You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
amg4psblas/mlprec/mld_dfact_bld.f90

461 lines
16 KiB
Fortran

!!$
!!$
!!$ MLD2P4 version 1.1
!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package
!!$ based on PSBLAS (Parallel Sparse BLAS version 2.3.1)
!!$
!!$ (C) Copyright 2008,2009
!!$
!!$ 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_dfact_bld.f90
!
! Subroutine: mld_dfact_bld
! Version: real
!
! This routine computes an LU or incomplete LU (ILU) factorization of the
! diagonal blocks of a distributed matrix, according to the value of
! p%iprcparm(iprcparm(sub_solve_), set by the user through mld_dprecinit
! or mld_dprecset.
! It may also compute an LU factorization of a distributed matrix, or split
! a distributed matrix into its block-diagonal and off block-diagonal parts,
! for the future application of multiple block-Jacobi sweeps.
!
! This routine is used by mld_as_bld, to build a 'base' block-Jacobi or
! Additive Schwarz (AS) preconditioner at any level of a multilevel preconditioner,
! or a block-Jacobi or LU or ILU solver at the coarsest level of a multilevel
! preconditioner. For the AS preconditioners, the diagonal blocks to be factorized
! are stored into the sparse matrix data structures a and blck, and blck contains
! the remote rows needed to build the extended local matrix as required by the
! AS preconditioner.
!
! More precisely, the routine performs one of the following tasks:
!
! 1. LU or ILU factorization of the diagonal blocks of the distributed matrix
! for the construction of block-Jacobi or AS preconditioners (allowed at
! any level of a multilevel preconditioner);
!
! 2. setup of block-Jacobi sweeps to compute an approximate solution of a
! linear system
! A*Y = X,
! distributed among the processes (allowed only at the coarsest level);
!
! 3. LU factorization of the matrix of a linear system
! A*Y = X,
! distributed among the processes (allowed only at the coarsest level);
!
! 4. LU or incomplete LU factorization of the matrix of a linear system
! A*Y = X,
! replicated on the processes (allowed only at the coarsest level).
!
! The following factorizations are available:
! - ILU(k), i.e. ILU factorization with fill-in level k;
! - MILU(k), i.e. modified ILU factorization with fill-in level k;
! - ILU(k,t), i.e. ILU with threshold (i.e. drop tolerance) t and k additional
! entries in each row of the L and U factors with respect to the initial
! sparsity pattern;
! - serial LU implemented in SuperLU version 3.0;
! - serial LU implemented in UMFPACK version 4.4;
! - distributed LU implemented in SuperLU_DIST version 2.0.
!
!
! Arguments:
! a - type(psb_dspmat_type), input.
! The sparse matrix structure containing the local part of the
! distributed matrix.
! p - type(mld_dbaseprec_type), input/output.
! The 'base preconditioner' data structure containing the local
! part of the preconditioner or solver at the current level.
! info - integer, output.
! Error code.
! upd - character, input.
! If upd='F' then the preconditioner is built from scratch;
! if upd=T' then the matrix to be preconditioned has the same
! sparsity pattern of a matrix that has been previously
! preconditioned, hence some information is reused in building
! the new preconditioner.
! blck - type(psb_dspmat_type), input, optional.
! The sparse matrix structure containing the remote rows of the
! distributed matrix, that have been retrieved by mld_as_bld
! to build an Additive Schwarz base preconditioner with overlap
! greater than 0. If the overlap is 0 blck is empty.
!
subroutine mld_dfact_bld(a,p,upd,info,blck)
use psb_sparse_mod
use mld_inner_mod, mld_protect_name => mld_dfact_bld
implicit none
! Arguments
type(psb_d_sparse_mat), intent(in), target :: a
type(mld_dbaseprec_type), intent(inout) :: p
integer, intent(out) :: info
character, intent(in) :: upd
type(psb_d_sparse_mat), intent(in), target, optional :: blck
! Local Variables
type(psb_d_sparse_mat), pointer :: blck_
type(psb_d_sparse_mat) :: atmp
integer :: ictxt,np,me,err_act
integer :: debug_level, debug_unit
integer :: k, m, int_err(5), n_row, nrow_a, n_col
character :: trans, unitd
character(len=20) :: name, ch_err
if(psb_get_errstatus().ne.0) return
info=psb_success_
name='mld_dfact_bld'
call psb_erractionsave(err_act)
debug_unit = psb_get_debug_unit()
debug_level = psb_get_debug_level()
ictxt = psb_cd_get_context(p%desc_data)
call psb_info(ictxt, me, np)
m = a%get_nrows()
if (m < 0) then
info = psb_err_iarg_neg_
int_err(1) = 1
int_err(2) = m
call psb_errpush(info,name,i_err=int_err)
goto 9999
endif
trans = 'N'
unitd = 'U'
if (present(blck)) then
blck_ => blck
else
allocate(blck_,stat=info)
if (info == psb_success_) call blck_%csall(0,0,info,1)
if(info /= psb_success_) then
info=psb_err_from_subroutine_
ch_err='psb_sp_all'
call psb_errpush(info,name,a_err=ch_err)
goto 9999
end if
end if
!
! Treat separately the case the local matrix has to be reordered
! and the case this is not required.
!
select case(p%iprcparm(mld_sub_ren_))
!
! A reordering of the local matrix is required.
!
case (1:)
!
! Reorder the rows and the columns of the local extended matrix,
! according to the value of p%iprcparm(sub_ren_). The reordered
! matrix is stored into atmp, using the COO format.
!
call mld_sp_renum(a,blck_,p,atmp,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_sp_renum')
goto 9999
end if
!
! Clip into p%av(ap_nd_) the off block-diagonal part of the local
! matrix. The clipped matrix is then stored in CSR format.
!
if (p%iprcparm(mld_smoother_sweeps_) > 1) then
call atmp%csclip(p%av(mld_ap_nd_),info,&
& jmin=atmp%get_nrows()+1,rscale=.false.,cscale=.false.)
if (info == psb_success_) call p%av(mld_ap_nd_)%cscnv(info,&
& type='csr',dupl=psb_dupl_add_)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_spcnv')
goto 9999
end if
k = p%av(mld_ap_nd_)%get_nzeros()
call psb_sum(ictxt,k)
if (k == 0) then
!
! If the off diagonal part is emtpy, there is no point in doing
! multiple Jacobi sweeps. This is certain to happen when running
! on a single processor.
!
p%iprcparm(mld_smoother_sweeps_) = 1
end if
end if
!
! Compute a factorization of the diagonal block of the local matrix,
! according to the choice made by the user by setting p%iprcparm(sub_solve_)
!
select case(p%iprcparm(mld_sub_solve_))
case(mld_ilu_n_,mld_milu_n_,mld_ilu_t_)
!
! ILU(k)/MILU(k)/ILU(k,t) factorization.
!
call atmp%cscnv(info,type='csr',dupl=psb_dupl_add_)
if (info == psb_success_) call mld_ilu_bld(atmp,p,upd,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_ilu_bld')
goto 9999
end if
case(mld_slu_)
!
! LU factorization through the SuperLU package.
!
call atmp%cscnv(info,type='csr',dupl=psb_dupl_add_)
if (info == psb_success_) call mld_slu_bld(atmp,p%desc_data,p,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_slu_bld')
goto 9999
end if
case(mld_sludist_)
!
! LU factorization through the SuperLU_DIST package. This works only
! when the matrix is distributed among the processes.
! NOTE: Should have NO overlap here!!!!
!
call a%cscnv(atmp,info,type='csr')
if (info == psb_success_) call mld_sludist_bld(atmp,p%desc_data,p,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_sludist_bld')
goto 9999
end if
case(mld_umf_)
!
! LU factorization through the UMFPACK package.
!
call atmp%cscnv(info,type='csc',dupl=psb_dupl_add_)
if (info == psb_success_) call mld_umf_bld(atmp,p%desc_data,p,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_umf_bld')
goto 9999
end if
case(mld_f_none_)
!
! Error: no factorization required.
!
info=psb_err_internal_error_
call psb_errpush(info,name,a_err='Inconsistent prec mld_f_none_')
goto 9999
case default
info=psb_err_internal_error_
call psb_errpush(info,name,a_err='Unknown mld_sub_solve_')
goto 9999
end select
call atmp%free()
!
! No reordering of the local matrix is required
!
case(0)
!
! In case of multiple block-Jacobi sweeps, clip into p%av(ap_nd_)
! the off block-diagonal part of the local extended matrix. The
! clipped matrix is then stored in CSR format.
!
if (p%iprcparm(mld_smoother_sweeps_) > 1) then
n_row = psb_cd_get_local_rows(p%desc_data)
n_col = psb_cd_get_local_cols(p%desc_data)
nrow_a = a%get_nrows()
! The following is known to work
! given that the output from CLIP is in COO.
call a%csclip(p%av(mld_ap_nd_),info,&
& jmin=nrow_a+1,rscale=.false.,cscale=.false.)
if (info == psb_success_) call blck_%csclip(atmp,info,&
& jmin=nrow_a+1,rscale=.false.,cscale=.false.)
if (info == psb_success_) call psb_rwextd(n_row,p%av(mld_ap_nd_),info,b=atmp)
if (info == psb_success_) call p%av(mld_ap_nd_)%cscnv(info,&
& type='csr',dupl=psb_dupl_add_)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='clip & psb_spcnv csr 4')
goto 9999
end if
k = p%av(mld_ap_nd_)%get_nzeros()
call psb_sum(ictxt,k)
if (k == 0) then
!
! If the off block-diagonal part is emtpy, there is no point in doing
! multiple Jacobi sweeps. This is certain to happen when running
! on a single processor.
!
p%iprcparm(mld_smoother_sweeps_) = 1
end if
call atmp%free()
end if
!
! Compute a factorization of the diagonal block of the local matrix,
! according to the choice made by the user by setting p%iprcparm(sub_solve_)
!
select case(p%iprcparm(mld_sub_solve_))
case(mld_ilu_n_,mld_milu_n_,mld_ilu_t_)
!
! ILU(k)/MILU(k)/ILU(k,t) factorization.
!
!
! Compute the incomplete LU factorization.
!
call mld_ilu_bld(a,p,upd,info,blck=blck_)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_ilu_bld')
goto 9999
end if
case(mld_slu_)
!
! LU factorization through the SuperLU package.
!
n_row = psb_cd_get_local_rows(p%desc_data)
n_col = psb_cd_get_local_cols(p%desc_data)
call a%cscnv(atmp,info,type='coo')
if (info == psb_success_) call psb_rwextd(n_row,atmp,info,b=blck_)
!
! Compute the LU factorization.
!
if (info == psb_success_) call atmp%cscnv(info,type='csr',dupl=psb_dupl_add_)
if (info == psb_success_) call mld_slu_bld(atmp,p%desc_data,p,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_slu_bld')
goto 9999
end if
call atmp%free()
case(mld_sludist_)
!
! LU factorization through the SuperLU_DIST package. This works only
! when the matrix is distributed among the processes.
! NOTE: Should have NO overlap here!!!!
!
call a%cscnv(atmp,info,type='csr')
if (info == psb_success_) call mld_sludist_bld(atmp,p%desc_data,p,info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_sludist_bld')
goto 9999
end if
call atmp%free()
case(mld_umf_)
!
! LU factorization through the UMFPACK package.
!
call a%cscnv(atmp,info,type='coo')
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_spcnv')
goto 9999
end if
n_row = psb_cd_get_local_rows(p%desc_data)
n_col = psb_cd_get_local_cols(p%desc_data)
call psb_rwextd(n_row,atmp,info,b=blck_)
!
! Compute the LU factorization.
!
if (info == psb_success_) call atmp%cscnv(info,type='csc',dupl=psb_dupl_add_)
if (info == psb_success_) call mld_umf_bld(atmp,p%desc_data,p,info)
if (debug_level >= psb_debug_outer_) &
& write(debug_unit,*) me,' ',trim(name),&
& ': Done mld_umf_bld ',info
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_umf_bld')
goto 9999
end if
call atmp%free()
case(mld_f_none_)
!
! Error: no factorization required.
!
info=psb_err_internal_error_
call psb_errpush(info,name,a_err='Inconsistent prec mld_f_none_')
goto 9999
case default
info=psb_err_internal_error_
call psb_errpush(info,name,a_err='Unknown mld_sub_solve_')
goto 9999
end select
case default
info=psb_err_internal_error_
call psb_errpush(info,name,a_err='Invalid renum_')
goto 9999
end select
if (.not.present(blck)) then
call blck_%free()
if (info == psb_success_) deallocate(blck_, stat=info)
if (info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_sp_free')
goto 9999
end if
end if
if (debug_level >= psb_debug_outer_) &
& write(debug_unit,*) me,' ',trim(name),'End '
call psb_erractionrestore(err_act)
return
9999 continue
call psb_erractionrestore(err_act)
if (err_act.eq.psb_act_abort_) then
call psb_error()
return
end if
return
end subroutine mld_dfact_bld