CTM: more robust adapt() timers

adding barriers before each timer ensures we get
the maximum duration over all ranks, and it also
ensures that imbalance from previous operations
don't affect timings for later operations.
also separate the region of interest into three
different timers.

src/CTM/CTM_Adapter.cpp

@@ -548,6 +548,7 @@ void Adapter::adapt(const double t_current) {
   printf("adapt(): before adapt - coming in: %d tets on cpu %d\n",
       M_numRegions(PM_mesh(sim_mesh, 0)), PMU_rank());
 
+  PCU_Barrier();
   double t0 = PCU_Time();
 
   // create the simmetrix adapter
@@ -573,19 +574,20 @@ void Adapter::adapt(const double t_current) {
   write_debug(debug, apf_mesh, "preadapt_", call_count);
   apf::destroyField(spr_size_field);
 
+  PCU_Barrier();
   double t1 = PCU_Time();
-  *out << "adapt(): preparing mesh adapt in " << t1-t0 << " seconds\n";
+  *out << "adapt(): prepared for adapt in " << t1-t0 << " seconds\n";
 
   // run the adapter
-  double t2 = PCU_Time();
   pProgress progress = Progress_new();
   MSA_adapt(adapter, progress);
   Progress_delete(progress);
   MSA_delete(adapter);
   MS_deleteMeshCase(mcase);
   write_debug(debug, apf_mesh, "postadapt_", call_count);
+  PCU_Barrier();
   double t3 = PCU_Time();
-  *out << "adapt(): mesh adapt in " << t3-t2 << " seconds\n";
+  *out << "adapt(): mesh adapted in " << t3-t1 << " seconds\n";
 
   // add the layer
   *out << "adapt(): adding layer: " << current_layer+1 << std::endl;
@@ -597,29 +599,38 @@ void Adapter::adapt(const double t_current) {
   // clean up
   PList_delete(sim_field_list);
 
-  // partition the mesh
+  PCU_Barrier();
   double t4 = PCU_Time();
+  *out << "adapt(): layer added in " << t4-t3 << " seconds\n";
+  // partition the mesh
   Parma_PrintPtnStats(apf_mesh, "pre load balance:");
   PM_partition(sim_mesh, 0, 0);
   Parma_PrintPtnStats(apf_mesh, "post load balance:");
+  PCU_Barrier();
   double t5 = PCU_Time();
-  *out << "adapt(): load balancing in " << t5-t4 << " seconds\n";
+  *out << "adapt(): load balanced in " << t5-t4 << " seconds\n";
 
   // rebuild the data structures needed for analysis
-  double t6 = PCU_Time();
   apf_mesh->verify();
+  PCU_Barrier();
+  double t7 = PCU_Time();
+  *out << "adapt(): mesh verified in " << t7-t5 << " seconds\n";
   auto t_sim_disc = rcp_dynamic_cast<Albany::SimDiscretization>(t_disc);
-  auto m_sim_disc = rcp_dynamic_cast<Albany::SimDiscretization>(m_disc);
   t_sim_disc->updateMesh(/* transfer ip = */ false, param_lib);
+  PCU_Barrier();
+  double t8 = PCU_Time();
+  *out << "adapt(): updated temperature discretization in " << t8-t7 << " seconds\n";
+  auto m_sim_disc = rcp_dynamic_cast<Albany::SimDiscretization>(m_disc);
   m_sim_disc->updateMesh(/* transfer ip = */ false, param_lib);
-  double t7 = PCU_Time();
-  *out << "adapt(): update albany structures in " << t7-t6 << " seconds\n";
+  PCU_Barrier();
+  double t9 = PCU_Time();
+  *out << "adapt(): updated mechanics discretization in " << t9-t8 << " seconds\n";
 
   // print stats after adaptation
   apf::printStats(apf_mesh);
 
   // print a final statement
-  *out << "adapt(): going out in " << PCU_Time()
+  *out << "adapt(): exiting at " << PCU_Time()
        << ", cpu = " << PMU_rank() << std::endl;
   printf("adapt(): leaving %d tets on cpu %d\n",
       M_numRegions(PM_mesh(sim_mesh,0)), PMU_rank());