GitLab

The place where the INSTALL() command was being run from did not have the
variables Trilinos_SOURCE_DIR or ${PROJECT_NAME}_INSTALL_LIB_DIR defined yet.

I must have had an existing cache file in place when I tested this.

Now this works from a fresh configure.

Build/Test Cases Summary
Enabled Packages: TeuchosCore
Disabled Packages: PyTrilinos,Pliris,Claps,STK,TriKota
0) MPI_DEBUG => passed: passed=52,notpassed=0 (2.49 min)
1) SERIAL_RELEASE => Test case SERIAL_RELEASE was not run! => Does not affect push readiness! (-1.00 min)

This is needed to allow xSDKTrilinos to build as its own CMake/TriBITS project
without having to point into the Trilinos source tree.  This allows
FindTPLHYPRE.cmake to get installed at the same time as TriBITS.

Since FindTPLHYPRE.cmake is now in TriBITS, it is not needed in Trilinos
anymore.

WIP: Squash against 'Add HYPRE to common_tpls ...

Build/Test Cases Summary
Enabled Packages: Teuchos
Disabled Packages: PyTrilinos,Pliris,Claps,STK,TriKota
0) MPI_DEBUG => passed: passed=122,notpassed=0 (5.96 min)
1) SERIAL_RELEASE => passed: passed=111,notpassed=0 (2.46 min)
Other local commits for this build/test group: f8ebfab4

With this, TriBITS now installs along with other enabled Trilinos packages.

The install is automatic and mandatory for now.  Later, we can add a variable
to allow the enable/disable of the install (like Trilinos_ENABLE_TriBITS or
Trilinos_INSTALL_TriBITS).

This is needed to allow xSDKTrilinos to build as a stand-alone CMake/TriBITS
project without having to point into the source tree.

The xSDKTrilinos package directly includes the file Tpetra_ETIHelperMacros.h.
When you build xSDKTrilinos as its own downstream CMake project, xSDKTrilinos
does not see the Trilinos source tree.  Therefore, you need to list
Tpetra_ETIHelperMacros.h in HEADERS so that it gets installed along with the
tpetracore library.

I am sure this was a simple omission.

To be safe for the addition of possible other *.h files, I just added the glob
of all *.h files in tpetra/core/src/.

Build/Test Cases Summary
Enabled Packages: TpetraCore
Disabled Packages: PyTrilinos,Pliris,Claps,STK,TriKota
0) MPI_DEBUG => passed: passed=80,notpassed=0 (16.30 min)
1) SERIAL_RELEASE => passed: passed=61,notpassed=0 (14.27 min)

Now that xSDKTrilinos is its own git repo, we might as well just add it as an
inserted package.  That requires no chagnes to anything under the xSDKTrilinos
at all.  An if xSDKTrilinos is cloned under Trilinos/packages/, then it will
be picked up automatically.

Build/Test Cases Summary
Enabled Packages: Teuchos
Disabled Packages: PyTrilinos,Pliris,Claps,STK,TriKota
0) MPI_DEBUG => passed: passed=122,notpassed=0 (6.05 min)
1) SERIAL_RELEASE => Test case SERIAL_RELEASE was not run! => Does not affect push readiness! (-1.00 min)

The dashboard is showing repeated failing tests for the householder
bordered solver unit tests.  I can't reproduce these, but I am
suspicious that it is the use of random bordering matrices.  I think it
is possible different matrices are being generated on different
processors.  This commit attempts to fix this by setting the seed
explicitly.

When using lids as an index, need to convert them from ZOLTAN_ID_PTR to lno_t.
@trilinos/zoltan2
(cherry picked from commit bcfc4c0c)

ParMETIS requires imbalance_tolerance to be > 1.0.
@trilinos/zoltan2
(cherry picked from commit 6a8136d8)

Added code to use this trait to AlgZoltan, allowing return on Zoltan
answer to Zoltan2 when Zoltan's Num_GID > 1 or Num_LID > 1.
Tests using lno_t=size_t revealed this bug.
(cherry picked from commit cc7288d1)

GIDs and LIDs are now copied correctly using ZOLTAN_SET_ID
(cherry picked from commit 1542909a)

After discussion with the Intrepid2 developers it was decided to delay the
release of Intrepid2.

Funded-by: IDEAS
Project: xSDK
Time: 2 hour including debugging and testing

Add a handful of simple scripts to deal with the multiple repos. These
aren't very fancy or generic, but they will get the job done for releases.

The TRILINOS_UNUSED_FUNCTION was being set improperly on windows due
to the intel compiler not supporting __unused__ on windows.

zoltan2:  release notes for Trilinos v12.6

the nsjac.mm file was being copied even from a tarball distribution.
This is a problem since that file is excluded from the tarball
intentionally. This prevents the copy when not in dev mode.

This file was previously being excluded for the tarball. It is now
a required part of the build so it needs to be distributed in the
tarball.

@trilinos/zoltan
(cherry picked from commit 9f27ca47)

Patch provided by SST team.
@trilinos/zoltan
(cherry picked from commit dec15961)

zoltan:  do not remove data directories from distribution if their data
is needed by zoltan2 tests

zoltan2:  when building test from a tarball release, exclude test that
uses many zoltan input files, as not all zoltan inputs are included in
tarball release.

@trilinos/zoltan2
@trilinos/zoltan
(cherry picked from commit 3c5f8f50)

-- Dirichlet conditions for arbitrary side ids.

-- Export/Import examples for converting between node-based and
   cell-based quantities.

-- File ASCII output for Matlab printing.

-- Runs with CompositeStep.

Requires some cleanup and modularization.  The reduced formulation
is very slow, for unknown reasons (will debug).

Fixed bug in ROL_OptimizationProblem.hpp.  Now
ROL_ObjectiveFromBoundConstraint.hpp is included.

- OptimizationProblem keeps track of whether it has slack variables with a bool
- OP's gradient and hessVec checks now work if there are slack variables
- InteriorPointStep interface with BoundConstraint now calls interface without bnd
  as the penalization is now handled through PenalizedObjective
- Upper bounds are ignored for elements with value >= ROL_INF
- Lower bounds are ignored for elements with value <= ROL_NINF
- Added Hock & Schittkowski problem 24 to ZOO (bound plus general inequality)
- Added test\step\test_11.cpp to test above problem.

StochasticProblem now build the appropriate vector based on the stored
ParameterList.  Changed stochastic objective function constructors to take in
const Teuchos::RCP instead of nonconst Teuchos::RCP.

the solution.  This change allows for easier use of the default solve.

Xyce has requested I allow Sacado to build with C++11 disabled.  It
appears the only unprotected usage of C++11 in Sacado is in the
implementation of mpl::vector, so I added guards around this too.  So
you will just not be able to use this class if C++11 is disabled.

Recent changes have made it possible to build Sacado without C++11
enabled. Removing it from the list of packages that require it since
a configure of Trilinos for Xyce fails on this check currently.

This test has been failing perpetually for quite a long time. There
is no reason to keep it on when it always fails.