trmm perf test nightly build error: clang/8+cuda tpl build

[ndellingwood]

The following nightly test failed on kokkos-dev-2 after merge of PR #720, @e10harvey can you take a look? From the compiler message looks like either a routine is missing a KOKKOS_INLINE_FUNCTION marking, or a host function is being incorrectly called from a kernel in a Kokkos::parallel_for

clang-8.0-Cuda_OpenMP-release (build failed)
https://jenkins-son.sandia.gov/job/KokkosKernels_KokkosDev2_SPOTCHECK_TPLS/83/console

Other impacted builds:
https://jenkins-son.sandia.gov/view/KokkosKernels/job/KokkosKernels_KokkosDev2_CLANG8_CUDA10_cpp17/
https://jenkins-son.sandia.gov/view/KokkosKernels/job/KokkosKernels_KokkosDev2_CLANG8_CUDA10_cpp17/
https://jenkins-son.sandia.gov/view/KokkosKernels/job/KokkosKernels_KokkosDev_CLANG7_CUDA9/84/console

First compile error

In file included from /home/jenkins/slave/workspace/KokkosKernels_KokkosDev2_SPOTCHECK_TPLS/kokkos-kernels/perf_test/blas/blas3/KokkosBlas3_perf_test.cpp:45:
/home/jenkins/slave/workspace/KokkosKernels_KokkosDev2_SPOTCHECK_TPLS/kokkos-kernels/perf_test/blas/blas3/KokkosBlas3_trmm_perf_test.hpp:303:17: error: reference to __host__ function 'trmm<Kokkos::View<double **, Kokkos::LayoutStride, Kokkos::Device<Kokkos::Cuda, Kokkos::CudaSpace>, Kokkos::MemoryTraits<0> >, Kokkos::View<double **, Kokkos::LayoutStride, Kokkos::Device<Kokkos::Cuda, Kokkos::CudaSpace>, Kokkos::MemoryTraits<0> > >' in __host__ __device__ function
 KokkosBlas::trmm(&trmm_args_.side, &trmm_args_.uplo, &trmm_args_.trans,

Reproducer instructions:

   #   Load modules:
         source /projects/sems/modulefiles/utils/sems-modules-init.sh ; module use /home/projects/x86-64/modulefiles/local
         module purge
         module load sems-env sems-cmake/3.12.2 kokkos-env clang/8.0 cuda/10.0
 
       $KOKKOSKERNELS_PATH/cm_generate_makefile.bash --with-devices=Cuda,OpenMP --arch=SNB,Volta70 --compiler=/home/projects/x86-64/clang/8.0/bin/clang++ --cxxflags="-O3 -Wall -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized -Wno-pass-failed " --cxxstandard="11" --ldflags="" --with-cuda=/home/projects/x86-64/cuda/10.0 --kokkos-path=$KOKKOS_PATH --kokkoskernels-path=$KOKKOSKERNELS_PATH --with-scalars='double,complex_double' --with-ordinals=int --with-offsets=int,size_t --with-layouts=LayoutLeft --with-tpls=blas    --with-options= --with-cuda-options= --no-examples 

Edit: Added links to other affected builds

[e10harvey]

@ndellingwood: Thank you! Looking into this now.