TRMM and TRTRI build failures with clang/7+cuda9+Cuda_OpenMP and gcc/5.3+OpenMP

[ndellingwood]

Merge of PR #622 is causing the following build failures:

gcc/5.3 OpenMP Build:

In file included from /home/ndellin/kokkos-kernels/src/blas/impl/KokkosBlas_trtri_spec.hpp:77:0,
                 from /home/ndellin/kokkos-kernels/src/blas/KokkosBlas_trtri.hpp:50,
                 from /home/ndellin/kokkos-kernels/unit_test/blas/Test_Blas_trtri.hpp:4,
                 from /home/ndellin/kokkos-kernels/unit_test/openmp/Test_OpenMP_Blas_trtri.cpp:3:
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:65:6: error: wrong number of template arguments (3, should be 2)
      >  { enum : bool { value = true }; };
      ^
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:69:1: note: in expansion of macro ‘KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL’
 KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL(SCALAR, LAYOUTA, MEMSPACE)
 ^
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:83:2: note: in expansion of macro ‘KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL_BLAS’
  KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL_BLAS( double,                  Kokkos::LayoutLeft, Kokkos::HostSpace)
  ^
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:53:8: note: provided for ‘template<class RVT, class AVT> struct KokkosBlas::Impl::trtri_tpl_spec_avail’
 struct trtri_tpl_spec_avail {
        ^

Reproducer sample:

  # Reproducer instructions:
  #   Load modules:
        module load sems-env sems-cmake/3.12.2 kokkos-env kokkos-hwloc/1.10.1/base sems-gcc/5.3.0

  #   Use generate_makefile line below to call cmake which generates makefile for this build:
        /home/ndellin/kokkos-kernels/cm_generate_makefile.bash --with-devices=OpenMP --arch=Kepler35 --compiler=/projects/sems/install/rhel6-x86_64/sems/compiler/gcc/5.3.0/base/bin/g++ --cxxflags="-O3 -Wall -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wignored-qualifiers -Wempty-body -Wclobbered -Wuninitialized " --cxxstandard="11" --ldflags=""  --kokkos-path=/home/ndellin/kokkos-kernels/../kokkos --kokkoskernels-path=/home/ndellin/kokkos-kernels --with-scalars='' --with-ordinals= --with-offsets= --with-layouts= --with-tpls=blas    --no-examples --with-options=enable_large_mem_tests

---

clang/7+cuda/9.2 Cuda_OpenMP Build

In file included from /home/ndellin/kokkos-kernels/unit_test/openmp/Test_OpenMP_Blas_trtri.cpp:3:
In file included from /home/ndellin/kokkos-kernels/unit_test/blas/Test_Blas_trtri.hpp:4:
In file included from /home/ndellin/kokkos-kernels/src/blas/KokkosBlas_trtri.hpp:50:
In file included from /home/ndellin/kokkos-kernels/src/blas/impl/KokkosBlas_trtri_spec.hpp:77:
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:83:2: error: expected expression
 KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL_BLAS( double,                  Kokkos::LayoutLeft, Kokkos::HostSpace)
 ^
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:68:76: note: expanded from macro 'KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL_BLAS'
#define KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL_BLAS( SCALAR , LAYOUTA, MEMSPACE ) \
                                                                           ^
/home/ndellin/kokkos-kernels/src/impl/tpls/KokkosBlas_trtri_tpl_spec_avail.hpp:65:6: note: expanded from macro '\
KOKKOSBLAS_TRTRI_TPL_SPEC_AVAIL'
     >  { enum : bool { value = true }; };
     ^

Reproducer sample:

  # Reproducer instructions:
  #   Load modules:
        module load sems-env sems-cmake/3.12.2 kokkos-env kokkos-hwloc/1.10.1/base sems-clang/7.0.1 sems-cuda/9.2

  #   Use generate_makefile line below to call cmake which generates makefile for this build:
        /home/ndellin/kokkos-kernels/cm_generate_makefile.bash --with-devices=Cuda,OpenMP --arch=Kepler35 --compiler=/projects/sems/install/rhel6-x86_64/sems/compiler/clang/7.0.1/base/bin/clang++ --cxxflags="-O3 -Wall -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized " --cxxstandard="11" --ldflags="" --with-cuda=/projects/sems/install/rhel6-x86_64/sems/compiler/cuda/9.2/base --kokkos-path=/home/ndellin/kokkos-kernels/../kokkos --kokkoskernels-path=/home/ndellin/kokkos-kernels --with-scalars='' --with-ordinals= --with-offsets= --with-layouts= --with-tpls=blas    --no-examples --with-options=enable_large_mem_tests

[ndellingwood]

@e10harvey it looks like OpenMP is the common culprit to address. If you don't have time to address this soon let me know so we can revert the PR and it can be resubmitted with the fixes necessary for this issue.