added -pc_hypre_boomeramg_nodal_coarsen <n> and -pc_hypre_boomeramg_vec_interp_variant <v> options
to allow PCType of PCHYPRE BoomerAMG to take advantage of MatSetNearNullSpace()

Requested by a vari

added -pc_hypre_boomeramg_nodal_coarsen <n> and -pc_hypre_boomeramg_vec_interp_variant <v> options
to allow PCType of PCHYPRE BoomerAMG to take advantage of MatSetNearNullSpace()

Requested by a variety of users

show more ...

Merge branch 'maint'

Fixed some PetscValidLogicalCollective calls that used the wrong typename
One of these caused problems for a user with 64 bit integers

Reported-by: Erik Andreassen <erand@mek.dtu.dk>

Merge branch 'barry/remove-pcsetup-mg-changeofpreonly/master'

fix location of ex10.rm in TESTEXAMPLES_HYPRE
allow for compressed poisson2.gz file
fix formatting in hypre.c
add comment in makefile regarding Euclid's handling of MPI communicator

Merge branch 'maint'

Changed option for AMS, reportedly with previous option hypre setup failed

Reported-by: Jürgen Zechner <juergen.zechner@tailsit.com>

Bugfix: Print string whenever smoothtype is undefined

If smoothtype has not been set (i.e. is the default value of -1), one of the standard smoothers is used so we should print out the string: "HYPR

Bugfix: Print string whenever smoothtype is undefined

If smoothtype has not been set (i.e. is the default value of -1), one of the standard smoothers is used so we should print out the string: "HYPRE BoomerAMG: Not using more complex smoothers". Before it was printed whenever an advanced smoother which is *not* Euclid was used

show more ...

Changed the /* */ style comments

Merge branch 'maint'

Need to dely attachment of null spaces from IS to matrices inside fieldsplit until after the matrices are available

Reported-by: Lawrence Mitchell <lawrence.mitchell@imperial.ac.uk>

removed eig est unused stuff

PCBDDC: compute column sum of zero block

PCBDDC: add command line switch for detecting saddle point

removed unused var

removed cheb eig in GAMG, added CG eig est to tests

previously PCSetUp_MG() would change the KSP type from preonly to gmres unless the PC was LU, Cholesky, redundant, SVD, or bjacobi (with one block per process). I have removed this code; it is now th

previously PCSetUp_MG() would change the KSP type from preonly to gmres unless the PC was LU, Cholesky, redundant, SVD, or bjacobi (with one block per process). I have removed this code; it is now the users responsibility if they change the PC and wish to use an iterative KSP they must change the KSP. The reason for the change was the check for bjacobi required that the matrices had already been provided and setup called on the submatrices which need not be the case; plus the code to change to gmres was ad hoc and doesn't match the style of PETSc which is not to change solvers unless directed by the user.

show more ...

Merge branch 'knepley/fix-pc-composite-dm'

* knepley/fix-pc-composite-dm:
PC: Fixed propagation of DM by PCCOMPOSITE Reported-by: Lawrence Mitchell <lawrence.mitchell@imperial.ac.uk>

Merge branch 'maint'

One now passes the nullspace in the matrix for a linear system as the first Amat argument to KSPSetOperators, SNESSetOperators, TSSetIJacobian() not the second pmat argument

This allows nesting of K

One now passes the nullspace in the matrix for a linear system as the first Amat argument to KSPSetOperators, SNESSetOperators, TSSetIJacobian() not the second pmat argument

This allows nesting of KSP/PC where the matrix is singular but the matrix passed to construct the preconditioner (the pmat) is not singular

Reported-by: Stephan Kramer <s.kramer@imperial.ac.uk>

show more ...

Merge branch 'stefano_zampini/matis-feature-localtoglobalmap-handling'

Merge branch 'master' of bitbucket.org:petsc/petsc into mark/gamg-crs

PCBDDC: fix set-but-not-used warning from nightly build

PCBDDC: raise an error if the preconditioning matrix is not square

MATIS: enable rectangular shape for MATIS