From 029c78bbf48e7c401b5114ba226b8ef8a35760e0 Mon Sep 17 00:00:00 2001 From: Salvatore Filippone Date: Wed, 11 Apr 2012 13:13:04 +0000 Subject: [PATCH] mld2p4-tpre-NewNL: mlprec/impl/mld_caggrmat_minnrg_asb.F90 mlprec/impl/mld_caggrmat_nosmth_asb.F90 mlprec/impl/mld_caggrmat_smth_asb.F90 mlprec/impl/mld_daggrmat_minnrg_asb.F90 mlprec/impl/mld_daggrmat_nosmth_asb.F90 mlprec/impl/mld_daggrmat_smth_asb.F90 mlprec/impl/mld_saggrmat_minnrg_asb.F90 mlprec/impl/mld_saggrmat_nosmth_asb.F90 mlprec/impl/mld_saggrmat_smth_asb.F90 mlprec/impl/mld_zaggrmat_minnrg_asb.F90 mlprec/impl/mld_zaggrmat_nosmth_asb.F90 mlprec/impl/mld_zaggrmat_smth_asb.F90 MPI no longer used directly. Step 1. --- mlprec/impl/mld_caggrmat_biz_asb.F90 | 523 +++++++++++++++++++++++++++ mlprec/impl/mld_daggrmat_biz_asb.F90 | 523 +++++++++++++++++++++++++++ mlprec/impl/mld_saggrmat_biz_asb.F90 | 523 +++++++++++++++++++++++++++ mlprec/impl/mld_zaggrmat_biz_asb.F90 | 523 +++++++++++++++++++++++++++ 4 files changed, 2092 insertions(+) create mode 100644 mlprec/impl/mld_caggrmat_biz_asb.F90 create mode 100644 mlprec/impl/mld_daggrmat_biz_asb.F90 create mode 100644 mlprec/impl/mld_saggrmat_biz_asb.F90 create mode 100644 mlprec/impl/mld_zaggrmat_biz_asb.F90 diff --git a/mlprec/impl/mld_caggrmat_biz_asb.F90 b/mlprec/impl/mld_caggrmat_biz_asb.F90 new file mode 100644 index 00000000..97d1ee4a --- /dev/null +++ b/mlprec/impl/mld_caggrmat_biz_asb.F90 @@ -0,0 +1,523 @@ +!!$ +!!$ +!!$ MLD2P4 version 2.0 +!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package +!!$ based on PSBLAS (Parallel Sparse BLAS version 3.0) +!!$ +!!$ (C) Copyright 2008,2009,2010 +!!$ +!!$ Salvatore Filippone University of Rome Tor Vergata +!!$ Alfredo Buttari CNRS-IRIT, Toulouse +!!$ Pasqua D'Ambra ICAR-CNR, Naples +!!$ Daniela di Serafino Second University of Naples +!!$ +!!$ 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 MLD2P4 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 MLD2P4 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: mld_caggrmat_biz_asb.F90 +! +! Subroutine: mld_caggrmat_biz_asb +! Version: complex +! +! This routine builds a coarse-level matrix A_C from a fine-level matrix A +! by using the Galerkin approach, i.e. +! +! A_C = P_C^T A P_C, +! +! where P_C is a prolongator from the coarse level to the fine one. +! +! This routine builds A_C according to a "bizarre" aggregation algorithm, +! using a "naive" prolongator proposed by the authors of MLD2P4. However, this +! prolongator still requires a deep analysis and testing and its use is not +! recommended. +! +! The coarse-level matrix A_C is distributed among the parallel processes or +! replicated on each of them, according to the value of p%parms%coarse_mat, +! specified by the user through mld_cprecinit and mld_zprecset. +! +! Arguments: +! a - type(psb_cspmat_type), input. +! The sparse matrix structure containing the local part of +! the fine-level matrix. +! desc_a - type(psb_desc_type), input. +! The communication descriptor of the fine-level matrix. +! p - type(mld_c_onelev_type), input/output. +! The 'one-level' data structure that will contain the local +! part of the matrix to be built as well as the information +! concerning the prolongator and its transpose. +! ilaggr - integer, dimension(:), allocatable. +! The mapping between the row indices of the coarse-level +! matrix and the row indices of the fine-level matrix. +! ilaggr(i)=j means that node i in the adjacency graph +! of the fine-level matrix is mapped onto node j in the +! adjacency graph of the coarse-level matrix. +! nlaggr - integer, dimension(:), allocatable. +! nlaggr(i) contains the aggregates held by process i. +! info - integer, output. +! Error code. +! +subroutine mld_caggrmat_biz_asb(a,desc_a,ilaggr,nlaggr,p,info) + use psb_base_mod + use mld_c_inner_mod, mld_protect_name => mld_caggrmat_biz_asb + + implicit none + + ! Arguments + type(psb_cspmat_type), intent(in) :: a + type(psb_desc_type), intent(in) :: desc_a + integer, intent(inout) :: ilaggr(:), nlaggr(:) + type(mld_c_onelev_type), intent(inout), target :: p + integer, intent(out) :: info + type(psb_cspmat_type) :: op_prol, op_restr, b + + ! Local variables + integer :: nrow, nglob, ncol, ntaggr, ip, ndx,& + & naggr, nzl,naggrm1,naggrp1, i, j, k, jd, icolF, nrw + integer ::ictxt, np, me, err_act + character(len=20) :: name + type(psb_cspmat_type) :: am3, am4 + type(psb_c_coo_sparse_mat) :: acoo, acoof, bcoo + type(psb_c_csr_sparse_mat) :: acsr1, acsr2, acsr3, acsrf, ptilde + complex(psb_spk_), allocatable :: adiag(:) + integer(psb_ipk_) :: ierr(5) + logical :: filter_mat + integer :: debug_level, debug_unit + integer, parameter :: ncmax=16 + real(psb_spk_) :: anorm, omega, tmp, dg, theta + + name='mld_aggrmat_biz_asb' + if(psb_get_errstatus().ne.0) return + info=psb_success_ + call psb_erractionsave(err_act) + debug_unit = psb_get_debug_unit() + debug_level = psb_get_debug_level() + + ictxt = desc_a%get_context() + ictxt = desc_a%get_context() + + call psb_info(ictxt, me, np) + + nglob = desc_a%get_global_rows() + nrow = desc_a%get_local_rows() + ncol = desc_a%get_local_cols() + + theta = p%parms%aggr_thresh + + naggr = nlaggr(me+1) + ntaggr = sum(nlaggr) + + naggrm1 = sum(nlaggr(1:me)) + naggrp1 = sum(nlaggr(1:me+1)) + + filter_mat = (p%parms%aggr_filter == mld_filter_mat_) + + ilaggr(1:nrow) = ilaggr(1:nrow) + naggrm1 + call psb_halo(ilaggr,desc_a,info) + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_halo') + goto 9999 + end if + + ! naggr: number of local aggregates + ! nrow: local rows. + ! + allocate(adiag(ncol),stat=info) + + if (info /= psb_success_) then + info=psb_err_alloc_request_; ierr(1)=nrow; + call psb_errpush(info,name,i_err=ierr,a_err='complex(psb_spk_)') + goto 9999 + end if + + ! Get the diagonal D + call a%get_diag(adiag,info) + if (info == psb_success_) & + & call psb_halo(adiag,desc_a,info) + + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag') + goto 9999 + end if + + ! 1. Allocate Ptilde in sparse matrix form + call acoo%allocate(ncol,naggr,ncol) + do i=1,nrow + acoo%val(i) = cone + acoo%ia(i) = i + acoo%ja(i) = ilaggr(i) + end do + call acoo%set_nzeros(nrow) + call acoo%set_dupl(psb_dupl_add_) + + call ptilde%mv_from_coo(acoo,info) + if (info == psb_success_) call a%cscnv(acsr3,info,dupl=psb_dupl_add_) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & ' Initial copies sone.' + + if (filter_mat) then + ! + ! Build the filtered matrix Af from A + ! + if (info == psb_success_) call a%cscnv(acsrf,info,dupl=psb_dupl_add_) + + do i=1,nrow + tmp = czero + jd = -1 + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) jd = j + if (abs(acsrf%val(j)) < theta*sqrt(abs(adiag(i)*adiag(acsrf%ja(j))))) then + tmp=tmp+acsrf%val(j) + acsrf%val(j)=czero + endif + + enddo + if (jd == -1) then + write(0,*) 'Wrong input: we need the diagonal!!!!', i + else + acsrf%val(jd)=acsrf%val(jd)-tmp + end if + enddo + ! Take out zeroed terms + call acsrf%mv_to_coo(acoof,info) + k = 0 + do j=1,acoof%get_nzeros() + if ((acoof%val(j) /= czero) .or. (acoof%ia(j) == acoof%ja(j))) then + k = k + 1 + acoof%val(k) = acoof%val(j) + acoof%ia(k) = acoof%ia(j) + acoof%ja(k) = acoof%ja(j) + end if + end do + call acoof%set_nzeros(k) + call acoof%set_dupl(psb_dupl_add_) + call acsrf%mv_from_coo(acoof,info) + end if + + + do i=1,size(adiag) + if (adiag(i) /= czero) then + adiag(i) = cone / adiag(i) + else + adiag(i) = cone + end if + end do + + if (filter_mat) call acsrf%scal(adiag,info) + if (info == psb_success_) call acsr3%scal(adiag,info) + if (info /= psb_success_) goto 9999 + + + if (p%parms%aggr_omega_alg == mld_eig_est_) then + + if (p%parms%aggr_eig == mld_max_norm_) then + + ! + ! This only works with CSR + ! + anorm = szero + dg = sone + nrw = acsr3%get_nrows() + do i=1, nrw + tmp = szero + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) <= nrw) then + tmp = tmp + abs(acsr3%val(j)) + endif + if (acsr3%ja(j) == i ) then + dg = abs(acsr3%val(j)) + end if + end do + anorm = max(anorm,tmp/dg) + enddo + + call psb_amx(ictxt,anorm) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Invalid AM3 storage format') + goto 9999 + end if + omega = 4.d0/(3.d0*anorm) + p%parms%aggr_omega_val = omega + + else + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_eig_') + goto 9999 + end if + + else if (p%parms%aggr_omega_alg == mld_user_choice_) then + + omega = p%parms%aggr_omega_val + + else if (p%parms%aggr_omega_alg /= mld_user_choice_) then + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_omega_alg_') + goto 9999 + end if + + if (filter_mat) then + ! + ! Build the smoothed prolongator using the filtered matrix + ! + do i=1,acsrf%get_nrows() + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) then + acsrf%val(j) = cone - omega*acsrf%val(j) + else + acsrf%val(j) = - omega*acsrf%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*Af)Ptilde + ! Doing it this way means to consider diag(Af_i) + ! + ! + call psb_symbmm(acsrf,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsrf,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + else + ! + ! Build the smoothed prolongator using the original matrix + ! + do i=1,acsr3%get_nrows() + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) == i) then + acsr3%val(j) = cone - omega*acsr3%val(j) + else + acsr3%val(j) = - omega*acsr3%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*A)Ptilde + ! Doing it this way means to consider diag(A_i) + ! + ! + call psb_symbmm(acsr3,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsr3,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + end if + call ptilde%free() + call acsr1%set_dupl(psb_dupl_add_) + + call op_prol%mv_from(acsr1) + + call psb_rwextd(ncol,op_prol,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Halo of op_prol') + goto 9999 + end if + + call psb_symbmm(a,op_prol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 2') + goto 9999 + end if + + call psb_numbmm(a,op_prol,am3) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 2',p%parms%aggr_kind, mld_smooth_prol_ + + call op_prol%transp(op_restr) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting sphalo/ rwxtd' + + call psb_rwextd(ncol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Extend am3') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting symbmm 3' + call psb_symbmm(op_restr,am3,b,info) + if (info == psb_success_) call psb_numbmm(op_restr,am3,b) + if (info == psb_success_) call am3%free() + if (info == psb_success_) call b%cscnv(info,type='coo',dupl=psb_dupl_add_) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Build b = op_restr x am3') + goto 9999 + end if + + + + + select case(p%parms%coarse_mat) + + case(mld_distr_mat_) + + nzl = b%get_nzeros() + call b%mv_to(bcoo) + + if (info == psb_success_) call psb_cdall(ictxt,p%desc_ac,info,nl=nlaggr(me+1)) + if (info == psb_success_) call psb_cdins(nzl,bcoo%ia,bcoo%ja,p%desc_ac,info) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) call psb_glob_to_loc(bcoo%ia(1:nzl),p%desc_ac,info,iact='I') + if (info == psb_success_) call psb_glob_to_loc(bcoo%ja(1:nzl),p%desc_ac,info,iact='I') + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,& + & a_err='Creating p%desc_ac and converting ac') + goto 9999 + end if + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Assembld aux descr. distr.' + call p%ac%mv_from(bcoo) + + call p%ac%set_nrows(p%desc_ac%get_local_rows()) + call p%ac%set_ncols(p%desc_ac%get_local_cols()) + call p%ac%set_asb() + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_sp_free') + goto 9999 + end if + + if (np>1) then + call op_prol%mv_to(acsr1) + nzl = acsr1%get_nzeros() + call psb_glob_to_loc(acsr1%ja(1:nzl),p%desc_ac,info,'I') + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_glob_to_loc') + goto 9999 + end if + call op_prol%mv_from(acsr1) + endif + call op_prol%set_ncols(p%desc_ac%get_local_cols()) + + if (np>1) then + call op_restr%cscnv(info,type='coo',dupl=psb_dupl_add_) + call op_restr%mv_to(acoo) + nzl = acoo%get_nzeros() + if (info == psb_success_) call psb_glob_to_loc(acoo%ia(1:nzl),p%desc_ac,info,'I') + call acoo%set_dupl(psb_dupl_add_) + if (info == psb_success_) call op_restr%mv_from(acoo) + if (info == psb_success_) call op_restr%cscnv(info,type='csr') + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Converting op_restr to local') + goto 9999 + end if + end if + call op_restr%set_nrows(p%desc_ac%get_local_cols()) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done ac ' + + case(mld_repl_mat_) + ! + ! + call psb_cdall(ictxt,p%desc_ac,info,mg=ntaggr,repl=.true.) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) & + & call psb_gather(p%ac,b,p%desc_ac,info,dupl=psb_dupl_add_,keeploc=.false.) + + if (info /= psb_success_) goto 9999 + + case default + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_coarse_mat_') + goto 9999 + end select + + + + call p%ac%cscnv(info,type='csr',dupl=psb_dupl_add_) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='spcnv') + goto 9999 + end if + + ! + ! Copy the prolongation/restriction matrices into the descriptor map. + ! op_restr => PR^T i.e. restriction operator + ! op_prol => PR i.e. prolongation operator + ! + + p%map = psb_linmap(psb_map_aggr_,desc_a,& + & p%desc_ac,op_restr,op_prol,ilaggr,nlaggr) + if (info == psb_success_) call op_prol%free() + if (info == psb_success_) call op_restr%free() + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_Free') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done smooth_aggregate ' + call psb_erractionrestore(err_act) + return + +9999 continue + call psb_errpush(info,name) + call psb_erractionrestore(err_act) + if (err_act.eq.psb_act_abort_) then + call psb_error() + return + end if + return + + + +end subroutine mld_caggrmat_biz_asb diff --git a/mlprec/impl/mld_daggrmat_biz_asb.F90 b/mlprec/impl/mld_daggrmat_biz_asb.F90 new file mode 100644 index 00000000..91e0178d --- /dev/null +++ b/mlprec/impl/mld_daggrmat_biz_asb.F90 @@ -0,0 +1,523 @@ +!!$ +!!$ +!!$ MLD2P4 version 2.0 +!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package +!!$ based on PSBLAS (Parallel Sparse BLAS version 3.0) +!!$ +!!$ (C) Copyright 2008,2009,2010 +!!$ +!!$ Salvatore Filippone University of Rome Tor Vergata +!!$ Alfredo Buttari CNRS-IRIT, Toulouse +!!$ Pasqua D'Ambra ICAR-CNR, Naples +!!$ Daniela di Serafino Second University of Naples +!!$ +!!$ 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 MLD2P4 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 MLD2P4 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: mld_daggrmat_biz_asb.F90 +! +! Subroutine: mld_daggrmat_biz_asb +! Version: real +! +! This routine builds a coarse-level matrix A_C from a fine-level matrix A +! by using the Galerkin approach, i.e. +! +! A_C = P_C^T A P_C, +! +! where P_C is a prolongator from the coarse level to the fine one. +! +! This routine builds A_C according to a "bizarre" aggregation algorithm, +! using a "naive" prolongator proposed by the authors of MLD2P4. However, this +! prolongator still requires a deep analysis and testing and its use is not +! recommended. +! +! The coarse-level matrix A_C is distributed among the parallel processes or +! replicated on each of them, according to the value of p%parms%coarse_mat, +! specified by the user through mld_dprecinit and mld_zprecset. +! +! Arguments: +! a - type(psb_dspmat_type), input. +! The sparse matrix structure containing the local part of +! the fine-level matrix. +! desc_a - type(psb_desc_type), input. +! The communication descriptor of the fine-level matrix. +! p - type(mld_d_onelev_type), input/output. +! The 'one-level' data structure that will contain the local +! part of the matrix to be built as well as the information +! concerning the prolongator and its transpose. +! ilaggr - integer, dimension(:), allocatable. +! The mapping between the row indices of the coarse-level +! matrix and the row indices of the fine-level matrix. +! ilaggr(i)=j means that node i in the adjacency graph +! of the fine-level matrix is mapped onto node j in the +! adjacency graph of the coarse-level matrix. +! nlaggr - integer, dimension(:), allocatable. +! nlaggr(i) contains the aggregates held by process i. +! info - integer, output. +! Error code. +! +subroutine mld_daggrmat_biz_asb(a,desc_a,ilaggr,nlaggr,p,info) + use psb_base_mod + use mld_d_inner_mod, mld_protect_name => mld_daggrmat_biz_asb + + implicit none + + ! Arguments + type(psb_dspmat_type), intent(in) :: a + type(psb_desc_type), intent(in) :: desc_a + integer, intent(inout) :: ilaggr(:), nlaggr(:) + type(mld_d_onelev_type), intent(inout), target :: p + integer, intent(out) :: info + type(psb_dspmat_type) :: op_prol, op_restr, b + + ! Local variables + integer :: nrow, nglob, ncol, ntaggr, ip, ndx,& + & naggr, nzl,naggrm1,naggrp1, i, j, k, jd, icolF, nrw + integer ::ictxt, np, me, err_act + character(len=20) :: name + type(psb_dspmat_type) :: am3, am4 + type(psb_d_coo_sparse_mat) :: acoo, acoof, bcoo + type(psb_d_csr_sparse_mat) :: acsr1, acsr2, acsr3, acsrf, ptilde + real(psb_dpk_), allocatable :: adiag(:) + integer(psb_ipk_) :: ierr(5) + logical :: filter_mat + integer :: debug_level, debug_unit + integer, parameter :: ncmax=16 + real(psb_dpk_) :: anorm, omega, tmp, dg, theta + + name='mld_aggrmat_biz_asb' + if(psb_get_errstatus().ne.0) return + info=psb_success_ + call psb_erractionsave(err_act) + debug_unit = psb_get_debug_unit() + debug_level = psb_get_debug_level() + + ictxt = desc_a%get_context() + ictxt = desc_a%get_context() + + call psb_info(ictxt, me, np) + + nglob = desc_a%get_global_rows() + nrow = desc_a%get_local_rows() + ncol = desc_a%get_local_cols() + + theta = p%parms%aggr_thresh + + naggr = nlaggr(me+1) + ntaggr = sum(nlaggr) + + naggrm1 = sum(nlaggr(1:me)) + naggrp1 = sum(nlaggr(1:me+1)) + + filter_mat = (p%parms%aggr_filter == mld_filter_mat_) + + ilaggr(1:nrow) = ilaggr(1:nrow) + naggrm1 + call psb_halo(ilaggr,desc_a,info) + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_halo') + goto 9999 + end if + + ! naggr: number of local aggregates + ! nrow: local rows. + ! + allocate(adiag(ncol),stat=info) + + if (info /= psb_success_) then + info=psb_err_alloc_request_; ierr(1)=nrow; + call psb_errpush(info,name,i_err=ierr,a_err='real(psb_dpk_)') + goto 9999 + end if + + ! Get the diagonal D + call a%get_diag(adiag,info) + if (info == psb_success_) & + & call psb_halo(adiag,desc_a,info) + + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag') + goto 9999 + end if + + ! 1. Allocate Ptilde in sparse matrix form + call acoo%allocate(ncol,naggr,ncol) + do i=1,nrow + acoo%val(i) = done + acoo%ia(i) = i + acoo%ja(i) = ilaggr(i) + end do + call acoo%set_nzeros(nrow) + call acoo%set_dupl(psb_dupl_add_) + + call ptilde%mv_from_coo(acoo,info) + if (info == psb_success_) call a%cscnv(acsr3,info,dupl=psb_dupl_add_) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & ' Initial copies sone.' + + if (filter_mat) then + ! + ! Build the filtered matrix Af from A + ! + if (info == psb_success_) call a%cscnv(acsrf,info,dupl=psb_dupl_add_) + + do i=1,nrow + tmp = dzero + jd = -1 + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) jd = j + if (abs(acsrf%val(j)) < theta*sqrt(abs(adiag(i)*adiag(acsrf%ja(j))))) then + tmp=tmp+acsrf%val(j) + acsrf%val(j)=dzero + endif + + enddo + if (jd == -1) then + write(0,*) 'Wrong input: we need the diagonal!!!!', i + else + acsrf%val(jd)=acsrf%val(jd)-tmp + end if + enddo + ! Take out zeroed terms + call acsrf%mv_to_coo(acoof,info) + k = 0 + do j=1,acoof%get_nzeros() + if ((acoof%val(j) /= dzero) .or. (acoof%ia(j) == acoof%ja(j))) then + k = k + 1 + acoof%val(k) = acoof%val(j) + acoof%ia(k) = acoof%ia(j) + acoof%ja(k) = acoof%ja(j) + end if + end do + call acoof%set_nzeros(k) + call acoof%set_dupl(psb_dupl_add_) + call acsrf%mv_from_coo(acoof,info) + end if + + + do i=1,size(adiag) + if (adiag(i) /= dzero) then + adiag(i) = done / adiag(i) + else + adiag(i) = done + end if + end do + + if (filter_mat) call acsrf%scal(adiag,info) + if (info == psb_success_) call acsr3%scal(adiag,info) + if (info /= psb_success_) goto 9999 + + + if (p%parms%aggr_omega_alg == mld_eig_est_) then + + if (p%parms%aggr_eig == mld_max_norm_) then + + ! + ! This only works with CSR + ! + anorm = dzero + dg = done + nrw = acsr3%get_nrows() + do i=1, nrw + tmp = dzero + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) <= nrw) then + tmp = tmp + abs(acsr3%val(j)) + endif + if (acsr3%ja(j) == i ) then + dg = abs(acsr3%val(j)) + end if + end do + anorm = max(anorm,tmp/dg) + enddo + + call psb_amx(ictxt,anorm) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Invalid AM3 storage format') + goto 9999 + end if + omega = 4.d0/(3.d0*anorm) + p%parms%aggr_omega_val = omega + + else + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_eig_') + goto 9999 + end if + + else if (p%parms%aggr_omega_alg == mld_user_choice_) then + + omega = p%parms%aggr_omega_val + + else if (p%parms%aggr_omega_alg /= mld_user_choice_) then + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_omega_alg_') + goto 9999 + end if + + if (filter_mat) then + ! + ! Build the smoothed prolongator using the filtered matrix + ! + do i=1,acsrf%get_nrows() + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) then + acsrf%val(j) = done - omega*acsrf%val(j) + else + acsrf%val(j) = - omega*acsrf%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*Af)Ptilde + ! Doing it this way means to consider diag(Af_i) + ! + ! + call psb_symbmm(acsrf,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsrf,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + else + ! + ! Build the smoothed prolongator using the original matrix + ! + do i=1,acsr3%get_nrows() + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) == i) then + acsr3%val(j) = done - omega*acsr3%val(j) + else + acsr3%val(j) = - omega*acsr3%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*A)Ptilde + ! Doing it this way means to consider diag(A_i) + ! + ! + call psb_symbmm(acsr3,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsr3,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + end if + call ptilde%free() + call acsr1%set_dupl(psb_dupl_add_) + + call op_prol%mv_from(acsr1) + + call psb_rwextd(ncol,op_prol,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Halo of op_prol') + goto 9999 + end if + + call psb_symbmm(a,op_prol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 2') + goto 9999 + end if + + call psb_numbmm(a,op_prol,am3) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 2',p%parms%aggr_kind, mld_smooth_prol_ + + call op_prol%transp(op_restr) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting sphalo/ rwxtd' + + call psb_rwextd(ncol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Extend am3') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting symbmm 3' + call psb_symbmm(op_restr,am3,b,info) + if (info == psb_success_) call psb_numbmm(op_restr,am3,b) + if (info == psb_success_) call am3%free() + if (info == psb_success_) call b%cscnv(info,type='coo',dupl=psb_dupl_add_) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Build b = op_restr x am3') + goto 9999 + end if + + + + + select case(p%parms%coarse_mat) + + case(mld_distr_mat_) + + nzl = b%get_nzeros() + call b%mv_to(bcoo) + + if (info == psb_success_) call psb_cdall(ictxt,p%desc_ac,info,nl=nlaggr(me+1)) + if (info == psb_success_) call psb_cdins(nzl,bcoo%ia,bcoo%ja,p%desc_ac,info) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) call psb_glob_to_loc(bcoo%ia(1:nzl),p%desc_ac,info,iact='I') + if (info == psb_success_) call psb_glob_to_loc(bcoo%ja(1:nzl),p%desc_ac,info,iact='I') + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,& + & a_err='Creating p%desc_ac and converting ac') + goto 9999 + end if + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Assembld aux descr. distr.' + call p%ac%mv_from(bcoo) + + call p%ac%set_nrows(p%desc_ac%get_local_rows()) + call p%ac%set_ncols(p%desc_ac%get_local_cols()) + call p%ac%set_asb() + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_sp_free') + goto 9999 + end if + + if (np>1) then + call op_prol%mv_to(acsr1) + nzl = acsr1%get_nzeros() + call psb_glob_to_loc(acsr1%ja(1:nzl),p%desc_ac,info,'I') + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_glob_to_loc') + goto 9999 + end if + call op_prol%mv_from(acsr1) + endif + call op_prol%set_ncols(p%desc_ac%get_local_cols()) + + if (np>1) then + call op_restr%cscnv(info,type='coo',dupl=psb_dupl_add_) + call op_restr%mv_to(acoo) + nzl = acoo%get_nzeros() + if (info == psb_success_) call psb_glob_to_loc(acoo%ia(1:nzl),p%desc_ac,info,'I') + call acoo%set_dupl(psb_dupl_add_) + if (info == psb_success_) call op_restr%mv_from(acoo) + if (info == psb_success_) call op_restr%cscnv(info,type='csr') + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Converting op_restr to local') + goto 9999 + end if + end if + call op_restr%set_nrows(p%desc_ac%get_local_cols()) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done ac ' + + case(mld_repl_mat_) + ! + ! + call psb_cdall(ictxt,p%desc_ac,info,mg=ntaggr,repl=.true.) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) & + & call psb_gather(p%ac,b,p%desc_ac,info,dupl=psb_dupl_add_,keeploc=.false.) + + if (info /= psb_success_) goto 9999 + + case default + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_coarse_mat_') + goto 9999 + end select + + + + call p%ac%cscnv(info,type='csr',dupl=psb_dupl_add_) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='spcnv') + goto 9999 + end if + + ! + ! Copy the prolongation/restriction matrices into the descriptor map. + ! op_restr => PR^T i.e. restriction operator + ! op_prol => PR i.e. prolongation operator + ! + + p%map = psb_linmap(psb_map_aggr_,desc_a,& + & p%desc_ac,op_restr,op_prol,ilaggr,nlaggr) + if (info == psb_success_) call op_prol%free() + if (info == psb_success_) call op_restr%free() + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_Free') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done smooth_aggregate ' + call psb_erractionrestore(err_act) + return + +9999 continue + call psb_errpush(info,name) + call psb_erractionrestore(err_act) + if (err_act.eq.psb_act_abort_) then + call psb_error() + return + end if + return + + + +end subroutine mld_daggrmat_biz_asb diff --git a/mlprec/impl/mld_saggrmat_biz_asb.F90 b/mlprec/impl/mld_saggrmat_biz_asb.F90 new file mode 100644 index 00000000..e1bf0552 --- /dev/null +++ b/mlprec/impl/mld_saggrmat_biz_asb.F90 @@ -0,0 +1,523 @@ +!!$ +!!$ +!!$ MLD2P4 version 2.0 +!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package +!!$ based on PSBLAS (Parallel Sparse BLAS version 3.0) +!!$ +!!$ (C) Copyright 2008,2009,2010 +!!$ +!!$ Salvatore Filippone University of Rome Tor Vergata +!!$ Alfredo Buttari CNRS-IRIT, Toulouse +!!$ Pasqua D'Ambra ICAR-CNR, Naples +!!$ Daniela di Serafino Second University of Naples +!!$ +!!$ 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 MLD2P4 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 MLD2P4 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: mld_saggrmat_biz_asb.F90 +! +! Subroutine: mld_saggrmat_biz_asb +! Version: real +! +! This routine builds a coarse-level matrix A_C from a fine-level matrix A +! by using the Galerkin approach, i.e. +! +! A_C = P_C^T A P_C, +! +! where P_C is a prolongator from the coarse level to the fine one. +! +! This routine builds A_C according to a "bizarre" aggregation algorithm, +! using a "naive" prolongator proposed by the authors of MLD2P4. However, this +! prolongator still requires a deep analysis and testing and its use is not +! recommended. +! +! The coarse-level matrix A_C is distributed among the parallel processes or +! replicated on each of them, according to the value of p%parms%coarse_mat, +! specified by the user through mld_sprecinit and mld_zprecset. +! +! Arguments: +! a - type(psb_sspmat_type), input. +! The sparse matrix structure containing the local part of +! the fine-level matrix. +! desc_a - type(psb_desc_type), input. +! The communication descriptor of the fine-level matrix. +! p - type(mld_s_onelev_type), input/output. +! The 'one-level' data structure that will contain the local +! part of the matrix to be built as well as the information +! concerning the prolongator and its transpose. +! ilaggr - integer, dimension(:), allocatable. +! The mapping between the row indices of the coarse-level +! matrix and the row indices of the fine-level matrix. +! ilaggr(i)=j means that node i in the adjacency graph +! of the fine-level matrix is mapped onto node j in the +! adjacency graph of the coarse-level matrix. +! nlaggr - integer, dimension(:), allocatable. +! nlaggr(i) contains the aggregates held by process i. +! info - integer, output. +! Error code. +! +subroutine mld_saggrmat_biz_asb(a,desc_a,ilaggr,nlaggr,p,info) + use psb_base_mod + use mld_s_inner_mod, mld_protect_name => mld_saggrmat_biz_asb + + implicit none + + ! Arguments + type(psb_sspmat_type), intent(in) :: a + type(psb_desc_type), intent(in) :: desc_a + integer, intent(inout) :: ilaggr(:), nlaggr(:) + type(mld_s_onelev_type), intent(inout), target :: p + integer, intent(out) :: info + type(psb_sspmat_type) :: op_prol, op_restr, b + + ! Local variables + integer :: nrow, nglob, ncol, ntaggr, ip, ndx,& + & naggr, nzl,naggrm1,naggrp1, i, j, k, jd, icolF, nrw + integer ::ictxt, np, me, err_act + character(len=20) :: name + type(psb_sspmat_type) :: am3, am4 + type(psb_s_coo_sparse_mat) :: acoo, acoof, bcoo + type(psb_s_csr_sparse_mat) :: acsr1, acsr2, acsr3, acsrf, ptilde + real(psb_spk_), allocatable :: adiag(:) + integer(psb_ipk_) :: ierr(5) + logical :: filter_mat + integer :: debug_level, debug_unit + integer, parameter :: ncmax=16 + real(psb_spk_) :: anorm, omega, tmp, dg, theta + + name='mld_aggrmat_biz_asb' + if(psb_get_errstatus().ne.0) return + info=psb_success_ + call psb_erractionsave(err_act) + debug_unit = psb_get_debug_unit() + debug_level = psb_get_debug_level() + + ictxt = desc_a%get_context() + ictxt = desc_a%get_context() + + call psb_info(ictxt, me, np) + + nglob = desc_a%get_global_rows() + nrow = desc_a%get_local_rows() + ncol = desc_a%get_local_cols() + + theta = p%parms%aggr_thresh + + naggr = nlaggr(me+1) + ntaggr = sum(nlaggr) + + naggrm1 = sum(nlaggr(1:me)) + naggrp1 = sum(nlaggr(1:me+1)) + + filter_mat = (p%parms%aggr_filter == mld_filter_mat_) + + ilaggr(1:nrow) = ilaggr(1:nrow) + naggrm1 + call psb_halo(ilaggr,desc_a,info) + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_halo') + goto 9999 + end if + + ! naggr: number of local aggregates + ! nrow: local rows. + ! + allocate(adiag(ncol),stat=info) + + if (info /= psb_success_) then + info=psb_err_alloc_request_; ierr(1)=nrow; + call psb_errpush(info,name,i_err=ierr,a_err='real(psb_spk_)') + goto 9999 + end if + + ! Get the diagonal D + call a%get_diag(adiag,info) + if (info == psb_success_) & + & call psb_halo(adiag,desc_a,info) + + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag') + goto 9999 + end if + + ! 1. Allocate Ptilde in sparse matrix form + call acoo%allocate(ncol,naggr,ncol) + do i=1,nrow + acoo%val(i) = sone + acoo%ia(i) = i + acoo%ja(i) = ilaggr(i) + end do + call acoo%set_nzeros(nrow) + call acoo%set_dupl(psb_dupl_add_) + + call ptilde%mv_from_coo(acoo,info) + if (info == psb_success_) call a%cscnv(acsr3,info,dupl=psb_dupl_add_) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & ' Initial copies sone.' + + if (filter_mat) then + ! + ! Build the filtered matrix Af from A + ! + if (info == psb_success_) call a%cscnv(acsrf,info,dupl=psb_dupl_add_) + + do i=1,nrow + tmp = szero + jd = -1 + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) jd = j + if (abs(acsrf%val(j)) < theta*sqrt(abs(adiag(i)*adiag(acsrf%ja(j))))) then + tmp=tmp+acsrf%val(j) + acsrf%val(j)=szero + endif + + enddo + if (jd == -1) then + write(0,*) 'Wrong input: we need the diagonal!!!!', i + else + acsrf%val(jd)=acsrf%val(jd)-tmp + end if + enddo + ! Take out zeroed terms + call acsrf%mv_to_coo(acoof,info) + k = 0 + do j=1,acoof%get_nzeros() + if ((acoof%val(j) /= szero) .or. (acoof%ia(j) == acoof%ja(j))) then + k = k + 1 + acoof%val(k) = acoof%val(j) + acoof%ia(k) = acoof%ia(j) + acoof%ja(k) = acoof%ja(j) + end if + end do + call acoof%set_nzeros(k) + call acoof%set_dupl(psb_dupl_add_) + call acsrf%mv_from_coo(acoof,info) + end if + + + do i=1,size(adiag) + if (adiag(i) /= szero) then + adiag(i) = sone / adiag(i) + else + adiag(i) = sone + end if + end do + + if (filter_mat) call acsrf%scal(adiag,info) + if (info == psb_success_) call acsr3%scal(adiag,info) + if (info /= psb_success_) goto 9999 + + + if (p%parms%aggr_omega_alg == mld_eig_est_) then + + if (p%parms%aggr_eig == mld_max_norm_) then + + ! + ! This only works with CSR + ! + anorm = szero + dg = sone + nrw = acsr3%get_nrows() + do i=1, nrw + tmp = szero + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) <= nrw) then + tmp = tmp + abs(acsr3%val(j)) + endif + if (acsr3%ja(j) == i ) then + dg = abs(acsr3%val(j)) + end if + end do + anorm = max(anorm,tmp/dg) + enddo + + call psb_amx(ictxt,anorm) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Invalid AM3 storage format') + goto 9999 + end if + omega = 4.d0/(3.d0*anorm) + p%parms%aggr_omega_val = omega + + else + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_eig_') + goto 9999 + end if + + else if (p%parms%aggr_omega_alg == mld_user_choice_) then + + omega = p%parms%aggr_omega_val + + else if (p%parms%aggr_omega_alg /= mld_user_choice_) then + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_omega_alg_') + goto 9999 + end if + + if (filter_mat) then + ! + ! Build the smoothed prolongator using the filtered matrix + ! + do i=1,acsrf%get_nrows() + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) then + acsrf%val(j) = sone - omega*acsrf%val(j) + else + acsrf%val(j) = - omega*acsrf%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*Af)Ptilde + ! Doing it this way means to consider diag(Af_i) + ! + ! + call psb_symbmm(acsrf,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsrf,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + else + ! + ! Build the smoothed prolongator using the original matrix + ! + do i=1,acsr3%get_nrows() + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) == i) then + acsr3%val(j) = sone - omega*acsr3%val(j) + else + acsr3%val(j) = - omega*acsr3%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*A)Ptilde + ! Doing it this way means to consider diag(A_i) + ! + ! + call psb_symbmm(acsr3,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsr3,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + end if + call ptilde%free() + call acsr1%set_dupl(psb_dupl_add_) + + call op_prol%mv_from(acsr1) + + call psb_rwextd(ncol,op_prol,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Halo of op_prol') + goto 9999 + end if + + call psb_symbmm(a,op_prol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 2') + goto 9999 + end if + + call psb_numbmm(a,op_prol,am3) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 2',p%parms%aggr_kind, mld_smooth_prol_ + + call op_prol%transp(op_restr) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting sphalo/ rwxtd' + + call psb_rwextd(ncol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Extend am3') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting symbmm 3' + call psb_symbmm(op_restr,am3,b,info) + if (info == psb_success_) call psb_numbmm(op_restr,am3,b) + if (info == psb_success_) call am3%free() + if (info == psb_success_) call b%cscnv(info,type='coo',dupl=psb_dupl_add_) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Build b = op_restr x am3') + goto 9999 + end if + + + + + select case(p%parms%coarse_mat) + + case(mld_distr_mat_) + + nzl = b%get_nzeros() + call b%mv_to(bcoo) + + if (info == psb_success_) call psb_cdall(ictxt,p%desc_ac,info,nl=nlaggr(me+1)) + if (info == psb_success_) call psb_cdins(nzl,bcoo%ia,bcoo%ja,p%desc_ac,info) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) call psb_glob_to_loc(bcoo%ia(1:nzl),p%desc_ac,info,iact='I') + if (info == psb_success_) call psb_glob_to_loc(bcoo%ja(1:nzl),p%desc_ac,info,iact='I') + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,& + & a_err='Creating p%desc_ac and converting ac') + goto 9999 + end if + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Assembld aux descr. distr.' + call p%ac%mv_from(bcoo) + + call p%ac%set_nrows(p%desc_ac%get_local_rows()) + call p%ac%set_ncols(p%desc_ac%get_local_cols()) + call p%ac%set_asb() + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_sp_free') + goto 9999 + end if + + if (np>1) then + call op_prol%mv_to(acsr1) + nzl = acsr1%get_nzeros() + call psb_glob_to_loc(acsr1%ja(1:nzl),p%desc_ac,info,'I') + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_glob_to_loc') + goto 9999 + end if + call op_prol%mv_from(acsr1) + endif + call op_prol%set_ncols(p%desc_ac%get_local_cols()) + + if (np>1) then + call op_restr%cscnv(info,type='coo',dupl=psb_dupl_add_) + call op_restr%mv_to(acoo) + nzl = acoo%get_nzeros() + if (info == psb_success_) call psb_glob_to_loc(acoo%ia(1:nzl),p%desc_ac,info,'I') + call acoo%set_dupl(psb_dupl_add_) + if (info == psb_success_) call op_restr%mv_from(acoo) + if (info == psb_success_) call op_restr%cscnv(info,type='csr') + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Converting op_restr to local') + goto 9999 + end if + end if + call op_restr%set_nrows(p%desc_ac%get_local_cols()) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done ac ' + + case(mld_repl_mat_) + ! + ! + call psb_cdall(ictxt,p%desc_ac,info,mg=ntaggr,repl=.true.) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) & + & call psb_gather(p%ac,b,p%desc_ac,info,dupl=psb_dupl_add_,keeploc=.false.) + + if (info /= psb_success_) goto 9999 + + case default + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_coarse_mat_') + goto 9999 + end select + + + + call p%ac%cscnv(info,type='csr',dupl=psb_dupl_add_) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='spcnv') + goto 9999 + end if + + ! + ! Copy the prolongation/restriction matrices into the descriptor map. + ! op_restr => PR^T i.e. restriction operator + ! op_prol => PR i.e. prolongation operator + ! + + p%map = psb_linmap(psb_map_aggr_,desc_a,& + & p%desc_ac,op_restr,op_prol,ilaggr,nlaggr) + if (info == psb_success_) call op_prol%free() + if (info == psb_success_) call op_restr%free() + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_Free') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done smooth_aggregate ' + call psb_erractionrestore(err_act) + return + +9999 continue + call psb_errpush(info,name) + call psb_erractionrestore(err_act) + if (err_act.eq.psb_act_abort_) then + call psb_error() + return + end if + return + + + +end subroutine mld_saggrmat_biz_asb diff --git a/mlprec/impl/mld_zaggrmat_biz_asb.F90 b/mlprec/impl/mld_zaggrmat_biz_asb.F90 new file mode 100644 index 00000000..3f43e05c --- /dev/null +++ b/mlprec/impl/mld_zaggrmat_biz_asb.F90 @@ -0,0 +1,523 @@ +!!$ +!!$ +!!$ MLD2P4 version 2.0 +!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package +!!$ based on PSBLAS (Parallel Sparse BLAS version 3.0) +!!$ +!!$ (C) Copyright 2008,2009,2010 +!!$ +!!$ Salvatore Filippone University of Rome Tor Vergata +!!$ Alfredo Buttari CNRS-IRIT, Toulouse +!!$ Pasqua D'Ambra ICAR-CNR, Naples +!!$ Daniela di Serafino Second University of Naples +!!$ +!!$ 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 MLD2P4 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 MLD2P4 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: mld_zaggrmat_biz_asb.F90 +! +! Subroutine: mld_zaggrmat_biz_asb +! Version: complex +! +! This routine builds a coarse-level matrix A_C from a fine-level matrix A +! by using the Galerkin approach, i.e. +! +! A_C = P_C^T A P_C, +! +! where P_C is a prolongator from the coarse level to the fine one. +! +! This routine builds A_C according to a "bizarre" aggregation algorithm, +! using a "naive" prolongator proposed by the authors of MLD2P4. However, this +! prolongator still requires a deep analysis and testing and its use is not +! recommended. +! +! The coarse-level matrix A_C is distributed among the parallel processes or +! replicated on each of them, according to the value of p%parms%coarse_mat, +! specified by the user through mld_zprecinit and mld_zprecset. +! +! Arguments: +! a - type(psb_zspmat_type), input. +! The sparse matrix structure containing the local part of +! the fine-level matrix. +! desc_a - type(psb_desc_type), input. +! The communication descriptor of the fine-level matrix. +! p - type(mld_z_onelev_type), input/output. +! The 'one-level' data structure that will contain the local +! part of the matrix to be built as well as the information +! concerning the prolongator and its transpose. +! ilaggr - integer, dimension(:), allocatable. +! The mapping between the row indices of the coarse-level +! matrix and the row indices of the fine-level matrix. +! ilaggr(i)=j means that node i in the adjacency graph +! of the fine-level matrix is mapped onto node j in the +! adjacency graph of the coarse-level matrix. +! nlaggr - integer, dimension(:), allocatable. +! nlaggr(i) contains the aggregates held by process i. +! info - integer, output. +! Error code. +! +subroutine mld_zaggrmat_biz_asb(a,desc_a,ilaggr,nlaggr,p,info) + use psb_base_mod + use mld_z_inner_mod, mld_protect_name => mld_zaggrmat_biz_asb + + implicit none + + ! Arguments + type(psb_zspmat_type), intent(in) :: a + type(psb_desc_type), intent(in) :: desc_a + integer, intent(inout) :: ilaggr(:), nlaggr(:) + type(mld_z_onelev_type), intent(inout), target :: p + integer, intent(out) :: info + type(psb_zspmat_type) :: op_prol, op_restr, b + + ! Local variables + integer :: nrow, nglob, ncol, ntaggr, ip, ndx,& + & naggr, nzl,naggrm1,naggrp1, i, j, k, jd, icolF, nrw + integer ::ictxt, np, me, err_act + character(len=20) :: name + type(psb_zspmat_type) :: am3, am4 + type(psb_z_coo_sparse_mat) :: acoo, acoof, bcoo + type(psb_z_csr_sparse_mat) :: acsr1, acsr2, acsr3, acsrf, ptilde + complex(psb_dpk_), allocatable :: adiag(:) + integer(psb_ipk_) :: ierr(5) + logical :: filter_mat + integer :: debug_level, debug_unit + integer, parameter :: ncmax=16 + real(psb_dpk_) :: anorm, omega, tmp, dg, theta + + name='mld_aggrmat_biz_asb' + if(psb_get_errstatus().ne.0) return + info=psb_success_ + call psb_erractionsave(err_act) + debug_unit = psb_get_debug_unit() + debug_level = psb_get_debug_level() + + ictxt = desc_a%get_context() + ictxt = desc_a%get_context() + + call psb_info(ictxt, me, np) + + nglob = desc_a%get_global_rows() + nrow = desc_a%get_local_rows() + ncol = desc_a%get_local_cols() + + theta = p%parms%aggr_thresh + + naggr = nlaggr(me+1) + ntaggr = sum(nlaggr) + + naggrm1 = sum(nlaggr(1:me)) + naggrp1 = sum(nlaggr(1:me+1)) + + filter_mat = (p%parms%aggr_filter == mld_filter_mat_) + + ilaggr(1:nrow) = ilaggr(1:nrow) + naggrm1 + call psb_halo(ilaggr,desc_a,info) + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_halo') + goto 9999 + end if + + ! naggr: number of local aggregates + ! nrow: local rows. + ! + allocate(adiag(ncol),stat=info) + + if (info /= psb_success_) then + info=psb_err_alloc_request_; ierr(1)=nrow; + call psb_errpush(info,name,i_err=ierr,a_err='complex(psb_dpk_)') + goto 9999 + end if + + ! Get the diagonal D + call a%get_diag(adiag,info) + if (info == psb_success_) & + & call psb_halo(adiag,desc_a,info) + + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag') + goto 9999 + end if + + ! 1. Allocate Ptilde in sparse matrix form + call acoo%allocate(ncol,naggr,ncol) + do i=1,nrow + acoo%val(i) = zone + acoo%ia(i) = i + acoo%ja(i) = ilaggr(i) + end do + call acoo%set_nzeros(nrow) + call acoo%set_dupl(psb_dupl_add_) + + call ptilde%mv_from_coo(acoo,info) + if (info == psb_success_) call a%cscnv(acsr3,info,dupl=psb_dupl_add_) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & ' Initial copies sone.' + + if (filter_mat) then + ! + ! Build the filtered matrix Af from A + ! + if (info == psb_success_) call a%cscnv(acsrf,info,dupl=psb_dupl_add_) + + do i=1,nrow + tmp = zzero + jd = -1 + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) jd = j + if (abs(acsrf%val(j)) < theta*sqrt(abs(adiag(i)*adiag(acsrf%ja(j))))) then + tmp=tmp+acsrf%val(j) + acsrf%val(j)=zzero + endif + + enddo + if (jd == -1) then + write(0,*) 'Wrong input: we need the diagonal!!!!', i + else + acsrf%val(jd)=acsrf%val(jd)-tmp + end if + enddo + ! Take out zeroed terms + call acsrf%mv_to_coo(acoof,info) + k = 0 + do j=1,acoof%get_nzeros() + if ((acoof%val(j) /= zzero) .or. (acoof%ia(j) == acoof%ja(j))) then + k = k + 1 + acoof%val(k) = acoof%val(j) + acoof%ia(k) = acoof%ia(j) + acoof%ja(k) = acoof%ja(j) + end if + end do + call acoof%set_nzeros(k) + call acoof%set_dupl(psb_dupl_add_) + call acsrf%mv_from_coo(acoof,info) + end if + + + do i=1,size(adiag) + if (adiag(i) /= zzero) then + adiag(i) = zone / adiag(i) + else + adiag(i) = zone + end if + end do + + if (filter_mat) call acsrf%scal(adiag,info) + if (info == psb_success_) call acsr3%scal(adiag,info) + if (info /= psb_success_) goto 9999 + + + if (p%parms%aggr_omega_alg == mld_eig_est_) then + + if (p%parms%aggr_eig == mld_max_norm_) then + + ! + ! This only works with CSR + ! + anorm = dzero + dg = done + nrw = acsr3%get_nrows() + do i=1, nrw + tmp = dzero + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) <= nrw) then + tmp = tmp + abs(acsr3%val(j)) + endif + if (acsr3%ja(j) == i ) then + dg = abs(acsr3%val(j)) + end if + end do + anorm = max(anorm,tmp/dg) + enddo + + call psb_amx(ictxt,anorm) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Invalid AM3 storage format') + goto 9999 + end if + omega = 4.d0/(3.d0*anorm) + p%parms%aggr_omega_val = omega + + else + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_eig_') + goto 9999 + end if + + else if (p%parms%aggr_omega_alg == mld_user_choice_) then + + omega = p%parms%aggr_omega_val + + else if (p%parms%aggr_omega_alg /= mld_user_choice_) then + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_aggr_omega_alg_') + goto 9999 + end if + + if (filter_mat) then + ! + ! Build the smoothed prolongator using the filtered matrix + ! + do i=1,acsrf%get_nrows() + do j=acsrf%irp(i),acsrf%irp(i+1)-1 + if (acsrf%ja(j) == i) then + acsrf%val(j) = zone - omega*acsrf%val(j) + else + acsrf%val(j) = - omega*acsrf%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*Af)Ptilde + ! Doing it this way means to consider diag(Af_i) + ! + ! + call psb_symbmm(acsrf,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsrf,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + else + ! + ! Build the smoothed prolongator using the original matrix + ! + do i=1,acsr3%get_nrows() + do j=acsr3%irp(i),acsr3%irp(i+1)-1 + if (acsr3%ja(j) == i) then + acsr3%val(j) = zone - omega*acsr3%val(j) + else + acsr3%val(j) = - omega*acsr3%val(j) + end if + end do + end do + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done gather, going for SYMBMM 1' + ! + ! Symbmm90 does the allocation for its result. + ! + ! acsrm1 = (I-w*D*A)Ptilde + ! Doing it this way means to consider diag(A_i) + ! + ! + call psb_symbmm(acsr3,ptilde,acsr1,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 1') + goto 9999 + end if + + call psb_numbmm(acsr3,ptilde,acsr1) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 1' + + end if + call ptilde%free() + call acsr1%set_dupl(psb_dupl_add_) + + call op_prol%mv_from(acsr1) + + call psb_rwextd(ncol,op_prol,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Halo of op_prol') + goto 9999 + end if + + call psb_symbmm(a,op_prol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='symbmm 2') + goto 9999 + end if + + call psb_numbmm(a,op_prol,am3) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done NUMBMM 2',p%parms%aggr_kind, mld_smooth_prol_ + + call op_prol%transp(op_restr) + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting sphalo/ rwxtd' + + call psb_rwextd(ncol,am3,info) + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Extend am3') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'starting symbmm 3' + call psb_symbmm(op_restr,am3,b,info) + if (info == psb_success_) call psb_numbmm(op_restr,am3,b) + if (info == psb_success_) call am3%free() + if (info == psb_success_) call b%cscnv(info,type='coo',dupl=psb_dupl_add_) + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Build b = op_restr x am3') + goto 9999 + end if + + + + + select case(p%parms%coarse_mat) + + case(mld_distr_mat_) + + nzl = b%get_nzeros() + call b%mv_to(bcoo) + + if (info == psb_success_) call psb_cdall(ictxt,p%desc_ac,info,nl=nlaggr(me+1)) + if (info == psb_success_) call psb_cdins(nzl,bcoo%ia,bcoo%ja,p%desc_ac,info) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) call psb_glob_to_loc(bcoo%ia(1:nzl),p%desc_ac,info,iact='I') + if (info == psb_success_) call psb_glob_to_loc(bcoo%ja(1:nzl),p%desc_ac,info,iact='I') + if (info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,& + & a_err='Creating p%desc_ac and converting ac') + goto 9999 + end if + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Assembld aux descr. distr.' + call p%ac%mv_from(bcoo) + + call p%ac%set_nrows(p%desc_ac%get_local_rows()) + call p%ac%set_ncols(p%desc_ac%get_local_cols()) + call p%ac%set_asb() + + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_sp_free') + goto 9999 + end if + + if (np>1) then + call op_prol%mv_to(acsr1) + nzl = acsr1%get_nzeros() + call psb_glob_to_loc(acsr1%ja(1:nzl),p%desc_ac,info,'I') + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_glob_to_loc') + goto 9999 + end if + call op_prol%mv_from(acsr1) + endif + call op_prol%set_ncols(p%desc_ac%get_local_cols()) + + if (np>1) then + call op_restr%cscnv(info,type='coo',dupl=psb_dupl_add_) + call op_restr%mv_to(acoo) + nzl = acoo%get_nzeros() + if (info == psb_success_) call psb_glob_to_loc(acoo%ia(1:nzl),p%desc_ac,info,'I') + call acoo%set_dupl(psb_dupl_add_) + if (info == psb_success_) call op_restr%mv_from(acoo) + if (info == psb_success_) call op_restr%cscnv(info,type='csr') + if(info /= psb_success_) then + call psb_errpush(psb_err_internal_error_,name,a_err='Converting op_restr to local') + goto 9999 + end if + end if + call op_restr%set_nrows(p%desc_ac%get_local_cols()) + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done ac ' + + case(mld_repl_mat_) + ! + ! + call psb_cdall(ictxt,p%desc_ac,info,mg=ntaggr,repl=.true.) + if (info == psb_success_) call psb_cdasb(p%desc_ac,info) + if (info == psb_success_) & + & call psb_gather(p%ac,b,p%desc_ac,info,dupl=psb_dupl_add_,keeploc=.false.) + + if (info /= psb_success_) goto 9999 + + case default + info = psb_err_internal_error_ + call psb_errpush(info,name,a_err='invalid mld_coarse_mat_') + goto 9999 + end select + + + + call p%ac%cscnv(info,type='csr',dupl=psb_dupl_add_) + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='spcnv') + goto 9999 + end if + + ! + ! Copy the prolongation/restriction matrices into the descriptor map. + ! op_restr => PR^T i.e. restriction operator + ! op_prol => PR i.e. prolongation operator + ! + + p%map = psb_linmap(psb_map_aggr_,desc_a,& + & p%desc_ac,op_restr,op_prol,ilaggr,nlaggr) + if (info == psb_success_) call op_prol%free() + if (info == psb_success_) call op_restr%free() + if(info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_Free') + goto 9999 + end if + + + if (debug_level >= psb_debug_outer_) & + & write(debug_unit,*) me,' ',trim(name),& + & 'Done smooth_aggregate ' + call psb_erractionrestore(err_act) + return + +9999 continue + call psb_errpush(info,name) + call psb_erractionrestore(err_act) + if (err_act.eq.psb_act_abort_) then + call psb_error() + return + end if + return + + + +end subroutine mld_zaggrmat_biz_asb