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.
459 lines
16 KiB
Fortran
459 lines
16 KiB
Fortran
!!$
|
|
!!$
|
|
!!$ MLD2P4
|
|
!!$ MultiLevel Domain Decomposition Parallel Preconditioners Package
|
|
!!$ based on PSBLAS (Parallel Sparse BLAS v.2.0)
|
|
!!$
|
|
!!$ (C) Copyright 2007 Alfredo Buttari University of Rome Tor Vergata
|
|
!!$ Pasqua D'Ambra ICAR-CNR, Naples
|
|
!!$ Daniela di Serafino Second University of Naples
|
|
!!$ Salvatore Filippone University of Rome Tor Vergata
|
|
!!$
|
|
!!$ 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_zbjac_aply.f90
|
|
!
|
|
! Subroutine: mld_zbjac_aply
|
|
! Version: complex
|
|
!
|
|
! This routine computes
|
|
!
|
|
! Y = beta*Y + alpha*op(K^(-1))*X,
|
|
!
|
|
! where
|
|
! - K is a suitable matrix, as specified below,
|
|
! - op(K^(-1)) is K^(-1) or its transpose, according to the value of trans,
|
|
! - X and Y are vectors,
|
|
! - alpha and beta are scalars.
|
|
!
|
|
! Depending on K, alpha, beta (and on the communication descriptor desc_data
|
|
! - see the arguments below), the above computation may correspond to one of
|
|
! the following tasks:
|
|
!
|
|
! 1. Application of a block-Jacobi preconditioner associated to a matrix A
|
|
! distributed among the processes. Here K is the preconditioner, op(K^(-1))
|
|
! = K^(-1), alpha = 1 and beta = 0.
|
|
!
|
|
! 2. Application of block-Jacobi sweeps to compute an approximate solution of
|
|
! a linear system
|
|
! A*Y = X,
|
|
!
|
|
! distributed among the processes (note that a single block-Jacobi sweep,
|
|
! with null starting guess, corresponds to the application of a block-Jacobi
|
|
! preconditioner). Here K^(-1) denotes the iteration matrix of the
|
|
! block-Jacobi solver, op(K^(-1)) = K^(-1), alpha = 1 and beta = 0.
|
|
!
|
|
! 3. Solution, through the LU factorization, of a linear system
|
|
!
|
|
! A*Y = X,
|
|
!
|
|
! distributed among the processes. Here K = L*U = A, op(K^(-1)) = K^(-1),
|
|
! alpha = 1 and beta = 0.
|
|
!
|
|
! 4. (Approximate) solution, through the LU or incomplete LU factorization, of
|
|
! a linear system
|
|
! A*Y = X,
|
|
!
|
|
! replicated on the processes. Here K = L*U = A or K = L*U ~ A, op(K^(-1)) =
|
|
! K^(-1), alpha = 1 and beta = 0.
|
|
!
|
|
! The block-Jacobi preconditioner or solver and the L and U factors of the LU
|
|
! or ILU factorizations have been built by the routine mld_dbjac_bld and stored
|
|
! into the 'base preconditioner' data structure prec. See mld_dbjac_bld for more
|
|
! details.
|
|
!
|
|
! This routine is used by mld_dbaseprec_aply, to apply 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.
|
|
!
|
|
! Inside mld_dbaseprec_aply, tasks 1, 3 and 4 may be selected if
|
|
! prec%iprcparm(smooth_sweeps_) = 1, while task 2 if prec%iprcparm(smooth_sweeps_)
|
|
! > 1. Furthermore, tasks 1, 2 and 3 may be performed if the matrix A is
|
|
! distributed among the processes (prec%iprcparm(mld_coarse_mat_) = mld_distr_mat_),
|
|
! while task 4 may be performed if A is replicated on the processes
|
|
! (prec%iprcparm(mld_coarse_mat_) = mld_repl_mat_). Note that the matrix A is
|
|
! distributed among the processes at each level of the multilevel preconditioner,
|
|
! except the coarsest one, where it may be either distributed or replicated on
|
|
! the processes. Furthermore, the tasks 2, 3 and 4 are performed only at the
|
|
! coarsest level. Note also that this routine manages implicitly the fact that
|
|
! the matrix is distributed or replicated, i.e. it does not make any explicit
|
|
! reference to the value of prec%iprcparm(mld_coarse_mat_).
|
|
!
|
|
!
|
|
! Arguments:
|
|
!
|
|
! alpha - complex(kind(0.d0)), input.
|
|
! The scalar alpha.
|
|
! prec - type(mld_zbaseprec_type), input.
|
|
! The 'base preconditioner' data structure containing the local
|
|
! part of the preconditioner or solver.
|
|
! x - complex(kind(0.d0)), dimension(:), input/output.
|
|
! The local part of the vector X.
|
|
! beta - complex(kind(0.d0)), input.
|
|
! The scalar beta.
|
|
! y - complex(kind(0.d0)), dimension(:), input/output.
|
|
! The local part of the vector Y.
|
|
! desc_data - type(psb_desc_type), input.
|
|
! The communication descriptor associated to the matrix to be
|
|
! preconditioned or 'inverted'.
|
|
! trans - character(len=1), input.
|
|
! If trans='N','n' then op(K^(-1)) = K^(-1);
|
|
! if trans='T','t' then op(K^(-1)) = K^(-T) (transpose of K^(-1)).
|
|
! If prec%iprcparm(smooth_sweeps_) > 1, the value of trans provided
|
|
! in input is ignored.
|
|
! work - complex(kind(0.d0)), dimension (:), target.
|
|
! Workspace. Its size must be at least 4*psb_cd_get_local_cols(desc_data).
|
|
! info - integer, output.
|
|
! Error code.
|
|
!
|
|
subroutine mld_zbjac_aply(alpha,prec,x,beta,y,desc_data,trans,work,info)
|
|
|
|
use psb_base_mod
|
|
use mld_prec_mod, mld_protect_name => mld_zbjac_aply
|
|
|
|
implicit none
|
|
|
|
! Arguments
|
|
type(psb_desc_type), intent(in) :: desc_data
|
|
type(mld_zbaseprc_type), intent(in) :: prec
|
|
complex(kind(0.d0)),intent(in) :: x(:)
|
|
complex(kind(0.d0)),intent(inout) :: y(:)
|
|
complex(kind(0.d0)),intent(in) :: alpha,beta
|
|
character(len=1) :: trans
|
|
complex(kind(0.d0)),target :: work(:)
|
|
integer, intent(out) :: info
|
|
|
|
! Local variables
|
|
integer :: n_row,n_col
|
|
complex(kind(1.d0)), pointer :: ww(:), aux(:), tx(:),ty(:)
|
|
integer :: ictxt,np,me,i, err_act, int_err(5)
|
|
logical,parameter :: debug=.false., debugprt=.false.
|
|
character(len=20) :: name
|
|
|
|
interface
|
|
subroutine mld_zumf_solve(flag,m,x,b,n,ptr,info)
|
|
integer, intent(in) :: flag,m,n,ptr
|
|
integer, intent(out) :: info
|
|
complex(kind(1.d0)), intent(in) :: b(*)
|
|
complex(kind(1.d0)), intent(inout) :: x(*)
|
|
end subroutine mld_zumf_solve
|
|
end interface
|
|
|
|
name='mld_zbjac_aply'
|
|
info = 0
|
|
call psb_erractionsave(err_act)
|
|
|
|
ictxt=psb_cd_get_context(desc_data)
|
|
call psb_info(ictxt, me, np)
|
|
|
|
select case(toupper(trans))
|
|
case('N')
|
|
case('T','C')
|
|
case default
|
|
call psb_errpush(40,name)
|
|
goto 9999
|
|
end select
|
|
|
|
|
|
n_row = psb_cd_get_local_rows(desc_data)
|
|
n_col = psb_cd_get_local_cols(desc_data)
|
|
|
|
if (n_col <= size(work)) then
|
|
ww => work(1:n_col)
|
|
if ((4*n_col+n_col) <= size(work)) then
|
|
aux => work(n_col+1:)
|
|
else
|
|
allocate(aux(4*n_col),stat=info)
|
|
if (info /= 0) then
|
|
info=4025
|
|
call psb_errpush(info,name,i_err=(/4*n_col,0,0,0,0/),&
|
|
& a_err='complex(kind(1.d0))')
|
|
goto 9999
|
|
end if
|
|
|
|
endif
|
|
else
|
|
allocate(ww(n_col),aux(4*n_col),stat=info)
|
|
if (info /= 0) then
|
|
info=4025
|
|
call psb_errpush(info,name,i_err=(/5*n_col,0,0,0,0/),&
|
|
& a_err='complex(kind(1.d0))')
|
|
goto 9999
|
|
end if
|
|
endif
|
|
|
|
if (debug) then
|
|
write(0,*) me,' mld_bjac_APLY: ',prec%iprcparm(mld_sub_solve_),prec%iprcparm(mld_smooth_sweeps_)
|
|
end if
|
|
|
|
if (prec%iprcparm(mld_smooth_sweeps_) == 1) then
|
|
!
|
|
! TASKS 1, 3 and 4
|
|
!
|
|
|
|
select case(prec%iprcparm(mld_sub_solve_))
|
|
case(mld_ilu_n_,mld_milu_n_,mld_ilu_t_)
|
|
!
|
|
! Apply a block-Jacobi preconditioner with ILU(k)/MILU(k)/ILU(k,t)
|
|
! factorization of the blocks (distributed matrix) or approximately
|
|
! solve a system through ILU(k)/MILU(k)/ILU(k,t) (replicated matrix).
|
|
!
|
|
|
|
select case(toupper(trans))
|
|
case('N')
|
|
|
|
call psb_spsm(zone,prec%av(mld_l_pr_),x,zzero,ww,desc_data,info,&
|
|
& trans='N',unit='L',diag=prec%d,choice=psb_none_,work=aux)
|
|
if(info /=0) goto 9999
|
|
call psb_spsm(alpha,prec%av(mld_u_pr_),ww,beta,y,desc_data,info,&
|
|
& trans='N',unit='U',choice=psb_none_, work=aux)
|
|
if(info /=0) goto 9999
|
|
|
|
case('T','C')
|
|
call psb_spsm(zone,prec%av(mld_u_pr_),x,zzero,ww,desc_data,info,&
|
|
& trans=trans,unit='L',diag=prec%d,choice=psb_none_, work=aux)
|
|
if(info /=0) goto 9999
|
|
call psb_spsm(alpha,prec%av(mld_l_pr_),ww,beta,y,desc_data,info,&
|
|
& trans=trans,unit='U',choice=psb_none_,work=aux)
|
|
if(info /=0) goto 9999
|
|
|
|
end select
|
|
|
|
case(mld_slu_)
|
|
!
|
|
! Apply a block-Jacobi preconditioner with LU factorization of the
|
|
! blocks (distributed matrix) or approximately solve a local linear
|
|
! system through LU (replicated matrix). The SuperLU package is used
|
|
! to apply the LU factorization in both cases.
|
|
!
|
|
|
|
ww(1:n_row) = x(1:n_row)
|
|
|
|
select case(toupper(trans))
|
|
case('N')
|
|
call mld_zslu_solve(0,n_row,1,ww,n_row,prec%iprcparm(mld_slu_ptr_),info)
|
|
case('T')
|
|
call mld_zslu_solve(1,n_row,1,ww,n_row,prec%iprcparm(mld_slu_ptr_),info)
|
|
case('C')
|
|
call mld_zslu_solve(2,n_row,1,ww,n_row,prec%iprcparm(mld_slu_ptr_),info)
|
|
end select
|
|
|
|
if(info /=0) goto 9999
|
|
call psb_geaxpby(alpha,ww,beta,y,desc_data,info)
|
|
|
|
case(mld_sludist_)
|
|
!
|
|
! Solve a distributed linear system with the LU factorization.
|
|
! The SuperLU_DIST package is used.
|
|
!
|
|
|
|
ww(1:n_row) = x(1:n_row)
|
|
|
|
select case(toupper(trans))
|
|
case('N')
|
|
call mld_zsludist_solve(0,n_row,1,ww,n_row,prec%iprcparm(mld_slud_ptr_),info)
|
|
case('T')
|
|
call mld_zsludist_solve(1,n_row,1,ww,n_row,prec%iprcparm(mld_slud_ptr_),info)
|
|
case('C')
|
|
call mld_zsludist_solve(2,n_row,1,ww,n_row,prec%iprcparm(mld_slud_ptr_),info)
|
|
end select
|
|
|
|
if(info /=0) goto 9999
|
|
call psb_geaxpby(alpha,ww,beta,y,desc_data,info)
|
|
|
|
case (mld_umf_)
|
|
!
|
|
! Apply a block-Jacobi preconditioner with LU factorization of the
|
|
! blocks (distributed matrix) or approximately solve a local linear
|
|
! system through LU (replicated matrix). The UMFPACK package is used
|
|
! to apply the LU factorization in both cases.
|
|
!
|
|
|
|
|
|
select case(toupper(trans))
|
|
case('N')
|
|
call mld_zumf_solve(0,n_row,ww,x,n_row,prec%iprcparm(mld_umf_numptr_),info)
|
|
case('T')
|
|
call mld_zumf_solve(1,n_row,ww,x,n_row,prec%iprcparm(mld_umf_numptr_),info)
|
|
case('C')
|
|
call mld_zumf_solve(2,n_row,ww,x,n_row,prec%iprcparm(mld_umf_numptr_),info)
|
|
end select
|
|
|
|
if(info /=0) goto 9999
|
|
|
|
call psb_geaxpby(alpha,ww,beta,y,desc_data,info)
|
|
|
|
case default
|
|
write(0,*) 'Unknown factorization type in mld_bjac_aply',prec%iprcparm(mld_sub_solve_)
|
|
end select
|
|
if (debugprt) write(0,*)' Y: ',y(:)
|
|
|
|
else if (prec%iprcparm(mld_smooth_sweeps_) > 1) then
|
|
|
|
!
|
|
! TASK 2
|
|
!
|
|
! Apply prec%iprcparm(smooth_sweeps_) sweeps of a block-Jacobi solver
|
|
! to compute an approximate solution of a linear system.
|
|
!
|
|
! Note: trans is always 'N' here.
|
|
!
|
|
|
|
if (size(prec%av) < mld_ap_nd_) then
|
|
info = 4011
|
|
goto 9999
|
|
endif
|
|
|
|
allocate(tx(n_col),ty(n_col),stat=info)
|
|
if (info /= 0) then
|
|
info=4025
|
|
call psb_errpush(info,name,i_err=(/2*n_col,0,0,0,0/),&
|
|
& a_err='complex(kind(1.d0))')
|
|
goto 9999
|
|
end if
|
|
|
|
tx = zzero
|
|
ty = zzero
|
|
select case(prec%iprcparm(mld_sub_solve_))
|
|
case(mld_ilu_n_,mld_milu_n_,mld_ilu_t_)
|
|
!
|
|
! Use ILU(k)/MILU(k)/ILU(k,t) on the blocks.
|
|
!
|
|
do i=1, prec%iprcparm(mld_smooth_sweeps_)
|
|
!
|
|
! Compute Y(j+1) = D^(-1)*(X-ND*Y(j)), where D and ND are the
|
|
! block diagonal part and the remaining part of the local matrix
|
|
! and Y(j) is the approximate solution at sweep j.
|
|
!
|
|
ty(1:n_row) = x(1:n_row)
|
|
call psb_spmm(-zone,prec%av(mld_ap_nd_),tx,zone,ty,&
|
|
& prec%desc_data,info,work=aux)
|
|
if(info /=0) goto 9999
|
|
call psb_spsm(zone,prec%av(mld_l_pr_),ty,zzero,ww,&
|
|
& prec%desc_data,info,&
|
|
& trans='N',unit='L',diag=prec%d,choice=psb_none_,work=aux)
|
|
if(info /=0) goto 9999
|
|
call psb_spsm(zone,prec%av(mld_u_pr_),ww,zzero,tx,&
|
|
& prec%desc_data,info,&
|
|
& trans='N',unit='U',choice=psb_none_,work=aux)
|
|
if(info /=0) goto 9999
|
|
end do
|
|
|
|
case(mld_sludist_)
|
|
!
|
|
! Wrong choice: SuperLU_DIST
|
|
!
|
|
write(0,*) 'No sense in having SuperLU_DIST with multiple Jacobi sweeps'
|
|
info=4010
|
|
goto 9999
|
|
|
|
case(mld_slu_)
|
|
!
|
|
! Use the LU factorization from SuperLU.
|
|
!
|
|
|
|
do i=1, prec%iprcparm(mld_smooth_sweeps_)
|
|
!
|
|
! Compute Y(k+1) = D^(-1)*(X-ND*Y(k)), where D and ND are the
|
|
! block diagonal part and the remaining part of the local matrix
|
|
! and Y(j) is the approximate solution at sweep j.
|
|
!
|
|
ty(1:n_row) = x(1:n_row)
|
|
call psb_spmm(-zone,prec%av(mld_ap_nd_),tx,zone,ty,&
|
|
& prec%desc_data,info,work=aux)
|
|
if(info /=0) goto 9999
|
|
|
|
call mld_zslu_solve(0,n_row,1,ty,n_row,prec%iprcparm(mld_slu_ptr_),info)
|
|
if(info /=0) goto 9999
|
|
tx(1:n_row) = ty(1:n_row)
|
|
end do
|
|
|
|
case(mld_umf_)
|
|
!
|
|
! Use the LU factorization from UMFPACK.
|
|
!
|
|
|
|
do i=1, prec%iprcparm(mld_smooth_sweeps_)
|
|
!
|
|
! Compute Y(k+1) = D^(-1)*(X-ND*Y(k)), where D and ND are the
|
|
! block diagonal part and the remaining part of the local matrix
|
|
! and Y(j) is the approximate solution at sweep j.
|
|
!
|
|
ty(1:n_row) = x(1:n_row)
|
|
call psb_spmm(-zone,prec%av(mld_ap_nd_),tx,zone,ty,&
|
|
& prec%desc_data,info,work=aux)
|
|
if(info /=0) goto 9999
|
|
|
|
call mld_zumf_solve(0,n_row,ww,ty,n_row,&
|
|
& prec%iprcparm(mld_umf_numptr_),info)
|
|
if(info /=0) goto 9999
|
|
tx(1:n_row) = ww(1:n_row)
|
|
end do
|
|
|
|
end select
|
|
|
|
!
|
|
! Put the result into the output vector Y.
|
|
!
|
|
call psb_geaxpby(alpha,tx,beta,y,desc_data,info)
|
|
|
|
|
|
deallocate(tx,ty)
|
|
|
|
else
|
|
|
|
info = 10
|
|
call psb_errpush(info,name,&
|
|
& i_err=(/2,prec%iprcparm(mld_smooth_sweeps_),0,0,0/))
|
|
goto 9999
|
|
|
|
endif
|
|
|
|
if (n_col <= size(work)) then
|
|
if ((4*n_col+n_col) <= size(work)) then
|
|
else
|
|
deallocate(aux)
|
|
endif
|
|
else
|
|
deallocate(ww,aux)
|
|
endif
|
|
|
|
call psb_erractionrestore(err_act)
|
|
return
|
|
|
|
9999 continue
|
|
call psb_errpush(info,name,i_err=int_err)
|
|
call psb_erractionrestore(err_act)
|
|
if (err_act.eq.psb_act_abort_) then
|
|
call psb_error()
|
|
return
|
|
end if
|
|
return
|
|
|
|
end subroutine mld_zbjac_aply
|
|
|