GPUs: Allow ViennaCL and VECCUDA to be used concurrently.

This also gets rid of the weird veccuda.py.

CUSP: Removed VECCUSP and AIJCUSP as well as preconditioners.

SA-AMG from CUSP will be re-added in a follow-up commit,
made to work with AIJCUSPARSE and VECCUDA.

PetscDatatype: If using 64bit indices, map MPI_INT -> PETSC_ENUM

Merge branch 'maint'

gcc-8: replace sprintf with PetscSNPrintf to fix -Wformat-overflow warnings

/home/balay/petsc/src/sys/objects/init.c:518:24: warning: ‘%d’ directive writing between 1 and 11 bytes into a region of s

gcc-8: replace sprintf with PetscSNPrintf to fix -Wformat-overflow warnings

/home/balay/petsc/src/sys/objects/init.c:518:24: warning: ‘%d’ directive writing between 1 and 11 bytes into a region of size between 0 and 4095 [-Wformat-overflow=]
sprintf(name,"%s.%d",mname,rank);
^~
/home/balay/petsc/src/sys/objects/init.c:518:7: note: ‘sprintf’ output between 3 and 4108 bytes into a destination of size 4096
sprintf(name,"%s.%d",mname,rank);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

show more ...

Change PetscStrncat() to PetscStrlcat() to match BSD API where l represents total length

Update many uses of PetscStrcpy() and PetscStrcat() to the safer PetscStrncpy() and PetscStrlcat()

Commit-ty

Change PetscStrncat() to PetscStrlcat() to match BSD API where l represents total length

Update many uses of PetscStrcpy() and PetscStrcat() to the safer PetscStrncpy() and PetscStrlcat()

Commit-type: style-fix, feature
Thanks-to: Jed Brown

show more ...

Redefine how the n in PetscStrncat() works to make it more useful and not require tracking the current string size

Commit-type: feature

Merge branch 'karlrupp/feature-lazy-cublas-init'

CUDA: Encapsulated formerly non-namespaced global cublasv2handle.

Commit-type: bug-fix

cleanup PETSC_DESIRE_FEATURE_TEST_MACROS

Commit-type: style-fix

previously options checks in PetscErrorPrintfInitialize() were never made because
routine was called before options database was setup

Commit-type: bug-fix

proper docs for -check_pointer_intensity

Commit-type: documentation

Merge branch 'stefano_zampini/feature-glvis-homesh-periodic'

CUDA: CUBLAS initialization now lazy (at first vector creation).

Unconditional initialization of CUBLAS in PetscInitialize() is very costly for a high number of MPI ranks.
With this commit CUBLAS is

CUDA: CUBLAS initialization now lazy (at first vector creation).

Unconditional initialization of CUBLAS in PetscInitialize() is very costly for a high number of MPI ranks.
With this commit CUBLAS is only initialized if an actual GPU vector is used.
Also, global barriers are now removed and instead the local GPU driver deals with concurrent initializations.

show more ...

Doc: Small fix for log_view
Suggested-by: Jose E. Roman <jroman@dsic.upv.es>

cosmetic fixes

Merge branch 'barry/add-fortran-features'

Merge branch 'jed/update-mpi3-removed-or-deprecated'

Merge branch 'barry/update-mpi-keyval-create'

Merge branch 'barry/all-new-test-harness-branches'

Added -with-fortran-type-initialize since Fortran common blocks don't work with initialization
Removed C incorrectly added to C@*/
Added Fortran null support for MatGetOwnershipRangeColumns()

Commit

Added -with-fortran-type-initialize since Fortran common blocks don't work with initialization
Removed C incorrectly added to C@*/
Added Fortran null support for MatGetOwnershipRangeColumns()

Commit-type: bug-fix, feature

show more ...

Added F90 interfaces for several PetscObjectXXX() operations; Add various F90 support for Mat operations

MatCreateSchurComplement is missing a custom wrapper with
CHKFORTRANNULLOBJECT, so I can't pr

Added F90 interfaces for several PetscObjectXXX() operations; Add various F90 support for Mat operations

MatCreateSchurComplement is missing a custom wrapper with
CHKFORTRANNULLOBJECT, so I can't provide a PETSC_NULL_MAT for the
optional A22 block, where previously I could use PETSC_NULL_OBJECT. As
a workaround it seems I can provide it with a tMat(0) argument

* PetscObjectReference also does not have a wrapper. Calling it with
the %v attribute of the derived type seems to work however.

* Previously we had the following calling sequence:

call MatGetNullSpace(mat, nullsp, ierr)
if (ierr==0 .and. nullsp/=PETSC_NULL_OBJECT) then
call MatNullSpaceRemove(nullsp, b, ierr)
end if

Now there doesn't seem to be a PETSC_NULL_MATNULLSPACE and even if
there were, MatGetNullSpace isn't wrapped to return it.

* there is also no PETSC_NULL_RANDOM so no way to leave out the Random
object in VecSetRandom()

* not really important, but MatGetOwnerShip no longer works with
PETSC_NULL_INTEGER as advertized.

Commit-type: feature
Reported-by: Stephan Kramer <s.kramer@imperial.ac.uk>

show more ...

Fix example that could produce random order of vector outputs; remove redundant Vec header information for MPI DA vectors

Commit-type: bug-fix, testing-fix

filter on DM names set deep inside DM stuff, add better docs for how objects are automatically named

ommit-type: bug-fix, testing-fix, documentation

Update to MPI-2.0 versions of attribute interfaces

Some of these interfaces were removed in MPI-3. This compiles cleanly
with Open MPI 3.0 and is clean according to the style checks.

src/contrib

Update to MPI-2.0 versions of attribute interfaces

Some of these interfaces were removed in MPI-3. This compiles cleanly
with Open MPI 3.0 and is clean according to the style checks.

src/contrib/style/checks/mpi3-removed-all.sh
src/contrib/style/checks/mpi3-deprecated-all.sh

show more ...