Closed phil-blain closed 4 years ago
will do #12 first.
added some in d154f90
must think if more is necessary.
@JFLemieux73 penses-tu que c'est suffisant ? une chose qu'on pourrait ajouter c'est s'assurer que tous les paramètres de namelist int_kind
et dbl_kind
du solveur :
integer (kind=int_kind), public :: &
maxits_nonlin , & ! max nb of iteration for nonlinear solver
im_fgmres , & ! size of fgmres Krylov subspace
im_pgmres , & ! size of pgmres Krylov subspace
maxits_fgmres , & ! max nb of iteration for fgmres
maxits_pgmres , & ! max nb of iteration for pgmres
fpfunc_andacc , & ! fixed point function for Anderson acceleration: 1: g(x) = FMGRES(A(x),b(x)), 2: g(x) = x - A(x)x + b(x)
im_andacc , & ! size of Anderson minimization matrix (number of saved previous residuals)
start_andacc ! acceleration delay factor (acceleration starts at this iteration)
real (kind=dbl_kind), public :: &
reltol_nonlin , & ! nonlinear stopping criterion: reltol_nonlin*res(k=0)
reltol_fgmres , & ! fgmres stopping criterion: reltol_fgmres*res(k)
reltol_pgmres , & ! pgmres stopping criterion: reltol_pgmres*res(k)
damping_andacc , & ! damping factor for Anderson acceleration
reltol_andacc ! relative tolerance for Anderson acceleration
sont positifs, mais ce genre de vérifications est en général pas fait pour les autres paramètres de namelist de CICE, donc pense-tu que ça vaut la peine ?
Conclusion: pas nécessaire.
Add VP solver-specific section at ice_init.F90:950.
Probably should move the code in
anderson_solver
that resetsim_andacc
to zero to that section, because if not the subroutine arguments are declared using theim_andacc
value from the namelist, before it is reset to 0.