Xpetra library build failures breaking all ATDM CUDA builds
Created by: bartlettroscoe
CC: @trilinos/xpetra, @csiefer2, @rppawlo
Description
The Xpetra libraries failed to build with the CUDA builds on 'shiller', 'white', and 'ride' this morning as shown at:
with build failure details shown at:
This showed errors starting out with:
/home/rabartl/WHITE/ATDM_Driver/Trilinos-atdm-white-ride-cuda-debug/SRC_AND_BUILD/Trilinos/packages/xpetra/src/MultiVector/Xpetra_MultiVectorFactory.hpp(156): error: object of abstract class type "Xpetra::TpetraMultiVector<double, int, int, Xpetra::EpetraNode>" is not allowed:
pure virtual function "Xpetra::MultiVector<Scalar, LocalOrdinal, GlobalOrdinal, Node>::isSameSize [with Scalar=double, LocalOrdinal=int, GlobalOrdinal=int, Node=Xpetra::EpetraNode]" has no overrider
...
and
/home/rabartl/WHITE/ATDM_Driver/Trilinos-atdm-white-ride-cuda-debug/SRC_AND_BUILD/Trilinos/packages/xpetra/src/MultiVector/Xpetra_MultiVectorFactory.hpp(156): error: object of abstract class type "Xpetra::TpetraMultiVector<double, int, int, Xpetra::EpetraNode>" is not allowed:
pure virtual function "Xpetra::MultiVector<Scalar, LocalOrdinal, GlobalOrdinal, Node>::isSameSize [with Scalar=double, LocalOrdinal=int, GlobalOrdinal=int, Node=Xpetra::EpetraNode]" has no overrider
...
Looking at the new commits pulled in these builds today at:
the likely breaking commit is from the set of commits:
cd1628f: Tpetra: Fixing merge issue in MultiVector
Author: Chris Siefert <csiefer@sandia.gov>
Date: Mon Mar 19 14:24:46 2018 -0600
M packages/tpetra/core/src/Tpetra_MultiVector_decl.hpp
M packages/tpetra/core/src/Tpetra_MultiVector_def.hpp
009fafc: Xpetra: Updates to BlockedMultiVector
Author: Chris Siefert <csiefer@sandia.gov>
Date: Mon Mar 19 13:29:33 2018 -0600
M packages/xpetra/src/BlockedMultiVector/Xpetra_BlockedMultiVector.hpp
d9771e2: Tpetra: Adding Stokhos-compatible size checker to MultiVector
Author: Chris Siefert <csiefer@sandia.gov>
Date: Thu Feb 22 12:49:18 2018 -0700
M packages/tpetra/core/src/Tpetra_MultiVector_decl.hpp
M packages/tpetra/core/src/Tpetra_MultiVector_def.hpp
Steps to Reproduce
Anyone with access to the machines 'white', 'ride', 'shiller' or 'hansen' should be able to reproduce these build failures using the cuda-opt
or cuda-debug
builds as described at:
More specifically, after cloning Trilinos on any of these four machines, do:
$ cd <some_build_dir>/
$ source $TRILINOS_DIR/cmake/std/atdm/load-env.sh cuda-opt
$ cmake \
-GNinja \
-DTrilinos_CONFIGURE_OPTIONS_FILE:STRING=cmake/std/atdm/ATDMDevEnv.cmake \
-DTrilinos_ENABLE_TESTS=ON -DTrilinos_ENABLE_Xpetra=ON \
$TRILINOS_DIR
$ make NP=16