ViennaCL: Updated VecMDot_SeqViennaCL() to use fast mdot() implementation.

The older implementation relied on an iterated application of the dot() product.
ViennaCL 1.5.0 and above provide an optimi

ViennaCL: Updated VecMDot_SeqViennaCL() to use fast mdot() implementation.

The older implementation relied on an iterated application of the dot() product.
ViennaCL 1.5.0 and above provide an optimized mdot() routine similar to what
has been implemented for the CUSP bindings already.

show more ...

tests: split up C++ tests [which can run with clanguage=c with C++ compiler build]

build: split cxx [legacy] build target via SOURCECXX

doc: fix errors from builddist.log

in ex6.c warn against using low level PetscBinaryRead/Write() and recommend VecView/VecLoad() instead

Merge branch 'knepley/fix-plex-refine-sf'

* knepley/fix-plex-refine-sf:
SF: Added additional checking
DMPlex: Fixed SF created during refinement - Remote edge had wrong starting point - Forgot a

Merge branch 'knepley/fix-plex-refine-sf'

* knepley/fix-plex-refine-sf:
SF: Added additional checking
DMPlex: Fixed SF created during refinement - Remote edge had wrong starting point - Forgot an entry for hybrid edge

show more ...

Mat & Vec: make block sizes lazy

In the past, MatSetUp eagerly set the block sizes (to 1) if not already
set. This prevents use of the "convenience" constructors because they
call MatSetUp before r

Mat & Vec: make block sizes lazy

In the past, MatSetUp eagerly set the block sizes (to 1) if not already
set. This prevents use of the "convenience" constructors because they
call MatSetUp before returning. We could add bs as an explicit argument
to those constructors, but block size is an extraneous concept for most
users. The default value of the block size is -1, so now we use
PetscAbs(map->bs) anywhere it is not known in advance that bs has been
explicitly set. This allows block sizes to be lazy, and they can be set
arbitrarily late, so long as the new size is compatible with the
existing distribution. For (S)BAIJ matrices, the block size is
structural so it gets set explicitly and can thus be referenced directly
rather than via MatGetBlockSize or wrapped in PetscAbs.

show more ...

PetscLayoutSetBlockSize: allow PETSC_DECIDE for use when block size has already been set

This slightly simplifies logic in some places, where otherwise we would
have to MatGetBlockSize() and pass it

PetscLayoutSetBlockSize: allow PETSC_DECIDE for use when block size has already been set

This slightly simplifies logic in some places, where otherwise we would
have to MatGetBlockSize() and pass it back in.

show more ...

Merged in jose/fix-examples (pull request #143)

Fix some C examples in Vec and KSP

Merge branch 'barry/fix-chaco'

vec/ex18.c: Correctly fill the last entry of vector xend

vec/ex1.c: Remove PetscFunctionBegin from main()

opstring was never initialized

SF: Added additional checking

Merge branch 'master' into balay/tao-to-petsc

Conflicts:
src/dm/examples/tests/ex42.c
src/dm/examples/tutorials/ex65dm.c
src/ksp/ksp/examples/tutorials/ex42.c
src/snes/examples/tutorials/cahn-hi

Merge branch 'master' into balay/tao-to-petsc

Conflicts:
src/dm/examples/tests/ex42.c
src/dm/examples/tutorials/ex65dm.c
src/ksp/ksp/examples/tutorials/ex42.c
src/snes/examples/tutorials/cahn-hilliard-schemes/ex1.c
src/snes/examples/tutorials/ex54.c
src/snes/examples/tutorials/ex55.c
src/snes/examples/tutorials/ex60.c
src/snes/examples/tutorials/ex61.c
src/snes/examples/tutorials/ex63.c
src/snes/examples/tutorials/ex633d_db.c
src/snes/examples/tutorials/ex64.c
src/snes/examples/tutorials/ex65.c
src/snes/examples/tutorials/ex653d.c
src/ts/examples/tutorials/ex23.c

show more ...

TAO: s/PETSC_NULL/NULL/ in manual

Touches a few non-TAO files.

Merge branch 'master' into balay/tao-to-petsc

fixes to tao to compile in single precision

Merge branch 'jose/mat-vec-fixes' into maint

* jose/mat-vec-fixes:
Correctly check input type of Mat argument in MatMPIAIJGetLocalMat
Fix Matlab viewer for parallel complex vectors

stop using taosolver.h includes

move tao vector operations over to Vec directory, fix a couple names and calling sequences

VecScatterCreateToZero: fix man page

Reported-by: Dave May <dave.mayhem23@gmail.com>

Section: Small doc fix

fixed quote marks for includes to <>

Merge branch 'knepley/fix-plex-closure-index'

* knepley/fix-plex-closure-index:
DMPlex: work around -Wmaybe-uninitialized
DMPlex: Small fix
PetscSection: Added PetscSectionGetField()
DMPlex:

Merge branch 'knepley/fix-plex-closure-index'

* knepley/fix-plex-closure-index:
DMPlex: work around -Wmaybe-uninitialized
DMPlex: Small fix
PetscSection: Added PetscSectionGetField()
DMPlex: Reorganization of closure operations - Rename DMPlexVecGetClosure_Static() —> DMPlexVecGetClosure_Depth1_Static() - Added special cases with/out fields - Removed the need for field offsets by interchanging loops over points and fields - Fixed bug where field offsets were not calculated if an array was passed in - Made all local functions static - Using SNES ex62 to verify changes, and DMPlex ex9 to time them
DMPlex: Fixed object check
DMPlex ex9: Now test VecClosure with index - Made functions static
DMPlex: Optimization for Get/SetClosure - Only need field offsets if there is more than one field
PetscSection: Change clIndices member to clPoints - Added some docs
SNES ex12: Use ViewFromOptions correctly
DMPlex: Use closure index in DMPlexVecSetClosure() and DMPlexMatSetClosure() - Moved depth 1 optimization to static function
DMPlex: Reorgnized closure index - Now index stores points and orientations instead of offsets, so it can be used for updates as well - Moved index creation to plexindices.c - Moved optimization for depth 1 mesh to a static function
DMPlex: Fixed two bugs with DMPlexCreateClosureIndex() - Must make index for entire DM chart - Need to update cloff when no fields are present

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

show more ...