mem tracing compile time option, simplify by MemoryTracer as member

CMakeLists.txt

@@ -33,6 +33,7 @@ option( BUILD_STUB_FILES "Build stub files for better autocompletion" ON)
 option( BUILD_FOR_CONDA "Link python libraries only to executables" OFF)
 
 option( USE_SUPERBUILD       "use ccache" ON)
+option( TRACE_MEMORY     "Enable memory tracing" OFF)
 
 set(NG_COMPILE_FLAGS "" CACHE STRING "Additional compile flags")
 

cmake/SuperBuild.cmake

@@ -143,6 +143,7 @@ set_vars( NETGEN_CMAKE_ARGS
   USE_SPDLOG
   DEBUG_LOG
   CHECK_RANGE
+  TRACE_MEMORY
   BUILD_STUB_FILES
   BUILD_FOR_CONDA
   NG_COMPILE_FLAGS

libsrc/core/CMakeLists.txt

@@ -42,6 +42,11 @@ if(CHECK_RANGE OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL
   target_compile_definitions(ngcore PUBLIC NETGEN_ENABLE_CHECK_RANGE)
 endif(CHECK_RANGE OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
 
+if(TRACE_MEMORY OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
+  target_compile_definitions(ngcore PUBLIC NETGEN_TRACE_MEMORY)
+endif(TRACE_MEMORY OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
+
+
 if(USE_SPDLOG)
   include_directories(${SPDLOG_INCLUDE_DIR})
   install(DIRECTORY ${SPDLOG_INCLUDE_DIR}

libsrc/core/array.hpp

@@ -655,7 +655,6 @@ namespace ngcore
     /// that's the data we have to delete, nullptr for not owning the memory
     T * mem_to_delete;
 
-    int mem_tracing_id = 0;
 
     using FlatArray<T,IndexType>::size;
     using FlatArray<T,IndexType>::data;
@@ -701,7 +700,7 @@ namespace ngcore
 
     NETGEN_INLINE Array (Array && a2) 
     {
-      TraceMemorySwap(mem_tracing_id, sizeof(T)*allocsize, a2.mem_tracing_id, sizeof(T)*a2.allocsize);
+      mt.Swap(sizeof(T) * allocsize, a2.mt, sizeof(T) * a2.allocsize);
 
       size = a2.size; 
       data = a2.data;
@@ -774,7 +773,7 @@ namespace ngcore
     NETGEN_INLINE ~Array()
     {
       delete [] mem_to_delete;
-      TraceMemoryFree(mem_tracing_id, sizeof(T)*allocsize);
+      mt.Free(sizeof(T)*allocsize);
     }
 
     // Only provide this function if T is archivable
@@ -829,7 +828,7 @@ namespace ngcore
     NETGEN_INLINE const Array & Assign (size_t asize, LocalHeap & lh)
     {
       delete [] mem_to_delete;
-      TraceMemoryFree(mem_tracing_id, sizeof(T)*allocsize);
+      mt.Free(sizeof(T)*allocsize);
       size = allocsize = asize;
       data = lh.Alloc<T> (asize);
       mem_to_delete = nullptr;
@@ -937,7 +936,7 @@ namespace ngcore
     NETGEN_INLINE void DeleteAll ()
     {
       delete [] mem_to_delete;
-      TraceMemoryFree(mem_tracing_id, sizeof(T)*allocsize);
+      mt.Free(sizeof(T)*allocsize);
       mem_to_delete = NULL;
       data = 0;
       size = allocsize = 0;
@@ -969,7 +968,7 @@ namespace ngcore
     /// steal array 
     NETGEN_INLINE Array & operator= (Array && a2)
     {
-      TraceMemorySwap(mem_tracing_id, sizeof(T)*allocsize, a2.mem_tracing_id, sizeof(T)*a2.allocsize);
+      mt.Swap(sizeof(T)*allocsize, a2.mt, sizeof(T)*a2.allocsize);
 
       ngcore::Swap (size, a2.size);
       ngcore::Swap (data, a2.data);
@@ -1044,7 +1043,7 @@ namespace ngcore
     
     NETGEN_INLINE void Swap (Array & b)
     {
-      TraceMemorySwap(mem_tracing_id, sizeof(T)*allocsize, b.mem_tracing_id, sizeof(T)*b.allocsize);
+      mt.Swap(sizeof(T) * allocsize, b.mt, sizeof(T) * b.allocsize);
 
       ngcore::Swap (size, b.size);
       ngcore::Swap (data, b.data);
@@ -1052,21 +1051,18 @@ namespace ngcore
       ngcore::Swap (mem_to_delete, b.mem_to_delete);
     }
 
-    NETGEN_INLINE void SetMemoryTracing (int mem_id)
+    NETGEN_INLINE void StartMemoryTracing () const
     {
-      if(!mem_tracing_id && mem_id)
+      mt.Alloc(sizeof(T) * allocsize);
-        TraceMemoryAlloc(mem_id, sizeof(T)*allocsize);
-
-      if(mem_tracing_id && !mem_id)
-        TraceMemoryFree(mem_tracing_id, sizeof(T)*allocsize);
-
-      mem_tracing_id = mem_id;
     }
 
+    const MemoryTracer& GetMemoryTracer() const { return mt; }
+
   private:
 
     /// resize array, at least to size minsize. copy contents
     NETGEN_INLINE void ReSize (size_t minsize);
+    MemoryTracer mt;
   };
 
   
@@ -1079,7 +1075,7 @@ namespace ngcore
     
     T * hdata = data;
     data = new T[nsize];
-    TraceMemoryAlloc(mem_tracing_id, sizeof(T)*nsize );
+    mt.Alloc(sizeof(T) * nsize);
 
     if (hdata)
       {
@@ -1093,7 +1089,7 @@ namespace ngcore
           for (size_t i = 0; i < mins; i++) data[i] = std::move(hdata[i]);
 #endif
         delete [] mem_to_delete;
-        TraceMemoryFree( mem_tracing_id, sizeof(T)*allocsize );
+        mt.Free(sizeof(T) * allocsize);
       }
 
     mem_to_delete = data;

libsrc/core/bitarray.cpp

@@ -36,10 +36,15 @@ namespace ngcore
   void BitArray :: SetSize (size_t asize)
   {
     if (size == asize) return;
-    if (owns_data) delete [] data;
+    if (owns_data)
+      {
+        delete [] data;
+        mt.Free(Addr(size)+1);
+      }
 
     size = asize;
     data = new unsigned char [Addr (size)+1];
+    mt.Alloc(Addr(size)+1);
   }
 
   BitArray & BitArray :: Set () throw()

libsrc/core/bitarray.hpp

@@ -150,6 +150,14 @@ public:
   NGCORE_API void DoArchive(Archive& archive);
   
   NGCORE_API auto * Data() const { return data; }
+
+  const MemoryTracer& GetMemoryTracer() const { return mt; }
+  void StartMemoryTracing() const
+  {
+    if(owns_data)
+      mt.Alloc(Addr(size)+1);
+  }
+
 private:
   ///
   unsigned char Mask (size_t i) const
@@ -159,6 +167,7 @@ private:
   size_t Addr (size_t i) const
   { return (i / CHAR_BIT); }
 
+  MemoryTracer mt;
 };
 
 

libsrc/core/paje_trace.cpp

@@ -743,7 +743,9 @@ namespace ngcore
             }
         }
       WriteTimingChart();
+#ifdef NETGEN_TRACE_MEMORY
       WriteMemoryChart("");
+#endif // NETGEN_TRACE_MEMORY
       paje.WriteEvents();
     }
 
@@ -930,6 +932,7 @@ namespace ngcore
 
   }
 
+#ifdef NETGEN_TRACE_MEMORY
   void PajeTrace::WriteMemoryChart( std::string fname )
   {
     if(fname=="")
@@ -1053,6 +1056,7 @@ namespace ngcore
     WriteSunburstHTML( root, fname, false );
 
   }
+#endif // NETGEN_TRACE_MEMORY
 
   void PajeTrace::WriteTimingChart( )
   {

libsrc/core/paje_trace.hpp

@@ -31,7 +31,9 @@ namespace ngcore
 
     public:
       NGCORE_API void WriteTimingChart();
+#ifdef NETGEN_TRACE_MEMORY
       NGCORE_API void WriteMemoryChart( std::string fname );
+#endif // NETGEN_TRACE_MEMORY
 
       // Approximate number of events to trace. Tracing will
       // be stopped if any thread reaches this number of events

libsrc/core/profiler.cpp

@@ -113,7 +113,9 @@ namespace ngcore
 
   NgProfiler prof; // NOLINT
 
+#ifdef NETGEN_TRACE_MEMORY
   std::vector<std::string> MemoryTracer::names{"root"};
   std::map< int, std::vector<int> > MemoryTracer::tree;
+#endif // NETGEN_TRACE_MEMORY
 
 } // namespace ngcore

libsrc/core/profiler.hpp

@@ -304,45 +304,29 @@ namespace ngcore
 
   namespace detail
   {
-    //Type trait to check if a class implements a 'const MemoryTracer& GetMemoryTracer()' function
-    template<typename T>
-    struct has_GetMemoryTracer
-    {
-    private:
-      template<typename T2>
-      static constexpr auto check(T2*) ->
-        typename std::is_same<decltype(std::declval<T2>().GetMemoryTracer()),const MemoryTracer &>::type;
-      template<typename>
-      static constexpr std::false_type check(...);
-      using type = decltype(check<T>(nullptr)); // NOLINT
-    public:
-      static constexpr bool value = type::value;
-    };
-
     //Type trait to check if a class implements a 'void SetMemoryTacing(int)' function
     template<typename T>
-    struct has_SetMemoryTracing
+    struct has_StartMemoryTracing
     {
     private:
       template<typename T2>
       static constexpr auto check(T2*) ->
-        typename std::is_same<decltype(std::declval<T2>().SetMemoryTracing(0)),void>::type;
+        typename std::is_same<decltype(std::declval<T2>().StartMemoryTracing()),void>::type;
       template<typename>
       static constexpr std::false_type check(...);
       using type = decltype(check<T>(nullptr)); // NOLINT
     public:
       static constexpr bool value = type::value;
     };
-
-
   } // namespace detail
 
   class MemoryTracer
   {
+    #ifdef NETGEN_TRACE_MEMORY
     NGCORE_API static std::vector<std::string> names;
     NGCORE_API static std::map< int, std::vector<int> > tree;
 
-    static int GetId(std::string name)
+    static int CreateId(const std::string& name)
     {
       int id = names.size();
       names.push_back(name);
@@ -350,48 +334,73 @@ namespace ngcore
         std::cerr << "Allocated " << id << " MemoryTracer objects" << std::endl;
       return id;
     }
-
-
     int id;
 
     public:
 
     MemoryTracer( std::string name )
     {
-      id = GetId(name);
+      id = CreateId(name);
     }
 
+    // not tracing
+    MemoryTracer() : id(0) {}
+
     template <typename... TRest>
     MemoryTracer( std::string name, TRest & ... rest )
     {
-      id = GetId(name);
+      id = CreateId(name);
       Track(rest...);
     }
 
+    NETGEN_INLINE void Alloc(size_t size) const
+    {
+      if(id && trace)
+        trace->AllocMemory(id, size);
+    }
+
+    void Free(size_t size) const
+    {
+      if(id && trace)
+        trace->FreeMemory(id, size);
+    }
+
+    void Swap(size_t mysize, MemoryTracer& other, size_t other_size) const
+    {
+      if(!trace || (id == 0 && other.id == 0))
+        return;
+      if(id == 0)
+        return trace->ChangeMemory(other.id, mysize - other_size);
+      if(other.id == 0)
+        return trace->ChangeMemory(id, other_size - mysize);
+
+      // first decrease memory, otherwise have artificial/wrong high peak memory usage
+      if(mysize<other_size)
+        {
+          trace->ChangeMemory(other.id, mysize-other_size);
+          trace->ChangeMemory(id, other_size-mysize);
+        }
+      else
+        {
+          trace->ChangeMemory(id, other_size-mysize);
+          trace->ChangeMemory(other.id, mysize-other_size);
+        }
+    }
+
+    int GetId() const { return id; }
+
     template <typename T1, typename... TRest>
-    void Track( T1 & obj, std::string name, TRest & ... rest ) const
+    void Track( T1 & obj, const std::string& name, TRest & ... rest ) const
     {
       Track(obj, name);
       Track(rest...);
     }
 
     template<typename T>
-    void Track( T & obj, std::string name ) const
+    void Track( T & obj, const std::string& name ) const
     {
-      if constexpr(detail::has_SetMemoryTracing<T>::value)
+      obj.GetMemoryTracer().Activate(obj, name);
-      {
+      tree[id].push_back(obj.GetMemoryTracer().GetId());
-        int child_id = GetId(name);
-        tree[id].push_back(child_id);
-        obj.SetMemoryTracing(child_id);
-      }
-      if constexpr(detail::has_GetMemoryTracer<T>::value)
-      {
-        auto & mt = obj.GetMemoryTracer();
-        int child_id = mt.id;
-        if(name!="")
-          names[mt.id] = name;
-        tree[id].push_back(child_id);
-      }
     }
 
     static std::string GetName(int id)
@@ -404,7 +413,20 @@ namespace ngcore
       return names[id];
     }
 
-    void SetName(std::string name) const
+    template<typename T>
+    void Activate(T& me, const std::string& name) const
+    {
+      if(!id)
+        {
+          const_cast<MemoryTracer*>(this)->id = CreateId(name);
+          if constexpr(detail::has_StartMemoryTracing<T>::value)
+            me.StartMemoryTracing();
+        }
+      else
+        SetName(name);
+    }
+
+    void SetName(const std::string& name) const
     {
       names[id] = name;
     }
@@ -412,49 +434,26 @@ namespace ngcore
 
     static const std::vector<std::string> & GetNames() { return names; }
     static const std::map<int, std::vector<int>> & GetTree() { return tree; }
+#else // NETGEN_TRACE_MEMORY
+  public:
+    MemoryTracer() {}
+    MemoryTracer( std::string name ) {}
+    template <typename... TRest>
+    MemoryTracer( std::string name, TRest & ... ) {}
 
+    void Alloc(size_t size) const {}
+    void Free(size_t size) const {}
+    void Swap(...) const {}
+    int GetId() const { return 0; }
+
+    template <typename... TRest>
+    void Track(TRest&...) const {}
+
+    static std::string GetName(int id) { return ""; }
+    std::string GetName() const { return ""; }
+    void SetName(std::string name) const {}
+#endif // NETGEN_TRACE_MEMORY
   };
-
-  NETGEN_INLINE void TraceMemoryAlloc( int mem_id, size_t size )
-  {
-    if(mem_id && trace)
-      trace->AllocMemory(mem_id, size);
-  }
-
-  NETGEN_INLINE void TraceMemoryFree( int mem_id, size_t size )
-  {
-    if(mem_id && trace)
-      trace->FreeMemory(mem_id, size);
-  }
-
-  NETGEN_INLINE void TraceMemoryChange( int mem_id, long long size )
-  {
-    if(mem_id && trace)
-      trace->ChangeMemory(mem_id, size);
-  }
-
-  NETGEN_INLINE void TraceMemorySwap( int mem_id, size_t size, int mem_id2, size_t size2 )
-  {
-    if(!trace || (mem_id==0 && mem_id2==0))
-      return;
-    if(mem_id == 0)
-      return trace->ChangeMemory(mem_id2, size-size2);
-    if(mem_id2 == 0)
-      return trace->ChangeMemory(mem_id, size2-size);
-
-    // first decrease memory, otherwise have artificial/wrong high peak memory usage
-    if(size<size2)
-    {
-      trace->ChangeMemory(mem_id2, size-size2);
-      trace->ChangeMemory(mem_id, size2-size);
-    }
-    else
-    {
-      trace->ChangeMemory(mem_id, size2-size);
-      trace->ChangeMemory(mem_id2, size-size2);
-    }
-  }
-
 } // namespace ngcore
 
 // Helper macro to easily add multiple timers in a function for profiling

libsrc/core/python_ngcore_export.cpp

@@ -266,7 +266,9 @@ threads : int
     .def_static("SetTraceThreads", &PajeTrace::SetTraceThreads)
     .def_static("SetTraceThreadCounter", &PajeTrace::SetTraceThreadCounter)
     .def_static("SetMaxTracefileSize", &PajeTrace::SetMaxTracefileSize)
+#ifdef NETGEN_TRACE_MEMORY
     .def_static("WriteMemoryChart", [](string filename){ if(trace) trace->WriteMemoryChart(filename); }, py::arg("filename")="memory" )
+#endif // NETGEN_TRACE_MEMORY
     ;
 
 

libsrc/core/table.hpp

@@ -159,6 +159,7 @@ namespace ngcore
     NETGEN_INLINE Table (Table && tab2)
       : FlatTable<T,IndexType>(0, nullptr, nullptr)
     {
+      tab2.mt.Free(tab2.GetMemUsage());
       Swap (size, tab2.size);
       Swap (index, tab2.index);
       Swap (data, tab2.data);
@@ -166,6 +167,7 @@ namespace ngcore
 
     NETGEN_INLINE Table & operator= (Table && tab2)
     {
+      mt.Swap(GetMemUsage(), tab2.mt, tab2.GetMemUsage());
       Swap (size, tab2.size);
       Swap (index, tab2.index);
       Swap (data, tab2.data);
@@ -177,6 +179,7 @@ namespace ngcore
     /// Delete data
     NETGEN_INLINE ~Table ()
     {
+      mt.Free(GetMemUsage());
       delete [] data;
       delete [] index;
     }
@@ -188,6 +191,16 @@ namespace ngcore
     NETGEN_INLINE size_t NElements() const { return index[size]; }
 
     using FlatTable<T,IndexType>::operator[];
+
+    NETGEN_INLINE void StartMemoryTracing (int mem_id)
+    {
+      mt.Alloc(GetMemUsage());
+    }
+    const MemoryTracer& GetMemoryTracer() const { return mt; }
+
+  private:
+    size_t GetMemUsage() const { return size == 0 ? 0 : sizeof(T)*index[size] + sizeof(IndexType) * size+1; }
+    MemoryTracer mt;
   };
 
 

libsrc/meshing/boundarylayer.cpp

@@ -90,6 +90,9 @@ namespace netgen
 
   void GenerateBoundaryLayer(Mesh& mesh, const BoundaryLayerParameters& blp)
   {
+    static Timer timer("Create Boundarylayers");
+    RegionTimer regt(timer);
+
     int max_edge_nr = -1;
     for(const auto& seg : mesh.LineSegments())
       if(seg.edgenr > max_edge_nr)