SF: shorten event names to fit within 16-char limit

There is no convention for including the Petsc namespace in event names.

Insure that all ASCII viewers for PETSc objects call PetscObjectPrintClassNamePrefixType()

IS: Fix compiler warning

SF: guard MPI_COMBINER_DUP

Also trust MPI_VERSION to avoid confusing memory corruption if a user
forgets to reconfigure after getting this patch.

IS: Forgot to protect HDF5

IS: Forgot to protect HDF5

IS: Added ISLoad() and an HDF5 implementation
- I know this only works for general now

IS: Added HDF5 viewer to general

IS: Changed over to using PetscLayout
- Could reorganize code to be cleaner now

Vec+HDF5: The HDF5 link test can only look 1 level deep. These hard working programmers are not paid enough to handle multiple levels.

simplify inversion of identity IS

PetscSection: Added an optional permutation for the chart
- Removed the unnecessary atlasLayout
- This permutation will allow people to segregate ghost unknowns at the end of their numbering,
and

PetscSection: Added an optional permutation for the chart
- Removed the unnecessary atlasLayout
- This permutation will allow people to segregate ghost unknowns at the end of their numbering,
and also to get around the stratification of point numbering in Plex

show more ...

ViewerHDF5: Try to set vector blocksize if its different from the HDF5 vector
- We changed how undefined blocksize is handled

PetscSection: Added PetscSectionHasConstraints()

SF: add arithmetic op types for contiguous types

This allows operations like MPI_SUM on
MPI_Type_contiguous(3,MPI_DOUBLE). The total sizes are classified by
their largest divisor in (1,2,3,4,5,7).

SF: add arithmetic op types for contiguous types

This allows operations like MPI_SUM on
MPI_Type_contiguous(3,MPI_DOUBLE). The total sizes are classified by
their largest divisor in (1,2,3,4,5,7). The small size is specified at
compile-time, allowing the compiler to unroll.

show more ...

Merge branch 'knepley/fix-fem-multifield'

* knepley/fix-fem-multifield:
DMDT: Fixed Fortran PetscQuadrature definition
DMPlex ex3: Added Q2 tests - Made DA tests only 2x2 - Changed output format

Merge branch 'knepley/fix-fem-multifield'

* knepley/fix-fem-multifield:
DMDT: Fixed Fortran PetscQuadrature definition
DMPlex ex3: Added Q2 tests - Made DA tests only 2x2 - Changed output format - Quad tests now actually work
DMDA: Added DMDAComputeL2GradientDiff()
PetscSpace: Fixed polynomial derivative evaluation for quads
DMDA: Fix closure operations in 2D
DMPlex: Interpolation cleanup - Remove unused PetscSpace - Fix uninitialized variable
PetscFE: Fix PetscBool handling
SNES ex62: Updated output
DMPlex ex3: Use PetscFECreateDefault() and correct tensor tests
PetscFE: Fix PetscFECreateDefault() - The space should default to tensor on a non-simplex - The quadrature should be tensor Gauss on a non-simplex
DT: Added PetscDTGaussTensorQuadrature()
DT: Added citation for Gloub+Welsch method
PetscSection: Now global section can also be viewed
DMPlex: Allow NULL function for ProjectFunction()
PetscDualSpace: Added doc for PetscDualSpaceApply() - Cherry-picked from df2d8bb
PetscFE: Fixed PetscFECreateDefault() - Setup FE - Minimum quad order - Destroy quad after setting
PetscFE: PetscFECreateDefault() now takes a number of components and a reference cell type - Cherry-picked from a255314
PetscFE: Added PetscFECreateDefault() - Cherry-picked from 8919f5a
PetscFE: Allow lowest order cell Lagrange dual basis functions
DMPlex: Fix preallocation with cell variables - I had an incorrect loop optimization

Conflicts:
src/dm/impls/plex/plexpreallocate.c

show more ...

PetscSection: Fixed PetscSFDistributeSection() for the case of an empty partition

PetscSection: Now global section can also be viewed

IS: Fix comm for ISLocalToGlobalMappingApplyIS()
Reported-by: Vincent Frans Maria De Groof <Vincent.De-Groof@uibk.ac.at>

Merge branch 'karlrupp/update-viennacl-1.5.1'

* karlrupp/update-viennacl-1.5.1: Updates the ViennaCL bindings to use
the latest 1.5.1 release. Performance improvements in VecMDot()
for ViennaC

Merge branch 'karlrupp/update-viennacl-1.5.1'

* karlrupp/update-viennacl-1.5.1: Updates the ViennaCL bindings to use
the latest 1.5.1 release. Performance improvements in VecMDot()
for ViennaCL types. Now providing an implementation for
compressed CSR, enabling full MPI runs.

show more ...

Merge branch 'jed/lazy-blocksize'

* jed/lazy-blocksize:
Mat & Vec: make block sizes lazy
PetscLayoutSetBlockSize: allow PETSC_DECIDE for use when block size has already been set

Merge branch 'qulogic/vecview-verbosity'

* qulogic/vecview-verbosity:
VecView_MPI_ASCII: avoid PetscObjectGetName unless really needed
VecView_*_ASCII: close leak for INFO and INFO_DETAIL
Chan

Merge branch 'qulogic/vecview-verbosity'

* qulogic/vecview-verbosity:
VecView_MPI_ASCII: avoid PetscObjectGetName unless really needed
VecView_*_ASCII: close leak for INFO and INFO_DETAIL
Change verbosity of VecView in parallel to match serial.

show more ...

Merge branch 'karlrupp/fix-cuda-viennacl-vecaypx'

karlrupp/fix-cuda-viennacl-vecaypx: Fixes VecAYPX_SeqCUSP and VecAYPX_SeqViennaCL
which did not handle the case alpha == 0 correctly (both operati

Merge branch 'karlrupp/fix-cuda-viennacl-vecaypx'

karlrupp/fix-cuda-viennacl-vecaypx: Fixes VecAYPX_SeqCUSP and VecAYPX_SeqViennaCL
which did not handle the case alpha == 0 correctly (both operation and FLOP count).

show more ...

ViennaCL: Added support for compressed CSR -> multiple MPI ranks possible.

ViennaCL 1.4.x did not provide a compressed CSR format.
With 1.5.x, there is now a dedicated type available, which is now u

ViennaCL: Added support for compressed CSR -> multiple MPI ranks possible.

ViennaCL 1.4.x did not provide a compressed CSR format.
With 1.5.x, there is now a dedicated type available, which is now used for
distributed matrix-vector products.
Slight performance enhancements in MatMultAdd_SeqAIJViennaCL() still possible
by fusing the two kernels involved -> ViennaCL 1.6.x

show more ...

Vec: Updated tests/ex43 to test results of dot() vs. mdot().

Since this is supposed to be a simple test of mdot() routines,
this commit adds additional checker logic to quickly evaluate that the
mdo

Vec: Updated tests/ex43 to test results of dot() vs. mdot().

Since this is supposed to be a simple test of mdot() routines,
this commit adds additional checker logic to quickly evaluate that the
mdot implementations are sane.
Particularly important for GPU implementations.

show more ...