PETSc

Documentation: Changes: 3.0.0

• Home
• Download
• Features
• Documentation
  • Manual pages and Users Manual
  • Referencing PETSc
  • Tutorials
  • Installation
  • SAWs
  • Changes
    • dev
    • 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
• Applications/Publications
• Miscellaneous
• External Software
• Developers Site

General:

• Changed the Matlab engine portion of the Matlab configure code to require a separate flag --with-matlab-engine
• Changed the Socket viewer to be the same as the binary viewer except for its creation
• Changed the Matlab Socket code to work the same as the Matlab binary reader except for its creation
• You can now replace the PetscVFPrintf() function pointer to redirect all stdout/stderr output in Windows for example.
• Removed MPICCommToFortran() and MPIFortranCommToC() since MPI_Comm_c2f() and f2c() are now always available.
• PetscOptionsInsertFile() now takes a MPI_Comm as the first argument and a final flag argument indicating if the file  MUST be located. It is also changed to be more scalable, only the 0th process opens it and broadcasts the input to the other processes.
• Added PetscObjectIncrementTabLevel() to allow proper indenting of monitor output for imbedded solvers for multigrid levels, for block Jacobi, for SNES inside TS and KSP inside SNES etc.

Logging:

• The int stage used in the PetscLogStageXXX() routines is now a PetscLogStage
• PetscLogEventRegister() and PetscLogStageRegister() now return the result in the final argument instead of the first
• PetscLogClassRegister() is changed to PetscClassIdRegister() and returns the classid in the final argument, not the first
• PetscEvent is now PetscLogEvent

config/configure.py:

• Changed location of base makefiles to conf from bmake/common
• Changed location of generated base makefiles to $PETSC_ARCH/conf instead of bmake/$PETSC_ARCH
• Changed location of generated include files to $PETSC_ARCH/include
• Changed location of generated libraries (including from external packages) to $PETSC_ARCH/lib
• You will need to change include ${PETSC_DIR}/bmake/common/base to ${PETSC_DIR}/conf/base

IS:

• ISPartitioningCount() now takes an additional argument, the length of the count array (that is the number of partitions).

Vec:

• Added logical argument to VecSetOption() to toggle option in either way
• The PetscMap inside the Vec object is now allocated and reference counted and shared between vectors and matrices, this will decrease memory usage on systems with a large size.

VecScatter:

Mat:

• MatSetOption() now takes a final PetscTruth flag to toggle the option. The redundant options like MAT_NO_SYMMETRIC are removed
• The MatSetOption() options MAT_ROWS_SORTED, MAT_COLUMNS_SORTED are no longer needed and are removed.
• MatConvert() now uses the matrix prefix when using option -matconvert_type.
• The bdiag matrix formats have been removed
• MatTranspose() now takes a new second argument MatReuse; also to do an inplace transpose use MAT_REUSE_MATRIX and pass the address of the first argument as the third argument (you can no longer pass 0).
• MATPLAPACK is removed; instead so long as PETSc is config/configure.py with --download-plapack the MATMPIDENSE matrix will automatically use the parallel plapack solvers
• The "parallel direct solver" matrix types like
  MATAIJSPOOLES are ALL gone. Now you use -pc_factor_mat_solver_package
  spooles etc or PCFactorSetMatSolverPackage() or if working directly with
  matrices, MatGetFactor(A,MATSPOOLES,...)
• Removed MatFDColoringSetRecompute(), now use MatFDSetFrequency(mat,-2) to have Jacobian computed at next Newton step but never again.
• Removed MatFDColoringSet/GetFrequency() and -mat_fd_coloring_freq since you can now use SNESSetLagJacobian() or -snes_lag_jacobian
• The PetscMap inside the Mat object is now allocated and reference counted and shared between vectors and matrices, this will decrease memory usage on systems with a large size.
• The first argument to MatXXXFactorSymbolic/Numeric() is now the factored matrix
• The fields global_rows,global_columns,local_rows,local_columns have been removed from the MatGetInfo() data. User MatGetSize(), MatGetLocalSize() to get that data

PC:

• PCHasApplyTranspose() was renamed to PCApplyTransposeExists().
• Added much faster matrix converter to use hypre preconditioners for PETSc AIJ matrices
• ILU preconditioner now defaults to use shift to prevent zero pivot
• ICC preconditioner now defaults to use Manteuffel shift to force positive definiteness
• Added -pc_factor_shift_in_blocks and PCFactorSetShiftInBlocks() made default for ILU
• PCApplyRichardson() now returns two additional arguments, the number of iterations performed and the reason it finished.

KSP:

• KSPSetConvergenceTest() now takes an optional final argument, destroy() that will be called to destroy the context passed in as the second to last argument
• KSPDefaultConverged() now has a KSPDefaultConvergedCreate() and KSPDefaultConvergedDestroy() that creates the default context to be passed KSPSetConvergenceTest() and the destroy to be passed in.
• Added KSPIBCGS a varient of Bi-CG-stab that has only 1 synchronization point rather than 3 in the usual.

SNES:

• SNESSetRhs() and SNESSetSolution() were removed. Use instead SNESSolve(snes, b, x) to pass right hand side (optional, can be PETSC_NULL) and solution (required).
• SNES now takes ownership of user-provided vector for right hand side and solution passed to SNESSolve(), and until the next SNESSolve() is called.
• SNES now takes ownership of user-provided vector for function evalutation passed to SNESSetFunction(), thus this vector can be safely destroyed after this call.
• SNESConverged_LS() and SNESConverged_LS() were removed. Use SNESDefaultConverged() or SNESSkipConverged() as convergence test.
• Option -snes_convergence_test was added, accepting values default|short, for setting SNESDefaultConverged() or SNESSkipConverged() as convergence test.
• Option -snes_no_convergence_test was removed. Use instead "-snes_convergence_test skip".
• Do not use the error PETSC_ERR_ARG_DOMAIN within your FormFunction() rather call SNESSetFunctionDomainError.
• The line search functions now take an additional argument xnorm.
• SNESSetConvergenceTest() takes another optional argument PetscErrorCode (*destroy)(void*) that will be called with SNESDestroy() to destroy the context.
• Added SNESSetLagPreconditioner(), -snes_lag_preconditioner, SNESGetLagPreconditioner()
• Added SNESSetLagJacobian(), -snes_lag_jacobian, SNESGetLagJacobian()

TS:

DA:

• Added DAGetNeighbors()
• Added DAGetOwnershipRanges()

DMMG:

• Added DMMGSetFromOptions(), this can only be used in conjunction with DMMGSetSNES(). You should call this sometime after DMMGSetSNES() is called, otherwise the SNES/KSPSetFromOptions() does not get called.
• Removed -dmmg_snes_monitor since now -snes_monitor and -snes_monitor_short will now do the proper indenting so this is no longer needed
• Removed -dmmg_ksp_monitor since now -ksp_monitor[_short) and -mg_levels_ksp_monitor[_short) etc will handle the proper indenting.
• Removed -dmmg_jacobian_period since you can use -snes_lag_jacobian

PetscViewer:

• added PETSC_VIEWER_STDOUT_(comm) for Fortran
• binary viewer now supports MPI IO for Vec loading and storing, use -viewer_binary_mpiio or PetscViewerBinarySetMPIIO() to enable it.
• Reworked the Matlab objects sreader and freader to be called PetscOpenSocket and PetscOpenFile, now PetscBinaryWrite and PetscBinaryRead can be used on the same socket connection to send matrix/vectors in both directions. See src/ksp/ksp/examples/tutorials/ex41.c and ex41.m

SYS:

• PetscLogFlopsNoCheck() no longer exists; you can simply call PetscLogFlops() it no longer has a return in it.
  

ExternalPackages: