Reduce with Teams broken for custom initialize

[khpierson]

Below is a patch that fixes the issue.

--- a/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_Parallel.hpp
+++ b/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_Parallel.hpp
@@ -1040,8 +1040,10 @@ public:

     pointer_type const result = (pointer_type) (m_unified_space ? m_unified_space : m_scratch_space) ;

+    value_type init;
+    ValueInit::init( ReducerConditional::select(m_functor , m_reducer) , &init);
     if(Impl::cuda_inter_block_reduction<FunctorType,ValueJoin,WorkTag>
-           (value,ValueJoin(ReducerConditional::select(m_functor , m_reducer)),m_scratch_space,result,m_scratch_flags,blockDim.y)) {
+           (value,init,ValueJoin(ReducerConditional::select(m_functor , m_reducer)),m_scratch_space,result,m_scratch_flags,blockDim.y)) {
       const unsigned id = threadIdx.y*blockDim.x + threadIdx.x;
       if(id==0) {
         Kokkos::Impl::FunctorFinal< ReducerTypeFwd , WorkTag >::final( ReducerConditional::select(m_functor , m_reducer) , (void*) &value );
diff --git a/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_ReduceScan.hpp b/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_ReduceScan.hpp
index cff183f..c9539a3 100644
--- a/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_ReduceScan.hpp
+++ b/TPLs_src/Trilinos/packages/kokkos/core/src/Cuda/Kokkos_Cuda_ReduceScan.hpp
@@ -133,6 +133,7 @@ inline void cuda_intra_block_reduction( ValueType& value,
 template< class FunctorType , class JoinOp , class ArgTag = void >
 __device__
 bool cuda_inter_block_reduction( typename FunctorValueTraits< FunctorType , ArgTag >::reference_type  value,
+                                 typename FunctorValueTraits< FunctorType , ArgTag >::reference_type  neutral,
                                  const JoinOp& join,
                                  Cuda::size_type * const m_scratch_space,
                                  typename FunctorValueTraits< FunctorType , ArgTag >::pointer_type const result,
@@ -170,7 +171,7 @@ bool cuda_inter_block_reduction( typename FunctorValueTraits< FunctorType , ArgT
       if(id == 0)
         *m_scratch_flags = 0;
       last_block = true;
-      value = value_type();
+      value = neutral;

       pointer_type const volatile global = (pointer_type) m_scratch_space ;