New BQNLS algorithm added based on the BNK kernels

Quasi-Newton-Krylov methods added that utilize MatMult on LMVM matrices inside a Krylov solver

BQNTR algorithm added

LMVM matrix changes propaged to TAO algorithms and test outputs updated

Merge branch 'master' into knepley/feature-pc-patch

* master: (562 commits)
fix clang-analyzer warning
Developers manual never use system randomizers in PETSc and example source code.
missing

Merge branch 'master' into knepley/feature-pc-patch

* master: (562 commits)
fix clang-analyzer warning
Developers manual never use system randomizers in PETSc and example source code.
missing CHKERRQ()
DT: Reverting the tgamma change broke MPFR - I guess we need a nightly build with it in
Add a test for ex21
Fix error in ex21
Add missing fortran interface that handles null objects
DMComposite: add missing guards
MatPreallocate: fix overpreallocation of local rows
style fixes
MatView_MPI_DA: do not print twice the class prefix
MatSetOperation: stash the original MatView_XXX in the viewnative function pointer
DMCreateInterpolation_DA gracefully error instead of catching FPE signals
MatIncreaseOverlap: preserve communicator of index sets
MatMPIAIJ: fix indentation and missing CHKERRQ
ISBuildTwoSided: remove tabs and fix style
Fix documentation
PetscViewerDraw: fix unininitialized warning from valgrind
MatConvert_SeqAIJ_SeqBAIJ: remove error for non square matrices and add MAT_REUSE_MATRIX case
Mat: MAT_INPLACE_MATRIX can be used for transposition with rectangular matrices
...

Conflicts:
src/dm/dt/interface/dtfe.c
src/dm/impls/plex/plex.c
src/dm/impls/plex/plexfem.c
src/snes/examples/tutorials/ex13.c
src/snes/examples/tutorials/ex62.c
src/snes/examples/tutorials/makefile
src/snes/examples/tutorials/output/ex62_18.out
src/snes/examples/tutorials/output/ex62_19.out
src/snes/examples/tutorials/output/ex62_20.out
src/snes/examples/tutorials/output/ex62_20_parmetis.out
src/snes/examples/tutorials/output/ex62_21.out
src/snes/examples/tutorials/output/ex62_22.out
src/snes/examples/tutorials/output/ex62_23.out
src/snes/examples/tutorials/output/ex62_24.out
src/snes/examples/tutorials/output/ex62_25.out
src/snes/examples/tutorials/output/ex62_26.out
src/snes/examples/tutorials/output/ex62_27.out
src/snes/examples/tutorials/output/ex62_28.out
src/snes/examples/tutorials/output/ex62_29.out
src/snes/examples/tutorials/output/ex62_2d_p3_1.out
src/snes/examples/tutorials/output/ex62_30.out
src/snes/examples/tutorials/output/ex62_31.out
src/snes/examples/tutorials/output/ex62_32.out
src/snes/examples/tutorials/output/ex62_33.out
src/snes/examples/tutorials/output/ex62_34.out
src/snes/examples/tutorials/output/ex62_35.out
src/snes/examples/tutorials/output/ex62_36.out
src/snes/examples/tutorials/output/ex62_pc_simple.out
src/snes/examples/tutorials/output/ex62_pc_simplec.out
src/snes/examples/tutorials/output/ex62_quad_q2p1_full.out
src/snes/examples/tutorials/output/ex62_quad_q2q1_full.out
src/snes/utils/dmplexsnes.c

show more ...

Merge branch 'knepley/feature-error-logging' of bitbucket.org:petsc/petsc into knepley/feature-error-logging

Merge branch 'master' into knepley/feature-error-logging

* master: (324 commits)
DT: Reverting the tgamma change broke MPFR - I guess we need a nightly build with it in
Revert "tgamma: icc on wi

Merge branch 'master' into knepley/feature-error-logging

* master: (324 commits)
DT: Reverting the tgamma change broke MPFR - I guess we need a nightly build with it in
Revert "tgamma: icc on windows requires mathimf.h for tgamma."
setup routine no longer calls PetscCalloc5 if the arrays have already been allocated in a previous setup call
clarify SNESGetFunction()
Move MOAB docs to submansec so as not to clutter DM mansec
doc: remove matlab from doc generation dependency - i.e remove 'alldoc3' and use 'html' to generate corresponding '*.m.html' files.
Remove matlab pounders related cruft that is not required. Document the matlab code better and include an appropriate message in the help text.
Fixed typo.
Remove the inclusion of matlab in testing.
Finalizing comments and the work for the test problems.
Finished the testing and generation of plots. Some simple bug fixes addressed.
Set -Wl,-rpath for Mac OS Matlab mkl directory; don't pass -nojvm to matlab since it causes crashes on Barry's machine and is not needed
Added list of known problems with the code.
Added a comment on ignoring the return from Tao.
Finish with initial implementation of matlab interface to test TAO pounders.
Working through the matlab interface for pounders testing.
Fix typo in manual Reported-by: Jaroslaw Piwonski (CAU) <jpi@informatik.uni-kiel.de>
Fixed mkl_sparse_sypr() calls that were using an unsupported value in the matrix descriptor.
test harness: simplify code that checks for sequential-only tests
Fix the doc for call-backs drdyf drdpf
...

show more ...

Merge remote-tracking branch 'origin/jed/tao-typedef'

TaoType: use a const typedef like other PETSc types

Merge branch 'maint'

Merge remote-tracking branch 'origin/jed/matsolverpackage-deprecate-stubs/maint' into maint

add more deprecated stubs MatColoringTestValid, MatSTRUMPACKSetHSSRelCompTol, MatSTRUMPACKSetHSSMinSize, TaoDefaultMonitor

Merge remote-tracking branch 'origin/hongzh/fix-hessian-test'

Fix the gradient test

Gradients are not always evaluated at tao->solution. Pass the vector as an additional argument to TaoTestGradient().

Funded-by:
Project: FASTMATH adjoint
Time: 0.2 hours
Repor

Fix the gradient test

Gradients are not always evaluated at tao->solution. Pass the vector as an additional argument to TaoTestGradient().

Funded-by:
Project: FASTMATH adjoint
Time: 0.2 hours
Reported-by:
Thanks-to:

show more ...

Merge remote-tracking branch 'origin/denera/tao-merge-pgd-into-bncg'

PGD removed and buried into BNCG as a step type option

Remove TAOTEST macro

Funded-by:
Project: FASTMATH adjoint
Time: 0.1 hours
Reported-by:
Thanks-to:

Add -tao_test_gradient and -tao_test_hessian

The old finite-difference tests (enabled with -tao_type test) have the following limitations:
* Can only be evaluated at three points and often fail
* No

Add -tao_test_gradient and -tao_test_hessian

The old finite-difference tests (enabled with -tao_type test) have the following limitations:
* Can only be evaluated at three points and often fail
* Not work for matrix-free Hessian

The new feature allows for on-the-fly gradient/Hessian validation, supports matrix-free Hessian and improves the output. It follows SNESTestJacobian().

Funded-by:
Project: FASTMATH adjoint
Time: 4.0 hours
Reported-by:
Thanks-to:

show more ...

Merge remote-tracking branch 'origin/denera/tao-bounded-newton'

New Tao monitor added that prints detailed globalization info alongside convergence

BNTL added and tested with TAO examples

Bounded NTR added and working for unconstrained cases.

Unconstrained NLS transferred to bounded folder and structured to support code-reuse with other Newton-Krylov methods

Merge branch 'master' into knepley/feature-error-logging

* master: (221 commits)
dashboard: count and list example compile failures like:
fix fix-pr879-jczhang-optimize-vecscatter-multi-copy
a

Merge branch 'master' into knepley/feature-error-logging

* master: (221 commits)
dashboard: count and list example compile failures like:
fix fix-pr879-jczhang-optimize-vecscatter-multi-copy
add manual page cross links for SNESLineSearchSetPostCheck() and related routines
fix help manual of MatCreateSubmatrices() Reported-by: Tim Steinhoff <kandanovian@gmail.com>
Add manual pages for PetscRealPart() and PetscImaginaryPart()
fix compile error with missing MPI_C_DOBULE_COMPLEX
fix typo Reported-by: Febrian Setianto <febrian.setianto@gmail.com>
plex tutorial ex5: updated test requirements
DMPlexLoad_HDF5_Xdmf_Internal: fix for complex PetscScalar.
DMPlexLoad_HDF5_Xdmf_Internal: fix for 64-bit PetscInt.
DMPlexBuildFromCellList_Internal: use DMPlexInvertCell_Internal (with PetscInt* arg)
petscviewer.h: removed notice which is no longer relevant
DMPlexCreateFromFile: document option
check only if PETSC_USE_DEBUG
DMPlexBuildFromCellList_Internal optionally inverts cells.
ex5: new test 4 with -compare without repartitioning
ex5: new test 3 using -compare
ex5: HDF5 write+read moved to separate function
ex5: option -format processed in ProcessOptions()
ex5: added missing hdf5 requirement
...

Conflicts:
include/petsclog.h

show more ...

Merge branch 'psanan/manpages-notes-newlines' [PR #932]

* psanan/manpages-notes-newlines:
A large number of man pages are improperly formatted because of the lack of a newline after "Notes:", "Fortr

Merge branch 'psanan/manpages-notes-newlines' [PR #932]

* psanan/manpages-notes-newlines:
A large number of man pages are improperly formatted because of the lack of a newline after "Notes:", "Fortran Notes:", or "Developer Notes:".
This batch adds those newlines, with 4 spaces of indentation afterwards.
The uniform indentation arguably makes the man pages a little less readable in the source itself,
but overall this hopefully has a large positive impact on the readability of the HTML pages.

show more ...