Merge branch 'development' into l1aggregation

This reverts commit a17f503486.

Make.inc.in

@@ -75,7 +75,7 @@ CDEFINES=$(AMGCDEFINES)
 AMGFDEFINES=@AMGFDEFINES@ $(PSBFDEFINES)
 FDEFINES=$(AMGFDEFINES)
 
-CXXDEFINES=@AMGCXXDEFINES@
+CXXDEFINES=@AMGCXXDEFINES@ $(PSBCXXDEFINES)
 
 @COMPILERULES@
 

Makefile

@@ -3,9 +3,9 @@ include Make.inc
 
 all:  objs lib
 
-objs: amgp cbnd
+objs: libdir amgp cbnd
 
-lib: libdir objs
+lib:  objs
 	cd amgprec && $(MAKE) lib
 	cd cbind && $(MAKE)  lib
 

amgprec/amg_base_prec_type.F90

@@ -326,6 +326,7 @@ module amg_base_prec_type
   integer(psb_ipk_), parameter :: amg_poly_lottes_      = 0
   integer(psb_ipk_), parameter :: amg_poly_lottes_beta_ = 1
   integer(psb_ipk_), parameter :: amg_poly_new_         = 2
+  integer(psb_ipk_), parameter :: amg_poly_dbg_         = 8
   
   integer(psb_ipk_), parameter :: amg_poly_rho_est_power_ = 0
 
@@ -575,6 +576,8 @@ contains
       val = amg_poly_lottes_beta_
     case('POLY_NEW')
       val = amg_poly_new_
+    case('POLY_DBG')
+      val = amg_poly_dbg_
     case('POLY_RHO_EST_POWER')
       val =  amg_poly_rho_est_power_
     case('A_NORMI')

amgprec/impl/aggregator/MatchBoxPC.cpp

@@ -42,7 +42,6 @@
 #include <stdlib.h>
 #if !defined(SERIAL_MPI)
 #include <mpi.h>
-#endif
 
 #include "MatchBoxPC.h"
 #ifdef __cplusplus
@@ -72,9 +71,8 @@ void dMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
         double tmr = MPI_Wtime();
     #endif
 
-// Rimosso per tornare al vecchio matching #define OMP
+#define OMP
 #ifdef OMP
-				fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
         dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(NLVer, NLEdge,
 							   verLocPtr, verLocInd, edgeLocWeight,
 							   verDistance,  Mate,
@@ -128,3 +126,4 @@ void sMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
 #ifdef __cplusplus
 }
 #endif
+#endif

amgprec/impl/aggregator/MatchBoxPC.h

@@ -59,11 +59,7 @@
 #include <assert.h>
 #include <map>
 #include <vector>
-#ifdef OMP
-// 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"
 
@@ -82,6 +78,8 @@ const int BundleTag = 9;  // Predefined tag
 
 static vector<MilanLongInt> DEFAULT_VECTOR;
 
+#if !defined(SERIAL_MPI)
+
 // MPI type map
 template <typename T>
 MPI_Datatype TypeMap();
@@ -93,6 +91,7 @@ template <>
 inline MPI_Datatype TypeMap<double>() { return MPI_DOUBLE; }
 template <>
 inline MPI_Datatype TypeMap<float>() { return MPI_FLOAT; }
+#endif
 
 #ifdef __cplusplus
 extern "C"
@@ -178,10 +177,6 @@ extern "C"
 #define MilanRealMin MINUS_INFINITY
 #endif
 
-#ifdef OMP
-/* These functions are only used in the experimental OMP implementation, if that
-is disabled there is no reason to actually compile or reference them. */
-
     // Function of find the owner of a ghost vertex using binary search:
     MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
                               MilanInt myRank, MilanInt numProcs);
@@ -428,14 +423,6 @@ is disabled there is no reason to actually compile or reference them. */
         MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
         MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
         MilanLongInt *ph1_card, MilanLongInt *ph2_card);
-#endif
-
-
-#ifndef OMP
-   //Function of find the owner of a ghost vertex using binary search:
-   inline MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
-                                        MilanInt myRank, MilanInt numProcs);
-#endif
 
     void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
         MilanLongInt NLVer, MilanLongInt NLEdge,

amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 // ***********************************************************************
 //
 //        MatchboxP: A C++ library for approximate weighted matching
@@ -222,7 +222,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
     cout << myRank << " Finished initialization" << endl;
     fflush(stdout);
 #endif
-
+   
     startTime = MPI_Wtime();
 
     /////////////////////////////////////////////////////////////////////////////////////////
@@ -391,7 +391,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
     cout << myRank << " Finished sendBundles" << endl;
     fflush(stdout);
 #endif
-
+ 
     *ph1_card = myCard;                 // Cardinality at the end of Phase-1
     startTime = MPI_Wtime();
     /////////////////////////////////////////////////////////////////////////////////////////
@@ -422,7 +422,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
         ///////////////////////////////////////////////////////////////////////////////////
         /////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
         ///////////////////////////////////////////////////////////////////////////////////
-
+      
       processMatchedVerticesAndSendMessages(NLVer,
 					    UChunkBeingProcessed,
 					    U,
@@ -456,7 +456,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
 					    comm,
 					    &msgActual,
 					    Message);
-
+      
       ///////////////////////// END OF PROCESS MATCHED VERTICES /////////////////////////
 
         //// BREAK IF NO MESSAGES EXPECTED /////////
@@ -552,4 +552,3 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
 #endif
 
 #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_daggrmat_smth_bld.f90

@@ -112,11 +112,11 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
   implicit none
 
   ! Arguments
-  type(psb_dspmat_type), intent(in)      :: a
+  type(psb_dspmat_type), intent(in)        :: a
   type(psb_desc_type), intent(inout)       :: desc_a
   integer(psb_lpk_), intent(inout)         :: ilaggr(:), nlaggr(:)
-  type(amg_dml_parms), intent(inout)      :: parms 
-  type(psb_dspmat_type), intent(inout)      :: op_prol,ac,op_restr
+  type(amg_dml_parms), intent(inout)       :: parms 
+  type(psb_dspmat_type), intent(inout)     :: op_prol,ac,op_restr
   type(psb_ldspmat_type), intent(inout)    :: t_prol
   type(psb_desc_type), intent(inout)       :: desc_ac
   integer(psb_ipk_), intent(out)           :: info
@@ -132,7 +132,7 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
   type(psb_d_coo_sparse_mat) :: coo_prol, coo_restr
   type(psb_d_csr_sparse_mat) :: acsr1,  acsrf, csr_prol, acsr
   real(psb_dpk_), allocatable :: adiag(:)
-  real(psb_dpk_), allocatable :: arwsum(:)
+  real(psb_dpk_), allocatable :: arwsum(:),l1rwsum(:)
   integer(psb_ipk_)  :: ierr(5)
   logical            :: filter_mat
   integer(psb_ipk_)            :: debug_level, debug_unit, err_act
@@ -141,6 +141,7 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
   logical, parameter :: debug_new=.false.
   character(len=80) :: filename
   logical, parameter :: do_timings=.false.
+  logical, parameter :: do_l1correction=.true.
   integer(psb_ipk_), save :: idx_spspmm=-1, idx_phase1=-1, idx_gtrans=-1, idx_phase2=-1, idx_refine=-1
   integer(psb_ipk_), save :: idx_phase3=-1, idx_cdasb=-1, idx_ptap=-1
 
@@ -200,6 +201,21 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
   if (info == psb_success_) &
        & call psb_halo(adiag,desc_a,info)
   if (info == psb_success_) call a%cp_to(acsr)
+  ! Get the l1-diagonal of D
+  if (do_l1correction) then
+    allocate(l1rwsum(nrow))
+    call acsr%arwsum(l1rwsum) 
+    if (info == psb_success_) &
+      & call psb_realloc(ncol,l1rwsum,info)
+    if (info == psb_success_) &
+      & call psb_halo(l1rwsum,desc_a,info)
+    ! \tilde{D}_{i,i} = \sum_{j \ne i} |a_{i,j}|
+    !$OMP parallel do private(i) schedule(static)
+    do i=1,size(adiag)
+      adiag(i) = adiag(i) + l1rwsum(i) - abs(adiag(i))
+    end do
+    !$OMP end parallel do 
+  end if
 
   if(info /= psb_success_) then
     call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag')
@@ -230,7 +246,7 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
 
       enddo
       if (jd == -1) then 
-        write(0,*) 'Wrong input: we need the diagonal!!!!', i
+        if (.not.do_l1correction) write(0,*) 'Wrong input: we need the diagonal!!!!', i
       else
         acsrf%val(jd)=acsrf%val(jd)-tmp
       end if
@@ -240,7 +256,6 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
     call acsrf%clean_zeros(info)
   end if
 
-
   !$OMP parallel do private(i) schedule(static)
   do i=1,size(adiag)
     if (adiag(i) /= dzero) then
@@ -249,7 +264,8 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
       adiag(i) = done
     end if
   end do
-  !$OMP end parallel do 
+  !$OMP end parallel do
+
   if (parms%aggr_omega_alg == amg_eig_est_) then 
 
     if (parms%aggr_eig == amg_max_norm_) then 
@@ -259,7 +275,9 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
       call psb_amx(ctxt,anorm)
       omega = 4.d0/(3.d0*anorm)
       parms%aggr_omega_val = omega 
-
+    else if (do_l1correction) then
+      ! For l1-Jacobi this can be estimated with 1
+      parms%aggr_omega_val = done
     else 
       info = psb_err_internal_error_
       call psb_errpush(info,name,a_err='invalid amg_aggr_eig_')
@@ -323,6 +341,9 @@ subroutine amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
        & write(debug_unit,*) me,' ',trim(name),&
        & 'Done smooth_aggregate '
   call psb_erractionrestore(err_act)
+
+  if (allocated(l1rwsum)) deallocate(l1rwsum)
+  if (allocated(arwsum)) deallocate(arwsum)
   return
 
 9999 continue

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

@@ -1,6 +1,7 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 // TODO comment
+#if !defined(SERIAL_MPI)
 
 void clean(MilanLongInt NLVer,
            MilanInt myRank,

amgprec/impl/aggregator/computeCandidateMate.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 /**
  * Execute the research fr the Candidate Mate without controlling if the vertices are already matched.
  * Returns the vertices with the highest weight
@@ -9,6 +9,8 @@
  * @param edgeLocWeight
  * @return
  */
+#if !defined(SERIAL_MPI)
+
 MilanLongInt firstComputeCandidateMate(MilanLongInt adj1,
                                        MilanLongInt adj2,
                                        MilanLongInt *verLocInd,
@@ -60,7 +62,7 @@ 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];
@@ -68,7 +70,7 @@ MilanLongInt computeCandidateMate(MilanLongInt adj1,
       }
     } // End of for loop
       //  End: PARALLEL_COMPUTE_CANDIDATE_MATE_B(v)
-
+    
     return w;
 }
 #endif

amgprec/impl/aggregator/extractUChunk.cpp

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

amgprec/impl/aggregator/findOwnerOfGhost.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 /// Find the owner of a ghost node:
 MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
                           MilanInt myRank, MilanInt numProcs)
@@ -27,4 +27,3 @@ MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
 
   return Current;
 } // End of findOwnerOfGhost()
-#endif

amgprec/impl/aggregator/initialize.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
                 MilanLongInt StartIndex, MilanLongInt EndIndex,
                 MilanLongInt *numGhostEdges,
@@ -291,7 +291,7 @@ 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);
@@ -302,4 +302,3 @@ void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
         } // End of single region
     }     // End of parallel region
 }
-#endif

amgprec/impl/aggregator/isAlreadyMatched.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 /**
  * //TODO documentation
  * @param k
@@ -32,7 +32,7 @@ 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
@@ -43,5 +43,4 @@ bool isAlreadyMatched(MilanLongInt node,
     val = Mate[node - StartIndex];
 
     return val >= 0; // Already matched
-}
-#endif
+}

amgprec/impl/aggregator/parallelComputeCandidateMateB.cpp

@@ -1,5 +1,6 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+#if !defined(SERIAL_MPI)
+
 void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
                                               MilanLongInt *verLocPtr,
                                               MilanLongInt *verLocInd,

amgprec/impl/aggregator/processCrossEdge.cpp

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

amgprec/impl/aggregator/processExposedVertex.cpp

@@ -1,5 +1,6 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+#if !defined(SERIAL_MPI)
+
 void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
                                        MilanLongInt *candidateMate,
                                        MilanLongInt *verLocInd,
@@ -29,7 +30,7 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
                                        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;
@@ -79,7 +80,7 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
 					       Ghost2LocalMap);
 		      candidateMate[v] = w;
                     }
-
+		    
                     if (w >= 0) {
 		      (*myCard)++;
 		      if ((w < StartIndex) || (w > EndIndex)) { // w is a ghost vertex
@@ -88,29 +89,29 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
 			  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 critical processExposed
-
+		
             } // End of if(w >=0)
             else  {
 	      // This piece of code is executed a really small amount of times
@@ -131,17 +132,17 @@ void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
 		  // assert(ghostOwner != -1);
 		  // assert(ghostOwner != myRank);
 		  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:

amgprec/impl/aggregator/processMatchedVertices.cpp

@@ -1,5 +1,6 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
+#if !defined(SERIAL_MPI)
 void processMatchedVertices(
     MilanLongInt NLVer,
     vector<MilanLongInt> &UChunkBeingProcessed,
@@ -58,29 +59,29 @@ 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);
@@ -102,27 +103,27 @@ void processMatchedVertices(
 						 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 >= 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
@@ -130,7 +131,7 @@ void processMatchedVertices(
 			      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);
@@ -146,7 +147,7 @@ void processMatchedVertices(
 		  }         // 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)
@@ -155,7 +156,7 @@ void processMatchedVertices(
 		      option = 5; // u is local
 		  }                   // End of critical
 		}                       // End of Else //A Ghost Vertex
-
+		
 		switch (option)
 		  {
 		  case -1:
@@ -165,7 +166,7 @@ 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 << ") ";
@@ -174,7 +175,7 @@ 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);
@@ -182,7 +183,7 @@ void processMatchedVertices(
 		    PCounter[ghostOwner]++;
 		    (*NumMessagesBundled)++;
 		    (*msgInd)++;
-
+		    
 		    privateQLocalVtx.push_back(v);
 		    privateQGhostVtx.push_back(w);
 		    privateQMsgType.push_back(REQUEST);
@@ -191,7 +192,7 @@ void processMatchedVertices(
 		  case 3:
 		    privateU.push_back(v);
 		    privateU.push_back(w);
-
+		    
 		    (*myCard)++;
 		    break;
 		  case 4:
@@ -201,94 +202,94 @@ 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);
-
+			
 			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());
 	  }
-
+	  
 	  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

@@ -1,6 +1,6 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
 //#define DEBUG_HANG_
+#if !defined(SERIAL_MPI)
 void processMatchedVerticesAndSendMessages(
 					   MilanLongInt NLVer,
 					   vector<MilanLongInt> &UChunkBeingProcessed,
@@ -64,29 +64,29 @@ 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);
@@ -108,28 +108,28 @@ void processMatchedVerticesAndSendMessages(
 						 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
@@ -137,7 +137,7 @@ void processMatchedVerticesAndSendMessages(
 			      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);
@@ -153,7 +153,7 @@ void processMatchedVerticesAndSendMessages(
 		  }         // 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 +162,7 @@ void processMatchedVerticesAndSendMessages(
 		      option = 5; // u is local
 		  }                   // End of critical
 		}                       // End of Else //A Ghost Vertex
-
+		
 		switch (option)
 		  {
 		  case -1:
@@ -180,20 +180,20 @@ 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 +211,94 @@ 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);
-
+	
 #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

@@ -1,6 +1,6 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
 //#define DEBUG_HANG_
+#if !defined(SERIAL_MPI)
 
 void processMessages(
     MilanLongInt NLVer,
@@ -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 )
@@ -250,7 +250,7 @@ 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 +311,7 @@ void processMessages(
 	}                                                       // End of else: CASE III
       }                                                           // End of else: CASE I
     }
-
+    
     return;
 }
 #endif

amgprec/impl/aggregator/queueTransfer.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+
 void queuesTransfer(vector<MilanLongInt> &U,
                     vector<MilanLongInt> &privateU,
                     vector<MilanLongInt> &QLocalVtx,
@@ -34,4 +34,3 @@ void queuesTransfer(vector<MilanLongInt> &U,
     privateQOwner.clear();
 
 }
-#endif

amgprec/impl/aggregator/sendBundledMessages.cpp

@@ -1,5 +1,5 @@
 #include "MatchBoxPC.h"
-#ifdef OMP
+#if !defined(SERIAL_MPI)
 void sendBundledMessages(MilanLongInt *numGhostEdges,
                                 MilanInt *BufferSize,
                                 MilanLongInt *Buffer,
@@ -62,7 +62,7 @@ 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 +84,7 @@ void sendBundledMessages(MilanLongInt *numGhostEdges,
     PCounter[QOwner[i]]++;
   }
  }
-
+ 
 // Send the Bundled Messages: Use ISend
 #pragma omp task depend(out \
                         : SRequest, SStatus)
@@ -101,7 +101,7 @@ void sendBundledMessages(MilanLongInt *numGhostEdges,
        exit(1);
      }
  }
-
+ 
 // Send the Messages
 #pragma omp task depend(inout                                                  \
                         : SRequest, PSizeInfoMessages, PCumulative) depend(out \

amgprec/impl/level/amg_c_base_onelev_csetc.F90

@@ -42,6 +42,8 @@ subroutine amg_c_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_c_base_aggregator_mod
   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
@@ -241,6 +243,8 @@ subroutine amg_c_base_onelev_csetc(lv,what,val,info,pos,idx)
       if (info == 0) deallocate(lv%aggr,stat=info)
       if (info /= 0) then
         info = psb_err_internal_error_
+        call psb_errpush(info,name,a_err='aggregator deallocation?')
+        goto 9999
         return
       end if
     end if
@@ -250,8 +254,16 @@ subroutine amg_c_base_onelev_csetc(lv,what,val,info,pos,idx)
       allocate(amg_c_dec_aggregator_type :: lv%aggr, stat=info)
     case('SYMDEC')
       allocate(amg_c_symdec_aggregator_type :: lv%aggr, stat=info)
-    case default
+#if !defined(SERIAL_MPI)
+#endif
+  case default
       info =  psb_err_internal_error_
+#if !defined(SERIAL_MPI)
+      call psb_errpush(info,name,a_err='Unsupported PAR_AGGR_ALG')
+#else
+      call psb_errpush(info,name,a_err='PAR_AGGR_ALG unsupported (SERIAL_MPI on)')
+#endif
+      goto 9999
     end select
     if (info == psb_success_) call lv%aggr%default()
 

amgprec/impl/level/amg_c_base_onelev_memory_use.f90

@@ -89,7 +89,7 @@ subroutine amg_c_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefi
   if (present(global)) then
     global_ = global
   else
-    global_ = .false.
+    global_ = .true.
   end if
 
   if (present(prefix)) then

amgprec/impl/level/amg_d_base_onelev_csetc.F90

@@ -42,7 +42,9 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_d_base_aggregator_mod
   use amg_d_dec_aggregator_mod
   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
@@ -267,6 +269,8 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
       if (info == 0) deallocate(lv%aggr,stat=info)
       if (info /= 0) then
         info = psb_err_internal_error_
+        call psb_errpush(info,name,a_err='aggregator deallocation?')
+        goto 9999
         return
       end if
     end if
@@ -276,10 +280,18 @@ subroutine amg_d_base_onelev_csetc(lv,what,val,info,pos,idx)
       allocate(amg_d_dec_aggregator_type :: lv%aggr, stat=info)
     case('SYMDEC')
       allocate(amg_d_symdec_aggregator_type :: lv%aggr, stat=info)
+#if !defined(SERIAL_MPI)
     case('COUP','COUPLED')
       allocate(amg_d_parmatch_aggregator_type :: lv%aggr, stat=info)
-    case default
+#endif
+  case default
       info =  psb_err_internal_error_
+#if !defined(SERIAL_MPI)
+      call psb_errpush(info,name,a_err='Unsupported PAR_AGGR_ALG')
+#else
+      call psb_errpush(info,name,a_err='PAR_AGGR_ALG unsupported (SERIAL_MPI on)')
+#endif
+      goto 9999
     end select
     if (info == psb_success_) call lv%aggr%default()
 

amgprec/impl/level/amg_d_base_onelev_memory_use.f90

@@ -89,7 +89,7 @@ subroutine amg_d_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefi
   if (present(global)) then
     global_ = global
   else
-    global_ = .false.
+    global_ = .true.
   end if
 
   if (present(prefix)) then

amgprec/impl/level/amg_s_base_onelev_csetc.F90

@@ -42,7 +42,9 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_s_base_aggregator_mod
   use amg_s_dec_aggregator_mod
   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
@@ -247,6 +249,8 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
       if (info == 0) deallocate(lv%aggr,stat=info)
       if (info /= 0) then
         info = psb_err_internal_error_
+        call psb_errpush(info,name,a_err='aggregator deallocation?')
+        goto 9999
         return
       end if
     end if
@@ -256,10 +260,18 @@ subroutine amg_s_base_onelev_csetc(lv,what,val,info,pos,idx)
       allocate(amg_s_dec_aggregator_type :: lv%aggr, stat=info)
     case('SYMDEC')
       allocate(amg_s_symdec_aggregator_type :: lv%aggr, stat=info)
+#if !defined(SERIAL_MPI)
     case('COUP','COUPLED')
       allocate(amg_s_parmatch_aggregator_type :: lv%aggr, stat=info)
-    case default
+#endif
+  case default
       info =  psb_err_internal_error_
+#if !defined(SERIAL_MPI)
+      call psb_errpush(info,name,a_err='Unsupported PAR_AGGR_ALG')
+#else
+      call psb_errpush(info,name,a_err='PAR_AGGR_ALG unsupported (SERIAL_MPI on)')
+#endif
+      goto 9999
     end select
     if (info == psb_success_) call lv%aggr%default()
 

amgprec/impl/level/amg_s_base_onelev_memory_use.f90

@@ -89,7 +89,7 @@ subroutine amg_s_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefi
   if (present(global)) then
     global_ = global
   else
-    global_ = .false.
+    global_ = .true.
   end if
 
   if (present(prefix)) then

amgprec/impl/level/amg_z_base_onelev_csetc.F90

@@ -42,6 +42,8 @@ subroutine amg_z_base_onelev_csetc(lv,what,val,info,pos,idx)
   use amg_z_base_aggregator_mod
   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
@@ -261,6 +263,8 @@ subroutine amg_z_base_onelev_csetc(lv,what,val,info,pos,idx)
       if (info == 0) deallocate(lv%aggr,stat=info)
       if (info /= 0) then
         info = psb_err_internal_error_
+        call psb_errpush(info,name,a_err='aggregator deallocation?')
+        goto 9999
         return
       end if
     end if
@@ -270,8 +274,16 @@ subroutine amg_z_base_onelev_csetc(lv,what,val,info,pos,idx)
       allocate(amg_z_dec_aggregator_type :: lv%aggr, stat=info)
     case('SYMDEC')
       allocate(amg_z_symdec_aggregator_type :: lv%aggr, stat=info)
-    case default
+#if !defined(SERIAL_MPI)
+#endif
+  case default
       info =  psb_err_internal_error_
+#if !defined(SERIAL_MPI)
+      call psb_errpush(info,name,a_err='Unsupported PAR_AGGR_ALG')
+#else
+      call psb_errpush(info,name,a_err='PAR_AGGR_ALG unsupported (SERIAL_MPI on)')
+#endif
+      goto 9999
     end select
     if (info == psb_success_) call lv%aggr%default()
 

amgprec/impl/level/amg_z_base_onelev_memory_use.f90

@@ -89,7 +89,7 @@ subroutine amg_z_base_onelev_memory_use(lv,il,nl,ilmin,info,iout,verbosity,prefi
   if (present(global)) then
     global_ = global
   else
-    global_ = .false.
+    global_ = .true.
   end if
 
   if (present(prefix)) then

amgprec/impl/smoother/amg_d_poly_smoother_apply_vect.f90

@@ -36,7 +36,7 @@
 !
 !
 subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
-     & sweeps,work,wv,info,init,initu)
+     & sweeps,work,wv,info,init,initu) 
 
   use psb_base_mod
   use amg_d_diag_solver
@@ -55,6 +55,10 @@ subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
   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
@@ -92,7 +96,19 @@ subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
     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()
 
@@ -125,38 +141,39 @@ subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
 
     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
+        do i=1, sm%pdegree-1
           !   B r_{k-1}
-          call sm%sv%apply(done,r,dzero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+          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 (.false.) then
-            ! z_k =  cz z_{k-1} + cr ty = cz z_{k-1} + cr Br_{k-1}
-            call psb_geaxpby(cr,ty,cz,tz,desc_data,info)
-            ! r_k =    b-Ax_k  = x -A tx
-            call psb_geaxpby(done,tz,done,tx,desc_data,info)
-          else
-            call psb_abgdxyz(cr,cz,done,done,ty,tz,tx,desc_data,info)
-          end if
-          if (.false.) then
-            call psb_geaxpby(done,x,dzero,r,desc_data,info)
-            call psb_spmm(-done,sm%pa,tx,done,r,desc_data,info,work=aux,trans=trans_)
-          else
-            call psb_spmm(-done,sm%pa,tz,done,r,desc_data,info,work=aux,trans=trans_)
-          end if
-!!$        res  = psb_genrm2(r,desc_data,info)
-!!$        write(0,*) 'Polynomial smoother  LOTTES',i,res
-          ! x_k =  x_{k-1} + z_k
+          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
@@ -170,32 +187,30 @@ subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
         end if
 
-        do i=1, sm%pdegree
+        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 (.false.) then
-            ! z_k =  cz z_{k-1} + cr ty = cz z_{k-1} + cr Br_{k-1}
-            call psb_geaxpby(cr,ty,cz,tz,desc_data,info)
-            ! r_k =    b-Ax_k  = x -A tx
-            call psb_geaxpby(sm%poly_beta(i),tz,done,tx,desc_data,info)
-          else
-            call psb_abgdxyz(cr,cz,sm%poly_beta(i),done,ty,tz,tx,desc_data,info)
-          end if
-          if (.false.) then
-            call psb_geaxpby(done,x,dzero,r,desc_data,info)
-            call psb_spmm(-done,sm%pa,tx,done,r,desc_data,info,work=aux,trans=trans_)
-          else
-            call psb_spmm(-done,sm%pa,tz,done,r,desc_data,info,work=aux,trans=trans_)
-          end if
-!!$        res  = psb_genrm2(r,desc_data,info)
-!!$        write(0,*) 'Polynomial smoother LOTTES_BETA',i,res
-          ! x_k =  x_{k-1} + z_k
+          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
@@ -206,40 +221,35 @@ subroutine amg_d_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
         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 (.false.) then
-          call psb_geaxpby((done/theta),r,dzero,tz,desc_data,info)
-          call psb_geaxpby(done,tz,done,tx,desc_data,info)
-        else
-          call psb_abgdxyz((done/theta),dzero,done,done,r,tz,tx,desc_data,info)
-        end if
+        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 (.false.) then
-            call psb_geaxpby((2*rho/delta),r,(rho*rho_old),tz,desc_data,info)
-            call psb_geaxpby(done,tz,done,tx,desc_data,info)
-          else
-            call psb_abgdxyz((2*rho/delta),(rho*rho_old),done,done,r,tz,tx,desc_data,info)
-          end if
-!!$          res  = psb_genrm2(r,desc_data,info)
-!!$          write(0,*) 'Polynomial smoother NEW ',i,res
-          ! x_k =  x_{k-1} + z_k
+          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,&

amgprec/impl/smoother/amg_s_poly_smoother_apply_vect.f90

@@ -36,7 +36,7 @@
 !
 !
 subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
-     & sweeps,work,wv,info,init,initu)
+     & sweeps,work,wv,info,init,initu) 
 
   use psb_base_mod
   use amg_s_diag_solver
@@ -55,6 +55,10 @@ subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
   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
@@ -92,7 +96,19 @@ subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
     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()
 
@@ -125,38 +141,39 @@ subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
 
     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
+        do i=1, sm%pdegree-1
           !   B r_{k-1}
-          call sm%sv%apply(sone,r,szero,ty,desc_data,trans_,aux,wv(5:),info,init='Z')
+          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 (.false.) then
-            ! z_k =  cz z_{k-1} + cr ty = cz z_{k-1} + cr Br_{k-1}
-            call psb_geaxpby(cr,ty,cz,tz,desc_data,info)
-            ! r_k =    b-Ax_k  = x -A tx
-            call psb_geaxpby(sone,tz,sone,tx,desc_data,info)
-          else
-            call psb_abgdxyz(cr,cz,sone,sone,ty,tz,tx,desc_data,info)
-          end if
-          if (.false.) then
-            call psb_geaxpby(sone,x,szero,r,desc_data,info)
-            call psb_spmm(-sone,sm%pa,tx,sone,r,desc_data,info,work=aux,trans=trans_)
-          else
-            call psb_spmm(-sone,sm%pa,tz,sone,r,desc_data,info,work=aux,trans=trans_)
-          end if
-!!$        res  = psb_genrm2(r,desc_data,info)
-!!$        write(0,*) 'Polynomial smoother  LOTTES',i,res
-          ! x_k =  x_{k-1} + z_k
+          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
@@ -170,32 +187,30 @@ subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
           sm%poly_beta(1:sm%pdegree) = amg_d_poly_beta_mat(1:sm%pdegree,sm%pdegree)
         end if
 
-        do i=1, sm%pdegree
+        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 (.false.) then
-            ! z_k =  cz z_{k-1} + cr ty = cz z_{k-1} + cr Br_{k-1}
-            call psb_geaxpby(cr,ty,cz,tz,desc_data,info)
-            ! r_k =    b-Ax_k  = x -A tx
-            call psb_geaxpby(sm%poly_beta(i),tz,sone,tx,desc_data,info)
-          else
-            call psb_abgdxyz(cr,cz,sm%poly_beta(i),sone,ty,tz,tx,desc_data,info)
-          end if
-          if (.false.) then
-            call psb_geaxpby(sone,x,szero,r,desc_data,info)
-            call psb_spmm(-sone,sm%pa,tx,sone,r,desc_data,info,work=aux,trans=trans_)
-          else
-            call psb_spmm(-sone,sm%pa,tz,sone,r,desc_data,info,work=aux,trans=trans_)
-          end if
-!!$        res  = psb_genrm2(r,desc_data,info)
-!!$        write(0,*) 'Polynomial smoother LOTTES_BETA',i,res
-          ! x_k =  x_{k-1} + z_k
+          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
@@ -206,40 +221,35 @@ subroutine amg_s_poly_smoother_apply_vect(alpha,sm,x,beta,y,desc_data,trans,&
         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 (.false.) then
-          call psb_geaxpby((sone/theta),r,szero,tz,desc_data,info)
-          call psb_geaxpby(sone,tz,sone,tx,desc_data,info)
-        else
-          call psb_abgdxyz((sone/theta),szero,sone,sone,r,tz,tx,desc_data,info)
-        end if
+        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 (.false.) then
-            call psb_geaxpby((2*rho/delta),r,(rho*rho_old),tz,desc_data,info)
-            call psb_geaxpby(sone,tz,sone,tx,desc_data,info)
-          else
-            call psb_abgdxyz((2*rho/delta),(rho*rho_old),sone,sone,r,tz,tx,desc_data,info)
-          end if
-!!$          res  = psb_genrm2(r,desc_data,info)
-!!$          write(0,*) 'Polynomial smoother NEW ',i,res
-          ! x_k =  x_{k-1} + z_k
+          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,&

samples/advanced/pdegen/amg_d_pde2d.F90

@@ -126,7 +126,7 @@ program amg_d_pde2d
     ! AMG cycles for ML
     ! general AMG data
     character(len=32)  :: mlcycle      ! AMG cycle type
-    integer(psb_ipk_)  :: maxlevs      ! maximum number of levels in AMG preconditioner
+    integer(psb_ipk_)  :: maxlevs     ! maximum number of levels in AMG preconditioner
 
     ! AMG aggregation
     character(len=32)  :: aggr_prol    ! aggregation type: SMOOTHED, NONSMOOTHED
@@ -174,15 +174,15 @@ program amg_d_pde2d
     real(psb_dpk_)      :: thr2         ! threshold for ILUT factorization
 
     ! coarsest-level solver
-    character(len=32)  :: cmat         ! coarsest matrix layout: REPL, DIST
-    character(len=32)  :: csolve       ! coarsest-lev solver: BJAC, SLUDIST (distr.
-                                       ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU
-                                       ! (repl. mat.)
-    character(len=32)  :: csbsolve     ! coarsest-lev local subsolver: ILU, ILUT,
-                                       ! MILU, UMF, MUMPS, SLU
-    integer(psb_ipk_)  :: cfill        ! fill-in for incomplete LU factorization
-    real(psb_dpk_)     :: cthres        ! threshold for ILUT factorization
-    integer(psb_ipk_)  :: cjswp        ! sweeps for GS or JAC coarsest-lev subsolver
+    character(len=32)  :: cmat        ! coarsest matrix layout: REPL, DIST
+    character(len=32)  :: csolve      ! coarsest-lev solver: BJAC, SLUDIST (distr.
+                                      ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU
+                                      ! (repl. mat.)
+    character(len=32)  :: csbsolve    ! coarsest-lev local subsolver: ILU, ILUT,
+                                      ! MILU, UMF, MUMPS, SLU
+    integer(psb_ipk_)  :: cfill       ! fill-in for incomplete LU factorization
+    real(psb_dpk_)     :: cthres      ! threshold for ILUT factorization
+    integer(psb_ipk_)  :: cjswp       ! sweeps for GS or JAC coarsest-lev subsolver
 
     ! Dump data
     logical            :: dump = .false.

samples/advanced/pdegen/amg_s_pde2d.F90

@@ -126,16 +126,16 @@ program amg_s_pde2d
     ! AMG cycles for ML
     ! general AMG data
     character(len=32)  :: mlcycle      ! AMG cycle type
-    integer(psb_ipk_)  :: maxlevs      ! maximum number of levels in AMG preconditioner
+    integer(psb_ipk_)  :: maxlevs     ! maximum number of levels in AMG preconditioner
 
     ! AMG aggregation
     character(len=32)  :: aggr_prol    ! aggregation type: SMOOTHED, NONSMOOTHED
     character(len=32)  :: par_aggr_alg ! parallel aggregation algorithm: DEC, SYMDEC
-    character(len=32)  :: aggr_type    ! Type of aggregation SOC1, SOC2, MATCHBOXP
-    integer(psb_ipk_)  :: aggr_size    ! Requested size of the aggregates for MATCHBOXP
-    character(len=32)  :: aggr_ord     ! ordering for aggregation: NATURAL, DEGREE
-    character(len=32)  :: aggr_filter  ! filtering: FILTER, NO_FILTER
-    real(psb_spk_)      :: mncrratio    ! minimum aggregation ratio
+    character(len=32)  :: aggr_type   ! Type of aggregation SOC1, SOC2, MATCHBOXP
+    integer(psb_ipk_)  :: aggr_size   ! Requested size of the aggregates for MATCHBOXP
+    character(len=32)  :: aggr_ord    ! ordering for aggregation: NATURAL, DEGREE
+    character(len=32)  :: aggr_filter ! filtering: FILTER, NO_FILTER
+    real(psb_spk_)     :: mncrratio  ! minimum aggregation ratio
     real(psb_spk_), allocatable :: athresv(:) ! smoothed aggregation threshold vector
     integer(psb_ipk_)  :: thrvsz       ! size of threshold vector
     real(psb_spk_)      :: athres       ! smoothed aggregation threshold
@@ -174,15 +174,15 @@ program amg_s_pde2d
     real(psb_spk_)      :: thr2         ! threshold for ILUT factorization
 
     ! coarsest-level solver
-    character(len=32)  :: cmat         ! coarsest matrix layout: REPL, DIST
-    character(len=32)  :: csolve       ! coarsest-lev solver: BJAC, SLUDIST (distr.
-                                       ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU
-                                       ! (repl. mat.)
-    character(len=32)  :: csbsolve     ! coarsest-lev local subsolver: ILU, ILUT,
-                                       ! MILU, UMF, MUMPS, SLU
-    integer(psb_ipk_)  :: cfill        ! fill-in for incomplete LU factorization
-    real(psb_spk_)     :: cthres        ! threshold for ILUT factorization
-    integer(psb_ipk_)  :: cjswp        ! sweeps for GS or JAC coarsest-lev subsolver
+    character(len=32)  :: cmat        ! coarsest matrix layout: REPL, DIST
+    character(len=32)  :: csolve      ! coarsest-lev solver: BJAC, SLUDIST (distr.
+                                      ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU
+                                      ! (repl. mat.)
+    character(len=32)  :: csbsolve    ! coarsest-lev local subsolver: ILU, ILUT,
+                                      ! MILU, UMF, MUMPS, SLU
+    integer(psb_ipk_)  :: cfill       ! fill-in for incomplete LU factorization
+    real(psb_spk_)     :: cthres      ! threshold for ILUT factorization
+    integer(psb_ipk_)  :: cjswp       ! sweeps for GS or JAC coarsest-lev subsolver
 
     ! Dump data
     logical            :: dump = .false.

samples/advanced/pdegen/runs/amg_pde3d.inp

@@ -1,10 +1,10 @@
 %%%%%%%%%%%  General  arguments % Lines starting with % are ignored.
 CSR                         ! Storage format CSR COO JAD
-0150                        ! IDIM; domain size. Linear system size is IDIM**3
-CONST                       ! PDECOEFF: CONST, EXP, BOX, GAUSS Coefficients of the PDE
+0040                        ! IDIM; domain size. Linear system size is IDIM**3
+CONST                       ! PDECOEFF: CONST, EXP, GAUSS Coefficients of the PDE
 CG                          ! Iterative method: BiCGSTAB BiCGSTABL BiCG CG CGS FCG GCR RGMRES
 2                           ! ISTOPC
-00500                       ! ITMAX
+00008                       ! ITMAX
 1                           ! ITRACE
 30                          ! IRST (restart for RGMRES and BiCGSTABL)
 1.d-6                       ! EPS