- fixed some compiler warnings of the cpp-runtime build with gcc

- added HPCOM_MEMORY_OPT-Flag to use experimental memory optimization
- some simple optimizations for the cpp-runtime
- papi-support to hpcom-codegen added

git-svn-id: https://openmodelica.org/svn/OpenModelica/trunk@21261 f25d12d1-65f4-0310-ae8a-bbce733d8d8e

Compiler/BackEnd/HpcOmMemory.mo

@@ -128,30 +128,30 @@ encapsulated package HpcOmMemory
     oMemoryMap := matchcontinue(iModelInfo, iTaskGraph, iTaskGraphMeta, iEqSystems, iFileNamePrefix, iSchedulerInfo, iSchedule, iSccSimEqMapping, iCriticalPaths, iCriticalPathsWoC, iCriticalPathInfo, iAllComponents)
       case(_,_,_,_,_,_,_,_,_,_,_,_)
         equation
-          true = Flags.isSet(Flags.HPCOM_ANALYZATION_MODE);
-          HpcOmTaskGraph.printTaskGraphMeta(iTaskGraphMeta);
+          true = Flags.isSet(Flags.HPCOM_MEMORY_OPT);
+          //HpcOmTaskGraph.printTaskGraphMeta(iTaskGraphMeta);
           //Create var hash table
           SimCode.MODELINFO(vars=simCodeVars) = iModelInfo;
           SimCode.SIMVARS(stateVars=stateVars, derivativeVars=derivativeVars, algVars=algVars, paramVars=paramVars) = simCodeVars;
           allVarsMapping = SimCodeUtil.createIdxSCVarMapping(simCodeVars);
-          SimCodeUtil.dumpIdxScVarMapping(allVarsMapping);
-          print("--------------------------------\n");
+          //SimCodeUtil.dumpIdxScVarMapping(allVarsMapping);
+          //print("--------------------------------\n");
           hashTable = HashTableCrILst.emptyHashTableSized(BaseHashTable.biggerBucketSize);
           hashTable = fillSimVarHashTable(stateVars,0,0,hashTable);
           hashTable = fillSimVarHashTable(derivativeVars,listLength(stateVars),0,hashTable);
           hashTable = fillSimVarHashTable(algVars,listLength(stateVars) + listLength(stateVars),0,hashTable);
           //hashTable = fillSimVarHashTable(paramVars,listLength(stateVars)*2 + listLength(algVars),0,hashTable);
-          print("-------------------------------------\n");
-          BaseHashTable.dumpHashTable(hashTable);
+          //print("-------------------------------------\n");
+          //BaseHashTable.dumpHashTable(hashTable);
           //Create CacheMap
           sccNodeMapping = HpcOmTaskGraph.getSccNodeMapping(arrayLength(iSccSimEqMapping), iTaskGraphMeta);
-          printSccNodeMapping(sccNodeMapping);
+          //printSccNodeMapping(sccNodeMapping);
           scVarTaskMapping = getSimCodeVarNodeMapping(iTaskGraphMeta,iEqSystems,listLength(stateVars)*2+listLength(algVars),sccNodeMapping,hashTable);
-          printScVarTaskMapping(scVarTaskMapping);
-          print("-------------------------------------\n");
+          //printScVarTaskMapping(scVarTaskMapping);
+          //print("-------------------------------------\n");
           nodeSimCodeVarMapping = getNodeSimCodeVarMapping(iTaskGraphMeta, iEqSystems, hashTable);
-          printNodeSimCodeVarMapping(nodeSimCodeVarMapping);
-          print("-------------------------------------\n");
+          //printNodeSimCodeVarMapping(nodeSimCodeVarMapping);
+          //print("-------------------------------------\n");
           (cacheMap,scVarCLMapping,numCL) = createCacheMapOptimized(allVarsMapping,stateVars,derivativeVars,algVars,paramVars,scVarTaskMapping,64,iAllComponents,iSchedule, nodeSimCodeVarMapping);
           eqSimCodeVarMapping = getEqSCVarMapping(iEqSystems,hashTable);
           (clTaskMapping,scVarTaskMapping) = getCacheLineTaskMapping(iTaskGraphMeta,iEqSystems,hashTable,numCL,scVarCLMapping);

Compiler/BackEnd/HpcOmSimCodeMain.mo

@@ -251,6 +251,7 @@ algorithm
 
       fileName = ("taskGraph"+&filenamePrefix+&"ODE_merged.graphml");
       HpcOmTaskGraph.dumpAsGraphMLSccLevel(taskGraphSimplified, taskGraphDataSimplified, inBackendDAE, fileName, criticalPathInfo, HpcOmTaskGraph.convertNodeListToEdgeTuples(List.first(criticalPaths)), HpcOmTaskGraph.convertNodeListToEdgeTuples(List.first(criticalPathsWoC)), sccSimEqMapping, schedulerInfo);
+      Debug.execStat("hpcom dump simplified TaskGraph", GlobalScript.RT_CLOCK_EXECSTAT_HPCOM_MODULES);
 
       //Create schedule
       //---------------
@@ -271,18 +272,24 @@ algorithm
       SimCode.SIMCODE(modelInfo, simCodeLiterals, simCodeRecordDecls, simCodeExternalFunctionIncludes, allEquations, odeEquations, algebraicEquations, residualEquations, useSymbolicInitialization, useHomotopy, initialEquations, removedInitialEquations, startValueEquations, nominalValueEquations, minValueEquations, maxValueEquations,
                  parameterEquations, removedEquations, algorithmAndEquationAsserts, zeroCrossingsEquations, jacobianEquations, stateSets, constraints, classAttributes, zeroCrossings, relations, timeEvents, whenClauses,
                  discreteModelVars, extObjInfo, makefileParams, delayedExps, jacobianMatrixes, simulationSettingsOpt, fileNamePrefix, _, _, crefToSimVarHT, backendMapping) = simCode;
-
+
+      //Check ODE-System size
+      //---------------------
+      System.realtimeTick(GlobalScript.RT_CLOCK_EXECSTAT_HPCOM_MODULES);
       checkOdeSystemSize(taskGraphOde,odeEquations);
       Debug.execStat("hpcom check ODE system size", GlobalScript.RT_CLOCK_EXECSTAT_HPCOM_MODULES);
-
+
+      //Create Memory-Map and Sim-Code
+      //------------------------------
       optTmpMemoryMap = HpcOmMemory.createMemoryMap(modelInfo, taskGraphSimplified, taskGraphDataSimplified, eqs, filenamePrefix, schedulerInfo, schedule, sccSimEqMapping, criticalPaths, criticalPathsWoC, criticalPathInfo, allComps);
-
+      Debug.execStat("hpcom create memory map", GlobalScript.RT_CLOCK_EXECSTAT_HPCOM_MODULES);
+
       simCode = SimCode.SIMCODE(modelInfo, simCodeLiterals, simCodeRecordDecls, simCodeExternalFunctionIncludes, allEquations, odeEquations, algebraicEquations, residualEquations, useSymbolicInitialization, useHomotopy, initialEquations, removedInitialEquations, startValueEquations, nominalValueEquations, minValueEquations, maxValueEquations,
                  parameterEquations, removedEquations, algorithmAndEquationAsserts, zeroCrossingsEquations, jacobianEquations, stateSets, constraints, classAttributes, zeroCrossings, relations, timeEvents, whenClauses,
                  discreteModelVars, extObjInfo, makefileParams, delayedExps, jacobianMatrixes, simulationSettingsOpt, fileNamePrefix, SOME(schedule), optTmpMemoryMap, crefToSimVarHT, backendMapping);
 
       //evaluateCacheBehaviour(schedulerInfo,taskGraphDataSimplified,clTaskMapping, transposeCacheLineTaskMapping(clTaskMapping, arrayLength(taskGraphSimplified)));
-
+      Debug.execStat("hpcom other", GlobalScript.RT_CLOCK_EXECSTAT_HPCOM_MODULES);
       print("HpcOm is still under construction.\n");
       then simCode;
     else equation
@@ -374,6 +381,11 @@ protected
   array<tuple<Integer,Integer>> schedulerInfo;
 algorithm
   (oTaskGraph,oTaskGraphMeta) := matchcontinue(iTaskGraph,iTaskGraphMeta,inBackendDAE,iApplyFilters)
+    //case(_,_,_,_)
+    //  equation
+    //    flagValue = Flags.getConfigString(Flags.HPCOM_SCHEDULER);
+    //    true = stringEq(flagValue, "level");
+    //  then (iTaskGraph, iTaskGraphMeta);
     case(_,_,_,true)
       equation
         //Merge simple and parent nodes

Compiler/Template/CodegenCpp.tpl

@@ -53,7 +53,7 @@ template simulationHeaderFile(SimCode simCode)
 match simCode
 case SIMCODE(__) then
   <<
-   <%generateHeaderInlcudeString(simCode)%>
+   <%generateHeaderIncludeString(simCode)%>
    <%generateClassDeclarationCode(simCode)%>
 
 
@@ -3697,7 +3697,7 @@ template writeoutput4(Integer index, Integer myindex2)
  >>
 end writeoutput4;
 
-template generateHeaderInlcudeString(SimCode simCode)
+template generateHeaderIncludeString(SimCode simCode)
  "Generates header part of simulation file."
 ::=
 match simCode
@@ -3718,7 +3718,7 @@ case SIMCODE(modelInfo=MODELINFO(__), extObjInfo=EXTOBJINFO(__)) then
   *
   *****************************************************************************/
    >>
-end generateHeaderInlcudeString;
+end generateHeaderIncludeString;
 
 
 
@@ -10066,16 +10066,15 @@ template checkForDiscreteEvents(list<ComponentRef> discreteModelVars,SimCode sim
 ::=
 
   let changediscreteVars = (discreteModelVars |> var => match var case CREF_QUAL(__) case CREF_IDENT(__) then
-       'if (_event_handling.changeDiscreteVar(<%cref(var)%>,"<%cref(var)%>")) {  restart=true; }'
+       'if (_event_handling.changeDiscreteVar(<%cref(var)%>,"<%cref(var)%>")) {  return true; }'
        ;separator="\n")
   match simCode
   case SIMCODE(modelInfo = MODELINFO(__)) then
   <<
   bool <%lastIdentOfPath(modelInfo.name)%>::checkForDiscreteEvents()
   {
-    bool restart = false;
     <%changediscreteVars%>
-    return restart;
+    return false;
   }
   >>
 end checkForDiscreteEvents;

Compiler/Template/CodegenCppHpcom.tpl

@@ -58,7 +58,7 @@ template simulationHeaderFile(SimCode simCode)
 match simCode
 case SIMCODE(__) then
   <<
-   <%generateHeaderInlcudeString(simCode)%>
+     <%generateHeaderIncludeString(simCode)%>
      #ifdef ANALYZATION_MODE
       #include <boost/shared_ptr.hpp>
       #include <boost/weak_ptr.hpp>
@@ -77,9 +77,12 @@ case SIMCODE(__) then
       #include <boost/numeric/ublas/matrix_sparse.hpp>
       typedef uBlas::compressed_matrix<double, uBlas::column_major, 0, uBlas::unbounded_array<int>, uBlas::unbounded_array<double> > SparseMatrix;
      #endif
-   <%generateHpcomSpecificIncludes(simCode)%>
-   <%generateClassDeclarationCode(simCode)%>
-
+     <%generateHpcomSpecificIncludes(simCode)%>
+     <%generateClassDeclarationCode(simCode)%>
+     #ifdef MEASURE_PAPI
+     #include <papi.h>
+     #define NUM_EVENTS 1
+     #endif
    >>
 end simulationHeaderFile;
 
@@ -296,7 +299,7 @@ template generateHpcomSpecificIncludes(SimCode simCode)
         >>
         case ("mixed") then
         <<
-        #include <omp.h>
+        #define NUM_EVENTS 1
         #include <boost/thread/mutex.hpp>
         #include <boost/thread.hpp>
         >>
@@ -730,6 +733,7 @@ template update2(list<SimEqSystem> allEquationsPlusWhen, list<list<SimEqSystem>>
           <<
           static bool state_var_reinitialized = false;
           static bool firstRun = true;
+
           void <%lastIdentOfPath(name)%>::evaluateThreadFunc0()
           {
             //if (omp_get_dynamic())
@@ -741,21 +745,51 @@ template update2(list<SimEqSystem> allEquationsPlusWhen, list<list<SimEqSystem>>
             {
                 while(!finished)
                 {
+                    #ifdef MEASURE_PAPI
+                    int event[NUM_EVENTS] = {PAPI_L2_TCM};
+                    long long values[NUM_EVENTS];
+                    #endif
+
                     #pragma omp master
                     {
                         <%function_HPCOM_assignLock("startEvaluateLock","","pthreads")%>
+                        #ifdef MEASURE_PAPI
+                        /* Start counting events */
+                        if (PAPI_start_counters(event, NUM_EVENTS) != PAPI_OK) {
+                            fprintf(stderr, "PAPI_start_counters - FAILED\n");
+                            exit(1);
+                        }
+                        #endif
                     }
+
                     #pragma omp barrier
                     if(finished)
                         <%function_HPCOM_releaseLock("finishedEvaluateLock","","pthreads")%>
 
                     <%odeEqs%>
-
+                    
                     #pragma omp barrier
+
                     #pragma omp master
                     {
+                        #ifdef MEASURE_PAPI
+                        /* Read the counters */
+                        if (PAPI_read_counters(values, NUM_EVENTS) != PAPI_OK) {
+                            fprintf(stderr, "PAPI_read_counters - FAILED\n");
+                            exit(1);
+                        }
+                        std::cerr << "L2 Cache misses: " << values[0] << std::endl;
+
+                        /* Stop counting events */
+                        if (PAPI_stop_counters(values, NUM_EVENTS) != PAPI_OK) {
+                            fprintf(stderr, "PAPI_stoped_counters - FAILED\n");
+                            exit(1);
+                        }
+                        #endif
                         <%function_HPCOM_releaseLock("finishedEvaluateLock","","pthreads")%>
                     }
+
+
                 }
             }
           }
@@ -779,10 +813,36 @@ template update2(list<SimEqSystem> allEquationsPlusWhen, list<list<SimEqSystem>>
           <<
           void <%lastIdentOfPath(name)%>::evaluateODE(const UPDATETYPE command)
           {
+             #ifdef MEASURE_PAPI
+             int event[NUM_EVENTS] = {PAPI_L2_TCM};
+             long long values[NUM_EVENTS];
+
+             /* Start counting events */
+             if (PAPI_start_counters(event, NUM_EVENTS) != PAPI_OK) {
+                fprintf(stderr, "PAPI_start_counters - FAILED\n");
+                exit(1);
+             }
+             #endif
+
              #pragma omp parallel num_threads(<%getConfigInt(NUM_PROC)%>)
              {
                 <%odeEqs%>
              }
+
+             #ifdef MEASURE_PAPI
+             /* Read the counters */
+             if (PAPI_read_counters(values, NUM_EVENTS) != PAPI_OK) {
+                fprintf(stderr, "PAPI_read_counters - FAILED\n");
+                exit(1);
+             }
+             std::cerr << "L2 Cache misses: " << values[0] << std::endl;
+
+             /* Stop counting events */
+             if (PAPI_stop_counters(values, NUM_EVENTS) != PAPI_OK) {
+                fprintf(stderr, "PAPI_stoped_counters - FAILED\n");
+                exit(1);
+             }
+             #endif
           }
           >>
         else ""
@@ -1394,7 +1454,7 @@ CFLAGS_BASED_ON_INIT_FILE=<%_extraCflags%> -I"<%makefileParams.omhome%>/../Simul
 CFLAGS=$(CFLAGS_BASED_ON_INIT_FILE) -Winvalid-pch $(SYSTEM_CFLAGS) -I"<%makefileParams.omhome%>/include/omc/cpp/Core" -I"<%makefileParams.omhome%>/include/omc/cpp/"   -I. <%makefileParams.includes%> -I"$(BOOST_INCLUDE)" <%makefileParams.includes ; separator=" "%> <%makefileParams.cflags%> <%match sopt case SOME(s as SIMULATION_SETTINGS(__)) then s.cflags %>
 LDSYTEMFLAGS=-L"<%makefileParams.omhome%>/lib/omc/cpp"    -L"$(BOOST_LIBS)"
 CPP_RUNTIME_LIBS=<%analyzationLibs%>
-LDMAINFLAGS=-L"<%makefileParams.omhome%>/lib/omc/cpp" <%simulationMainDLLib(simCode)%> -L"<%makefileParams.omhome%>/bin" <%schedulerLibs%> $(CPP_RUNTIME_LIBS) -L"$(BOOST_LIBS)" $(BOOST_SYSTEM_LIB) $(BOOST_FILESYSTEM_LIB) $(BOOST_PROGRAM_OPTIONS_LIB) $(BOOST_SERIALIZATION_LIB) $(BOOST_THREAD_LIB) $(LINUX_LIB_DL)
+LDMAINFLAGS=-L"<%makefileParams.omhome%>/lib/omc/cpp" <%simulationMainDLLib(simCode)%> -L"<%makefileParams.omhome%>/bin" <%schedulerLibs%> $(CPP_RUNTIME_LIBS) -L"$(BOOST_LIBS)" $(BOOST_SYSTEM_LIB) $(BOOST_FILESYSTEM_LIB) $(BOOST_PROGRAM_OPTIONS_LIB) $(BOOST_SERIALIZATION_LIB) $(BOOST_THREAD_LIB) $(LINUX_LIB_DL) -lpapi
 CPPFLAGS = $(CFLAGS) -DOMC_BUILD -DBOOST_SYSTEM_NO_DEPRICATED
 SYSTEMFILE=OMCpp<%fileNamePrefix%><% if acceptMetaModelicaGrammar() then ".conv"%>.cpp
 FUNCTIONFILE=OMCpp<%fileNamePrefix%>Functions.cpp

Compiler/Util/Flags.mo

@@ -391,7 +391,9 @@ constant DebugFlag ITERATION_VARS = DEBUG_FLAG(112, "iterationVars", false,
   Util.gettext("Shows a list of all iteration variables."));
 constant DebugFlag ALLOW_RECORD_TOO_MANY_FIELDS = DEBUG_FLAG(113, "acceptTooManyFields", false,
   Util.gettext("Accepts passing records with more fields than expected to a function. This is not allowed, but is used in Fluid.Dissipation. See https://trac.modelica.org/Modelica/ticket/1245 for details."));
-
+constant DebugFlag HPCOM_MEMORY_OPT = DEBUG_FLAG(114, "hpcomMemoryOpt", false,
+  Util.gettext("Optimize the memory structure regarding the selected scheduler"));
+
 // This is a list of all debug flags, to keep track of which flags are used. A
 // flag can not be used unless it's in this list, and the list is checked at
 // initialization so that all flags are sorted by index (and thus have unique
@@ -509,7 +511,8 @@ constant list<DebugFlag> allDebugFlags = {
   EVAL_FUNC_DUMP,
   PRINT_STRUCTURAL,
   ITERATION_VARS,
-  ALLOW_RECORD_TOO_MANY_FIELDS
+  ALLOW_RECORD_TOO_MANY_FIELDS,
+  HPCOM_MEMORY_OPT
 };
 
 // CONFIGURATION FLAGS

SimulationRuntime/cpp/Core/Solver/SystemStateSelection.cpp

@@ -7,8 +7,8 @@
 
 SystemStateSelection::SystemStateSelection(IMixedSystem* system)
       :_system(system)
-      ,_colPivot(NULL)
-      ,_rowPivot(NULL)
+      ,_colPivot()
+      ,_rowPivot()
       ,_initialized(false)
 {
       _state_selection = dynamic_cast<IStateSelection*>(system);

SimulationRuntime/cpp/Core/System/EventHandling.cpp

@@ -10,14 +10,18 @@ Constructor
 \param system Modelica system object
 \param dim Dimenson of help variables
 */
-EventHandling::EventHandling()
+EventHandling::EventHandling() : _countinous_system(NULL), _mixed_system(NULL), _conditions0(NULL), _conditions1(NULL)
 //:_h(NULL)
 {
 }
 
 EventHandling::~EventHandling(void)
 {
    // if(_h) delete [] _h;
+  if(_conditions0)
+    delete[] _conditions0;
+  if(_conditions1)
+    delete[] _conditions1;
 }
 /**
 Inits the event variables
@@ -26,6 +30,9 @@ void EventHandling::initialize(IEvent* system,int dim,init_prevars_type init_pre
 {
    // _dimH=dim;
     _event_system=system;
+    _countinous_system = dynamic_cast<IContinuous*>(_event_system);
+    _mixed_system= dynamic_cast<IMixedSystem*>(_event_system);
+
     init_prevars(_pre_vars_idx,_pre_discrete_vars_idx);
     _pre_vars.resize((boost::extents[_pre_vars_idx.size()]));
     _pre_discrete_vars.resize((boost::extents[_pre_discrete_vars_idx.size()]));
@@ -37,6 +44,13 @@ void EventHandling::initialize(IEvent* system,int dim,init_prevars_type init_pre
         memset(_h,0,(_dimH)*sizeof(double));
     }
     */
+    if(_conditions0)
+      delete[] _conditions0;
+    if(_conditions1)
+      delete[] _conditions1;
+
+    _conditions0 = new bool[dim];
+    _conditions1 = new bool[dim];
 }
 /**
 Returns the help vector
@@ -165,29 +179,24 @@ Handles  all events occured a the same time. These are stored  the eventqueue
 
 bool EventHandling::IterateEventQueue(bool& state_vars_reinitialized)
 {
-    IContinuous*  countinous_system = dynamic_cast<IContinuous*>(_event_system);
-    IMixedSystem* mixed_system= dynamic_cast<IMixedSystem*>(_event_system);
-
     //save discrete varibales
     _event_system->saveDiscreteVars(); // store values of discrete vars vor next check
 
     unsigned int dim = _event_system->getDimZeroFunc();
-    bool* conditions0 = new bool[dim];
-    bool* conditions1 = new bool[dim];
-    _event_system->getConditions(conditions0);
+
+    _event_system->getConditions(_conditions0);
     //Handle all events
 
-    state_vars_reinitialized =     countinous_system->evaluateAll();
+    state_vars_reinitialized = _countinous_system->evaluateAll();
 
 
     //check if discrete variables changed
     bool drestart= _event_system->checkForDiscreteEvents();
 
 
-    _event_system->getConditions(conditions1);
-    bool crestart = !std::equal (conditions1, conditions1+dim,conditions0);
-    delete[] conditions0;
-    delete [] conditions1;
+    _event_system->getConditions(_conditions1);
+    bool crestart = !std::equal (_conditions1, _conditions1+dim,_conditions0);
+
     return((drestart||crestart)); //returns true if new events occured
 }