SF tutorial ex1 tests new function and shows motivation

PetscSFComputeMultiRootOriginalNumbering() impl.

Merge branch 'jose/cuda-duplicate/maint' [PR #1147]

* jose/cuda-duplicate/maint:
Fix uninitialized memory on VecDuplicate_MPICUDA

fortran type checking

not clear what prevents auto-generation

Fix uninitialized memory on VecDuplicate_MPICUDA

Merge branch 'knepley/feature-section-field-major' [PR #1131]

* knepley/feature-section-field-major:
PetscSection: Cleanup point iteration in SetUp()
Section: Add a point major flag to allow differe

Merge branch 'knepley/feature-section-field-major' [PR #1131]

* knepley/feature-section-field-major:
PetscSection: Cleanup point iteration in SetUp()
Section: Add a point major flag to allow different dog orderings
Does not do anything right now
Section: Added PetscSectionSetFromOptions()
Plex: When Plex creates a section automatically, call SetFromOptions
Could condition this on whether the DM has been set from options
Section: During SetUp(), change dog ordering based on the pointMajor flag
Plex ex27: Check Section dof ordering

show more ...

Vec: raise error if get array is not present

Fix LOCDIR paths

Section: Use correct format string for PetscInt

Change outdated HAVE_VECCUDA macro to simply HAVE_CUDA for clarity/ease of maintenance.

Commit-type: style-fix

Merge branch 'barry/refactor-vecscatter-2'

Remove memory leaks of XXXRegister() is called but XXX object is never used

This means adding YYYInitialPackage() at the beginning of XXXRegister() routines

Minor fixes of missing prototypes

Commi

Remove memory leaks of XXXRegister() is called but XXX object is never used

This means adding YYYInitialPackage() at the beginning of XXXRegister() routines

Minor fixes of missing prototypes

Commit-type: style-fix
Reported-by: Jed Brown <jed@jedbrown.org>
Thanks-to: Fande Kong <fande.kong@inl.gov>

show more ...

Only register MPI3 VecScatter if they are built

Part 7 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: bug-fix, portability-fix
Reported-by: nightly tests

Add missing makefiles for VecScatter implementations

Part 6 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: bug-fix

Changed VecScatterCreate() to take a communicator and return scatter as other first class PETSc objects do.
Changed old VecScatterCreate() to VecScatterCreateWithData()

Completed basic refactorizati

Changed VecScatterCreate() to take a communicator and return scatter as other first class PETSc objects do.
Changed old VecScatterCreate() to VecScatterCreateWithData()

Completed basic refactorization of VecScatter code, ready for new implementations to be added.

(There is still some code duplication between the VecScatter MPI1 and MPI3 code but it can be fixed at anytime without requiring changes to the VecScatter API.

Part 5 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: style-fix

show more ...

Added VecScatterSetUp() and converted implementations to use it to compute scatter information

Part 4 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: sty

Added VecScatterSetUp() and converted implementations to use it to compute scatter information

Part 4 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: style-fix

show more ...

Pull "MPI1" and "MPI3" specific code out of common VecScatter impls code
This allows future implementations to use the common code also without changing the common code

Part 3 of a refactorization o

Pull "MPI1" and "MPI3" specific code out of common VecScatter impls code
This allows future implementations to use the common code also without changing the common code

Part 3 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: style-fix

show more ...

Seperate VecScatter implementation code into three subdirectories seq, mpi1, and mpi3

Part 2 of a refactorization of VecScatter to be treated as a first class PETSc object

Commit-type: style-fix

doc: fix LOCDIR wrt broken URL from MatHYPRESetPreallocation.html

Reported-by: Fande Kong <fdkong.jd@gmail.com>

Tested with:
make ACTION=sourcelist BASE_DIR=$PWD alltree_makefile
ls `cat files` > /

doc: fix LOCDIR wrt broken URL from MatHYPRESetPreallocation.html

Reported-by: Fande Kong <fdkong.jd@gmail.com>

Tested with:
make ACTION=sourcelist BASE_DIR=$PWD alltree_makefile
ls `cat files` > /dev/null

show more ...

Merge branch 'master' into arcowie-rem/feature-error-logging

* master: (345 commits)
Update dev notes for the coming release
Increase patchlevel to 3.9.4
doc fixes
Added some AIJSELL tests t

Merge branch 'master' into arcowie-rem/feature-error-logging

* master: (345 commits)
Update dev notes for the coming release
Increase patchlevel to 3.9.4
doc fixes
Added some AIJSELL tests that mirror the SELL tests.
Bib: More refs
Removed use of hardwired -1 to represent PETSC_NULL_XXX object, since the null value is now 0.
fix memleaks flagged by valgrind
ex62: disable long running test and workaround differences in iteration counts on various builds.
Add FAQ we can point to when users get error messages such as: "PETSc was configured with one OpenMPI mpi.h version but now appears to be compiling using a different OpenMPI mpi.h version"
swith test from mpich to openmpi as mpich ch3:sock is unuserable per PetscShmCommGet()
DM: Forgot to move include statement
fix nightlybuild issues - disable hypre [for openmp] and fix uninitialized variables
tests: enable openmp and hwloc with one of the mumps tests
Default BoomerAMG smoothers
Configure: Revert changes from debb90c, so that we are back to the original Libmesh
Configure: Reverted change for Libmesh from 83cd5d7
Split vecimpl.h into vecimpl.h and vecscatterimpl.h
SNES ex62: Turned off some tests for other precisions and shortened output of sensitive tests
Use mmap to allocate the shared pthread barrier if available
Add configure test for functional mmap
...

Conflicts:
include/petscconvest.h
src/snes/examples/tutorials/ex13.c
src/snes/interface/snes.c
src/snes/utils/convest.c

show more ...

VecScatter: add tests for VIENNACL and VECCUDA

VecScatter_MPI1: fix bug

Need to check for CUDA types, not VIENNACL!

VECCUDA: fix bug in NORM_INFINITY

cublasIXamax gives back a 1-based index
Also, fix PetscScalar->PetscReal conversion
Added test

VecGetSubVector: Use VecPlaceArray instead of VecCreateXXXWithArray

This enables VECCUDA and VECVIENNACL: added test

Merge remote-tracking branch 'origin/jczhang/feature-mumps-omp'