Build error on Ride + CUDA Kokkos node due to issues in TpetraExt_MatrixMatrix_ExtraKernels_def.hpp

[ikalash]

The Albany CUDA build failed last night due to issues in TpetraExt_MatrixMatrix_ExtraKernels_def.hpp :

http://cdash.sandia.gov/CDash-2-3-0/viewBuildError.php?buildid=66026

/.../repos/Trilinos/packages/tpetra/core/ext/TpetraExt_MatrixMatrix_ExtraKernels_def.hpp(979): error: no instance of overloaded function "Tpetra::MatrixMatrix::add" matches the argument list

Can someone please look into and resolve the problem?

@trilinos/tpetra

[mhoemmen]

@csiefer2

[mhoemmen]

See also #2183 (comment) .
@lucbv

[csiefer2]

Weird. It built on white and artemis.

[ikalash]

It looks like there is now a compilation error having to do with TpetraExt_MatrixMatrix_ExtraKernels_def.hpp in the Albany OpenMP build on my workstation:

/home/ikalash/nightlyAlbanyTests/Results/Trilinos/packages/tpetra/core/ext/TpetraExt_MatrixMatrix_ExtraKernels_def.hpp:979:28: error: no matching function for call to ‘add(Sacado::MP::Vector<Stokhos::StaticFixedStorage<int, double, 4, Kokkos::OpenMP> >&, bool, const Tpetra::CrsMatrix<Sacado::MP::Vector<Stokhos::StaticFixedStorage<int, double, 4, Kokkos::OpenMP> >, int, int, Kokkos::Compat::KokkosDeviceWrapperNode<Kokkos::OpenMP> >&, Sacado::MP::UnaryMinusOp<Sacado::MP::Vector<Stokhos::StaticFixedStorage<int, double, 4, Kokkos::OpenMP> > >, bool, Tpetra::CrsMatrix<Sacado::MP::Vector<Stokhos::StaticFixedStorage<int, double, 4, Kokkos::OpenMP> >, int, int, Kokkos::Compat::KokkosDeviceWrapperNode<Kokkos::OpenMP> >&, Tpetra::CrsMatrix<Sacado::MP::Vector<Stokhos::StaticFixedStorage<int, double, 4, Kokkos::OpenMP> >, int, int, Kokkos::Compat::KokkosDeviceWrapperNode<Kokkos::OpenMP> >&, Teuchos::RCP<const Tpetra::Map<int, int, Kokkos::Compat::KokkosDeviceWrapperNode<Kokkos::OpenMP> > >, Teuchos::RCP<const Tpetra::Map<int, int, Kokkos::Compat::KokkosDeviceWrapperNode<Kokkos::OpenMP> > >, const Teuchos::RCP<Teuchos::ParameterList>&)’
   Tpetra::MatrixMatrix::add(one,false,*Bview.origMatrix,-omega,false,*AB,C,AB->getDomainMap(),AB->getRangeMap(),params);

[mhoemmen]

@ikalash @csiefer2 @etphipp Looks like we need to add a new instantiation for that Sacado type to Stokhos (which is where all the instantiations go). Let me see if I can figure it out now.

[mhoemmen]

Hm, TpetraExt_MatrixMatrix_ExtraKernels_def.hpp line 979 refers to one of the Tpetra::MatrixMatrix::add overloads that the TPETRA_MATRIXMATRIX_INSTANT macro includes:

Trilinos/packages/tpetra/core/ext/TpetraExt_MatrixMatrix_def.hpp

Line 3809 in a138b30

MatrixMatrix::add<SCALAR, LO, GO, NODE> \

I don't see that Trilinos uses that macro anywhere else:

$ git grep "TPETRA_MATRIXMATRIX_INSTANT" *
packages/tpetra/core/ext/TpetraExt_MatrixMatrix.cpp:  TPETRA_INSTANTIATE_SLGN(TPETRA_MATRIXMATRIX_INSTANT)
...
packages/tpetra/core/ext/TpetraExt_MatrixMatrix_def.hpp:#define TPETRA_MATRIXMATRIX_INSTANT(SCALAR,LO,GO,NODE) \

@ikalash , have you just started running Sacado types through MueLu? Is that a new thing for Albany?

[ikalash]

@mhoemmen : if we have, I am not aware of it. @ibaned may know more than me.

[csiefer2]

FYI - the magic stokhos rebase scripts don't seem to do anything for this (though it looks like someone missed the rebase stuff on BlackBox). Will bug @etphipp

[etphipp]

Because of the way Tpetra::MatrixMatrix::add() is declared, you need to cast -omega to the right type, i.e., replace

Tpetra::MatrixMatrix::add(one,false,*Bview.origMatrix,-omega,false,*AB,C,AB->getDomainMap(),AB->getRangeMap(),params);

with

Tpetra::MatrixMatrix::add(one,false,*Bview.origMatrix,Scalar(-omega),false,*AB,C,AB->getDomainMap(),AB->getRangeMap(),params);

[csiefer2]

Checkin script baking on both this & blackbox fix.

[ibaned]

@csiefer2 you can put up pull requests and use the automatic pull request testing system if thats more convenient for you

[csiefer2]

That may be the first time I've heard "convenient" and "pull request" in the same sentence 😛

[csiefer2]

Pushed. @ikalash Feel free to close if Albany builds tonight

[ikalash]

Thanks, @csiefer2 ! I'll know by tomorrow morning if your fix solves the issues for Albany from the nightlies.

[mhoemmen]

Thanks @csiefer2 ! :D

[ikalash]

This issue is resolved - thanks for fixing it @csiefer2 ! There is still a MueLu + CUDA issue, though, namely #2161 .