v
+			    }     // End of if a Ghost Vertex
+			    else    { // w is a local vertex
+			      if (candidateMate[w - StartIndex] == v)  {
+				option = 3;
+				Mate[v - StartIndex] = w; // v is a local vertex
+				Mate[w - StartIndex] = v; // w is a local vertex
+				
+#ifdef PRINT_DEBUG_INFO_
+				cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+				fflush(stdout);
+#endif
+			      } // End of if(CandidateMate(w) = v
+			    }     // End of Else
+			  }         // End of if(w >=0)
+			  else
+			    option = 4; // End of Else: w == -1
+			  // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			} // End of If (candidateMate[v-StartIndex] == u
+		      }
+		    }     // End of task
+		  }         // mateval < 0
+		}             // End of if ( (v >= StartIndex) && (v <= EndIndex) ) //If Local Vertex:
+		else  { // Neighbor is a ghost vertex
+
+#pragma omp critical
+		  {
+		    if (candidateMate[NLVer + Ghost2LocalMap[v]] == u)
+		      candidateMate[NLVer + Ghost2LocalMap[v]] = -1;
+		    if (v != Mate[u - StartIndex])
+		      option = 5; // u is local
+		  }                   // End of critical
+		}                       // End of Else //A Ghost Vertex
+
+		switch (option)
+		  {
+		  case -1:
+		    // No things to do
+		    break;
+		  case 1:
+		    // Found a dominating edge, it is a ghost and candidateMate[NLVer + Ghost2LocalMap[w]] == v
+		    privateU.push_back(v);
+		    privateU.push_back(w);
+
+		    (*myCard)++;
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+		    fflush(stdout);
+#endif
+		    // Decrement the counter:
+		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], SPtr);
+		  case 2:
+
+		    // Found a dominating edge, it is a ghost
+		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+		    // assert(ghostOwner != -1);
+		    // assert(ghostOwner != myRank);
+#pragma omp atomic		    
+		    PCounter[ghostOwner]++;
+		    (*NumMessagesBundled)++;
+		    (*msgInd)++;
+
+		    privateQLocalVtx.push_back(v);
+		    privateQGhostVtx.push_back(w);
+		    privateQMsgType.push_back(REQUEST);
+		    privateQOwner.push_back(ghostOwner);
+		    break;
+		  case 3:
+		    privateU.push_back(v);
+		    privateU.push_back(w);
+
+		    (*myCard)++;
+		    break;
+		  case 4:
+		    // Could not find a dominating vertex
+		    adj11 = verLocPtr[v - StartIndex];
+		    adj12 = verLocPtr[v - StartIndex + 1];
+		    for (k1 = adj11; k1 < adj12; k1++) {
+		      w = verLocInd[k1];
+		      if ((w < StartIndex) || (w > EndIndex)) { // A ghost
+
+#ifdef PRINT_DEBUG_INFO_
+			cout << "\n(" << myRank << ")Sending a failure message: ";
+			cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+			fflush(stdout);
+#endif
+
+			ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+			// assert(ghostOwner != -1);
+			// assert(ghostOwner != myRank);
+#pragma omp atomic		    
+			PCounter[ghostOwner]++;
+			(*NumMessagesBundled)++;
+			(*msgInd)++;
+
+			privateQLocalVtx.push_back(v);
+			privateQGhostVtx.push_back(w);
+			privateQMsgType.push_back(FAILURE);
+			privateQOwner.push_back(ghostOwner);
+
+		      } // End of if(GHOST)
+		    }     // End of for loop
+		    break;
+		  case 5:
+		  default:
+
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")Sending a success message: ";
+		    cout << "\n(" << myRank << ")Ghost is " << v << " Owner is: " << findOwnerOfGhost(v, verDistance, myRank, numProcs) << "\n";
+		    fflush(stdout);
+#endif
+
+		    ghostOwner = findOwnerOfGhost(v, verDistance, myRank, numProcs);
+		    // assert(ghostOwner != -1);
+		    // assert(ghostOwner != myRank);
+
+		    (*NumMessagesBundled)++;
+		    PCounter[ghostOwner]++;
+		    (*msgInd)++;
+
+		    privateQLocalVtx.push_back(u);
+		    privateQGhostVtx.push_back(v);
+		    privateQMsgType.push_back(SUCCESS);
+		    privateQOwner.push_back(ghostOwner);
+
+		    break;
+		  } // End of switch
+	      } // End of inner for
+	    }
+	  } // End of outer for
+
+	  queuesTransfer(U, privateU, QLocalVtx,
+			 QGhostVtx,
+			 QMsgType, QOwner, privateQLocalVtx,
+			 privateQGhostVtx,
+			 privateQMsgType,
+			 privateQOwner);
+
+#pragma omp critical(U)
+	  {
+	    U.insert(U.end(), privateU.begin(), privateU.end());
+	  }
+
+#pragma omp critical(sendMessageTransfer)
+	  {
+	    QLocalVtx.insert(QLocalVtx.end(), privateQLocalVtx.begin(), privateQLocalVtx.end());
+	    QGhostVtx.insert(QGhostVtx.end(), privateQGhostVtx.begin(), privateQGhostVtx.end());
+	    QMsgType.insert(QMsgType.end(), privateQMsgType.begin(), privateQMsgType.end());
+	    QOwner.insert(QOwner.end(), privateQOwner.begin(), privateQOwner.end());
+	  }
+
+	  privateU.clear();
+	  privateQLocalVtx.clear();
+	  privateQGhostVtx.clear();
+	  privateQMsgType.clear();
+	  privateQOwner.clear();
+
+        } // End of while ( !U.empty() )
+
+#ifdef COUNT_LOCAL_VERTEX
+        printf("Count local vertexes: %ld for thread %d of processor %d\n",
+               localVertices,
+               omp_get_thread_num(),
+               myRank);
+
 #endif
     } // End of parallel region
 }

amgprec/impl/aggregator/processMatchedVerticesAndSendMessages.cpp

@@ -1,7 +1,6 @@
 #include "MatchBoxPC.h"
 //#define DEBUG_HANG_
-#if !defined(SERIAL_MPI)
-void processMatchedVerticesAndSendMessages(
+void processMatchedVerticesAndSendMessagesD(
 					   MilanLongInt NLVer,
 					   vector<MilanLongInt> &UChunkBeingProcessed,
 					   vector<MilanLongInt> &U,
@@ -27,6 +26,302 @@ void processMatchedVerticesAndSendMessages(
 					   vector<MilanLongInt> &QLocalVtx,
 					   vector<MilanLongInt> &QGhostVtx,
 					   vector<MilanLongInt> &QMsgType,
+ 					   vector<MilanInt> &QOwner,
+					   vector<MilanLongInt> &privateQLocalVtx,
+					   vector<MilanLongInt> &privateQGhostVtx,
+					   vector<MilanLongInt> &privateQMsgType,
+					   vector<MilanInt> &privateQOwner,
+					   MPI_Comm comm,
+					   MilanLongInt *msgActual,
+					   vector<MilanLongInt> &Message)
+{
+
+    MilanLongInt initialSize = QLocalVtx.size();
+    MilanLongInt adj1, adj2, adj11, adj12, k, k1, v = -1, w = -1, ghostOwner;
+    int option;
+    MilanLongInt mateVal;
+
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+
+#ifdef COUNT_LOCAL_VERTEX
+    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, privateQMsgType, privateQOwner, UChunkBeingProcessed) \
+  default(shared)							\
+      num_threads(NUM_THREAD)						\
+  reduction(+								\
+	    : msgInd[:1], PCounter					\
+	    [:numProcs], myCard						\
+	    [:1], NumMessagesBundled					\
+	    [:1], msgActual						\
+	    [:1])
+    {
+
+        while (!U.empty()) {
+
+	  extractUChunk(UChunkBeingProcessed, U, privateU);
+
+	  for (MilanLongInt u : UChunkBeingProcessed) {
+#ifdef PRINT_DEBUG_INFO_
+	    cout << "\n(" << myRank << ")u: " << u;
+	    fflush(stdout);
+#endif
+	    if ((u >= StartIndex) && (u <= EndIndex)) { // Process Only the Local Vertices
+
+#ifdef COUNT_LOCAL_VERTEX
+	      localVertices++;
+#endif
+	      // Get the Adjacency list for u
+	      adj1 = verLocPtr[u - StartIndex]; // Pointer
+	      adj2 = verLocPtr[u - StartIndex + 1];
+	      for (k = adj1; k < adj2; k++) {
+		option = -1;
+		v = verLocInd[k];
+
+		if ((v >= StartIndex) && (v <= EndIndex)) { // If Local Vertex:
+
+#ifdef PRINT_DEBUG_INFO_
+		  cout << "\n(" << myRank << ")v: " << v << " c(v)= " << candidateMate[v - StartIndex] << " Mate[v]: " << Mate[v];
+		  fflush(stdout);
+#endif
+#pragma omp atomic read
+		  mateVal = Mate[v - StartIndex];
+		  // If the current vertex is pointing to a matched vertex and is not matched
+		  if (mateVal < 0) {
+#pragma omp critical
+		    {
+#pragma omp atomic read
+		      mateVal = Mate[v - StartIndex];
+		      // If the current vertex is pointing to a matched vertex and is not matched
+		      if (mateVal < 0) {
+			
+			if (candidateMate[v - StartIndex] == u)  {
+			  // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			  w = computeCandidateMateD(verLocPtr[v - StartIndex],
+						   verLocPtr[v - StartIndex + 1],
+						   edgeLocWeight, 0,
+						   verLocInd, StartIndex, EndIndex,
+						   GMate, Mate, Ghost2LocalMap);
+			  candidateMate[v - StartIndex] = w;
+#ifdef PRINT_DEBUG_INFO_
+			  cout << "\n(" << myRank << ")" << v << " Points to: " << w;
+			  fflush(stdout);
+#endif
+			  // If found a dominating edge:
+			  if (w >= 0)  {
+			    
+			    if ((w < StartIndex) || (w > EndIndex))  { // A ghost
+#ifdef PRINT_DEBUG_INFO_
+			      cout << "\n(" << myRank << ")Sending a request message:";
+			      cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+#endif
+			      option = 2;
+			      
+			      if (candidateMate[NLVer + Ghost2LocalMap[w]] == v) {
+				option = 1;
+				Mate[v - StartIndex] = w;     // v is a local vertex
+				GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
+				
+			      } // End of if CandidateMate[w] = v
+			    }     // End of if a Ghost Vertex
+			    else   { // w is a local vertex
+			      if (candidateMate[w - StartIndex] == v)  {
+				option = 3;
+				Mate[v - StartIndex] = w; // v is a local vertex
+				Mate[w - StartIndex] = v; // w is a local vertex
+#ifdef PRINT_DEBUG_INFO_
+				cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+				fflush(stdout);
+#endif
+			      } // End of if(CandidateMate(w) = v
+			    }     // End of Else
+			  }         // End of if(w >=0)
+			  else
+			    option = 4; // End of Else: w == -1
+			  // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			} // End of If (candidateMate[v-StartIndex] == u
+		      }
+		    }     // End of task
+		  }         // mateval < 0
+		}             // End of if ( (v >= StartIndex) && (v <= EndIndex) ) //If Local Vertex:
+		else { // Neighbor is a ghost vertex
+
+#pragma omp critical
+		  {
+		    if (candidateMate[NLVer + Ghost2LocalMap[v]] == u)
+		      candidateMate[NLVer + Ghost2LocalMap[v]] = -1;
+		    if (v != Mate[u - StartIndex])
+		      option = 5; // u is local
+		  }                   // End of critical
+		}                       // End of Else //A Ghost Vertex
+
+		switch (option)
+		  {
+		  case -1:
+		    // No things to do
+		    break;
+		  case 1:
+		    // Found a dominating edge, it is a ghost and candidateMate[NLVer + Ghost2LocalMap[w]] == v
+		    privateU.push_back(v);
+		    privateU.push_back(w);
+		    (*myCard)++;
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+		    fflush(stdout);
+#endif
+		    // Decrement the counter:
+		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], SPtr);
+		  case 2:
+
+		    // Found a dominating edge, it is a ghost
+		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+
+		    // Build the Message Packet:
+		    // Message[0] = v;       // LOCAL
+		    // Message[1] = w;       // GHOST
+		    // Message[2] = REQUEST; // TYPE
+		    // Send a Request (Asynchronous)
+		    // MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+
+		    (*msgActual)++;
+		    (*msgInd)++;
+
+		    privateQLocalVtx.push_back(v);
+		    privateQGhostVtx.push_back(w);
+		    privateQMsgType.push_back(REQUEST);
+		    privateQOwner.push_back(ghostOwner);
+		    break;
+		  case 3:
+		    privateU.push_back(v);
+		    privateU.push_back(w);
+		    (*myCard)++;
+		    break;
+		  case 4:
+		    // Could not find a dominating vertex
+		    adj11 = verLocPtr[v - StartIndex];
+		    adj12 = verLocPtr[v - StartIndex + 1];
+		    for (k1 = adj11; k1 < adj12; k1++) {
+		      w = verLocInd[k1];
+		      if ((w < StartIndex) || (w > EndIndex))  { // A ghost
+#ifdef PRINT_DEBUG_INFO_
+			cout << "\n(" << myRank << ")Sending a failure message: ";
+			cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
+			fflush(stdout);
+#endif
+			ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+			// Build the Message Packet:
+			// Message[0] = v;       // LOCAL
+			// Message[1] = w;       // GHOST
+			// Message[2] = FAILURE; // TYPE
+			// Send a Request (Asynchronous)
+			// MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+
+			(*msgActual)++;
+			(*msgInd)++;
+			privateQLocalVtx.push_back(v);
+			privateQGhostVtx.push_back(w);
+			privateQMsgType.push_back(FAILURE);
+			privateQOwner.push_back(ghostOwner);
+
+		      } // End of if(GHOST)
+		    }     // End of for loop
+		    break;
+		  case 5:
+		  default:
+
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")Sending a success message: ";
+		    cout << "\n(" << myRank << ")Ghost is " << v << " Owner is: " << findOwnerOfGhost(v, verDistance, myRank, numProcs) << "\n";
+		    fflush(stdout);
+#endif
+
+		    ghostOwner = findOwnerOfGhost(v, verDistance, myRank, numProcs);
+		    // Build the Message Packet:
+		    // Message[0] = u;       // LOCAL
+		    // Message[1] = v;       // GHOST
+		    // Message[2] = SUCCESS; // TYPE
+		    // Send a Request (Asynchronous)
+		    // MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+		    (*msgActual)++;
+		    (*msgInd)++;
+		    privateQLocalVtx.push_back(u);
+		    privateQGhostVtx.push_back(v);
+		    privateQMsgType.push_back(SUCCESS);
+		    privateQOwner.push_back(ghostOwner);
+
+		    break;
+		  } // End of switch
+	      }     // End of inner for
+	    }
+	  } // End of outer for
+
+	  queuesTransfer(U, privateU, QLocalVtx,
+			 QGhostVtx,
+			 QMsgType, QOwner, privateQLocalVtx,
+			 privateQGhostVtx,
+			 privateQMsgType,
+			 privateQOwner);
+
+        } // End of while ( !U.empty() )
+
+#ifdef COUNT_LOCAL_VERTEX
+        printf("Count local vertexes: %ld for thread %d of processor %d\n",
+               localVertices, mp_get_thread_num(), myRank);
+#endif
+    } // End of parallel region
+
+    // Send the messages
+#ifdef DEBUG_HANG_
+    cout << myRank<<" Sending: "<<QOwner.size()-initialSize<<" messages" <<endl;
+#endif
+    for (int i = initialSize; i < QOwner.size(); i++)   {
+      Message[0] = QLocalVtx[i];
+      Message[1] = QGhostVtx[i];
+      Message[2] = QMsgType[i];
+      ghostOwner = QOwner[i];
+      //MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+      //cout << myRank<<" Sending to "<<ghostOwner<<endl;
+      MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+    }
+#ifdef DEBUG_HANG_
+    cout << myRank<<" Done sending messages"<<endl;
+#endif
+}
+
+
+
+void processMatchedVerticesAndSendMessagesS(
+					   MilanLongInt NLVer,
+					   vector<MilanLongInt> &UChunkBeingProcessed,
+					   vector<MilanLongInt> &U,
+					   vector<MilanLongInt> &privateU,
+					   MilanLongInt StartIndex,
+					   MilanLongInt EndIndex,
+					   MilanLongInt *myCard,
+					   MilanLongInt *msgInd,
+					   MilanLongInt *NumMessagesBundled,
+					   MilanLongInt *SPtr,
+					   MilanLongInt *verLocPtr,
+					   MilanLongInt *verLocInd,
+					   MilanLongInt *verDistance,
+					   MilanLongInt *PCounter,
+					   vector<MilanLongInt> &Counter,
+					   MilanInt myRank,
+					   MilanInt numProcs,
+					   MilanLongInt *candidateMate,
+					   vector<MilanLongInt> &GMate,
+					   MilanLongInt *Mate,
+					   map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+					   MilanFloat *edgeLocWeight,
+					   vector<MilanLongInt> &QLocalVtx,
+					   vector<MilanLongInt> &QGhostVtx,
+					   vector<MilanLongInt> &QMsgType,
 					   vector<MilanInt> &QOwner,
 					   vector<MilanLongInt> &privateQLocalVtx,
 					   vector<MilanLongInt> &privateQGhostVtx,
@@ -64,29 +359,28 @@ void processMatchedVerticesAndSendMessages(
     {
 
         while (!U.empty()) {
-	  
+
 	  extractUChunk(UChunkBeingProcessed, U, privateU);
-	  
+
 	  for (MilanLongInt u : UChunkBeingProcessed) {
 #ifdef PRINT_DEBUG_INFO_
 	    cout << "\n(" << myRank << ")u: " << u;
 	    fflush(stdout);
 #endif
 	    if ((u >= StartIndex) && (u <= EndIndex)) { // Process Only the Local Vertices
-	      
+
 #ifdef COUNT_LOCAL_VERTEX
 	      localVertices++;
 #endif
-	      
 	      // Get the Adjacency list for u
 	      adj1 = verLocPtr[u - StartIndex]; // Pointer
 	      adj2 = verLocPtr[u - StartIndex + 1];
 	      for (k = adj1; k < adj2; k++) {
 		option = -1;
 		v = verLocInd[k];
-		
+
 		if ((v >= StartIndex) && (v <= EndIndex)) { // If Local Vertex:
-		  
+
 #ifdef PRINT_DEBUG_INFO_
 		  cout << "\n(" << myRank << ")v: " << v << " c(v)= " << candidateMate[v - StartIndex] << " Mate[v]: " << Mate[v];
 		  fflush(stdout);
@@ -97,63 +391,62 @@ void processMatchedVerticesAndSendMessages(
 		  if (mateVal < 0) {
 #pragma omp critical
 		    {
-		      if (candidateMate[v - StartIndex] == u)  {
-			// Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
-			w = computeCandidateMate(verLocPtr[v - StartIndex],
-						 verLocPtr[v - StartIndex + 1],
-						 edgeLocWeight, 0,
-						 verLocInd,
-						 StartIndex,
-						 EndIndex,
-						 GMate,
-						 Mate,
-						 Ghost2LocalMap);
-			
-			candidateMate[v - StartIndex] = w;
+#pragma omp atomic read
+		      mateVal = Mate[v - StartIndex];
+		      // If the current vertex is pointing to a matched vertex and is not matched
+		      if (mateVal < 0) {
 			
+			if (candidateMate[v - StartIndex] == u)  {
+			  // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			  w = computeCandidateMateS(verLocPtr[v - StartIndex],
+						   verLocPtr[v - StartIndex + 1],
+						   edgeLocWeight, 0,
+						   verLocInd, StartIndex, EndIndex,
+						   GMate, Mate, Ghost2LocalMap);
+			  candidateMate[v - StartIndex] = w;
 #ifdef PRINT_DEBUG_INFO_
-			cout << "\n(" << myRank << ")" << v << " Points to: " << w;
-			fflush(stdout);
+			  cout << "\n(" << myRank << ")" << v << " Points to: " << w;
+			  fflush(stdout);
 #endif
-			// If found a dominating edge:
-			if (w >= 0)  {
-			  
-			  if ((w < StartIndex) || (w > EndIndex))  { // A ghost
-#ifdef PRINT_DEBUG_INFO_
-			    cout << "\n(" << myRank << ")Sending a request message:";
-			    cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
-#endif
-			    option = 2;
+			  // If found a dominating edge:
+			  if (w >= 0)  {
 			    
-			    if (candidateMate[NLVer + Ghost2LocalMap[w]] == v) {
-			      option = 1;
-			      Mate[v - StartIndex] = w;     // v is a local vertex
-			      GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
-			      
-			    } // End of if CandidateMate[w] = v
-			  }     // End of if a Ghost Vertex
-			  else   { // w is a local vertex
-			    if (candidateMate[w - StartIndex] == v)  {
-			      option = 3;
-			      Mate[v - StartIndex] = w; // v is a local vertex
-			      Mate[w - StartIndex] = v; // w is a local vertex
-			      
+			    if ((w < StartIndex) || (w > EndIndex))  { // A ghost
 #ifdef PRINT_DEBUG_INFO_
-			      cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
-			      fflush(stdout);
+			      cout << "\n(" << myRank << ")Sending a request message:";
+			      cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
 #endif
-			    } // End of if(CandidateMate(w) = v
-			  }     // End of Else
-			}         // End of if(w >=0)
-			else
-			  option = 4; // End of Else: w == -1
-			// End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
-		      } // End of If (candidateMate[v-StartIndex] == u
+			      option = 2;
+			      
+			      if (candidateMate[NLVer + Ghost2LocalMap[w]] == v) {
+				option = 1;
+				Mate[v - StartIndex] = w;     // v is a local vertex
+				GMate[Ghost2LocalMap[w]] = v; // w is a ghost vertex
+				
+			      } // End of if CandidateMate[w] = v
+			    }     // End of if a Ghost Vertex
+			    else   { // w is a local vertex
+			      if (candidateMate[w - StartIndex] == v)  {
+				option = 3;
+				Mate[v - StartIndex] = w; // v is a local vertex
+				Mate[w - StartIndex] = v; // w is a local vertex
+#ifdef PRINT_DEBUG_INFO_
+				cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") ";
+				fflush(stdout);
+#endif
+			      } // End of if(CandidateMate(w) = v
+			    }     // End of Else
+			  }         // End of if(w >=0)
+			  else
+			    option = 4; // End of Else: w == -1
+			  // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+			} // End of If (candidateMate[v-StartIndex] == u
+		      }
 		    }     // End of task
 		  }         // mateval < 0
 		}             // End of if ( (v >= StartIndex) && (v <= EndIndex) ) //If Local Vertex:
 		else { // Neighbor is a ghost vertex
-		  
+
 #pragma omp critical
 		  {
 		    if (candidateMate[NLVer + Ghost2LocalMap[v]] == u)
@@ -162,7 +455,7 @@ void processMatchedVerticesAndSendMessages(
 		      option = 5; // u is local
 		  }                   // End of critical
 		}                       // End of Else //A Ghost Vertex
-		
+
 		switch (option)
 		  {
 		  case -1:
@@ -180,20 +473,20 @@ void processMatchedVerticesAndSendMessages(
 		    // Decrement the counter:
 		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], SPtr);
 		  case 2:
-		    
+
 		    // Found a dominating edge, it is a ghost
 		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
-		    
+
 		    // Build the Message Packet:
 		    // Message[0] = v;       // LOCAL
 		    // Message[1] = w;       // GHOST
 		    // Message[2] = REQUEST; // TYPE
 		    // Send a Request (Asynchronous)
 		    // MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
-		    
+
 		    (*msgActual)++;
 		    (*msgInd)++;
-		    
+
 		    privateQLocalVtx.push_back(v);
 		    privateQGhostVtx.push_back(w);
 		    privateQMsgType.push_back(REQUEST);
@@ -211,94 +504,82 @@ void processMatchedVerticesAndSendMessages(
 		    for (k1 = adj11; k1 < adj12; k1++) {
 		      w = verLocInd[k1];
 		      if ((w < StartIndex) || (w > EndIndex))  { // A ghost
-			
 #ifdef PRINT_DEBUG_INFO_
 			cout << "\n(" << myRank << ")Sending a failure message: ";
 			cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs);
 			fflush(stdout);
 #endif
-			
 			ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
-			
 			// Build the Message Packet:
 			// Message[0] = v;       // LOCAL
 			// Message[1] = w;       // GHOST
 			// Message[2] = FAILURE; // TYPE
 			// Send a Request (Asynchronous)
 			// MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
-			
+
 			(*msgActual)++;
 			(*msgInd)++;
-			
 			privateQLocalVtx.push_back(v);
 			privateQGhostVtx.push_back(w);
 			privateQMsgType.push_back(FAILURE);
 			privateQOwner.push_back(ghostOwner);
-			
+
 		      } // End of if(GHOST)
 		    }     // End of for loop
 		    break;
 		  case 5:
 		  default:
-		    
+
 #ifdef PRINT_DEBUG_INFO_
 		    cout << "\n(" << myRank << ")Sending a success message: ";
 		    cout << "\n(" << myRank << ")Ghost is " << v << " Owner is: " << findOwnerOfGhost(v, verDistance, myRank, numProcs) << "\n";
 		    fflush(stdout);
 #endif
-		    
+
 		    ghostOwner = findOwnerOfGhost(v, verDistance, myRank, numProcs);
-		    
 		    // Build the Message Packet:
 		    // Message[0] = u;       // LOCAL
 		    // Message[1] = v;       // GHOST
 		    // Message[2] = SUCCESS; // TYPE
 		    // Send a Request (Asynchronous)
 		    // MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
-		    
 		    (*msgActual)++;
 		    (*msgInd)++;
-		    
 		    privateQLocalVtx.push_back(u);
 		    privateQGhostVtx.push_back(v);
 		    privateQMsgType.push_back(SUCCESS);
 		    privateQOwner.push_back(ghostOwner);
-		    
+
 		    break;
 		  } // End of switch
 	      }     // End of inner for
 	    }
 	  } // End of outer for
-	  
+
 	  queuesTransfer(U, privateU, QLocalVtx,
 			 QGhostVtx,
 			 QMsgType, QOwner, privateQLocalVtx,
 			 privateQGhostVtx,
 			 privateQMsgType,
 			 privateQOwner);
-	  
+
         } // End of while ( !U.empty() )
-	
+
 #ifdef COUNT_LOCAL_VERTEX
         printf("Count local vertexes: %ld for thread %d of processor %d\n",
-               localVertices,
-               omp_get_thread_num(),
-               myRank);
-	
+               localVertices, mp_get_thread_num(), myRank);
 #endif
     } // End of parallel region
-    
+
     // Send the messages
 #ifdef DEBUG_HANG_
     cout << myRank<<" Sending: "<<QOwner.size()-initialSize<<" messages" <<endl;
 #endif
     for (int i = initialSize; i < QOwner.size(); i++)   {
-      
       Message[0] = QLocalVtx[i];
       Message[1] = QGhostVtx[i];
       Message[2] = QMsgType[i];
       ghostOwner = QOwner[i];
-      
       //MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
       //cout << myRank<<" Sending to "<<ghostOwner<<endl;
       MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);

amgprec/impl/aggregator/processMessages.cpp

@@ -2,7 +2,7 @@
 //#define DEBUG_HANG_
 #if !defined(SERIAL_MPI)
 
-void processMessages(
+void processMessagesD(
     MilanLongInt NLVer,
     MilanLongInt *Mate,
     MilanLongInt *candidateMate,
@@ -139,12 +139,12 @@ void processMessages(
       if (!ReceiveBuffer.empty())
 	ReceiveBuffer.clear();            // Empty it out first
       ReceiveBuffer.resize(bundleSize, -1); // Initialize
-      
+
       ReceiveBuffer[0] = Message[0]; // u
       ReceiveBuffer[1] = Message[1]; // v
       ReceiveBuffer[2] = Message[2]; // message_type
     }
-    
+
 #ifdef DEBUG_GHOST_
     if ((v < StartIndex) || (v > EndIndex)) {
       cout << "\n(" << myRank << ") From ReceiveBuffer: This should not happen: u= " << u << " v= " << v << " Type= " << message_type << " StartIndex " << StartIndex << " EndIndex " << EndIndex << endl;
@@ -161,7 +161,7 @@ void processMessages(
       u = ReceiveBuffer[bundleCounter - 3];            // GHOST
       v = ReceiveBuffer[bundleCounter - 2];            // LOCAL
       message_type = ReceiveBuffer[bundleCounter - 1]; // TYPE
-      
+
       // CASE I: REQUEST
       if (message_type == REQUEST)  {
 #ifdef PRINT_DEBUG_INFO_
@@ -189,7 +189,7 @@ void processMessages(
 	    cout << "\n(" << myRank << ")MATCH: (" << v << "," << u << ") " << endl;
 	    fflush(stdout);
 #endif
-	    
+
 	    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[u]], S);
 	  } // End of if ( candidateMate[v-StartIndex] == u )e
 	}     // End of if ( Mate[v] == -1 )
@@ -212,8 +212,9 @@ void processMessages(
 	    // Process only if not already matched ( v is local)
 	    if (candidateMate[v - StartIndex] == u)  {
 	      // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
-	      w = computeCandidateMate(verLocPtr[v - StartIndex], verLocPtr[v - StartIndex + 1], edgeLocWeight, k,
-				       verLocInd, StartIndex, EndIndex, GMate, Mate, Ghost2LocalMap);
+	      w = computeCandidateMateD(verLocPtr[v - StartIndex], verLocPtr[v - StartIndex + 1],
+					edgeLocWeight, k,verLocInd, StartIndex, EndIndex,
+					GMate, Mate, Ghost2LocalMap);
 	      candidateMate[v - StartIndex] = w;
 #ifdef PRINT_DEBUG_INFO_
 	      cout << "\n(" << myRank << ")" << v << " Points to: " << w << endl;
@@ -250,7 +251,7 @@ void processMessages(
 		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") " << endl;
 		    fflush(stdout);
 #endif
-		    
+
 		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], S);
 		  } // End of if CandidateMate[w] = v
 		}     // End of if a Ghost Vertex
@@ -311,7 +312,322 @@ void processMessages(
 	}                                                       // End of else: CASE III
       }                                                           // End of else: CASE I
     }
-    
+
+    return;
+}
+
+
+void processMessagesS(
+    MilanLongInt NLVer,
+    MilanLongInt *Mate,
+    MilanLongInt *candidateMate,
+    map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
+    vector<MilanLongInt> &GMate,
+    vector<MilanLongInt> &Counter,
+    MilanLongInt StartIndex,
+    MilanLongInt EndIndex,
+    MilanLongInt *myCard,
+    MilanLongInt *msgInd,
+    MilanLongInt *msgActual,
+    MilanFloat *edgeLocWeight,
+    MilanLongInt *verDistance,
+    MilanLongInt *verLocPtr,
+    MilanLongInt k,
+    MilanLongInt *verLocInd,
+    MilanInt numProcs,
+    MilanInt myRank,
+    MPI_Comm comm,
+    vector<MilanLongInt> &Message,
+    MilanLongInt numGhostEdges,
+    MilanLongInt u,
+    MilanLongInt v,
+    MilanLongInt *S,
+    vector<MilanLongInt> &U)
+{
+
+    //#define PRINT_DEBUG_INFO_
+
+    MilanInt Sender;
+    MPI_Status computeStatus;
+    MilanLongInt bundleSize, w;
+    MilanLongInt adj11, adj12, k1;
+    MilanLongInt ghostOwner;
+    int error_codeC;
+    error_codeC = MPI_Comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN);
+    char error_message[MPI_MAX_ERROR_STRING];
+    int message_length;
+    MilanLongInt message_type = 0;
+
+    // Buffer to receive bundled messages
+    // Maximum messages that can be received from any processor is
+    // twice the edge cut: REQUEST; REQUEST+(FAILURE/SUCCESS)
+    vector<MilanLongInt> ReceiveBuffer;
+    try
+    {
+        ReceiveBuffer.reserve(numGhostEdges * 2 * 3); // Three integers per cross edge
+    }
+    catch (length_error)
+    {
+        cout << "Error in function algoDistEdgeApproxDominatingEdgesMessageBundling: \n";
+        cout << "Not enough memory to allocate the internal variables \n";
+        exit(1);
+    }
+
+#ifdef PRINT_DEBUG_INFO_
+    cout
+        << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")About to begin Message processing phase ... *S=" << *S << endl;
+    fflush(stdout);
+#endif
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << "=========================************===============================" << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+    // BLOCKING RECEIVE:
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << " Waiting for blocking receive..." << endl;
+    fflush(stdout);
+    fflush(stdout);
+#endif
+
+    //cout << myRank<<" Receiving ...";
+    error_codeC = MPI_Recv(&Message[0], 3, TypeMap<MilanLongInt>(), MPI_ANY_SOURCE, ComputeTag, comm, &computeStatus);
+    if (error_codeC != MPI_SUCCESS)
+    {
+        MPI_Error_string(error_codeC, error_message, &message_length);
+        cout << "\n*Error in call to MPI_Receive on Slave: " << error_message << "\n";
+        fflush(stdout);
+    }
+    Sender = computeStatus.MPI_SOURCE;
+    //cout << " ...from "<<Sender << endl;
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")Received message from Process " << Sender << " Type= " << Message[2] << endl;
+    fflush(stdout);
+#endif
+
+    if (Message[2] == SIZEINFO)  {
+
+#ifdef PRINT_DEBUG_INFO_
+      cout << "\n(" << myRank << ")Received bundled message from Process " << Sender << " Size= " << Message[0] << endl;
+      fflush(stdout);
+#endif
+      bundleSize = Message[0]; //#of integers in the message
+      // Build the Message Buffer:
+      if (!ReceiveBuffer.empty())
+	ReceiveBuffer.clear();            // Empty it out first
+      ReceiveBuffer.resize(bundleSize, -1); // Initialize
+#ifdef PRINT_DEBUG_INFO_
+      cout << "\n(" << myRank << ")Message Bundle Before: " << endl;
+      for (int i = 0; i < bundleSize; i++)
+	cout << ReceiveBuffer[i] << ",";
+      cout << endl;
+      fflush(stdout);
+#endif
+      // Receive the message
+      //cout << myRank<<" Receiving from "<<Sender<<endl;
+      error_codeC = MPI_Recv(&ReceiveBuffer[0], bundleSize, TypeMap<MilanLongInt>(), Sender, BundleTag, comm, &computeStatus);
+      if (error_codeC != MPI_SUCCESS) {
+	MPI_Error_string(error_codeC, error_message, &message_length);
+	cout << "\n*Error in call to MPI_Receive on processor " << myRank << " Error: " << error_message << "\n";
+	fflush(stdout);
+      }
+#ifdef PRINT_DEBUG_INFO_
+      cout << "\n(" << myRank << ")Message Bundle After: " << endl;
+      for (int i = 0; i < bundleSize; i++)
+	cout << ReceiveBuffer[i] << ",";
+      cout << endl;
+      fflush(stdout);
+#endif
+    } else   { // Just a single message:
+#ifdef PRINT_DEBUG_INFO_
+      cout << "\n(" << myRank << ")Received regular message from Process " << Sender << " u= " << Message[0] << " v= " << Message[1] << endl;
+      fflush(stdout);
+#endif
+      // Add the current message to Queue:
+      bundleSize = 3; //#of integers in the message
+      // Build the Message Buffer:
+      if (!ReceiveBuffer.empty())
+	ReceiveBuffer.clear();            // Empty it out first
+      ReceiveBuffer.resize(bundleSize, -1); // Initialize
+
+      ReceiveBuffer[0] = Message[0]; // u
+      ReceiveBuffer[1] = Message[1]; // v
+      ReceiveBuffer[2] = Message[2]; // message_type
+    }
+
+#ifdef DEBUG_GHOST_
+    if ((v < StartIndex) || (v > EndIndex)) {
+      cout << "\n(" << myRank << ") From ReceiveBuffer: This should not happen: u= " << u << " v= " << v << " Type= " << message_type << " StartIndex " << StartIndex << " EndIndex " << EndIndex << endl;
+      fflush(stdout);
+    }
+#endif
+#ifdef PRINT_DEBUG_INFO_
+    cout << "\n(" << myRank << ")Processing message: u= " << u << " v= " << v << " Type= " << message_type << endl;
+    fflush(stdout);
+#endif
+
+    // Most of the time bundleSize == 3, thus, it's not worth parallelizing thi loop
+    for (MilanLongInt bundleCounter = 3; bundleCounter < bundleSize + 3; bundleCounter += 3)  {
+      u = ReceiveBuffer[bundleCounter - 3];            // GHOST
+      v = ReceiveBuffer[bundleCounter - 2];            // LOCAL
+      message_type = ReceiveBuffer[bundleCounter - 1]; // TYPE
+
+      // CASE I: REQUEST
+      if (message_type == REQUEST)  {
+#ifdef PRINT_DEBUG_INFO_
+	cout << "\n(" << myRank << ")Message type is REQUEST" << endl;
+	fflush(stdout);
+#endif
+#ifdef DEBUG_GHOST_
+	if ((v < 0) || (v < StartIndex) || ((v - StartIndex) > NLVer)) {
+	  cout << "\n(" << myRank << ") case 1 Bad address " << v << " " << StartIndex << " " << v - StartIndex << " " << NLVer << endl;
+	  fflush(stdout);
+	}
+
+#endif
+
+	if (Mate[v - StartIndex] == -1) {
+	  // Process only if not already matched  (v is local)
+	  candidateMate[NLVer + Ghost2LocalMap[u]] = v; // Set CandidateMate for the ghost
+	  if (candidateMate[v - StartIndex] == u)   {
+	    GMate[Ghost2LocalMap[u]] = v; // u is ghost
+	    Mate[v - StartIndex] = u;     // v is local
+	    U.push_back(v);
+	    U.push_back(u);
+	    (*myCard)++;
+#ifdef PRINT_DEBUG_INFO_
+	    cout << "\n(" << myRank << ")MATCH: (" << v << "," << u << ") " << endl;
+	    fflush(stdout);
+#endif
+
+	    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[u]], S);
+	  } // End of if ( candidateMate[v-StartIndex] == u )e
+	}     // End of if ( Mate[v] == -1 )
+      }         // End of REQUEST
+      else  { // CASE II: SUCCESS
+	if (message_type == SUCCESS) {
+#ifdef PRINT_DEBUG_INFO_
+	  cout << "\n(" << myRank << ")Message type is SUCCESS" << endl;
+	  fflush(stdout);
+#endif
+	  GMate[Ghost2LocalMap[u]] = EndIndex + 1; // Set a Dummy Mate to make sure that we do not (u is a ghost) process it again
+	  PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[u]], S);
+#ifdef DEBUG_GHOST_
+	  if ((v < 0) || (v < StartIndex) || ((v - StartIndex) > NLVer)) {
+	    cout << "\n(" << myRank << ") case 2  Bad address " << v << " " << StartIndex << " " << v - StartIndex << " " << NLVer << endl;
+	    fflush(stdout);
+	  }
+#endif
+	  if (Mate[v - StartIndex] == -1) {
+	    // Process only if not already matched ( v is local)
+	    if (candidateMate[v - StartIndex] == u)  {
+	      // Start: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+	      w = computeCandidateMateS(verLocPtr[v - StartIndex], verLocPtr[v - StartIndex + 1],
+					edgeLocWeight, k,verLocInd, StartIndex, EndIndex,
+					GMate, Mate, Ghost2LocalMap);
+	      candidateMate[v - StartIndex] = w;
+#ifdef PRINT_DEBUG_INFO_
+	      cout << "\n(" << myRank << ")" << v << " Points to: " << w << endl;
+	      fflush(stdout);
+#endif
+	      // If found a dominating edge:
+	      if (w >= 0)  {
+		if ((w < StartIndex) || (w > EndIndex))   {
+		  // w is a ghost
+		  // Build the Message Packet:
+		  Message[0] = v;       // LOCAL
+		  Message[1] = w;       // GHOST
+		  Message[2] = REQUEST; // TYPE
+		  // Send a Request (Asynchronous)
+#ifdef PRINT_DEBUG_INFO_
+		  cout << "\n(" << myRank << ")Sending a request message: ";
+		  cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs) << endl;
+		  fflush(stdout);
+#endif
+		  ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+		  //assert(ghostOwner != -1);
+		  //assert(ghostOwner != myRank);
+		  //cout << myRank<<" Sending to "<<ghostOwner<<endl;
+		  MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+		  (*msgInd)++;
+		  (*msgActual)++;
+		  if (candidateMate[NLVer + Ghost2LocalMap[w]] == v) {
+		    Mate[v - StartIndex] = w;     // v is local
+		    GMate[Ghost2LocalMap[w]] = v; // w is ghost
+		    U.push_back(v);
+		    U.push_back(w);
+		    (*myCard)++;
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") " << endl;
+		    fflush(stdout);
+#endif
+
+		    PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[w]], S);
+		  } // End of if CandidateMate[w] = v
+		}     // End of if a Ghost Vertex
+		else   { // w is a local vertex
+		  if (candidateMate[w - StartIndex] == v) {
+		    Mate[v - StartIndex] = w; // v is local
+		    Mate[w - StartIndex] = v; // w is local
+		    // Q.push_back(u);
+		    U.push_back(v);
+		    U.push_back(w);
+		    (*myCard)++;
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")MATCH: (" << v << "," << w << ") " << endl;
+		    fflush(stdout);
+#endif
+		  } // End of if(CandidateMate(w) = v
+		}     // End of Else
+	      }         // End of if(w >=0)
+	      else  { // No dominant edge found
+		adj11 = verLocPtr[v - StartIndex];
+		adj12 = verLocPtr[v - StartIndex + 1];
+		for (k1 = adj11; k1 < adj12; k1++) {
+		  w = verLocInd[k1];
+		  if ((w < StartIndex) || (w > EndIndex))  {
+		    // A ghost
+		    // Build the Message Packet:
+		    Message[0] = v;       // LOCAL
+		    Message[1] = w;       // GHOST
+		    Message[2] = FAILURE; // TYPE
+		    // Send a Request (Asynchronous)
+#ifdef PRINT_DEBUG_INFO_
+		    cout << "\n(" << myRank << ")Sending a failure message: ";
+		    cout << "\n(" << myRank << ")Ghost is " << w << " Owner is: " << findOwnerOfGhost(w, verDistance, myRank, numProcs) << endl;
+		    fflush(stdout);
+#endif
+		    ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
+		    //assert(ghostOwner != -1);
+		    //assert(ghostOwner != myRank);
+		    //cout << myRank<<" Sending to "<<ghostOwner<<endl;
+		    MPI_Bsend(&Message[0], 3, TypeMap<MilanLongInt>(), ghostOwner, ComputeTag, comm);
+		    (*msgInd)++;
+		    (*msgActual)++;
+		  } // End of if(GHOST)
+		}     // End of for loop
+	      }         // End of Else: w == -1
+                        // End:   PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
+	    } // End of if ( candidateMate[v-StartIndex] == u )
+	  }     // End of if ( Mate[v] == -1 )
+	}         // End of if ( message_type == SUCCESS )
+	else  {
+	  // CASE III: FAILURE
+#ifdef PRINT_DEBUG_INFO_
+	  cout << "\n(" << myRank << ")Message type is FAILURE" << endl;
+	  fflush(stdout);
+#endif
+	  GMate[Ghost2LocalMap[u]] = EndIndex + 1;            // Set a Dummy Mate to make sure that we do not (u is a ghost) process this anymore
+	  PROCESS_CROSS_EDGE(&Counter[Ghost2LocalMap[u]], S); // Decrease the counter
+	}                                                       // End of else: CASE III
+      }                                                           // End of else: CASE I
+    }
+
     return;
 }
 #endif

amgprec/impl/aggregator/queueTransfer.cpp

@@ -1,5 +1,4 @@
 #include "MatchBoxPC.h"
-#ifdef OPENMP
 void queuesTransfer(vector<MilanLongInt> &U,
                     vector<MilanLongInt> &privateU,
                     vector<MilanLongInt> &QLocalVtx,

amgprec/impl/aggregator/sendBundledMessages.cpp

@@ -1,24 +1,23 @@
 #include "MatchBoxPC.h"
-#if !defined(SERIAL_MPI)
 void sendBundledMessages(MilanLongInt *numGhostEdges,
-                                MilanInt *BufferSize,
-                                MilanLongInt *Buffer,
-                                vector<MilanLongInt> &PCumulative,
-                                vector<MilanLongInt> &PMessageBundle,
-                                vector<MilanLongInt> &PSizeInfoMessages,
-                                MilanLongInt *PCounter,
-                                MilanLongInt NumMessagesBundled,
-                                MilanLongInt *msgActual,
-                                MilanLongInt *msgInd,
-                                MilanInt numProcs,
-                                MilanInt myRank,
-                                MPI_Comm comm,
-                                vector<MilanLongInt> &QLocalVtx,
-                                vector<MilanLongInt> &QGhostVtx,
-                                vector<MilanLongInt> &QMsgType,
-                                vector<MilanInt> &QOwner,
-                                vector<MPI_Request> &SRequest,
-                                vector<MPI_Status> &SStatus)
+			 MilanInt *BufferSize,
+			 MilanLongInt *Buffer,
+			 vector<MilanLongInt> &PCumulative,
+			 vector<MilanLongInt> &PMessageBundle,
+			 vector<MilanLongInt> &PSizeInfoMessages,
+			 MilanLongInt *PCounter,
+			 MilanLongInt NumMessagesBundled,
+			 MilanLongInt *msgActual,
+			 MilanLongInt *msgInd,
+			 MilanInt numProcs,
+			 MilanInt myRank,
+			 MPI_Comm comm,
+			 vector<MilanLongInt> &QLocalVtx,
+			 vector<MilanLongInt> &QGhostVtx,
+			 vector<MilanLongInt> &QMsgType,
+			 vector<MilanInt> &QOwner,
+			 vector<MPI_Request> &SRequest,
+			 vector<MPI_Status> &SStatus)
 {
 
     MilanLongInt myIndex = 0, numMessagesToSend;
@@ -62,7 +61,7 @@ void sendBundledMessages(MilanLongInt *numGhostEdges,
 	      for (i = 0; i < numProcs; i++)
 		PCumulative[i + 1] = PCumulative[i] + PCounter[i];
 	    }
-	    
+
 #pragma omp task depend(inout \
                         : PCounter)
 	    {
@@ -84,7 +83,7 @@ void sendBundledMessages(MilanLongInt *numGhostEdges,
     PCounter[QOwner[i]]++;
   }
  }
- 
+
 // Send the Bundled Messages: Use ISend
 #pragma omp task depend(out \
                         : SRequest, SStatus)
@@ -101,7 +100,7 @@ void sendBundledMessages(MilanLongInt *numGhostEdges,
        exit(1);
      }
  }
- 
+
 // Send the Messages
 #pragma omp task depend(inout                                                  \
                         : SRequest, PSizeInfoMessages, PCumulative) depend(out \

amgprec/impl/amg_c_hierarchy_bld.F90

@@ -507,6 +507,71 @@ subroutine amg_c_hierarchy_bld(a,desc_a,prec,info)
   return
 
 contains
+
+#if ( __GNUC__ == 13 && __GNUC_MINOR__ == 3) 
+  ! gfortran 13.3.0 generates a strange error here with MOLD
+  ! moving to SOURCE but only for this version, since it's heavier
+  subroutine save_smoothers(level,save1, save2,info)
+    type(amg_c_onelev_type), intent(inout) :: level
+    class(amg_c_base_smoother_type), allocatable , intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0 
+    if (allocated(save1)) then
+      call save1%free(info)
+      if (info  == 0) deallocate(save1,stat=info)
+      if (info /= 0) return
+    end if
+    if (allocated(save2)) then
+      call save2%free(info)
+      if (info  == 0) deallocate(save2,stat=info)
+      if (info /= 0) return
+    end if
+    allocate(save1, source=level%sm,stat=info)
+    if (info == 0) call level%sm%clone_settings(save1,info)
+    if ((info == 0).and.allocated(level%sm2a)) then
+      allocate(save2, source=level%sm2a,stat=info) 
+      if (info == 0) call level%sm2a%clone_settings(save2,info)
+    end if
+
+    return
+  end subroutine save_smoothers
+
+  subroutine restore_smoothers(level,save1, save2,info)
+    type(amg_c_onelev_type), intent(inout), target :: level
+    class(amg_c_base_smoother_type), allocatable, intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0
+
+    if (allocated(level%sm)) then
+      if (info  == 0) call level%sm%free(info)
+      if (info  == 0) deallocate(level%sm,stat=info)
+    end if
+    if (allocated(save1)) then
+      if (info  == 0) allocate(level%sm,source=save1,stat=info)
+      if (info == 0) call save1%clone_settings(level%sm,info)
+    end if
+
+    if (info /= 0) return
+
+    if (allocated(level%sm2a)) then
+      if (info  == 0) call level%sm2a%free(info)
+      if (info  == 0) deallocate(level%sm2a,stat=info)
+    end if
+    if (allocated(save2)) then
+      if (info  == 0) allocate(level%sm2a,source=save2,stat=info)
+      if (info == 0) call save2%clone_settings(level%sm2a,info)
+      if (info  == 0) level%sm2 => level%sm2a
+    else
+      if (allocated(level%sm)) level%sm2 => level%sm
+    end if
+
+    return
+  end subroutine restore_smoothers
+
+#else
+
   subroutine save_smoothers(level,save1, save2,info)
     type(amg_c_onelev_type), intent(inout) :: level
     class(amg_c_base_smoother_type), allocatable , intent(inout) :: save1, save2
@@ -565,5 +630,5 @@ contains
 
     return
   end subroutine restore_smoothers
-
+#endif
 end subroutine amg_c_hierarchy_bld

amgprec/impl/amg_c_smoothers_bld.f90

@@ -186,8 +186,8 @@ subroutine amg_c_smoothers_bld(a,desc_a,prec,info,amold,vmold,imold)
              & ' but it has been changed to distributed.'
       end if
 
-    case(psb_ilu_n_, psb_ilu_t_,psb_milu_n_)
-      if (prec%precv(iszv)%sm%sv%get_id() /= psb_ilu_n_) then
+    case(amg_ilu_n_, amg_ilu_t_,amg_milu_n_)
+      if (prec%precv(iszv)%sm%sv%get_id() /= amg_ilu_n_) then
         write(psb_err_unit,*) &
              & 'AMG4PSBLAS: Warning: original coarse solver was requested as ',&
              & amg_fact_names(coarse_solve_id)

amgprec/impl/amg_ccprecset.F90

@@ -485,7 +485,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -499,7 +499,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -522,7 +522,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -530,21 +530,21 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
           case('ILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('ILUT')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('MILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
@@ -559,7 +559,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -582,7 +582,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -606,7 +606,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -714,7 +714,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -727,7 +727,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -750,7 +750,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -758,21 +758,21 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
         case('ILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('ILUT')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('MILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
@@ -787,7 +787,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -810,7 +810,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -834,7 +834,7 @@ subroutine amg_ccprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)

amgprec/impl/amg_cfile_prec_descr.f90

@@ -170,7 +170,7 @@ subroutine amg_cfile_prec_descr(prec,info,iout,root, verbosity,prefix)
             call prec%precv(1)%sm%descr(info,iout=iout_,prefix=prefix)
             nswps = prec%precv(1)%parms%sweeps_pre
           end if
-          write(iout_,*) trim(prefix_), '  Number of  sweeps : ',nswps
+          write(iout_,*) trim(prefix_), '  Number of  sweeps/degree : ',nswps
           write(iout_,*) trim(prefix_)
 
         else if (nlev > 1) then

amgprec/impl/amg_cfile_prec_memory_use.f90

@@ -0,0 +1,152 @@
+!   
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+! File: amg_dfile_prec_memory_use.f90
+!
+!
+! Subroutine: amg_file_prec_memory_use
+! Version: complex
+!
+!  This routine prints a memory_useiption of the preconditioner to the standard 
+!  output or to a file. It must be called after the preconditioner has been
+!  built by amg_precbld.
+!
+! Arguments:
+!  p       -  type(amg_Tprec_type), input.
+!             The preconditioner data structure to be printed out.
+!  info    -  integer, output.
+!             error code.
+!  iout    -  integer, input, optional.
+!             The id of the file where the preconditioner description
+!             will be printed. If iout is not present, then the standard
+!             output is condidered.
+!  root    -  integer, input, optional.
+!             The id of the process printing the message; -1 acts as a wildcard.
+!             Default is psb_root_
+!
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_cfile_prec_memory_use(prec,info,iout,root, verbosity,prefix,global)
+  use psb_base_mod
+  use amg_c_prec_mod, amg_protect_name => amg_cfile_prec_memory_use
+  use amg_c_inner_mod
+  use amg_c_gs_solver
+
+  implicit none 
+  ! Arguments
+  class(amg_cprec_type), intent(in)     :: prec
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  integer(psb_ipk_), intent(in), optional :: root
+  integer(psb_ipk_), intent(in), optional :: verbosity
+  character(len=*), intent(in), optional  :: prefix
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  integer(psb_ipk_)   :: ilev, nlev, ilmin, nswps
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: me, np
+  logical             :: is_symgs
+  character(len=20), parameter :: name='amg_file_prec_memory_use'
+  integer(psb_ipk_)  :: iout_, root_, verbosity_
+  logical            :: global_
+  character(1024)    :: prefix_
+
+  info = psb_success_
+  if (present(iout)) then 
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  end if
+  if (iout_ < 0) iout_ = psb_out_unit
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+
+  ctxt = prec%ctxt
+  call psb_info(ctxt,me,np)
+  prefix_ = ""
+  if (verbosity_ == 0) then 
+    if (present(prefix)) then
+      prefix_ = prefix
+    end if
+  else if (verbosity_ > 0) then 
+    if (present(prefix)) then
+      write(prefix_,'(a,a,i5,a)') prefix,' from process ',me,': '
+    else
+      write(prefix_,'(a,i5,a)') 'Process ',me,': '
+    end if
+
+  end if
+  
+  if (present(root)) then 
+    root_ = root
+  else
+    root_ = psb_root_
+  end if
+  if (root_ == -1) root_ = me
+
+  if (allocated(prec%precv)) then
+    if (verbosity_ >=0) then 
+      if (me == root_) then
+        write(iout_,*) 
+        write(iout_,'(a,1x,a)') trim(prefix_),'Preconditioner memory usage'
+      end if
+      nlev = size(prec%precv)
+      do ilev=1,nlev
+        call prec%precv(ilev)%memory_use(ilev,nlev,ilmin,info, &
+             & iout=iout_,verbosity=verbosity_,prefix=trim(prefix_),global=global)
+      end do
+
+    else
+      write(iout_,*) trim(name), &
+           & ': Error: no base preconditioner available, something is wrong!'
+      info = -2
+      return
+    endif
+  end if
+9998 continue
+end subroutine amg_cfile_prec_memory_use

amgprec/impl/amg_cprecinit.F90

@@ -98,6 +98,7 @@ subroutine amg_cprecinit(ctxt,prec,ptype,info)
   use amg_c_diag_solver
   use amg_c_ilu_solver
   use amg_c_gs_solver
+
 #if defined(HAVE_SLU_)
   use amg_c_slu_solver
 #endif
@@ -152,7 +153,6 @@ subroutine amg_cprecinit(ctxt,prec,ptype,info)
     if (info /= psb_success_) return
     allocate(amg_c_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
     call prec%precv(ilev_)%default()
-
   case ('L1-DIAG','L1-JACOBI','L1_DIAG','L1_JACOBI')
     nlev_ = 1
     ilev_ = 1

amgprec/impl/amg_d_hierarchy_bld.F90

@@ -507,6 +507,71 @@ subroutine amg_d_hierarchy_bld(a,desc_a,prec,info)
   return
 
 contains
+
+#if ( __GNUC__ == 13 && __GNUC_MINOR__ == 3) 
+  ! gfortran 13.3.0 generates a strange error here with MOLD
+  ! moving to SOURCE but only for this version, since it's heavier
+  subroutine save_smoothers(level,save1, save2,info)
+    type(amg_d_onelev_type), intent(inout) :: level
+    class(amg_d_base_smoother_type), allocatable , intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0 
+    if (allocated(save1)) then
+      call save1%free(info)
+      if (info  == 0) deallocate(save1,stat=info)
+      if (info /= 0) return
+    end if
+    if (allocated(save2)) then
+      call save2%free(info)
+      if (info  == 0) deallocate(save2,stat=info)
+      if (info /= 0) return
+    end if
+    allocate(save1, source=level%sm,stat=info)
+    if (info == 0) call level%sm%clone_settings(save1,info)
+    if ((info == 0).and.allocated(level%sm2a)) then
+      allocate(save2, source=level%sm2a,stat=info) 
+      if (info == 0) call level%sm2a%clone_settings(save2,info)
+    end if
+
+    return
+  end subroutine save_smoothers
+
+  subroutine restore_smoothers(level,save1, save2,info)
+    type(amg_d_onelev_type), intent(inout), target :: level
+    class(amg_d_base_smoother_type), allocatable, intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0
+
+    if (allocated(level%sm)) then
+      if (info  == 0) call level%sm%free(info)
+      if (info  == 0) deallocate(level%sm,stat=info)
+    end if
+    if (allocated(save1)) then
+      if (info  == 0) allocate(level%sm,source=save1,stat=info)
+      if (info == 0) call save1%clone_settings(level%sm,info)
+    end if
+
+    if (info /= 0) return
+
+    if (allocated(level%sm2a)) then
+      if (info  == 0) call level%sm2a%free(info)
+      if (info  == 0) deallocate(level%sm2a,stat=info)
+    end if
+    if (allocated(save2)) then
+      if (info  == 0) allocate(level%sm2a,source=save2,stat=info)
+      if (info == 0) call save2%clone_settings(level%sm2a,info)
+      if (info  == 0) level%sm2 => level%sm2a
+    else
+      if (allocated(level%sm)) level%sm2 => level%sm
+    end if
+
+    return
+  end subroutine restore_smoothers
+
+#else
+
   subroutine save_smoothers(level,save1, save2,info)
     type(amg_d_onelev_type), intent(inout) :: level
     class(amg_d_base_smoother_type), allocatable , intent(inout) :: save1, save2
@@ -565,5 +630,5 @@ contains
 
     return
   end subroutine restore_smoothers
-
+#endif
 end subroutine amg_d_hierarchy_bld

amgprec/impl/amg_d_smoothers_bld.f90

@@ -186,8 +186,8 @@ subroutine amg_d_smoothers_bld(a,desc_a,prec,info,amold,vmold,imold)
              & ' but it has been changed to distributed.'
       end if
 
-    case(psb_ilu_n_, psb_ilu_t_,psb_milu_n_)
-      if (prec%precv(iszv)%sm%sv%get_id() /= psb_ilu_n_) then
+    case(amg_ilu_n_, amg_ilu_t_,amg_milu_n_)
+      if (prec%precv(iszv)%sm%sv%get_id() /= amg_ilu_n_) then
         write(psb_err_unit,*) &
              & 'AMG4PSBLAS: Warning: original coarse solver was requested as ',&
              & amg_fact_names(coarse_solve_id)

amgprec/impl/amg_dcprecset.F90

@@ -499,7 +499,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -515,7 +515,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -538,7 +538,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -546,21 +546,21 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
           case('ILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('ILUT')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('MILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
@@ -575,7 +575,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -605,7 +605,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -643,7 +643,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -753,7 +753,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -768,7 +768,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -791,7 +791,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -799,21 +799,21 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
         case('ILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('ILUT')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('MILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
@@ -828,7 +828,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -858,7 +858,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -896,7 +896,7 @@ subroutine amg_dcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)

amgprec/impl/amg_dfile_prec_descr.f90

@@ -170,7 +170,7 @@ subroutine amg_dfile_prec_descr(prec,info,iout,root, verbosity,prefix)
             call prec%precv(1)%sm%descr(info,iout=iout_,prefix=prefix)
             nswps = prec%precv(1)%parms%sweeps_pre
           end if
-          write(iout_,*) trim(prefix_), '  Number of  sweeps : ',nswps
+          write(iout_,*) trim(prefix_), '  Number of  sweeps/degree : ',nswps
           write(iout_,*) trim(prefix_)
 
         else if (nlev > 1) then

amgprec/impl/amg_dfile_prec_memory_use.f90

@@ -0,0 +1,152 @@
+!   
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+! File: amg_dfile_prec_memory_use.f90
+!
+!
+! Subroutine: amg_file_prec_memory_use
+! Version: real
+!
+!  This routine prints a memory_useiption of the preconditioner to the standard 
+!  output or to a file. It must be called after the preconditioner has been
+!  built by amg_precbld.
+!
+! Arguments:
+!  p       -  type(amg_Tprec_type), input.
+!             The preconditioner data structure to be printed out.
+!  info    -  integer, output.
+!             error code.
+!  iout    -  integer, input, optional.
+!             The id of the file where the preconditioner description
+!             will be printed. If iout is not present, then the standard
+!             output is condidered.
+!  root    -  integer, input, optional.
+!             The id of the process printing the message; -1 acts as a wildcard.
+!             Default is psb_root_
+!
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_dfile_prec_memory_use(prec,info,iout,root, verbosity,prefix,global)
+  use psb_base_mod
+  use amg_d_prec_mod, amg_protect_name => amg_dfile_prec_memory_use
+  use amg_d_inner_mod
+  use amg_d_gs_solver
+
+  implicit none 
+  ! Arguments
+  class(amg_dprec_type), intent(in)     :: prec
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  integer(psb_ipk_), intent(in), optional :: root
+  integer(psb_ipk_), intent(in), optional :: verbosity
+  character(len=*), intent(in), optional  :: prefix
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  integer(psb_ipk_)   :: ilev, nlev, ilmin, nswps
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: me, np
+  logical             :: is_symgs
+  character(len=20), parameter :: name='amg_file_prec_memory_use'
+  integer(psb_ipk_)  :: iout_, root_, verbosity_
+  logical            :: global_
+  character(1024)    :: prefix_
+
+  info = psb_success_
+  if (present(iout)) then 
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  end if
+  if (iout_ < 0) iout_ = psb_out_unit
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+
+  ctxt = prec%ctxt
+  call psb_info(ctxt,me,np)
+  prefix_ = ""
+  if (verbosity_ == 0) then 
+    if (present(prefix)) then
+      prefix_ = prefix
+    end if
+  else if (verbosity_ > 0) then 
+    if (present(prefix)) then
+      write(prefix_,'(a,a,i5,a)') prefix,' from process ',me,': '
+    else
+      write(prefix_,'(a,i5,a)') 'Process ',me,': '
+    end if
+
+  end if
+  
+  if (present(root)) then 
+    root_ = root
+  else
+    root_ = psb_root_
+  end if
+  if (root_ == -1) root_ = me
+
+  if (allocated(prec%precv)) then
+    if (verbosity_ >=0) then 
+      if (me == root_) then
+        write(iout_,*) 
+        write(iout_,'(a,1x,a)') trim(prefix_),'Preconditioner memory usage'
+      end if
+      nlev = size(prec%precv)
+      do ilev=1,nlev
+        call prec%precv(ilev)%memory_use(ilev,nlev,ilmin,info, &
+             & iout=iout_,verbosity=verbosity_,prefix=trim(prefix_),global=global)
+      end do
+
+    else
+      write(iout_,*) trim(name), &
+           & ': Error: no base preconditioner available, something is wrong!'
+      info = -2
+      return
+    endif
+  end if
+9998 continue
+end subroutine amg_dfile_prec_memory_use

amgprec/impl/amg_dprecinit.F90

@@ -98,6 +98,8 @@ subroutine amg_dprecinit(ctxt,prec,ptype,info)
   use amg_d_diag_solver
   use amg_d_ilu_solver
   use amg_d_gs_solver
+  use amg_d_poly_smoother
+
 #if defined(HAVE_UMF_)
   use amg_d_umf_solver
 #endif
@@ -155,7 +157,14 @@ subroutine amg_dprecinit(ctxt,prec,ptype,info)
     if (info /= psb_success_) return
     allocate(amg_d_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
     call prec%precv(ilev_)%default()
-
+  case ('POLY')
+    nlev_ = 1
+    ilev_ = 1
+    allocate(prec%precv(nlev_),stat=info)
+    allocate(amg_d_poly_smoother_type :: prec%precv(ilev_)%sm, stat=info)
+    if (info /= psb_success_) return
+    allocate(amg_d_l1_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
+    call prec%precv(ilev_)%default()
   case ('L1-DIAG','L1-JACOBI','L1_DIAG','L1_JACOBI')
     nlev_ = 1
     ilev_ = 1

amgprec/impl/amg_s_hierarchy_bld.F90

@@ -507,6 +507,71 @@ subroutine amg_s_hierarchy_bld(a,desc_a,prec,info)
   return
 
 contains
+
+#if ( __GNUC__ == 13 && __GNUC_MINOR__ == 3) 
+  ! gfortran 13.3.0 generates a strange error here with MOLD
+  ! moving to SOURCE but only for this version, since it's heavier
+  subroutine save_smoothers(level,save1, save2,info)
+    type(amg_s_onelev_type), intent(inout) :: level
+    class(amg_s_base_smoother_type), allocatable , intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0 
+    if (allocated(save1)) then
+      call save1%free(info)
+      if (info  == 0) deallocate(save1,stat=info)
+      if (info /= 0) return
+    end if
+    if (allocated(save2)) then
+      call save2%free(info)
+      if (info  == 0) deallocate(save2,stat=info)
+      if (info /= 0) return
+    end if
+    allocate(save1, source=level%sm,stat=info)
+    if (info == 0) call level%sm%clone_settings(save1,info)
+    if ((info == 0).and.allocated(level%sm2a)) then
+      allocate(save2, source=level%sm2a,stat=info) 
+      if (info == 0) call level%sm2a%clone_settings(save2,info)
+    end if
+
+    return
+  end subroutine save_smoothers
+
+  subroutine restore_smoothers(level,save1, save2,info)
+    type(amg_s_onelev_type), intent(inout), target :: level
+    class(amg_s_base_smoother_type), allocatable, intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0
+
+    if (allocated(level%sm)) then
+      if (info  == 0) call level%sm%free(info)
+      if (info  == 0) deallocate(level%sm,stat=info)
+    end if
+    if (allocated(save1)) then
+      if (info  == 0) allocate(level%sm,source=save1,stat=info)
+      if (info == 0) call save1%clone_settings(level%sm,info)
+    end if
+
+    if (info /= 0) return
+
+    if (allocated(level%sm2a)) then
+      if (info  == 0) call level%sm2a%free(info)
+      if (info  == 0) deallocate(level%sm2a,stat=info)
+    end if
+    if (allocated(save2)) then
+      if (info  == 0) allocate(level%sm2a,source=save2,stat=info)
+      if (info == 0) call save2%clone_settings(level%sm2a,info)
+      if (info  == 0) level%sm2 => level%sm2a
+    else
+      if (allocated(level%sm)) level%sm2 => level%sm
+    end if
+
+    return
+  end subroutine restore_smoothers
+
+#else
+
   subroutine save_smoothers(level,save1, save2,info)
     type(amg_s_onelev_type), intent(inout) :: level
     class(amg_s_base_smoother_type), allocatable , intent(inout) :: save1, save2
@@ -565,5 +630,5 @@ contains
 
     return
   end subroutine restore_smoothers
-
+#endif
 end subroutine amg_s_hierarchy_bld

amgprec/impl/amg_s_smoothers_bld.f90

@@ -186,8 +186,8 @@ subroutine amg_s_smoothers_bld(a,desc_a,prec,info,amold,vmold,imold)
              & ' but it has been changed to distributed.'
       end if
 
-    case(psb_ilu_n_, psb_ilu_t_,psb_milu_n_)
-      if (prec%precv(iszv)%sm%sv%get_id() /= psb_ilu_n_) then
+    case(amg_ilu_n_, amg_ilu_t_,amg_milu_n_)
+      if (prec%precv(iszv)%sm%sv%get_id() /= amg_ilu_n_) then
         write(psb_err_unit,*) &
              & 'AMG4PSBLAS: Warning: original coarse solver was requested as ',&
              & amg_fact_names(coarse_solve_id)

amgprec/impl/amg_scprecset.F90

@@ -485,7 +485,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -499,7 +499,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -522,7 +522,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -530,21 +530,21 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
           case('ILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('ILUT')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('MILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
@@ -559,7 +559,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -582,7 +582,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -606,7 +606,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -714,7 +714,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -727,7 +727,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -750,7 +750,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -758,21 +758,21 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
         case('ILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('ILUT')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('MILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
@@ -787,7 +787,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -810,7 +810,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -834,7 +834,7 @@ subroutine amg_scprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)

amgprec/impl/amg_sfile_prec_descr.f90

@@ -170,7 +170,7 @@ subroutine amg_sfile_prec_descr(prec,info,iout,root, verbosity,prefix)
             call prec%precv(1)%sm%descr(info,iout=iout_,prefix=prefix)
             nswps = prec%precv(1)%parms%sweeps_pre
           end if
-          write(iout_,*) trim(prefix_), '  Number of  sweeps : ',nswps
+          write(iout_,*) trim(prefix_), '  Number of  sweeps/degree : ',nswps
           write(iout_,*) trim(prefix_)
 
         else if (nlev > 1) then

amgprec/impl/amg_sfile_prec_memory_use.f90

@@ -0,0 +1,152 @@
+!   
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+! File: amg_dfile_prec_memory_use.f90
+!
+!
+! Subroutine: amg_file_prec_memory_use
+! Version: real
+!
+!  This routine prints a memory_useiption of the preconditioner to the standard 
+!  output or to a file. It must be called after the preconditioner has been
+!  built by amg_precbld.
+!
+! Arguments:
+!  p       -  type(amg_Tprec_type), input.
+!             The preconditioner data structure to be printed out.
+!  info    -  integer, output.
+!             error code.
+!  iout    -  integer, input, optional.
+!             The id of the file where the preconditioner description
+!             will be printed. If iout is not present, then the standard
+!             output is condidered.
+!  root    -  integer, input, optional.
+!             The id of the process printing the message; -1 acts as a wildcard.
+!             Default is psb_root_
+!
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_sfile_prec_memory_use(prec,info,iout,root, verbosity,prefix,global)
+  use psb_base_mod
+  use amg_s_prec_mod, amg_protect_name => amg_sfile_prec_memory_use
+  use amg_s_inner_mod
+  use amg_s_gs_solver
+
+  implicit none 
+  ! Arguments
+  class(amg_sprec_type), intent(in)     :: prec
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  integer(psb_ipk_), intent(in), optional :: root
+  integer(psb_ipk_), intent(in), optional :: verbosity
+  character(len=*), intent(in), optional  :: prefix
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  integer(psb_ipk_)   :: ilev, nlev, ilmin, nswps
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: me, np
+  logical             :: is_symgs
+  character(len=20), parameter :: name='amg_file_prec_memory_use'
+  integer(psb_ipk_)  :: iout_, root_, verbosity_
+  logical            :: global_
+  character(1024)    :: prefix_
+
+  info = psb_success_
+  if (present(iout)) then 
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  end if
+  if (iout_ < 0) iout_ = psb_out_unit
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+
+  ctxt = prec%ctxt
+  call psb_info(ctxt,me,np)
+  prefix_ = ""
+  if (verbosity_ == 0) then 
+    if (present(prefix)) then
+      prefix_ = prefix
+    end if
+  else if (verbosity_ > 0) then 
+    if (present(prefix)) then
+      write(prefix_,'(a,a,i5,a)') prefix,' from process ',me,': '
+    else
+      write(prefix_,'(a,i5,a)') 'Process ',me,': '
+    end if
+
+  end if
+  
+  if (present(root)) then 
+    root_ = root
+  else
+    root_ = psb_root_
+  end if
+  if (root_ == -1) root_ = me
+
+  if (allocated(prec%precv)) then
+    if (verbosity_ >=0) then 
+      if (me == root_) then
+        write(iout_,*) 
+        write(iout_,'(a,1x,a)') trim(prefix_),'Preconditioner memory usage'
+      end if
+      nlev = size(prec%precv)
+      do ilev=1,nlev
+        call prec%precv(ilev)%memory_use(ilev,nlev,ilmin,info, &
+             & iout=iout_,verbosity=verbosity_,prefix=trim(prefix_),global=global)
+      end do
+
+    else
+      write(iout_,*) trim(name), &
+           & ': Error: no base preconditioner available, something is wrong!'
+      info = -2
+      return
+    endif
+  end if
+9998 continue
+end subroutine amg_sfile_prec_memory_use

amgprec/impl/amg_sprecinit.F90

@@ -98,6 +98,8 @@ subroutine amg_sprecinit(ctxt,prec,ptype,info)
   use amg_s_diag_solver
   use amg_s_ilu_solver
   use amg_s_gs_solver
+  use amg_s_poly_smoother
+
 #if defined(HAVE_SLU_)
   use amg_s_slu_solver
 #endif
@@ -152,7 +154,14 @@ subroutine amg_sprecinit(ctxt,prec,ptype,info)
     if (info /= psb_success_) return
     allocate(amg_s_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
     call prec%precv(ilev_)%default()
-
+  case ('POLY')
+    nlev_ = 1
+    ilev_ = 1
+    allocate(prec%precv(nlev_),stat=info)
+    allocate(amg_s_poly_smoother_type :: prec%precv(ilev_)%sm, stat=info)
+    if (info /= psb_success_) return
+    allocate(amg_s_l1_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
+    call prec%precv(ilev_)%default()
   case ('L1-DIAG','L1-JACOBI','L1_DIAG','L1_JACOBI')
     nlev_ = 1
     ilev_ = 1

amgprec/impl/amg_z_hierarchy_bld.F90

@@ -507,6 +507,71 @@ subroutine amg_z_hierarchy_bld(a,desc_a,prec,info)
   return
 
 contains
+
+#if ( __GNUC__ == 13 && __GNUC_MINOR__ == 3) 
+  ! gfortran 13.3.0 generates a strange error here with MOLD
+  ! moving to SOURCE but only for this version, since it's heavier
+  subroutine save_smoothers(level,save1, save2,info)
+    type(amg_z_onelev_type), intent(inout) :: level
+    class(amg_z_base_smoother_type), allocatable , intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0 
+    if (allocated(save1)) then
+      call save1%free(info)
+      if (info  == 0) deallocate(save1,stat=info)
+      if (info /= 0) return
+    end if
+    if (allocated(save2)) then
+      call save2%free(info)
+      if (info  == 0) deallocate(save2,stat=info)
+      if (info /= 0) return
+    end if
+    allocate(save1, source=level%sm,stat=info)
+    if (info == 0) call level%sm%clone_settings(save1,info)
+    if ((info == 0).and.allocated(level%sm2a)) then
+      allocate(save2, source=level%sm2a,stat=info) 
+      if (info == 0) call level%sm2a%clone_settings(save2,info)
+    end if
+
+    return
+  end subroutine save_smoothers
+
+  subroutine restore_smoothers(level,save1, save2,info)
+    type(amg_z_onelev_type), intent(inout), target :: level
+    class(amg_z_base_smoother_type), allocatable, intent(inout) :: save1, save2
+    integer(psb_ipk_), intent(out) :: info
+
+    info  = 0
+
+    if (allocated(level%sm)) then
+      if (info  == 0) call level%sm%free(info)
+      if (info  == 0) deallocate(level%sm,stat=info)
+    end if
+    if (allocated(save1)) then
+      if (info  == 0) allocate(level%sm,source=save1,stat=info)
+      if (info == 0) call save1%clone_settings(level%sm,info)
+    end if
+
+    if (info /= 0) return
+
+    if (allocated(level%sm2a)) then
+      if (info  == 0) call level%sm2a%free(info)
+      if (info  == 0) deallocate(level%sm2a,stat=info)
+    end if
+    if (allocated(save2)) then
+      if (info  == 0) allocate(level%sm2a,source=save2,stat=info)
+      if (info == 0) call save2%clone_settings(level%sm2a,info)
+      if (info  == 0) level%sm2 => level%sm2a
+    else
+      if (allocated(level%sm)) level%sm2 => level%sm
+    end if
+
+    return
+  end subroutine restore_smoothers
+
+#else
+
   subroutine save_smoothers(level,save1, save2,info)
     type(amg_z_onelev_type), intent(inout) :: level
     class(amg_z_base_smoother_type), allocatable , intent(inout) :: save1, save2
@@ -565,5 +630,5 @@ contains
 
     return
   end subroutine restore_smoothers
-
+#endif
 end subroutine amg_z_hierarchy_bld

amgprec/impl/amg_z_smoothers_bld.f90

@@ -186,8 +186,8 @@ subroutine amg_z_smoothers_bld(a,desc_a,prec,info,amold,vmold,imold)
              & ' but it has been changed to distributed.'
       end if
 
-    case(psb_ilu_n_, psb_ilu_t_,psb_milu_n_)
-      if (prec%precv(iszv)%sm%sv%get_id() /= psb_ilu_n_) then
+    case(amg_ilu_n_, amg_ilu_t_,amg_milu_n_)
+      if (prec%precv(iszv)%sm%sv%get_id() /= amg_ilu_n_) then
         write(psb_err_unit,*) &
              & 'AMG4PSBLAS: Warning: original coarse solver was requested as ',&
              & amg_fact_names(coarse_solve_id)

amgprec/impl/amg_zcprecset.F90

@@ -499,7 +499,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -515,7 +515,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
             call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -538,7 +538,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -546,21 +546,21 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
           case('ILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('ILUT')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
             call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
           case('MILU')
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_repl_mat_)
@@ -575,7 +575,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -605,7 +605,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -643,7 +643,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
             call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
             call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-            call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+            call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
             if (hier_asb) &
                  & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                  & amg_distr_mat_)
@@ -753,7 +753,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -768,7 +768,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #elif defined(HAVE_MUMPS_)
           call p%precv(nlev_)%set('SUB_SOLVE',amg_mumps_,info,pos=pos)
 #else
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
 #endif
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
@@ -791,7 +791,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -799,21 +799,21 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
 #endif
         case('ILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('ILUT')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_t_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_t_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
           call p%precv(nlev_)%set('COARSE_MAT',amg_repl_mat_,info,pos=pos)
         case('MILU')
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_milu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_milu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_repl_mat_)
@@ -828,7 +828,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -858,7 +858,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)
@@ -896,7 +896,7 @@ subroutine amg_zcprecsetc(p,what,string,info,ilev,ilmax,pos,idx)
           call p%precv(nlev_)%set('COARSE_MAT',amg_distr_mat_,info,pos=pos)
 #else
           call p%precv(nlev_)%set('SMOOTHER_TYPE',amg_bjac_,info,pos=pos)
-          call p%precv(nlev_)%set('SUB_SOLVE',psb_ilu_n_,info,pos=pos)
+          call p%precv(nlev_)%set('SUB_SOLVE',amg_ilu_n_,info,pos=pos)
           if (hier_asb) &
                & call amg_warn_coarse_mat(p%precv(nlev_)%parms%get_coarse_mat(),&
                & amg_distr_mat_)

amgprec/impl/amg_zfile_prec_descr.f90

@@ -170,7 +170,7 @@ subroutine amg_zfile_prec_descr(prec,info,iout,root, verbosity,prefix)
             call prec%precv(1)%sm%descr(info,iout=iout_,prefix=prefix)
             nswps = prec%precv(1)%parms%sweeps_pre
           end if
-          write(iout_,*) trim(prefix_), '  Number of  sweeps : ',nswps
+          write(iout_,*) trim(prefix_), '  Number of  sweeps/degree : ',nswps
           write(iout_,*) trim(prefix_)
 
         else if (nlev > 1) then

amgprec/impl/amg_zfile_prec_memory_use.f90

@@ -0,0 +1,152 @@
+!   
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+! File: amg_dfile_prec_memory_use.f90
+!
+!
+! Subroutine: amg_file_prec_memory_use
+! Version: complex
+!
+!  This routine prints a memory_useiption of the preconditioner to the standard 
+!  output or to a file. It must be called after the preconditioner has been
+!  built by amg_precbld.
+!
+! Arguments:
+!  p       -  type(amg_Tprec_type), input.
+!             The preconditioner data structure to be printed out.
+!  info    -  integer, output.
+!             error code.
+!  iout    -  integer, input, optional.
+!             The id of the file where the preconditioner description
+!             will be printed. If iout is not present, then the standard
+!             output is condidered.
+!  root    -  integer, input, optional.
+!             The id of the process printing the message; -1 acts as a wildcard.
+!             Default is psb_root_
+!
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_zfile_prec_memory_use(prec,info,iout,root, verbosity,prefix,global)
+  use psb_base_mod
+  use amg_z_prec_mod, amg_protect_name => amg_zfile_prec_memory_use
+  use amg_z_inner_mod
+  use amg_z_gs_solver
+
+  implicit none 
+  ! Arguments
+  class(amg_zprec_type), intent(in)     :: prec
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  integer(psb_ipk_), intent(in), optional :: root
+  integer(psb_ipk_), intent(in), optional :: verbosity
+  character(len=*), intent(in), optional  :: prefix
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  integer(psb_ipk_)   :: ilev, nlev, ilmin, nswps
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: me, np
+  logical             :: is_symgs
+  character(len=20), parameter :: name='amg_file_prec_memory_use'
+  integer(psb_ipk_)  :: iout_, root_, verbosity_
+  logical            :: global_
+  character(1024)    :: prefix_
+
+  info = psb_success_
+  if (present(iout)) then 
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  end if
+  if (iout_ < 0) iout_ = psb_out_unit
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+
+  ctxt = prec%ctxt
+  call psb_info(ctxt,me,np)
+  prefix_ = ""
+  if (verbosity_ == 0) then 
+    if (present(prefix)) then
+      prefix_ = prefix
+    end if
+  else if (verbosity_ > 0) then 
+    if (present(prefix)) then
+      write(prefix_,'(a,a,i5,a)') prefix,' from process ',me,': '
+    else
+      write(prefix_,'(a,i5,a)') 'Process ',me,': '
+    end if
+
+  end if
+  
+  if (present(root)) then 
+    root_ = root
+  else
+    root_ = psb_root_
+  end if
+  if (root_ == -1) root_ = me
+
+  if (allocated(prec%precv)) then
+    if (verbosity_ >=0) then 
+      if (me == root_) then
+        write(iout_,*) 
+        write(iout_,'(a,1x,a)') trim(prefix_),'Preconditioner memory usage'
+      end if
+      nlev = size(prec%precv)
+      do ilev=1,nlev
+        call prec%precv(ilev)%memory_use(ilev,nlev,ilmin,info, &
+             & iout=iout_,verbosity=verbosity_,prefix=trim(prefix_),global=global)
+      end do
+
+    else
+      write(iout_,*) trim(name), &
+           & ': Error: no base preconditioner available, something is wrong!'
+      info = -2
+      return
+    endif
+  end if
+9998 continue
+end subroutine amg_zfile_prec_memory_use

amgprec/impl/amg_zprecinit.F90

@@ -98,6 +98,7 @@ subroutine amg_zprecinit(ctxt,prec,ptype,info)
   use amg_z_diag_solver
   use amg_z_ilu_solver
   use amg_z_gs_solver
+
 #if defined(HAVE_UMF_)
   use amg_z_umf_solver
 #endif
@@ -155,7 +156,6 @@ subroutine amg_zprecinit(ctxt,prec,ptype,info)
     if (info /= psb_success_) return
     allocate(amg_z_diag_solver_type :: prec%precv(ilev_)%sm%sv, stat=info)
     call prec%precv(ilev_)%default()
-
   case ('L1-DIAG','L1-JACOBI','L1_DIAG','L1_JACOBI')
     nlev_ = 1
     ilev_ = 1

amgprec/impl/level/Makefile

@@ -15,6 +15,7 @@ amg_c_base_onelev_csetc.o \
 amg_c_base_onelev_cseti.o \
 amg_c_base_onelev_csetr.o \
 amg_c_base_onelev_descr.o \
+amg_c_base_onelev_memory_use.o \
 amg_c_base_onelev_dump.o \
 amg_c_base_onelev_free.o \
 amg_c_base_onelev_free_smoothers.o \
@@ -31,6 +32,7 @@ amg_d_base_onelev_csetc.o \
 amg_d_base_onelev_cseti.o \
 amg_d_base_onelev_csetr.o \
 amg_d_base_onelev_descr.o \
+amg_d_base_onelev_memory_use.o \
 amg_d_base_onelev_dump.o \
 amg_d_base_onelev_free.o \
 amg_d_base_onelev_free_smoothers.o \
@@ -47,6 +49,7 @@ amg_s_base_onelev_csetc.o \
 amg_s_base_onelev_cseti.o \
 amg_s_base_onelev_csetr.o \
 amg_s_base_onelev_descr.o \
+amg_s_base_onelev_memory_use.o \
 amg_s_base_onelev_dump.o \
 amg_s_base_onelev_free.o \
 amg_s_base_onelev_free_smoothers.o \
@@ -63,6 +66,7 @@ amg_z_base_onelev_csetc.o \
 amg_z_base_onelev_cseti.o \
 amg_z_base_onelev_csetr.o \
 amg_z_base_onelev_descr.o \
+amg_z_base_onelev_memory_use.o \
 amg_z_base_onelev_dump.o \
 amg_z_base_onelev_free.o \
 amg_z_base_onelev_free_smoothers.o \

amgprec/impl/level/amg_c_base_onelev_csetc.F90

@@ -43,7 +43,6 @@ subroutine amg_c_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_c_dec_aggregator_mod
   use amg_c_symdec_aggregator_mod
 #if !defined(SERIAL_MPI)
-#endif
   use amg_c_jac_smoother
   use amg_c_as_smoother
   use amg_c_diag_solver
@@ -190,16 +189,11 @@ subroutine amg_c_base_onelev_csetc(lv,what,val,info,pos,idx)
     case ('NONE','NOPREC','FACT_NONE')
       call lv%set(amg_c_id_solver_mold,info,pos=pos)
 
-    case ('DIAG')
+    case ('DIAG','JACOBI')
       call lv%set(amg_c_diag_solver_mold,info,pos=pos)
 
-    case ('JACOBI')
-      call lv%set(amg_c_jac_solver_mold,info,pos=pos)
-
-    case ('L1-DIAG')
+    case ('L1-DIAG','L1-JACOBI')
       call lv%set(amg_c_l1_diag_solver_mold,info,pos=pos)
-    case ('L1-JACOBI')
-      call lv%set(amg_c_l1_jac_solver_mold,info,pos=pos)
 
     case ('GS','FGS','FWGS')
       call lv%set(amg_c_gs_solver_mold,info,pos=pos)

amgprec/impl/level/amg_c_base_onelev_cseti.F90

@@ -164,7 +164,7 @@ subroutine amg_c_base_onelev_cseti(lv,what,val,info,pos,idx)
     case (amg_bwgs_)
       call lv%set(amg_c_bwgs_solver_mold,info,pos=pos)
 
-    case (psb_ilu_n_,psb_milu_n_,psb_ilu_t_)
+    case (amg_ilu_n_,amg_milu_n_,amg_ilu_t_)
       call lv%set(amg_c_ilu_solver_mold,info,pos=pos)
       if (info == 0) then
         if ((ipos_==amg_smooth_pre_) .or.(ipos_==amg_smooth_both_)) then

amgprec/impl/level/amg_c_base_onelev_memory_use.f90

@@ -0,0 +1,150 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_c_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefix,global)
+
+  use psb_base_mod
+  use amg_c_onelev_mod, amg_protect_name => amg_c_base_onelev_memory_use
+  Implicit None
+  ! Arguments
+  class(amg_c_onelev_type), intent(in)    :: lv
+  integer(psb_ipk_), intent(in)           :: il,nl,ilmin
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  character(len=*), intent(in), optional  :: prefix
+      integer(psb_ipk_), intent(in), optional :: verbosity
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)  :: err_act ,me, np
+  character(len=20), parameter :: name='amg_c_base_onelev_memory_use'
+  integer(psb_ipk_)  :: iout_, verbosity_
+  logical      :: coarse, global_
+  character(1024)    :: prefix_
+  integer(psb_epk_), allocatable  :: sz(:)
+
+
+  call psb_erractionsave(err_act)
+
+  ctxt = lv%base_desc%get_ctxt()
+  call psb_info(ctxt,me,np)
+  
+  coarse = (il==nl)
+
+  if (present(iout)) then 
+    iout_ = iout
+  else 
+    iout_ = psb_out_unit
+  end if
+
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+  if (present(global)) then
+    global_ = global
+  else
+    global_ = .true.
+  end if
+
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  if ((me == 0).or.(verbosity_>0)) write(iout_,*) trim(prefix_)
+
+  if (global_) then
+    allocate(sz(6))
+    sz(:) = 0 
+    sz(1) =  lv%base_a%sizeof()
+    sz(2) =  lv%base_desc%sizeof()
+    if (il >1)              sz(3) =  lv%linmap%sizeof()
+    if (allocated(lv%sm))   sz(4) =  lv%sm%sizeof()
+    if (allocated(lv%sm2a)) sz(5) =  lv%sm2a%sizeof()
+    if (allocated(lv%wrk))  sz(6) =  lv%wrk%sizeof()
+    call psb_sum(ctxt,sz)
+    if (me == 0) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:',                         sz(1) 
+      write(iout_,*) trim(prefix_), '       Descriptor:',                         sz(2) 
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:',              sz(3) 
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:',   sz(4) 
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', sz(5) 
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:',  sz(6)
+    end if
+    
+  else
+    if ((me == 0).or.(verbosity_>0)) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:', lv%base_a%sizeof()
+      write(iout_,*) trim(prefix_), '       Descriptor:', lv%base_desc%sizeof()
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:', lv%linmap%sizeof()
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:', lv%sm%sizeof()
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', lv%sm2a%sizeof()
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:', lv%wrk%sizeof()
+    end if
+  endif
+
+9998 continue
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+
+end subroutine amg_c_base_onelev_memory_use

amgprec/impl/level/amg_d_base_onelev_csetc.F90

@@ -44,7 +44,7 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_d_symdec_aggregator_mod
 #if !defined(SERIAL_MPI)
   use amg_d_parmatch_aggregator_mod
-#endif
+  use amg_d_poly_smoother
   use amg_d_jac_smoother
   use amg_d_as_smoother
   use amg_d_diag_solver
@@ -97,6 +97,7 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
   type(amg_d_ainv_solver_type)     ::  amg_d_ainv_solver_mold
   type(amg_d_invk_solver_type)     ::  amg_d_invk_solver_mold
   type(amg_d_invt_solver_type)     ::  amg_d_invt_solver_mold
+  type(amg_d_poly_smoother_type)   ::  amg_d_poly_smoother_mold
 #if defined(HAVE_UMF_)
   type(amg_d_umf_solver_type)     ::  amg_d_umf_solver_mold
 #endif
@@ -158,6 +159,9 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
       call lv%set(amg_d_as_smoother_mold,info,pos=pos)
       if (info == 0) call lv%set(amg_d_ilu_solver_mold,info,pos=pos)
 
+    case ('POLY')
+      call lv%set(amg_d_poly_smoother_mold,info,pos=pos)
+      if (info == 0) call lv%set(amg_d_l1_diag_solver_mold,info,pos=pos)
     case ('GS','FWGS')
       call lv%set(amg_d_jac_smoother_mold,info,pos='pre')
       if (info == 0) call lv%set(amg_d_gs_solver_mold,info,pos='pre')
@@ -203,16 +207,11 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
     case ('NONE','NOPREC','FACT_NONE')
       call lv%set(amg_d_id_solver_mold,info,pos=pos)
 
-    case ('DIAG')
+    case ('DIAG','JACOBI')
       call lv%set(amg_d_diag_solver_mold,info,pos=pos)
 
-    case ('JACOBI')
-      call lv%set(amg_d_jac_solver_mold,info,pos=pos)
-
-    case ('L1-DIAG')
+    case ('L1-DIAG','L1-JACOBI')
       call lv%set(amg_d_l1_diag_solver_mold,info,pos=pos)
-    case ('L1-JACOBI')
-      call lv%set(amg_d_l1_jac_solver_mold,info,pos=pos)
 
     case ('GS','FGS','FWGS')
       call lv%set(amg_d_gs_solver_mold,info,pos=pos)

amgprec/impl/level/amg_d_base_onelev_cseti.F90

@@ -177,7 +177,7 @@ subroutine amg_d_base_onelev_cseti(lv,what,val,info,pos,idx)
     case (amg_bwgs_)
       call lv%set(amg_d_bwgs_solver_mold,info,pos=pos)
 
-    case (psb_ilu_n_,psb_milu_n_,psb_ilu_t_)
+    case (amg_ilu_n_,amg_milu_n_,amg_ilu_t_)
       call lv%set(amg_d_ilu_solver_mold,info,pos=pos)
       if (info == 0) then
         if ((ipos_==amg_smooth_pre_) .or.(ipos_==amg_smooth_both_)) then

amgprec/impl/level/amg_d_base_onelev_memory_use.f90

@@ -0,0 +1,150 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_d_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefix,global)
+
+  use psb_base_mod
+  use amg_d_onelev_mod, amg_protect_name => amg_d_base_onelev_memory_use
+  Implicit None
+  ! Arguments
+  class(amg_d_onelev_type), intent(in)    :: lv
+  integer(psb_ipk_), intent(in)           :: il,nl,ilmin
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  character(len=*), intent(in), optional  :: prefix
+      integer(psb_ipk_), intent(in), optional :: verbosity
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)  :: err_act ,me, np
+  character(len=20), parameter :: name='amg_d_base_onelev_memory_use'
+  integer(psb_ipk_)  :: iout_, verbosity_
+  logical      :: coarse, global_
+  character(1024)    :: prefix_
+  integer(psb_epk_), allocatable  :: sz(:)
+
+
+  call psb_erractionsave(err_act)
+
+  ctxt = lv%base_desc%get_ctxt()
+  call psb_info(ctxt,me,np)
+  
+  coarse = (il==nl)
+
+  if (present(iout)) then 
+    iout_ = iout
+  else 
+    iout_ = psb_out_unit
+  end if
+
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+  if (present(global)) then
+    global_ = global
+  else
+    global_ = .true.
+  end if
+
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  if ((me == 0).or.(verbosity_>0)) write(iout_,*) trim(prefix_)
+
+  if (global_) then
+    allocate(sz(6))
+    sz(:) = 0 
+    sz(1) =  lv%base_a%sizeof()
+    sz(2) =  lv%base_desc%sizeof()
+    if (il >1)              sz(3) =  lv%linmap%sizeof()
+    if (allocated(lv%sm))   sz(4) =  lv%sm%sizeof()
+    if (allocated(lv%sm2a)) sz(5) =  lv%sm2a%sizeof()
+    if (allocated(lv%wrk))  sz(6) =  lv%wrk%sizeof()
+    call psb_sum(ctxt,sz)
+    if (me == 0) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:',                         sz(1) 
+      write(iout_,*) trim(prefix_), '       Descriptor:',                         sz(2) 
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:',              sz(3) 
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:',   sz(4) 
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', sz(5) 
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:',  sz(6)
+    end if
+    
+  else
+    if ((me == 0).or.(verbosity_>0)) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:', lv%base_a%sizeof()
+      write(iout_,*) trim(prefix_), '       Descriptor:', lv%base_desc%sizeof()
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:', lv%linmap%sizeof()
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:', lv%sm%sizeof()
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', lv%sm2a%sizeof()
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:', lv%wrk%sizeof()
+    end if
+  endif
+
+9998 continue
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+
+end subroutine amg_d_base_onelev_memory_use

amgprec/impl/level/amg_s_base_onelev_csetc.F90

@@ -44,7 +44,7 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_s_symdec_aggregator_mod
 #if !defined(SERIAL_MPI)
   use amg_s_parmatch_aggregator_mod
-#endif
+  use amg_s_poly_smoother
   use amg_s_jac_smoother
   use amg_s_as_smoother
   use amg_s_diag_solver
@@ -91,6 +91,7 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
   type(amg_s_ainv_solver_type)     ::  amg_s_ainv_solver_mold
   type(amg_s_invk_solver_type)     ::  amg_s_invk_solver_mold
   type(amg_s_invt_solver_type)     ::  amg_s_invt_solver_mold
+  type(amg_s_poly_smoother_type)   ::  amg_s_poly_smoother_mold
 #if defined(HAVE_SLU_)
   type(amg_s_slu_solver_type)   ::  amg_s_slu_solver_mold
 #endif
@@ -146,6 +147,9 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
       call lv%set(amg_s_as_smoother_mold,info,pos=pos)
       if (info == 0) call lv%set(amg_s_ilu_solver_mold,info,pos=pos)
 
+    case ('POLY')
+      call lv%set(amg_s_poly_smoother_mold,info,pos=pos)
+      if (info == 0) call lv%set(amg_s_l1_diag_solver_mold,info,pos=pos)
     case ('GS','FWGS')
       call lv%set(amg_s_jac_smoother_mold,info,pos='pre')
       if (info == 0) call lv%set(amg_s_gs_solver_mold,info,pos='pre')
@@ -191,16 +195,11 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
     case ('NONE','NOPREC','FACT_NONE')
       call lv%set(amg_s_id_solver_mold,info,pos=pos)
 
-    case ('DIAG')
+    case ('DIAG','JACOBI')
       call lv%set(amg_s_diag_solver_mold,info,pos=pos)
 
-    case ('JACOBI')
-      call lv%set(amg_s_jac_solver_mold,info,pos=pos)
-
-    case ('L1-DIAG')
+    case ('L1-DIAG','L1-JACOBI')
       call lv%set(amg_s_l1_diag_solver_mold,info,pos=pos)
-    case ('L1-JACOBI')
-      call lv%set(amg_s_l1_jac_solver_mold,info,pos=pos)
 
     case ('GS','FGS','FWGS')
       call lv%set(amg_s_gs_solver_mold,info,pos=pos)

amgprec/impl/level/amg_s_base_onelev_cseti.F90

@@ -165,7 +165,7 @@ subroutine amg_s_base_onelev_cseti(lv,what,val,info,pos,idx)
     case (amg_bwgs_)
       call lv%set(amg_s_bwgs_solver_mold,info,pos=pos)
 
-    case (psb_ilu_n_,psb_milu_n_,psb_ilu_t_)
+    case (amg_ilu_n_,amg_milu_n_,amg_ilu_t_)
       call lv%set(amg_s_ilu_solver_mold,info,pos=pos)
       if (info == 0) then
         if ((ipos_==amg_smooth_pre_) .or.(ipos_==amg_smooth_both_)) then

amgprec/impl/level/amg_s_base_onelev_memory_use.f90

@@ -0,0 +1,150 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_s_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefix,global)
+
+  use psb_base_mod
+  use amg_s_onelev_mod, amg_protect_name => amg_s_base_onelev_memory_use
+  Implicit None
+  ! Arguments
+  class(amg_s_onelev_type), intent(in)    :: lv
+  integer(psb_ipk_), intent(in)           :: il,nl,ilmin
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  character(len=*), intent(in), optional  :: prefix
+      integer(psb_ipk_), intent(in), optional :: verbosity
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)  :: err_act ,me, np
+  character(len=20), parameter :: name='amg_s_base_onelev_memory_use'
+  integer(psb_ipk_)  :: iout_, verbosity_
+  logical      :: coarse, global_
+  character(1024)    :: prefix_
+  integer(psb_epk_), allocatable  :: sz(:)
+
+
+  call psb_erractionsave(err_act)
+
+  ctxt = lv%base_desc%get_ctxt()
+  call psb_info(ctxt,me,np)
+  
+  coarse = (il==nl)
+
+  if (present(iout)) then 
+    iout_ = iout
+  else 
+    iout_ = psb_out_unit
+  end if
+
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+  if (present(global)) then
+    global_ = global
+  else
+    global_ = .true.
+  end if
+
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  if ((me == 0).or.(verbosity_>0)) write(iout_,*) trim(prefix_)
+
+  if (global_) then
+    allocate(sz(6))
+    sz(:) = 0 
+    sz(1) =  lv%base_a%sizeof()
+    sz(2) =  lv%base_desc%sizeof()
+    if (il >1)              sz(3) =  lv%linmap%sizeof()
+    if (allocated(lv%sm))   sz(4) =  lv%sm%sizeof()
+    if (allocated(lv%sm2a)) sz(5) =  lv%sm2a%sizeof()
+    if (allocated(lv%wrk))  sz(6) =  lv%wrk%sizeof()
+    call psb_sum(ctxt,sz)
+    if (me == 0) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:',                         sz(1) 
+      write(iout_,*) trim(prefix_), '       Descriptor:',                         sz(2) 
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:',              sz(3) 
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:',   sz(4) 
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', sz(5) 
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:',  sz(6)
+    end if
+    
+  else
+    if ((me == 0).or.(verbosity_>0)) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:', lv%base_a%sizeof()
+      write(iout_,*) trim(prefix_), '       Descriptor:', lv%base_desc%sizeof()
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:', lv%linmap%sizeof()
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:', lv%sm%sizeof()
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', lv%sm2a%sizeof()
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:', lv%wrk%sizeof()
+    end if
+  endif
+
+9998 continue
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+
+end subroutine amg_s_base_onelev_memory_use

amgprec/impl/level/amg_z_base_onelev_csetc.F90

@@ -43,7 +43,6 @@ subroutine amg_z_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_z_dec_aggregator_mod
   use amg_z_symdec_aggregator_mod
 #if !defined(SERIAL_MPI)
-#endif
   use amg_z_jac_smoother
   use amg_z_as_smoother
   use amg_z_diag_solver
@@ -202,16 +201,11 @@ subroutine amg_z_base_onelev_csetc(lv,what,val,info,pos,idx)
     case ('NONE','NOPREC','FACT_NONE')
       call lv%set(amg_z_id_solver_mold,info,pos=pos)
 
-    case ('DIAG')
+    case ('DIAG','JACOBI')
       call lv%set(amg_z_diag_solver_mold,info,pos=pos)
 
-    case ('JACOBI')
-      call lv%set(amg_z_jac_solver_mold,info,pos=pos)
-
-    case ('L1-DIAG')
+    case ('L1-DIAG','L1-JACOBI')
       call lv%set(amg_z_l1_diag_solver_mold,info,pos=pos)
-    case ('L1-JACOBI')
-      call lv%set(amg_z_l1_jac_solver_mold,info,pos=pos)
 
     case ('GS','FGS','FWGS')
       call lv%set(amg_z_gs_solver_mold,info,pos=pos)

amgprec/impl/level/amg_z_base_onelev_cseti.F90

@@ -176,7 +176,7 @@ subroutine amg_z_base_onelev_cseti(lv,what,val,info,pos,idx)
     case (amg_bwgs_)
       call lv%set(amg_z_bwgs_solver_mold,info,pos=pos)
 
-    case (psb_ilu_n_,psb_milu_n_,psb_ilu_t_)
+    case (amg_ilu_n_,amg_milu_n_,amg_ilu_t_)
       call lv%set(amg_z_ilu_solver_mold,info,pos=pos)
       if (info == 0) then
         if ((ipos_==amg_smooth_pre_) .or.(ipos_==amg_smooth_both_)) then

amgprec/impl/level/amg_z_base_onelev_memory_use.f90

@@ -0,0 +1,150 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+!
+!
+! verbosity:
+!        <0: suppress all messages
+!         0: normal
+!        >1: increased details 
+!
+subroutine amg_z_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefix,global)
+
+  use psb_base_mod
+  use amg_z_onelev_mod, amg_protect_name => amg_z_base_onelev_memory_use
+  Implicit None
+  ! Arguments
+  class(amg_z_onelev_type), intent(in)    :: lv
+  integer(psb_ipk_), intent(in)           :: il,nl,ilmin
+  integer(psb_ipk_), intent(out)          :: info
+  integer(psb_ipk_), intent(in), optional :: iout
+  character(len=*), intent(in), optional  :: prefix
+      integer(psb_ipk_), intent(in), optional :: verbosity
+  logical, intent(in), optional           :: global
+
+
+  ! Local variables
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)  :: err_act ,me, np
+  character(len=20), parameter :: name='amg_z_base_onelev_memory_use'
+  integer(psb_ipk_)  :: iout_, verbosity_
+  logical      :: coarse, global_
+  character(1024)    :: prefix_
+  integer(psb_epk_), allocatable  :: sz(:)
+
+
+  call psb_erractionsave(err_act)
+
+  ctxt = lv%base_desc%get_ctxt()
+  call psb_info(ctxt,me,np)
+  
+  coarse = (il==nl)
+
+  if (present(iout)) then 
+    iout_ = iout
+  else 
+    iout_ = psb_out_unit
+  end if
+
+  if (present(verbosity)) then
+    verbosity_ = verbosity
+  else
+    verbosity_ = 0
+  end if
+  if (verbosity_ < 0) goto 9998
+  if (present(global)) then
+    global_ = global
+  else
+    global_ = .true.
+  end if
+
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  if ((me == 0).or.(verbosity_>0)) write(iout_,*) trim(prefix_)
+
+  if (global_) then
+    allocate(sz(6))
+    sz(:) = 0 
+    sz(1) =  lv%base_a%sizeof()
+    sz(2) =  lv%base_desc%sizeof()
+    if (il >1)              sz(3) =  lv%linmap%sizeof()
+    if (allocated(lv%sm))   sz(4) =  lv%sm%sizeof()
+    if (allocated(lv%sm2a)) sz(5) =  lv%sm2a%sizeof()
+    if (allocated(lv%wrk))  sz(6) =  lv%wrk%sizeof()
+    call psb_sum(ctxt,sz)
+    if (me == 0) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:',                         sz(1) 
+      write(iout_,*) trim(prefix_), '       Descriptor:',                         sz(2) 
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:',              sz(3) 
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:',   sz(4) 
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', sz(5) 
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:',  sz(6)
+    end if
+    
+  else
+    if ((me == 0).or.(verbosity_>0)) then 
+      if (coarse)  then 
+        write(iout_,*) trim(prefix_), ' Level ',il,' (coarse)'
+      else
+        write(iout_,*) trim(prefix_), ' Level ',il
+      end if
+      write(iout_,*) trim(prefix_), '           Matrix:', lv%base_a%sizeof()
+      write(iout_,*) trim(prefix_), '       Descriptor:', lv%base_desc%sizeof()
+      if (il >1) write(iout_,*) trim(prefix_), '       Linear map:', lv%linmap%sizeof()
+      if (allocated(lv%sm)) write(iout_,*) trim(prefix_), '         Smoother:', lv%sm%sizeof()
+      if (allocated(lv%sm2a)) write(iout_,*) trim(prefix_), '       Smoother 2:', lv%sm2a%sizeof()
+      if (allocated(lv%wrk)) write(iout_,*) trim(prefix_), '        Workspace:', lv%wrk%sizeof()
+    end if
+  endif
+
+9998 continue
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+
+end subroutine amg_z_base_onelev_memory_use

amgprec/impl/smoother/Makefile

@@ -97,6 +97,17 @@ amg_d_jac_smoother_csetr.o \
 amg_d_l1_jac_smoother_bld.o \
 amg_d_l1_jac_smoother_descr.o \
 amg_d_l1_jac_smoother_clone.o \
+amg_d_poly_smoother_apply_vect.o \
+amg_d_poly_smoother_bld.o \
+amg_d_poly_smoother_cnv.o \
+amg_d_poly_smoother_clone.o \
+amg_d_poly_smoother_clone_settings.o \
+amg_d_poly_smoother_clear_data.o \
+amg_d_poly_smoother_descr.o \
+amg_d_poly_smoother_dmp.o \
+amg_d_poly_smoother_csetc.o \
+amg_d_poly_smoother_cseti.o \
+amg_d_poly_smoother_csetr.o \
 amg_s_as_smoother_apply.o \
 amg_s_as_smoother_apply_vect.o \
 amg_s_as_smoother_bld.o \
@@ -142,6 +153,17 @@ amg_s_jac_smoother_csetr.o \
 amg_s_l1_jac_smoother_bld.o \
 amg_s_l1_jac_smoother_descr.o \
 amg_s_l1_jac_smoother_clone.o \
+amg_s_poly_smoother_apply_vect.o \
+amg_s_poly_smoother_bld.o \
+amg_s_poly_smoother_cnv.o \
+amg_s_poly_smoother_clone.o \
+amg_s_poly_smoother_clone_settings.o \
+amg_s_poly_smoother_clear_data.o \
+amg_s_poly_smoother_descr.o \
+amg_s_poly_smoother_dmp.o \
+amg_s_poly_smoother_csetc.o \
+amg_s_poly_smoother_cseti.o \
+amg_s_poly_smoother_csetr.o \
 amg_z_as_smoother_apply.o \
 amg_z_as_smoother_apply_vect.o \
 amg_z_as_smoother_bld.o \

amgprec/impl/smoother/amg_c_jac_smoother_apply_vect.f90

@@ -175,7 +175,7 @@ subroutine amg_c_jac_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           if (info /= psb_success_) exit
 
           if ( sm%checkres.and.(mod(i,sm%checkiter) == 0) ) then
-            call psb_geaxpby(cone,x,czero,r,r,desc_data,info)
+            call psb_geaxpby(cone,x,czero,r,desc_data,info)
             call psb_spmm(-cone,sm%pa,ty,cone,r,desc_data,info)
             res  = psb_genrm2(r,desc_data,info)
             if( sm%printres ) then

amgprec/impl/smoother/amg_d_jac_smoother_apply_vect.f90

@@ -175,7 +175,7 @@ subroutine amg_d_jac_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           if (info /= psb_success_) exit
 
           if ( sm%checkres.and.(mod(i,sm%checkiter) == 0) ) then
-            call psb_geaxpby(done,x,dzero,r,r,desc_data,info)
+            call psb_geaxpby(done,x,dzero,r,desc_data,info)
             call psb_spmm(-done,sm%pa,ty,done,r,desc_data,info)
             res  = psb_genrm2(r,desc_data,info)
             if( sm%printres ) then

amgprec/impl/smoother/amg_d_poly_smoother_apply_vect.f90

@@ -0,0 +1,281 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
+     & sweeps,work,wv,info,init,initu) 
+
+  use psb_base_mod
+  use amg_d_diag_solver
+  use psb_base_krylov_conv_mod, only : log_conv
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_apply_vect
+  implicit none
+  type(psb_desc_type), intent(in)                 :: desc_data
+  class(amg_d_poly_smoother_type), intent(inout) :: sm
+  type(psb_d_vect_type),intent(inout)           :: x
+  type(psb_d_vect_type),intent(inout)           :: y
+  real(psb_dpk_),intent(in)                      :: alpha,beta
+  character(len=1),intent(in)                     :: trans
+  integer(psb_ipk_), intent(in)                   :: sweeps! this is ignored here, the polynomial degree dictates the value
+  real(psb_dpk_),target, intent(inout)           :: work(:)
+  type(psb_d_vect_type),intent(inout)           :: wv(:)
+  integer(psb_ipk_), intent(out)                  :: info
+  character, intent(in), optional                :: init
+  type(psb_d_vect_type),intent(inout), optional   :: initu
+  ! Timers
+  logical, parameter  :: do_timings=.true.
+  integer(psb_ipk_), save  :: poly_1=-1, poly_2=-1, poly_3=-1
+  integer(psb_ipk_), save  :: poly_mv=-1, poly_sv=-1, poly_vect=-1
+  !
+  integer(psb_ipk_)    :: n_row,n_col
+  type(psb_d_vect_type)  :: tx, ty, tz, r
+  real(psb_dpk_), pointer :: aux(:)
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: np, me, i, err_act
+  character           :: trans_, init_
+  real(psb_dpk_)      :: res, resdenum
+  character(len=20)   :: name='d_poly_smoother_apply_v'
+
+  call psb_erractionsave(err_act)
+
+  info = psb_success_
+  ctxt = desc_data%get_context()
+  call psb_info(ctxt,me,np)
+
+
+  if (present(init)) then
+    init_ = psb_toupper(init)
+  else
+    init_='Z'
+  end if
+
+  trans_ = psb_toupper(trans)
+  select case(trans_)
+  case('N')
+  case('T','C')
+  case default
+    call psb_errpush(psb_err_iarg_invalid_i_,name)
+    goto 9999
+  end select
+
+  if (.not.allocated(sm%sv)) then
+    info = 1121
+    call psb_errpush(info,name)
+    goto 9999
+  end if
+  
+  if ((do_timings).and.(poly_1==-1))       &
+       & poly_1 = psb_get_timer_idx("POLY: Chebychev4")
+  if ((do_timings).and.(poly_2==-1))       &
+       & poly_2 = psb_get_timer_idx("POLY: OptChebychev4")
+  if ((do_timings).and.(poly_3==-1))       &
+       & poly_3 = psb_get_timer_idx("POLY: OptChebychev1")
+  if ((do_timings).and.(poly_mv==-1))       &
+       & poly_mv = psb_get_timer_idx("POLY: spMV")
+  if ((do_timings).and.(poly_vect==-1))       &
+       & poly_vect = psb_get_timer_idx("POLY: Vectors")
+  if ((do_timings).and.(poly_sv==-1))       &
+       & poly_sv = psb_get_timer_idx("POLY: solver")
+  n_row = desc_data%get_local_rows()
+  n_col = desc_data%get_local_cols()
+
+  if (4*n_col <= size(work)) then
+    aux => work(:)
+  else
+    allocate(aux(4*n_col),stat=info)
+    if (info /= psb_success_) then
+      info=psb_err_alloc_request_
+      call psb_errpush(info,name,&
+           & i_err=(/4*n_col,izero,izero,izero,izero/),&
+           & a_err='real(psb_dpk_)')
+      goto 9999
+    end if
+  endif
+
+  if (size(wv) < 4) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='invalid wv size in smoother_apply')
+    goto 9999
+  end if
+
+  associate(tx => wv(1), ty => wv(2), tz => wv(3), r => wv(4))
+
+    call psb_geaxpby(done,x,dzero,r,desc_data,info)
+    call tx%zero()
+    call ty%zero()
+    call tz%zero()
+
+    select case(sm%variant)
+    case(amg_poly_lottes_)
+      if (do_timings) call psb_tic(poly_1)
+      block
+        real(psb_dpk_)      :: cz, cr
+        !  b == x
+        !  x == tx
+        !
+        do i=1, sm%pdegree-1
+          !   B r_{k-1}
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z') ! ty = M^{-1} r
+          if (do_timings) call psb_toc(poly_sv)
+          cz = (2*i*done-3)/(2*i*done+done)
+          cr = (8*i*done-4)/((2*i*done+done)*sm%rho_ba)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz(cr,cz,done,done,ty,tz,tx,desc_data,info) ! zk = cz * zk-1 + cr * rk-1 
+          if (do_timings) call psb_toc(poly_vect)
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(-done,sm%pa,tz,done,r,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+        end do
+        if (do_timings) call psb_tic(poly_sv)
+        call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z') ! ty = M^{-1} r
+        if (do_timings) call psb_toc(poly_sv)
+        cz = (2*sm%pdegree*done-3)/(2*sm%pdegree*done+done)
+        cr = (8*sm%pdegree*done-4)/((2*sm%pdegree*done+done)*sm%rho_ba)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz(cr,cz,done,done,ty,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+      end block
+      if (do_timings) call psb_toc(poly_1)
+
+    case(amg_poly_lottes_beta_)
+      if (do_timings) call psb_tic(poly_2)
+      block
+        real(psb_dpk_)      :: cz, cr
+        !  b == x
+        !  x == tx
+        !
+        if (allocated(sm%poly_beta)) then
+          if (size(sm%poly_beta) /= sm%pdegree) deallocate(sm%poly_beta)
+        end if
+        if (.not.allocated(sm%poly_beta)) then
+          call psb_realloc(sm%pdegree,sm%poly_beta,info)
+          sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
+        end if
+
+        do i=1, sm%pdegree-1
+          !   B r_{k-1}
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+          if (do_timings) call psb_toc(poly_sv)
+          cz = (2*i*done-3)/(2*i*done+done)
+          cr = (8*i*done-4)/((2*i*done+done)*sm%rho_ba)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz(cr,cz,sm%poly_beta(i),done,ty,tz,tx,desc_data,info)
+          if (do_timings) call psb_toc(poly_vect)
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(-done,sm%pa,tz,done,r,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+        end do
+        call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+        cz = (2*sm%pdegree*done-3)/(2*sm%pdegree*done+done)
+        cr = (8*sm%pdegree*done-4)/((2*sm%pdegree*done+done)*sm%rho_ba)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz(cr,cz,sm%poly_beta(sm%pdegree),done,ty,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+      end block
+      if (do_timings) call psb_toc(poly_2)
+    case(amg_poly_new_)
+      if (do_timings) call psb_tic(poly_3)
+      block
+        real(psb_dpk_)      :: sigma, theta, delta, rho_old, rho
+        !  b == x
+        !  x == tx
+        !
+
+        theta = (done+sm%cf_a)/2
+        delta = (done-sm%cf_a)/2
+        sigma = theta/delta
+        rho_old = done/sigma
+        if (do_timings) call psb_tic(poly_sv)
+        call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+        if (do_timings) call psb_toc(poly_sv)
+        call psb_geaxpby((done/sm%rho_ba),ty,dzero,r,desc_data,info)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz((done/theta),dzero,done,done,r,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+
+        ! tz == d
+        do i=1, sm%pdegree-1
+          !
+          !
+          ! r_{k-1} = r_k - (1/rho(BA)) B A d_k
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(done,sm%pa,tz,dzero,ty,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(-(done/sm%rho_ba),ty,done,r,desc_data,trans_,aux,wv(5:),info,init='Z')
+          if (do_timings) call psb_toc(poly_sv)
+          !
+          ! d_{k+1} = (rho rho_old) d_k + 2(rho/delta) r_{k+1}
+          rho = done/(2*sigma - rho_old)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz((2*rho/delta),(rho*rho_old),done,done,r,tz,tx,desc_data,info)
+          if (do_timings) call psb_toc(poly_vect)
+          rho_old = rho
+        end do
+      end block
+      if (do_timings) call psb_toc(poly_3)
+    case default
+      info=psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='wrong polynomial variant')
+      goto 9999
+    end select
+
+    if (info == psb_success_) call psb_geaxpby(alpha,tx,beta,y,desc_data,info)
+
+    if (info /= psb_success_) then
+      info=psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='polynomial smoother')
+      goto 9999
+    end if
+  end associate
+
+  if (.not.(4*n_col <= size(work))) then
+    deallocate(aux)
+  endif
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_d_poly_smoother_apply_vect

amgprec/impl/smoother/amg_d_poly_smoother_bld.f90

@@ -0,0 +1,179 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_bld(a,desc_a,sm,info,amold,vmold,imold)
+
+  use psb_base_mod
+  use amg_d_diag_solver
+  use amg_d_l1_diag_solver
+  use amg_d_poly_coeff_mod
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_bld
+  Implicit None
+
+  ! Arguments
+  type(psb_dspmat_type), intent(in), target          :: a
+  Type(psb_desc_type), Intent(inout)                 :: desc_a
+  class(amg_d_poly_smoother_type), intent(inout)      :: sm
+  integer(psb_ipk_), intent(out)                     :: info
+  class(psb_d_base_sparse_mat), intent(in), optional :: amold
+  class(psb_d_base_vect_type), intent(in), optional  :: vmold
+  class(psb_i_base_vect_type), intent(in), optional  :: imold
+  ! Local variables
+  type(psb_dspmat_type) :: tmpa
+  integer(psb_ipk_)   :: n_row,n_col, nrow_a, nztota, nzeros
+  type(psb_ctxt_type) :: ctxt
+  real(psb_dpk_), allocatable :: da(:), dsv(:)
+  integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
+  character(len=20)   :: name='d_poly_smoother_bld', ch_err
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+  debug_unit  = psb_get_debug_unit()
+  debug_level = psb_get_debug_level()
+  ctxt       = desc_a%get_context()
+  call psb_info(ctxt, me, np)
+  if (debug_level >= psb_debug_outer_) &
+       & write(debug_unit,*) me,' ',trim(name),' start'
+
+
+  n_row  = desc_a%get_local_rows()
+  n_col  = desc_a%get_local_cols()
+  nrow_a = a%get_nrows()
+  nztota = a%get_nzeros()
+  select case(sm%variant)
+  case(amg_poly_lottes_)
+    ! do nothing
+  case(amg_poly_lottes_beta_)
+    if ((1<=sm%pdegree).and.(sm%pdegree<=30)) then
+      call psb_realloc(sm%pdegree,sm%poly_beta,info)
+      sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
+    else
+      info = psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='invalid sm%degree for poly_beta')
+      goto 9999
+    end if
+  case(amg_poly_new_)
+
+    if ((1<=sm%pdegree).and.(sm%pdegree<=30)) then
+      !Ok
+      sm%cf_a = amg_d_poly_a_vect(sm%pdegree)
+    else
+      info = psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='invalid sm%degree for poly_a')
+      goto 9999
+    end if
+
+  case default  
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='invalid sm%variant')
+    goto 9999
+  end select
+
+  sm%pa => a
+  if (.not.allocated(sm%sv)) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='unallocated sm%sv')
+    goto 9999
+  end if
+  call sm%sv%build(a,desc_a,info,amold=amold,vmold=vmold)
+  if (info /= psb_success_) then
+    call psb_errpush(psb_err_from_subroutine_,name,&
+         & a_err='sv%build')
+    goto 9999
+  end if
+
+!!$  if (.false.) then
+!!$    select type(ssv => sm%sv)
+!!$    class is(amg_d_l1_diag_solver_type)
+!!$      da  = a%arwsum(info)
+!!$      dsv = ssv%dv%get_vect()
+!!$      sm%rho_ba = maxval(da(1:n_row)*dsv(1:n_row))
+!!$    class default
+!!$      write(0,*) 'PolySmoother BUILD: only L1-Jacobi/L1-DIAG for now ',ssv%get_fmt()
+!!$      sm%rho_ba = done
+!!$    end select
+!!$  else
+  if (sm%rho_ba <= dzero) then
+    select case(sm%rho_estimate)
+    case(amg_poly_rho_est_power_)
+      block
+        type(psb_d_vect_type) :: tq, tt, tz,wv(2)
+        real(psb_dpk_)        :: znrm, lambda
+        real(psb_dpk_),allocatable :: work(:)
+        integer(psb_ipk_)     :: i, n_cols
+        n_cols = desc_a%get_local_cols()
+        allocate(work(4*n_cols))
+        call psb_geasb(tz,desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(tt,desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(wv(1),desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(wv(2),desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geall(tq,desc_a,info)
+        call tq%set(done)
+        call psb_geasb(tq,desc_a,info,mold=vmold)
+        call psb_spmm(done,a,tq,dzero,tt,desc_a,info) !
+        call sm%sv%apply_v(done,tt,dzero,tz,desc_a,'NoTrans',work,wv,info) ! z_{k+1} = BA q_k
+        do i=1,sm%rho_estimate_iterations
+          znrm = psb_genrm2(tz,desc_a,info)               ! znrm = |z_k|_2
+          call psb_geaxpby((done/znrm),tz,dzero,tq,desc_a,info)  ! q_k = z_k/znrm
+          call psb_spmm(done,a,tq,dzero,tt,desc_a,info) ! t_{k+1} = BA q_k
+          call sm%sv%apply_v(done,tt,dzero,tz,desc_a,'NoTrans',work,wv,info) ! z_{k+1} = B t_{k+1}
+          lambda = psb_gedot(tq,tz,desc_a,info)      ! lambda = q_k^T z_{k+1} = q_k^T BA q_k
+          !write(0,*) 'BLD: lambda estimate ',i,lambda
+        end do
+        sm%rho_ba = lambda
+      end block
+    case default
+      write(0,*) ' Unknown algorithm for RHO(BA) estimate, defaulting to a value of 1.0 '
+      sm%rho_ba = done
+    end select
+  end if
+
+  if (debug_level >= psb_debug_outer_) &
+       & write(debug_unit,*) me,' ',trim(name),' end'
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_d_poly_smoother_bld

amgprec/impl/smoother/amg_d_poly_smoother_clear_data.f90

@@ -0,0 +1,70 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_d_poly_smoother_clear_data(sm,info)
+  
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_name =>  amg_d_poly_smoother_clear_data
+  Implicit None
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout) :: sm
+  integer(psb_ipk_), intent(out)                   :: info
+  integer(psb_ipk_)  :: err_act
+  character(len=20) :: name='amg_d_poly_smoother_clear_data'
+
+  call psb_erractionsave(err_act)
+
+  info = 0
+  sm%pdegree = 0
+  if (allocated(sm%poly_beta)) deallocate(sm%poly_beta)
+  sm%pa      => null()  
+  if ((info==0).and.allocated(sm%sv)) then
+    call sm%sv%clear_data(info)
+  end if
+  if (info /= 0) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name)
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_clear_data

amgprec/impl/smoother/amg_d_poly_smoother_clone.f90

@@ -0,0 +1,90 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_clone(sm,smout,info)
+
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_clone
+
+  Implicit None
+
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout)               :: sm
+  class(amg_d_base_smoother_type), allocatable, intent(inout) :: smout
+  integer(psb_ipk_), intent(out)                :: info
+  ! Local variables
+  integer(psb_ipk_) :: err_act
+
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+
+  if (allocated(smout)) then
+    call smout%free(info)
+    if (info == psb_success_) deallocate(smout, stat=info)
+  end if
+  if (info == psb_success_) &
+       & allocate(amg_d_poly_smoother_type :: smout, stat=info)
+  if (info /= 0) then
+    info = psb_err_alloc_dealloc_
+    goto 9999
+  end if
+
+  select type(smo => smout)
+  type is (amg_d_poly_smoother_type)
+    smo%pdegree    = sm%pdegree
+    smo%rho_ba     = sm%rho_ba
+    smo%poly_beta  = sm%poly_beta
+    smo%pa         => sm%pa
+    if ((info==psb_success_).and.(allocated(sm%sv))) then
+      allocate(smout%sv,mold=sm%sv,stat=info)
+      if (info == psb_success_) call sm%sv%clone(smo%sv,info)
+    end if
+
+  class default
+    info = psb_err_internal_error_
+  end select
+
+  if (info /= 0) goto 9999
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_clone

amgprec/impl/smoother/amg_d_poly_smoother_clone_settings.f90

@@ -0,0 +1,102 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               asd on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_d_poly_smoother_clone_settings(sm,smout,info)
+
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_name =>  amg_d_poly_smoother_clone_settings
+  Implicit None
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout)              :: sm
+  class(amg_d_base_smoother_type), intent(inout) :: smout
+  integer(psb_ipk_), intent(out)                 :: info
+  integer(psb_ipk_)  :: err_act
+  character(len=20) :: name='d_poly_smoother_clone_settings'
+
+  call psb_erractionsave(err_act)
+
+  info = psb_success_
+  
+  select type(smout)
+  class is(amg_d_poly_smoother_type)
+
+    smout%pa => null()
+    smout%pdegree   = sm%pdegree
+    smout%variant   = sm%variant
+    smout%cf_a      = sm%cf_a
+    smout%rho_ba    = sm%rho_ba
+    smout%rho_estimate  = sm%rho_estimate
+    smout%rho_estimate_iterations  = sm%rho_estimate_iterations
+    smout%poly_beta = sm%poly_beta
+    
+    if (allocated(smout%sv)) then
+      if (.not.same_type_as(sm%sv,smout%sv)) then
+        call smout%sv%free(info)
+        if (info == 0) deallocate(smout%sv,stat=info)
+      end if
+    end if
+    if (info /= 0) then
+      info = psb_err_internal_error_
+    else
+      if (allocated(smout%sv)) then
+        if (same_type_as(sm%sv,smout%sv)) then
+          call sm%sv%clone_settings(smout%sv,info)
+        else
+          info = psb_err_internal_error_
+        end if
+      else
+        allocate(smout%sv,mold=sm%sv,stat=info)
+        if (info == 0) call sm%sv%clone_settings(smout%sv,info)
+        if (info /= 0) info = psb_err_internal_error_
+      end if
+    end if
+  class default
+    info = psb_err_internal_error_
+  end select
+  
+  if (info /= 0) then
+    call psb_errpush(info,name) 
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_clone_settings

amgprec/impl/smoother/amg_d_poly_smoother_cnv.f90

@@ -0,0 +1,77 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_d_poly_smoother_cnv(sm,info,amold,vmold,imold)
+
+  use psb_base_mod
+  use amg_d_diag_solver
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_cnv
+  Implicit None
+
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout)      :: sm
+  integer(psb_ipk_), intent(out)                     :: info
+  class(psb_d_base_sparse_mat), intent(in), optional :: amold
+  class(psb_d_base_vect_type), intent(in), optional  :: vmold
+  class(psb_i_base_vect_type), intent(in), optional  :: imold
+  ! Local variables
+  integer(psb_ipk_) :: i, err_act, debug_unit, debug_level
+  character(len=20) :: name='d_poly_smoother_cnv', ch_err
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+  debug_unit  = psb_get_debug_unit()
+  debug_level = psb_get_debug_level()
+
+
+
+  if (allocated(sm%sv)) &
+       & call sm%sv%cnv(info,amold=amold,vmold=vmold,imold=imold)
+  if (info /= psb_success_) then
+    call psb_errpush(psb_err_from_subroutine_,name,&
+         & a_err='solver cnv')
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_d_poly_smoother_cnv

amgprec/impl/smoother/amg_d_poly_smoother_csetc.f90

@@ -0,0 +1,76 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_csetc(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_csetc
+  Implicit None
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                     :: what
+  character(len=*), intent(in)                   :: val
+  integer(psb_ipk_), intent(out)                 :: info
+  integer(psb_ipk_), intent(in), optional       :: idx
+  integer(psb_ipk_) :: err_act, ival
+  character(len=20) :: name='d_poly_smoother_csetc'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(trim(what)))
+  case('POLY_VARIANT')
+    call sm%set(what,amg_stringval(val),info,idx=idx)
+  case('POLY_RHO_ESTIMATE')
+    call sm%set(what,amg_stringval(val),info,idx=idx)
+  case default
+    call sm%amg_d_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  if (info /= psb_success_) then
+    info = psb_err_from_subroutine_
+    call psb_errpush(info, name)
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_csetc

amgprec/impl/smoother/amg_d_poly_smoother_cseti.f90

@@ -0,0 +1,92 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_cseti(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_nam => amg_d_poly_smoother_cseti
+  Implicit None
+
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                 :: what
+  integer(psb_ipk_), intent(in)                :: val
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: idx
+  integer(psb_ipk_) :: err_act
+  character(len=20)  :: name='d_poly_smoother_cseti'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(what))
+  case('POLY_DEGREE')
+    sm%pdegree = val
+  case('POLY_VARIANT')
+    select case(val)
+    case(amg_poly_lottes_,amg_poly_lottes_beta_,amg_poly_new_)
+      sm%variant = val
+    case default
+      write(0,*) 'Invalid choice for POLY_VARIANT, defaulting to amg_poly_lottes_',val
+      sm%variant = amg_poly_lottes_
+    end select
+  case('POLY_RHO_ESTIMATE')
+    select case(val)
+    case (amg_poly_rho_est_power_)
+      sm%rho_estimate = val
+    case default
+      write(0,*) 'Invalid choice for POLY_RHO_ESTIMATE, defaulting to amg_poly_rho_power'
+      sm%variant = amg_poly_rho_est_power_
+    end select      
+  case('POLY_RHO_ESTIMATE_ITERATIONS')
+    if (val>0) then 
+      sm%rho_estimate_iterations = val
+    else
+      write(0,*) 'Invalid choice for POLY_RHO_ESTIMATE_ITERATIONS, defaulting to 20'
+      sm%variant = 20
+    end if
+  case default
+    call sm%amg_d_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_cseti

amgprec/impl/smoother/amg_d_poly_smoother_csetr.f90

@@ -0,0 +1,74 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_csetr(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_nam => amg_d_poly_smoother_csetr
+  Implicit None
+
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                 :: what
+  real(psb_dpk_), intent(in)                 :: val
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: idx
+  integer(psb_ipk_) :: err_act
+  character(len=20)  :: name='d_poly_smoother_csetr'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(what))
+  case('POLY_RHO_BA')
+    if ((dzero<val).and.(val<=done)) then 
+      sm%rho_ba  = val
+    else
+      write(0,*) 'Invalid choice for POLY_RHO_BA, defaulting to compute estimate'
+      sm%rho_ba = -done
+    end if
+  case default
+    call sm%amg_d_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_d_poly_smoother_csetr

amgprec/impl/smoother/amg_d_poly_smoother_descr.f90

@@ -0,0 +1,108 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Fabio Durastante
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_d_poly_smoother_descr(sm,info,iout,coarse,prefix)
+
+  use psb_base_mod
+  use amg_d_diag_solver
+  use amg_d_poly_smoother, amg_protect_name => amg_d_poly_smoother_descr
+
+  Implicit None
+
+  ! Arguments
+  class(amg_d_poly_smoother_type), intent(in) :: sm
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: iout
+  logical, intent(in), optional              :: coarse
+  character(len=*), intent(in), optional  :: prefix
+
+  ! Local variables
+  integer(psb_ipk_)      :: err_act
+  character(len=20), parameter :: name='amg_d_poly_smoother_descr'
+  integer(psb_ipk_)      :: iout_
+  logical      :: coarse_
+  character(1024)    :: prefix_
+
+  call psb_erractionsave(err_act)
+  info = psb_success_
+  if (present(coarse)) then
+    coarse_ = coarse
+  else
+    coarse_ = .false.
+  end if
+  if (present(iout)) then
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  endif
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  write(iout_,*) trim(prefix_), '  Polynomial smoother   '
+  select case(sm%variant)
+  case(amg_poly_lottes_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_LOTTES'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+  case(amg_poly_lottes_beta_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_LOTTES_BETA'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+    if (allocated(sm%poly_beta)) write(iout_,*) trim(prefix_), '   Coefficients:        ',sm%poly_beta(1:sm%pdegree)
+  case(amg_poly_new_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_NEW'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+    write(iout_,*) trim(prefix_), '    Coefficient:        ',sm%cf_a
+  case default
+    write(iout_,*) trim(prefix_), '        variant:        ','UNKNOWN???'
+  end select
+  if (allocated(sm%sv)) then
+    write(iout_,*) trim(prefix_), '       Local solver details:'
+    call sm%sv%descr(info,iout_,coarse=coarse,prefix=prefix)
+  end if
+
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+end subroutine amg_d_poly_smoother_descr

amgprec/impl/smoother/amg_d_poly_smoother_dmp.f90

@@ -0,0 +1,90 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_d_poly_smoother_dmp(sm,desc,level,info,prefix,head,smoother,solver,global_num)
+  
+  use psb_base_mod
+  use amg_d_poly_smoother, amg_protect_nam => amg_d_poly_smoother_dmp
+  implicit none 
+  class(amg_d_poly_smoother_type), intent(in) :: sm
+  type(psb_desc_type), intent(in)               :: desc
+  integer(psb_ipk_), intent(in)               :: level
+  integer(psb_ipk_), intent(out)              :: info
+  character(len=*), intent(in), optional :: prefix, head
+  logical, optional, intent(in)    :: smoother, solver, global_num
+  integer(psb_ipk_)  :: i, j, il1, iln, lname, lev
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: iam, np
+  character(len=80)   :: prefix_
+  character(len=120)  :: fname ! len should be at least 20 more than
+  integer(psb_lpk_), allocatable :: iv(:)
+  logical :: smoother_, global_num_
+  !  len of prefix_ 
+
+  info = 0
+
+  if (present(prefix)) then 
+    prefix_ = trim(prefix(1:min(len(prefix),len(prefix_))))
+  else
+    prefix_ = "dump_smth_d"
+  end if
+  ctxt = desc%get_context()
+  call psb_info(ctxt,iam,np)
+
+  if (present(smoother)) then 
+    smoother_ = smoother
+  else
+    smoother_ = .false. 
+  end if
+  if (present(global_num)) then 
+    global_num_ = global_num
+  else
+    global_num_ = .false. 
+  end if
+  lname = len_trim(prefix_)
+  fname = trim(prefix_)
+  write(fname(lname+1:lname+5),'(a,i3.3)') '_poly',iam
+  lname = lname + 8
+  ! to be completed
+  
+
+
+  ! At base level do nothing for the smoother
+  if (allocated(sm%sv)) &
+       & call sm%sv%dump(desc,level,info,solver=solver,prefix=prefix,global_num=global_num)
+
+end subroutine amg_d_poly_smoother_dmp

amgprec/impl/smoother/amg_s_jac_smoother_apply_vect.f90

@@ -175,7 +175,7 @@ subroutine amg_s_jac_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           if (info /= psb_success_) exit
 
           if ( sm%checkres.and.(mod(i,sm%checkiter) == 0) ) then
-            call psb_geaxpby(sone,x,szero,r,r,desc_data,info)
+            call psb_geaxpby(sone,x,szero,r,desc_data,info)
             call psb_spmm(-sone,sm%pa,ty,sone,r,desc_data,info)
             res  = psb_genrm2(r,desc_data,info)
             if( sm%printres ) then

amgprec/impl/smoother/amg_s_poly_smoother_apply_vect.f90

@@ -0,0 +1,281 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
+     & sweeps,work,wv,info,init,initu) 
+
+  use psb_base_mod
+  use amg_s_diag_solver
+  use psb_base_krylov_conv_mod, only : log_conv
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_apply_vect
+  implicit none
+  type(psb_desc_type), intent(in)                 :: desc_data
+  class(amg_s_poly_smoother_type), intent(inout) :: sm
+  type(psb_s_vect_type),intent(inout)           :: x
+  type(psb_s_vect_type),intent(inout)           :: y
+  real(psb_spk_),intent(in)                      :: alpha,beta
+  character(len=1),intent(in)                     :: trans
+  integer(psb_ipk_), intent(in)                   :: sweeps! this is ignored here, the polynomial degree dictates the value
+  real(psb_spk_),target, intent(inout)           :: work(:)
+  type(psb_s_vect_type),intent(inout)           :: wv(:)
+  integer(psb_ipk_), intent(out)                  :: info
+  character, intent(in), optional                :: init
+  type(psb_s_vect_type),intent(inout), optional   :: initu
+  ! Timers
+  logical, parameter  :: do_timings=.true.
+  integer(psb_ipk_), save  :: poly_1=-1, poly_2=-1, poly_3=-1
+  integer(psb_ipk_), save  :: poly_mv=-1, poly_sv=-1, poly_vect=-1
+  !
+  integer(psb_ipk_)    :: n_row,n_col
+  type(psb_s_vect_type)  :: tx, ty, tz, r
+  real(psb_spk_), pointer :: aux(:)
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: np, me, i, err_act
+  character           :: trans_, init_
+  real(psb_spk_)      :: res, resdenum
+  character(len=20)   :: name='d_poly_smoother_apply_v'
+
+  call psb_erractionsave(err_act)
+
+  info = psb_success_
+  ctxt = desc_data%get_context()
+  call psb_info(ctxt,me,np)
+
+
+  if (present(init)) then
+    init_ = psb_toupper(init)
+  else
+    init_='Z'
+  end if
+
+  trans_ = psb_toupper(trans)
+  select case(trans_)
+  case('N')
+  case('T','C')
+  case default
+    call psb_errpush(psb_err_iarg_invalid_i_,name)
+    goto 9999
+  end select
+
+  if (.not.allocated(sm%sv)) then
+    info = 1121
+    call psb_errpush(info,name)
+    goto 9999
+  end if
+  
+  if ((do_timings).and.(poly_1==-1))       &
+       & poly_1 = psb_get_timer_idx("POLY: Chebychev4")
+  if ((do_timings).and.(poly_2==-1))       &
+       & poly_2 = psb_get_timer_idx("POLY: OptChebychev4")
+  if ((do_timings).and.(poly_3==-1))       &
+       & poly_3 = psb_get_timer_idx("POLY: OptChebychev1")
+  if ((do_timings).and.(poly_mv==-1))       &
+       & poly_mv = psb_get_timer_idx("POLY: spMV")
+  if ((do_timings).and.(poly_vect==-1))       &
+       & poly_vect = psb_get_timer_idx("POLY: Vectors")
+  if ((do_timings).and.(poly_sv==-1))       &
+       & poly_sv = psb_get_timer_idx("POLY: solver")
+  n_row = desc_data%get_local_rows()
+  n_col = desc_data%get_local_cols()
+
+  if (4*n_col <= size(work)) then
+    aux => work(:)
+  else
+    allocate(aux(4*n_col),stat=info)
+    if (info /= psb_success_) then
+      info=psb_err_alloc_request_
+      call psb_errpush(info,name,&
+           & i_err=(/4*n_col,izero,izero,izero,izero/),&
+           & a_err='real(psb_spk_)')
+      goto 9999
+    end if
+  endif
+
+  if (size(wv) < 4) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='invalid wv size in smoother_apply')
+    goto 9999
+  end if
+
+  associate(tx => wv(1), ty => wv(2), tz => wv(3), r => wv(4))
+
+    call psb_geaxpby(sone,x,szero,r,desc_data,info)
+    call tx%zero()
+    call ty%zero()
+    call tz%zero()
+
+    select case(sm%variant)
+    case(amg_poly_lottes_)
+      if (do_timings) call psb_tic(poly_1)
+      block
+        real(psb_spk_)      :: cz, cr
+        !  b == x
+        !  x == tx
+        !
+        do i=1, sm%pdegree-1
+          !   B r_{k-1}
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z') ! ty = M^{-1} r
+          if (do_timings) call psb_toc(poly_sv)
+          cz = (2*i*sone-3)/(2*i*sone+sone)
+          cr = (8*i*sone-4)/((2*i*sone+sone)*sm%rho_ba)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz(cr,cz,sone,sone,ty,tz,tx,desc_data,info) ! zk = cz * zk-1 + cr * rk-1 
+          if (do_timings) call psb_toc(poly_vect)
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(-sone,sm%pa,tz,sone,r,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+        end do
+        if (do_timings) call psb_tic(poly_sv)
+        call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z') ! ty = M^{-1} r
+        if (do_timings) call psb_toc(poly_sv)
+        cz = (2*sm%pdegree*sone-3)/(2*sm%pdegree*sone+sone)
+        cr = (8*sm%pdegree*sone-4)/((2*sm%pdegree*sone+sone)*sm%rho_ba)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz(cr,cz,sone,sone,ty,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+      end block
+      if (do_timings) call psb_toc(poly_1)
+
+    case(amg_poly_lottes_beta_)
+      if (do_timings) call psb_tic(poly_2)
+      block
+        real(psb_spk_)      :: cz, cr
+        !  b == x
+        !  x == tx
+        !
+        if (allocated(sm%poly_beta)) then
+          if (size(sm%poly_beta) /= sm%pdegree) deallocate(sm%poly_beta)
+        end if
+        if (.not.allocated(sm%poly_beta)) then
+          call psb_realloc(sm%pdegree,sm%poly_beta,info)
+          sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
+        end if
+
+        do i=1, sm%pdegree-1
+          !   B r_{k-1}
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+          if (do_timings) call psb_toc(poly_sv)
+          cz = (2*i*sone-3)/(2*i*sone+sone)
+          cr = (8*i*sone-4)/((2*i*sone+sone)*sm%rho_ba)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz(cr,cz,sm%poly_beta(i),sone,ty,tz,tx,desc_data,info)
+          if (do_timings) call psb_toc(poly_vect)
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(-sone,sm%pa,tz,sone,r,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+        end do
+        call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+        cz = (2*sm%pdegree*sone-3)/(2*sm%pdegree*sone+sone)
+        cr = (8*sm%pdegree*sone-4)/((2*sm%pdegree*sone+sone)*sm%rho_ba)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz(cr,cz,sm%poly_beta(sm%pdegree),sone,ty,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+      end block
+      if (do_timings) call psb_toc(poly_2)
+    case(amg_poly_new_)
+      if (do_timings) call psb_tic(poly_3)
+      block
+        real(psb_spk_)      :: sigma, theta, delta, rho_old, rho
+        !  b == x
+        !  x == tx
+        !
+
+        theta = (sone+sm%cf_a)/2
+        delta = (sone-sm%cf_a)/2
+        sigma = theta/delta
+        rho_old = sone/sigma
+        if (do_timings) call psb_tic(poly_sv)
+        call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+        if (do_timings) call psb_toc(poly_sv)
+        call psb_geaxpby((sone/sm%rho_ba),ty,szero,r,desc_data,info)
+        if (do_timings) call psb_tic(poly_vect)
+        call psb_abgdxyz((sone/theta),szero,sone,sone,r,tz,tx,desc_data,info)
+        if (do_timings) call psb_toc(poly_vect)
+
+        ! tz == d
+        do i=1, sm%pdegree-1
+          !
+          !
+          ! r_{k-1} = r_k - (1/rho(BA)) B A d_k
+          if (do_timings) call psb_tic(poly_mv)
+          call psb_spmm(sone,sm%pa,tz,szero,ty,desc_data,info,work=aux,trans=trans_)
+          if (do_timings) call psb_toc(poly_mv)
+          if (do_timings) call psb_tic(poly_sv)
+          call sm%sv%apply(-(sone/sm%rho_ba),ty,sone,r,desc_data,trans_,aux,wv(5:),info,init='Z')
+          if (do_timings) call psb_toc(poly_sv)
+          !
+          ! d_{k+1} = (rho rho_old) d_k + 2(rho/delta) r_{k+1}
+          rho = sone/(2*sigma - rho_old)
+          if (do_timings) call psb_tic(poly_vect)
+          call psb_abgdxyz((2*rho/delta),(rho*rho_old),sone,sone,r,tz,tx,desc_data,info)
+          if (do_timings) call psb_toc(poly_vect)
+          rho_old = rho
+        end do
+      end block
+      if (do_timings) call psb_toc(poly_3)
+    case default
+      info=psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='wrong polynomial variant')
+      goto 9999
+    end select
+
+    if (info == psb_success_) call psb_geaxpby(alpha,tx,beta,y,desc_data,info)
+
+    if (info /= psb_success_) then
+      info=psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='polynomial smoother')
+      goto 9999
+    end if
+  end associate
+
+  if (.not.(4*n_col <= size(work))) then
+    deallocate(aux)
+  endif
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_s_poly_smoother_apply_vect

amgprec/impl/smoother/amg_s_poly_smoother_bld.f90

@@ -0,0 +1,179 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_bld(a,desc_a,sm,info,amold,vmold,imold)
+
+  use psb_base_mod
+  use amg_s_diag_solver
+  use amg_s_l1_diag_solver
+  use amg_d_poly_coeff_mod
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_bld
+  Implicit None
+
+  ! Arguments
+  type(psb_sspmat_type), intent(in), target          :: a
+  Type(psb_desc_type), Intent(inout)                 :: desc_a
+  class(amg_s_poly_smoother_type), intent(inout)      :: sm
+  integer(psb_ipk_), intent(out)                     :: info
+  class(psb_s_base_sparse_mat), intent(in), optional :: amold
+  class(psb_s_base_vect_type), intent(in), optional  :: vmold
+  class(psb_i_base_vect_type), intent(in), optional  :: imold
+  ! Local variables
+  type(psb_sspmat_type) :: tmpa
+  integer(psb_ipk_)   :: n_row,n_col, nrow_a, nztota, nzeros
+  type(psb_ctxt_type) :: ctxt
+  real(psb_spk_), allocatable :: da(:), dsv(:)
+  integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
+  character(len=20)   :: name='d_poly_smoother_bld', ch_err
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+  debug_unit  = psb_get_debug_unit()
+  debug_level = psb_get_debug_level()
+  ctxt       = desc_a%get_context()
+  call psb_info(ctxt, me, np)
+  if (debug_level >= psb_debug_outer_) &
+       & write(debug_unit,*) me,' ',trim(name),' start'
+
+
+  n_row  = desc_a%get_local_rows()
+  n_col  = desc_a%get_local_cols()
+  nrow_a = a%get_nrows()
+  nztota = a%get_nzeros()
+  select case(sm%variant)
+  case(amg_poly_lottes_)
+    ! do nothing
+  case(amg_poly_lottes_beta_)
+    if ((1<=sm%pdegree).and.(sm%pdegree<=30)) then
+      call psb_realloc(sm%pdegree,sm%poly_beta,info)
+      sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
+    else
+      info = psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='invalid sm%degree for poly_beta')
+      goto 9999
+    end if
+  case(amg_poly_new_)
+
+    if ((1<=sm%pdegree).and.(sm%pdegree<=30)) then
+      !Ok
+      sm%cf_a = amg_d_poly_a_vect(sm%pdegree)
+    else
+      info = psb_err_internal_error_
+      call psb_errpush(info,name,&
+           & a_err='invalid sm%degree for poly_a')
+      goto 9999
+    end if
+
+  case default  
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='invalid sm%variant')
+    goto 9999
+  end select
+
+  sm%pa => a
+  if (.not.allocated(sm%sv)) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name,&
+         & a_err='unallocated sm%sv')
+    goto 9999
+  end if
+  call sm%sv%build(a,desc_a,info,amold=amold,vmold=vmold)
+  if (info /= psb_success_) then
+    call psb_errpush(psb_err_from_subroutine_,name,&
+         & a_err='sv%build')
+    goto 9999
+  end if
+
+!!$  if (.false.) then
+!!$    select type(ssv => sm%sv)
+!!$    class is(amg_s_l1_diag_solver_type)
+!!$      da  = a%arwsum(info)
+!!$      dsv = ssv%dv%get_vect()
+!!$      sm%rho_ba = maxval(da(1:n_row)*dsv(1:n_row))
+!!$    class default
+!!$      write(0,*) 'PolySmoother BUILD: only L1-Jacobi/L1-DIAG for now ',ssv%get_fmt()
+!!$      sm%rho_ba = sone
+!!$    end select
+!!$  else
+  if (sm%rho_ba <= szero) then
+    select case(sm%rho_estimate)
+    case(amg_poly_rho_est_power_)
+      block
+        type(psb_s_vect_type) :: tq, tt, tz,wv(2)
+        real(psb_spk_)        :: znrm, lambda
+        real(psb_spk_),allocatable :: work(:)
+        integer(psb_ipk_)     :: i, n_cols
+        n_cols = desc_a%get_local_cols()
+        allocate(work(4*n_cols))
+        call psb_geasb(tz,desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(tt,desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(wv(1),desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geasb(wv(2),desc_a,info,mold=vmold,scratch=.true.)
+        call psb_geall(tq,desc_a,info)
+        call tq%set(sone)
+        call psb_geasb(tq,desc_a,info,mold=vmold)
+        call psb_spmm(sone,a,tq,szero,tt,desc_a,info) !
+        call sm%sv%apply_v(sone,tt,szero,tz,desc_a,'NoTrans',work,wv,info) ! z_{k+1} = BA q_k
+        do i=1,sm%rho_estimate_iterations
+          znrm = psb_genrm2(tz,desc_a,info)               ! znrm = |z_k|_2
+          call psb_geaxpby((sone/znrm),tz,szero,tq,desc_a,info)  ! q_k = z_k/znrm
+          call psb_spmm(sone,a,tq,szero,tt,desc_a,info) ! t_{k+1} = BA q_k
+          call sm%sv%apply_v(sone,tt,szero,tz,desc_a,'NoTrans',work,wv,info) ! z_{k+1} = B t_{k+1}
+          lambda = psb_gedot(tq,tz,desc_a,info)      ! lambda = q_k^T z_{k+1} = q_k^T BA q_k
+          !write(0,*) 'BLD: lambda estimate ',i,lambda
+        end do
+        sm%rho_ba = lambda
+      end block
+    case default
+      write(0,*) ' Unknown algorithm for RHO(BA) estimate, defaulting to a value of 1.0 '
+      sm%rho_ba = sone
+    end select
+  end if
+
+  if (debug_level >= psb_debug_outer_) &
+       & write(debug_unit,*) me,' ',trim(name),' end'
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_s_poly_smoother_bld

amgprec/impl/smoother/amg_s_poly_smoother_clear_data.f90

@@ -0,0 +1,70 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_s_poly_smoother_clear_data(sm,info)
+  
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_name =>  amg_s_poly_smoother_clear_data
+  Implicit None
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout) :: sm
+  integer(psb_ipk_), intent(out)                   :: info
+  integer(psb_ipk_)  :: err_act
+  character(len=20) :: name='amg_s_poly_smoother_clear_data'
+
+  call psb_erractionsave(err_act)
+
+  info = 0
+  sm%pdegree = 0
+  if (allocated(sm%poly_beta)) deallocate(sm%poly_beta)
+  sm%pa      => null()  
+  if ((info==0).and.allocated(sm%sv)) then
+    call sm%sv%clear_data(info)
+  end if
+  if (info /= 0) then
+    info = psb_err_internal_error_
+    call psb_errpush(info,name)
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_clear_data

amgprec/impl/smoother/amg_s_poly_smoother_clone.f90

@@ -0,0 +1,90 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_clone(sm,smout,info)
+
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_clone
+
+  Implicit None
+
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout)               :: sm
+  class(amg_s_base_smoother_type), allocatable, intent(inout) :: smout
+  integer(psb_ipk_), intent(out)                :: info
+  ! Local variables
+  integer(psb_ipk_) :: err_act
+
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+
+  if (allocated(smout)) then
+    call smout%free(info)
+    if (info == psb_success_) deallocate(smout, stat=info)
+  end if
+  if (info == psb_success_) &
+       & allocate(amg_s_poly_smoother_type :: smout, stat=info)
+  if (info /= 0) then
+    info = psb_err_alloc_dealloc_
+    goto 9999
+  end if
+
+  select type(smo => smout)
+  type is (amg_s_poly_smoother_type)
+    smo%pdegree    = sm%pdegree
+    smo%rho_ba     = sm%rho_ba
+    smo%poly_beta  = sm%poly_beta
+    smo%pa         => sm%pa
+    if ((info==psb_success_).and.(allocated(sm%sv))) then
+      allocate(smout%sv,mold=sm%sv,stat=info)
+      if (info == psb_success_) call sm%sv%clone(smo%sv,info)
+    end if
+
+  class default
+    info = psb_err_internal_error_
+  end select
+
+  if (info /= 0) goto 9999
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_clone

amgprec/impl/smoother/amg_s_poly_smoother_clone_settings.f90

@@ -0,0 +1,102 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               asd on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_s_poly_smoother_clone_settings(sm,smout,info)
+
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_name =>  amg_s_poly_smoother_clone_settings
+  Implicit None
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout)              :: sm
+  class(amg_s_base_smoother_type), intent(inout) :: smout
+  integer(psb_ipk_), intent(out)                 :: info
+  integer(psb_ipk_)  :: err_act
+  character(len=20) :: name='d_poly_smoother_clone_settings'
+
+  call psb_erractionsave(err_act)
+
+  info = psb_success_
+  
+  select type(smout)
+  class is(amg_s_poly_smoother_type)
+
+    smout%pa => null()
+    smout%pdegree   = sm%pdegree
+    smout%variant   = sm%variant
+    smout%cf_a      = sm%cf_a
+    smout%rho_ba    = sm%rho_ba
+    smout%rho_estimate  = sm%rho_estimate
+    smout%rho_estimate_iterations  = sm%rho_estimate_iterations
+    smout%poly_beta = sm%poly_beta
+    
+    if (allocated(smout%sv)) then
+      if (.not.same_type_as(sm%sv,smout%sv)) then
+        call smout%sv%free(info)
+        if (info == 0) deallocate(smout%sv,stat=info)
+      end if
+    end if
+    if (info /= 0) then
+      info = psb_err_internal_error_
+    else
+      if (allocated(smout%sv)) then
+        if (same_type_as(sm%sv,smout%sv)) then
+          call sm%sv%clone_settings(smout%sv,info)
+        else
+          info = psb_err_internal_error_
+        end if
+      else
+        allocate(smout%sv,mold=sm%sv,stat=info)
+        if (info == 0) call sm%sv%clone_settings(smout%sv,info)
+        if (info /= 0) info = psb_err_internal_error_
+      end if
+    end if
+  class default
+    info = psb_err_internal_error_
+  end select
+  
+  if (info /= 0) then
+    call psb_errpush(info,name) 
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_clone_settings

amgprec/impl/smoother/amg_s_poly_smoother_cnv.f90

@@ -0,0 +1,77 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_s_poly_smoother_cnv(sm,info,amold,vmold,imold)
+
+  use psb_base_mod
+  use amg_s_diag_solver
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_cnv
+  Implicit None
+
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout)      :: sm
+  integer(psb_ipk_), intent(out)                     :: info
+  class(psb_s_base_sparse_mat), intent(in), optional :: amold
+  class(psb_s_base_vect_type), intent(in), optional  :: vmold
+  class(psb_i_base_vect_type), intent(in), optional  :: imold
+  ! Local variables
+  integer(psb_ipk_) :: i, err_act, debug_unit, debug_level
+  character(len=20) :: name='d_poly_smoother_cnv', ch_err
+
+  info=psb_success_
+  call psb_erractionsave(err_act)
+  debug_unit  = psb_get_debug_unit()
+  debug_level = psb_get_debug_level()
+
+
+
+  if (allocated(sm%sv)) &
+       & call sm%sv%cnv(info,amold=amold,vmold=vmold,imold=imold)
+  if (info /= psb_success_) then
+    call psb_errpush(psb_err_from_subroutine_,name,&
+         & a_err='solver cnv')
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+
+end subroutine amg_s_poly_smoother_cnv

amgprec/impl/smoother/amg_s_poly_smoother_csetc.f90

@@ -0,0 +1,76 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_csetc(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_csetc
+  Implicit None
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                     :: what
+  character(len=*), intent(in)                   :: val
+  integer(psb_ipk_), intent(out)                 :: info
+  integer(psb_ipk_), intent(in), optional       :: idx
+  integer(psb_ipk_) :: err_act, ival
+  character(len=20) :: name='d_poly_smoother_csetc'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(trim(what)))
+  case('POLY_VARIANT')
+    call sm%set(what,amg_stringval(val),info,idx=idx)
+  case('POLY_RHO_ESTIMATE')
+    call sm%set(what,amg_stringval(val),info,idx=idx)
+  case default
+    call sm%amg_s_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  if (info /= psb_success_) then
+    info = psb_err_from_subroutine_
+    call psb_errpush(info, name)
+    goto 9999
+  end if
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_csetc

amgprec/impl/smoother/amg_s_poly_smoother_cseti.f90

@@ -0,0 +1,92 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_cseti(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_nam => amg_s_poly_smoother_cseti
+  Implicit None
+
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                 :: what
+  integer(psb_ipk_), intent(in)                :: val
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: idx
+  integer(psb_ipk_) :: err_act
+  character(len=20)  :: name='d_poly_smoother_cseti'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(what))
+  case('POLY_DEGREE')
+    sm%pdegree = val
+  case('POLY_VARIANT')
+    select case(val)
+    case(amg_poly_lottes_,amg_poly_lottes_beta_,amg_poly_new_)
+      sm%variant = val
+    case default
+      write(0,*) 'Invalid choice for POLY_VARIANT, defaulting to amg_poly_lottes_',val
+      sm%variant = amg_poly_lottes_
+    end select
+  case('POLY_RHO_ESTIMATE')
+    select case(val)
+    case (amg_poly_rho_est_power_)
+      sm%rho_estimate = val
+    case default
+      write(0,*) 'Invalid choice for POLY_RHO_ESTIMATE, defaulting to amg_poly_rho_power'
+      sm%variant = amg_poly_rho_est_power_
+    end select      
+  case('POLY_RHO_ESTIMATE_ITERATIONS')
+    if (val>0) then 
+      sm%rho_estimate_iterations = val
+    else
+      write(0,*) 'Invalid choice for POLY_RHO_ESTIMATE_ITERATIONS, defaulting to 20'
+      sm%variant = 20
+    end if
+  case default
+    call sm%amg_s_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_cseti

amgprec/impl/smoother/amg_s_poly_smoother_csetr.f90

@@ -0,0 +1,74 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Daniela di Serafino
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_csetr(sm,what,val,info,idx)
+
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_nam => amg_s_poly_smoother_csetr
+  Implicit None
+
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(inout) :: sm
+  character(len=*), intent(in)                 :: what
+  real(psb_spk_), intent(in)                 :: val
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: idx
+  integer(psb_ipk_) :: err_act
+  character(len=20)  :: name='d_poly_smoother_csetr'
+
+  info = psb_success_
+  call psb_erractionsave(err_act)
+
+  select case(psb_toupper(what))
+  case('POLY_RHO_BA')
+    if ((szero<val).and.(val<=sone)) then 
+      sm%rho_ba  = val
+    else
+      write(0,*) 'Invalid choice for POLY_RHO_BA, defaulting to compute estimate'
+      sm%rho_ba = -sone
+    end if
+  case default
+    call sm%amg_s_base_smoother_type%set(what,val,info,idx=idx)
+  end select
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+
+  return
+end subroutine amg_s_poly_smoother_csetr

amgprec/impl/smoother/amg_s_poly_smoother_descr.f90

@@ -0,0 +1,108 @@
+!
+!
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!
+!    (C) Copyright 2021
+!
+!        Salvatore Filippone
+!        Pasqua D'Ambra
+!        Fabio Durastante
+!
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!
+!
+subroutine amg_s_poly_smoother_descr(sm,info,iout,coarse,prefix)
+
+  use psb_base_mod
+  use amg_s_diag_solver
+  use amg_s_poly_smoother, amg_protect_name => amg_s_poly_smoother_descr
+
+  Implicit None
+
+  ! Arguments
+  class(amg_s_poly_smoother_type), intent(in) :: sm
+  integer(psb_ipk_), intent(out)               :: info
+  integer(psb_ipk_), intent(in), optional      :: iout
+  logical, intent(in), optional              :: coarse
+  character(len=*), intent(in), optional  :: prefix
+
+  ! Local variables
+  integer(psb_ipk_)      :: err_act
+  character(len=20), parameter :: name='amg_s_poly_smoother_descr'
+  integer(psb_ipk_)      :: iout_
+  logical      :: coarse_
+  character(1024)    :: prefix_
+
+  call psb_erractionsave(err_act)
+  info = psb_success_
+  if (present(coarse)) then
+    coarse_ = coarse
+  else
+    coarse_ = .false.
+  end if
+  if (present(iout)) then
+    iout_ = iout
+  else
+    iout_ = psb_out_unit
+  endif
+  if (present(prefix)) then
+    prefix_ = prefix
+  else
+    prefix_ = ""
+  end if
+
+  write(iout_,*) trim(prefix_), '  Polynomial smoother   '
+  select case(sm%variant)
+  case(amg_poly_lottes_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_LOTTES'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+  case(amg_poly_lottes_beta_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_LOTTES_BETA'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+    if (allocated(sm%poly_beta)) write(iout_,*) trim(prefix_), '   Coefficients:        ',sm%poly_beta(1:sm%pdegree)
+  case(amg_poly_new_)
+    write(iout_,*) trim(prefix_), '        variant:        ','POLY_NEW'
+    write(iout_,*) trim(prefix_), '         Degree:        ',sm%pdegree
+    write(iout_,*) trim(prefix_), '         rho_ba:        ',sm%rho_ba
+    write(iout_,*) trim(prefix_), '    Coefficient:        ',sm%cf_a
+  case default
+    write(iout_,*) trim(prefix_), '        variant:        ','UNKNOWN???'
+  end select
+  if (allocated(sm%sv)) then
+    write(iout_,*) trim(prefix_), '       Local solver details:'
+    call sm%sv%descr(info,iout_,coarse=coarse,prefix=prefix)
+  end if
+
+
+  call psb_erractionrestore(err_act)
+  return
+
+9999 call psb_error_handler(err_act)
+  return
+end subroutine amg_s_poly_smoother_descr

amgprec/impl/smoother/amg_s_poly_smoother_dmp.f90

@@ -0,0 +1,90 @@
+!  
+!   
+!                             AMG4PSBLAS version 1.0
+!    Algebraic Multigrid Package
+!               based on PSBLAS (Parallel Sparse BLAS version 3.7)
+!    
+!    (C) Copyright 2021 
+!  
+!        Salvatore Filippone  
+!        Pasqua D'Ambra   
+!        Fabio Durastante        
+!   
+!    Redistribution and use in source and binary forms, with or without
+!    modification, are permitted provided that the following conditions
+!    are met:
+!      1. Redistributions of source code must retain the above copyright
+!         notice, this list of conditions and the following disclaimer.
+!      2. Redistributions in binary form must reproduce the above copyright
+!         notice, this list of conditions, and the following disclaimer in the
+!         documentation and/or other materials provided with the distribution.
+!      3. The name of the AMG4PSBLAS group or the names of its contributors may
+!         not be used to endorse or promote products derived from this
+!         software without specific written permission.
+!   
+!    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+!    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+!    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+!    PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
+!    BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+!    CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+!    SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+!    INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+!    CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+!    ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+!    POSSIBILITY OF SUCH DAMAGE.
+!   
+!  
+subroutine amg_s_poly_smoother_dmp(sm,desc,level,info,prefix,head,smoother,solver,global_num)
+  
+  use psb_base_mod
+  use amg_s_poly_smoother, amg_protect_nam => amg_s_poly_smoother_dmp
+  implicit none 
+  class(amg_s_poly_smoother_type), intent(in) :: sm
+  type(psb_desc_type), intent(in)               :: desc
+  integer(psb_ipk_), intent(in)               :: level
+  integer(psb_ipk_), intent(out)              :: info
+  character(len=*), intent(in), optional :: prefix, head
+  logical, optional, intent(in)    :: smoother, solver, global_num
+  integer(psb_ipk_)  :: i, j, il1, iln, lname, lev
+  type(psb_ctxt_type) :: ctxt
+  integer(psb_ipk_)   :: iam, np
+  character(len=80)   :: prefix_
+  character(len=120)  :: fname ! len should be at least 20 more than
+  integer(psb_lpk_), allocatable :: iv(:)
+  logical :: smoother_, global_num_
+  !  len of prefix_ 
+
+  info = 0
+
+  if (present(prefix)) then 
+    prefix_ = trim(prefix(1:min(len(prefix),len(prefix_))))
+  else
+    prefix_ = "dump_smth_d"
+  end if
+  ctxt = desc%get_context()
+  call psb_info(ctxt,iam,np)
+
+  if (present(smoother)) then 
+    smoother_ = smoother
+  else
+    smoother_ = .false. 
+  end if
+  if (present(global_num)) then 
+    global_num_ = global_num
+  else
+    global_num_ = .false. 
+  end if
+  lname = len_trim(prefix_)
+  fname = trim(prefix_)
+  write(fname(lname+1:lname+5),'(a,i3.3)') '_poly',iam
+  lname = lname + 8
+  ! to be completed
+  
+
+
+  ! At base level do nothing for the smoother
+  if (allocated(sm%sv)) &
+       & call sm%sv%dump(desc,level,info,solver=solver,prefix=prefix,global_num=global_num)
+
+end subroutine amg_s_poly_smoother_dmp

amgprec/impl/smoother/amg_z_jac_smoother_apply_vect.f90

@@ -175,7 +175,7 @@ subroutine amg_z_jac_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           if (info /= psb_success_) exit
 
           if ( sm%checkres.and.(mod(i,sm%checkiter) == 0) ) then
-            call psb_geaxpby(zone,x,zzero,r,r,desc_data,info)
+            call psb_geaxpby(zone,x,zzero,r,desc_data,info)
             call psb_spmm(-zone,sm%pa,ty,zone,r,desc_data,info)
             res  = psb_genrm2(r,desc_data,info)
             if( sm%printres ) then

amgprec/impl/solver/amg_c_ilu_solver_bld.f90

@@ -52,7 +52,7 @@ subroutine amg_c_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
   class(psb_c_base_vect_type), intent(in), optional   :: vmold
   class(psb_i_base_vect_type), intent(in), optional   :: imold
   ! Local variables
-  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota
+  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota, psb_fctype
 !!$    complex(psb_spk_), pointer :: ww(:), aux(:), tx(:),ty(:)
   type(psb_ctxt_type) :: ctxt
   integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
@@ -97,10 +97,24 @@ subroutine amg_c_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
     goto 9999      
   endif
 
-
+  select case(sv%fact_type)
+  case (amg_ilu_n_)
+    psb_fctype = psb_ilu_n_
+  case (amg_milu_n_)
+    psb_fctype = psb_milu_n_
+  case (amg_ilu_t_)
+    psb_fctype = psb_ilu_t_
+  case default
+    ! If we end up here, something was wrong up in the call chain. 
+    info = psb_err_input_value_invalid_i_
+    call psb_errpush(psb_err_input_value_invalid_i_,name,&
+         & i_err=(/ithree,sv%fact_type,izero,izero,izero/))
+    goto 9999
+  end select
+  
   select case(sv%fact_type)
 
-  case (psb_ilu_t_)
+  case (amg_ilu_t_)
     !
     ! ILU(k,t)
     !
@@ -124,7 +138,7 @@ subroutine amg_c_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       goto 9999
     end if
 
-  case(psb_ilu_n_,psb_milu_n_) 
+  case(amg_ilu_n_,amg_milu_n_) 
     !
     ! ILU(k) and MILU(k)
     !
@@ -140,17 +154,17 @@ subroutine amg_c_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       ! There seems to be a problem with the separate implementation of MILU(0),
       ! contained into psb_ilu0_fact. This must be investigated. For the time being,
       ! resort to the implementation of MILU(k) with k=0.
-      if (sv%fact_type == psb_ilu_n_) then 
-        call psb_ilu0_fact(sv%fact_type,a,sv%l,sv%u,&
+      if (sv%fact_type == amg_ilu_n_) then 
+        call psb_ilu0_fact(psb_fctype,a,sv%l,sv%u,&
              & sv%d,info,blck=b)
       else
-        call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+        call psb_iluk_fact(sv%fill_in,psb_fctype,&
              & a,sv%l,sv%u,sv%d,info,blck=b)
       endif
     case(1:)
       ! Fill-in >= 1
       ! The same routine implements both ILU(k) and MILU(k)
-      call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+      call psb_iluk_fact(sv%fill_in,psb_fctype,&
            & a,sv%l,sv%u,sv%d,info,blck=b)
     end select
     if (info /= psb_success_) then

amgprec/impl/solver/amg_d_ilu_solver_bld.f90

@@ -52,7 +52,7 @@ subroutine amg_d_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
   class(psb_d_base_vect_type), intent(in), optional   :: vmold
   class(psb_i_base_vect_type), intent(in), optional   :: imold
   ! Local variables
-  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota
+  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota, psb_fctype
 !!$    real(psb_dpk_), pointer :: ww(:), aux(:), tx(:),ty(:)
   type(psb_ctxt_type) :: ctxt
   integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
@@ -97,10 +97,24 @@ subroutine amg_d_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
     goto 9999      
   endif
 
-
+  select case(sv%fact_type)
+  case (amg_ilu_n_)
+    psb_fctype = psb_ilu_n_
+  case (amg_milu_n_)
+    psb_fctype = psb_milu_n_
+  case (amg_ilu_t_)
+    psb_fctype = psb_ilu_t_
+  case default
+    ! If we end up here, something was wrong up in the call chain. 
+    info = psb_err_input_value_invalid_i_
+    call psb_errpush(psb_err_input_value_invalid_i_,name,&
+         & i_err=(/ithree,sv%fact_type,izero,izero,izero/))
+    goto 9999
+  end select
+  
   select case(sv%fact_type)
 
-  case (psb_ilu_t_)
+  case (amg_ilu_t_)
     !
     ! ILU(k,t)
     !
@@ -124,7 +138,7 @@ subroutine amg_d_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       goto 9999
     end if
 
-  case(psb_ilu_n_,psb_milu_n_) 
+  case(amg_ilu_n_,amg_milu_n_) 
     !
     ! ILU(k) and MILU(k)
     !
@@ -140,17 +154,17 @@ subroutine amg_d_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       ! There seems to be a problem with the separate implementation of MILU(0),
       ! contained into psb_ilu0_fact. This must be investigated. For the time being,
       ! resort to the implementation of MILU(k) with k=0.
-      if (sv%fact_type == psb_ilu_n_) then 
-        call psb_ilu0_fact(sv%fact_type,a,sv%l,sv%u,&
+      if (sv%fact_type == amg_ilu_n_) then 
+        call psb_ilu0_fact(psb_fctype,a,sv%l,sv%u,&
              & sv%d,info,blck=b)
       else
-        call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+        call psb_iluk_fact(sv%fill_in,psb_fctype,&
              & a,sv%l,sv%u,sv%d,info,blck=b)
       endif
     case(1:)
       ! Fill-in >= 1
       ! The same routine implements both ILU(k) and MILU(k)
-      call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+      call psb_iluk_fact(sv%fill_in,psb_fctype,&
            & a,sv%l,sv%u,sv%d,info,blck=b)
     end select
     if (info /= psb_success_) then

amgprec/impl/solver/amg_s_ilu_solver_bld.f90

@@ -52,7 +52,7 @@ subroutine amg_s_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
   class(psb_s_base_vect_type), intent(in), optional   :: vmold
   class(psb_i_base_vect_type), intent(in), optional   :: imold
   ! Local variables
-  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota
+  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota, psb_fctype
 !!$    real(psb_spk_), pointer :: ww(:), aux(:), tx(:),ty(:)
   type(psb_ctxt_type) :: ctxt
   integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
@@ -97,10 +97,24 @@ subroutine amg_s_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
     goto 9999      
   endif
 
-
+  select case(sv%fact_type)
+  case (amg_ilu_n_)
+    psb_fctype = psb_ilu_n_
+  case (amg_milu_n_)
+    psb_fctype = psb_milu_n_
+  case (amg_ilu_t_)
+    psb_fctype = psb_ilu_t_
+  case default
+    ! If we end up here, something was wrong up in the call chain. 
+    info = psb_err_input_value_invalid_i_
+    call psb_errpush(psb_err_input_value_invalid_i_,name,&
+         & i_err=(/ithree,sv%fact_type,izero,izero,izero/))
+    goto 9999
+  end select
+  
   select case(sv%fact_type)
 
-  case (psb_ilu_t_)
+  case (amg_ilu_t_)
     !
     ! ILU(k,t)
     !
@@ -124,7 +138,7 @@ subroutine amg_s_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       goto 9999
     end if
 
-  case(psb_ilu_n_,psb_milu_n_) 
+  case(amg_ilu_n_,amg_milu_n_) 
     !
     ! ILU(k) and MILU(k)
     !
@@ -140,17 +154,17 @@ subroutine amg_s_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       ! There seems to be a problem with the separate implementation of MILU(0),
       ! contained into psb_ilu0_fact. This must be investigated. For the time being,
       ! resort to the implementation of MILU(k) with k=0.
-      if (sv%fact_type == psb_ilu_n_) then 
-        call psb_ilu0_fact(sv%fact_type,a,sv%l,sv%u,&
+      if (sv%fact_type == amg_ilu_n_) then 
+        call psb_ilu0_fact(psb_fctype,a,sv%l,sv%u,&
              & sv%d,info,blck=b)
       else
-        call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+        call psb_iluk_fact(sv%fill_in,psb_fctype,&
              & a,sv%l,sv%u,sv%d,info,blck=b)
       endif
     case(1:)
       ! Fill-in >= 1
       ! The same routine implements both ILU(k) and MILU(k)
-      call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+      call psb_iluk_fact(sv%fill_in,psb_fctype,&
            & a,sv%l,sv%u,sv%d,info,blck=b)
     end select
     if (info /= psb_success_) then

amgprec/impl/solver/amg_z_ilu_solver_bld.f90

@@ -52,7 +52,7 @@ subroutine amg_z_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
   class(psb_z_base_vect_type), intent(in), optional   :: vmold
   class(psb_i_base_vect_type), intent(in), optional   :: imold
   ! Local variables
-  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota
+  integer(psb_ipk_) :: n_row,n_col, nrow_a, nztota, psb_fctype
 !!$    complex(psb_dpk_), pointer :: ww(:), aux(:), tx(:),ty(:)
   type(psb_ctxt_type) :: ctxt
   integer(psb_ipk_)   :: np, me, i, err_act, debug_unit, debug_level
@@ -97,10 +97,24 @@ subroutine amg_z_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
     goto 9999      
   endif
 
-
+  select case(sv%fact_type)
+  case (amg_ilu_n_)
+    psb_fctype = psb_ilu_n_
+  case (amg_milu_n_)
+    psb_fctype = psb_milu_n_
+  case (amg_ilu_t_)
+    psb_fctype = psb_ilu_t_
+  case default
+    ! If we end up here, something was wrong up in the call chain. 
+    info = psb_err_input_value_invalid_i_
+    call psb_errpush(psb_err_input_value_invalid_i_,name,&
+         & i_err=(/ithree,sv%fact_type,izero,izero,izero/))
+    goto 9999
+  end select
+  
   select case(sv%fact_type)
 
-  case (psb_ilu_t_)
+  case (amg_ilu_t_)
     !
     ! ILU(k,t)
     !
@@ -124,7 +138,7 @@ subroutine amg_z_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       goto 9999
     end if
 
-  case(psb_ilu_n_,psb_milu_n_) 
+  case(amg_ilu_n_,amg_milu_n_) 
     !
     ! ILU(k) and MILU(k)
     !
@@ -140,17 +154,17 @@ subroutine amg_z_ilu_solver_bld(a,desc_a,sv,info,b,amold,vmold,imold)
       ! There seems to be a problem with the separate implementation of MILU(0),
       ! contained into psb_ilu0_fact. This must be investigated. For the time being,
       ! resort to the implementation of MILU(k) with k=0.
-      if (sv%fact_type == psb_ilu_n_) then 
-        call psb_ilu0_fact(sv%fact_type,a,sv%l,sv%u,&
+      if (sv%fact_type == amg_ilu_n_) then 
+        call psb_ilu0_fact(psb_fctype,a,sv%l,sv%u,&
              & sv%d,info,blck=b)
       else
-        call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+        call psb_iluk_fact(sv%fill_in,psb_fctype,&
              & a,sv%l,sv%u,sv%d,info,blck=b)
       endif
     case(1:)
       ! Fill-in >= 1
       ! The same routine implements both ILU(k) and MILU(k)
-      call psb_iluk_fact(sv%fill_in,sv%fact_type,&
+      call psb_iluk_fact(sv%fill_in,psb_fctype,&
            & a,sv%l,sv%u,sv%d,info,blck=b)
     end select
     if (info /= psb_success_) then