Sparse: fix CrsMatrix traversal on GPU

Obvious typos and mistakes left-over from CPU development.

Author: lucbv

sparse/impl/KokkosSparse_CrsMatrix_traversal_impl.hpp

@@ -149,7 +149,7 @@ void crsmatrix_traversal_on_gpu(const execution_space& space,
   // Wrap user functor with crsmatrix_traversal_functor
   int64_t rows_per_thread = 0;
   int team_size = 0, vector_length = 0;
-  const int64_t rows_per_team = crsmatrix_traversal_launch_parameters(A.numRows(), A.nnz(), rows_per_thread, team_size, vector_length);
+  const int64_t rows_per_team = crsmatrix_traversal_launch_parameters<execution_space>(A.numRows(), A.nnz(), rows_per_thread, team_size, vector_length);
   const int nteams = (static_cast<int>(A.numRows()) + rows_per_team - 1) / rows_per_team;
   crsmatrix_traversal_functor<execution_space, crsmatrix_type, functor_type> traversal_func(A, func, rows_per_team);
 

sparse/src/KokkosSparse_CrsMatrix_traversal.hpp

@@ -37,7 +37,7 @@ void crsmatrix_traversal(const execution_space& space, const crsmatrix_type& mat
 
   // Choose between device and host implementation
   if constexpr (KokkosKernels::Impl::kk_is_gpu_exec_space<execution_space>()) {
-
+    KokkosSparse::Impl::crsmatrix_traversal_on_gpu(space, matrix, functor);
   } else {
     KokkosSparse::Impl::crsmatrix_traversal_on_host(space, matrix, functor);
   }