Add explanation on the use of __synchtreads()

diff --git a/include/deal.II/matrix_free/cuda_hanging_nodes_internal.h b/include/deal.II/matrix_free/cuda_hanging_nodes_internal.h
index 9a9b828749e4ba366e0bfcfd3368a357201c9af3..8abf9b68ec6268747918234421c9ba32d2e86330 100644 (file)
--- a/include/deal.II/matrix_free/cuda_hanging_nodes_internal.h
+++ b/include/deal.II/matrix_free/cuda_hanging_nodes_internal.h
@@ -830,6 +830,8 @@ namespace CUDAWrappers
               }
           }
 
+        // The synchronization is done for all the threads in one block with
+        // each block being assigned to one element.
         __syncthreads();
 
         if (constrained_face && constrained_dof)