PETSc

Documentation: Changes: 2.0.18-2.0.21

• Home
• Download
• Features
• Documentation
  • Manual pages and Users Manual
  • Referencing PETSc
  • Tutorials
  • Installation
  • SAWs
  • Changes
    • dev
    • 3.15
    • 3.14
    • 3.13
    • 3.12
    • 3.11
    • 3.10
    • 3.9
    • 3.8
    • 3.7
    • 3.6
    • 3.5
    • 3.4
    • 3.3
    • 3.2
    • 3.1
    • 3.0.0
    • 2.3.3
    • 2.3.2
    • 2.3.1
    • 2.3.0
    • 2.2.1
    • 2.2.0
    • 2.1.6
    • 2.1.5
    • 2.1.3
    • 2.1.2
    • 2.1.1
    • 2.1.0
    • 2.0.29
    • 2.0.28
    • 2.0.24
    • 2.0.22
    • 2.0.18-21
    • 2.0.17
    • 2.0.16
    • 2.0.15
  • Bug Reporting
  • Code Management
  • FAQ
  • License
  • Linear Solver Table
  • Nonlinear Solver Table
  • TAO Solver Table
• Applications/Publications
• Miscellaneous
• External Software
• Developers Site

General:

• Complex numbers performance upgrade: Added support for using optimized Fortran kernels for some key complex numbers numerical routines (such as matrix-vector products, vector norms, etc.) instead of the default C++ routines. This implementation exploits the maturity of Fortran compilers while retaining the identical user interface. For example, on rs6000 machines, the base single-node performance when using the Fortran kernels is 4-5 times faster than the default C++ code.
• Changed the names of various compiler flags, e.g., changed PETSC_COMPLEX to USE_PETSC_COMPLEX.
• Changed PetscObjectInherit() to PetscObjectCompose() since it really denotes a "has-a" relationship, not an "is-a" relationship.

AO (Application Orderings):

• Changed AOCreateDebug() to AOCreateBasic(); changed AOCreateDebugIS() to AOCreateBasicIS().
• Removed the MPI_Comm argument from AOCreateBasicIS() since it is contained in the IS arguments.
• Now the AOxxxToxxxXX() remapping routines will not map negative entries in the input arrays. This allows, for example, the mapping of neighbor lists that use negative entries to indicate non-existent neighbors due to boundaries, etc.

TS (Timestepping Solvers):

• Added an interface to PVODE, the stiff integrator package of Hindmarsh et al.

SNES (Nonlinear Solvers):

• Added support for using matrix colorings within finite difference Jacobian approximations. See the section "Finite Difference Jacobian Approximations" of the users manual for more details. Also see the man pages for SNESDefaultComputeJacobianWithColoring().
• Fixed a bug in method SNES_EQ_NLS.
• Increased the default maximum number of function evaluations to 100000.

SLES (Linear Solvers):

• See PC and KSP

KSP (Krylov Subspace Methods):

• Added the routine KSPGetResidualNorm().

PC (Preconditioners):

• Added -pc_lu_fill and -pc_ilu_fill to replace -mat_lu_fill and -mat_ilu_fill; also added the commands PCLUSetFill() and PCILUSetFill().
• Added PCLUSetMatReordering() and PCILUSetMatReordering().

MAT (Matrices):

• Added support for matrix colorings, which are intended primarily for use in finite difference Jacobian approximations. See the SNES section above for more info. New routines include:
  • MatFDColoringCreate()
  • MatFDColoringSetParameters()
  • MatFDColoringSetFrequency()
  • MatFDColoringSetFunction()
  • MatFDColoringSetFromOptions()
  • MatFDColoringView()
  • MatFDColoringApply()
  • MatFDColoringDestroy()
• Added the matrix option MatSetOption(mat,MAT_NEW_NONZERO_ALLOCATION_ERROR) that will cause an error if a new entry that has not been preallocated is generated in a sparse matrix. (currently implemented for AIJ and BAIJ matrices only). This is a useful flag when debugging memory preallocation.
• Replaced the options -mat_lu_fill and -mat_ilu_fill with -pc_lu_fill and -pc_ilu_fill.
• Added the routine MatSetValuesBlockedLocal() for BAIJ matrices.
• Changed the final argument of MatGetTypeFromOptions() from type int* to PetscTruth*.
• Added MatCreateSeqAdj() for supplying adjacency matrices to PETSc to do reordering on (for example RCM to reduce bandwidth and thus get better cache performance) and eventually partitioners.
• MatSetLocalToGlobalMapping() and MatSetLocalToGlobalMappingBlocked() now take a ISLocalToGlobalMapping object rather than a list of indices.
• Added the routine MatGetSubMatrix(), which extracts a parallel matrix from a parallel matrix (currently implemented only for the MPIAIJ format).

DA (Distributed Arrays):

• When used with the DA_STENCIL_STAR stencil, type, the routine DAGetGlobalIndices() returns local-to-global mapping indices that now include the inactive corner ghost nodes. This is useful, e.g., when using MatSetValuesLocal() to set matrix elements, including corner boundary nodes.

VEC (Vectors):

• VecSetLocalToGlobalMapping() now takes a ISLocalToGlobalMapping object rather than a list of indices.
• Added the routine VecCreateMPIWithArray().
• Changed the calling sequence for VecCreateGhost(); added VecCreateGhostWithArray(), VecGhostUpdate[Begin/End](), and VecGhost[Get/Restore]LocalRepresentations().

IS (Index Sets):

• Added ISGlobalToLocalMappingApply() to allow one to convert lists that are in the global numbering to a local numbering.
• Added a communicator as the first argument to ISLocalToGlobalMappingCreate().

Draw (Graphics):

• Added routines for drawing simple histograms. See DrawHistCreate().
• Removed the option -draw_x_private_colormap and made a private colormap the default.
• Added the option -draw_x_shared_colormap to indicate not to use a private colormap. If you use Netscape on your machine and are also doing contour plots, you generally don't want to use a shared colormap.
• Improved the colors used in the contour plotting.
• Changed some routine names:
  • DrawText() to DrawString()
  • DrawTextVertical() to DrawStringVertical()
  • DrawTextSetSize() to DrawStringSetSize()
  • DrawTextGetSize() to DrawStringGetSize()
  • DrawSyncClear() to DrawSynchronizedClear()
  • DrawSyncFlush() to DrawSynchronizedFlush()
  • DrawSyncGetMouseButton() to DrawSynchronizedGetMouseButton().

Viewers:

• Added VIEWER_STDOUT_() and VIEWER_STDERR_().

System Routines:

• Added the routine OptionsClearValue().
• Added the option -get_resident_set_size that causes the program to call PetscGetResidentSetSize() at the end of the run and print how much physical memory each process has used.
• Changed OptionsGetProgramName() to PetscGetProgramName() and changed the calling sequence to match PetscGetHostname(), etc.
• Changed BINARY_INT and BINARY_SCALAR to PETSC_INT and PETSC_SCALAR.

Error Handling:

Event Logging:

Fortran Interface:

• Added routines to map between C and Fortran representations of communicators
  • extern int MPICCommToFortranComm(MPI_Comm,int *);
  • extern int MPIFortranCommToCComm(int,MPI_Comm*);
  These provide the same functionality that
  • extern int PetscCObjectToFortranObject(void *,int *);
  • extern int PetscFortranObjectToCObject(int,void *);
  do for PETSc objects.
• Removed the macros Double, DBLE, PetscDoubleExp as they are no longer required. PETSc now compiles on the Cray T3D/T3E with the -dp option that correctly handles Fortran code using double precision.
• Added support for MatGetRow() MatRestoreRow() from Fortran; see manpage for Fortran calling sequence.
• Added PetscBinaryOpen(), PetscBinaryClose(), PetscBinaryRead() and PetscBinaryWrite() for binary IO from Fortran; see src/vec/examples/tests/ex20.F. Most users should not need this functionality.