Sparse: fix some issues raised in PR review

sparse/impl/KokkosSparse_CrsMatrix_traversal_impl.hpp

@@ -121,14 +121,6 @@ int64_t crsmatrix_traversal_launch_parameters(int64_t numRows, int64_t nnz,
 
   rows_per_team = rows_per_thread * team_size;
 
-  if (rows_per_team < 0) {
-    int64_t nnz_per_team = 4096;
-    int64_t conc         = execution_space().concurrency();
-    while ((conc * nnz_per_team * 4 > nnz) && (nnz_per_team > 256))
-      nnz_per_team /= 2;
-    rows_per_team = (nnz_per_team + nnz_per_row - 1) / nnz_per_row;
-  }
-
   return rows_per_team;
 }
 

sparse/src/KokkosSparse_CrsMatrix_traversal.hpp

@@ -17,7 +17,9 @@
 /// \file KokkosSparse_CrsMatrix_traversal.hpp
 /// \brief Traversal method to access all entries in a CrsMatrix
 ///
-/// blah
+/// This file provides a public interface to traversal
+/// methods that are used as a common and efficient way
+/// to access entries in a matrix on host and/or device.
 
 #ifndef KOKKOSSPARSE_CRSMATRIX_TRAVERSAL_HPP
 #define KOKKOSSPARSE_CRSMATRIX_TRAVERSAL_HPP
@@ -32,9 +34,30 @@
 namespace KokkosSparse {
 namespace Experimental {
 
+
+/// \brief Public interface to sparse matrix traversal algorithm.
+///
+/// Loop over the entries of the input matrix and apply the functor
+/// to them. The functor itself may contain its own data to save results
+/// after the traversal completes.
+///
+/// \tparam execution_space
+/// \tparam crsmatrix_type
+/// \tparam functor_type
+///
+/// \param space [in] execution space instance that specifies where the kernel
+///   will be executed.
+/// \param matrix [in] the matrix to be traversed.
+/// \param functor [in] a functor that is being called on each local entries
+/// of the crsmatrix and that implement a user defined capabilities.
+///
 template <class execution_space, class crsmatrix_type, class functor_type>
 void crsmatrix_traversal(const execution_space& space,
                          const crsmatrix_type& matrix, functor_type& functor) {
+
+  // Check if a quick return can be performed
+  if(!matrix.nnz()) { return; }
+
   // 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);
@@ -43,6 +66,19 @@ void crsmatrix_traversal(const execution_space& space,
   }
 }
 
+/// \brief Public interface to sparse matrix traversal algorithm.
+///
+/// Loop over the entries of the input matrix and apply the functor
+/// to them. The functor itself may contain its own data to save results
+/// after the traversal completes.
+///
+/// \tparam crsmatrix_type
+/// \tparam functor_type
+///
+/// \param matrix [in] the matrix to be traversed.
+/// \param functor [in] a functor that is being called on each local entries
+/// of the crsmatrix and that implement a user defined capabilities.
+///
 template <class crsmatrix_type, class functor_type>
 void crsmatrix_traversal(const crsmatrix_type& matrix, functor_type& functor) {
   using execution_space = typename crsmatrix_type::execution_space;