| 0a94a983 | 03-Mar-2022 |
Jed Brown <jed@jedbrown.org> |
PCGAMG/KSPChebyshev: clean up sticky logic
Chebyshev was re-estimating eigenvalues after the first solve even when
PCGAMG was providing SA eigenvalues. The output was also confusing. Now
we only sto
PCGAMG/KSPChebyshev: clean up sticky logic
Chebyshev was re-estimating eigenvalues after the first solve even when
PCGAMG was providing SA eigenvalues. The output was also confusing. Now
we only store what is specified directly by the user and report how we
got our estimates.
MG smoother output now explains the eigenvalue targets and transform,
e.g., for the typical case where it's provided by GAMG.
KSP Object: (mg_levels_1_) 1 MPI processes
type: chebyshev
eigenvalue targets used: min 0.197657, max 2.17423
eigenvalues provided (min 0.0234084, max 1.97657) with transform: [0. 0.1; 0. 1.1]
maximum iterations=2, nonzero initial guess
tolerances: relative=1e-05, absolute=1e-50, divergence=10000.
left preconditioning
using NONE norm type for convergence test
show more ...
|
| 2de708cb | 03-Mar-2022 |
Jed Brown <jed@jedbrown.org> |
PCGAMG: fix invalid access from redundent KSPSetFromOptions(smoother)
If the user writes -mg_levels_ksp_type richardson, for example, then the
following code would be referencing invalid cheb->membe
PCGAMG: fix invalid access from redundent KSPSetFromOptions(smoother)
If the user writes -mg_levels_ksp_type richardson, for example, then the
following code would be referencing invalid cheb->members after
KSPSetFromOptions(smoother) changed the smoother from Chebyshev to
Richardson. But we don't need this call because KSPSetFromOptions is
called later in PCSetUp_MG, thus it's harmless for us to handle
Chebyshev here and allow for it to be changed later.
show more ...
|