Closed philbucher closed 5 years ago
Did you use the new preconditioner_type
option with amgcl solver? This switches from amg to a single level relaxation for preconditioning. When you do that, the usual amg parameters are no longer applicable, hence the warnings. The warnings themselves are nothing to worry about except the extra verbosity. Your options are:
#define AMGCL_PARAM_UNKNOWN(name) (void)0
preconditioner_type
option.@sunethwarna I think this is applicable to your issues too.
I think the amgcl_solver.h needs to be updated. We are getting a lot of "unknown parameter M" warnings. I closed #3148 since this thread came first, but it is still an open issue for us.
@RiccardoRossi , what do you think about disabling amgcl warnings completely?
Looks like uncommenting the following line should be enough:
Just tested it, but still getting the warning about "M" being unknown. My settings are:
"solver_type" : "AMGCL",
"max_iteration" : 200,
"tolerance" : 1e-9,
"provide_coordinates" : false,
"smoother_type" : "ilu0",
"krylov_type" : "gmres",
"coarsening_type" : "aggregation",
"scaling" : false
I think the problem is related to this if-statement: https://github.com/KratosMultiphysics/Kratos/blob/dd66d8da4fb621575f673eb2177ccc5b20d83275/kratos/linear_solvers/amgcl_solver.h#L298
gmres-type solvers should expect 'M' as a parameter. Are you sure you are not using bicgstab_with_gmres_fallback
? It sets 'M' but uses bicgstab solvers which does not have the parameter.
@philbucher @ddemidov this is on me. I'll take a look asap
there are two problems:
1 - M is always set even when gmres is not used. The AMGCL correctly warns about this 2 - amg settings are passed even when amg is not used as a preconditioner
@ddemidov i think that the warnings issued by AMGCL are very useful, and i would prefer not to deactivate them. I'll try to correct this on the user side
The trilinos-version still prints some of those, see #3343
AMGCL: Unknown parameter class
@RiccardoRossi ....
I cleaned up the OpenMP version but not the MPI version. I will do also the MPI when i can...
however, @msandre are u getting this with the "serial" version? because that is expected to be clean
My warnings (in serial) have disappeared. Thanks!
ping @RiccardoRossi
@RiccardoRossi @ddemidov after #2807 AMGCL throws a lot of warnings about unknown parameters
This is some tests-output:
I looked around a bit but I couldn't figure out how to fix it
can you please have a look?