Improve further.

git-svn-id: https://svn.dealii.org/trunk@31774 0785d39b-7218-0410-832d-ea1e28bc413d

diff --git a/deal.II/source/numerics/data_out.cc b/deal.II/source/numerics/data_out.cc
index 8ac46cd8e57068ef6a9d8b706504cf0b92deba5b..d242eaa816d6fabc34da0dd66954936a5dab17fd 100644 (file)
--- a/deal.II/source/numerics/data_out.cc
+++ b/deal.II/source/numerics/data_out.cc
@@ -455,11 +455,13 @@ void DataOut<dim,DH>::build_patches (const Mapping<DH::dimension,DH::space_dimen
                     std_cxx1x::function<void (const ::dealii::DataOutBase::Patch<DH::dimension, DH::space_dimension> &)>(),
                      thread_data,
                      sample_patch,
-                    2*multithread_info.n_threads(),
                     // experimenting shows that we can make things run a bit
-                    // faster if we increase the number of items in flight at
-                    // any given time (about 10% on the testcase discussed in
-                    // @ref workstream_paper, on 32 cores)
+                    // faster if we increase the number of cells we work on
+                    // per item (i.e., WorkStream's chunk_size argument,
+                    // about 10% improvement) and the items in flight at any
+                    // given time (another 5% on the testcase discussed in
+                    // @ref workstream_paper, on 32 cores) and if
+                    8*multithread_info.n_threads(),
                     64);
 }