From a29a7f5dcfd7b3dcd9c436fcf859f6bc6b3eec00 Mon Sep 17 00:00:00 2001 From: Salvatore Filippone Date: Tue, 30 May 2017 13:42:58 +0000 Subject: [PATCH] mld2p4-2: tests/fileread/data_input.f90 tests/fileread/mld_cf_sample.f90 tests/fileread/mld_df_sample.f90 tests/fileread/mld_sf_sample.f90 tests/fileread/mld_zf_sample.f90 tests/fileread/runs/cfs.inp tests/fileread/runs/dfs.inp tests/fileread/runs/sfs.inp tests/fileread/runs/zfs.inp Fixed new Xf_sample. --- tests/fileread/data_input.f90 | 100 ++++- tests/fileread/mld_cf_sample.f90 | 669 +++++++++++++++++++++---------- tests/fileread/mld_df_sample.f90 | 667 ++++++++++++++++++++---------- tests/fileread/mld_sf_sample.f90 | 669 +++++++++++++++++++++---------- tests/fileread/mld_zf_sample.f90 | 667 ++++++++++++++++++++---------- tests/fileread/runs/cfs.inp | 65 +-- tests/fileread/runs/dfs.inp | 51 ++- tests/fileread/runs/sfs.inp | 67 ++-- tests/fileread/runs/zfs.inp | 69 ++-- 9 files changed, 2113 insertions(+), 911 deletions(-) diff --git a/tests/fileread/data_input.f90 b/tests/fileread/data_input.f90 index 829db94b..1e756e0e 100644 --- a/tests/fileread/data_input.f90 +++ b/tests/fileread/data_input.f90 @@ -41,10 +41,12 @@ module data_input use psb_base_mod, only : psb_spk_, psb_dpk_, psb_ipk_ interface read_data - module procedure read_char, read_int,& - & read_double, read_single, read_logical,& - & string_read_char, string_read_int,& - & string_read_double, string_read_single, & + module procedure read_char, read_int, read_int_array,& + & read_double, read_double_array, & + & read_single, read_single_array, read_logical,& + & string_read_char, string_read_int, string_read_int_array,& + & string_read_double, string_read_double_array,& + & string_read_single, string_read_single_array, & & string_read_logical end interface read_data interface trim_string @@ -85,6 +87,17 @@ contains call read_data(val,charbuf,marker) end subroutine read_int + + subroutine read_int_array(val,file,marker) + integer(psb_ipk_), intent(out) :: val(:) + integer(psb_ipk_), intent(in) :: file + character(len=1), optional, intent(in) :: marker + + read(file,'(a)')charbuf + call read_data(val,charbuf,marker) + + end subroutine read_int_array + subroutine read_single(val,file,marker) real(psb_spk_), intent(out) :: val integer(psb_ipk_), intent(in) :: file @@ -94,6 +107,17 @@ contains call read_data(val,charbuf,marker) end subroutine read_single + + subroutine read_single_array(val,file,marker) + real(psb_spk_), intent(out) :: val(:) + integer(psb_ipk_), intent(in) :: file + character(len=1), optional, intent(in) :: marker + + read(file,'(a)')charbuf + call read_data(val,charbuf,marker) + + end subroutine read_single_array + subroutine read_double(val,file,marker) real(psb_dpk_), intent(out) :: val integer(psb_ipk_), intent(in) :: file @@ -104,6 +128,16 @@ contains end subroutine read_double + subroutine read_double_array(val,file,marker) + real(psb_dpk_), intent(out) :: val(:) + integer(psb_ipk_), intent(in) :: file + character(len=1), optional, intent(in) :: marker + + read(file,'(a)')charbuf + call read_data(val,charbuf,marker) + + end subroutine read_double_array + subroutine string_read_char(val,file,marker) character(len=*), intent(out) :: val character(len=*), intent(in) :: file @@ -142,6 +176,25 @@ contains read(charbuf(1:idx-1),*) val end subroutine string_read_int + subroutine string_read_int_array(val,file,marker) + integer(psb_ipk_), intent(out) :: val(:) + character(len=*), intent(in) :: file + character(len=1), optional, intent(in) :: marker + character(len=1) :: marker_ + character(len=1024) :: charbuf + integer(psb_ipk_) :: idx + if (present(marker)) then + marker_ = marker + else + marker_ = def_marker + end if + read(file,'(a)')charbuf + charbuf = adjustl(charbuf) + idx=index(charbuf,marker_) + if (idx == 0) idx = len(charbuf)+1 + read(charbuf(1:idx-1),*) val(:) + end subroutine string_read_int_array + subroutine string_read_single(val,file,marker) real(psb_spk_), intent(out) :: val character(len=*), intent(in) :: file @@ -161,6 +214,25 @@ contains read(charbuf(1:idx-1),*) val end subroutine string_read_single + subroutine string_read_single_array(val,file,marker) + real(psb_spk_), intent(out) :: val(:) + character(len=*), intent(in) :: file + character(len=1), optional, intent(in) :: marker + character(len=1) :: marker_ + character(len=1024) :: charbuf + integer(psb_ipk_) :: idx + if (present(marker)) then + marker_ = marker + else + marker_ = def_marker + end if + read(file,'(a)')charbuf + charbuf = adjustl(charbuf) + idx=index(charbuf,marker_) + if (idx == 0) idx = len(charbuf)+1 + read(charbuf(1:idx-1),*) val(:) + end subroutine string_read_single_array + subroutine string_read_double(val,file,marker) real(psb_dpk_), intent(out) :: val character(len=*), intent(in) :: file @@ -180,6 +252,25 @@ contains read(charbuf(1:idx-1),*) val end subroutine string_read_double + subroutine string_read_double_array(val,file,marker) + real(psb_dpk_), intent(out) :: val(:) + character(len=*), intent(in) :: file + character(len=1), optional, intent(in) :: marker + character(len=1) :: marker_ + character(len=1024) :: charbuf + integer(psb_ipk_) :: idx + if (present(marker)) then + marker_ = marker + else + marker_ = def_marker + end if + read(file,'(a)')charbuf + charbuf = adjustl(charbuf) + idx=index(charbuf,marker_) + if (idx == 0) idx = len(charbuf)+1 + read(charbuf(1:idx-1),*) val(:) + end subroutine string_read_double_array + subroutine string_read_logical(val,file,marker) logical, intent(out) :: val character(len=*), intent(in) :: file @@ -196,7 +287,6 @@ contains charbuf = adjustl(charbuf) idx=index(charbuf,marker_) if (idx == 0) idx = len(charbuf)+1 - write(0,*) ' From string_read_logical: ',idx read(charbuf(1:idx-1),*) val end subroutine string_read_logical diff --git a/tests/fileread/mld_cf_sample.f90 b/tests/fileread/mld_cf_sample.f90 index b741c0a2..57b8f514 100644 --- a/tests/fileread/mld_cf_sample.f90 +++ b/tests/fileread/mld_cf_sample.f90 @@ -4,7 +4,7 @@ ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) ! -! (C) Copyright 2008, 2010, 2012, 2015, 2017 , 2016 +! (C) Copyright 2008, 2010, 2012, 2015, 2017 ! ! Salvatore Filippone Cranfield University, UK ! Ambra Abdullahi Hassan University of Rome Tor Vergata, IT @@ -47,35 +47,83 @@ program mld_cf_sample ! input parameters - character(len=40) :: kmethd, mtrx_file, rhs_file + + character(len=40) :: kmethd, mtrx_file, rhs_file, guess_file, sol_file character(len=2) :: filefmt + + ! Krylov solver data + type solverdata + character(len=40) :: kmethd ! Krylov solver + integer(psb_ipk_) :: istopc ! stopping criterion + integer(psb_ipk_) :: itmax ! maximum number of iterations + integer(psb_ipk_) :: itrace ! tracing + integer(psb_ipk_) :: irst ! restart + real(psb_spk_) :: eps ! stopping tolerance + end type solverdata + type(solverdata) :: s_choice + + ! preconditioner data type precdata - character(len=20) :: descr ! verbose description of the prec - character(len=10) :: prec ! overall prectype + + ! preconditioner type + character(len=40) :: descr ! verbose description of the prec + character(len=10) :: ptype ! preconditioner type + + ! general AMG data + character(len=16) :: mltype ! AMG cycle type + integer(psb_ipk_) :: otr_sweeps ! number of AMG cycles + integer(psb_ipk_) :: maxlevs ! maximum number of levels in AMG preconditioner + + ! AMG aggregation + character(len=16) :: aggrkind ! aggregation type: SMOOTHED, NONSMOOTHED + character(len=16) :: aggr_alg ! parallel aggregation algorithm: DEC, SYMDEC + character(len=16) :: aggr_ord ! ordering for aggregation: NATURAL, DEGREE + character(len=16) :: aggr_filter ! filtering: FILTER, NO_FILTER + real(psb_spk_) :: mnaggratio ! minimum aggregation ratio + real(psb_spk_), allocatable :: athresv(:) ! smoothed aggregation threshold vector + integer(psb_ipk_) :: thrvsz ! size of threshold vector + real(psb_spk_) :: athres ! smoothed aggregation threshold + real(psb_spk_) :: ascale ! smoothed aggregation scale factor for threshold + character(len=16) :: aggr_omalg ! algorithm for estimating omega parameter + character(len=16) :: aggr_eig ! Eigenvalue estimation procedure + real(psb_spk_) :: omega_val ! Eigenvalue estimate value + integer(psb_ipk_) :: csize ! minimum size of coarsest matrix + + ! AMG smoother or pre-smoother; also 1-lev preconditioner + character(len=16) :: smther ! (pre-)smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps ! (pre-)smoother / 1-lev prec. sweeps integer(psb_ipk_) :: novr ! number of overlap layers - integer(psb_ipk_) :: jsweeps ! Jacobi/smoother sweeps character(len=16) :: restr ! restriction over application of AS character(len=16) :: prol ! prolongation over application of AS - character(len=16) :: solve ! factorization type: ILU, SuperLU, UMFPACK - integer(psb_ipk_) :: fill ! fillin for factorization - real(psb_spk_) :: thr ! threshold for fact. ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: nlev ! number of levels in multilevel prec. - character(len=16) :: aggrkind ! smoothed, raw aggregation - character(len=16) :: aggr_alg ! aggregation algorithm (currently only decoupled) - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: mltype ! additive or multiplicative multi-level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - character(len=16) :: cmat ! coarse mat: distributed, replicated - character(len=16) :: csolve ! coarse solver: bjac, umf, slu, sludist - character(len=16) :: csbsolve ! coarse subsolver: ILU, ILU(T), SuperLU, UMFPACK - integer(psb_ipk_) :: cfill ! fillin for coarse factorization - real(psb_spk_) :: cthres ! threshold for coarse fact. ILU(T) - integer(psb_ipk_) :: cjswp ! block-Jacobi sweeps - real(psb_spk_) :: athres ! smoothed aggregation threshold - real(psb_spk_) :: ascale ! smoothed aggregation scale factor + character(len=16) :: solve ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill ! fill-in for incomplete LU factorization + real(psb_spk_) :: thr ! threshold for ILUT factorization + + ! AMG post-smoother; ignored by 1-lev preconditioner + character(len=16) :: smther2 ! post-smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps2 ! post-smoother sweeps + integer(psb_ipk_) :: novr2 ! number of overlap layers + character(len=16) :: restr2 ! restriction over application of AS + character(len=16) :: prol2 ! prolongation over application of AS + character(len=16) :: solve2 ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill2 ! fill-in for incomplete LU factorization + real(psb_spk_) :: thr2 ! threshold for ILUT factorization + + ! coarsest-level solver + character(len=16) :: cmat ! coarsest matrix layout: REPL, DIST + character(len=16) :: csolve ! coarsest-lev solver: BJAC, SLUDIST (distr. + ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU + ! (repl. mat.) + character(len=16) :: csbsolve ! coarsest-lev local subsolver: ILU, ILUT, + ! MILU, UMF, MUMPS, SLU + integer(psb_ipk_) :: cfill ! fill-in for incomplete LU factorization + real(psb_spk_) :: cthres ! threshold for ILUT factorization + integer(psb_ipk_) :: cjswp ! sweeps for GS or JAC coarsest-lev subsolver + end type precdata - type(precdata) :: prec_choice + type(precdata) :: p_choice ! sparse matrices type(psb_cspmat_type) :: a, aux_a @@ -83,10 +131,10 @@ program mld_cf_sample ! preconditioner data type(mld_cprec_type) :: prec ! dense matrices - complex(psb_spk_), allocatable, target :: aux_b(:,:), d(:) + complex(psb_spk_), allocatable, target :: aux_b(:,:), d(:), aux_g(:,:), aux_x(:,:) complex(psb_spk_), allocatable , save :: x_col_glob(:), r_col_glob(:) - complex(psb_spk_), pointer :: b_col_glob(:) - type(psb_c_vect_type) :: b_col, x_col, r_col + complex(psb_spk_), pointer :: b_col_glob(:), ref_col_glob(:), guess_col_glob(:) + type(psb_c_vect_type) :: b_col, x_col, r_col, ref_col ! communications data structure type(psb_desc_type):: desc_a @@ -94,24 +142,23 @@ program mld_cf_sample integer(psb_ipk_) :: ictxt, iam, np ! solver paramters - integer(psb_ipk_) :: iter, itmax, ierr, itrace, ircode, ipart,& - & methd, istopc, irst, nlv + integer(psb_ipk_) :: iter, ircode, ipart, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_spk_) :: err, eps + real(psb_spk_) :: err character(len=5) :: afmt character(len=20) :: name, renum integer(psb_ipk_), parameter :: iunit=12 integer(psb_ipk_) :: iparm(20) - character(len=40) :: fprefix ! other variables - integer(psb_ipk_) :: i,info,j,m_problem - integer(psb_ipk_) :: internal, m,ii,nnzero - real(psb_spk_) :: t1, t2, tprec, thier, tslv - real(psb_spk_) :: r_amax, b_amax, scale,resmx,resmxp - integer(psb_ipk_) :: nrhs, nrow, n_row, dim, nv, ne - integer(psb_ipk_), allocatable :: ivg(:), ipv(:) + integer(psb_ipk_) :: i, info, j, k, m_problem + integer(psb_ipk_) :: lbw, ubw, prf + real(psb_dpk_) :: t1, t2, tprec, thier, tslv + real(psb_spk_) :: resmx, resmxp, xdiffn2, xdiffni, xni, xn2 + integer(psb_ipk_) :: nrhs, nv + integer(psb_ipk_), allocatable :: ivg(:), ipv(:), perm(:) + logical :: have_guess=.false., have_ref=.false. call psb_init(ictxt) call psb_info(ictxt,iam,np) @@ -131,41 +178,55 @@ program mld_cf_sample ! Hello world ! if (iam == psb_root_) then - write(*,*) 'Welcome to MLD2P4 version: ',mld_version_string_ - write(*,*) 'This is the ',trim(name),' sample program' + write(psb_out_unit,*) ' ' + write(psb_out_unit,*) 'Welcome to MLD2P4 version: ',mld_version_string_ + write(psb_out_unit,*) 'This is the ',trim(name),' sample test program' + write(psb_out_unit,*) ' ' end if ! - ! get parameters + ! get parameters ! - call get_parms(ictxt,mtrx_file,rhs_file,filefmt,kmethd,& - & prec_choice,ipart,afmt,istopc,itmax,itrace,irst,eps) + call get_parms(ictxt,mtrx_file,rhs_file,guess_file,sol_file,filefmt, & + & ipart,afmt,s_choice,p_choice) call psb_barrier(ictxt) t1 = psb_wtime() - ! read the input matrix to be processed and (possibly) the rhs + ! read the input matrix to be processed and (possibly) the rhs, + ! the initial guess and the reference solution nrhs = 1 if (iam == psb_root_) then - select case(psb_toupper(filefmt)) + select case(psb_toupper(filefmt)) + case('MM') ! For Matrix Market we have an input file for the matrix - ! and an (optional) second file for the RHS. + ! and (optional) separate files for the rhs, the initial guess + ! and the reference solution call mm_mat_read(aux_a,info,iunit=iunit,filename=mtrx_file) - if (info == psb_success_) then - if (rhs_file /= 'NONE') then - call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) - end if + if ((info == psb_success_).and.(rhs_file /= 'NONE')) & + & call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) + if ((info == psb_success_).and.(guess_file /= 'NONE')) then + call mm_array_read(aux_g,info,iunit=iunit,filename=guess_file) + have_guess = .true. + end if + if ((info == psb_success_).and.(sol_file /= 'NONE')) then + call mm_array_read(aux_x,info,iunit=iunit,filename=sol_file) + have_ref = .true. end if case ('HB') ! For Harwell-Boeing we have a single file which may or may not - ! contain an RHS. - call hb_read(aux_a,info,iunit=iunit,b=aux_b,filename=mtrx_file) + ! contain rhs, initial guess and reference solution. + call hb_read(aux_a,info,iunit=iunit,b=aux_b,& + & g=aux_g,x=aux_x,filename=mtrx_file) + have_guess = allocated(aux_g) + have_ref = allocated(aux_x) case default info = -1 write(psb_err_unit,*) 'Wrong choice for fileformat ', filefmt end select + if (info /= psb_success_) then write(psb_err_unit,*) 'Error while reading input matrix ' call psb_abort(ictxt) @@ -173,6 +234,8 @@ program mld_cf_sample m_problem = aux_a%get_nrows() call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) ! At this point aux_b may still be unallocated if (psb_size(aux_b,dim=ione) == m_problem) then @@ -181,24 +244,82 @@ program mld_cf_sample b_col_glob =>aux_b(:,1) else write(psb_out_unit,'("Generating an rhs...")') - write(psb_out_unit,'(" ")') call psb_realloc(m_problem,1,aux_b,ircode) if (ircode /= 0) then call psb_errpush(psb_err_alloc_dealloc_,name) goto 9999 endif - b_col_glob => aux_b(:,1) do i=1, m_problem b_col_glob(i) = 1.d0 enddo endif + + if ((have_guess).and.(psb_size(aux_g,dim=ione) == m_problem)) then + ! if any initial guess were present, broadcast the first one + write(psb_err_unit,'("Ok, got an initial guess ")') + guess_col_glob =>aux_g(:,1) + else + write(psb_out_unit,'("Generating an initial guess...")') + call psb_realloc(m_problem,1,aux_g,ircode) + if (ircode /= 0) then + call psb_errpush(psb_err_alloc_dealloc_,name) + goto 9999 + endif + guess_col_glob => aux_g(:,1) + do i=1, m_problem + guess_col_glob(i) = 0.d0 + enddo + endif + + if ((have_ref).and.(psb_size(aux_x,dim=ione) == m_problem)) then + ! if any reference were present, broadcast the first one + write(psb_err_unit,'("Ok, got a reference solution ")') + ref_col_glob =>aux_x(:,1) + else + write(psb_out_unit,'("No reference solution...")') + endif + + ! clean zeros in the input matrix + call aux_a%clean_zeros(info) + else call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) + end if + + + ! + ! Renumbering (NONE for the moment) + ! + if (iam==psb_root_) then + renum='NONE' + + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,'(" ")') + write(psb_out_unit,*) 'Bandwidth and profile: ',lbw,ubw,prf + write(psb_out_unit,*) 'Renumbering algorithm: ',psb_toupper(renum) + + if (trim(psb_toupper(renum))/='NONE') then + call psb_mat_renum(renum,aux_a,info,perm=perm) + if (info /= 0) then + write(psb_err_unit,*) 'Error from RENUM',info + goto 9999 + end if + call psb_gelp('N',perm(1:m_problem),b_col_glob(1:m_problem),info) + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,*) 'Bandwidth and profile (renumberd):',lbw,ubw,prf + end if + + write(psb_out_unit,'(" ")') + end if + ! ! switch over different partition types - if (ipart == 0) then + ! + if (ipart == 0) then call psb_barrier(ictxt) if (iam == psb_root_) write(psb_out_unit,'("Partition type: block")') allocate(ivg(m_problem),ipv(np)) @@ -211,10 +332,8 @@ program mld_cf_sample if (iam == psb_root_) then write(psb_out_unit,'("Partition type: graph")') write(psb_out_unit,'(" ")') - ! write(psb_err_unit,'("Build type: graph")') call build_mtpart(aux_a,np) endif -!!$ call psb_barrier(ictxt) call distr_mtpart(psb_root_,ictxt) call getv_mtpart(ivg) call psb_matdist(aux_a, a, ictxt,desc_a,info,fmt=afmt,v=ivg) @@ -223,89 +342,136 @@ program mld_cf_sample call psb_matdist(aux_a, a, ictxt, desc_a,info,fmt=afmt,parts=part_block) end if - call psb_scatter(b_col_glob,b_col,desc_a,info,root=psb_root_) + ! + ! Scatter rhs, initial guess and reference solution + ! + call psb_geall(b_col,desc_a,info) call psb_geall(x_col,desc_a,info) - call x_col%zero() - call psb_geasb(x_col,desc_a,info) - call psb_geall(r_col,desc_a,info) - call r_col%zero() - call psb_geasb(r_col,desc_a,info) - t2 = psb_wtime() - t1 - + if (have_ref) call psb_geall(ref_col,desc_a,info) + + if (iam == psb_root_) write(psb_out_unit,'("Scatter rhs")') + call psb_scatter(b_col_glob,b_col,desc_a,info) + if (iam == psb_root_) write(psb_out_unit,'("Scatter initial guess")') + call psb_scatter(guess_col_glob,x_col,desc_a,info) + if (have_ref) then + if (iam == psb_root_) write(psb_out_unit,'("Scatter reference solution")') + call psb_scatter(ref_col_glob,ref_col,desc_a,info) + end if + t2 = psb_wtime() - t1 call psb_amx(ictxt, t2) if (iam == psb_root_) then write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Time to read and partition matrix : ",es12.5)')t2 + write(psb_out_unit,'("Time to read and partition matrix, rhs(, guess, ref sol) : ",es12.5)')t2 write(psb_out_unit,'(" ")') - write(psb_out_unit,*) 'Preconditioner: ',prec_choice%descr end if - ! - - if (psb_toupper(prec_choice%prec) == 'ML') then - call mld_precinit(prec,prec_choice%prec, info) - if (prec_choice%nlev > 0) & - & call mld_precset(prec,'max_prec_levs', prec_choice%nlev, info) - call mld_precset(prec,'aggr_kind', prec_choice%aggrkind,info) - call mld_precset(prec,'aggr_alg', prec_choice%aggr_alg,info) - call mld_precset(prec,'aggr_ord', prec_choice%aggr_ord,info) - call mld_precset(prec,'aggr_scale', prec_choice%ascale, info) - call mld_precset(prec,'aggr_thresh', prec_choice%athres, info) - call mld_precset(prec,'smoother_type', prec_choice%smther, info) - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) - call mld_precset(prec,'ml_type', prec_choice%mltype, info) - call mld_precset(prec,'smoother_pos', prec_choice%smthpos, info) - call mld_precset(prec,'coarse_solve', prec_choice%csolve, info) - call mld_precset(prec,'coarse_subsolve', prec_choice%csbsolve,info) - call mld_precset(prec,'coarse_mat', prec_choice%cmat, info) - call mld_precset(prec,'coarse_fillin', prec_choice%cfill, info) - call mld_precset(prec,'coarse_iluthrs', prec_choice%cthres, info) - call mld_precset(prec,'coarse_sweeps', prec_choice%cjswp, info) - ! building the preconditioner + ! + ! initialize the preconditioner + ! + if (psb_toupper(p_choice%ptype) == 'ML') then + ! multilevel preconditioner + call prec%init(p_choice%ptype,info) + + call prec%set('ml_type', p_choice%mltype, info) + call prec%set('outer_sweeps', p_choice%otr_sweeps,info) + if (p_choice%csize>0)& + & call prec%set('coarse_aggr_size', p_choice%csize, info) + if (p_choice%mnaggratio>0)& + & call prec%set('min_aggr_ratio', p_choice%mnaggratio, info) + if (p_choice%maxlevs>0)& + & call prec%set('max_prec_levs', p_choice%maxlevs, info) + if (p_choice%ascale > dzero) & + & call prec%set('aggr_scale', p_choice%ascale, info) + if (p_choice%athres >= dzero) & + & call prec%set('aggr_thresh', p_choice%athres, info) + if (p_choice%thrvsz>0) then + do k=1,min(p_choice%thrvsz,size(prec%precv)-1) + call prec%set('aggr_thresh', p_choice%athresv(k), info,ilev=(k+1)) + end do + end if + + call prec%set('aggr_kind', p_choice%aggrkind, info) + call prec%set('aggr_alg', p_choice%aggr_alg, info) + call prec%set('aggr_ord', p_choice%aggr_ord, info) + call prec%set('aggr_filter', p_choice%aggr_filter,info) + call prec%set('aggr_omega_alg', p_choice%aggr_omalg, info) + if (psb_toupper(p_choice%aggr_omalg) == 'EIG_EST') then + call prec%set('aggr_eig', p_choice%aggr_eig, info) + else if (psb_toupper(p_choice%aggr_omalg) == 'USER_CHOICE') then + call prec%set('aggr_omega_val', p_choice%omega_val, info) + end if + call prec%set('coarse_solve', p_choice%csolve, info) + if (psb_toupper(p_choice%csolve) == 'BJAC') & + & call prec%set('coarse_subsolve', p_choice%csbsolve, info) + call prec%set('coarse_mat', p_choice%cmat, info) + call prec%set('coarse_fillin', p_choice%cfill, info) + call prec%set('coarse_iluthrs', p_choice%cthres, info) + call prec%set('coarse_sweeps', p_choice%cjswp, info) + + + call prec%set('smoother_type', p_choice%smther, info) + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + + if (psb_toupper(p_choice%smther2) /= 'NONE') then + call prec%set('smoother_type', p_choice%smther2, info,pos='post') + call prec%set('smoother_sweeps', p_choice%jsweeps2, info,pos='post') + call prec%set('sub_ovr', p_choice%novr2, info,pos='post') + call prec%set('sub_restr', p_choice%restr2, info,pos='post') + call prec%set('sub_prol', p_choice%prol2, info,pos='post') + call prec%set('sub_solve', p_choice%solve2, info,pos='post') + call prec%set('sub_fillin', p_choice%fill2, info,pos='post') + call prec%set('sub_iluthrs', p_choice%thr2, info,pos='post') + end if + + ! build the preconditioner call psb_barrier(ictxt) t1 = psb_wtime() - call mld_hierarchy_bld(a,desc_a,prec,info) + call prec%hierarchy_build(a,desc_a,info) thier = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if call psb_barrier(ictxt) t1 = psb_wtime() - call mld_smoothers_bld(a,desc_a,prec,info) + call prec%smoothers_build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if else + ! 1-level preconditioner nlv = 1 - call mld_precinit(prec,prec_choice%prec,info) - if (psb_toupper(prec_choice%prec) /= 'NONE') then - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) + call prec%init(p_choice%ptype,info) + + if (psb_toupper(p_choice%ptype) /= 'NONE') then + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + !!! call prec%set('solver_sweeps', p_choice%svsweeps, info) end if - ! building the preconditioner + + ! build the preconditioner thier = dzero t1 = psb_wtime() - call mld_precbld(a,desc_a,prec,info) + call prec%build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if end if @@ -314,6 +480,8 @@ program mld_cf_sample call psb_amx(ictxt, tprec) if(iam == psb_root_) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'("Preconditioner: ",a)') trim(p_choice%descr) write(psb_out_unit,'("Preconditioner time: ",es12.5)')thier+tprec write(psb_out_unit,'(" ")') end if @@ -321,43 +489,71 @@ program mld_cf_sample iparm = 0 call psb_barrier(ictxt) t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b_col,x_col,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + call psb_krylov(s_choice%kmethd,a,prec,b_col,x_col,s_choice%eps,& + & desc_a,info,itmax=s_choice%itmax,iter=iter,err=err,itrace=s_choice%itrace,& + & istop=s_choice%istopc,irst=s_choice%irst) call psb_barrier(ictxt) tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + + ! compute residual norms + call psb_geall(r_col,desc_a,info) + call r_col%zero() + call psb_geasb(r_col,desc_a,info) call psb_geaxpby(cone,b_col,czero,r_col,desc_a,info) call psb_spmm(-cone,a,x_col,cone,r_col,desc_a,info) resmx = psb_genrm2(r_col,desc_a,info) resmxp = psb_geamax(r_col,desc_a,info) - + + ! compute error in solution + if (have_ref) then + call psb_geaxpby(-cone,x_col,cone,ref_col,desc_a,info) + xdiffn2 = psb_genrm2(ref_col,desc_a,info) + xdiffni = psb_geamax(ref_col,desc_a,info) + xn2 = psb_genrm2(ref_col,desc_a,info) + xni = psb_geamax(ref_col,desc_a,info) + end if + amatsize = a%sizeof() descsize = desc_a%sizeof() precsize = prec%sizeof() call psb_sum(ictxt,amatsize) call psb_sum(ictxt,descsize) call psb_sum(ictxt,precsize) + call prec%descr(info) if (iam == psb_root_) then - call mld_precdescr(prec,info) write(psb_out_unit,'("Matrix: ",a)')mtrx_file write(psb_out_unit,'("Computed solution on ",i8," processors")')np - write(psb_out_unit,'("Iterations to convergence : ",i6)')iter - write(psb_out_unit,'("Error estimate on exit : ",es12.5)')err + write(psb_out_unit,'("Krylov method : ",a)') trim(s_choice%kmethd) + write(psb_out_unit,'("Preconditioner : ",a)') trim(p_choice%descr) + write(psb_out_unit,'("Iterations to convergence : ",i12)')iter + write(psb_out_unit,'("Relative error estimate on exit : ",es12.5)') err write(psb_out_unit,'("Number of levels in hierarchy : ",i12)') prec%get_nlevs() write(psb_out_unit,'("Time to build hierarchy : ",es12.5)')thier write(psb_out_unit,'("Time to build smoothers : ",es12.5)')tprec write(psb_out_unit,'("Total time for preconditioner : ",es12.5)')tprec+thier write(psb_out_unit,'("Time to solve system : ",es12.5)')tslv - write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/(iter) - write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec - write(psb_out_unit,'("Residual norm 2 : ",es12.5)')resmx - write(psb_out_unit,'("Residual norm inf : ",es12.5)')resmxp + write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/iter + write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec+thier + write(psb_out_unit,'("Residual 2-norm : ",es12.5)')resmx + write(psb_out_unit,'("Residual inf-norm : ",es12.5)')resmxp write(psb_out_unit,'("Total memory occupation for A : ",i12)')amatsize write(psb_out_unit,'("Total memory occupation for DESC_A : ",i12)')descsize write(psb_out_unit,'("Total memory occupation for PREC : ",i12)')precsize write(psb_out_unit,'("Storage format for A : ",a )')a%get_fmt() write(psb_out_unit,'("Storage format for DESC_A : ",a )')desc_a%get_fmt() + if (have_ref) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,5x,a8)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(2-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffn2,xn2,xdiffn2/xn2 + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,4x,a10)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(inf-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffni,xni,xdiffni/xni + end if + end if call psb_gather(x_col_glob,x_col,desc_a,info,root=psb_root_) @@ -367,13 +563,15 @@ program mld_cf_sample if (iam == psb_root_) then write(psb_err_unit,'(" ")') write(psb_err_unit,'("Saving x on file")') - write(20,*) 'matrix: ',mtrx_file - write(20,*) 'computed solution on ',np,' processors.' - write(20,*) 'iterations to convergence: ',iter - write(20,*) 'error estimate (infinity norm) on exit:', & - & ' ||r||/(||a||||x||+||b||) = ',err - write(20,'("Residual norm 2 : ",es12.5)')resmx - write(20,'("Residual norm inf : ",es12.5)')resmxp + write(20,*) 'Matrix: ',mtrx_file + write(20,*) 'Krylov method:',trim(s_choice%kmethd) + write(20,*) 'Preconditioner:',trim(p_choice%descr) + write(20,*) 'Computed solution on ',np,' processors.' + write(20,*) 'Iterations to convergence: ',iter + write(20,*) 'Error estimate (infinity norm) on exit:', & + & ' ||r||/||b|| (inf-norm) = ',err + write(20,'(" Residual 2-norm 2 : ",es12.5)')resmx + write(20,'(" Residual inf-norm : ",es12.5)')resmxp write(20,'(a8,4(2x,a20))') 'I','X(I)','R(I)','B(I)' do i=1,m_problem write(20,998) i,x_col_glob(i),r_col_glob(i),b_col_glob(i) @@ -383,10 +581,12 @@ program mld_cf_sample 993 format(i6,4(1x,e12.6)) - call psb_gefree(b_col, desc_a,info) - call psb_gefree(x_col, desc_a,info) + call psb_gefree(b_col,desc_a,info) + call psb_gefree(x_col,desc_a,info) + call psb_gefree(r_col,desc_a,info) + call psb_gefree(ref_col,desc_a,info) call psb_spfree(a, desc_a,info) - call mld_precfree(prec,info) + call prec%free(info) call psb_cdfree(desc_a,info) call psb_exit(ictxt) @@ -399,101 +599,161 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(icontxt,mtrx,rhs,filefmt,kmethd,& - & prec, ipart,afmt,istopc,itmax,itrace,irst,eps) + subroutine get_parms(icontxt,mtrx,rhs,guess,sol,filefmt,ipart,afmt,solve,prec) use psb_base_mod implicit none - integer(psb_ipk_) :: icontxt - character(len=*) :: kmethd, mtrx, rhs, afmt,filefmt + integer(psb_ipk_) :: icontxt, ipart + character(len=*) :: mtrx, rhs, guess, sol, filefmt, afmt + type(solverdata) :: solve type(precdata) :: prec - real(psb_spk_) :: eps - integer(psb_ipk_) :: iret, istopc,itmax,itrace, ipart, irst - integer(psb_ipk_) :: iam, nm, np, i + integer(psb_ipk_) :: iam, nm, np call psb_info(icontxt,iam,np) if (iam == psb_root_) then - ! read input parameters - call read_data(mtrx,psb_inp_unit) - call read_data(rhs,psb_inp_unit) - call read_data(filefmt,psb_inp_unit) - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(ipart,psb_inp_unit) - call read_data(istopc,psb_inp_unit) - call read_data(itmax,psb_inp_unit) - call read_data(itrace,psb_inp_unit) - call read_data(irst,psb_inp_unit) - call read_data(eps,psb_inp_unit) + ! read input data + ! + ! input files + call read_data(mtrx,psb_inp_unit) ! matrix file + call read_data(rhs,psb_inp_unit) ! rhs file + call read_data(guess,psb_inp_unit) ! starting guess file + call read_data(sol,psb_inp_unit) ! solution file (for comparison) + call read_data(filefmt,psb_inp_unit) ! format of files + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(ipart,psb_inp_unit) ! partition type + ! Krylov solver data + call read_data(solve%kmethd,psb_inp_unit) ! Krylov solver + call read_data(solve%istopc,psb_inp_unit) ! stopping criterion + call read_data(solve%itmax,psb_inp_unit) ! max num iterations + call read_data(solve%itrace,psb_inp_unit) ! tracing + call read_data(solve%irst,psb_inp_unit) ! restart + call read_data(solve%eps,psb_inp_unit) ! tolerance + ! preconditioner type call read_data(prec%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prec%prec,psb_inp_unit) ! overall prectype - call read_data(prec%novr,psb_inp_unit) ! number of overlap layers - call read_data(prec%restr,psb_inp_unit) ! restriction over application of as - call read_data(prec%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prec%solve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%fill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%thr,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%jsweeps,psb_inp_unit) ! Jacobi sweeps for PJAC - if (psb_toupper(prec%prec) == 'ML') then - call read_data(prec%nlev,psb_inp_unit) ! Number of levels in multilevel prec. - call read_data(prec%smther,psb_inp_unit) ! Smoother type. - call read_data(prec%aggrkind,psb_inp_unit) ! smoothed/raw aggregatin - call read_data(prec%aggr_alg,psb_inp_unit) ! local or global aggregation - call read_data(prec%aggr_ord,psb_inp_unit) ! Ordering for aggregation - call read_data(prec%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prec%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prec%cmat,psb_inp_unit) ! coarse mat - call read_data(prec%csolve,psb_inp_unit) ! Factorization type: BJAC, SuperLU, UMFPACK. - call read_data(prec%csbsolve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%cfill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%cthres,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%cjswp,psb_inp_unit) ! Jacobi sweeps - call read_data(prec%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prec%ascale,psb_inp_unit) ! smoother aggr thresh + call read_data(prec%ptype,psb_inp_unit) ! preconditioner type + ! general AMG data + call read_data(prec%mltype,psb_inp_unit) ! AMG cycle type + call read_data(prec%otr_sweeps,psb_inp_unit) ! number of AMG cycles + call read_data(prec%maxlevs,psb_inp_unit) ! max number of levels in AMG prec + call read_data(prec%csize,psb_inp_unit) ! min size coarsest mat + ! aggregation + call read_data(prec%aggrkind,psb_inp_unit) ! aggregation type + call read_data(prec%aggr_alg,psb_inp_unit) ! parallel aggregation alg + call read_data(prec%aggr_ord,psb_inp_unit) ! ordering for aggregation + call read_data(prec%aggr_filter,psb_inp_unit) ! filtering + call read_data(prec%mnaggratio,psb_inp_unit) ! minimum aggregation ratio + call read_data(prec%thrvsz,psb_inp_unit) ! size of aggr thresh vector + if (prec%thrvsz > 0) then + call psb_realloc(prec%thrvsz,prec%athresv,info) + call read_data(prec%athresv,psb_inp_unit) ! aggr thresh vector + else + read(psb_inp_unit,*) ! dummy read to skip a record end if + call read_data(prec%athres,psb_inp_unit) ! smoothed aggr thresh + call read_data(prec%aggr_omalg,psb_inp_unit) ! alg for estimating omega + call read_data(prec%aggr_eig,psb_inp_unit) ! alg for estimating omega + call read_data(prec%omega_val,psb_inp_unit) ! alg for estimating omega + ! AMG smoother (or pre-smoother) / 1-lev preconditioner + call read_data(prec%smther,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps,psb_inp_unit) ! (pre-)smoother / 1-lev prec sweeps + call read_data(prec%novr,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve,psb_inp_unit) ! local subsolver + call read_data(prec%fill,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr,psb_inp_unit) ! threshold for ILUT + ! AMG post-smoother + call read_data(prec%smther2,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps2,psb_inp_unit) ! (post-)smoother sweeps + call read_data(prec%novr2,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr2,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol2,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve2,psb_inp_unit) ! local subsolver + call read_data(prec%fill2,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr2,psb_inp_unit) ! threshold for ILUT + ! coasest-level solver + call read_data(prec%csolve,psb_inp_unit) ! coarsest-lev solver + call read_data(prec%cmat,psb_inp_unit) ! coarsest mat layout + call read_data(prec%csbsolve,psb_inp_unit) ! coarsest-lev subsolver + call read_data(prec%cfill,psb_inp_unit) ! fill-in for incompl LU + call read_data(prec%cthres,psb_inp_unit) ! Threshold for ILUT + call read_data(prec%cjswp,psb_inp_unit) ! sweeps for GS/JAC subsolver end if call psb_bcast(icontxt,mtrx) call psb_bcast(icontxt,rhs) + call psb_bcast(icontxt,guess) + call psb_bcast(icontxt,sol) call psb_bcast(icontxt,filefmt) - call psb_bcast(icontxt,kmethd) call psb_bcast(icontxt,afmt) call psb_bcast(icontxt,ipart) - call psb_bcast(icontxt,istopc) - call psb_bcast(icontxt,itmax) - call psb_bcast(icontxt,itrace) - call psb_bcast(icontxt,irst) - call psb_bcast(icontxt,eps) - call psb_bcast(icontxt,prec%descr) ! verbose description of the prec - call psb_bcast(icontxt,prec%prec) ! overall prectype - call psb_bcast(icontxt,prec%novr) ! number of overlap layers - call psb_bcast(icontxt,prec%restr) ! restriction over application of as - call psb_bcast(icontxt,prec%prol) ! prolongation over application of as - call psb_bcast(icontxt,prec%solve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%fill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%thr) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%jsweeps) ! Jacobi sweeps - if (psb_toupper(prec%prec) == 'ML') then - call psb_bcast(icontxt,prec%smther) ! Smoother type. - call psb_bcast(icontxt,prec%nlev) ! Number of levels in multilevel prec. - call psb_bcast(icontxt,prec%aggrkind) ! smoothed/raw aggregatin - call psb_bcast(icontxt,prec%aggr_alg) ! local or global aggregation - call psb_bcast(icontxt,prec%aggr_ord) ! Ordering for aggregation - call psb_bcast(icontxt,prec%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(icontxt,prec%smthpos) ! side: pre, post, both smoothing - call psb_bcast(icontxt,prec%cmat) ! coarse mat - call psb_bcast(icontxt,prec%csolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%csbsolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%cfill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%cthres) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%cjswp) ! Jacobi sweeps - call psb_bcast(icontxt,prec%athres) ! smoother aggr thresh - call psb_bcast(icontxt,prec%ascale) ! smoother aggr scale factor + + call psb_bcast(icontxt,solve%kmethd) + call psb_bcast(icontxt,solve%istopc) + call psb_bcast(icontxt,solve%itmax) + call psb_bcast(icontxt,solve%itrace) + call psb_bcast(icontxt,solve%irst) + call psb_bcast(icontxt,solve%eps) + + call psb_bcast(icontxt,prec%descr) + call psb_bcast(icontxt,prec%ptype) + + ! broadcast first (pre-)smoother / 1-lev prec data + call psb_bcast(icontxt,prec%smther) ! actually not needed for 1-lev precs + call psb_bcast(icontxt,prec%jsweeps) + call psb_bcast(icontxt,prec%novr) + call psb_bcast(icontxt,prec%restr) + call psb_bcast(icontxt,prec%prol) + call psb_bcast(icontxt,prec%solve) + call psb_bcast(icontxt,prec%fill) + call psb_bcast(icontxt,prec%thr) + + ! broadcast (other) AMG parameters + if (psb_toupper(prec%ptype) == 'ML') then + + call psb_bcast(icontxt,prec%mltype) + call psb_bcast(icontxt,prec%otr_sweeps) + call psb_bcast(icontxt,prec%maxlevs) + + call psb_bcast(icontxt,prec%smther2) + call psb_bcast(icontxt,prec%jsweeps2) + call psb_bcast(icontxt,prec%novr2) + call psb_bcast(icontxt,prec%restr2) + call psb_bcast(icontxt,prec%prol2) + call psb_bcast(icontxt,prec%solve2) + call psb_bcast(icontxt,prec%fill2) + call psb_bcast(icontxt,prec%thr2) + + call psb_bcast(icontxt,prec%aggrkind) + call psb_bcast(icontxt,prec%aggr_alg) + call psb_bcast(icontxt,prec%aggr_ord) + call psb_bcast(icontxt,prec%aggr_filter) + call psb_bcast(icontxt,prec%mnaggratio) + call psb_bcast(ictxt,prec%thrvsz) + if (prec%thrvsz > 0) then + if (iam /= psb_root_) call psb_realloc(prec%thrvsz,prec%athresv,info) + call psb_bcast(ictxt,prec%athresv) + end if + call psb_bcast(ictxt,prec%athres) + call psb_bcast(ictxt,prec%ascale) + call psb_bcast(ictxt,prec%aggr_omalg) + call psb_bcast(ictxt,prec%aggr_eig) + call psb_bcast(ictxt,prec%omega_val) + + call psb_bcast(icontxt,prec%csize) + call psb_bcast(icontxt,prec%cmat) + call psb_bcast(icontxt,prec%csolve) + call psb_bcast(icontxt,prec%csbsolve) + call psb_bcast(icontxt,prec%cfill) + call psb_bcast(icontxt,prec%cthres) + call psb_bcast(icontxt,prec%cjswp) + end if end subroutine get_parms + subroutine pr_usage(iout) integer(psb_ipk_) iout write(iout, *) ' number of parameters is incorrect!' @@ -511,4 +771,5 @@ contains write(iout, *) ' ptype partition strategy default 0' write(iout, *) ' 0: block partition ' end subroutine pr_usage + end program mld_cf_sample diff --git a/tests/fileread/mld_df_sample.f90 b/tests/fileread/mld_df_sample.f90 index fc05d14e..9fe8863e 100644 --- a/tests/fileread/mld_df_sample.f90 +++ b/tests/fileread/mld_df_sample.f90 @@ -4,7 +4,7 @@ ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) ! -! (C) Copyright 2008, 2010, 2012, 2015, 2017 , 2016 +! (C) Copyright 2008, 2010, 2012, 2015, 2017 ! ! Salvatore Filippone Cranfield University, UK ! Ambra Abdullahi Hassan University of Rome Tor Vergata, IT @@ -47,35 +47,83 @@ program mld_df_sample ! input parameters - character(len=40) :: kmethd, mtrx_file, rhs_file + + character(len=40) :: kmethd, mtrx_file, rhs_file, guess_file, sol_file character(len=2) :: filefmt + + ! Krylov solver data + type solverdata + character(len=40) :: kmethd ! Krylov solver + integer(psb_ipk_) :: istopc ! stopping criterion + integer(psb_ipk_) :: itmax ! maximum number of iterations + integer(psb_ipk_) :: itrace ! tracing + integer(psb_ipk_) :: irst ! restart + real(psb_dpk_) :: eps ! stopping tolerance + end type solverdata + type(solverdata) :: s_choice + + ! preconditioner data type precdata - character(len=20) :: descr ! verbose description of the prec - character(len=10) :: prec ! overall prectype + + ! preconditioner type + character(len=40) :: descr ! verbose description of the prec + character(len=10) :: ptype ! preconditioner type + + ! general AMG data + character(len=16) :: mltype ! AMG cycle type + integer(psb_ipk_) :: otr_sweeps ! number of AMG cycles + integer(psb_ipk_) :: maxlevs ! maximum number of levels in AMG preconditioner + + ! AMG aggregation + character(len=16) :: aggrkind ! aggregation type: SMOOTHED, NONSMOOTHED + character(len=16) :: aggr_alg ! parallel aggregation algorithm: DEC, SYMDEC + character(len=16) :: aggr_ord ! ordering for aggregation: NATURAL, DEGREE + character(len=16) :: aggr_filter ! filtering: FILTER, NO_FILTER + real(psb_dpk_) :: mnaggratio ! minimum aggregation ratio + real(psb_dpk_), allocatable :: athresv(:) ! smoothed aggregation threshold vector + integer(psb_ipk_) :: thrvsz ! size of threshold vector + real(psb_dpk_) :: athres ! smoothed aggregation threshold + real(psb_dpk_) :: ascale ! smoothed aggregation scale factor for threshold + character(len=16) :: aggr_omalg ! algorithm for estimating omega parameter + character(len=16) :: aggr_eig ! Eigenvalue estimation procedure + real(psb_dpk_) :: omega_val ! Eigenvalue estimate value + integer(psb_ipk_) :: csize ! minimum size of coarsest matrix + + ! AMG smoother or pre-smoother; also 1-lev preconditioner + character(len=16) :: smther ! (pre-)smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps ! (pre-)smoother / 1-lev prec. sweeps integer(psb_ipk_) :: novr ! number of overlap layers - integer(psb_ipk_) :: jsweeps ! Jacobi/smoother sweeps character(len=16) :: restr ! restriction over application of AS character(len=16) :: prol ! prolongation over application of AS - character(len=16) :: solve ! factorization type: ILU, SuperLU, UMFPACK - integer(psb_ipk_) :: fill ! fillin for factorization - real(psb_dpk_) :: thr ! threshold for fact. ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: nlev ! number of levels in multilevel prec. - character(len=16) :: aggrkind ! smoothed, raw aggregation - character(len=16) :: aggr_alg ! aggregation algorithm (currently only decoupled) - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: mltype ! additive or multiplicative multi-level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - character(len=16) :: cmat ! coarse mat: distributed, replicated - character(len=16) :: csolve ! coarse solver: bjac, umf, slu, sludist - character(len=16) :: csbsolve ! coarse subsolver: ILU, ILU(T), SuperLU, UMFPACK - integer(psb_ipk_) :: cfill ! fillin for coarse factorization - real(psb_dpk_) :: cthres ! threshold for coarse fact. ILU(T) - integer(psb_ipk_) :: cjswp ! block-Jacobi sweeps - real(psb_dpk_) :: athres ! smoothed aggregation threshold - real(psb_dpk_) :: ascale ! smoothed aggregation scale factor + character(len=16) :: solve ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill ! fill-in for incomplete LU factorization + real(psb_dpk_) :: thr ! threshold for ILUT factorization + + ! AMG post-smoother; ignored by 1-lev preconditioner + character(len=16) :: smther2 ! post-smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps2 ! post-smoother sweeps + integer(psb_ipk_) :: novr2 ! number of overlap layers + character(len=16) :: restr2 ! restriction over application of AS + character(len=16) :: prol2 ! prolongation over application of AS + character(len=16) :: solve2 ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill2 ! fill-in for incomplete LU factorization + real(psb_dpk_) :: thr2 ! threshold for ILUT factorization + + ! coarsest-level solver + character(len=16) :: cmat ! coarsest matrix layout: REPL, DIST + character(len=16) :: csolve ! coarsest-lev solver: BJAC, SLUDIST (distr. + ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU + ! (repl. mat.) + character(len=16) :: csbsolve ! coarsest-lev local subsolver: ILU, ILUT, + ! MILU, UMF, MUMPS, SLU + integer(psb_ipk_) :: cfill ! fill-in for incomplete LU factorization + real(psb_dpk_) :: cthres ! threshold for ILUT factorization + integer(psb_ipk_) :: cjswp ! sweeps for GS or JAC coarsest-lev subsolver + end type precdata - type(precdata) :: prec_choice + type(precdata) :: p_choice ! sparse matrices type(psb_dspmat_type) :: a, aux_a @@ -83,10 +131,10 @@ program mld_df_sample ! preconditioner data type(mld_dprec_type) :: prec ! dense matrices - real(psb_dpk_), allocatable, target :: aux_b(:,:), d(:) + real(psb_dpk_), allocatable, target :: aux_b(:,:), d(:), aux_g(:,:), aux_x(:,:) real(psb_dpk_), allocatable , save :: x_col_glob(:), r_col_glob(:) - real(psb_dpk_), pointer :: b_col_glob(:) - type(psb_d_vect_type) :: b_col, x_col, r_col + real(psb_dpk_), pointer :: b_col_glob(:), ref_col_glob(:), guess_col_glob(:) + type(psb_d_vect_type) :: b_col, x_col, r_col, ref_col ! communications data structure type(psb_desc_type):: desc_a @@ -94,24 +142,23 @@ program mld_df_sample integer(psb_ipk_) :: ictxt, iam, np ! solver paramters - integer(psb_ipk_) :: iter, itmax, ierr, itrace, ircode, ipart,& - & methd, istopc, irst, nlv + integer(psb_ipk_) :: iter, ircode, ipart, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_dpk_) :: err, eps + real(psb_dpk_) :: err character(len=5) :: afmt character(len=20) :: name, renum integer(psb_ipk_), parameter :: iunit=12 integer(psb_ipk_) :: iparm(20) - character(len=40) :: fprefix ! other variables - integer(psb_ipk_) :: i,info,j,m_problem - integer(psb_ipk_) :: internal, m,ii,nnzero + integer(psb_ipk_) :: i, info, j, k, m_problem + integer(psb_ipk_) :: lbw, ubw, prf real(psb_dpk_) :: t1, t2, tprec, thier, tslv - real(psb_dpk_) :: r_amax, b_amax, scale,resmx,resmxp - integer(psb_ipk_) :: nrhs, nrow, n_row, dim, nv, ne - integer(psb_ipk_), allocatable :: ivg(:), ipv(:) + real(psb_dpk_) :: resmx, resmxp, xdiffn2, xdiffni, xni, xn2 + integer(psb_ipk_) :: nrhs, nv + integer(psb_ipk_), allocatable :: ivg(:), ipv(:), perm(:) + logical :: have_guess=.false., have_ref=.false. call psb_init(ictxt) call psb_info(ictxt,iam,np) @@ -131,41 +178,55 @@ program mld_df_sample ! Hello world ! if (iam == psb_root_) then - write(*,*) 'Welcome to MLD2P4 version: ',mld_version_string_ - write(*,*) 'This is the ',trim(name),' sample program' + write(psb_out_unit,*) ' ' + write(psb_out_unit,*) 'Welcome to MLD2P4 version: ',mld_version_string_ + write(psb_out_unit,*) 'This is the ',trim(name),' sample test program' + write(psb_out_unit,*) ' ' end if ! - ! get parameters + ! get parameters ! - call get_parms(ictxt,mtrx_file,rhs_file,filefmt,kmethd,& - & prec_choice,ipart,afmt,istopc,itmax,itrace,irst,eps) + call get_parms(ictxt,mtrx_file,rhs_file,guess_file,sol_file,filefmt, & + & ipart,afmt,s_choice,p_choice) call psb_barrier(ictxt) t1 = psb_wtime() - ! read the input matrix to be processed and (possibly) the rhs + ! read the input matrix to be processed and (possibly) the rhs, + ! the initial guess and the reference solution nrhs = 1 if (iam == psb_root_) then - select case(psb_toupper(filefmt)) + select case(psb_toupper(filefmt)) + case('MM') ! For Matrix Market we have an input file for the matrix - ! and an (optional) second file for the RHS. + ! and (optional) separate files for the rhs, the initial guess + ! and the reference solution call mm_mat_read(aux_a,info,iunit=iunit,filename=mtrx_file) - if (info == psb_success_) then - if (rhs_file /= 'NONE') then - call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) - end if + if ((info == psb_success_).and.(rhs_file /= 'NONE')) & + & call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) + if ((info == psb_success_).and.(guess_file /= 'NONE')) then + call mm_array_read(aux_g,info,iunit=iunit,filename=guess_file) + have_guess = .true. + end if + if ((info == psb_success_).and.(sol_file /= 'NONE')) then + call mm_array_read(aux_x,info,iunit=iunit,filename=sol_file) + have_ref = .true. end if case ('HB') ! For Harwell-Boeing we have a single file which may or may not - ! contain an RHS. - call hb_read(aux_a,info,iunit=iunit,b=aux_b,filename=mtrx_file) + ! contain rhs, initial guess and reference solution. + call hb_read(aux_a,info,iunit=iunit,b=aux_b,& + & g=aux_g,x=aux_x,filename=mtrx_file) + have_guess = allocated(aux_g) + have_ref = allocated(aux_x) case default info = -1 write(psb_err_unit,*) 'Wrong choice for fileformat ', filefmt end select + if (info /= psb_success_) then write(psb_err_unit,*) 'Error while reading input matrix ' call psb_abort(ictxt) @@ -173,6 +234,8 @@ program mld_df_sample m_problem = aux_a%get_nrows() call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) ! At this point aux_b may still be unallocated if (psb_size(aux_b,dim=ione) == m_problem) then @@ -181,24 +244,82 @@ program mld_df_sample b_col_glob =>aux_b(:,1) else write(psb_out_unit,'("Generating an rhs...")') - write(psb_out_unit,'(" ")') call psb_realloc(m_problem,1,aux_b,ircode) if (ircode /= 0) then call psb_errpush(psb_err_alloc_dealloc_,name) goto 9999 endif - b_col_glob => aux_b(:,1) do i=1, m_problem b_col_glob(i) = 1.d0 enddo endif + + if ((have_guess).and.(psb_size(aux_g,dim=ione) == m_problem)) then + ! if any initial guess were present, broadcast the first one + write(psb_err_unit,'("Ok, got an initial guess ")') + guess_col_glob =>aux_g(:,1) + else + write(psb_out_unit,'("Generating an initial guess...")') + call psb_realloc(m_problem,1,aux_g,ircode) + if (ircode /= 0) then + call psb_errpush(psb_err_alloc_dealloc_,name) + goto 9999 + endif + guess_col_glob => aux_g(:,1) + do i=1, m_problem + guess_col_glob(i) = 0.d0 + enddo + endif + + if ((have_ref).and.(psb_size(aux_x,dim=ione) == m_problem)) then + ! if any reference were present, broadcast the first one + write(psb_err_unit,'("Ok, got a reference solution ")') + ref_col_glob =>aux_x(:,1) + else + write(psb_out_unit,'("No reference solution...")') + endif + + ! clean zeros in the input matrix + call aux_a%clean_zeros(info) + else call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) + end if + + + ! + ! Renumbering (NONE for the moment) + ! + if (iam==psb_root_) then + renum='NONE' + + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,'(" ")') + write(psb_out_unit,*) 'Bandwidth and profile: ',lbw,ubw,prf + write(psb_out_unit,*) 'Renumbering algorithm: ',psb_toupper(renum) + + if (trim(psb_toupper(renum))/='NONE') then + call psb_mat_renum(renum,aux_a,info,perm=perm) + if (info /= 0) then + write(psb_err_unit,*) 'Error from RENUM',info + goto 9999 + end if + call psb_gelp('N',perm(1:m_problem),b_col_glob(1:m_problem),info) + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,*) 'Bandwidth and profile (renumberd):',lbw,ubw,prf + end if + + write(psb_out_unit,'(" ")') + end if + ! ! switch over different partition types - if (ipart == 0) then + ! + if (ipart == 0) then call psb_barrier(ictxt) if (iam == psb_root_) write(psb_out_unit,'("Partition type: block")') allocate(ivg(m_problem),ipv(np)) @@ -211,10 +332,8 @@ program mld_df_sample if (iam == psb_root_) then write(psb_out_unit,'("Partition type: graph")') write(psb_out_unit,'(" ")') - ! write(psb_err_unit,'("Build type: graph")') call build_mtpart(aux_a,np) endif -!!$ call psb_barrier(ictxt) call distr_mtpart(psb_root_,ictxt) call getv_mtpart(ivg) call psb_matdist(aux_a, a, ictxt,desc_a,info,fmt=afmt,v=ivg) @@ -223,89 +342,136 @@ program mld_df_sample call psb_matdist(aux_a, a, ictxt, desc_a,info,fmt=afmt,parts=part_block) end if - call psb_scatter(b_col_glob,b_col,desc_a,info,root=psb_root_) + ! + ! Scatter rhs, initial guess and reference solution + ! + call psb_geall(b_col,desc_a,info) call psb_geall(x_col,desc_a,info) - call x_col%zero() - call psb_geasb(x_col,desc_a,info) - call psb_geall(r_col,desc_a,info) - call r_col%zero() - call psb_geasb(r_col,desc_a,info) - t2 = psb_wtime() - t1 - + if (have_ref) call psb_geall(ref_col,desc_a,info) + + if (iam == psb_root_) write(psb_out_unit,'("Scatter rhs")') + call psb_scatter(b_col_glob,b_col,desc_a,info) + if (iam == psb_root_) write(psb_out_unit,'("Scatter initial guess")') + call psb_scatter(guess_col_glob,x_col,desc_a,info) + if (have_ref) then + if (iam == psb_root_) write(psb_out_unit,'("Scatter reference solution")') + call psb_scatter(ref_col_glob,ref_col,desc_a,info) + end if + t2 = psb_wtime() - t1 call psb_amx(ictxt, t2) if (iam == psb_root_) then write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Time to read and partition matrix : ",es12.5)')t2 + write(psb_out_unit,'("Time to read and partition matrix, rhs(, guess, ref sol) : ",es12.5)')t2 write(psb_out_unit,'(" ")') - write(psb_out_unit,*) 'Preconditioner: ',prec_choice%descr end if - ! - - if (psb_toupper(prec_choice%prec) == 'ML') then - call mld_precinit(prec,prec_choice%prec, info) - if (prec_choice%nlev > 0) & - & call mld_precset(prec,'max_prec_levs', prec_choice%nlev, info) - call mld_precset(prec,'aggr_kind', prec_choice%aggrkind,info) - call mld_precset(prec,'aggr_alg', prec_choice%aggr_alg,info) - call mld_precset(prec,'aggr_ord', prec_choice%aggr_ord,info) - call mld_precset(prec,'aggr_scale', prec_choice%ascale, info) - call mld_precset(prec,'aggr_thresh', prec_choice%athres, info) - call mld_precset(prec,'smoother_type', prec_choice%smther, info) - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) - call mld_precset(prec,'ml_type', prec_choice%mltype, info) - call mld_precset(prec,'smoother_pos', prec_choice%smthpos, info) - call mld_precset(prec,'coarse_solve', prec_choice%csolve, info) - call mld_precset(prec,'coarse_subsolve', prec_choice%csbsolve,info) - call mld_precset(prec,'coarse_mat', prec_choice%cmat, info) - call mld_precset(prec,'coarse_fillin', prec_choice%cfill, info) - call mld_precset(prec,'coarse_iluthrs', prec_choice%cthres, info) - call mld_precset(prec,'coarse_sweeps', prec_choice%cjswp, info) - ! building the preconditioner + ! + ! initialize the preconditioner + ! + if (psb_toupper(p_choice%ptype) == 'ML') then + ! multilevel preconditioner + call prec%init(p_choice%ptype,info) + + call prec%set('ml_type', p_choice%mltype, info) + call prec%set('outer_sweeps', p_choice%otr_sweeps,info) + if (p_choice%csize>0)& + & call prec%set('coarse_aggr_size', p_choice%csize, info) + if (p_choice%mnaggratio>0)& + & call prec%set('min_aggr_ratio', p_choice%mnaggratio, info) + if (p_choice%maxlevs>0)& + & call prec%set('max_prec_levs', p_choice%maxlevs, info) + if (p_choice%ascale > dzero) & + & call prec%set('aggr_scale', p_choice%ascale, info) + if (p_choice%athres >= dzero) & + & call prec%set('aggr_thresh', p_choice%athres, info) + if (p_choice%thrvsz>0) then + do k=1,min(p_choice%thrvsz,size(prec%precv)-1) + call prec%set('aggr_thresh', p_choice%athresv(k), info,ilev=(k+1)) + end do + end if + + call prec%set('aggr_kind', p_choice%aggrkind, info) + call prec%set('aggr_alg', p_choice%aggr_alg, info) + call prec%set('aggr_ord', p_choice%aggr_ord, info) + call prec%set('aggr_filter', p_choice%aggr_filter,info) + call prec%set('aggr_omega_alg', p_choice%aggr_omalg, info) + if (psb_toupper(p_choice%aggr_omalg) == 'EIG_EST') then + call prec%set('aggr_eig', p_choice%aggr_eig, info) + else if (psb_toupper(p_choice%aggr_omalg) == 'USER_CHOICE') then + call prec%set('aggr_omega_val', p_choice%omega_val, info) + end if + call prec%set('coarse_solve', p_choice%csolve, info) + if (psb_toupper(p_choice%csolve) == 'BJAC') & + & call prec%set('coarse_subsolve', p_choice%csbsolve, info) + call prec%set('coarse_mat', p_choice%cmat, info) + call prec%set('coarse_fillin', p_choice%cfill, info) + call prec%set('coarse_iluthrs', p_choice%cthres, info) + call prec%set('coarse_sweeps', p_choice%cjswp, info) + + + call prec%set('smoother_type', p_choice%smther, info) + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + + if (psb_toupper(p_choice%smther2) /= 'NONE') then + call prec%set('smoother_type', p_choice%smther2, info,pos='post') + call prec%set('smoother_sweeps', p_choice%jsweeps2, info,pos='post') + call prec%set('sub_ovr', p_choice%novr2, info,pos='post') + call prec%set('sub_restr', p_choice%restr2, info,pos='post') + call prec%set('sub_prol', p_choice%prol2, info,pos='post') + call prec%set('sub_solve', p_choice%solve2, info,pos='post') + call prec%set('sub_fillin', p_choice%fill2, info,pos='post') + call prec%set('sub_iluthrs', p_choice%thr2, info,pos='post') + end if + + ! build the preconditioner call psb_barrier(ictxt) t1 = psb_wtime() - call mld_hierarchy_bld(a,desc_a,prec,info) + call prec%hierarchy_build(a,desc_a,info) thier = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if call psb_barrier(ictxt) t1 = psb_wtime() - call mld_smoothers_bld(a,desc_a,prec,info) + call prec%smoothers_build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if else + ! 1-level preconditioner nlv = 1 - call mld_precinit(prec,prec_choice%prec,info) - if (psb_toupper(prec_choice%prec) /= 'NONE') then - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) + call prec%init(p_choice%ptype,info) + + if (psb_toupper(p_choice%ptype) /= 'NONE') then + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + !!! call prec%set('solver_sweeps', p_choice%svsweeps, info) end if - ! building the preconditioner + + ! build the preconditioner thier = dzero t1 = psb_wtime() - call mld_precbld(a,desc_a,prec,info) + call prec%build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if end if @@ -314,6 +480,8 @@ program mld_df_sample call psb_amx(ictxt, tprec) if(iam == psb_root_) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'("Preconditioner: ",a)') trim(p_choice%descr) write(psb_out_unit,'("Preconditioner time: ",es12.5)')thier+tprec write(psb_out_unit,'(" ")') end if @@ -321,43 +489,71 @@ program mld_df_sample iparm = 0 call psb_barrier(ictxt) t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b_col,x_col,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + call psb_krylov(s_choice%kmethd,a,prec,b_col,x_col,s_choice%eps,& + & desc_a,info,itmax=s_choice%itmax,iter=iter,err=err,itrace=s_choice%itrace,& + & istop=s_choice%istopc,irst=s_choice%irst) call psb_barrier(ictxt) tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + + ! compute residual norms + call psb_geall(r_col,desc_a,info) + call r_col%zero() + call psb_geasb(r_col,desc_a,info) call psb_geaxpby(done,b_col,dzero,r_col,desc_a,info) call psb_spmm(-done,a,x_col,done,r_col,desc_a,info) resmx = psb_genrm2(r_col,desc_a,info) resmxp = psb_geamax(r_col,desc_a,info) - + + ! compute error in solution + if (have_ref) then + call psb_geaxpby(-done,x_col,done,ref_col,desc_a,info) + xdiffn2 = psb_genrm2(ref_col,desc_a,info) + xdiffni = psb_geamax(ref_col,desc_a,info) + xn2 = psb_genrm2(ref_col,desc_a,info) + xni = psb_geamax(ref_col,desc_a,info) + end if + amatsize = a%sizeof() descsize = desc_a%sizeof() precsize = prec%sizeof() call psb_sum(ictxt,amatsize) call psb_sum(ictxt,descsize) call psb_sum(ictxt,precsize) + call prec%descr(info) if (iam == psb_root_) then - call mld_precdescr(prec,info) write(psb_out_unit,'("Matrix: ",a)')mtrx_file write(psb_out_unit,'("Computed solution on ",i8," processors")')np - write(psb_out_unit,'("Iterations to convergence : ",i6)')iter - write(psb_out_unit,'("Error estimate on exit : ",es12.5)')err + write(psb_out_unit,'("Krylov method : ",a)') trim(s_choice%kmethd) + write(psb_out_unit,'("Preconditioner : ",a)') trim(p_choice%descr) + write(psb_out_unit,'("Iterations to convergence : ",i12)')iter + write(psb_out_unit,'("Relative error estimate on exit : ",es12.5)') err write(psb_out_unit,'("Number of levels in hierarchy : ",i12)') prec%get_nlevs() write(psb_out_unit,'("Time to build hierarchy : ",es12.5)')thier write(psb_out_unit,'("Time to build smoothers : ",es12.5)')tprec write(psb_out_unit,'("Total time for preconditioner : ",es12.5)')tprec+thier write(psb_out_unit,'("Time to solve system : ",es12.5)')tslv - write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/(iter) - write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec - write(psb_out_unit,'("Residual norm 2 : ",es12.5)')resmx - write(psb_out_unit,'("Residual norm inf : ",es12.5)')resmxp + write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/iter + write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec+thier + write(psb_out_unit,'("Residual 2-norm : ",es12.5)')resmx + write(psb_out_unit,'("Residual inf-norm : ",es12.5)')resmxp write(psb_out_unit,'("Total memory occupation for A : ",i12)')amatsize write(psb_out_unit,'("Total memory occupation for DESC_A : ",i12)')descsize write(psb_out_unit,'("Total memory occupation for PREC : ",i12)')precsize write(psb_out_unit,'("Storage format for A : ",a )')a%get_fmt() write(psb_out_unit,'("Storage format for DESC_A : ",a )')desc_a%get_fmt() + if (have_ref) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,5x,a8)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(2-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffn2,xn2,xdiffn2/xn2 + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,4x,a10)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(inf-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffni,xni,xdiffni/xni + end if + end if call psb_gather(x_col_glob,x_col,desc_a,info,root=psb_root_) @@ -367,13 +563,15 @@ program mld_df_sample if (iam == psb_root_) then write(psb_err_unit,'(" ")') write(psb_err_unit,'("Saving x on file")') - write(20,*) 'matrix: ',mtrx_file - write(20,*) 'computed solution on ',np,' processors.' - write(20,*) 'iterations to convergence: ',iter - write(20,*) 'error estimate (infinity norm) on exit:', & - & ' ||r||/(||a||||x||+||b||) = ',err - write(20,'("Residual norm 2 : ",es12.5)')resmx - write(20,'("Residual norm inf : ",es12.5)')resmxp + write(20,*) 'Matrix: ',mtrx_file + write(20,*) 'Krylov method:',trim(s_choice%kmethd) + write(20,*) 'Preconditioner:',trim(p_choice%descr) + write(20,*) 'Computed solution on ',np,' processors.' + write(20,*) 'Iterations to convergence: ',iter + write(20,*) 'Error estimate (infinity norm) on exit:', & + & ' ||r||/||b|| (inf-norm) = ',err + write(20,'(" Residual 2-norm 2 : ",es12.5)')resmx + write(20,'(" Residual inf-norm : ",es12.5)')resmxp write(20,'(a8,4(2x,a20))') 'I','X(I)','R(I)','B(I)' do i=1,m_problem write(20,998) i,x_col_glob(i),r_col_glob(i),b_col_glob(i) @@ -383,10 +581,12 @@ program mld_df_sample 993 format(i6,4(1x,e12.6)) - call psb_gefree(b_col, desc_a,info) - call psb_gefree(x_col, desc_a,info) + call psb_gefree(b_col,desc_a,info) + call psb_gefree(x_col,desc_a,info) + call psb_gefree(r_col,desc_a,info) + call psb_gefree(ref_col,desc_a,info) call psb_spfree(a, desc_a,info) - call mld_precfree(prec,info) + call prec%free(info) call psb_cdfree(desc_a,info) call psb_exit(ictxt) @@ -399,101 +599,161 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(icontxt,mtrx,rhs,filefmt,kmethd,& - & prec, ipart,afmt,istopc,itmax,itrace,irst,eps) + subroutine get_parms(icontxt,mtrx,rhs,guess,sol,filefmt,ipart,afmt,solve,prec) use psb_base_mod implicit none - integer(psb_ipk_) :: icontxt - character(len=*) :: kmethd, mtrx, rhs, afmt,filefmt + integer(psb_ipk_) :: icontxt, ipart + character(len=*) :: mtrx, rhs, guess, sol, filefmt, afmt + type(solverdata) :: solve type(precdata) :: prec - real(psb_dpk_) :: eps - integer(psb_ipk_) :: iret, istopc,itmax,itrace, ipart, irst - integer(psb_ipk_) :: iam, nm, np, i + integer(psb_ipk_) :: iam, nm, np call psb_info(icontxt,iam,np) if (iam == psb_root_) then - ! read input parameters - call read_data(mtrx,psb_inp_unit) - call read_data(rhs,psb_inp_unit) - call read_data(filefmt,psb_inp_unit) - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(ipart,psb_inp_unit) - call read_data(istopc,psb_inp_unit) - call read_data(itmax,psb_inp_unit) - call read_data(itrace,psb_inp_unit) - call read_data(irst,psb_inp_unit) - call read_data(eps,psb_inp_unit) + ! read input data + ! + ! input files + call read_data(mtrx,psb_inp_unit) ! matrix file + call read_data(rhs,psb_inp_unit) ! rhs file + call read_data(guess,psb_inp_unit) ! starting guess file + call read_data(sol,psb_inp_unit) ! solution file (for comparison) + call read_data(filefmt,psb_inp_unit) ! format of files + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(ipart,psb_inp_unit) ! partition type + ! Krylov solver data + call read_data(solve%kmethd,psb_inp_unit) ! Krylov solver + call read_data(solve%istopc,psb_inp_unit) ! stopping criterion + call read_data(solve%itmax,psb_inp_unit) ! max num iterations + call read_data(solve%itrace,psb_inp_unit) ! tracing + call read_data(solve%irst,psb_inp_unit) ! restart + call read_data(solve%eps,psb_inp_unit) ! tolerance + ! preconditioner type call read_data(prec%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prec%prec,psb_inp_unit) ! overall prectype - call read_data(prec%novr,psb_inp_unit) ! number of overlap layers - call read_data(prec%restr,psb_inp_unit) ! restriction over application of as - call read_data(prec%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prec%solve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%fill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%thr,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%jsweeps,psb_inp_unit) ! Jacobi sweeps for PJAC - if (psb_toupper(prec%prec) == 'ML') then - call read_data(prec%nlev,psb_inp_unit) ! Number of levels in multilevel prec. - call read_data(prec%smther,psb_inp_unit) ! Smoother type. - call read_data(prec%aggrkind,psb_inp_unit) ! smoothed/raw aggregatin - call read_data(prec%aggr_alg,psb_inp_unit) ! local or global aggregation - call read_data(prec%aggr_ord,psb_inp_unit) ! Ordering for aggregation - call read_data(prec%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prec%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prec%cmat,psb_inp_unit) ! coarse mat - call read_data(prec%csolve,psb_inp_unit) ! Factorization type: BJAC, SuperLU, UMFPACK. - call read_data(prec%csbsolve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%cfill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%cthres,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%cjswp,psb_inp_unit) ! Jacobi sweeps - call read_data(prec%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prec%ascale,psb_inp_unit) ! smoother aggr thresh + call read_data(prec%ptype,psb_inp_unit) ! preconditioner type + ! general AMG data + call read_data(prec%mltype,psb_inp_unit) ! AMG cycle type + call read_data(prec%otr_sweeps,psb_inp_unit) ! number of AMG cycles + call read_data(prec%maxlevs,psb_inp_unit) ! max number of levels in AMG prec + call read_data(prec%csize,psb_inp_unit) ! min size coarsest mat + ! aggregation + call read_data(prec%aggrkind,psb_inp_unit) ! aggregation type + call read_data(prec%aggr_alg,psb_inp_unit) ! parallel aggregation alg + call read_data(prec%aggr_ord,psb_inp_unit) ! ordering for aggregation + call read_data(prec%aggr_filter,psb_inp_unit) ! filtering + call read_data(prec%mnaggratio,psb_inp_unit) ! minimum aggregation ratio + call read_data(prec%thrvsz,psb_inp_unit) ! size of aggr thresh vector + if (prec%thrvsz > 0) then + call psb_realloc(prec%thrvsz,prec%athresv,info) + call read_data(prec%athresv,psb_inp_unit) ! aggr thresh vector + else + read(psb_inp_unit,*) ! dummy read to skip a record end if + call read_data(prec%athres,psb_inp_unit) ! smoothed aggr thresh + call read_data(prec%aggr_omalg,psb_inp_unit) ! alg for estimating omega + call read_data(prec%aggr_eig,psb_inp_unit) ! alg for estimating omega + call read_data(prec%omega_val,psb_inp_unit) ! alg for estimating omega + ! AMG smoother (or pre-smoother) / 1-lev preconditioner + call read_data(prec%smther,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps,psb_inp_unit) ! (pre-)smoother / 1-lev prec sweeps + call read_data(prec%novr,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve,psb_inp_unit) ! local subsolver + call read_data(prec%fill,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr,psb_inp_unit) ! threshold for ILUT + ! AMG post-smoother + call read_data(prec%smther2,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps2,psb_inp_unit) ! (post-)smoother sweeps + call read_data(prec%novr2,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr2,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol2,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve2,psb_inp_unit) ! local subsolver + call read_data(prec%fill2,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr2,psb_inp_unit) ! threshold for ILUT + ! coasest-level solver + call read_data(prec%csolve,psb_inp_unit) ! coarsest-lev solver + call read_data(prec%cmat,psb_inp_unit) ! coarsest mat layout + call read_data(prec%csbsolve,psb_inp_unit) ! coarsest-lev subsolver + call read_data(prec%cfill,psb_inp_unit) ! fill-in for incompl LU + call read_data(prec%cthres,psb_inp_unit) ! Threshold for ILUT + call read_data(prec%cjswp,psb_inp_unit) ! sweeps for GS/JAC subsolver end if call psb_bcast(icontxt,mtrx) call psb_bcast(icontxt,rhs) + call psb_bcast(icontxt,guess) + call psb_bcast(icontxt,sol) call psb_bcast(icontxt,filefmt) - call psb_bcast(icontxt,kmethd) call psb_bcast(icontxt,afmt) call psb_bcast(icontxt,ipart) - call psb_bcast(icontxt,istopc) - call psb_bcast(icontxt,itmax) - call psb_bcast(icontxt,itrace) - call psb_bcast(icontxt,irst) - call psb_bcast(icontxt,eps) - call psb_bcast(icontxt,prec%descr) ! verbose description of the prec - call psb_bcast(icontxt,prec%prec) ! overall prectype - call psb_bcast(icontxt,prec%novr) ! number of overlap layers - call psb_bcast(icontxt,prec%restr) ! restriction over application of as - call psb_bcast(icontxt,prec%prol) ! prolongation over application of as - call psb_bcast(icontxt,prec%solve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%fill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%thr) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%jsweeps) ! Jacobi sweeps - if (psb_toupper(prec%prec) == 'ML') then - call psb_bcast(icontxt,prec%smther) ! Smoother type. - call psb_bcast(icontxt,prec%nlev) ! Number of levels in multilevel prec. - call psb_bcast(icontxt,prec%aggrkind) ! smoothed/raw aggregatin - call psb_bcast(icontxt,prec%aggr_alg) ! local or global aggregation - call psb_bcast(icontxt,prec%aggr_ord) ! Ordering for aggregation - call psb_bcast(icontxt,prec%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(icontxt,prec%smthpos) ! side: pre, post, both smoothing - call psb_bcast(icontxt,prec%cmat) ! coarse mat - call psb_bcast(icontxt,prec%csolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%csbsolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%cfill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%cthres) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%cjswp) ! Jacobi sweeps - call psb_bcast(icontxt,prec%athres) ! smoother aggr thresh - call psb_bcast(icontxt,prec%ascale) ! smoother aggr scale factor + + call psb_bcast(icontxt,solve%kmethd) + call psb_bcast(icontxt,solve%istopc) + call psb_bcast(icontxt,solve%itmax) + call psb_bcast(icontxt,solve%itrace) + call psb_bcast(icontxt,solve%irst) + call psb_bcast(icontxt,solve%eps) + + call psb_bcast(icontxt,prec%descr) + call psb_bcast(icontxt,prec%ptype) + + ! broadcast first (pre-)smoother / 1-lev prec data + call psb_bcast(icontxt,prec%smther) ! actually not needed for 1-lev precs + call psb_bcast(icontxt,prec%jsweeps) + call psb_bcast(icontxt,prec%novr) + call psb_bcast(icontxt,prec%restr) + call psb_bcast(icontxt,prec%prol) + call psb_bcast(icontxt,prec%solve) + call psb_bcast(icontxt,prec%fill) + call psb_bcast(icontxt,prec%thr) + + ! broadcast (other) AMG parameters + if (psb_toupper(prec%ptype) == 'ML') then + + call psb_bcast(icontxt,prec%mltype) + call psb_bcast(icontxt,prec%otr_sweeps) + call psb_bcast(icontxt,prec%maxlevs) + + call psb_bcast(icontxt,prec%smther2) + call psb_bcast(icontxt,prec%jsweeps2) + call psb_bcast(icontxt,prec%novr2) + call psb_bcast(icontxt,prec%restr2) + call psb_bcast(icontxt,prec%prol2) + call psb_bcast(icontxt,prec%solve2) + call psb_bcast(icontxt,prec%fill2) + call psb_bcast(icontxt,prec%thr2) + + call psb_bcast(icontxt,prec%aggrkind) + call psb_bcast(icontxt,prec%aggr_alg) + call psb_bcast(icontxt,prec%aggr_ord) + call psb_bcast(icontxt,prec%aggr_filter) + call psb_bcast(icontxt,prec%mnaggratio) + call psb_bcast(ictxt,prec%thrvsz) + if (prec%thrvsz > 0) then + if (iam /= psb_root_) call psb_realloc(prec%thrvsz,prec%athresv,info) + call psb_bcast(ictxt,prec%athresv) + end if + call psb_bcast(ictxt,prec%athres) + call psb_bcast(ictxt,prec%ascale) + call psb_bcast(ictxt,prec%aggr_omalg) + call psb_bcast(ictxt,prec%aggr_eig) + call psb_bcast(ictxt,prec%omega_val) + + call psb_bcast(icontxt,prec%csize) + call psb_bcast(icontxt,prec%cmat) + call psb_bcast(icontxt,prec%csolve) + call psb_bcast(icontxt,prec%csbsolve) + call psb_bcast(icontxt,prec%cfill) + call psb_bcast(icontxt,prec%cthres) + call psb_bcast(icontxt,prec%cjswp) + end if end subroutine get_parms + subroutine pr_usage(iout) integer(psb_ipk_) iout write(iout, *) ' number of parameters is incorrect!' @@ -511,4 +771,5 @@ contains write(iout, *) ' ptype partition strategy default 0' write(iout, *) ' 0: block partition ' end subroutine pr_usage + end program mld_df_sample diff --git a/tests/fileread/mld_sf_sample.f90 b/tests/fileread/mld_sf_sample.f90 index 1e72a074..0571c201 100644 --- a/tests/fileread/mld_sf_sample.f90 +++ b/tests/fileread/mld_sf_sample.f90 @@ -4,7 +4,7 @@ ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) ! -! (C) Copyright 2008, 2010, 2012, 2015, 2017 , 2016 +! (C) Copyright 2008, 2010, 2012, 2015, 2017 ! ! Salvatore Filippone Cranfield University, UK ! Ambra Abdullahi Hassan University of Rome Tor Vergata, IT @@ -47,35 +47,83 @@ program mld_sf_sample ! input parameters - character(len=40) :: kmethd, mtrx_file, rhs_file + + character(len=40) :: kmethd, mtrx_file, rhs_file, guess_file, sol_file character(len=2) :: filefmt + + ! Krylov solver data + type solverdata + character(len=40) :: kmethd ! Krylov solver + integer(psb_ipk_) :: istopc ! stopping criterion + integer(psb_ipk_) :: itmax ! maximum number of iterations + integer(psb_ipk_) :: itrace ! tracing + integer(psb_ipk_) :: irst ! restart + real(psb_spk_) :: eps ! stopping tolerance + end type solverdata + type(solverdata) :: s_choice + + ! preconditioner data type precdata - character(len=20) :: descr ! verbose description of the prec - character(len=10) :: prec ! overall prectype + + ! preconditioner type + character(len=40) :: descr ! verbose description of the prec + character(len=10) :: ptype ! preconditioner type + + ! general AMG data + character(len=16) :: mltype ! AMG cycle type + integer(psb_ipk_) :: otr_sweeps ! number of AMG cycles + integer(psb_ipk_) :: maxlevs ! maximum number of levels in AMG preconditioner + + ! AMG aggregation + character(len=16) :: aggrkind ! aggregation type: SMOOTHED, NONSMOOTHED + character(len=16) :: aggr_alg ! parallel aggregation algorithm: DEC, SYMDEC + character(len=16) :: aggr_ord ! ordering for aggregation: NATURAL, DEGREE + character(len=16) :: aggr_filter ! filtering: FILTER, NO_FILTER + real(psb_spk_) :: mnaggratio ! minimum aggregation ratio + real(psb_spk_), allocatable :: athresv(:) ! smoothed aggregation threshold vector + integer(psb_ipk_) :: thrvsz ! size of threshold vector + real(psb_spk_) :: athres ! smoothed aggregation threshold + real(psb_spk_) :: ascale ! smoothed aggregation scale factor for threshold + character(len=16) :: aggr_omalg ! algorithm for estimating omega parameter + character(len=16) :: aggr_eig ! Eigenvalue estimation procedure + real(psb_spk_) :: omega_val ! Eigenvalue estimate value + integer(psb_ipk_) :: csize ! minimum size of coarsest matrix + + ! AMG smoother or pre-smoother; also 1-lev preconditioner + character(len=16) :: smther ! (pre-)smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps ! (pre-)smoother / 1-lev prec. sweeps integer(psb_ipk_) :: novr ! number of overlap layers - integer(psb_ipk_) :: jsweeps ! Jacobi/smoother sweeps character(len=16) :: restr ! restriction over application of AS character(len=16) :: prol ! prolongation over application of AS - character(len=16) :: solve ! factorization type: ILU, SuperLU, UMFPACK - integer(psb_ipk_) :: fill ! fillin for factorization - real(psb_spk_) :: thr ! threshold for fact. ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: nlev ! number of levels in multilevel prec. - character(len=16) :: aggrkind ! smoothed, raw aggregation - character(len=16) :: aggr_alg ! aggregation algorithm (currently only decoupled) - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: mltype ! additive or multiplicative multi-level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - character(len=16) :: cmat ! coarse mat: distributed, replicated - character(len=16) :: csolve ! coarse solver: bjac, umf, slu, sludist - character(len=16) :: csbsolve ! coarse subsolver: ILU, ILU(T), SuperLU, UMFPACK - integer(psb_ipk_) :: cfill ! fillin for coarse factorization - real(psb_spk_) :: cthres ! threshold for coarse fact. ILU(T) - integer(psb_ipk_) :: cjswp ! block-Jacobi sweeps - real(psb_spk_) :: athres ! smoothed aggregation threshold - real(psb_spk_) :: ascale ! smoothed aggregation scale factor + character(len=16) :: solve ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill ! fill-in for incomplete LU factorization + real(psb_spk_) :: thr ! threshold for ILUT factorization + + ! AMG post-smoother; ignored by 1-lev preconditioner + character(len=16) :: smther2 ! post-smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps2 ! post-smoother sweeps + integer(psb_ipk_) :: novr2 ! number of overlap layers + character(len=16) :: restr2 ! restriction over application of AS + character(len=16) :: prol2 ! prolongation over application of AS + character(len=16) :: solve2 ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill2 ! fill-in for incomplete LU factorization + real(psb_spk_) :: thr2 ! threshold for ILUT factorization + + ! coarsest-level solver + character(len=16) :: cmat ! coarsest matrix layout: REPL, DIST + character(len=16) :: csolve ! coarsest-lev solver: BJAC, SLUDIST (distr. + ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU + ! (repl. mat.) + character(len=16) :: csbsolve ! coarsest-lev local subsolver: ILU, ILUT, + ! MILU, UMF, MUMPS, SLU + integer(psb_ipk_) :: cfill ! fill-in for incomplete LU factorization + real(psb_spk_) :: cthres ! threshold for ILUT factorization + integer(psb_ipk_) :: cjswp ! sweeps for GS or JAC coarsest-lev subsolver + end type precdata - type(precdata) :: prec_choice + type(precdata) :: p_choice ! sparse matrices type(psb_sspmat_type) :: a, aux_a @@ -83,10 +131,10 @@ program mld_sf_sample ! preconditioner data type(mld_sprec_type) :: prec ! dense matrices - real(psb_spk_), allocatable, target :: aux_b(:,:), d(:) + real(psb_spk_), allocatable, target :: aux_b(:,:), d(:), aux_g(:,:), aux_x(:,:) real(psb_spk_), allocatable , save :: x_col_glob(:), r_col_glob(:) - real(psb_spk_), pointer :: b_col_glob(:) - type(psb_s_vect_type) :: b_col, x_col, r_col + real(psb_spk_), pointer :: b_col_glob(:), ref_col_glob(:), guess_col_glob(:) + type(psb_s_vect_type) :: b_col, x_col, r_col, ref_col ! communications data structure type(psb_desc_type):: desc_a @@ -94,24 +142,23 @@ program mld_sf_sample integer(psb_ipk_) :: ictxt, iam, np ! solver paramters - integer(psb_ipk_) :: iter, itmax, ierr, itrace, ircode, ipart,& - & methd, istopc, irst, nlv + integer(psb_ipk_) :: iter, ircode, ipart, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_spk_) :: err, eps + real(psb_spk_) :: err character(len=5) :: afmt character(len=20) :: name, renum integer(psb_ipk_), parameter :: iunit=12 integer(psb_ipk_) :: iparm(20) - character(len=40) :: fprefix ! other variables - integer(psb_ipk_) :: i,info,j,m_problem - integer(psb_ipk_) :: internal, m,ii,nnzero - real(psb_spk_) :: t1, t2, tprec, thier, tslv - real(psb_spk_) :: r_amax, b_amax, scale,resmx,resmxp - integer(psb_ipk_) :: nrhs, nrow, n_row, dim, nv, ne - integer(psb_ipk_), allocatable :: ivg(:), ipv(:) + integer(psb_ipk_) :: i, info, j, k, m_problem + integer(psb_ipk_) :: lbw, ubw, prf + real(psb_dpk_) :: t1, t2, tprec, thier, tslv + real(psb_spk_) :: resmx, resmxp, xdiffn2, xdiffni, xni, xn2 + integer(psb_ipk_) :: nrhs, nv + integer(psb_ipk_), allocatable :: ivg(:), ipv(:), perm(:) + logical :: have_guess=.false., have_ref=.false. call psb_init(ictxt) call psb_info(ictxt,iam,np) @@ -131,41 +178,55 @@ program mld_sf_sample ! Hello world ! if (iam == psb_root_) then - write(*,*) 'Welcome to MLD2P4 version: ',mld_version_string_ - write(*,*) 'This is the ',trim(name),' sample program' + write(psb_out_unit,*) ' ' + write(psb_out_unit,*) 'Welcome to MLD2P4 version: ',mld_version_string_ + write(psb_out_unit,*) 'This is the ',trim(name),' sample test program' + write(psb_out_unit,*) ' ' end if ! - ! get parameters + ! get parameters ! - call get_parms(ictxt,mtrx_file,rhs_file,filefmt,kmethd,& - & prec_choice,ipart,afmt,istopc,itmax,itrace,irst,eps) + call get_parms(ictxt,mtrx_file,rhs_file,guess_file,sol_file,filefmt, & + & ipart,afmt,s_choice,p_choice) call psb_barrier(ictxt) t1 = psb_wtime() - ! read the input matrix to be processed and (possibly) the rhs + ! read the input matrix to be processed and (possibly) the rhs, + ! the initial guess and the reference solution nrhs = 1 if (iam == psb_root_) then - select case(psb_toupper(filefmt)) + select case(psb_toupper(filefmt)) + case('MM') ! For Matrix Market we have an input file for the matrix - ! and an (optional) second file for the RHS. + ! and (optional) separate files for the rhs, the initial guess + ! and the reference solution call mm_mat_read(aux_a,info,iunit=iunit,filename=mtrx_file) - if (info == psb_success_) then - if (rhs_file /= 'NONE') then - call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) - end if + if ((info == psb_success_).and.(rhs_file /= 'NONE')) & + & call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) + if ((info == psb_success_).and.(guess_file /= 'NONE')) then + call mm_array_read(aux_g,info,iunit=iunit,filename=guess_file) + have_guess = .true. + end if + if ((info == psb_success_).and.(sol_file /= 'NONE')) then + call mm_array_read(aux_x,info,iunit=iunit,filename=sol_file) + have_ref = .true. end if case ('HB') ! For Harwell-Boeing we have a single file which may or may not - ! contain an RHS. - call hb_read(aux_a,info,iunit=iunit,b=aux_b,filename=mtrx_file) + ! contain rhs, initial guess and reference solution. + call hb_read(aux_a,info,iunit=iunit,b=aux_b,& + & g=aux_g,x=aux_x,filename=mtrx_file) + have_guess = allocated(aux_g) + have_ref = allocated(aux_x) case default info = -1 write(psb_err_unit,*) 'Wrong choice for fileformat ', filefmt end select + if (info /= psb_success_) then write(psb_err_unit,*) 'Error while reading input matrix ' call psb_abort(ictxt) @@ -173,6 +234,8 @@ program mld_sf_sample m_problem = aux_a%get_nrows() call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) ! At this point aux_b may still be unallocated if (psb_size(aux_b,dim=ione) == m_problem) then @@ -181,24 +244,82 @@ program mld_sf_sample b_col_glob =>aux_b(:,1) else write(psb_out_unit,'("Generating an rhs...")') - write(psb_out_unit,'(" ")') call psb_realloc(m_problem,1,aux_b,ircode) if (ircode /= 0) then call psb_errpush(psb_err_alloc_dealloc_,name) goto 9999 endif - b_col_glob => aux_b(:,1) do i=1, m_problem b_col_glob(i) = 1.d0 enddo endif + + if ((have_guess).and.(psb_size(aux_g,dim=ione) == m_problem)) then + ! if any initial guess were present, broadcast the first one + write(psb_err_unit,'("Ok, got an initial guess ")') + guess_col_glob =>aux_g(:,1) + else + write(psb_out_unit,'("Generating an initial guess...")') + call psb_realloc(m_problem,1,aux_g,ircode) + if (ircode /= 0) then + call psb_errpush(psb_err_alloc_dealloc_,name) + goto 9999 + endif + guess_col_glob => aux_g(:,1) + do i=1, m_problem + guess_col_glob(i) = 0.d0 + enddo + endif + + if ((have_ref).and.(psb_size(aux_x,dim=ione) == m_problem)) then + ! if any reference were present, broadcast the first one + write(psb_err_unit,'("Ok, got a reference solution ")') + ref_col_glob =>aux_x(:,1) + else + write(psb_out_unit,'("No reference solution...")') + endif + + ! clean zeros in the input matrix + call aux_a%clean_zeros(info) + else call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) + end if + + + ! + ! Renumbering (NONE for the moment) + ! + if (iam==psb_root_) then + renum='NONE' + + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,'(" ")') + write(psb_out_unit,*) 'Bandwidth and profile: ',lbw,ubw,prf + write(psb_out_unit,*) 'Renumbering algorithm: ',psb_toupper(renum) + + if (trim(psb_toupper(renum))/='NONE') then + call psb_mat_renum(renum,aux_a,info,perm=perm) + if (info /= 0) then + write(psb_err_unit,*) 'Error from RENUM',info + goto 9999 + end if + call psb_gelp('N',perm(1:m_problem),b_col_glob(1:m_problem),info) + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,*) 'Bandwidth and profile (renumberd):',lbw,ubw,prf + end if + + write(psb_out_unit,'(" ")') + end if + ! ! switch over different partition types - if (ipart == 0) then + ! + if (ipart == 0) then call psb_barrier(ictxt) if (iam == psb_root_) write(psb_out_unit,'("Partition type: block")') allocate(ivg(m_problem),ipv(np)) @@ -211,10 +332,8 @@ program mld_sf_sample if (iam == psb_root_) then write(psb_out_unit,'("Partition type: graph")') write(psb_out_unit,'(" ")') - ! write(psb_err_unit,'("Build type: graph")') call build_mtpart(aux_a,np) endif -!!$ call psb_barrier(ictxt) call distr_mtpart(psb_root_,ictxt) call getv_mtpart(ivg) call psb_matdist(aux_a, a, ictxt,desc_a,info,fmt=afmt,v=ivg) @@ -223,89 +342,136 @@ program mld_sf_sample call psb_matdist(aux_a, a, ictxt, desc_a,info,fmt=afmt,parts=part_block) end if - call psb_scatter(b_col_glob,b_col,desc_a,info,root=psb_root_) + ! + ! Scatter rhs, initial guess and reference solution + ! + call psb_geall(b_col,desc_a,info) call psb_geall(x_col,desc_a,info) - call x_col%zero() - call psb_geasb(x_col,desc_a,info) - call psb_geall(r_col,desc_a,info) - call r_col%zero() - call psb_geasb(r_col,desc_a,info) - t2 = psb_wtime() - t1 - + if (have_ref) call psb_geall(ref_col,desc_a,info) + + if (iam == psb_root_) write(psb_out_unit,'("Scatter rhs")') + call psb_scatter(b_col_glob,b_col,desc_a,info) + if (iam == psb_root_) write(psb_out_unit,'("Scatter initial guess")') + call psb_scatter(guess_col_glob,x_col,desc_a,info) + if (have_ref) then + if (iam == psb_root_) write(psb_out_unit,'("Scatter reference solution")') + call psb_scatter(ref_col_glob,ref_col,desc_a,info) + end if + t2 = psb_wtime() - t1 call psb_amx(ictxt, t2) if (iam == psb_root_) then write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Time to read and partition matrix : ",es12.5)')t2 + write(psb_out_unit,'("Time to read and partition matrix, rhs(, guess, ref sol) : ",es12.5)')t2 write(psb_out_unit,'(" ")') - write(psb_out_unit,*) 'Preconditioner: ',prec_choice%descr end if - ! - - if (psb_toupper(prec_choice%prec) == 'ML') then - call mld_precinit(prec,prec_choice%prec, info) - if (prec_choice%nlev > 0) & - & call mld_precset(prec,'max_prec_levs', prec_choice%nlev, info) - call mld_precset(prec,'aggr_kind', prec_choice%aggrkind,info) - call mld_precset(prec,'aggr_alg', prec_choice%aggr_alg,info) - call mld_precset(prec,'aggr_ord', prec_choice%aggr_ord,info) - call mld_precset(prec,'aggr_scale', prec_choice%ascale, info) - call mld_precset(prec,'aggr_thresh', prec_choice%athres, info) - call mld_precset(prec,'smoother_type', prec_choice%smther, info) - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) - call mld_precset(prec,'ml_type', prec_choice%mltype, info) - call mld_precset(prec,'smoother_pos', prec_choice%smthpos, info) - call mld_precset(prec,'coarse_solve', prec_choice%csolve, info) - call mld_precset(prec,'coarse_subsolve', prec_choice%csbsolve,info) - call mld_precset(prec,'coarse_mat', prec_choice%cmat, info) - call mld_precset(prec,'coarse_fillin', prec_choice%cfill, info) - call mld_precset(prec,'coarse_iluthrs', prec_choice%cthres, info) - call mld_precset(prec,'coarse_sweeps', prec_choice%cjswp, info) - ! building the preconditioner + ! + ! initialize the preconditioner + ! + if (psb_toupper(p_choice%ptype) == 'ML') then + ! multilevel preconditioner + call prec%init(p_choice%ptype,info) + + call prec%set('ml_type', p_choice%mltype, info) + call prec%set('outer_sweeps', p_choice%otr_sweeps,info) + if (p_choice%csize>0)& + & call prec%set('coarse_aggr_size', p_choice%csize, info) + if (p_choice%mnaggratio>0)& + & call prec%set('min_aggr_ratio', p_choice%mnaggratio, info) + if (p_choice%maxlevs>0)& + & call prec%set('max_prec_levs', p_choice%maxlevs, info) + if (p_choice%ascale > dzero) & + & call prec%set('aggr_scale', p_choice%ascale, info) + if (p_choice%athres >= dzero) & + & call prec%set('aggr_thresh', p_choice%athres, info) + if (p_choice%thrvsz>0) then + do k=1,min(p_choice%thrvsz,size(prec%precv)-1) + call prec%set('aggr_thresh', p_choice%athresv(k), info,ilev=(k+1)) + end do + end if + + call prec%set('aggr_kind', p_choice%aggrkind, info) + call prec%set('aggr_alg', p_choice%aggr_alg, info) + call prec%set('aggr_ord', p_choice%aggr_ord, info) + call prec%set('aggr_filter', p_choice%aggr_filter,info) + call prec%set('aggr_omega_alg', p_choice%aggr_omalg, info) + if (psb_toupper(p_choice%aggr_omalg) == 'EIG_EST') then + call prec%set('aggr_eig', p_choice%aggr_eig, info) + else if (psb_toupper(p_choice%aggr_omalg) == 'USER_CHOICE') then + call prec%set('aggr_omega_val', p_choice%omega_val, info) + end if + call prec%set('coarse_solve', p_choice%csolve, info) + if (psb_toupper(p_choice%csolve) == 'BJAC') & + & call prec%set('coarse_subsolve', p_choice%csbsolve, info) + call prec%set('coarse_mat', p_choice%cmat, info) + call prec%set('coarse_fillin', p_choice%cfill, info) + call prec%set('coarse_iluthrs', p_choice%cthres, info) + call prec%set('coarse_sweeps', p_choice%cjswp, info) + + + call prec%set('smoother_type', p_choice%smther, info) + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + + if (psb_toupper(p_choice%smther2) /= 'NONE') then + call prec%set('smoother_type', p_choice%smther2, info,pos='post') + call prec%set('smoother_sweeps', p_choice%jsweeps2, info,pos='post') + call prec%set('sub_ovr', p_choice%novr2, info,pos='post') + call prec%set('sub_restr', p_choice%restr2, info,pos='post') + call prec%set('sub_prol', p_choice%prol2, info,pos='post') + call prec%set('sub_solve', p_choice%solve2, info,pos='post') + call prec%set('sub_fillin', p_choice%fill2, info,pos='post') + call prec%set('sub_iluthrs', p_choice%thr2, info,pos='post') + end if + + ! build the preconditioner call psb_barrier(ictxt) t1 = psb_wtime() - call mld_hierarchy_bld(a,desc_a,prec,info) + call prec%hierarchy_build(a,desc_a,info) thier = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if call psb_barrier(ictxt) t1 = psb_wtime() - call mld_smoothers_bld(a,desc_a,prec,info) + call prec%smoothers_build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if else + ! 1-level preconditioner nlv = 1 - call mld_precinit(prec,prec_choice%prec,info) - if (psb_toupper(prec_choice%prec) /= 'NONE') then - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) + call prec%init(p_choice%ptype,info) + + if (psb_toupper(p_choice%ptype) /= 'NONE') then + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + !!! call prec%set('solver_sweeps', p_choice%svsweeps, info) end if - ! building the preconditioner + + ! build the preconditioner thier = dzero t1 = psb_wtime() - call mld_precbld(a,desc_a,prec,info) + call prec%build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if end if @@ -314,6 +480,8 @@ program mld_sf_sample call psb_amx(ictxt, tprec) if(iam == psb_root_) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'("Preconditioner: ",a)') trim(p_choice%descr) write(psb_out_unit,'("Preconditioner time: ",es12.5)')thier+tprec write(psb_out_unit,'(" ")') end if @@ -321,43 +489,71 @@ program mld_sf_sample iparm = 0 call psb_barrier(ictxt) t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b_col,x_col,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + call psb_krylov(s_choice%kmethd,a,prec,b_col,x_col,s_choice%eps,& + & desc_a,info,itmax=s_choice%itmax,iter=iter,err=err,itrace=s_choice%itrace,& + & istop=s_choice%istopc,irst=s_choice%irst) call psb_barrier(ictxt) tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + + ! compute residual norms + call psb_geall(r_col,desc_a,info) + call r_col%zero() + call psb_geasb(r_col,desc_a,info) call psb_geaxpby(sone,b_col,szero,r_col,desc_a,info) call psb_spmm(-sone,a,x_col,sone,r_col,desc_a,info) resmx = psb_genrm2(r_col,desc_a,info) resmxp = psb_geamax(r_col,desc_a,info) - + + ! compute error in solution + if (have_ref) then + call psb_geaxpby(-sone,x_col,sone,ref_col,desc_a,info) + xdiffn2 = psb_genrm2(ref_col,desc_a,info) + xdiffni = psb_geamax(ref_col,desc_a,info) + xn2 = psb_genrm2(ref_col,desc_a,info) + xni = psb_geamax(ref_col,desc_a,info) + end if + amatsize = a%sizeof() descsize = desc_a%sizeof() precsize = prec%sizeof() call psb_sum(ictxt,amatsize) call psb_sum(ictxt,descsize) call psb_sum(ictxt,precsize) + call prec%descr(info) if (iam == psb_root_) then - call mld_precdescr(prec,info) write(psb_out_unit,'("Matrix: ",a)')mtrx_file write(psb_out_unit,'("Computed solution on ",i8," processors")')np - write(psb_out_unit,'("Iterations to convergence : ",i6)')iter - write(psb_out_unit,'("Error estimate on exit : ",es12.5)')err + write(psb_out_unit,'("Krylov method : ",a)') trim(s_choice%kmethd) + write(psb_out_unit,'("Preconditioner : ",a)') trim(p_choice%descr) + write(psb_out_unit,'("Iterations to convergence : ",i12)')iter + write(psb_out_unit,'("Relative error estimate on exit : ",es12.5)') err write(psb_out_unit,'("Number of levels in hierarchy : ",i12)') prec%get_nlevs() write(psb_out_unit,'("Time to build hierarchy : ",es12.5)')thier write(psb_out_unit,'("Time to build smoothers : ",es12.5)')tprec write(psb_out_unit,'("Total time for preconditioner : ",es12.5)')tprec+thier write(psb_out_unit,'("Time to solve system : ",es12.5)')tslv - write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/(iter) - write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec - write(psb_out_unit,'("Residual norm 2 : ",es12.5)')resmx - write(psb_out_unit,'("Residual norm inf : ",es12.5)')resmxp + write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/iter + write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec+thier + write(psb_out_unit,'("Residual 2-norm : ",es12.5)')resmx + write(psb_out_unit,'("Residual inf-norm : ",es12.5)')resmxp write(psb_out_unit,'("Total memory occupation for A : ",i12)')amatsize write(psb_out_unit,'("Total memory occupation for DESC_A : ",i12)')descsize write(psb_out_unit,'("Total memory occupation for PREC : ",i12)')precsize write(psb_out_unit,'("Storage format for A : ",a )')a%get_fmt() write(psb_out_unit,'("Storage format for DESC_A : ",a )')desc_a%get_fmt() + if (have_ref) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,5x,a8)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(2-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffn2,xn2,xdiffn2/xn2 + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,4x,a10)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(inf-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffni,xni,xdiffni/xni + end if + end if call psb_gather(x_col_glob,x_col,desc_a,info,root=psb_root_) @@ -367,13 +563,15 @@ program mld_sf_sample if (iam == psb_root_) then write(psb_err_unit,'(" ")') write(psb_err_unit,'("Saving x on file")') - write(20,*) 'matrix: ',mtrx_file - write(20,*) 'computed solution on ',np,' processors.' - write(20,*) 'iterations to convergence: ',iter - write(20,*) 'error estimate (infinity norm) on exit:', & - & ' ||r||/(||a||||x||+||b||) = ',err - write(20,'("Residual norm 2 : ",es12.5)')resmx - write(20,'("Residual norm inf : ",es12.5)')resmxp + write(20,*) 'Matrix: ',mtrx_file + write(20,*) 'Krylov method:',trim(s_choice%kmethd) + write(20,*) 'Preconditioner:',trim(p_choice%descr) + write(20,*) 'Computed solution on ',np,' processors.' + write(20,*) 'Iterations to convergence: ',iter + write(20,*) 'Error estimate (infinity norm) on exit:', & + & ' ||r||/||b|| (inf-norm) = ',err + write(20,'(" Residual 2-norm 2 : ",es12.5)')resmx + write(20,'(" Residual inf-norm : ",es12.5)')resmxp write(20,'(a8,4(2x,a20))') 'I','X(I)','R(I)','B(I)' do i=1,m_problem write(20,998) i,x_col_glob(i),r_col_glob(i),b_col_glob(i) @@ -383,10 +581,12 @@ program mld_sf_sample 993 format(i6,4(1x,e12.6)) - call psb_gefree(b_col, desc_a,info) - call psb_gefree(x_col, desc_a,info) + call psb_gefree(b_col,desc_a,info) + call psb_gefree(x_col,desc_a,info) + call psb_gefree(r_col,desc_a,info) + call psb_gefree(ref_col,desc_a,info) call psb_spfree(a, desc_a,info) - call mld_precfree(prec,info) + call prec%free(info) call psb_cdfree(desc_a,info) call psb_exit(ictxt) @@ -399,101 +599,161 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(icontxt,mtrx,rhs,filefmt,kmethd,& - & prec, ipart,afmt,istopc,itmax,itrace,irst,eps) + subroutine get_parms(icontxt,mtrx,rhs,guess,sol,filefmt,ipart,afmt,solve,prec) use psb_base_mod implicit none - integer(psb_ipk_) :: icontxt - character(len=*) :: kmethd, mtrx, rhs, afmt,filefmt + integer(psb_ipk_) :: icontxt, ipart + character(len=*) :: mtrx, rhs, guess, sol, filefmt, afmt + type(solverdata) :: solve type(precdata) :: prec - real(psb_spk_) :: eps - integer(psb_ipk_) :: iret, istopc,itmax,itrace, ipart, irst - integer(psb_ipk_) :: iam, nm, np, i + integer(psb_ipk_) :: iam, nm, np call psb_info(icontxt,iam,np) if (iam == psb_root_) then - ! read input parameters - call read_data(mtrx,psb_inp_unit) - call read_data(rhs,psb_inp_unit) - call read_data(filefmt,psb_inp_unit) - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(ipart,psb_inp_unit) - call read_data(istopc,psb_inp_unit) - call read_data(itmax,psb_inp_unit) - call read_data(itrace,psb_inp_unit) - call read_data(irst,psb_inp_unit) - call read_data(eps,psb_inp_unit) + ! read input data + ! + ! input files + call read_data(mtrx,psb_inp_unit) ! matrix file + call read_data(rhs,psb_inp_unit) ! rhs file + call read_data(guess,psb_inp_unit) ! starting guess file + call read_data(sol,psb_inp_unit) ! solution file (for comparison) + call read_data(filefmt,psb_inp_unit) ! format of files + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(ipart,psb_inp_unit) ! partition type + ! Krylov solver data + call read_data(solve%kmethd,psb_inp_unit) ! Krylov solver + call read_data(solve%istopc,psb_inp_unit) ! stopping criterion + call read_data(solve%itmax,psb_inp_unit) ! max num iterations + call read_data(solve%itrace,psb_inp_unit) ! tracing + call read_data(solve%irst,psb_inp_unit) ! restart + call read_data(solve%eps,psb_inp_unit) ! tolerance + ! preconditioner type call read_data(prec%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prec%prec,psb_inp_unit) ! overall prectype - call read_data(prec%novr,psb_inp_unit) ! number of overlap layers - call read_data(prec%restr,psb_inp_unit) ! restriction over application of as - call read_data(prec%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prec%solve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%fill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%thr,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%jsweeps,psb_inp_unit) ! Jacobi sweeps for PJAC - if (psb_toupper(prec%prec) == 'ML') then - call read_data(prec%nlev,psb_inp_unit) ! Number of levels in multilevel prec. - call read_data(prec%smther,psb_inp_unit) ! Smoother type. - call read_data(prec%aggrkind,psb_inp_unit) ! smoothed/raw aggregatin - call read_data(prec%aggr_alg,psb_inp_unit) ! local or global aggregation - call read_data(prec%aggr_ord,psb_inp_unit) ! Ordering for aggregation - call read_data(prec%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prec%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prec%cmat,psb_inp_unit) ! coarse mat - call read_data(prec%csolve,psb_inp_unit) ! Factorization type: BJAC, SuperLU, UMFPACK. - call read_data(prec%csbsolve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%cfill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%cthres,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%cjswp,psb_inp_unit) ! Jacobi sweeps - call read_data(prec%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prec%ascale,psb_inp_unit) ! smoother aggr thresh + call read_data(prec%ptype,psb_inp_unit) ! preconditioner type + ! general AMG data + call read_data(prec%mltype,psb_inp_unit) ! AMG cycle type + call read_data(prec%otr_sweeps,psb_inp_unit) ! number of AMG cycles + call read_data(prec%maxlevs,psb_inp_unit) ! max number of levels in AMG prec + call read_data(prec%csize,psb_inp_unit) ! min size coarsest mat + ! aggregation + call read_data(prec%aggrkind,psb_inp_unit) ! aggregation type + call read_data(prec%aggr_alg,psb_inp_unit) ! parallel aggregation alg + call read_data(prec%aggr_ord,psb_inp_unit) ! ordering for aggregation + call read_data(prec%aggr_filter,psb_inp_unit) ! filtering + call read_data(prec%mnaggratio,psb_inp_unit) ! minimum aggregation ratio + call read_data(prec%thrvsz,psb_inp_unit) ! size of aggr thresh vector + if (prec%thrvsz > 0) then + call psb_realloc(prec%thrvsz,prec%athresv,info) + call read_data(prec%athresv,psb_inp_unit) ! aggr thresh vector + else + read(psb_inp_unit,*) ! dummy read to skip a record end if + call read_data(prec%athres,psb_inp_unit) ! smoothed aggr thresh + call read_data(prec%aggr_omalg,psb_inp_unit) ! alg for estimating omega + call read_data(prec%aggr_eig,psb_inp_unit) ! alg for estimating omega + call read_data(prec%omega_val,psb_inp_unit) ! alg for estimating omega + ! AMG smoother (or pre-smoother) / 1-lev preconditioner + call read_data(prec%smther,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps,psb_inp_unit) ! (pre-)smoother / 1-lev prec sweeps + call read_data(prec%novr,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve,psb_inp_unit) ! local subsolver + call read_data(prec%fill,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr,psb_inp_unit) ! threshold for ILUT + ! AMG post-smoother + call read_data(prec%smther2,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps2,psb_inp_unit) ! (post-)smoother sweeps + call read_data(prec%novr2,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr2,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol2,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve2,psb_inp_unit) ! local subsolver + call read_data(prec%fill2,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr2,psb_inp_unit) ! threshold for ILUT + ! coasest-level solver + call read_data(prec%csolve,psb_inp_unit) ! coarsest-lev solver + call read_data(prec%cmat,psb_inp_unit) ! coarsest mat layout + call read_data(prec%csbsolve,psb_inp_unit) ! coarsest-lev subsolver + call read_data(prec%cfill,psb_inp_unit) ! fill-in for incompl LU + call read_data(prec%cthres,psb_inp_unit) ! Threshold for ILUT + call read_data(prec%cjswp,psb_inp_unit) ! sweeps for GS/JAC subsolver end if call psb_bcast(icontxt,mtrx) call psb_bcast(icontxt,rhs) + call psb_bcast(icontxt,guess) + call psb_bcast(icontxt,sol) call psb_bcast(icontxt,filefmt) - call psb_bcast(icontxt,kmethd) call psb_bcast(icontxt,afmt) call psb_bcast(icontxt,ipart) - call psb_bcast(icontxt,istopc) - call psb_bcast(icontxt,itmax) - call psb_bcast(icontxt,itrace) - call psb_bcast(icontxt,irst) - call psb_bcast(icontxt,eps) - call psb_bcast(icontxt,prec%descr) ! verbose description of the prec - call psb_bcast(icontxt,prec%prec) ! overall prectype - call psb_bcast(icontxt,prec%novr) ! number of overlap layers - call psb_bcast(icontxt,prec%restr) ! restriction over application of as - call psb_bcast(icontxt,prec%prol) ! prolongation over application of as - call psb_bcast(icontxt,prec%solve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%fill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%thr) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%jsweeps) ! Jacobi sweeps - if (psb_toupper(prec%prec) == 'ML') then - call psb_bcast(icontxt,prec%smther) ! Smoother type. - call psb_bcast(icontxt,prec%nlev) ! Number of levels in multilevel prec. - call psb_bcast(icontxt,prec%aggrkind) ! smoothed/raw aggregatin - call psb_bcast(icontxt,prec%aggr_alg) ! local or global aggregation - call psb_bcast(icontxt,prec%aggr_ord) ! Ordering for aggregation - call psb_bcast(icontxt,prec%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(icontxt,prec%smthpos) ! side: pre, post, both smoothing - call psb_bcast(icontxt,prec%cmat) ! coarse mat - call psb_bcast(icontxt,prec%csolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%csbsolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%cfill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%cthres) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%cjswp) ! Jacobi sweeps - call psb_bcast(icontxt,prec%athres) ! smoother aggr thresh - call psb_bcast(icontxt,prec%ascale) ! smoother aggr scale factor + + call psb_bcast(icontxt,solve%kmethd) + call psb_bcast(icontxt,solve%istopc) + call psb_bcast(icontxt,solve%itmax) + call psb_bcast(icontxt,solve%itrace) + call psb_bcast(icontxt,solve%irst) + call psb_bcast(icontxt,solve%eps) + + call psb_bcast(icontxt,prec%descr) + call psb_bcast(icontxt,prec%ptype) + + ! broadcast first (pre-)smoother / 1-lev prec data + call psb_bcast(icontxt,prec%smther) ! actually not needed for 1-lev precs + call psb_bcast(icontxt,prec%jsweeps) + call psb_bcast(icontxt,prec%novr) + call psb_bcast(icontxt,prec%restr) + call psb_bcast(icontxt,prec%prol) + call psb_bcast(icontxt,prec%solve) + call psb_bcast(icontxt,prec%fill) + call psb_bcast(icontxt,prec%thr) + + ! broadcast (other) AMG parameters + if (psb_toupper(prec%ptype) == 'ML') then + + call psb_bcast(icontxt,prec%mltype) + call psb_bcast(icontxt,prec%otr_sweeps) + call psb_bcast(icontxt,prec%maxlevs) + + call psb_bcast(icontxt,prec%smther2) + call psb_bcast(icontxt,prec%jsweeps2) + call psb_bcast(icontxt,prec%novr2) + call psb_bcast(icontxt,prec%restr2) + call psb_bcast(icontxt,prec%prol2) + call psb_bcast(icontxt,prec%solve2) + call psb_bcast(icontxt,prec%fill2) + call psb_bcast(icontxt,prec%thr2) + + call psb_bcast(icontxt,prec%aggrkind) + call psb_bcast(icontxt,prec%aggr_alg) + call psb_bcast(icontxt,prec%aggr_ord) + call psb_bcast(icontxt,prec%aggr_filter) + call psb_bcast(icontxt,prec%mnaggratio) + call psb_bcast(ictxt,prec%thrvsz) + if (prec%thrvsz > 0) then + if (iam /= psb_root_) call psb_realloc(prec%thrvsz,prec%athresv,info) + call psb_bcast(ictxt,prec%athresv) + end if + call psb_bcast(ictxt,prec%athres) + call psb_bcast(ictxt,prec%ascale) + call psb_bcast(ictxt,prec%aggr_omalg) + call psb_bcast(ictxt,prec%aggr_eig) + call psb_bcast(ictxt,prec%omega_val) + + call psb_bcast(icontxt,prec%csize) + call psb_bcast(icontxt,prec%cmat) + call psb_bcast(icontxt,prec%csolve) + call psb_bcast(icontxt,prec%csbsolve) + call psb_bcast(icontxt,prec%cfill) + call psb_bcast(icontxt,prec%cthres) + call psb_bcast(icontxt,prec%cjswp) + end if end subroutine get_parms + subroutine pr_usage(iout) integer(psb_ipk_) iout write(iout, *) ' number of parameters is incorrect!' @@ -511,4 +771,5 @@ contains write(iout, *) ' ptype partition strategy default 0' write(iout, *) ' 0: block partition ' end subroutine pr_usage + end program mld_sf_sample diff --git a/tests/fileread/mld_zf_sample.f90 b/tests/fileread/mld_zf_sample.f90 index 7c57621c..16316b5e 100644 --- a/tests/fileread/mld_zf_sample.f90 +++ b/tests/fileread/mld_zf_sample.f90 @@ -4,7 +4,7 @@ ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) ! -! (C) Copyright 2008, 2010, 2012, 2015, 2017 , 2016 +! (C) Copyright 2008, 2010, 2012, 2015, 2017 ! ! Salvatore Filippone Cranfield University, UK ! Ambra Abdullahi Hassan University of Rome Tor Vergata, IT @@ -47,35 +47,83 @@ program mld_zf_sample ! input parameters - character(len=40) :: kmethd, mtrx_file, rhs_file + + character(len=40) :: kmethd, mtrx_file, rhs_file, guess_file, sol_file character(len=2) :: filefmt + + ! Krylov solver data + type solverdata + character(len=40) :: kmethd ! Krylov solver + integer(psb_ipk_) :: istopc ! stopping criterion + integer(psb_ipk_) :: itmax ! maximum number of iterations + integer(psb_ipk_) :: itrace ! tracing + integer(psb_ipk_) :: irst ! restart + real(psb_dpk_) :: eps ! stopping tolerance + end type solverdata + type(solverdata) :: s_choice + + ! preconditioner data type precdata - character(len=20) :: descr ! verbose description of the prec - character(len=10) :: prec ! overall prectype + + ! preconditioner type + character(len=40) :: descr ! verbose description of the prec + character(len=10) :: ptype ! preconditioner type + + ! general AMG data + character(len=16) :: mltype ! AMG cycle type + integer(psb_ipk_) :: otr_sweeps ! number of AMG cycles + integer(psb_ipk_) :: maxlevs ! maximum number of levels in AMG preconditioner + + ! AMG aggregation + character(len=16) :: aggrkind ! aggregation type: SMOOTHED, NONSMOOTHED + character(len=16) :: aggr_alg ! parallel aggregation algorithm: DEC, SYMDEC + character(len=16) :: aggr_ord ! ordering for aggregation: NATURAL, DEGREE + character(len=16) :: aggr_filter ! filtering: FILTER, NO_FILTER + real(psb_dpk_) :: mnaggratio ! minimum aggregation ratio + real(psb_dpk_), allocatable :: athresv(:) ! smoothed aggregation threshold vector + integer(psb_ipk_) :: thrvsz ! size of threshold vector + real(psb_dpk_) :: athres ! smoothed aggregation threshold + real(psb_dpk_) :: ascale ! smoothed aggregation scale factor for threshold + character(len=16) :: aggr_omalg ! algorithm for estimating omega parameter + character(len=16) :: aggr_eig ! Eigenvalue estimation procedure + real(psb_dpk_) :: omega_val ! Eigenvalue estimate value + integer(psb_ipk_) :: csize ! minimum size of coarsest matrix + + ! AMG smoother or pre-smoother; also 1-lev preconditioner + character(len=16) :: smther ! (pre-)smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps ! (pre-)smoother / 1-lev prec. sweeps integer(psb_ipk_) :: novr ! number of overlap layers - integer(psb_ipk_) :: jsweeps ! Jacobi/smoother sweeps character(len=16) :: restr ! restriction over application of AS character(len=16) :: prol ! prolongation over application of AS - character(len=16) :: solve ! factorization type: ILU, SuperLU, UMFPACK - integer(psb_ipk_) :: fill ! fillin for factorization - real(psb_dpk_) :: thr ! threshold for fact. ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: nlev ! number of levels in multilevel prec. - character(len=16) :: aggrkind ! smoothed, raw aggregation - character(len=16) :: aggr_alg ! aggregation algorithm (currently only decoupled) - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: mltype ! additive or multiplicative multi-level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - character(len=16) :: cmat ! coarse mat: distributed, replicated - character(len=16) :: csolve ! coarse solver: bjac, umf, slu, sludist - character(len=16) :: csbsolve ! coarse subsolver: ILU, ILU(T), SuperLU, UMFPACK - integer(psb_ipk_) :: cfill ! fillin for coarse factorization - real(psb_dpk_) :: cthres ! threshold for coarse fact. ILU(T) - integer(psb_ipk_) :: cjswp ! block-Jacobi sweeps - real(psb_dpk_) :: athres ! smoothed aggregation threshold - real(psb_dpk_) :: ascale ! smoothed aggregation scale factor + character(len=16) :: solve ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill ! fill-in for incomplete LU factorization + real(psb_dpk_) :: thr ! threshold for ILUT factorization + + ! AMG post-smoother; ignored by 1-lev preconditioner + character(len=16) :: smther2 ! post-smoother type: BJAC, AS + integer(psb_ipk_) :: jsweeps2 ! post-smoother sweeps + integer(psb_ipk_) :: novr2 ! number of overlap layers + character(len=16) :: restr2 ! restriction over application of AS + character(len=16) :: prol2 ! prolongation over application of AS + character(len=16) :: solve2 ! local subsolver type: ILU, MILU, ILUT, + ! UMF, MUMPS, SLU, FWGS, BWGS, JAC + integer(psb_ipk_) :: fill2 ! fill-in for incomplete LU factorization + real(psb_dpk_) :: thr2 ! threshold for ILUT factorization + + ! coarsest-level solver + character(len=16) :: cmat ! coarsest matrix layout: REPL, DIST + character(len=16) :: csolve ! coarsest-lev solver: BJAC, SLUDIST (distr. + ! mat.); UMF, MUMPS, SLU, ILU, ILUT, MILU + ! (repl. mat.) + character(len=16) :: csbsolve ! coarsest-lev local subsolver: ILU, ILUT, + ! MILU, UMF, MUMPS, SLU + integer(psb_ipk_) :: cfill ! fill-in for incomplete LU factorization + real(psb_dpk_) :: cthres ! threshold for ILUT factorization + integer(psb_ipk_) :: cjswp ! sweeps for GS or JAC coarsest-lev subsolver + end type precdata - type(precdata) :: prec_choice + type(precdata) :: p_choice ! sparse matrices type(psb_zspmat_type) :: a, aux_a @@ -83,10 +131,10 @@ program mld_zf_sample ! preconditioner data type(mld_zprec_type) :: prec ! dense matrices - complex(psb_dpk_), allocatable, target :: aux_b(:,:), d(:) + complex(psb_dpk_), allocatable, target :: aux_b(:,:), d(:), aux_g(:,:), aux_x(:,:) complex(psb_dpk_), allocatable , save :: x_col_glob(:), r_col_glob(:) - complex(psb_dpk_), pointer :: b_col_glob(:) - type(psb_z_vect_type) :: b_col, x_col, r_col + complex(psb_dpk_), pointer :: b_col_glob(:), ref_col_glob(:), guess_col_glob(:) + type(psb_z_vect_type) :: b_col, x_col, r_col, ref_col ! communications data structure type(psb_desc_type):: desc_a @@ -94,24 +142,23 @@ program mld_zf_sample integer(psb_ipk_) :: ictxt, iam, np ! solver paramters - integer(psb_ipk_) :: iter, itmax, ierr, itrace, ircode, ipart,& - & methd, istopc, irst, nlv + integer(psb_ipk_) :: iter, ircode, ipart, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_dpk_) :: err, eps + real(psb_dpk_) :: err character(len=5) :: afmt character(len=20) :: name, renum integer(psb_ipk_), parameter :: iunit=12 integer(psb_ipk_) :: iparm(20) - character(len=40) :: fprefix ! other variables - integer(psb_ipk_) :: i,info,j,m_problem - integer(psb_ipk_) :: internal, m,ii,nnzero + integer(psb_ipk_) :: i, info, j, k, m_problem + integer(psb_ipk_) :: lbw, ubw, prf real(psb_dpk_) :: t1, t2, tprec, thier, tslv - real(psb_dpk_) :: r_amax, b_amax, scale,resmx,resmxp - integer(psb_ipk_) :: nrhs, nrow, n_row, dim, nv, ne - integer(psb_ipk_), allocatable :: ivg(:), ipv(:) + real(psb_dpk_) :: resmx, resmxp, xdiffn2, xdiffni, xni, xn2 + integer(psb_ipk_) :: nrhs, nv + integer(psb_ipk_), allocatable :: ivg(:), ipv(:), perm(:) + logical :: have_guess=.false., have_ref=.false. call psb_init(ictxt) call psb_info(ictxt,iam,np) @@ -131,41 +178,55 @@ program mld_zf_sample ! Hello world ! if (iam == psb_root_) then - write(*,*) 'Welcome to MLD2P4 version: ',mld_version_string_ - write(*,*) 'This is the ',trim(name),' sample program' + write(psb_out_unit,*) ' ' + write(psb_out_unit,*) 'Welcome to MLD2P4 version: ',mld_version_string_ + write(psb_out_unit,*) 'This is the ',trim(name),' sample test program' + write(psb_out_unit,*) ' ' end if ! - ! get parameters + ! get parameters ! - call get_parms(ictxt,mtrx_file,rhs_file,filefmt,kmethd,& - & prec_choice,ipart,afmt,istopc,itmax,itrace,irst,eps) + call get_parms(ictxt,mtrx_file,rhs_file,guess_file,sol_file,filefmt, & + & ipart,afmt,s_choice,p_choice) call psb_barrier(ictxt) t1 = psb_wtime() - ! read the input matrix to be processed and (possibly) the rhs + ! read the input matrix to be processed and (possibly) the rhs, + ! the initial guess and the reference solution nrhs = 1 if (iam == psb_root_) then - select case(psb_toupper(filefmt)) + select case(psb_toupper(filefmt)) + case('MM') ! For Matrix Market we have an input file for the matrix - ! and an (optional) second file for the RHS. + ! and (optional) separate files for the rhs, the initial guess + ! and the reference solution call mm_mat_read(aux_a,info,iunit=iunit,filename=mtrx_file) - if (info == psb_success_) then - if (rhs_file /= 'NONE') then - call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) - end if + if ((info == psb_success_).and.(rhs_file /= 'NONE')) & + & call mm_array_read(aux_b,info,iunit=iunit,filename=rhs_file) + if ((info == psb_success_).and.(guess_file /= 'NONE')) then + call mm_array_read(aux_g,info,iunit=iunit,filename=guess_file) + have_guess = .true. + end if + if ((info == psb_success_).and.(sol_file /= 'NONE')) then + call mm_array_read(aux_x,info,iunit=iunit,filename=sol_file) + have_ref = .true. end if case ('HB') ! For Harwell-Boeing we have a single file which may or may not - ! contain an RHS. - call hb_read(aux_a,info,iunit=iunit,b=aux_b,filename=mtrx_file) + ! contain rhs, initial guess and reference solution. + call hb_read(aux_a,info,iunit=iunit,b=aux_b,& + & g=aux_g,x=aux_x,filename=mtrx_file) + have_guess = allocated(aux_g) + have_ref = allocated(aux_x) case default info = -1 write(psb_err_unit,*) 'Wrong choice for fileformat ', filefmt end select + if (info /= psb_success_) then write(psb_err_unit,*) 'Error while reading input matrix ' call psb_abort(ictxt) @@ -173,6 +234,8 @@ program mld_zf_sample m_problem = aux_a%get_nrows() call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) ! At this point aux_b may still be unallocated if (psb_size(aux_b,dim=ione) == m_problem) then @@ -181,24 +244,82 @@ program mld_zf_sample b_col_glob =>aux_b(:,1) else write(psb_out_unit,'("Generating an rhs...")') - write(psb_out_unit,'(" ")') call psb_realloc(m_problem,1,aux_b,ircode) if (ircode /= 0) then call psb_errpush(psb_err_alloc_dealloc_,name) goto 9999 endif - b_col_glob => aux_b(:,1) do i=1, m_problem b_col_glob(i) = 1.d0 enddo endif + + if ((have_guess).and.(psb_size(aux_g,dim=ione) == m_problem)) then + ! if any initial guess were present, broadcast the first one + write(psb_err_unit,'("Ok, got an initial guess ")') + guess_col_glob =>aux_g(:,1) + else + write(psb_out_unit,'("Generating an initial guess...")') + call psb_realloc(m_problem,1,aux_g,ircode) + if (ircode /= 0) then + call psb_errpush(psb_err_alloc_dealloc_,name) + goto 9999 + endif + guess_col_glob => aux_g(:,1) + do i=1, m_problem + guess_col_glob(i) = 0.d0 + enddo + endif + + if ((have_ref).and.(psb_size(aux_x,dim=ione) == m_problem)) then + ! if any reference were present, broadcast the first one + write(psb_err_unit,'("Ok, got a reference solution ")') + ref_col_glob =>aux_x(:,1) + else + write(psb_out_unit,'("No reference solution...")') + endif + + ! clean zeros in the input matrix + call aux_a%clean_zeros(info) + else call psb_bcast(ictxt,m_problem) + call psb_bcast(ictxt,have_guess) + call psb_bcast(ictxt,have_ref) + end if + + + ! + ! Renumbering (NONE for the moment) + ! + if (iam==psb_root_) then + renum='NONE' + + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,'(" ")') + write(psb_out_unit,*) 'Bandwidth and profile: ',lbw,ubw,prf + write(psb_out_unit,*) 'Renumbering algorithm: ',psb_toupper(renum) + + if (trim(psb_toupper(renum))/='NONE') then + call psb_mat_renum(renum,aux_a,info,perm=perm) + if (info /= 0) then + write(psb_err_unit,*) 'Error from RENUM',info + goto 9999 + end if + call psb_gelp('N',perm(1:m_problem),b_col_glob(1:m_problem),info) + call psb_cmp_bwpf(aux_a,lbw,ubw,prf,info) + write(psb_out_unit,*) 'Bandwidth and profile (renumberd):',lbw,ubw,prf + end if + + write(psb_out_unit,'(" ")') + end if + ! ! switch over different partition types - if (ipart == 0) then + ! + if (ipart == 0) then call psb_barrier(ictxt) if (iam == psb_root_) write(psb_out_unit,'("Partition type: block")') allocate(ivg(m_problem),ipv(np)) @@ -211,10 +332,8 @@ program mld_zf_sample if (iam == psb_root_) then write(psb_out_unit,'("Partition type: graph")') write(psb_out_unit,'(" ")') - ! write(psb_err_unit,'("Build type: graph")') call build_mtpart(aux_a,np) endif -!!$ call psb_barrier(ictxt) call distr_mtpart(psb_root_,ictxt) call getv_mtpart(ivg) call psb_matdist(aux_a, a, ictxt,desc_a,info,fmt=afmt,v=ivg) @@ -223,89 +342,136 @@ program mld_zf_sample call psb_matdist(aux_a, a, ictxt, desc_a,info,fmt=afmt,parts=part_block) end if - call psb_scatter(b_col_glob,b_col,desc_a,info,root=psb_root_) + ! + ! Scatter rhs, initial guess and reference solution + ! + call psb_geall(b_col,desc_a,info) call psb_geall(x_col,desc_a,info) - call x_col%zero() - call psb_geasb(x_col,desc_a,info) - call psb_geall(r_col,desc_a,info) - call r_col%zero() - call psb_geasb(r_col,desc_a,info) - t2 = psb_wtime() - t1 - + if (have_ref) call psb_geall(ref_col,desc_a,info) + + if (iam == psb_root_) write(psb_out_unit,'("Scatter rhs")') + call psb_scatter(b_col_glob,b_col,desc_a,info) + if (iam == psb_root_) write(psb_out_unit,'("Scatter initial guess")') + call psb_scatter(guess_col_glob,x_col,desc_a,info) + if (have_ref) then + if (iam == psb_root_) write(psb_out_unit,'("Scatter reference solution")') + call psb_scatter(ref_col_glob,ref_col,desc_a,info) + end if + t2 = psb_wtime() - t1 call psb_amx(ictxt, t2) if (iam == psb_root_) then write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Time to read and partition matrix : ",es12.5)')t2 + write(psb_out_unit,'("Time to read and partition matrix, rhs(, guess, ref sol) : ",es12.5)')t2 write(psb_out_unit,'(" ")') - write(psb_out_unit,*) 'Preconditioner: ',prec_choice%descr end if - ! - - if (psb_toupper(prec_choice%prec) == 'ML') then - call mld_precinit(prec,prec_choice%prec, info) - if (prec_choice%nlev > 0) & - & call mld_precset(prec,'max_prec_levs', prec_choice%nlev, info) - call mld_precset(prec,'aggr_kind', prec_choice%aggrkind,info) - call mld_precset(prec,'aggr_alg', prec_choice%aggr_alg,info) - call mld_precset(prec,'aggr_ord', prec_choice%aggr_ord,info) - call mld_precset(prec,'aggr_scale', prec_choice%ascale, info) - call mld_precset(prec,'aggr_thresh', prec_choice%athres, info) - call mld_precset(prec,'smoother_type', prec_choice%smther, info) - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) - call mld_precset(prec,'ml_type', prec_choice%mltype, info) - call mld_precset(prec,'smoother_pos', prec_choice%smthpos, info) - call mld_precset(prec,'coarse_solve', prec_choice%csolve, info) - call mld_precset(prec,'coarse_subsolve', prec_choice%csbsolve,info) - call mld_precset(prec,'coarse_mat', prec_choice%cmat, info) - call mld_precset(prec,'coarse_fillin', prec_choice%cfill, info) - call mld_precset(prec,'coarse_iluthrs', prec_choice%cthres, info) - call mld_precset(prec,'coarse_sweeps', prec_choice%cjswp, info) - ! building the preconditioner + ! + ! initialize the preconditioner + ! + if (psb_toupper(p_choice%ptype) == 'ML') then + ! multilevel preconditioner + call prec%init(p_choice%ptype,info) + + call prec%set('ml_type', p_choice%mltype, info) + call prec%set('outer_sweeps', p_choice%otr_sweeps,info) + if (p_choice%csize>0)& + & call prec%set('coarse_aggr_size', p_choice%csize, info) + if (p_choice%mnaggratio>0)& + & call prec%set('min_aggr_ratio', p_choice%mnaggratio, info) + if (p_choice%maxlevs>0)& + & call prec%set('max_prec_levs', p_choice%maxlevs, info) + if (p_choice%ascale > dzero) & + & call prec%set('aggr_scale', p_choice%ascale, info) + if (p_choice%athres >= dzero) & + & call prec%set('aggr_thresh', p_choice%athres, info) + if (p_choice%thrvsz>0) then + do k=1,min(p_choice%thrvsz,size(prec%precv)-1) + call prec%set('aggr_thresh', p_choice%athresv(k), info,ilev=(k+1)) + end do + end if + + call prec%set('aggr_kind', p_choice%aggrkind, info) + call prec%set('aggr_alg', p_choice%aggr_alg, info) + call prec%set('aggr_ord', p_choice%aggr_ord, info) + call prec%set('aggr_filter', p_choice%aggr_filter,info) + call prec%set('aggr_omega_alg', p_choice%aggr_omalg, info) + if (psb_toupper(p_choice%aggr_omalg) == 'EIG_EST') then + call prec%set('aggr_eig', p_choice%aggr_eig, info) + else if (psb_toupper(p_choice%aggr_omalg) == 'USER_CHOICE') then + call prec%set('aggr_omega_val', p_choice%omega_val, info) + end if + call prec%set('coarse_solve', p_choice%csolve, info) + if (psb_toupper(p_choice%csolve) == 'BJAC') & + & call prec%set('coarse_subsolve', p_choice%csbsolve, info) + call prec%set('coarse_mat', p_choice%cmat, info) + call prec%set('coarse_fillin', p_choice%cfill, info) + call prec%set('coarse_iluthrs', p_choice%cthres, info) + call prec%set('coarse_sweeps', p_choice%cjswp, info) + + + call prec%set('smoother_type', p_choice%smther, info) + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + + if (psb_toupper(p_choice%smther2) /= 'NONE') then + call prec%set('smoother_type', p_choice%smther2, info,pos='post') + call prec%set('smoother_sweeps', p_choice%jsweeps2, info,pos='post') + call prec%set('sub_ovr', p_choice%novr2, info,pos='post') + call prec%set('sub_restr', p_choice%restr2, info,pos='post') + call prec%set('sub_prol', p_choice%prol2, info,pos='post') + call prec%set('sub_solve', p_choice%solve2, info,pos='post') + call prec%set('sub_fillin', p_choice%fill2, info,pos='post') + call prec%set('sub_iluthrs', p_choice%thr2, info,pos='post') + end if + + ! build the preconditioner call psb_barrier(ictxt) t1 = psb_wtime() - call mld_hierarchy_bld(a,desc_a,prec,info) + call prec%hierarchy_build(a,desc_a,info) thier = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if call psb_barrier(ictxt) t1 = psb_wtime() - call mld_smoothers_bld(a,desc_a,prec,info) + call prec%smoothers_build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if else + ! 1-level preconditioner nlv = 1 - call mld_precinit(prec,prec_choice%prec,info) - if (psb_toupper(prec_choice%prec) /= 'NONE') then - call mld_precset(prec,'smoother_sweeps', prec_choice%jsweeps, info) - call mld_precset(prec,'sub_ovr', prec_choice%novr, info) - call mld_precset(prec,'sub_restr', prec_choice%restr, info) - call mld_precset(prec,'sub_prol', prec_choice%prol, info) - call mld_precset(prec,'sub_solve', prec_choice%solve, info) - call mld_precset(prec,'sub_fillin', prec_choice%fill, info) - call mld_precset(prec,'sub_iluthrs', prec_choice%thr, info) + call prec%init(p_choice%ptype,info) + + if (psb_toupper(p_choice%ptype) /= 'NONE') then + call prec%set('smoother_sweeps', p_choice%jsweeps, info) + call prec%set('sub_ovr', p_choice%novr, info) + call prec%set('sub_restr', p_choice%restr, info) + call prec%set('sub_prol', p_choice%prol, info) + call prec%set('sub_solve', p_choice%solve, info) + call prec%set('sub_fillin', p_choice%fill, info) + call prec%set('sub_iluthrs', p_choice%thr, info) + !!! call prec%set('solver_sweeps', p_choice%svsweeps, info) end if - ! building the preconditioner + + ! build the preconditioner thier = dzero t1 = psb_wtime() - call mld_precbld(a,desc_a,prec,info) + call prec%build(a,desc_a,info) tprec = psb_wtime()-t1 if (info /= psb_success_) then - call psb_errpush(psb_err_from_subroutine_,name,a_err='psb_precbld') + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if end if @@ -314,6 +480,8 @@ program mld_zf_sample call psb_amx(ictxt, tprec) if(iam == psb_root_) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'("Preconditioner: ",a)') trim(p_choice%descr) write(psb_out_unit,'("Preconditioner time: ",es12.5)')thier+tprec write(psb_out_unit,'(" ")') end if @@ -321,43 +489,71 @@ program mld_zf_sample iparm = 0 call psb_barrier(ictxt) t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b_col,x_col,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + call psb_krylov(s_choice%kmethd,a,prec,b_col,x_col,s_choice%eps,& + & desc_a,info,itmax=s_choice%itmax,iter=iter,err=err,itrace=s_choice%itrace,& + & istop=s_choice%istopc,irst=s_choice%irst) call psb_barrier(ictxt) tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + + ! compute residual norms + call psb_geall(r_col,desc_a,info) + call r_col%zero() + call psb_geasb(r_col,desc_a,info) call psb_geaxpby(zone,b_col,zzero,r_col,desc_a,info) call psb_spmm(-zone,a,x_col,zone,r_col,desc_a,info) resmx = psb_genrm2(r_col,desc_a,info) resmxp = psb_geamax(r_col,desc_a,info) - + + ! compute error in solution + if (have_ref) then + call psb_geaxpby(-zone,x_col,zone,ref_col,desc_a,info) + xdiffn2 = psb_genrm2(ref_col,desc_a,info) + xdiffni = psb_geamax(ref_col,desc_a,info) + xn2 = psb_genrm2(ref_col,desc_a,info) + xni = psb_geamax(ref_col,desc_a,info) + end if + amatsize = a%sizeof() descsize = desc_a%sizeof() precsize = prec%sizeof() call psb_sum(ictxt,amatsize) call psb_sum(ictxt,descsize) call psb_sum(ictxt,precsize) + call prec%descr(info) if (iam == psb_root_) then - call mld_precdescr(prec,info) write(psb_out_unit,'("Matrix: ",a)')mtrx_file write(psb_out_unit,'("Computed solution on ",i8," processors")')np - write(psb_out_unit,'("Iterations to convergence : ",i6)')iter - write(psb_out_unit,'("Error estimate on exit : ",es12.5)')err + write(psb_out_unit,'("Krylov method : ",a)') trim(s_choice%kmethd) + write(psb_out_unit,'("Preconditioner : ",a)') trim(p_choice%descr) + write(psb_out_unit,'("Iterations to convergence : ",i12)')iter + write(psb_out_unit,'("Relative error estimate on exit : ",es12.5)') err write(psb_out_unit,'("Number of levels in hierarchy : ",i12)') prec%get_nlevs() write(psb_out_unit,'("Time to build hierarchy : ",es12.5)')thier write(psb_out_unit,'("Time to build smoothers : ",es12.5)')tprec write(psb_out_unit,'("Total time for preconditioner : ",es12.5)')tprec+thier write(psb_out_unit,'("Time to solve system : ",es12.5)')tslv - write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/(iter) - write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec - write(psb_out_unit,'("Residual norm 2 : ",es12.5)')resmx - write(psb_out_unit,'("Residual norm inf : ",es12.5)')resmxp + write(psb_out_unit,'("Time per iteration : ",es12.5)')tslv/iter + write(psb_out_unit,'("Total time : ",es12.5)')tslv+tprec+thier + write(psb_out_unit,'("Residual 2-norm : ",es12.5)')resmx + write(psb_out_unit,'("Residual inf-norm : ",es12.5)')resmxp write(psb_out_unit,'("Total memory occupation for A : ",i12)')amatsize write(psb_out_unit,'("Total memory occupation for DESC_A : ",i12)')descsize write(psb_out_unit,'("Total memory occupation for PREC : ",i12)')precsize write(psb_out_unit,'("Storage format for A : ",a )')a%get_fmt() write(psb_out_unit,'("Storage format for DESC_A : ",a )')desc_a%get_fmt() + if (have_ref) then + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,5x,a8)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(2-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffn2,xn2,xdiffn2/xn2 + write(psb_out_unit,'(" ")') + write(psb_out_unit,'(2x,a10,9x,a8,4x,a20,4x,a10)') & + & '||X-XREF||','||XREF||','||X-XREF||/||XREF||','(inf-norm)' + write(psb_out_unit,'(1x,3(e12.6,6x))') xdiffni,xni,xdiffni/xni + end if + end if call psb_gather(x_col_glob,x_col,desc_a,info,root=psb_root_) @@ -367,13 +563,15 @@ program mld_zf_sample if (iam == psb_root_) then write(psb_err_unit,'(" ")') write(psb_err_unit,'("Saving x on file")') - write(20,*) 'matrix: ',mtrx_file - write(20,*) 'computed solution on ',np,' processors.' - write(20,*) 'iterations to convergence: ',iter - write(20,*) 'error estimate (infinity norm) on exit:', & - & ' ||r||/(||a||||x||+||b||) = ',err - write(20,'("Residual norm 2 : ",es12.5)')resmx - write(20,'("Residual norm inf : ",es12.5)')resmxp + write(20,*) 'Matrix: ',mtrx_file + write(20,*) 'Krylov method:',trim(s_choice%kmethd) + write(20,*) 'Preconditioner:',trim(p_choice%descr) + write(20,*) 'Computed solution on ',np,' processors.' + write(20,*) 'Iterations to convergence: ',iter + write(20,*) 'Error estimate (infinity norm) on exit:', & + & ' ||r||/||b|| (inf-norm) = ',err + write(20,'(" Residual 2-norm 2 : ",es12.5)')resmx + write(20,'(" Residual inf-norm : ",es12.5)')resmxp write(20,'(a8,4(2x,a20))') 'I','X(I)','R(I)','B(I)' do i=1,m_problem write(20,998) i,x_col_glob(i),r_col_glob(i),b_col_glob(i) @@ -383,10 +581,12 @@ program mld_zf_sample 993 format(i6,4(1x,e12.6)) - call psb_gefree(b_col, desc_a,info) - call psb_gefree(x_col, desc_a,info) + call psb_gefree(b_col,desc_a,info) + call psb_gefree(x_col,desc_a,info) + call psb_gefree(r_col,desc_a,info) + call psb_gefree(ref_col,desc_a,info) call psb_spfree(a, desc_a,info) - call mld_precfree(prec,info) + call prec%free(info) call psb_cdfree(desc_a,info) call psb_exit(ictxt) @@ -399,101 +599,161 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(icontxt,mtrx,rhs,filefmt,kmethd,& - & prec, ipart,afmt,istopc,itmax,itrace,irst,eps) + subroutine get_parms(icontxt,mtrx,rhs,guess,sol,filefmt,ipart,afmt,solve,prec) use psb_base_mod implicit none - integer(psb_ipk_) :: icontxt - character(len=*) :: kmethd, mtrx, rhs, afmt,filefmt + integer(psb_ipk_) :: icontxt, ipart + character(len=*) :: mtrx, rhs, guess, sol, filefmt, afmt + type(solverdata) :: solve type(precdata) :: prec - real(psb_dpk_) :: eps - integer(psb_ipk_) :: iret, istopc,itmax,itrace, ipart, irst - integer(psb_ipk_) :: iam, nm, np, i + integer(psb_ipk_) :: iam, nm, np call psb_info(icontxt,iam,np) if (iam == psb_root_) then - ! read input parameters - call read_data(mtrx,psb_inp_unit) - call read_data(rhs,psb_inp_unit) - call read_data(filefmt,psb_inp_unit) - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(ipart,psb_inp_unit) - call read_data(istopc,psb_inp_unit) - call read_data(itmax,psb_inp_unit) - call read_data(itrace,psb_inp_unit) - call read_data(irst,psb_inp_unit) - call read_data(eps,psb_inp_unit) + ! read input data + ! + ! input files + call read_data(mtrx,psb_inp_unit) ! matrix file + call read_data(rhs,psb_inp_unit) ! rhs file + call read_data(guess,psb_inp_unit) ! starting guess file + call read_data(sol,psb_inp_unit) ! solution file (for comparison) + call read_data(filefmt,psb_inp_unit) ! format of files + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(ipart,psb_inp_unit) ! partition type + ! Krylov solver data + call read_data(solve%kmethd,psb_inp_unit) ! Krylov solver + call read_data(solve%istopc,psb_inp_unit) ! stopping criterion + call read_data(solve%itmax,psb_inp_unit) ! max num iterations + call read_data(solve%itrace,psb_inp_unit) ! tracing + call read_data(solve%irst,psb_inp_unit) ! restart + call read_data(solve%eps,psb_inp_unit) ! tolerance + ! preconditioner type call read_data(prec%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prec%prec,psb_inp_unit) ! overall prectype - call read_data(prec%novr,psb_inp_unit) ! number of overlap layers - call read_data(prec%restr,psb_inp_unit) ! restriction over application of as - call read_data(prec%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prec%solve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%fill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%thr,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%jsweeps,psb_inp_unit) ! Jacobi sweeps for PJAC - if (psb_toupper(prec%prec) == 'ML') then - call read_data(prec%nlev,psb_inp_unit) ! Number of levels in multilevel prec. - call read_data(prec%smther,psb_inp_unit) ! Smoother type. - call read_data(prec%aggrkind,psb_inp_unit) ! smoothed/raw aggregatin - call read_data(prec%aggr_alg,psb_inp_unit) ! local or global aggregation - call read_data(prec%aggr_ord,psb_inp_unit) ! Ordering for aggregation - call read_data(prec%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prec%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prec%cmat,psb_inp_unit) ! coarse mat - call read_data(prec%csolve,psb_inp_unit) ! Factorization type: BJAC, SuperLU, UMFPACK. - call read_data(prec%csbsolve,psb_inp_unit) ! Factorization type: ILU, SuperLU, UMFPACK. - call read_data(prec%cfill,psb_inp_unit) ! Fill-in for factorization - call read_data(prec%cthres,psb_inp_unit) ! Threshold for fact. ILU(T) - call read_data(prec%cjswp,psb_inp_unit) ! Jacobi sweeps - call read_data(prec%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prec%ascale,psb_inp_unit) ! smoother aggr thresh + call read_data(prec%ptype,psb_inp_unit) ! preconditioner type + ! general AMG data + call read_data(prec%mltype,psb_inp_unit) ! AMG cycle type + call read_data(prec%otr_sweeps,psb_inp_unit) ! number of AMG cycles + call read_data(prec%maxlevs,psb_inp_unit) ! max number of levels in AMG prec + call read_data(prec%csize,psb_inp_unit) ! min size coarsest mat + ! aggregation + call read_data(prec%aggrkind,psb_inp_unit) ! aggregation type + call read_data(prec%aggr_alg,psb_inp_unit) ! parallel aggregation alg + call read_data(prec%aggr_ord,psb_inp_unit) ! ordering for aggregation + call read_data(prec%aggr_filter,psb_inp_unit) ! filtering + call read_data(prec%mnaggratio,psb_inp_unit) ! minimum aggregation ratio + call read_data(prec%thrvsz,psb_inp_unit) ! size of aggr thresh vector + if (prec%thrvsz > 0) then + call psb_realloc(prec%thrvsz,prec%athresv,info) + call read_data(prec%athresv,psb_inp_unit) ! aggr thresh vector + else + read(psb_inp_unit,*) ! dummy read to skip a record end if + call read_data(prec%athres,psb_inp_unit) ! smoothed aggr thresh + call read_data(prec%aggr_omalg,psb_inp_unit) ! alg for estimating omega + call read_data(prec%aggr_eig,psb_inp_unit) ! alg for estimating omega + call read_data(prec%omega_val,psb_inp_unit) ! alg for estimating omega + ! AMG smoother (or pre-smoother) / 1-lev preconditioner + call read_data(prec%smther,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps,psb_inp_unit) ! (pre-)smoother / 1-lev prec sweeps + call read_data(prec%novr,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve,psb_inp_unit) ! local subsolver + call read_data(prec%fill,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr,psb_inp_unit) ! threshold for ILUT + ! AMG post-smoother + call read_data(prec%smther2,psb_inp_unit) ! smoother type + call read_data(prec%jsweeps2,psb_inp_unit) ! (post-)smoother sweeps + call read_data(prec%novr2,psb_inp_unit) ! number of overlap layers + call read_data(prec%restr2,psb_inp_unit) ! restriction over application of AS + call read_data(prec%prol2,psb_inp_unit) ! prolongation over application of AS + call read_data(prec%solve2,psb_inp_unit) ! local subsolver + call read_data(prec%fill2,psb_inp_unit) ! fill-in for incomplete LU + call read_data(prec%thr2,psb_inp_unit) ! threshold for ILUT + ! coasest-level solver + call read_data(prec%csolve,psb_inp_unit) ! coarsest-lev solver + call read_data(prec%cmat,psb_inp_unit) ! coarsest mat layout + call read_data(prec%csbsolve,psb_inp_unit) ! coarsest-lev subsolver + call read_data(prec%cfill,psb_inp_unit) ! fill-in for incompl LU + call read_data(prec%cthres,psb_inp_unit) ! Threshold for ILUT + call read_data(prec%cjswp,psb_inp_unit) ! sweeps for GS/JAC subsolver end if call psb_bcast(icontxt,mtrx) call psb_bcast(icontxt,rhs) + call psb_bcast(icontxt,guess) + call psb_bcast(icontxt,sol) call psb_bcast(icontxt,filefmt) - call psb_bcast(icontxt,kmethd) call psb_bcast(icontxt,afmt) call psb_bcast(icontxt,ipart) - call psb_bcast(icontxt,istopc) - call psb_bcast(icontxt,itmax) - call psb_bcast(icontxt,itrace) - call psb_bcast(icontxt,irst) - call psb_bcast(icontxt,eps) - call psb_bcast(icontxt,prec%descr) ! verbose description of the prec - call psb_bcast(icontxt,prec%prec) ! overall prectype - call psb_bcast(icontxt,prec%novr) ! number of overlap layers - call psb_bcast(icontxt,prec%restr) ! restriction over application of as - call psb_bcast(icontxt,prec%prol) ! prolongation over application of as - call psb_bcast(icontxt,prec%solve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%fill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%thr) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%jsweeps) ! Jacobi sweeps - if (psb_toupper(prec%prec) == 'ML') then - call psb_bcast(icontxt,prec%smther) ! Smoother type. - call psb_bcast(icontxt,prec%nlev) ! Number of levels in multilevel prec. - call psb_bcast(icontxt,prec%aggrkind) ! smoothed/raw aggregatin - call psb_bcast(icontxt,prec%aggr_alg) ! local or global aggregation - call psb_bcast(icontxt,prec%aggr_ord) ! Ordering for aggregation - call psb_bcast(icontxt,prec%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(icontxt,prec%smthpos) ! side: pre, post, both smoothing - call psb_bcast(icontxt,prec%cmat) ! coarse mat - call psb_bcast(icontxt,prec%csolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%csbsolve) ! Factorization type: ILU, SuperLU, UMFPACK. - call psb_bcast(icontxt,prec%cfill) ! Fill-in for factorization - call psb_bcast(icontxt,prec%cthres) ! Threshold for fact. ILU(T) - call psb_bcast(icontxt,prec%cjswp) ! Jacobi sweeps - call psb_bcast(icontxt,prec%athres) ! smoother aggr thresh - call psb_bcast(icontxt,prec%ascale) ! smoother aggr scale factor + + call psb_bcast(icontxt,solve%kmethd) + call psb_bcast(icontxt,solve%istopc) + call psb_bcast(icontxt,solve%itmax) + call psb_bcast(icontxt,solve%itrace) + call psb_bcast(icontxt,solve%irst) + call psb_bcast(icontxt,solve%eps) + + call psb_bcast(icontxt,prec%descr) + call psb_bcast(icontxt,prec%ptype) + + ! broadcast first (pre-)smoother / 1-lev prec data + call psb_bcast(icontxt,prec%smther) ! actually not needed for 1-lev precs + call psb_bcast(icontxt,prec%jsweeps) + call psb_bcast(icontxt,prec%novr) + call psb_bcast(icontxt,prec%restr) + call psb_bcast(icontxt,prec%prol) + call psb_bcast(icontxt,prec%solve) + call psb_bcast(icontxt,prec%fill) + call psb_bcast(icontxt,prec%thr) + + ! broadcast (other) AMG parameters + if (psb_toupper(prec%ptype) == 'ML') then + + call psb_bcast(icontxt,prec%mltype) + call psb_bcast(icontxt,prec%otr_sweeps) + call psb_bcast(icontxt,prec%maxlevs) + + call psb_bcast(icontxt,prec%smther2) + call psb_bcast(icontxt,prec%jsweeps2) + call psb_bcast(icontxt,prec%novr2) + call psb_bcast(icontxt,prec%restr2) + call psb_bcast(icontxt,prec%prol2) + call psb_bcast(icontxt,prec%solve2) + call psb_bcast(icontxt,prec%fill2) + call psb_bcast(icontxt,prec%thr2) + + call psb_bcast(icontxt,prec%aggrkind) + call psb_bcast(icontxt,prec%aggr_alg) + call psb_bcast(icontxt,prec%aggr_ord) + call psb_bcast(icontxt,prec%aggr_filter) + call psb_bcast(icontxt,prec%mnaggratio) + call psb_bcast(ictxt,prec%thrvsz) + if (prec%thrvsz > 0) then + if (iam /= psb_root_) call psb_realloc(prec%thrvsz,prec%athresv,info) + call psb_bcast(ictxt,prec%athresv) + end if + call psb_bcast(ictxt,prec%athres) + call psb_bcast(ictxt,prec%ascale) + call psb_bcast(ictxt,prec%aggr_omalg) + call psb_bcast(ictxt,prec%aggr_eig) + call psb_bcast(ictxt,prec%omega_val) + + call psb_bcast(icontxt,prec%csize) + call psb_bcast(icontxt,prec%cmat) + call psb_bcast(icontxt,prec%csolve) + call psb_bcast(icontxt,prec%csbsolve) + call psb_bcast(icontxt,prec%cfill) + call psb_bcast(icontxt,prec%cthres) + call psb_bcast(icontxt,prec%cjswp) + end if end subroutine get_parms + subroutine pr_usage(iout) integer(psb_ipk_) iout write(iout, *) ' number of parameters is incorrect!' @@ -511,4 +771,5 @@ contains write(iout, *) ' ptype partition strategy default 0' write(iout, *) ' 0: block partition ' end subroutine pr_usage + end program mld_zf_sample diff --git a/tests/fileread/runs/cfs.inp b/tests/fileread/runs/cfs.inp index 2e446cf2..c10c5e9d 100644 --- a/tests/fileread/runs/cfs.inp +++ b/tests/fileread/runs/cfs.inp @@ -1,35 +1,52 @@ -/opt/matrici/qc2534.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or +/opt/matrici/qc2534.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or NONE ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html -MM ! File format: MM (Matrix Market) HB (Harwell-Boeing). -BICGSTAB ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG +NONE ! Initial guess +NONE ! Reference solution +MM ! File format: MatrixMarket or Harwell-Boeing CSR ! Storage format: CSR COO JAD -0 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +2 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC -01000 ! ITMAX --1 ! ITRACE +00500 ! ITMAX +1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) -1.d-5 ! EPS -3L-M-RAS-I-D4 ! Longer descriptive name for preconditioner (up to 20 chars) +1.d-6 ! EPS +ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) ML ! Preconditioner type: NONE JACOBI BJAC AS ML +VCYCLE ! Type of multilevel correction: ADD MULT +1 ! Number of outer sweeps +-3 ! Max Number of levels in a multilevel preconditioner; if <0, lib default +-3 ! Target coarse matrix size; if <0, lib default +SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED +DEC ! Parallel aggregation: DEC, SYMDEC +NATURAL ! Ordering of aggregation NATURAL DEGREE +NOFILTER ! Filtering of matrix ? FILTER NOFILTER +-1.5 ! Coarsening ratio, if < 0 use library default +2 ! Number of thresholds in vector, next line ignored if <= 0 +0.05 0.025 ! Thresholds +0.0100d0 ! Smoothed aggregation threshold: >= 0.0 +EIG_EST ! Omega algorithm: EIG_EST USER_CHOICE +A_NORMI ! Estimate procedure if EIG_EST +1.0 ! Estimate value (if USER_CHOICE) +FBGS ! Smoother type JACOBI BJAC AS ignored for non-ML +1 ! Number of sweeps for (pre) smoother 0 ! Number of overlap layers for AS preconditioner HALO ! AS restriction operator: NONE HALO NONE ! AS prolongation operator: NONE SUM AVG -ILU ! AS subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS -1 ! Fill level P for ILU(P) and ILU(T,P) +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) 1.d-4 ! Threshold T for ILU(T,P) -1 ! Jacobi sweeps for base smoother -3 ! Number of levels in a multilevel preconditioner -AS ! Smoother type JACOBI BJAC AS ignored for non-ML -SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED -DEC ! Type of aggregation: DEC -NATURAL ! Ordering of aggregation NATURAL DEGREE -MULT ! Type of multilevel correction: ADD MULT -POST ! Side of multiplicative correction PRE POST TWOSIDE (ignored for ADD) +NONE ! Second (post) smoother, ignored if NONE +1 ! Number of sweeps for (post) smoother +0 ! Number of overlap layers for AS preconditioner +HALO ! AS restriction operator: NONE HALO +NONE ! AS prolongation operator: NONE SUM AVG +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) +1.d-4 ! Threshold T for ILU(T,P) +SLU ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS REPL ! Coarsest-level matrix distribution: DIST REPL -BJAC ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS -ILU ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST (DSCALE for JACOBI) -0 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) +SLU ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST +1 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) 1.d-4 ! Coarsest-level threshold T for ILU(T,P) -4 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver -0.0125d0 ! Smoothed aggregation threshold: >= 0.0 -0.5 ! Smoothed aggregation scaling factor. +1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver diff --git a/tests/fileread/runs/dfs.inp b/tests/fileread/runs/dfs.inp index d85b015b..7730602f 100644 --- a/tests/fileread/runs/dfs.inp +++ b/tests/fileread/runs/dfs.inp @@ -1,9 +1,11 @@ -A1M-hetero3D.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or -B1M-hetero3D.mtx ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html -MM ! -FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG +A1M-hetero3D.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or +B1M-hetero3D.mtx ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html +NONE ! Initial guess +NONE ! Reference solution +MM ! File format: MatrixMarket or Harwell-Boeing CSR ! Storage format: CSR COO JAD 2 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC 00500 ! ITMAX 1 ! ITRACE @@ -11,25 +13,40 @@ CSR ! Storage format: CSR COO JAD 1.d-6 ! EPS ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) ML ! Preconditioner type: NONE JACOBI BJAC AS ML +VCYCLE ! Type of multilevel correction: ADD MULT +1 ! Number of outer sweeps +-3 ! Max Number of levels in a multilevel preconditioner; if <0, lib default +-3 ! Target coarse matrix size; if <0, lib default +SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED +DEC ! Parallel aggregation: DEC, SYMDEC +NATURAL ! Ordering of aggregation NATURAL DEGREE +NOFILTER ! Filtering of matrix ? FILTER NOFILTER +-1.5 ! Coarsening ratio, if < 0 use library default +2 ! Number of thresholds in vector, next line ignored if <= 0 +0.05 0.025 ! Thresholds +0.0100d0 ! Smoothed aggregation threshold: >= 0.0 +EIG_EST ! Omega algorithm: EIG_EST USER_CHOICE +A_NORMI ! Estimate procedure if EIG_EST +1.0 ! Estimate value (if USER_CHOICE) +FBGS ! Smoother type JACOBI BJAC AS ignored for non-ML +1 ! Number of sweeps for (pre) smoother 0 ! Number of overlap layers for AS preconditioner HALO ! AS restriction operator: NONE HALO NONE ! AS prolongation operator: NONE SUM AVG -DSCALE ! AS subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS 0 ! Fill level P for ILU(P) and ILU(T,P) 1.d-4 ! Threshold T for ILU(T,P) -1 ! Number of Jacobi sweeps for base smoother --3 ! Number of levels in a multilevel preconditioner; if <0, lib default -FBGS ! Smoother type JACOBI BJAC AS ignored for non-ML -SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED -DEC ! Type of aggregation: DEC -NATURAL ! Ordering of aggregation NATURAL DEGREE -VCYCLE ! Type of multilevel correction: ADD MULT -TWOSIDE ! Side of multiplicative correction PRE POST TWOSIDE (ignored for ADD) +NONE ! Second (post) smoother, ignored if NONE +1 ! Number of sweeps for (post) smoother +0 ! Number of overlap layers for AS preconditioner +HALO ! AS restriction operator: NONE HALO +NONE ! AS prolongation operator: NONE SUM AVG +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) +1.d-4 ! Threshold T for ILU(T,P) +SLUDIST ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS DIST ! Coarsest-level matrix distribution: DIST REPL -MUMPS ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS -MUMPS ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST +SLUDIST ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST 1 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) 1.d-4 ! Coarsest-level threshold T for ILU(T,P) 1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver -0.0100d0 ! Smoothed aggregation threshold: >= 0.0 -1.0 ! Smoothed aggregation scaling factor. diff --git a/tests/fileread/runs/sfs.inp b/tests/fileread/runs/sfs.inp index 9dbe7aab..78495dfa 100644 --- a/tests/fileread/runs/sfs.inp +++ b/tests/fileread/runs/sfs.inp @@ -1,35 +1,52 @@ -young1r.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or -NONE ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html -MM ! File format: MM (Matrix Market) HB (Harwell-Boeing). -BICGSTAB ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG +A1M-hetero3D.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or +B1M-hetero3D.mtx ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html +NONE ! Initial guess +NONE ! Reference solution +MM ! File format: MatrixMarket or Harwell-Boeing CSR ! Storage format: CSR COO JAD -0 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +2 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC -01000 ! ITMAX --1 ! ITRACE +00500 ! ITMAX +1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) -1.d-5 ! EPS -3L-M-RAS-I-D4 ! Longer descriptive name for preconditioner (up to 20 chars) +1.d-6 ! EPS +ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) ML ! Preconditioner type: NONE JACOBI BJAC AS ML +VCYCLE ! Type of multilevel correction: ADD MULT +1 ! Number of outer sweeps +-3 ! Max Number of levels in a multilevel preconditioner; if <0, lib default +-3 ! Target coarse matrix size; if <0, lib default +SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED +DEC ! Parallel aggregation: DEC, SYMDEC +NATURAL ! Ordering of aggregation NATURAL DEGREE +NOFILTER ! Filtering of matrix ? FILTER NOFILTER +-1.5 ! Coarsening ratio, if < 0 use library default +2 ! Number of thresholds in vector, next line ignored if <= 0 +0.05 0.025 ! Thresholds +0.0100d0 ! Smoothed aggregation threshold: >= 0.0 +EIG_EST ! Omega algorithm: EIG_EST USER_CHOICE +A_NORMI ! Estimate procedure if EIG_EST +1.0 ! Estimate value (if USER_CHOICE) +FBGS ! Smoother type JACOBI BJAC AS ignored for non-ML +1 ! Number of sweeps for (pre) smoother 0 ! Number of overlap layers for AS preconditioner HALO ! AS restriction operator: NONE HALO NONE ! AS prolongation operator: NONE SUM AVG -ILU ! AS subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS -1 ! Fill level P for ILU(P) and ILU(T,P) +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) 1.d-4 ! Threshold T for ILU(T,P) -4 ! Number of Jacobi sweeps for base smoother -4 ! Number of levels in a multilevel preconditioner -BJAC ! Smoother type JACOBI BJAC AS ignored for non-ML -SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED -DEC ! Type of aggregation: DEC -NATURAL ! Ordering of aggregation NATURAL DEGREE -MULT ! Type of multilevel correction: ADD MULT -POST ! Side of multiplicative correction PRE POST TWOSIDE (ignored for ADD) +NONE ! Second (post) smoother, ignored if NONE +1 ! Number of sweeps for (post) smoother +0 ! Number of overlap layers for AS preconditioner +HALO ! AS restriction operator: NONE HALO +NONE ! AS prolongation operator: NONE SUM AVG +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) +1.d-4 ! Threshold T for ILU(T,P) +SLU ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS REPL ! Coarsest-level matrix distribution: DIST REPL -BJAC ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS -ILU ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST (DSCALE for JACOBI) -0 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) +SLU ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST +1 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) 1.d-4 ! Coarsest-level threshold T for ILU(T,P) -4 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver -0.0125d0 ! Smoothed aggregation threshold: >= 0.0 -0.5 ! Smoothed aggregation scaling factor. +1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver diff --git a/tests/fileread/runs/zfs.inp b/tests/fileread/runs/zfs.inp index 3abbed35..7730602f 100644 --- a/tests/fileread/runs/zfs.inp +++ b/tests/fileread/runs/zfs.inp @@ -1,35 +1,52 @@ -young1c.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or -NONE ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html -MM ! File format: MM (Matrix Market) HB (Harwell-Boeing). -BICGSTAB ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG +A1M-hetero3D.mtx ! This matrix (and others) from: http://math.nist.gov/MatrixMarket/ or +B1M-hetero3D.mtx ! rhs | http://www.cise.ufl.edu/research/sparse/matrices/index.html +NONE ! Initial guess +NONE ! Reference solution +MM ! File format: MatrixMarket or Harwell-Boeing CSR ! Storage format: CSR COO JAD -0 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +2 ! IPART (partition method): 0 (block) 2 (graph, with Metis) +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC -00200 ! ITMAX -02 ! ITRACE +00500 ! ITMAX +1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) -1.d-5 ! EPS -3L-M-RAS-I-D4 ! Longer descriptive name for preconditioner (up to 20 chars) +1.d-6 ! EPS +ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) ML ! Preconditioner type: NONE JACOBI BJAC AS ML +VCYCLE ! Type of multilevel correction: ADD MULT +1 ! Number of outer sweeps +-3 ! Max Number of levels in a multilevel preconditioner; if <0, lib default +-3 ! Target coarse matrix size; if <0, lib default +SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED +DEC ! Parallel aggregation: DEC, SYMDEC +NATURAL ! Ordering of aggregation NATURAL DEGREE +NOFILTER ! Filtering of matrix ? FILTER NOFILTER +-1.5 ! Coarsening ratio, if < 0 use library default +2 ! Number of thresholds in vector, next line ignored if <= 0 +0.05 0.025 ! Thresholds +0.0100d0 ! Smoothed aggregation threshold: >= 0.0 +EIG_EST ! Omega algorithm: EIG_EST USER_CHOICE +A_NORMI ! Estimate procedure if EIG_EST +1.0 ! Estimate value (if USER_CHOICE) +FBGS ! Smoother type JACOBI BJAC AS ignored for non-ML +1 ! Number of sweeps for (pre) smoother 0 ! Number of overlap layers for AS preconditioner HALO ! AS restriction operator: NONE HALO NONE ! AS prolongation operator: NONE SUM AVG -ILU ! AS subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS -1 ! Fill level P for ILU(P) and ILU(T,P) +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) 1.d-4 ! Threshold T for ILU(T,P) -4 ! Number of Jacobi sweeps for base smoother -4 ! Number of levels in a multilevel preconditioner -BJAC ! Smoother type JACOBI BJAC AS ignored for non-ML -SMOOTHED ! Type of aggregation: SMOOTHED NONSMOOTHED -DEC ! Type of aggregation: DEC -NATURAL ! Ordering of aggregation NATURAL DEGREE -MULT ! Type of multilevel correction: ADD MULT -POST ! Side of multiplicative correction PRE POST TWOSIDE (ignored for ADD) -REPL ! Coarsest-level matrix distribution: DIST REPL -BJAC ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS -ILU ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST (DSCALE for JACOBI) -0 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) +NONE ! Second (post) smoother, ignored if NONE +1 ! Number of sweeps for (post) smoother +0 ! Number of overlap layers for AS preconditioner +HALO ! AS restriction operator: NONE HALO +NONE ! AS prolongation operator: NONE SUM AVG +DSCALE ! Subdomain solver: DSCALE ILU MILU ILUT UMF SLU MUMPS +0 ! Fill level P for ILU(P) and ILU(T,P) +1.d-4 ! Threshold T for ILU(T,P) +SLUDIST ! Coarsest-level solver: JACOBI BJAC UMF SLU SLUDIST MUMPS +DIST ! Coarsest-level matrix distribution: DIST REPL +SLUDIST ! Coarsest-level subsolver: ILU UMF SLU MUMPS SLUDIST +1 ! Coarsest-level fillin P for ILU(P) and ILU(T,P) 1.d-4 ! Coarsest-level threshold T for ILU(T,P) -4 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver -0.0125d0 ! Smoothed aggregation threshold: >= 0.0 -0.5 ! Smoothed aggregation scaling factor. +1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver