petsc_3_14_5-gnu-openmpi4-hpc-3.14.5-bp153.1.7

- Update to version 3.14.5.
  * Documentation & man page fixes, removal of non-existing functions.
  * CPARDISO: stick to OpenMPI BLACS when needed
  * bugfix for MatMatMultSymbolic_MPIAIJ_MPIDense() when Bbn1 = 0.
    Bb (column block size) zero lead to infinite loop in
    MatMatMultNumeric_MPIAIJ_MPIDense() with n=0.
  * petscviewer: fix the fortran bindings of XXXViewFromOptions
    when called with obj=PETSC_NULL_XXX.
  * PetscOptionsPrefixPush: Accept numbers if the stack is not
    empty.
  * PetscOptionsClearValue: Also clear 'help_intro' slot.
  * PetscOptionsDestroyDefault: Fix possible NULL-dereference.
  * PetscOptionsUsed: Fix for case-insensitive string comparison.
  * Do not use MPI_Bcast() on a single rank - This allows direct
    reading  of much larger files, in, for example, MatLoad_SeqAIJ_Binary.
  * PCHPDDM: fix for KSPLSQR.
  * DMPlexVTKWriteAll_VTU: numerous bugfixes.
  * MatBAIJ: FIx specialization for size 9.

- Updated to version 3.14.4 (jsc#SLE-14959)
  * versionToTuple: ignore leading strings in package versions.
    i.e 5.3.3consortium -> 5.3.3
  * Add error checks for integer overflow for
    MatSeqXAIJSetPreallocation when the user provides nnz[].
    Otherwise PetscMalloc() can produce error messages with requests to
    absurds amount of memory.
    Improve the error message for PetscIntMultError() and other casting
    routines and add more seealso entries to the manual pages.
  * Update dmlocalts.c to prevent zeroing out of locX_t after being computed
    in TSComputeIJacobian_DMLocal.
  * nargs.Arg: Remove interactiveness; provide useful error when key is not
    found.
  * DMPlexLandau: fix implicit conversion.
  * Allow users to mistaken use openmpi or mpich when setting mpi or mpi-dir
    and consistently use local variable 'name'
  * Consume -help in PetscOptionsSetValue() so it can be set from RC file etc.
  * Enable search patterns to have 'src*' at start.
  * Fix Matlab int32 to match changed changes in PETSc data types.
  * Generate error instead of crashing when estimating eigenvalues for
    Chebyshev, no right hand side vector is available and the user requested
    to not use a noisy right hand side.
  * Fix pcmgsetlevels() for Fortran to properly convert the C MPI communictors
    to Fortran.
  * Fix PCMGSetLevels_MG() to support the use of MPI_COMM_NULL for
    communicators that that live on a subset of the orginal communicator.
  * Add additional documentation material to PCMGSetLevels() explaining how
    one must handle the restriction and interpolation operations that move
    between different levels which have a different number of MPI ranks.
  * Call PetscCUDAInitializeCheck() in
    MatConvert_SeqAIJ_SeqAIJCUSPARSE since it uses cuSparse().
  * Missing this sometimes produces errors in the CU tests, but other times
    it passes. I don't understand why it is not needed.
  * checkbadSource now prints line numbers of where it finds problems.
  * Add support to Fortran interface for PCASMGetSubKSP() and PCGASMGetSubKSP()
    to use PETSC_NULL_INT for first two arguments.
  * Missing define in landaucu.cu.
  * Fix for PCView_BJacobi() case when neither if case is statisfied.
  * DMCreateInterpolation() requires a custom Fortran stub because it RETURNS
    an optional vector.
  * Add -da_refine to manual page for DMDASetRefinementFactor().
  * AIJMKL: cannot print a scalar in case of complex.
  * Fix memory leaks in TSTrajectoryCreate_Memory() and possibly bugs
  * Don't generate an error if the user does not free all objects while using
    SuperLU_DIST.
  * -ksp_view -ksp_type gmres -pc_type gasm fails on one process. Fixed by
    removing special case handling of size 1 that tried to reuse
    orginal viewer as subviewer which does not work.
  * Add KSPGMRESSetBreakdownTolerance() and -ksp_gmres_breakdown_tolerance
  - to restore old GMRES behaviour for MOOSE.
  * Fix bug introduced by PCASMSetSubMatType() when MAT_REUSE_MATRIX is
    used with a different type of pmat compared to pmat[i].
  * Always destroy previously created pmat[i] and use   MAT_INITIAL_MATRIX
  * Clarify integers set with to ISCreateBlock() and ISBlockSetIndices() in
    manual pages.
  * Remove ierr = VecSetInf(ksp->vec_sol);CHKERRQ(ierr); from
    KSPSetUp_Chebyshev() since ksp->vec_sol may not yet exist and it is not
    needed since VecSetInf(ksp->vec_sol) is set in KSPSolve() just before
    the actual solve if ksp->reason = KSP_DIVERGED_PC_FAILED.
  * Remove mention of DMDAGetLocalInfoF90() since it does not exist, Fortran
    version has same name as C version.
  * MATSHELL: matrix sizes may not be known at creation time.
  * Pass PetscDrawSetSave() and PetscDrawSetSaveFinalImage() to
    PetscViewerGetSubViewer_Draw(). Otherwise MatView() on parallel matrices,
    for example, will not generate an image file.
  * config MPI: fix type of self.libdir (which is passed to os.path.join).
  * MUMPS: add -DBLR_MT when configuring --with-openmp.
  * missing PDIPM and typo in BQNLS for fortran imports.
  * Fixed issue where scalapack matrix is created with a communicator different
    from MPI_COMM_WORLD.
  * Fix PetscErrorCode typos in Fortran stubs.
  * Remove interface declarations for functions that don't exist.
  * SNESMultiblock: add back into build (had been missing -  still untested).
  * MATDAAD: remove vestigial references to deleted DAAD.
  * Fix device atomics with 64-bit indices and prefer long long over int64_t
  * petsc4py.py: set download-petsc4py-commit to None to overcome missing key
    error.
  * petsc4py: Update Cython build
  * Set `language_level` to `3str` if supported (Cython >= 0.29)
  * Initial preparations for upcoming release Cython 3.0
  * Require Cython >= 0.24, older versions do not longer work
  * Improve python-related entries in configure --help.
  * add missing --download-petsc4py option.
  * remove python options other than --with-python-exec
    which have no effect.
  * move --with-python-exec to PETSc section.
  * PetscPythonFindLibrary: One more way to get pythonlib.
    Hotfix for case when LDLIBRARY yields libpython<version>.a
    erroneously.
  * SLEPc: pass PYTHONPATH for slepc4py.
  * Add manual page for MATMPIDENSE.
  * Fix outdated mention of "-snes_ksp_ew_conv" in users manual.
  * Improve manual pages for MatGetValuesLocal() and MatGetValues() to clarify
    what values may be obtained with these routines.
  * Add to manual pages the use of call XXXView(x,0) from the debugger.
  * KSPGCR: add man pages for setting and getting restart.
  * Put the docs for -snes_test_jacobian in SNESSetFromOptions() instead of
    hidden away in developer routine.
  * Improve clarity of docs for PETSC_VIEWER_SOCKET with Matlab.
- Refactored patches:
  * petsc-3.3-p2-no-rpath.patch
  * petsc-3.7-fix-pastix-detection.patch

- Add build support for gcc10 to HPC build (bsc#1174439).
- Add openmpi4 flavors.

- Change 'Requires:' to other HPC packages to %requires_eq to depend
  on the exact version. This should take care of HPC packages ignoring
  proper ABI versioning.

- Update to version 3.14.1
  * https://www.mcs.anl.gov/petsc/documentation/changes/314.html
- Refresh patches

- Make sure, doc package is built for SLE (bsc#1173269).

- Force use of python3 on Tumbleweed. As the python2 executable was
  also set in petsc's configuration, dependent packages no longer
  depend on python2.
- Fix she-bangs in python scripts
- Only require python{2,3}-base packages instead of full packages.

- Remove unneeded dependency from HPC devel package.

- Update to version 3.12.2 (jsc#SLE-8505)
  * https://www.mcs.anl.gov/petsc/documentation/changes/312.html
- Remove petsc-3.3-p2-fix-shared-libs-sonames.patch
  link created directly in %install section
- Remove petsc-3.3-p2-dont-check-for-option-mistakes.patch
  not needed.
- Remove petsc-3.3-fix-error-detection-in-makefile.patch
  not needed.
- Add non-HPC build with openmpi3
- Add HPC builds for gcc8 & gcc 9 (jsc#SLE-7766 & jsc#SLE-8604).
- Disable openmpi1 builds for SLE/Leap > 15.1.
- Enable openmpi3 builds for Leap and SLE > 15.1 (jsc#SLE-7773).

- Reflect renaming of openmpi packages to openmpi1 in Tumbleweed,
  fixes unresolvable packages.
- Fix copy'n'paste error, mpi_vers vs. mpi_ver
- Add non-HPC openmpi2 flavor.

- Update to version 3.11.3 (jsc#SLE-8505)
  - https://www.mcs.anl.gov/petsc/documentation/changes/311.html
  - Binaries have changed path from bin/ to lib/petsc/bin/
- Patches removed, merged by upstream
  * Python-Fix-a-number-of-scripts-to-by-Python-3-compliant.patch
  * petsc-fix-nonmpi.patch
- Patches refreshed
  * petsc-3.3-p2-fix-shared-libs-sonames.patch
  * petsc-3.3-p2-no-rpath.patch
  * petsc-3.3-p2-dont-check-for-option-mistakes.patch
  * petsc-3.3-fix-error-detection-in-makefile.patch
  * petsc-3.7-fix-pastix-detection.patch
- Set -fPIC also on x86_64

- Upped memory constraints to 6G for non-ppc systems as well,
  leaving the special ppc treatment in place in case we need to up it
  some more there.

- update _constraints for PowerPC more than 4G memory
- Add BuildRequires libopenblas-gnu-hpc >=  0.3.6 for hpc to  avoid
  to build when openblas library not available.

- Disable pastix also in openmpi flavor, same as for serial and HPC flavors.
  Pastix is only built with openmpi, and not available in Factory.

- Fix building non-MPI flavors of packages depending on petsc
  Add petsc-fix-nonmpi.patch
- Do not move headers to /usr/include and symlink, causes problems with
  - isystem

- Fix spec for non-HPC, reenable serial and non-HPC openmpi build.
- Disable HYPRE for non-HPC, same as for HPC.

- Consolidate use of openmpi1, openmpi2, openmpi3.
- Do not build non-HPC on SLE.

- (Temporarily) disable fpr S390* as a check test for netcdf fails
  currently on BE (bsc#1091474).

- Add _constraints to avoid OOM errors

- Do not require 'Modules' for HPC build: seetings from
  /etc/profile.d/lmod.[c]sh will be overwritten by
  /etc/profile.d/modules.[c]sh installed by Modules
  (bsc#1089865).