Changes: 2.0.29#

Changes#

(See Features below)

General:

AO (Application Orderings):

TS (Timestepping Solvers):

SNES (Nonlinear Solvers):

  • The second argument to MatSNESMFSetType() is now a MatSNESMFType, rather than a char*. This should not affect any user code. Current built in choices are MATSNESMF_DEFAULT “default” and MATSNESMF_WP “wp”

  • SNESGetJacobian() now has an additional optional final argument that can return the function pointer

  • SNESGetFunction() now has an additional optional final argument that can return the function pointer

SLES (Linear Solvers):

KSP (Krylov Subspace Methods):

PC (Preconditioners):

  • PCILUSetReuseFill() changed to PCILUDTSetReuseFill() and -pc_ilu_reuse_fill changed to -pc_iludt_reuse_fill Note that currently these functions don’t work.

  • PCNullSpaceXXX changed to MatNullSpaceXXX except for PCNullSpaceAttach()

  • MGSetLevels() now takes an additional optional final argument that is an array of communicators to use on the SLES on each level

MAT (Matrices):

  • Changed MatDiagonalShift(Mat,Vec) to MatDiagonalSet(Mat,Vec,InsertMode);

  • Dropped the second to last argument to MatCreateRowbs()

  • -mat_ordering_type now should be either -pc_ilu_mat_ordering_type, -pc_lu_mat_ordering_type, -pc_icc_mat_ordering_type or -pc_cholesky_mat_ordering_type

DA (Distributed Arrays):

VEC (Vectors):

IS (Index Sets):

  • ISEqual() now is collective and returns true only if each processor has the same indices

  • ISGetSize() now returns global parallel size, ISGetLocalSize() returns local size

General:

  • All include files like mat.h vec.h sles.h snes.h ts.h etc are now changed to petscmat.h petscvec.h petscsles.h petscsnes.h petscts.h

Draw (Graphics):

Viewers:

System:

  • FListDestroy now takes FList * instead of FList

Error Handling:

Event Logging:

  • PLogEventBegin(), PLogEventEnd(), PLogFlops(), PLogEventBarrierBegin() and PLogEventBarrierEnd() now return error codes.

Fortran Interface:

Features#

(See Changes above)

General:

AO (Application Orderings):

TS (Timestepping Solvers):

SNES (Nonlinear Solvers):

SLES (Linear Solvers):

KSP (Krylov Subspace Methods):

  • Minres solver contibuted by: Robert Scheichl: maprs@maths.bath.ac.uk

  • SYMMLQ solver contributed by Hong Zhang

PC (Preconditioners):

  • ICC(k) and Cholesky() preconditioners are now available for the new SeqSBAIJ matrices.

  • added -pc_ilu_damping and -pc_lu_damping, PCILUSetDamping(), PCLUSetDamping() for handling singular or matrices that generate a zero pivot.

  • Added DAMGCreate() etc to help easily write parallel multigrid codes that use rectangular grids

  • Paulo Goldfeld contributed new PCType of nn, a version of Mandel’s balancing Neumann-Neumann preconditioner, see src/contrib/oberman/laplacian_q1 for an example of its use

MAT (Matrices):

  • added new matrix types SeqSBAIJ and MPISBAIJ for symmetric matrices. These store only the upper triangular portion of the matrix. We also provide ICC(k) and Cholesky factorization (currently only for block size of 1) for the SeqSBAIJ matrices

  • augmented the SeqAIJ matrices to support -mat_aij_matlab that causes the matrices to use the MATLAB engine for sparse LU, and ILUDT factorization and solves. Which have pivoting and thus are more robust than PETSc’s default.

DA (Distributed Arrays):

  • added -da_noao for DACreateXXX() makes them more scalable, but cannot use the natural numbering DA routines with this option.

  • Added DAMGCreate() etc to help easily write parallel multigrid codes that use rectangular grids

VEC (Vectors):

IS (Index Sets):

PF:

  • added the type PFMATLAB with -pf_type matlab -pf_matlab matlab_commands

Draw (Graphics):

Viewers:

System:

  • Add PetscMatlabEngineCreate() which allows calling Matlab functions from PETSc on each processor. Good for graphics or numerical routines that PETSc does not have. for example for the local evaluation of the parallel nonlinear function required for SNES; see src/snes/examples/tutorials/ex5m.c. This functionality is accessed via the commands PetscMatlabEngineCreate(),Put(),Get(),Evaluate() and Destroy() It even imports the MATLAB flop counts correctly. Obviously you need enough MATLAB licenses for each node. Note: MATLAB itself is not directly doing any parallel computing.

  • added PetscSetHelpVersionFunctions() for developers of other libraries that use PETSc.

  • added PetscPoUpSelect(), see src/sys/examples/tests/ex15.c

  • added int PetscShowMemoryUsage(Viewer viewer,char *message)

  • added PreLoadBegin(), PreLoadStage() and PreLoadEnd() to simplify getting accurate timings by peloading. See src/snes/examples/tutorials/ex10.c

  • added PetscSynchronizedFGets()

Error Handling:

Event Logging:

Fortran Interface: