dynamics: add implicit VP solver#491

Merged

apcraig merged 196 commits into

CICE-Consortium:masterfrom

phil-blain:parallel-picard

Sep 22, 2020

Commits on Aug 25, 2020

initial coding of ice_dyn_vp.F90

JFLemieux73
authored and
phil-blain
committed
code (matvec, bvec and residual) now compiles after debugging

JFLemieux73
authored and
phil-blain
committed
introduced calc_vrel_Cb and calc_visc_coeff subroutines and precond.F90 file

JFLemieux73
authored and
phil-blain
committed
after debugging...now compiles

JFLemieux73
authored and
phil-blain
committed
in process of adding precondD

JFLemieux73
authored and
phil-blain
committed
precondD is in progress

JFLemieux73
authored and
phil-blain
committed
I finished coding precondD...it compiles

JFLemieux73
authored and
phil-blain
committed
added calculation for L2norm

JFLemieux73
authored and
phil-blain
committed
modif to ice_step_mod.F90 for call to imp_solver

JFLemieux73
authored and
phil-blain
committed
now form complete vectors by alternating u(i,j) and v(i,j) and so on

JFLemieux73
authored and
phil-blain
committed
added simple routine to form a vector

JFLemieux73
authored and
phil-blain
committed
corrected bug for vectors of size ntot

JFLemieux73
authored and
phil-blain
committed
added vectors of size ntot to bvec and matvec

JFLemieux73
authored and
phil-blain
committed
step back...not so easy to form all the vectors for the blocks...

JFLemieux73
authored and
phil-blain
committed
new subroutine for assembling ntot vectors

JFLemieux73
authored and
phil-blain
committed
In the process of adding the call to fgmres

JFLemieux73
authored and
phil-blain
committed
new subroutine to convert vector to the max_blocks arrays

JFLemieux73
authored and
phil-blain
committed
ice_dyn_vp compiles...I now need to make fgmres.F90 compile...

JFLemieux73
authored and
phil-blain
committed
added BLAS routines needed by fgmres

JFLemieux73
authored and
phil-blain
committed
added the fgmres.F90 routine...does not compile yet...

JFLemieux73
authored and
phil-blain
committed
minor modif to fgmres.F90

JFLemieux73
authored and
phil-blain
committed
bug in call to arrays_to_vec instead of vec_to_arrays

JFLemieux73
authored and
phil-blain
committed
Remove fgmres.F90 (MPI version, will re-add GEM version later)
phil-blain
committed
added serial fgmres for testing

JFLemieux73
authored and
phil-blain
committed
modified code for serial tests

JFLemieux73
authored and
phil-blain
committed
after debugging...

JFLemieux73
authored and
phil-blain
committed
added code to put sol vector in uvel and vvel arrays at the end of fgmres

JFLemieux73
authored and
phil-blain
committed
modified tinyarea calc for VP (punyVP=2e-9)

JFLemieux73
authored and
phil-blain
committed
modified location of allocation for some variables...

JFLemieux73
authored and
phil-blain
committed
small bug corrected in allocating variables

JFLemieux73
authored and
phil-blain
committed
introduced capping of the viscous coeff following Kreysher

JFLemieux73
authored and
phil-blain
committed
added a relaxation to uvel,vvel to try to get nonlinear conv

JFLemieux73
authored and
phil-blain
committed
modif to relaxation calc...minor

JFLemieux73
authored and
phil-blain
committed
I found a bug in my 1st implementation. The dsig/dx term related to the rep pressurer should be on the RHS...in the process of correcting this

JFLemieux73
authored and
phil-blain
committed
Pr is now calc with zeta

JFLemieux73
authored and
phil-blain
committed
dPrdx is now in calc_bvec

JFLemieux73
authored and
phil-blain
committed
removed Pr part from stress_vp calc

JFLemieux73
authored and
phil-blain
committed
minor modifs to improve print of res norm in fgmres

JFLemieux73
authored and
phil-blain
committed
added linear conv criterion based on a gamma

JFLemieux73
authored and
phil-blain
committed
added NL conv criterion

JFLemieux73
authored and
phil-blain
committed
in the process of adding precond diag

JFLemieux73
authored and
phil-blain
committed
now forms the vector diagvec

JFLemieux73
authored and
phil-blain
committed
precond diag is completed and compiles

JFLemieux73
authored and
phil-blain
committed
diagvec was not allocated...now ok

JFLemieux73
authored and
phil-blain
committed
diagvec was not deallocated...now ok

JFLemieux73
authored and
phil-blain
committed
fixed bug in formDiag_step2

JFLemieux73
authored and
phil-blain
committed
small modif

JFLemieux73
authored and
phil-blain
committed
small modif...again

JFLemieux73
authored and
phil-blain
committed
modified formation of diagonal for precond

JFLemieux73
authored and
phil-blain
committed
just removed a comment

JFLemieux73
authored and
phil-blain
committed
minor modif to formDiag_step1

JFLemieux73
authored and
phil-blain
committed
added an option for precond (free drift diag or complete diag

JFLemieux73
authored and
phil-blain
committed
minor modif

JFLemieux73
authored and
phil-blain
committed
minor modif

JFLemieux73
authored and
phil-blain
committed
removed OLDformDiag_step1

JFLemieux73
authored and
phil-blain
committed
Ok I think I know what was wrong with the precond diag...I have to use icellu and not icellt

JFLemieux73
authored and
phil-blain
committed
precond 3 works but not as good as precond 2 (free drift)...removed OLD code

JFLemieux73
authored and
phil-blain
committed
ok found the problem in step2...now fixed

JFLemieux73
authored and
phil-blain
committed
added routine to calc deformations for mech redistribution...for the evp this is done for k=ndte in stress...here it is done once we have NL convergence

JFLemieux73
authored and
phil-blain
committed
chnaged names of evp_prep1 to dyn_prep1...

JFLemieux73
authored and
phil-blain
committed
in process of adding gmres as precond to fgmres...

JFLemieux73
authored and
phil-blain
committed
matvec is now done in one routine instead of in two steps before

JFLemieux73
authored and
phil-blain
committed
in the process of adding gmres precond

JFLemieux73
authored and
phil-blain
committed
finalizing implementation of gmres for precond

JFLemieux73
authored and
phil-blain
committed
pgmres seems to work...I used it as a solver and it works fine

JFLemieux73
authored and
phil-blain
committed
minor modif: introduction of maxits_pgmres

JFLemieux73
authored and
phil-blain
committed
modif to comments...very minor

JFLemieux73
authored and
phil-blain
committed
something was inconsistent for precond choice...now fixed and removed one precond option (free drift)

JFLemieux73
authored and
phil-blain
committed
removed krelax...not needed as a more complex method will be added

JFLemieux73
authored and
phil-blain
committed
in the process of adding the RRE1 acceleration method

JFLemieux73
authored and
phil-blain
committed
fixed problem in halo update (spherical grid...wraps on itself) before matvec

JFLemieux73
authored and
phil-blain
committed
cicecore: add puny_dyn variable to initialize tinyarea, remove temporary punyVP variable
phil-blain
committed
ice_dyn_vp: move solver algorithm and output parameters to namelist
phil-blain
committed
{f,}gmres: use nu_diag for output
phil-blain
committed
dynamics: correct 2 bugs found using debugging flags
phil-blain
committed
ice_dyn_vp: rename VP namelist variables to be more descriptive
phil-blain
committed
pgmres: streamlined monitoring output
phil-blain
committed
ice_dyn_vp: add computation of nonlinear and fixed point residuals
phil-blain
committed
ice_dyn_vp: remove references to RRE acceleration method
phil-blain
committed
ice_dyn_vp: remove unused 'ulin', 'vlin' variables
phil-blain
committed
ice_dyn_vp: move Picard iteration to separate subroutine
phil-blain
committed
ice_dyn_vp: add Anderson acceleration for Picard iteration
phil-blain
committed
options: add options files for implicit solver
phil-blain
committed
ice_dyn_vp: correct residual computations if max_blocks > 1
phil-blain
committed
ice_dyn_vp: differentiate fixed point and progress residuals
phil-blain
committed
ice_dyn_vp: fix convergence issues in Picard and Anderson solvers
phil-blain
committed
ice_dyn_vp: add (ulin,vlin) to compute vrel based on 2 previous iterates
phil-blain
committed
machines: cesium: add LAPACK and BLAS library to Macros
phil-blain
committed
ice_dyn_vp: add 'subname' variable to each subroutine
phil-blain
committed
ice_dyn_vp: cleanup 'icepack_warnings_*' calls
phil-blain
committed
ice_dyn_vp, pgmres: remove unused arguments, variables and 'use' statements variables
phil-blain
committed
ice_dyn_vp: refactor 'puny' treatment for implicit solver
phil-blain
committed
ice_dyn_vp: make 'calc_bvec' use already computed 'vrel'
phil-blain
committed
ice_dyn_vp: remove references to fgmres2 subroutine
phil-blain
committed
dynamics: remove 'BLAS_routines.F90'
phil-blain
committed
ice_dyn_vp: synchronize 'imp_solver' with ice_dyn_evp::evp
phil-blain
committed
dynamics: refactor strain rates and deformation calculations
phil-blain
committed
dynamics: add sol_fgmres2d from GEM as ice_krylov.F90 (does not compile)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to inner loop (WIP)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to preconditioner call (WIP)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to before CGS (WIP)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to end of Arnoldi (except 1st GS loop) (WIP)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to updating of solution (WIP)
phil-blain
committed
ice_krylov: adapt FGMRES algo up to residual calculation (WIP)
phil-blain
committed
ice_krylov: correct CICE kinds in function almost_zero
phil-blain
committed
comm: add 'global_sums' module procedure
phil-blain
committed
ice_krylov: adapt FGMRES including Gram-Schmidt (WIP)
phil-blain
committed
ice_krylov: add 'pgmres' subroutine (mostly the same as 'fgmres')
phil-blain
committed
ice_krylov: move subroutines to ice_dyn_vp to work around circular deps
phil-blain
committed
ice_dyn_vp: add 'subname' to subroutine 'almost_zero' and correct indentation
phil-blain
committed
ice_dyn_vp: add 'ice_HaloUpdate_vel' subroutine
phil-blain
committed
ice_dyn_vp: make 'fld2' a module variable
phil-blain
committed
ice_dyn_vp: move 'ice_halo' to module 'use' statement
phil-blain
committed
ice_dyn_vp: add workspace initialization to 0 and haloupdate before matvec in fgmres (WIP)
phil-blain
committed
ice_dyn_vp: add choice between classical and modified Gram-Schmidt orthogonalization
phil-blain
committed
ice_dyn_vp: change order of arguments for 'precondition' subroutine
phil-blain
committed
ice_dyn_vp: fgmres: correctly initialize workspace_[xy] and arnoldi_basis_[xy]
phil-blain
committed
ice_dyn_vp: matvec: change Au,Av from 'intent(out)' to 'intent(inout)'
phil-blain
committed
ice_dyn_vp: clarify decriptions of FGMRES variables
phil-blain
committed
ice_dyn_vp: add subroutines description and clarify existing ones
phil-blain
committed
ice_dyn_vp: fix comments in FGMRES algorithm
phil-blain
committed
ice_dyn_vp: fgmres: fix misleading check for return
phil-blain
committed
ice_dyn_vp: fgmres: fix error in the last loop computing the residual
phil-blain
committed
ice_dyn_vp: change the definition of F(x) from 'A.x - b' to 'b - A.x'
phil-blain
committed
ice_dyn_vp: fgmres: comment out early return (WIP)
phil-blain
committed
ice_dyn_vp: pgmres: synchronize with fgmres (WIP)
phil-blain
committed
ice_dyn_vp: picard_solver: re-enable legacy solver
phil-blain
committed
cicecore: remove remaining 'SVN:$Id:' lines
phil-blain
committed
ice_dyn_vp: remove 'picard_solver' subroutine
phil-blain
committed
ice_dyn_vp: remove legacy FGMRES solver
phil-blain
committed
ice_dyn_vp: pgmres: use Classical Gram-Schmidt
phil-blain
committed
ice_in: change maxits_{f,p}gmres to be in line with new meaning
phil-blain
committed
ice_dyn_vp: uniformize naming of namelist parameters for solver tolerances
phil-blain
committed
ice_dyn_vp: anderson_solver: adapt residual norm computations for MPI
phil-blain
committed
ice_dyn_vp: write solver diagnostics only for master task
phil-blain
committed
ice_dyn_vp: remove references to EVP
phil-blain
committed
ice_dyn_vp: rename 'puny_vp' to 'min_strain_rate'
phil-blain
committed
ice_dyn_vp: remove "signature" comments from JF
phil-blain
committed
ice_dyn_vp: write stresses at end of time step
phil-blain
committed
ice_dyn_vp: cleanup 'intent's
phil-blain
committed
dynamics: correct the definition of 'shearing strain rate' in comments
phil-blain
committed
ice_dyn_vp: correctly use 'subname' in 'abort_ice' calls
phil-blain
committed
ice_dyn_vp: add 'calc_seabed_stress' subroutine
phil-blain
committed
dynamics: move basal stress residual velocity ('u0') to ice_dyn_shared
phil-blain
committed
ice_dyn_vp: use 'deformations' from ice_dyn_shared
phil-blain
committed
ice_dyn_vp: simplify 'calc_vrel_Cb' subroutine
phil-blain
committed
ice_dyn_vp: rename 'Diag[uv]' to 'diag[xy]' for consistency
phil-blain
committed
ice_dyn_vp: introduce 'CICE_USE_LAPACK' preprocessor macro
phil-blain
committed
ice_dyn_vp: convert 'algo_nonlin' to string
phil-blain
committed
ice_dyn_vp: convert 'precond' to string
phil-blain
committed
ice_dyn_vp: convert 'monitor_{f,p}gmres' to logical
phil-blain
committed
ice_dyn_vp: remove unimplemented 'fpfunc_andacc' from the namelist
phil-blain
committed
ice_dyn_vp: add input validation for string namelist variables
phil-blain
committed
ice_dyn_vp: abort if Anderson solver is used in parallel
phil-blain
committed
ice_dyn_vp: reimplement monitor_{f,p}gmres
phil-blain
committed
ice_dyn_vp: use 'ice_HaloUpdate_vel' everywhere
phil-blain
committed
ice_dyn_vp: move ice_HaloUpdate_vel subroutine to ice_dyn_shared
phil-blain
committed
ice_dyn_evp: use ice_HaloUpdate_vel
phil-blain
committed
ice_dyn_eap: use ice_HaloUpdate_vel
phil-blain
committed
options: set 'use_mean_vrel' in 'dynpicard'
phil-blain
committed
conda: add 'libapack' to environment spec and Macros
phil-blain
committed
ice_dyn_vp: initialize 'L2norm' and 'norm_squared' to zero
phil-blain
committed
ice_dyn_vp: initialize 'stPr' and 'zetaD' to zero
phil-blain
committed
doc: document implicit solver
phil-blain
committed
ice_dyn_vp: default 'use_mean_vrel' to 'true'
phil-blain
committed
ice_in: remove Anderson solver parameters
phil-blain
committed
ice_dyn_vp: add article references
phil-blain
committed
ice_dyn_vp: remove unused subroutines
phil-blain
committed
ice_dyn_vp: remove unused local variables
phil-blain
committed
ice_dyn_vp: calc_bvec: remove unused arguments
phil-blain
committed
ice_dyn_vp: anderson_solver: remove unused arguments
phil-blain
committed
ice_dyn_vp: remove unused 'use'-imported variables in subroutines
phil-blain
committed
ice_dyn_vp: fix trailing whitespace errors
phil-blain
committed
ice_dyn_vp: uniformize indentation and whitespace
phil-blain
committed
ice_dyn_vp: put 'intent's on same lines as types
phil-blain
committed
ice_dyn_vp: use '==' instead of '.eq.'
phil-blain
committed
ice_dyn_vp: uniformize whitespace in subroutine arguments
phil-blain
committed
ice_dyn_vp: calc_bvec: remove unneeded arguments
phil-blain
committed
ice_dyn_vp: clean up and clarify code comments
phil-blain
committed
ice_dyn_vp: matvec: remove unneeded local variable
phil-blain
committed
ice_dyn_vp: rename 'stPrtmp' to 'stress_Pr' and 'Dstrtmp' to 'diag_rheo'
phil-blain
committed
ice_dyn_vp: rename 'calc_zeta_Pr' to 'calc_zeta_dPr'
phil-blain
committed
ice_dyn_vp: rename 'ww[xy]' to 'orig_basis_[xy]'
phil-blain
committed
ice_dyn_vp: [fp]gmres: removed unused 'r0' variable and 'conv' argument
phil-blain
committed
ice_dyn_vp: make 'maxits_nonlin' default to 4
phil-blain
committed
ice_dyn_vp: rename 'im_{fgmres,pgmres,andacc}' to 'dim_*'
phil-blain
committed
tests: add 'dynpicard' test to base_suite
phil-blain
committed
dynamics: remove proprietary vector directives
phil-blain
committed
ice_init: validate implicit solver arguments only if solver is active
phil-blain
committed
ice_dyn_vp: rename 'imp_solver' to 'implicit_solver'
phil-blain
committed
doc: add caveat for VP solver (tx1 grid, threading)
phil-blain
committed
ice_dyn_vp: rename 'CICE_USE_LAPACK' macro to 'USE_LAPACK'
phil-blain
committed
comm: rename 'global_sums' to 'global_allreduce_sum'
phil-blain
committed
dynamics: remove 'ice_HaloUpdate_vel' and introduce '(un)stack_velocity_field'
phil-blain
committed

Commits on Aug 26, 2020

dynamics: rename 'init_evp' to 'init_dyn'
phil-blain
committed
drivers: call 'init_dyn' unconditionnally
phil-blain
committed