CPU profiling implemented

include/libgmshdg/gmsh_io.h

@@ -383,6 +383,7 @@ public:
     size_t  num_cells() const;
     size_t  num_faces() const;
     size_t  num_entities() const;
+    size_t  num_orientations() const;
 
     std::vector<entity>::const_iterator begin() const;
     std::vector<entity>::const_iterator end() const;

src/libgmshdg/gmsh_io.cpp

@@ -1087,6 +1087,16 @@ model::num_entities(void) const
     return entities.size();
 }
 
+size_t
+model::num_orientations(void) const
+{
+    size_t ret = 0;
+    for (const auto& e : entities)
+        ret += e.num_cell_orientations();
+
+    return ret;
+}
+
 
 
 model::entofs_pair

src/maxwell/maxwell_solver.cpp

@@ -69,74 +69,116 @@ void initialize_solver(const model& mod, State& state, const maxwell::parameter_
 template<typename State>
 void initialize_profiling(const model& mod, State& state, time_integrator_type ti) 
 {
+    auto num_entities = mod.num_entities();
+    auto num_cells = mod.num_cells_world();
+    auto num_all_faces = mod.num_faces();
     auto cell_dofs = num_dofs_3D(mod.approximation_order);
+    auto cell_fluxes = num_dofs_2D(mod.approximation_order);
     auto num_all_dofs = mod.num_dofs();
     auto num_all_fluxes = mod.num_fluxes();
-    auto num_cells = mod.num_cells_world();
-    auto num_entities = mod.num_entities();
+    auto num_all_orientations = mod.num_orientations();
+
+    std::cout << "num_entities: " << num_entities << std::endl
+        << "num_cells: " << num_cells << std::endl
+        << "num_all_faces: " << num_all_faces << std::endl
+        << "cell_dofs: " << cell_dofs << std::endl
+        << "cell_fluxes: " << cell_fluxes << std::endl
+        << "num_all_dofs: " << num_all_dofs << std::endl
+        << "num_all_fluxes: " << num_all_fluxes << std::endl;
 
     state.logger.register_profiler("solver", TYPE)
-        .set_quantity("DoFs", mod.num_dofs()*6);
+        .set_quantity("DoFs", num_all_dofs*6);
 
+    auto jacobians_read = 3*3*num_cells;
+    auto dm_read = 3*cell_dofs*cell_dofs;
+    auto curl_readwrite = 6*num_all_dofs;
+    auto sources_application = 3*num_all_dofs;
     if (ti == time_integrator_type::LEAPFROG)
     {
 #ifdef TM_NEW_CURLS
         state.logger.register_profiler("compute_curls_H", TYPE)
             .set_quantity("GFlops", ((45*cell_dofs+9)*num_all_dofs)/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", (jacobians_read+dm_read+curl_readwrite+sources_application)*8/1e9);
         state.logger.register_profiler("compute_curls_E", TYPE)
             .set_quantity("GFlops", ((45*cell_dofs+6)*num_all_dofs)/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", (jacobians_read+dm_read+curl_readwrite)*8/1e9);
 #else
+        auto derivatives = jacobians_read+dm_read+4*num_all_dofs;
+        auto curls_h = 4*num_all_dofs;
+        auto curls_e = 3*num_all_dofs;
         state.logger.register_profiler("compute_curls_H", TYPE)
             .set_quantity("GFlops", ((21*cell_dofs*num_all_dofs)*3 + 6*num_all_dofs)/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", (derivatives*3+curls_h)*8/1e9);
         state.logger.register_profiler("compute_curls_E", TYPE)
             .set_quantity("GFlops", ((21*cell_dofs*num_all_dofs)*3 + 3*num_all_dofs)/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB",  (derivatives*3+curls_e)*8/1e9);
 #endif /* TM_NEW_CURLS */
 
         state.logger.register_profiler("compute_field_jumps", TYPE)
             .set_quantity("GFlops", 3*num_all_fluxes/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", 9*num_all_fluxes*8/1e9);
  
+        auto normals_read = num_all_faces;
+        auto dets_read = num_all_faces;
+        auto sources_imposition = 18*num_all_fluxes;
+        auto matparams_readwrite = 4*num_all_fluxes;
+        auto fluxes_write = 3*num_all_fluxes;
         state.logger.register_profiler("compute_fluxes_planar", TYPE)
             .set_quantity("GFlops", 37*num_all_fluxes/1e9)
-            .set_quantity("GB", 0); 
+            .set_quantity("GB", (normals_read+dets_read+sources_imposition+matparams_readwrite+fluxes_write)*8/1e9); 
         
+        dets_read = num_cells;
+        auto lm_read = num_all_orientations*cell_dofs*4*cell_fluxes;
+        auto lifting_readwrite = num_all_fluxes+num_all_dofs;
         state.logger.register_profiler("compute_flux_lifting", TYPE)
             .set_quantity("GFlops", 21*cell_dofs*num_all_dofs*3/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", 3*(dets_read+lm_read+lifting_readwrite)*8/1e9);
     }
     else
     {
 #ifdef TM_NEW_CURLS
+
         state.logger.register_profiler("compute_curls", TYPE)
             .set_quantity("GFlops", ((45*cell_dofs+6)*num_all_dofs + (45*cell_dofs+9)*num_all_dofs)/1e9)
-            .set_quantity("GB", 2*(6*num_all_dofs + num_entities*(3*cell_dofs*cell_dofs+3*3*num_cells))*8/1e9);
+            .set_quantity("GB", (2*(jacobians_read+dm_read+curl_readwrite)+sources_application)*8/1e9);
 #else
+        auto derivatives = jacobians_read+dm_read+4*num_all_dofs;
+        auto curls = 4*num_all_dofs + 3*num_all_dofs;
         state.logger.register_profiler("compute_curls", TYPE)
             .set_quantity("GFlops", ((21*cell_dofs*num_all_dofs)*6 + 9*num_all_dofs)/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", (derivatives*6+curls)*8/1e9);
 #endif /* TM_NEW_CURLS */
 
 #ifdef USE_MPI
         state.logger.register_profiler("compute_field_jumps", TYPE)
             .set_quantity("GFlops", 6*num_all_fluxes/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", 18*num_all_fluxes*8/1e9);
  
+        auto normals_read = num_all_faces;
+        auto dets_read = num_all_faces;
+        auto sources_imposition = 18*num_all_fluxes;
+        auto matparams_readwrite = 8*num_all_fluxes;
+        auto fluxes_write = 6*num_all_fluxes;
         state.logger.register_profiler("compute_fluxes_planar", TYPE)
             .set_quantity("GFlops", 68*num_all_fluxes/1e9)
-            .set_quantity("GB", 0); 
+            .set_quantity("GB", (normals_read+dets_read+sources_imposition+matparams_readwrite+fluxes_write)*8/1e9); 
 #else
+        auto normals_read = num_all_faces;
+        auto dets_read = num_all_faces;
+        auto jumps_and_sources = 24*num_all_fluxes;
+        auto matparams_readwrite = 8*num_all_fluxes;
+        auto fluxes_write = 6*num_all_fluxes;
         state.logger.register_profiler("compute_field_jumps_and_fluxes", TYPE)
             .set_quantity("GFlops", 72*num_all_fluxes/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", (normals_read+dets_read+jumps_and_sources+matparams_readwrite+fluxes_write)*8/1e9); 
 #endif /* USE_MPI */
         
+        dets_read = num_cells;
+        auto lm_read = num_all_orientations*cell_dofs*4*cell_fluxes;
+        auto lifting_readwrite = num_all_fluxes+num_all_dofs;
         state.logger.register_profiler("compute_flux_lifting", TYPE)
             .set_quantity("GFlops", 21*cell_dofs*num_all_dofs*6/1e9)
-            .set_quantity("GB", 0);
+            .set_quantity("GB", 6*(dets_read+lm_read+lifting_readwrite)*8/1e9);
     }
 }
 #endif /* ENABLE_PROFILING */