remove memory leak in PCSemiRedundant

PCBDDC: disable local subdomain info before refining connected components

PCBDDC: major improvement in benign code

Allows for multiple local benign dofs (in case any of the local matrices represents disconnected subdomains)
Extended accordingly BDDC class and BDDCGraph cl

PCBDDC: major improvement in benign code

Allows for multiple local benign dofs (in case any of the local matrices represents disconnected subdomains)
Extended accordingly BDDC class and BDDCGraph class
Added MatDetectDisconnectedComponent to identify them

show more ...

Merge branch 'master' into pr352/dmay/pctelescope/master

PCBDDC: avoid checks if the option has been set in the matrices and spd check if the benign trick has been requested

PCBDDC: usage of benign trick even without change of basis on velocities

this could happen if the user has already changed the basis on the velocities for local matrices before calling KSPSetUp

Merge branch 'master' into mark/eigen-rand

fix some issues with PetscViewerFlush() and PetscViewerASCIIPushSynchronized() within the plex object viewing

cleanup up of use of randomization in GAMG and Cheby, merge in Matt's portable rander48 and use it
instead of hokey hash based randomization

compile fix: complex build

PCBDDC: complex build fixes

when I swapped the role of coulping and rest in fieldsplit I missed a place

cleaned up cheby eig estimates and gamg rand

fixed C comments with no space after /* possibly broke bfort

Never, never never have some random comment that does not have a space after /*
it breaks sowing and hence configure big time

Take real part when comparing PetscScalar to PetscInt.

We need to product all the CUDA code from the spinlock definitions since the CUDA compiler
crashes on them

PCBDDC: Workaround for rhs in benign subspace

When the rhs is already in the benign space for Darcy, CG suffers a lot (still need to write down some formulas to understand why); instead GMRES conver

PCBDDC: Workaround for rhs in benign subspace

When the rhs is already in the benign space for Darcy, CG suffers a lot (still need to write down some formulas to understand why); instead GMRES converges in 1 (!) iteration.
With this, CG converges without iterating

show more ...

PCBDDC: nullspace info for the benign trick

PCBDDC: remove unneeded lines of code

Merge branch 'master' into barry/fix-petscviewer-attempt-2

Merge branch 'maint'

Fix bug introduced in previous patch

Reported-by: Shao-Ching Huang <huangsc@gmail.com>

Cleaned up comments

PCBDDC: handle possible non-zero diagonal terms in pressure blocks (wells)

By default, the code uses the zero diagonal dofs as pressure dofs.
If a local dof split is present, assumes that the last f

PCBDDC: handle possible non-zero diagonal terms in pressure blocks (wells)

By default, the code uses the zero diagonal dofs as pressure dofs.
If a local dof split is present, assumes that the last field represents pressures
Then checks the two fields and avoid change of basis on pressures if the two fields are not the same

show more ...