31       integer           :: numtask,i,j,k,m,p,ki,kj,krowp,ierr  local
57           allocate(lhsGP%p(redun,annz_tot),stat=ierr)
58           allocate(lhsGProwp%p(annz_tot))
59           allocate(lhsGPcolm%p(nshg+1))
60           lhsGP%p(:,:) = alhsP(:,:)
61           lhsGProwp%p(:) = arowp(:)
62           lhsGPcolm%p(:) = acolm(:)
81       allocate(sub_nnz%p(numtask))
82       allocate(sub_nnz2%p(numtask))
83       allocate(sub_nshg%p(numtask))
97          sub_nshg%p(i) = numseg
99          allocate(sub_map%pp(i)%p(numseg))
100          allocate(sub_revmap%pp(i)%p(nshg))
102          sub_map%pp(i)%p = 0
103          sub_revmap%pp(i)%p = 0
108             sub_map%pp(i)%p(j) = k
109             sub_revmap%pp(i)%p(k) = j
115          allocate(sub_colm%pp(i)%p(numseg+1))
116          allocate(sub_colm2%pp(i)%p(numseg+1))
119          sub_nnz%p(i) = 0
121             sub_colm%pp(i)%p(j) = sub_nnz%p(i) + 1
122             ki = sub_map%pp(i)%p(j)
125                if (sub_revmap%pp(i)%p(krowp).ne.0) then
133                    sub_nnz%p(i) = sub_nnz%p(i) + 1
138          sub_colm%pp(i)%p(numseg+1) = sub_nnz%p(i) + 1
141             call MPI_ISEND(sub_colm%pp(i)%p(1),numseg+1,MPI_INTEGER,
143             call MPI_IRECV(sub_colm2%pp(i)%p(1),numseg+1,MPI_INTEGER,
147             call MPI_IRECV(sub_colm2%pp(i)%p(1),numseg+1,MPI_INTEGER,
149             call MPI_ISEND(sub_colm%pp(i)%p(1),numseg+1,MPI_INTEGER,
164          sub_nnz2%p(i) = sub_colm2%pp(i)%p(sub_nshg%p(i)+1)-1
181          allocate(sub_rowp%pp(i)%p(sub_nnz%p(i)))
182          allocate(sub_rowp2%pp(i)%p(sub_nnz2%p(i)))
183          allocate(sub_rowpmap%pp(i)%p(sub_nnz%p(i)))
184          allocate(sub_mtx%pp(i)%p(redun,sub_nnz%p(i)))
185          allocate(sub_mtx2%pp(i)%p(redun,sub_nnz2%p(i)))
190             ki = sub_map%pp(i)%p(j)
193                if (sub_revmap%pp(i)%p(krowp).ne.0) then
195                    sub_rowp%pp(i)%p(k) = sub_revmap%pp(i)%p(krowp)
196                    sub_rowpmap%pp(i)%p(k) = kj
197                    do p=1,redun
198                       sub_mtx%pp(i)%p(p,k) = alhsP(p,kj)
205         call MPI_ISEND(sub_rowp%pp(i)%p(1),sub_nnz%p(i),MPI_INTEGER,
207        call MPI_IRECV(sub_rowp2%pp(i)%p(1),sub_nnz2%p(i),MPI_INTEGER,
211        call MPI_IRECV(sub_rowp2%pp(i)%p(1),sub_nnz2%p(i),MPI_INTEGER,
213         call MPI_ISEND(sub_rowp%pp(i)%p(1),sub_nnz%p(i),MPI_INTEGER,
218         call MPI_ISEND(sub_mtx%pp(i)%p(1,1),redun*sub_nnz%p(i),
221        call MPI_IRECV(sub_mtx2%pp(i)%p(1,1),redun*sub_nnz2%p(i),
226         call MPI_IRECV(sub_mtx2%pp(i)%p(1,1),redun*sub_nnz2%p(i),
229         call MPI_ISEND(sub_mtx%pp(i)%p(1,1),redun*sub_nnz%p(i),
272          do j=1,sub_nshg%p(i) ! each line
280             ki = sub_map%pp(i)%p(j)
293                    kj = Prowp%pp(ki)%p(k)
296                    tmp_rmtx(:,kj) = Pmtx%pp(ki)%p(:,k)
299                 deallocate(Prowp%pp(ki)%p)
300                 deallocate(Pmtx%pp(ki)%p)
304             do ki = sub_colm2%pp(i)%p(j),sub_colm2%pp(i)%p(j+1)-1
306                krowp = sub_rowp2%pp(i)%p(ki)
307                kj = sub_map%pp(i)%p(krowp)
320               do p=1,redun
321                tmp_rmtx(p,kj) = tmp_rmtx(p,kj)+sub_mtx2%pp(i)%p(p,ki)
326             ki = sub_map%pp(i)%p(j)
328             allocate(Prowp%pp(ki)%p(rownnz))
329             allocate(Pmtx%pp(ki)%p(redun,rownnz))
332                Prowp%pp(ki)%p(k) = kj
333                Pmtx%pp(ki)%p(:,k) = tmp_rmtx(:,kj)
345                if (Prowp%pp(ki)%p(gi).gt.Prowp%pp(ki)%p(gj)) then
346                    gk = Prowp%pp(ki)%p(gj)
347                    Prowp%pp(ki)%p(gj) = Prowp%pp(ki)%p(gi)
348                    Prowp%pp(ki)%p(gi) = gk
349                    swaptemp(:) = Pmtx%pp(ki)%p(:,gj)
350                    Pmtx%pp(ki)%p(:,gj) = Pmtx%pp(ki)%p(:,gi)
351                    Pmtx%pp(ki)%p(:,gi) = swaptemp(:)
359       allocate(lhsGPcolm%p(nshg+1))
363       lhsGPcolm%p(1) = 1
372          lhsGPcolm%p(i+1)=rownnz+1
377       allocate(lhsGProwp%p(rownnz))
378       allocate(lhsGP%p(redun,rownnz))
385                 lhsGProwp%p(rownnz) = arowp(j)
386                 lhsGP%p(:,rownnz) = alhsP(:,j)
391                 lhsGProwp%p(rownnz) = Prowp%pp(i)%p(j)
392                 lhsGP%p(:,rownnz) = Pmtx%pp(i)%p(:,j)
401          gi = lhsGPcolm%p(i)
402          if (lhsGProwp%p(gi).ne.i) then
403          do j=gi+1,lhsGPcolm%p(i+1)-1
404             k = lhsGProwp%p(j)
407                 lhsGProwp%p(j) = lhsGProwp%p(gi)
408                 lhsGProwp%p(gi) = i
410                 swaptemp(:) = lhsGP%p(:,gi)
411                 lhsGP%p(:,gi) = lhsGP%p(:,j)
412                 lhsGP%p(:,j) = swaptemp(:)
422          do j=lhsGPcolm%p(i),lhsGPcolm%p(i+1)-1
423             k = lhsGProwp%p(j)
424             gi = amg_paramap(1)%p(i)
425             gk = amg_paramap(1)%p(k)
443              deallocate(Prowp%pp(i)%p)
444              deallocate(Pmtx%pp(i)%p)
454            deallocate(sub_map%pp(i)%p)
455            deallocate(sub_revmap%pp(i)%p)
456            deallocate(sub_colm%pp(i)%p)
457            deallocate(sub_colm2%pp(i)%p)
458            deallocate(sub_rowp%pp(i)%p)
459            deallocate(sub_rowp2%pp(i)%p)
460            deallocate(sub_rowpmap%pp(i)%p)
461            deallocate(sub_mtx%pp(i)%p)
462            deallocate(sub_mtx2%pp(i)%p)
469         deallocate(sub_nnz%p)
470         deallocate(sub_nnz2%p)
471         deallocate(sub_nshg%p)
486       subroutine ramg_PPEAp(q,p,  argument
493       real(kind=8),intent(in),dimension(nshg)         :: p  local
508       diag = ramg_flowDiag%p
511       call fMtxVdimVecMult(p,amg_ppeDiag(1)%p,t1a,1,1,1,1,nshg)
532       call fMtxVdimVecMult(t1a,amg_ppeDiag(1)%p,q,1,1,1,1,nshg)
542       subroutine ramg_PPEAps(q,p,diag,  argument
549       real(kind=8),intent(in),dimension(nshg)         :: p  local
565       call fMtxVdimVecMult(p,amg_ppeDiag(1)%p,t1a,1,1,1,1,nshg)
585       call fMtxVdimVecMult(t1a,amg_ppeDiag(1)%p,q,1,1,1,1,nshg)
709          allocate(amg_ilwork(lvl)%p(tasknnztot+1+2*numtask))
710          amg_ilwork(lvl)%p = 0
712          amg_ilwork(lvl)%p(1) = numtask
720             amg_ilwork(lvl)%p(kk) = iother  ! first put iother
722             amg_ilwork(lvl)%p(kk) = tasknnz(i) ! then numseg
728                    amg_ilwork(lvl)%p(kk)=lvlmap(k)
921       numtask = amg_ilwork(level)%p(1)
931          iother(i)=amg_ilwork(level)%p(itkbeg+1)
932          numseg(i)=amg_ilwork(level)%p(itkbeg+2)
933          allocate(tmparray(i)%p(numseg(i),redun))
937          tmparray(i)%p(j,:)=global(amg_ilwork(level)%p(itkbeg),:)
955             call MPI_ISEND(tmparray(i)%p(1,1),redun*numseg(i),
960             call MPI_IRECV(tmparray(i)%p(1,1),redun*numseg(i),
975          numseg(i)=amg_ilwork(level)%p(itkbeg+2)
979             global(amg_ilwork(level)%p(itkbeg),:)=tmparray(i)%p(j,:)
984          deallocate(tmparray(i)%p)
1017       numtask = amg_ilwork(level)%p(1)
1027          iother(i)=amg_ilwork(level)%p(itkbeg+1)
1028          numseg(i)=amg_ilwork(level)%p(itkbeg+2)
1029          allocate(tmparray(i)%p(numseg(i),redun))
1033          tmparray(i)%p(j,:)=global(amg_ilwork(level)%p(itkbeg),:)
1043             call MPI_IRECV(tmparray(i)%p(1,1),redun*numseg(i),
1047             call MPI_ISEND(tmparray(i)%p(1,1),redun*numseg(i),
1062          numseg(i)=amg_ilwork(level)%p(itkbeg+2)
1067             global(amg_ilwork(level)%p(itkbeg),:)=
1068      &      global(amg_ilwork(level)%p(itkbeg),:)+tmparray(i)%p(j,:)
1070             global(amg_ilwork(level)%p(itkbeg),:)=0
1076          deallocate(tmparray(i)%p)