From 1371fe19c0e97d1ef64392a7ac6ef30a19b94001 Mon Sep 17 00:00:00 2001 From: Salvatore Filippone Date: Tue, 30 May 2017 16:06:42 +0000 Subject: [PATCH] mld2p4-2: tests/fileread/runs/cfs.inp tests/fileread/runs/dfs.inp tests/fileread/runs/sfs.inp tests/fileread/runs/zfs.inp tests/pdegen/data_input.f90 tests/pdegen/mld_d_pde2d.f90 tests/pdegen/mld_d_pde3d.f90 tests/pdegen/mld_s_pde2d.f90 tests/pdegen/mld_s_pde3d.f90 tests/pdegen/runs/mld_pde2d.inp tests/pdegen/runs/mld_pde3d.inp Fixes for new test programs. --- tests/fileread/runs/cfs.inp | 2 +- tests/fileread/runs/dfs.inp | 2 +- tests/fileread/runs/sfs.inp | 2 +- tests/fileread/runs/zfs.inp | 2 +- tests/pdegen/data_input.f90 | 99 +++++- tests/pdegen/mld_d_pde2d.f90 | 586 ++++++++++++++++++------------- tests/pdegen/mld_d_pde3d.f90 | 587 ++++++++++++++++++------------- tests/pdegen/mld_s_pde2d.f90 | 588 ++++++++++++++++++------------- tests/pdegen/mld_s_pde3d.f90 | 589 +++++++++++++++++++------------- tests/pdegen/runs/mld_pde2d.inp | 76 +++-- tests/pdegen/runs/mld_pde3d.inp | 72 ++-- 11 files changed, 1595 insertions(+), 1010 deletions(-) diff --git a/tests/fileread/runs/cfs.inp b/tests/fileread/runs/cfs.inp index c10c5e9d..d12bd4f0 100644 --- a/tests/fileread/runs/cfs.inp +++ b/tests/fileread/runs/cfs.inp @@ -11,7 +11,7 @@ FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTA 1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) +ML-VCYCLE-FBGS-SLU ! 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 diff --git a/tests/fileread/runs/dfs.inp b/tests/fileread/runs/dfs.inp index 7730602f..d71e4a56 100644 --- a/tests/fileread/runs/dfs.inp +++ b/tests/fileread/runs/dfs.inp @@ -11,7 +11,7 @@ FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTA 1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) +ML-VCYCLE-FBGS-SLUD ! 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 diff --git a/tests/fileread/runs/sfs.inp b/tests/fileread/runs/sfs.inp index 78495dfa..8be8e324 100644 --- a/tests/fileread/runs/sfs.inp +++ b/tests/fileread/runs/sfs.inp @@ -11,7 +11,7 @@ FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTA 1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) +ML-VCYCLE-FBGS-SLU ! 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 diff --git a/tests/fileread/runs/zfs.inp b/tests/fileread/runs/zfs.inp index 7730602f..d71e4a56 100644 --- a/tests/fileread/runs/zfs.inp +++ b/tests/fileread/runs/zfs.inp @@ -11,7 +11,7 @@ FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTA 1 ! ITRACE 30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -ML-VCYCLE-FBGS-MUMPS ! Longer descriptive name for preconditioner (up to 20 chars) +ML-VCYCLE-FBGS-SLUD ! 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 diff --git a/tests/pdegen/data_input.f90 b/tests/pdegen/data_input.f90 index 72ea21b8..1e756e0e 100644 --- a/tests/pdegen/data_input.f90 +++ b/tests/pdegen/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 diff --git a/tests/pdegen/mld_d_pde2d.f90 b/tests/pdegen/mld_d_pde2d.f90 index 2d39888c..1983a921 100644 --- a/tests/pdegen/mld_d_pde2d.f90 +++ b/tests/pdegen/mld_d_pde2d.f90 @@ -1,4 +1,4 @@ -!!!$ +! ! ! MLD2P4 version 2.1 ! MultiLevel Domain Decomposition Parallel Preconditioners Package @@ -135,50 +135,91 @@ program mld_d_pde2d ! descriptor type(psb_desc_type) :: desc_a ! dense vectors - type(psb_d_vect_type) :: x,b + type(psb_d_vect_type) :: x,b,r ! parallel environment integer(psb_ipk_) :: ictxt, iam, np ! solver parameters integer(psb_ipk_) :: iter, itmax,itrace, istopc, irst, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_dpk_) :: err, eps - + real(psb_dpk_) :: err, resmx, resmxp + + ! 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 ! Solver type: ILU, SuperLU, UMFPACK. - integer(psb_ipk_) :: fill1 ! Fill-in for factorization 1 - integer(psb_ipk_) :: svsweeps ! Solver sweeps for GS - real(psb_dpk_) :: thr1 ! Threshold for fact. 1 ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: maxlevs ! Maximum number of levels in multilevel prec. - character(len=16) :: aggrprol ! smoothed/raw aggregatin - character(len=16) :: par_aggr_alg ! decoupled aggregation - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: aggr_filter ! Use filtering? - character(len=16) :: mltype ! additive or multiplicative 2nd level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - integer(psb_ipk_) :: csize ! aggregation size at which to stop. - character(len=16) :: cmat ! coarse mat - 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 ! Fill-in for factorization 1 - real(psb_dpk_) :: cthres ! Threshold for fact. 1 ILU(T) - integer(psb_ipk_) :: cjswp ! Jacobi sweeps - real(psb_dpk_) :: athres ! smoother aggregation threshold - real(psb_dpk_) :: mncrratio ! Minimum aggregation ratio + character(len=16) :: restr ! restriction over application of AS + character(len=16) :: prol ! prolongation over application of AS + 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) :: prectype - type(psb_d_coo_sparse_mat) :: acoo + type(precdata) :: p_choice + ! other variables - logical :: dump_prec - character(len=40) :: dump_prefix - integer(psb_ipk_) :: info, i + integer(psb_ipk_) :: info, i, k character(len=20) :: name,ch_err info=psb_success_ @@ -206,8 +247,7 @@ program mld_d_pde2d ! ! get parameters ! - call get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) + call get_parms(ictxt,afmt,idim,s_choice,p_choice) ! ! allocate and fill in the coefficient matrix, rhs and initial guess @@ -230,108 +270,136 @@ program mld_d_pde2d if (iam == psb_root_) & & write(psb_out_unit,'(" ")') ! - ! prepare the preconditioner. - ! - if (psb_toupper(prectype%prec) == 'ML') then - call prec%init(prectype%prec, info) - if (prectype%csize>0)& - & call prec%set('min_coarse_size', prectype%csize, info) - if (prectype%maxlevs>0)& - & call prec%set('max_levs', prectype%maxlevs, info) - if (prectype%mncrratio>0)& - & call prec%set('min_cr_ratio', prectype%mncrratio, info) - if (prectype%athres >= dzero) & - & call prec%set('aggr_thresh', prectype%athres, info) - call prec%set('aggr_prol', prectype%aggrprol,info) - call prec%set('par_aggr_alg', prectype%par_aggr_alg,info) - call prec%set('aggr_ord', prectype%aggr_ord,info) - call prec%set('aggr_filter', prectype%aggr_filter, info) + ! 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 prec%hierarchy_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + thier = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if - thier = psb_wtime()-t1 - - - call prec%set('smoother_type', prectype%smther, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call prec%set('ml_type', prectype%mltype, info) - call prec%set('smoother_pos', prectype%smthpos, info) - call prec%set('coarse_solve', prectype%csolve, info) - call prec%set('coarse_subsolve', prectype%csbsolve,info) - call prec%set('coarse_mat', prectype%cmat, info) - call prec%set('coarse_fillin', prectype%cfill, info) - call prec%set('coarse_iluthrs', prectype%cthres, info) - call prec%set('coarse_sweeps', prectype%cjswp, info) - call psb_barrier(ictxt) t1 = psb_wtime() call prec%smoothers_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if - tprec = psb_wtime()-t1 else + ! 1-level preconditioner nlv = 1 - call prec%init(prectype%prec, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call psb_barrier(ictxt) + 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 + + ! build the preconditioner thier = dzero t1 = psb_wtime() call prec%build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if - tprec = psb_wtime()-t1 end if -!!$ call prec%dump(info,prefix='test-ml',ac=.true.,solver=.true.,smoother=.true.) + call psb_amx(ictxt, thier) + call psb_amx(ictxt, tprec) - call psb_amx(ictxt,thier) - call psb_amx(ictxt,tprec) - - if (iam == psb_root_) & - & write(psb_out_unit,'("Preconditioner time : ",es12.5)') tprec+thier - if (iam == psb_root_) call mld_precdescr(prec,info) - if (iam == psb_root_) & - & write(psb_out_unit,'(" ")') + 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 ! ! iterative method parameters ! - if(iam == psb_root_) & - & write(psb_out_unit,'("Calling iterative method ",a)')kmethd call psb_barrier(ictxt) - t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b,x,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + t1 = psb_wtime() + call psb_krylov(s_choice%kmethd,a,prec,b,x,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) if(info /= psb_success_) then info=psb_err_from_subroutine_ @@ -344,33 +412,44 @@ program mld_d_pde2d tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + ! compute residual norms + call psb_geall(r,desc_a,info) + call r%zero() + call psb_geasb(r,desc_a,info) + call psb_geaxpby(done,b,dzero,r,desc_a,info) + call psb_spmm(-done,a,x,done,r,desc_a,info) + resmx = psb_genrm2(r,desc_a,info) + resmxp = psb_geamax(r,desc_a,info) + 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) - if (iam == psb_root_) then - write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Numer of levels of aggr. hierarchy: ",i12)') prec%get_nlevs() - write(psb_out_unit,'("Time to build aggr. hierarchy : ",es12.5)') thier - write(psb_out_unit,'("Time to build smoothers : ",es12.5)') tprec - write(psb_out_unit,'("Total preconditioner time : ",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,'("Number of iterations : ",i0)') iter - write(psb_out_unit,'("Convergence indicator on exit : ",es12.5)') err - write(psb_out_unit,'("Info on exit : ",i0)') info - write(psb_out_unit,'("Total memory occupation for A: ",i12)') amatsize - write(psb_out_unit,'("Storage format for A: ",a)') trim(a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for DESC_A: ",i12)') descsize - write(psb_out_unit,'("Storage format for DESC_A: ",a)') trim(desc_a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for PREC: ",i12)') precsize - end if - - if (dump_prec) call prec%dump(info,prefix=trim(dump_prefix),& - & ac=.true.,solver=.true.,smoother=.true.,rp=.true.,global_num=.true.) + call prec%descr(info) + if (iam == psb_root_) then + write(psb_out_unit,'("Computed solution on ",i8," processors")') np + 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+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() + end if ! ! cleanup storage and exit @@ -396,114 +475,153 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) - - integer(psb_ipk_) :: ictxt - type(precdata) :: prectype - character(len=*) :: kmethd, afmt - integer(psb_ipk_) :: idim, istopc,itmax,itrace,irst - integer(psb_ipk_) :: np, iam, info - real(psb_dpk_) :: eps - logical :: dump_prec - character(len=*) :: dump_prefix - character(len=20) :: buffer - - call psb_info(ictxt, iam, np) + ! + ! get iteration parameters from standard input + ! + subroutine get_parms(icontxt,afmt,idim,solve,prec) + + use psb_base_mod + implicit none + + integer(psb_ipk_) :: icontxt, idim + character(len=*) :: afmt + type(solverdata) :: solve + type(precdata) :: prec + integer(psb_ipk_) :: iam, nm, np + + call psb_info(icontxt,iam,np) if (iam == psb_root_) then - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(idim,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) - call read_data(dump_prec,psb_inp_unit) - call read_data(dump_prefix,psb_inp_unit) - call read_data(prectype%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prectype%prec,psb_inp_unit) ! overall prectype - call read_data(prectype%csize,psb_inp_unit) ! coarse size - call read_data(prectype%mncrratio,psb_inp_unit) ! Minimum aggregation ratio - call read_data(prectype%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prectype%maxlevs,psb_inp_unit) ! Maximum number of levels - call read_data(prectype%aggrprol,psb_inp_unit) ! smoothed/nonsmoothed/minenergy aggregatin - call read_data(prectype%par_aggr_alg,psb_inp_unit) ! decoupled or sym. decoupled aggregation - call read_data(prectype%aggr_ord,psb_inp_unit) ! aggregation ordering: natural, node degree - call read_data(prectype%aggr_filter,psb_inp_unit) ! aggregation filtering: filter, no_filter - call read_data(prectype%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prectype%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prectype%jsweeps,psb_inp_unit) ! Smoother sweeps - call read_data(prectype%smther,psb_inp_unit) ! Smoother type. - call read_data(prectype%novr,psb_inp_unit) ! number of overlap layers - call read_data(prectype%restr,psb_inp_unit) ! restriction over application of as - call read_data(prectype%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prectype%solve,psb_inp_unit) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call read_data(prectype%svsweeps,psb_inp_unit) ! Solver sweeps (GS) - call read_data(prectype%fill1,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%thr1,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cmat,psb_inp_unit) ! coarse mat - call read_data(prectype%csolve,psb_inp_unit) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call read_data(prectype%csbsolve,psb_inp_unit) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call read_data(prectype%cfill,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%cthres,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cjswp,psb_inp_unit) ! Jacobi sweeps + ! read input data + ! + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(idim,psb_inp_unit) ! Discretization grid size + ! 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%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 - ! broadcast parameters to all processors - call psb_bcast(ictxt,kmethd) - call psb_bcast(ictxt,afmt) - call psb_bcast(ictxt,idim) - call psb_bcast(ictxt,istopc) - call psb_bcast(ictxt,itmax) - call psb_bcast(ictxt,itrace) - call psb_bcast(ictxt,irst) - call psb_bcast(ictxt,eps) - call psb_bcast(ictxt,dump_prec) - call psb_bcast(ictxt,dump_prefix) - call psb_bcast(ictxt,prectype%descr) ! verbose description of the prec - call psb_bcast(ictxt,prectype%prec) ! overall prectype - call psb_bcast(ictxt,prectype%csize) ! coarse size - call psb_bcast(ictxt,prectype%mncrratio) ! Minimum aggregation ratio - call psb_bcast(ictxt,prectype%athres) ! smoother aggr thresh - call psb_bcast(ictxt,prectype%maxlevs) ! Maximum number of levels - call psb_bcast(ictxt,prectype%aggrprol) ! smoothed/nonsmoothed/minenergy aggregatin - call psb_bcast(ictxt,prectype%par_aggr_alg) ! decoupled or sym. decoupled aggregation - call psb_bcast(ictxt,prectype%aggr_ord) ! aggregation ordering: natural, node degree - call psb_bcast(ictxt,prectype%aggr_filter) ! aggregation filtering: filter, no_filter - call psb_bcast(ictxt,prectype%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(ictxt,prectype%smthpos) ! side: pre, post, both smoothing - call psb_bcast(ictxt,prectype%jsweeps) ! Smoother sweeps - call psb_bcast(ictxt,prectype%smther) ! Smoother type. - call psb_bcast(ictxt,prectype%novr) ! number of overlap layers - call psb_bcast(ictxt,prectype%restr) ! restriction over application of as - call psb_bcast(ictxt,prectype%prol) ! prolongation over application of as - call psb_bcast(ictxt,prectype%solve) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call psb_bcast(ictxt,prectype%svsweeps) ! Solver sweeps (GS) - call psb_bcast(ictxt,prectype%fill1) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%thr1) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cmat) ! coarse mat - call psb_bcast(ictxt,prectype%csolve) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%csbsolve) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%cfill) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%cthres) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cjswp) ! Jacobi sweeps - - - if (iam == psb_root_) then - write(psb_out_unit,'("Solving matrix : ell1")') - write(psb_out_unit,'("Grid dimensions : ",i4,"x",i4)')idim,idim - write(psb_out_unit,'("Number of processors : ",i0)') np - write(psb_out_unit,'("Data distribution : BLOCK")') - write(psb_out_unit,'("Preconditioner : ",a)') prectype%descr - write(psb_out_unit,'("Iterative method : ",a)') kmethd - write(psb_out_unit,'(" ")') - endif - - return + call psb_bcast(icontxt,afmt) + call psb_bcast(icontxt,idim) + + 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 + ! ! print an error message ! diff --git a/tests/pdegen/mld_d_pde3d.f90 b/tests/pdegen/mld_d_pde3d.f90 index b15357b4..e32e5d57 100644 --- a/tests/pdegen/mld_d_pde3d.f90 +++ b/tests/pdegen/mld_d_pde3d.f90 @@ -1,4 +1,5 @@ ! +! ! MLD2P4 version 2.1 ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) @@ -146,50 +147,91 @@ program mld_d_pde3d ! descriptor type(psb_desc_type) :: desc_a ! dense vectors - type(psb_d_vect_type) :: x,b + type(psb_d_vect_type) :: x,b,r ! parallel environment integer(psb_ipk_) :: ictxt, iam, np ! solver parameters integer(psb_ipk_) :: iter, itmax,itrace, istopc, irst, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_dpk_) :: err, eps - + real(psb_dpk_) :: err, resmx, resmxp + + ! 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 ! Solver type: ILU, SuperLU, UMFPACK. - integer(psb_ipk_) :: fill1 ! Fill-in for factorization 1 - integer(psb_ipk_) :: svsweeps ! Solver sweeps for GS - real(psb_dpk_) :: thr1 ! Threshold for fact. 1 ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: maxlevs ! Maximum number of levels in multilevel prec. - character(len=16) :: aggrprol ! smoothed/raw aggregatin - character(len=16) :: par_aggr_alg ! decoupled aggregation - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: aggr_filter ! Use filtering? - character(len=16) :: mltype ! additive or multiplicative 2nd level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - integer(psb_ipk_) :: csize ! aggregation size at which to stop. - character(len=16) :: cmat ! coarse mat - 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 ! Fill-in for factorization 1 - real(psb_dpk_) :: cthres ! Threshold for fact. 1 ILU(T) - integer(psb_ipk_) :: cjswp ! Jacobi sweeps - real(psb_dpk_) :: athres ! smoother aggregation threshold - real(psb_dpk_) :: mncrratio ! Minimum aggregation ratio + character(len=16) :: restr ! restriction over application of AS + character(len=16) :: prol ! prolongation over application of AS + 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) :: prectype - type(psb_d_coo_sparse_mat) :: acoo + type(precdata) :: p_choice + ! other variables - logical :: dump_prec - character(len=40) :: dump_prefix - integer(psb_ipk_) :: info, i + integer(psb_ipk_) :: info, i, k character(len=20) :: name,ch_err info=psb_success_ @@ -217,8 +259,7 @@ program mld_d_pde3d ! ! get parameters ! - call get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - &dump_prec,dump_prefix) + call get_parms(ictxt,afmt,idim,s_choice,p_choice) ! ! allocate and fill in the coefficient matrix, rhs and initial guess @@ -232,7 +273,7 @@ program mld_d_pde3d t2 = psb_wtime() - t1 if(info /= psb_success_) then info=psb_err_from_subroutine_ - ch_err='create_matrix' + ch_err='psb_gen_pdeed' call psb_errpush(info,name,a_err=ch_err) goto 9999 end if @@ -242,108 +283,136 @@ program mld_d_pde3d if (iam == psb_root_) & & write(psb_out_unit,'(" ")') ! - ! prepare the preconditioner. - ! - if (psb_toupper(prectype%prec) == 'ML') then - call prec%init(prectype%prec, info) - if (prectype%csize>0)& - & call prec%set('min_coarse_size', prectype%csize, info) - if (prectype%maxlevs>0)& - & call prec%set('max_levs', prectype%maxlevs, info) - if (prectype%mncrratio>0)& - & call prec%set('min_cr_ratio', prectype%mncrratio, info) - if (prectype%athres >= dzero) & - & call prec%set('aggr_thresh', prectype%athres, info) - call prec%set('aggr_prol', prectype%aggrprol,info) - call prec%set('par_aggr_alg', prectype%par_aggr_alg,info) - call prec%set('aggr_ord', prectype%aggr_ord,info) - call prec%set('aggr_filter', prectype%aggr_filter, info) + ! 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 prec%hierarchy_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + thier = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if - thier = psb_wtime()-t1 - - - call prec%set('smoother_type', prectype%smther, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call prec%set('ml_type', prectype%mltype, info) - call prec%set('smoother_pos', prectype%smthpos, info) - call prec%set('coarse_solve', prectype%csolve, info) - call prec%set('coarse_subsolve', prectype%csbsolve,info) - call prec%set('coarse_mat', prectype%cmat, info) - call prec%set('coarse_fillin', prectype%cfill, info) - call prec%set('coarse_iluthrs', prectype%cthres, info) - call prec%set('coarse_sweeps', prectype%cjswp, info) - call psb_barrier(ictxt) t1 = psb_wtime() call prec%smoothers_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if - tprec = psb_wtime()-t1 else + ! 1-level preconditioner nlv = 1 - call prec%init(prectype%prec, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call psb_barrier(ictxt) + 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 + + ! build the preconditioner thier = dzero t1 = psb_wtime() call prec%build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if - tprec = psb_wtime()-t1 end if -!!$ call prec%dump(info,prefix='test-ml',ac=.true.,solver=.true.,smoother=.true.) + call psb_amx(ictxt, thier) + call psb_amx(ictxt, tprec) - call psb_amx(ictxt,thier) - call psb_amx(ictxt,tprec) - - if (iam == psb_root_) & - & write(psb_out_unit,'("Preconditioner time : ",es12.5)') tprec+thier - if (iam == psb_root_) call mld_precdescr(prec,info) - if (iam == psb_root_) & - & write(psb_out_unit,'(" ")') + 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 ! ! iterative method parameters ! - if(iam == psb_root_) & - & write(psb_out_unit,'("Calling iterative method ",a)')kmethd call psb_barrier(ictxt) - t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b,x,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + t1 = psb_wtime() + call psb_krylov(s_choice%kmethd,a,prec,b,x,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) if(info /= psb_success_) then info=psb_err_from_subroutine_ @@ -356,33 +425,44 @@ program mld_d_pde3d tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + ! compute residual norms + call psb_geall(r,desc_a,info) + call r%zero() + call psb_geasb(r,desc_a,info) + call psb_geaxpby(done,b,dzero,r,desc_a,info) + call psb_spmm(-done,a,x,done,r,desc_a,info) + resmx = psb_genrm2(r,desc_a,info) + resmxp = psb_geamax(r,desc_a,info) + 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) - if (iam == psb_root_) then - write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Numer of levels of aggr. hierarchy: ",i12)') prec%get_nlevs() - write(psb_out_unit,'("Time to build aggr. hierarchy : ",es12.5)') thier - write(psb_out_unit,'("Time to build smoothers : ",es12.5)') tprec - write(psb_out_unit,'("Total preconditioner time : ",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,'("Number of iterations : ",i0)') iter - write(psb_out_unit,'("Convergence indicator on exit : ",es12.5)') err - write(psb_out_unit,'("Info on exit : ",i0)') info - write(psb_out_unit,'("Total memory occupation for A: ",i12)') amatsize - write(psb_out_unit,'("Storage format for A: ",a)') trim(a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for DESC_A: ",i12)') descsize - write(psb_out_unit,'("Storage format for DESC_A: ",a)') trim(desc_a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for PREC: ",i12)') precsize - end if - - if (dump_prec) call prec%dump(info,prefix=trim(dump_prefix),& - & ac=.true.,solver=.true.,smoother=.true.,rp=.true.,global_num=.true.) + call prec%descr(info) + if (iam == psb_root_) then + write(psb_out_unit,'("Computed solution on ",i8," processors")') np + 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+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() + end if ! ! cleanup storage and exit @@ -408,114 +488,153 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) - - integer(psb_ipk_) :: ictxt - type(precdata) :: prectype - character(len=*) :: kmethd, afmt - integer(psb_ipk_) :: idim, istopc,itmax,itrace,irst - integer(psb_ipk_) :: np, iam, info - real(psb_dpk_) :: eps - logical :: dump_prec - character(len=*) :: dump_prefix - character(len=20) :: buffer - - call psb_info(ictxt, iam, np) + ! + ! get iteration parameters from standard input + ! + subroutine get_parms(icontxt,afmt,idim,solve,prec) + + use psb_base_mod + implicit none + + integer(psb_ipk_) :: icontxt, idim + character(len=*) :: afmt + type(solverdata) :: solve + type(precdata) :: prec + integer(psb_ipk_) :: iam, nm, np + + call psb_info(icontxt,iam,np) if (iam == psb_root_) then - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(idim,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) - call read_data(dump_prec,psb_inp_unit) - call read_data(dump_prefix,psb_inp_unit) - call read_data(prectype%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prectype%prec,psb_inp_unit) ! overall prectype - call read_data(prectype%csize,psb_inp_unit) ! coarse size - call read_data(prectype%mncrratio,psb_inp_unit) ! Minimum aggregation ratio - call read_data(prectype%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prectype%maxlevs,psb_inp_unit) ! Maximum number of levels - call read_data(prectype%aggrprol,psb_inp_unit) ! smoothed/nonsmoothed/minenergy aggregatin - call read_data(prectype%par_aggr_alg,psb_inp_unit) ! decoupled or sym. decoupled aggregation - call read_data(prectype%aggr_ord,psb_inp_unit) ! aggregation ordering: natural, node degree - call read_data(prectype%aggr_filter,psb_inp_unit) ! aggregation filtering: filter, no_filter - call read_data(prectype%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prectype%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prectype%jsweeps,psb_inp_unit) ! Smoother sweeps - call read_data(prectype%smther,psb_inp_unit) ! Smoother type. - call read_data(prectype%novr,psb_inp_unit) ! number of overlap layers - call read_data(prectype%restr,psb_inp_unit) ! restriction over application of as - call read_data(prectype%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prectype%solve,psb_inp_unit) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call read_data(prectype%svsweeps,psb_inp_unit) ! Solver sweeps (GS) - call read_data(prectype%fill1,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%thr1,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cmat,psb_inp_unit) ! coarse mat - call read_data(prectype%csolve,psb_inp_unit) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call read_data(prectype%csbsolve,psb_inp_unit) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call read_data(prectype%cfill,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%cthres,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cjswp,psb_inp_unit) ! Jacobi sweeps + ! read input data + ! + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(idim,psb_inp_unit) ! Discretization grid size + ! 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%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 - ! broadcast parameters to all processors - call psb_bcast(ictxt,kmethd) - call psb_bcast(ictxt,afmt) - call psb_bcast(ictxt,idim) - call psb_bcast(ictxt,istopc) - call psb_bcast(ictxt,itmax) - call psb_bcast(ictxt,itrace) - call psb_bcast(ictxt,irst) - call psb_bcast(ictxt,eps) - call psb_bcast(ictxt,dump_prec) - call psb_bcast(ictxt,dump_prefix) - call psb_bcast(ictxt,prectype%descr) ! verbose description of the prec - call psb_bcast(ictxt,prectype%prec) ! overall prectype - call psb_bcast(ictxt,prectype%csize) ! coarse size - call psb_bcast(ictxt,prectype%mncrratio) ! Minimum aggregation ratio - call psb_bcast(ictxt,prectype%athres) ! smoother aggr thresh - call psb_bcast(ictxt,prectype%maxlevs) ! Maximum number of levels - call psb_bcast(ictxt,prectype%aggrprol) ! smoothed/nonsmoothed/minenergy aggregatin - call psb_bcast(ictxt,prectype%par_aggr_alg) ! decoupled or sym. decoupled aggregation - call psb_bcast(ictxt,prectype%aggr_ord) ! aggregation ordering: natural, node degree - call psb_bcast(ictxt,prectype%aggr_filter) ! aggregation filtering: filter, no_filter - call psb_bcast(ictxt,prectype%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(ictxt,prectype%smthpos) ! side: pre, post, both smoothing - call psb_bcast(ictxt,prectype%jsweeps) ! Smoother sweeps - call psb_bcast(ictxt,prectype%smther) ! Smoother type. - call psb_bcast(ictxt,prectype%novr) ! number of overlap layers - call psb_bcast(ictxt,prectype%restr) ! restriction over application of as - call psb_bcast(ictxt,prectype%prol) ! prolongation over application of as - call psb_bcast(ictxt,prectype%solve) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call psb_bcast(ictxt,prectype%svsweeps) ! Solver sweeps (GS) - call psb_bcast(ictxt,prectype%fill1) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%thr1) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cmat) ! coarse mat - call psb_bcast(ictxt,prectype%csolve) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%csbsolve) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%cfill) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%cthres) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cjswp) ! Jacobi sweeps - - - if (iam == psb_root_) then - write(psb_out_unit,'("Solving matrix : ell1")') - write(psb_out_unit,'("Grid dimensions : ",i4,"x",i4,"x",i4)')idim,idim,idim - write(psb_out_unit,'("Number of processors : ",i0)') np - write(psb_out_unit,'("Data distribution : BLOCK")') - write(psb_out_unit,'("Preconditioner : ",a)') prectype%descr - write(psb_out_unit,'("Iterative method : ",a)') kmethd - write(psb_out_unit,'(" ")') - endif - - return + call psb_bcast(icontxt,afmt) + call psb_bcast(icontxt,idim) + + 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 + ! ! print an error message ! diff --git a/tests/pdegen/mld_s_pde2d.f90 b/tests/pdegen/mld_s_pde2d.f90 index f8fa9e8a..ba5a6097 100644 --- a/tests/pdegen/mld_s_pde2d.f90 +++ b/tests/pdegen/mld_s_pde2d.f90 @@ -1,4 +1,4 @@ -!!!$ +! ! ! MLD2P4 version 2.1 ! MultiLevel Domain Decomposition Parallel Preconditioners Package @@ -135,50 +135,91 @@ program mld_s_pde2d ! descriptor type(psb_desc_type) :: desc_a ! dense vectors - type(psb_s_vect_type) :: x,b + type(psb_s_vect_type) :: x,b,r ! parallel environment integer(psb_ipk_) :: ictxt, iam, np ! solver parameters integer(psb_ipk_) :: iter, itmax,itrace, istopc, irst, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_spk_) :: err, eps - + real(psb_spk_) :: err, resmx, resmxp + + ! 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 ! Solver type: ILU, SuperLU, UMFPACK. - integer(psb_ipk_) :: fill1 ! Fill-in for factorization 1 - integer(psb_ipk_) :: svsweeps ! Solver sweeps for GS - real(psb_spk_) :: thr1 ! Threshold for fact. 1 ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: maxlevs ! Maximum number of levels in multilevel prec. - character(len=16) :: aggrprol ! smoothed/raw aggregatin - character(len=16) :: par_aggr_alg ! decoupled aggregation - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: aggr_filter ! Use filtering? - character(len=16) :: mltype ! additive or multiplicative 2nd level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - integer(psb_ipk_) :: csize ! aggregation size at which to stop. - character(len=16) :: cmat ! coarse mat - 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 ! Fill-in for factorization 1 - real(psb_spk_) :: cthres ! Threshold for fact. 1 ILU(T) - integer(psb_ipk_) :: cjswp ! Jacobi sweeps - real(psb_spk_) :: athres ! smoother aggregation threshold - real(psb_spk_) :: mncrratio ! Minimum aggregation ratio + character(len=16) :: restr ! restriction over application of AS + character(len=16) :: prol ! prolongation over application of AS + 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) :: prectype - type(psb_s_coo_sparse_mat) :: acoo + type(precdata) :: p_choice + ! other variables - logical :: dump_prec - character(len=40) :: dump_prefix - integer(psb_ipk_) :: info, i + integer(psb_ipk_) :: info, i, k character(len=20) :: name,ch_err info=psb_success_ @@ -206,8 +247,7 @@ program mld_s_pde2d ! ! get parameters ! - call get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) + call get_parms(ictxt,afmt,idim,s_choice,p_choice) ! ! allocate and fill in the coefficient matrix, rhs and initial guess @@ -230,108 +270,136 @@ program mld_s_pde2d if (iam == psb_root_) & & write(psb_out_unit,'(" ")') ! - ! prepare the preconditioner. - ! - if (psb_toupper(prectype%prec) == 'ML') then - call prec%init(prectype%prec, info) - if (prectype%csize>0)& - & call prec%set('min_coarse_size', prectype%csize, info) - if (prectype%maxlevs>0)& - & call prec%set('max_levs', prectype%maxlevs, info) - if (prectype%mncrratio>0)& - & call prec%set('min_cr_ratio', prectype%mncrratio, info) - if (prectype%athres >= szero) & - & call prec%set('aggr_thresh', prectype%athres, info) - call prec%set('aggr_prol', prectype%aggrprol,info) - call prec%set('par_aggr_alg', prectype%par_aggr_alg,info) - call prec%set('aggr_ord', prectype%aggr_ord,info) - call prec%set('aggr_filter', prectype%aggr_filter, info) + ! 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 prec%hierarchy_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + thier = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if - thier = psb_wtime()-t1 - - - call prec%set('smoother_type', prectype%smther, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call prec%set('ml_type', prectype%mltype, info) - call prec%set('smoother_pos', prectype%smthpos, info) - call prec%set('coarse_solve', prectype%csolve, info) - call prec%set('coarse_subsolve', prectype%csbsolve,info) - call prec%set('coarse_mat', prectype%cmat, info) - call prec%set('coarse_fillin', prectype%cfill, info) - call prec%set('coarse_iluthrs', prectype%cthres, info) - call prec%set('coarse_sweeps', prectype%cjswp, info) - call psb_barrier(ictxt) t1 = psb_wtime() call prec%smoothers_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if - tprec = psb_wtime()-t1 else + ! 1-level preconditioner nlv = 1 - call prec%init(prectype%prec, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call psb_barrier(ictxt) - thier = szero + 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 + + ! build the preconditioner + thier = dzero t1 = psb_wtime() call prec%build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if - tprec = psb_wtime()-t1 end if -!!$ call prec%dump(info,prefix='test-ml',ac=.true.,solver=.true.,smoother=.true.) + call psb_amx(ictxt, thier) + call psb_amx(ictxt, tprec) - call psb_amx(ictxt,thier) - call psb_amx(ictxt,tprec) - - if (iam == psb_root_) & - & write(psb_out_unit,'("Preconditioner time : ",es12.5)') tprec+thier - if (iam == psb_root_) call mld_precdescr(prec,info) - if (iam == psb_root_) & - & write(psb_out_unit,'(" ")') + 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 ! ! iterative method parameters ! - if(iam == psb_root_) & - & write(psb_out_unit,'("Calling iterative method ",a)')kmethd call psb_barrier(ictxt) - t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b,x,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + t1 = psb_wtime() + call psb_krylov(s_choice%kmethd,a,prec,b,x,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) if(info /= psb_success_) then info=psb_err_from_subroutine_ @@ -344,33 +412,44 @@ program mld_s_pde2d tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + ! compute residual norms + call psb_geall(r,desc_a,info) + call r%zero() + call psb_geasb(r,desc_a,info) + call psb_geaxpby(sone,b,szero,r,desc_a,info) + call psb_spmm(-sone,a,x,sone,r,desc_a,info) + resmx = psb_genrm2(r,desc_a,info) + resmxp = psb_geamax(r,desc_a,info) + 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) - if (iam == psb_root_) then - write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Numer of levels of aggr. hierarchy: ",i12)') prec%get_nlevs() - write(psb_out_unit,'("Time to build aggr. hierarchy : ",es12.5)') thier - write(psb_out_unit,'("Time to build smoothers : ",es12.5)') tprec - write(psb_out_unit,'("Total preconditioner time : ",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,'("Number of iterations : ",i0)') iter - write(psb_out_unit,'("Convergence indicator on exit : ",es12.5)') err - write(psb_out_unit,'("Info on exit : ",i0)') info - write(psb_out_unit,'("Total memory occupation for A: ",i12)') amatsize - write(psb_out_unit,'("Storage format for A: ",a)') trim(a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for DESC_A: ",i12)') descsize - write(psb_out_unit,'("Storage format for DESC_A: ",a)') trim(desc_a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for PREC: ",i12)') precsize - end if - - if (dump_prec) call prec%dump(info,prefix=trim(dump_prefix),& - & ac=.true.,solver=.true.,smoother=.true.,rp=.true.,global_num=.true.) + call prec%descr(info) + if (iam == psb_root_) then + write(psb_out_unit,'("Computed solution on ",i8," processors")') np + 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+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() + end if ! ! cleanup storage and exit @@ -396,114 +475,153 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) - - integer(psb_ipk_) :: ictxt - type(precdata) :: prectype - character(len=*) :: kmethd, afmt - integer(psb_ipk_) :: idim, istopc,itmax,itrace,irst - integer(psb_ipk_) :: np, iam, info - real(psb_spk_) :: eps - logical :: dump_prec - character(len=*) :: dump_prefix - character(len=20) :: buffer - - call psb_info(ictxt, iam, np) + ! + ! get iteration parameters from standard input + ! + subroutine get_parms(icontxt,afmt,idim,solve,prec) + + use psb_base_mod + implicit none + + integer(psb_ipk_) :: icontxt, idim + character(len=*) :: afmt + type(solverdata) :: solve + type(precdata) :: prec + integer(psb_ipk_) :: iam, nm, np + + call psb_info(icontxt,iam,np) if (iam == psb_root_) then - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(idim,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) - call read_data(dump_prec,psb_inp_unit) - call read_data(dump_prefix,psb_inp_unit) - call read_data(prectype%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prectype%prec,psb_inp_unit) ! overall prectype - call read_data(prectype%csize,psb_inp_unit) ! coarse size - call read_data(prectype%mncrratio,psb_inp_unit) ! Minimum aggregation ratio - call read_data(prectype%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prectype%maxlevs,psb_inp_unit) ! Maximum number of levels - call read_data(prectype%aggrprol,psb_inp_unit) ! smoothed/nonsmoothed/minenergy aggregatin - call read_data(prectype%par_aggr_alg,psb_inp_unit) ! decoupled or sym. decoupled aggregation - call read_data(prectype%aggr_ord,psb_inp_unit) ! aggregation ordering: natural, node degree - call read_data(prectype%aggr_filter,psb_inp_unit) ! aggregation filtering: filter, no_filter - call read_data(prectype%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prectype%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prectype%jsweeps,psb_inp_unit) ! Smoother sweeps - call read_data(prectype%smther,psb_inp_unit) ! Smoother type. - call read_data(prectype%novr,psb_inp_unit) ! number of overlap layers - call read_data(prectype%restr,psb_inp_unit) ! restriction over application of as - call read_data(prectype%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prectype%solve,psb_inp_unit) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call read_data(prectype%svsweeps,psb_inp_unit) ! Solver sweeps (GS) - call read_data(prectype%fill1,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%thr1,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cmat,psb_inp_unit) ! coarse mat - call read_data(prectype%csolve,psb_inp_unit) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call read_data(prectype%csbsolve,psb_inp_unit) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call read_data(prectype%cfill,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%cthres,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cjswp,psb_inp_unit) ! Jacobi sweeps + ! read input data + ! + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(idim,psb_inp_unit) ! Discretization grid size + ! 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%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 - ! broadcast parameters to all processors - call psb_bcast(ictxt,kmethd) - call psb_bcast(ictxt,afmt) - call psb_bcast(ictxt,idim) - call psb_bcast(ictxt,istopc) - call psb_bcast(ictxt,itmax) - call psb_bcast(ictxt,itrace) - call psb_bcast(ictxt,irst) - call psb_bcast(ictxt,eps) - call psb_bcast(ictxt,dump_prec) - call psb_bcast(ictxt,dump_prefix) - call psb_bcast(ictxt,prectype%descr) ! verbose description of the prec - call psb_bcast(ictxt,prectype%prec) ! overall prectype - call psb_bcast(ictxt,prectype%csize) ! coarse size - call psb_bcast(ictxt,prectype%mncrratio) ! Minimum aggregation ratio - call psb_bcast(ictxt,prectype%athres) ! smoother aggr thresh - call psb_bcast(ictxt,prectype%maxlevs) ! Maximum number of levels - call psb_bcast(ictxt,prectype%aggrprol) ! smoothed/nonsmoothed/minenergy aggregatin - call psb_bcast(ictxt,prectype%par_aggr_alg) ! decoupled or sym. decoupled aggregation - call psb_bcast(ictxt,prectype%aggr_ord) ! aggregation ordering: natural, node degree - call psb_bcast(ictxt,prectype%aggr_filter) ! aggregation filtering: filter, no_filter - call psb_bcast(ictxt,prectype%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(ictxt,prectype%smthpos) ! side: pre, post, both smoothing - call psb_bcast(ictxt,prectype%jsweeps) ! Smoother sweeps - call psb_bcast(ictxt,prectype%smther) ! Smoother type. - call psb_bcast(ictxt,prectype%novr) ! number of overlap layers - call psb_bcast(ictxt,prectype%restr) ! restriction over application of as - call psb_bcast(ictxt,prectype%prol) ! prolongation over application of as - call psb_bcast(ictxt,prectype%solve) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call psb_bcast(ictxt,prectype%svsweeps) ! Solver sweeps (GS) - call psb_bcast(ictxt,prectype%fill1) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%thr1) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cmat) ! coarse mat - call psb_bcast(ictxt,prectype%csolve) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%csbsolve) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%cfill) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%cthres) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cjswp) ! Jacobi sweeps - - - if (iam == psb_root_) then - write(psb_out_unit,'("Solving matrix : ell1")') - write(psb_out_unit,'("Grid dimensions : ",i4,"x",i4)')idim,idim - write(psb_out_unit,'("Number of processors : ",i0)') np - write(psb_out_unit,'("Data distribution : BLOCK")') - write(psb_out_unit,'("Preconditioner : ",a)') prectype%descr - write(psb_out_unit,'("Iterative method : ",a)') kmethd - write(psb_out_unit,'(" ")') - endif - - return + call psb_bcast(icontxt,afmt) + call psb_bcast(icontxt,idim) + + 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 + ! ! print an error message ! diff --git a/tests/pdegen/mld_s_pde3d.f90 b/tests/pdegen/mld_s_pde3d.f90 index 05c41e7b..22b806e2 100644 --- a/tests/pdegen/mld_s_pde3d.f90 +++ b/tests/pdegen/mld_s_pde3d.f90 @@ -1,4 +1,5 @@ ! +! ! MLD2P4 version 2.1 ! MultiLevel Domain Decomposition Parallel Preconditioners Package ! based on PSBLAS (Parallel Sparse BLAS version 3.5) @@ -146,50 +147,91 @@ program mld_s_pde3d ! descriptor type(psb_desc_type) :: desc_a ! dense vectors - type(psb_s_vect_type) :: x,b + type(psb_s_vect_type) :: x,b,r ! parallel environment integer(psb_ipk_) :: ictxt, iam, np ! solver parameters integer(psb_ipk_) :: iter, itmax,itrace, istopc, irst, nlv integer(psb_long_int_k_) :: amatsize, precsize, descsize - real(psb_spk_) :: err, eps - + real(psb_spk_) :: err, resmx, resmxp + + ! 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 ! Solver type: ILU, SuperLU, UMFPACK. - integer(psb_ipk_) :: fill1 ! Fill-in for factorization 1 - integer(psb_ipk_) :: svsweeps ! Solver sweeps for GS - real(psb_spk_) :: thr1 ! Threshold for fact. 1 ILU(T) - character(len=16) :: smther ! Smoother - integer(psb_ipk_) :: maxlevs ! Maximum number of levels in multilevel prec. - character(len=16) :: aggrprol ! smoothed/raw aggregatin - character(len=16) :: par_aggr_alg ! decoupled aggregation - character(len=16) :: aggr_ord ! Ordering for aggregation - character(len=16) :: aggr_filter ! Use filtering? - character(len=16) :: mltype ! additive or multiplicative 2nd level prec - character(len=16) :: smthpos ! side: pre, post, both smoothing - integer(psb_ipk_) :: csize ! aggregation size at which to stop. - character(len=16) :: cmat ! coarse mat - 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 ! Fill-in for factorization 1 - real(psb_spk_) :: cthres ! Threshold for fact. 1 ILU(T) - integer(psb_ipk_) :: cjswp ! Jacobi sweeps - real(psb_spk_) :: athres ! smoother aggregation threshold - real(psb_spk_) :: mncrratio ! Minimum aggregation ratio + character(len=16) :: restr ! restriction over application of AS + character(len=16) :: prol ! prolongation over application of AS + 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) :: prectype - type(psb_s_coo_sparse_mat) :: acoo + type(precdata) :: p_choice + ! other variables - logical :: dump_prec - character(len=40) :: dump_prefix - integer(psb_ipk_) :: info, i + integer(psb_ipk_) :: info, i, k character(len=20) :: name,ch_err info=psb_success_ @@ -217,8 +259,7 @@ program mld_s_pde3d ! ! get parameters ! - call get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - &dump_prec,dump_prefix) + call get_parms(ictxt,afmt,idim,s_choice,p_choice) ! ! allocate and fill in the coefficient matrix, rhs and initial guess @@ -232,7 +273,7 @@ program mld_s_pde3d t2 = psb_wtime() - t1 if(info /= psb_success_) then info=psb_err_from_subroutine_ - ch_err='create_matrix' + ch_err='psb_gen_pdeed' call psb_errpush(info,name,a_err=ch_err) goto 9999 end if @@ -242,108 +283,136 @@ program mld_s_pde3d if (iam == psb_root_) & & write(psb_out_unit,'(" ")') ! - ! prepare the preconditioner. - ! - if (psb_toupper(prectype%prec) == 'ML') then - call prec%init(prectype%prec, info) - if (prectype%csize>0)& - & call prec%set('min_coarse_size', prectype%csize, info) - if (prectype%maxlevs>0)& - & call prec%set('max_levs', prectype%maxlevs, info) - if (prectype%mncrratio>0)& - & call prec%set('min_cr_ratio', prectype%mncrratio, info) - if (prectype%athres >= szero) & - & call prec%set('aggr_thresh', prectype%athres, info) - call prec%set('aggr_prol', prectype%aggrprol,info) - call prec%set('par_aggr_alg', prectype%par_aggr_alg,info) - call prec%set('aggr_ord', prectype%aggr_ord,info) - call prec%set('aggr_filter', prectype%aggr_filter, info) + ! 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 prec%hierarchy_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + thier = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_hierarchy_bld') goto 9999 end if - thier = psb_wtime()-t1 - - - call prec%set('smoother_type', prectype%smther, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call prec%set('ml_type', prectype%mltype, info) - call prec%set('smoother_pos', prectype%smthpos, info) - call prec%set('coarse_solve', prectype%csolve, info) - call prec%set('coarse_subsolve', prectype%csbsolve,info) - call prec%set('coarse_mat', prectype%cmat, info) - call prec%set('coarse_fillin', prectype%cfill, info) - call prec%set('coarse_iluthrs', prectype%cthres, info) - call prec%set('coarse_sweeps', prectype%cjswp, info) - call psb_barrier(ictxt) t1 = psb_wtime() call prec%smoothers_build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_smoothers_bld') goto 9999 end if - tprec = psb_wtime()-t1 else + ! 1-level preconditioner nlv = 1 - call prec%init(prectype%prec, info) - call prec%set('smoother_sweeps', prectype%jsweeps, info) - call prec%set('sub_ovr', prectype%novr, info) - call prec%set('sub_restr', prectype%restr, info) - call prec%set('sub_prol', prectype%prol, info) - call prec%set('sub_solve', prectype%solve, info) - call prec%set('sub_fillin', prectype%fill1, info) - call prec%set('solver_sweeps', prectype%svsweeps, info) - call prec%set('sub_iluthrs', prectype%thr1, info) - call psb_barrier(ictxt) - thier = szero + 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 + + ! build the preconditioner + thier = dzero t1 = psb_wtime() call prec%build(a,desc_a,info) - if(info /= psb_success_) then - info=psb_err_from_subroutine_ - ch_err='psb_precbld' - call psb_errpush(info,name,a_err=ch_err) + tprec = psb_wtime()-t1 + if (info /= psb_success_) then + call psb_errpush(psb_err_from_subroutine_,name,a_err='mld_precbld') goto 9999 end if - tprec = psb_wtime()-t1 end if -!!$ call prec%dump(info,prefix='test-ml',ac=.true.,solver=.true.,smoother=.true.) + call psb_amx(ictxt, thier) + call psb_amx(ictxt, tprec) - call psb_amx(ictxt,thier) - call psb_amx(ictxt,tprec) - - if (iam == psb_root_) & - & write(psb_out_unit,'("Preconditioner time : ",es12.5)') tprec+thier - if (iam == psb_root_) call mld_precdescr(prec,info) - if (iam == psb_root_) & - & write(psb_out_unit,'(" ")') + 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 ! ! iterative method parameters ! - if(iam == psb_root_) & - & write(psb_out_unit,'("Calling iterative method ",a)')kmethd call psb_barrier(ictxt) - t1 = psb_wtime() - call psb_krylov(kmethd,a,prec,b,x,eps,desc_a,info,& - & itmax=itmax,iter=iter,err=err,itrace=itrace,istop=istopc,irst=irst) + t1 = psb_wtime() + call psb_krylov(s_choice%kmethd,a,prec,b,x,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) if(info /= psb_success_) then info=psb_err_from_subroutine_ @@ -356,33 +425,44 @@ program mld_s_pde3d tslv = psb_wtime() - t1 call psb_amx(ictxt,tslv) + ! compute residual norms + call psb_geall(r,desc_a,info) + call r%zero() + call psb_geasb(r,desc_a,info) + call psb_geaxpby(sone,b,szero,r,desc_a,info) + call psb_spmm(-sone,a,x,sone,r,desc_a,info) + resmx = psb_genrm2(r,desc_a,info) + resmxp = psb_geamax(r,desc_a,info) + 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) - if (iam == psb_root_) then - write(psb_out_unit,'(" ")') - write(psb_out_unit,'("Numer of levels of aggr. hierarchy: ",i12)') prec%get_nlevs() - write(psb_out_unit,'("Time to build aggr. hierarchy : ",es12.5)') thier - write(psb_out_unit,'("Time to build smoothers : ",es12.5)') tprec - write(psb_out_unit,'("Total preconditioner time : ",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,'("Number of iterations : ",i0)') iter - write(psb_out_unit,'("Convergence indicator on exit : ",es12.5)') err - write(psb_out_unit,'("Info on exit : ",i0)') info - write(psb_out_unit,'("Total memory occupation for A: ",i12)') amatsize - write(psb_out_unit,'("Storage format for A: ",a)') trim(a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for DESC_A: ",i12)') descsize - write(psb_out_unit,'("Storage format for DESC_A: ",a)') trim(desc_a%get_fmt()) - write(psb_out_unit,'("Total memory occupation for PREC: ",i12)') precsize - end if - - if (dump_prec) call prec%dump(info,prefix=trim(dump_prefix),& - & ac=.true.,solver=.true.,smoother=.true.,rp=.true.,global_num=.true.) + call prec%descr(info) + if (iam == psb_root_) then + write(psb_out_unit,'("Computed solution on ",i8," processors")') np + 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+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() + end if ! ! cleanup storage and exit @@ -408,114 +488,153 @@ contains ! ! get iteration parameters from standard input ! - subroutine get_parms(ictxt,kmethd,prectype,afmt,idim,istopc,itmax,itrace,irst,eps,& - & dump_prec,dump_prefix) - - integer(psb_ipk_) :: ictxt - type(precdata) :: prectype - character(len=*) :: kmethd, afmt - integer(psb_ipk_) :: idim, istopc,itmax,itrace,irst - integer(psb_ipk_) :: np, iam, info - real(psb_spk_) :: eps - logical :: dump_prec - character(len=*) :: dump_prefix - character(len=20) :: buffer - - call psb_info(ictxt, iam, np) + ! + ! get iteration parameters from standard input + ! + subroutine get_parms(icontxt,afmt,idim,solve,prec) + + use psb_base_mod + implicit none + + integer(psb_ipk_) :: icontxt, idim + character(len=*) :: afmt + type(solverdata) :: solve + type(precdata) :: prec + integer(psb_ipk_) :: iam, nm, np + + call psb_info(icontxt,iam,np) if (iam == psb_root_) then - call read_data(kmethd,psb_inp_unit) - call read_data(afmt,psb_inp_unit) - call read_data(idim,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) - call read_data(dump_prec,psb_inp_unit) - call read_data(dump_prefix,psb_inp_unit) - call read_data(prectype%descr,psb_inp_unit) ! verbose description of the prec - call read_data(prectype%prec,psb_inp_unit) ! overall prectype - call read_data(prectype%csize,psb_inp_unit) ! coarse size - call read_data(prectype%mncrratio,psb_inp_unit) ! Minimum aggregation ratio - call read_data(prectype%athres,psb_inp_unit) ! smoother aggr thresh - call read_data(prectype%maxlevs,psb_inp_unit) ! Maximum number of levels - call read_data(prectype%aggrprol,psb_inp_unit) ! smoothed/nonsmoothed/minenergy aggregatin - call read_data(prectype%par_aggr_alg,psb_inp_unit) ! decoupled or sym. decoupled aggregation - call read_data(prectype%aggr_ord,psb_inp_unit) ! aggregation ordering: natural, node degree - call read_data(prectype%aggr_filter,psb_inp_unit) ! aggregation filtering: filter, no_filter - call read_data(prectype%mltype,psb_inp_unit) ! additive or multiplicative 2nd level prec - call read_data(prectype%smthpos,psb_inp_unit) ! side: pre, post, both smoothing - call read_data(prectype%jsweeps,psb_inp_unit) ! Smoother sweeps - call read_data(prectype%smther,psb_inp_unit) ! Smoother type. - call read_data(prectype%novr,psb_inp_unit) ! number of overlap layers - call read_data(prectype%restr,psb_inp_unit) ! restriction over application of as - call read_data(prectype%prol,psb_inp_unit) ! prolongation over application of as - call read_data(prectype%solve,psb_inp_unit) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call read_data(prectype%svsweeps,psb_inp_unit) ! Solver sweeps (GS) - call read_data(prectype%fill1,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%thr1,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cmat,psb_inp_unit) ! coarse mat - call read_data(prectype%csolve,psb_inp_unit) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call read_data(prectype%csbsolve,psb_inp_unit) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call read_data(prectype%cfill,psb_inp_unit) ! Fill-in for factorization 1 - call read_data(prectype%cthres,psb_inp_unit) ! Threshold for fact. 1 ILU(T) - call read_data(prectype%cjswp,psb_inp_unit) ! Jacobi sweeps + ! read input data + ! + call read_data(afmt,psb_inp_unit) ! matrix storage format + call read_data(idim,psb_inp_unit) ! Discretization grid size + ! 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%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 - ! broadcast parameters to all processors - call psb_bcast(ictxt,kmethd) - call psb_bcast(ictxt,afmt) - call psb_bcast(ictxt,idim) - call psb_bcast(ictxt,istopc) - call psb_bcast(ictxt,itmax) - call psb_bcast(ictxt,itrace) - call psb_bcast(ictxt,irst) - call psb_bcast(ictxt,eps) - call psb_bcast(ictxt,dump_prec) - call psb_bcast(ictxt,dump_prefix) - call psb_bcast(ictxt,prectype%descr) ! verbose description of the prec - call psb_bcast(ictxt,prectype%prec) ! overall prectype - call psb_bcast(ictxt,prectype%csize) ! coarse size - call psb_bcast(ictxt,prectype%mncrratio) ! Minimum aggregation ratio - call psb_bcast(ictxt,prectype%athres) ! smoother aggr thresh - call psb_bcast(ictxt,prectype%maxlevs) ! Maximum number of levels - call psb_bcast(ictxt,prectype%aggrprol) ! smoothed/nonsmoothed/minenergy aggregatin - call psb_bcast(ictxt,prectype%par_aggr_alg) ! decoupled or sym. decoupled aggregation - call psb_bcast(ictxt,prectype%aggr_ord) ! aggregation ordering: natural, node degree - call psb_bcast(ictxt,prectype%aggr_filter) ! aggregation filtering: filter, no_filter - call psb_bcast(ictxt,prectype%mltype) ! additive or multiplicative 2nd level prec - call psb_bcast(ictxt,prectype%smthpos) ! side: pre, post, both smoothing - call psb_bcast(ictxt,prectype%jsweeps) ! Smoother sweeps - call psb_bcast(ictxt,prectype%smther) ! Smoother type. - call psb_bcast(ictxt,prectype%novr) ! number of overlap layers - call psb_bcast(ictxt,prectype%restr) ! restriction over application of as - call psb_bcast(ictxt,prectype%prol) ! prolongation over application of as - call psb_bcast(ictxt,prectype%solve) ! Subdomain solver: DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU - call psb_bcast(ictxt,prectype%svsweeps) ! Solver sweeps (GS) - call psb_bcast(ictxt,prectype%fill1) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%thr1) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cmat) ! coarse mat - call psb_bcast(ictxt,prectype%csolve) ! Coarse solver: JACOBI BJAC UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%csbsolve) ! subsolver: DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS - call psb_bcast(ictxt,prectype%cfill) ! Fill-in for factorization 1 - call psb_bcast(ictxt,prectype%cthres) ! Threshold for fact. 1 ILU(T) - call psb_bcast(ictxt,prectype%cjswp) ! Jacobi sweeps - - - if (iam == psb_root_) then - write(psb_out_unit,'("Solving matrix : ell1")') - write(psb_out_unit,'("Grid dimensions : ",i4,"x",i4,"x",i4)')idim,idim,idim - write(psb_out_unit,'("Number of processors : ",i0)') np - write(psb_out_unit,'("Data distribution : BLOCK")') - write(psb_out_unit,'("Preconditioner : ",a)') prectype%descr - write(psb_out_unit,'("Iterative method : ",a)') kmethd - write(psb_out_unit,'(" ")') - endif - - return + call psb_bcast(icontxt,afmt) + call psb_bcast(icontxt,idim) + + 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 + ! ! print an error message ! diff --git a/tests/pdegen/runs/mld_pde2d.inp b/tests/pdegen/runs/mld_pde2d.inp index 182f1a44..a1156ac4 100644 --- a/tests/pdegen/runs/mld_pde2d.inp +++ b/tests/pdegen/runs/mld_pde2d.inp @@ -1,37 +1,47 @@ -BICGSTAB ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG CSR ! Storage format CSR COO JAD -0100 ! IDIM; domain size is idim**2 +0080 ! IDIM; domain size is idim**3 +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC -2000 ! ITMAX -10 ! ITRACE -30 ! IRST (restart for RGMRES and BiCGSTABL) +00500 ! ITMAX +1 ! ITRACE +30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -F ! Dump preconditioner on file T F -test-ml-unsm-our ! File prefix for preconditioner dump -ML-MUL-RAS-ILU ! Descriptive name for preconditioner (up to 40 chars) -ML ! Preconditioner NONE JACOBI BJAC AS ML --8000 ! If ML: Target coarse size. If <0, then use library default --1.5d0 ! If ML: Minimum aggregation ratio; if <0 use library default --0.08d0 ! If ML: Smoother Aggregation Threshold: >= 0.0 default if <0 --20 ! If ML: Maximum acceptable number of levels; if <0 use library default -SMOOTHED ! Type of aggregation: SMOOTHED, UNSMOOTHED, MINENERGY -DEC ! Type of aggregation: DEC SYMDEC -NATURAL ! Ordering of aggregation: NATURAL DEGREE -FILTER ! Filtering aggregation: FILTER NOFILTER -MULT ! Type of multilevel correction: ADD MULT KCYCLE VCYCLE WCYCLE KCYCLESYM -TWOSIDE ! Side of correction: PRE POST TWOSIDE (ignored for ADD) -2 ! Smoother sweeps -BJAC ! Smoother type JACOBI BJAC AS; ignored for non-ML -0 ! Number of overlap layers for AS preconditioner (at finest level) -HALO ! AS Restriction operator NONE HALO -NONE ! AS Prolongation operator NONE SUM AVG -ILU ! Subdomain solver DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU -1 ! Solver sweeps for GS -0 ! Level-set N for ILU(N), and P for ILUT +ML-VCYCLE-FBGS-SLUD ! 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 ! 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) -REPL ! Coarse level: matrix distribution DIST REPL -SLU ! Coarse level: solver JACOBI BJAC UMF SLU SLUDIST MUMPS -SLU ! Coarse level: subsolver DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS -1 ! Coarse level: Level-set N for ILU(N) -1.d-4 ! Coarse level: Threshold T for ILU(T,P) -2 ! Coarse level: Number of Jacobi sweeps +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) +1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver diff --git a/tests/pdegen/runs/mld_pde3d.inp b/tests/pdegen/runs/mld_pde3d.inp index 89867afb..a1156ac4 100644 --- a/tests/pdegen/runs/mld_pde3d.inp +++ b/tests/pdegen/runs/mld_pde3d.inp @@ -1,37 +1,47 @@ -BICGSTAB ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG CSR ! Storage format CSR COO JAD 0080 ! IDIM; domain size is idim**3 +FCG ! Iterative method: BiCGSTAB BiCG CGS RGMRES BiCGSTABL CG 2 ! ISTOPC -2000 ! ITMAX +00500 ! ITMAX 1 ! ITRACE -30 ! IRST (restart for RGMRES and BiCGSTABL) +30 ! IRST (restart for RGMRES and BiCGSTABL) 1.d-6 ! EPS -F ! Dump preconditioner on file T F -test-ml-unsm-our ! File prefix for preconditioner dump -ML-MUL-RAS-ILU ! Descriptive name for preconditioner (up to 40 chars) -ML ! Preconditioner NONE JACOBI BJAC AS ML --8000 ! If ML: Target coarse size. If <0, then use library default --1.5d0 ! If ML: Minimum aggregation ratio; if <0 use library default --0.10d0 ! If ML: Smoother Aggregation Threshold: >= 0.0 default if <0 --20 ! If ML: Maximum acceptable number of levels; if <0 use library default -SMOOTHED ! Type of aggregation: SMOOTHED, UNSMOOTHED, MINENERGY -DEC ! Type of aggregation: DEC SYMDEC -NATURAL ! Ordering of aggregation: NATURAL DEGREE -FILTER ! Filtering aggregation: FILTER NOFILTER -VCYCLE ! Type of multilevel correction: ADD MULT KCYCLE VCYCLE WCYCLE KCYCLESYM -TWOSIDE ! Side of correction: PRE POST TWOSIDE (ignored for ADD) -4 ! Smoother sweeps -BJAC ! Smoother type JACOBI BJAC AS; ignored for non-ML -0 ! Number of overlap layers for AS preconditioner (at finest level) -HALO ! AS Restriction operator NONE HALO -NONE ! AS Prolongation operator NONE SUM AVG -ILU ! Subdomain solver DSCALE ILU MILU ILUT FWGS BWGS MUMPS UMF SLU -1 ! Solver sweeps for GS -0 ! Level-set N for ILU(N), and P for ILUT +ML-VCYCLE-FBGS-SLUD ! 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 ! 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) -DIST ! Coarse level: matrix distribution DIST REPL -BJAC ! Coarse level: solver JACOBI BJAC UMF SLU SLUDIST MUMPS -UMFPACK ! Coarse level: subsolver DSCALE GS BWGS ILU UMF SLU SLUDIST MUMPS -1 ! Coarse level: Level-set N for ILU(N) -1.d-4 ! Coarse level: Threshold T for ILU(T,P) -2 ! Coarse level: Number of Jacobi sweeps +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) +1 ! Number of Jacobi sweeps for BJAC/PJAC coarsest-level solver