Beispiel #1
0
inline UMatrix2D<T>& UMatrix2D<T>::operator  = (UMatrix2D<T>& M)
                                        {
#ifdef _SAFE_ACCESS_
    CheckLocker cl1(GetLocker());
    CheckLocker cl2(M.GetLocker());
#endif //_SAFE_ACCESS_
    nX  = M.GetX();
    nY  = M.GetY();
    if(mt == MXT_MEM) {
        if(Ptr !=  NULL) 
#ifdef __ICC
            _mm_free(Ptr);
#else
            free(Ptr);
#endif //__ICC        
#ifdef __ICC
            Ptr = (T*)(_mm_malloc(sizeof(T)*nX*nY,_ALIGN));
#else         
            Ptr = (T*)(malloc(sizeof(T)*nX*nY));
#endif         
        memcpy(Ptr,M.GetMatrixPtr(),sizeof(T)*nX*nY);
    } else {
        Ptr = M.GetMatrixPtr();
    }

    ms = M.GetMatrixState();

    return *this;
}
int main( int argc, char **argv )
{
    const  FP  ver=_VER;
    static char    inFile[256];
#ifdef _MPI
    FlowNode2D<FP,NUM_COMPONENTS>* TmpMatrixPtr;
    int TmpMaxX;
    UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >*            TmpSubDomain    = NULL;
    UMatrix2D< FlowNodeCore2D<FP,NUM_COMPONENTS> >*        TmpCoreSubDomain= NULL;
#endif // _MPI
    ostream*     o_stream = &cout;
#ifndef   _WIN32
#ifdef    _DEBUG_0
    __ExceptLib.SetSystemException(SIGINT, AT_HANDLED);
    __ExceptLib.SetSystemException(SIGFPE, AT_HANDLED);
    __ExceptLib.SetSystemException(SIGSEGV,AT_HANDLED);
    __ExceptLib.SetSystemException(SIGBUS, AT_HANDLED);
#endif //_DEBUG_0
#endif //_WIN32

#ifdef _DEBUG_0
#ifndef _WIN32
    ___try {
#endif  // _WIN32
#endif  // _DEBUG_0
        if (argc < 2) {
            printf("OpenHyperFLOW2D/DEEPS/FP%u solver v %'.2f ",(unsigned int)(sizeof(FP)*8),ver);
#ifndef _MPI
#ifndef _OPEN_MP
            printf(" (serial version)\n");
#else
            printf(" (parallel OpenMP version)\n");
#endif // OPEN_MP
#else
            printf("(parallel MPI version)\n");
#endif // _MPI
            printf("Copyright (C) 1995-2016 by Serge A. Suchkov\nCopyright policy: LGPL V3\nUsage: %s [{input_data_file}]\n",argv[0]);

            printf("\n\t* Density-based 2D-Navier-Stokes solver for ");
#ifdef _UNIFORM_MESH_
            printf("uniform cartesian mesh");
#else
            printf("non-uniform mesh");
#endif //_UNIFORM_MESH_
            printf("\n\n\tFlowNode2D size = %d bytes\n\n",(int)(sizeof(FlowNode2D<FP,NUM_COMPONENTS>)));
            exit(0);
        } else {
#ifdef _MPI
            MPI::Init(argc, argv);
rank      = MPI::COMM_WORLD.Get_rank();
#endif // _MPI
            sprintf(inFile,"%s",argv[1]);
            Data = new InputData(inFile,DS_FILE,o_stream,0,10
#ifdef _MPI
                                 ,rank
#endif // _MPI
                                 );
            if (Data->GetDataError()!=0) {
#ifdef _MPI
                if( rank == 0 ) {
#endif // _MPI
                    *o_stream << "\nInput data error.\n" ;
                    o_stream->flush();
#ifdef _MPI

                }
                Exit_OpenHyperFLOW2D();
#else
                exit(0);
#endif // _MPI
            }
        }
#ifdef _MPI
//------------------------- MPI version ------------------------------------
	    last_rank = MPI::COMM_WORLD.Get_size()-1;
           if(last_rank == 0) {
	       printf("At least 2 CPU should be used.\n Check mpiexec or mpirun parameters\n");
	       Exit_OpenHyperFLOW2D();
	    }
	        InitSharedData(Data,&chemical_reactions,rank);        // Init shared data
            MPI::COMM_WORLD.Barrier();
            
            if (rank == 0) {
               ArraySubDomain = new UArray< UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >* >();
               InitDEEPS2D((void*)o_stream);                       // Init solver (only for rank=0)
               
               gettimeofday(&mark2,NULL);

               if(ProblemType == SM_NS) {
                   // Scan area for seek wall nodes      
                   *o_stream << "\nSolver Mode: Navier-Stokes/FP" << 8*sizeof(FP) <<"\n" << endl;
                   WallNodes = GetWallNodes((ofstream*)o_stream,J,Data->GetIntVal((char*)"isVerboseOutput")); 

 #ifndef     _PARALLEL_RECALC_Y_PLUS_
                   *o_stream << "\nSerial calc min distance to wall..." << flush;
                   SetMinDistanceToWall2D(J,WallNodes);
                   *o_stream << "OK\n" << flush;
                   Recalc_y_plus(J,WallNodes);                        // Calculate initial y+ value         
                   
                   if(!PreloadFlag)
                      SetInitBoundaryLayer(J,delta_bl);               // Set Initial boundary layer profile
 #else
                   *o_stream << "\nParallel calc min distance to wall..." << endl;
                   
 #endif //   _PARALLEL_RECALC_Y_PLUS_
                   NumWallNodes = WallNodes->GetNumElements();
               } else {
                   *o_stream << "\nSolver Mode: Euler/FP" << 8*sizeof(FP) <<"\n" << endl;
               }
               
               TmpMatrixPtr=J->GetMatrixPtr();
               int SubStartIndex, SubMaxX,
                            r_Overlap=0,
                            l_Overlap=0;
               *o_stream << "Allocate SubDomain:\n";
               SubStartIndex = 0;
               
               for (unsigned int i=0;i<GlobalSubDomain->GetNumElements();i++) {
                   SubStartIndex = GlobalSubDomain->GetElementPtr(i)->GetX();  
                   SubMaxX = GlobalSubDomain->GetElementPtr(i)->GetY();
               if(i == GlobalSubDomain->GetNumElements()-1)
                 r_Overlap = 0;
               else
                 r_Overlap = 1; 
               if(i == 0)
                 l_Overlap = 0;
               else
                 l_Overlap = 1; 

               TmpMaxX = (SubMaxX-SubStartIndex)+r_Overlap;
               TmpMatrixPtr = (FlowNode2D<FP,NUM_COMPONENTS>*)((ulong)J->GetMatrixPtr()+(ulong)(sizeof(FlowNode2D<FP,NUM_COMPONENTS>)*(SubStartIndex)*MaxY));                      
               
               x0 = SubStartIndex*FlowNode2D<FP,NUM_COMPONENTS>::dx;
               
               *o_stream << "SubDomain("<<i<<")[" << TmpMaxX << "x" << MaxY << "]  Size=" << (ulong)(sizeof(FlowNode2D<FP,NUM_COMPONENTS>)*TmpMaxX*MaxY)/(1024*1024) << " Mb\n"; 
               TmpSubDomain = new UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >(TmpMatrixPtr,TmpMaxX,MaxY);
               
               if(ProblemType == SM_NS) {
#ifdef _PARALLEL_RECALC_Y_PLUS_
                  SetMinDistanceToWall2D(TmpSubDomain,WallNodes,x0);
#endif // _PARALLEL_RECALC_Y_PLUS_
               }

               ArraySubDomain->AddElement(&TmpSubDomain);
               o_stream->flush();
               
               if(i>0) {
                  MPI::COMM_WORLD.Send(&MaxY,1,MPI::INT,i,tag_MaxY);
                  MPI::COMM_WORLD.Send(&TmpMaxX,1,MPI::INT,i,tag_MaxX);
#ifdef _IMPI_
                  LongMatrixSend(i, TmpSubDomain->GetMatrixPtr(), TmpSubDomain->GetMatrixSize());// Low Mem Send subdomain
#else
                  MPI::COMM_WORLD.Send(TmpSubDomain->GetMatrixPtr(),
                                       TmpSubDomain->GetMatrixSize(),
                                       MPI::BYTE,i,tag_Matrix); 
#endif // _IMPI_
                  
                  if(ProblemType == SM_NS) {
                     MPI::COMM_WORLD.Send(&NumWallNodes,1,MPI::INT,i,tag_NumWallNodes);           // Send wall nodes array size
                     MPI::COMM_WORLD.Send(WallNodes->GetArrayPtr(),NumWallNodes*sizeof(XY<int>),  // Send wall nodes array
                                           MPI::BYTE,i,tag_WallNodesArray);
                  }

                  MPI::COMM_WORLD.Send(&x0,1,MPI::DOUBLE,i,tag_X0);                               // Send x0 for SubDomain
                }
               if(MonitorPointsArray) {
                   for(int ii_monitor=0;ii_monitor<(int)MonitorPointsArray->GetNumElements();ii_monitor++) {
                           if(MonitorPointsArray->GetElement(ii_monitor).MonitorXY.GetX() >= x0 &&
                              MonitorPointsArray->GetElement(ii_monitor).MonitorXY.GetX() < x0 + FlowNode2D<FP,NUM_COMPONENTS>::dx*TmpSubDomain->GetX()) {
                              MonitorPointsArray->GetElement(ii_monitor).rank = i; 
                           }
                   }
               }
           }
           TmpMaxX      = ArraySubDomain->GetElement(0)->GetX();
           MaxY         = ArraySubDomain->GetElement(0)->GetY();
           TmpSubDomain = ArraySubDomain->GetElement(0);
        } else {
           
           MPI::COMM_WORLD.Recv(&MaxY,1,MPI::INT,0,tag_MaxY);
           MPI::COMM_WORLD.Recv(&TmpMaxX,1,MPI::INT,0,tag_MaxX);
           TmpSubDomain = new UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >(TmpMaxX,MaxY);
#ifdef _IMPI_
             LongMatrixRecv(0,TmpSubDomain->GetMatrixPtr(),TmpSubDomain->GetMatrixSize());
#else
             MPI::COMM_WORLD.Recv(TmpSubDomain->GetMatrixPtr(),
                                  TmpSubDomain->GetMatrixSize(),
                                  MPI::BYTE,0,tag_Matrix);
#endif // _IMPI_
             if(ProblemType == SM_NS) {
                 MPI::COMM_WORLD.Recv(&NumWallNodes,1,MPI::INT,0,tag_NumWallNodes);                      // Recive wall nodes array size
                 WallNodes = new UArray< XY<int> >(NumWallNodes,-1);                                     // Create wall nodes array
                 MPI::COMM_WORLD.Recv(WallNodes->GetArrayPtr(),NumWallNodes*sizeof(XY<int>),             // Recive wall nodes array
                                      MPI::BYTE,0,tag_WallNodesArray);
                 WallNodesUw_2D = new UArray<FP>(NumWallNodes,-1);                                       // Create friction velosity array
             }

             MPI::COMM_WORLD.Recv(&x0,1,MPI::DOUBLE,0,tag_X0);                                           // Recive x0 for SubDomain
       }
        
        if(MonitorPointsArray && 
           MonitorPointsArray->GetNumElements() > 0) {
                   for(int ii_monitor=0;ii_monitor<(int)MonitorPointsArray->GetNumElements();ii_monitor++) {
                       MPI::COMM_WORLD.Bcast(&MonitorPointsArray->GetElement(ii_monitor).rank,
                                             sizeof(int),
                                             MPI::INT,
                                             0);
                   }
        }

        
        MPI::COMM_WORLD.Barrier();
        TmpCoreSubDomain = new UMatrix2D< FlowNodeCore2D<FP,NUM_COMPONENTS> >(TmpMaxX,MaxY);
        
#ifdef _PARALLEL_RECALC_Y_PLUS_
        if(ProblemType == SM_NS) {
            if(!PreloadFlag)
               SetInitBoundaryLayer(TmpSubDomain,delta_bl);                                               // Set Initial boundary layer linear profile
            if( rank == 0 ) {
                if(WallNodes && !WallNodesUw_2D && J) 
                    WallNodesUw_2D = GetWallFrictionVelocityArray2D(J,WallNodes);
                else if(WallNodes && J)
                    RecalcWallFrictionVelocityArray2D(J,WallNodesUw_2D,WallNodes);
                *o_stream << "Parallel recalc y+...";
                for(int ii = 1; (int)ii < last_rank + 1; ii++ ) {
                    MPI::COMM_WORLD.Send(WallNodesUw_2D->GetArrayPtr(),
                                         WallNodesUw_2D->GetNumElements()*WallNodesUw_2D->GetElementSize(),
                                         MPI::BYTE,ii,tag_WallFrictionVelocity);
                }
            } else {
                    MPI::COMM_WORLD.Recv(WallNodesUw_2D->GetArrayPtr(),
                                         WallNodesUw_2D->GetNumElements()*WallNodesUw_2D->GetElementSize(),
                                         MPI::BYTE,0,tag_WallFrictionVelocity);
            }
          ParallelRecalc_y_plus(TmpSubDomain,WallNodes,WallNodesUw_2D,x0);
        }
#endif // _PARALLEL_RECALC_Y_PLUS_
     
        if(rank == 0) {
            gettimeofday(&mark1,NULL);
            
            *o_stream << "OK\n" << "Time: " << (FP)(mark1.tv_sec-mark2.tv_sec)+(FP)(mark1.tv_usec-mark2.tv_usec)*1.e-6 << " sec." << endl; 
            *o_stream << "\nStart computation...\n" << flush;
            
            x0 = 0;

            //>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
            //DataSnapshot(OutFileName,WM_REWRITE);    //
            //>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
     
        }
        MPI::COMM_WORLD.Barrier();

        DEEPS2D_Run((ofstream*)o_stream, 
                      TmpSubDomain,
                      TmpCoreSubDomain,
                      rank, last_rank, x0);
//------------------------- MPI version ------------------------------------
#else
//---------------------- OpenMP/Single thread version ----------------------
       InitSharedData(Data,&chemical_reactions);          // Init shared data
       InitDEEPS2D((void*)o_stream);                      // Init solver 
       
       if(ProblemType == SM_NS) {
           // Scan area for seek wall nodes
           WallNodes = GetWallNodes((ofstream*)o_stream,J,Data->GetIntVal((char*)"isVerboseOutput")); 
           SetMinDistanceToWall2D(J,WallNodes);
           Recalc_y_plus(J,WallNodes);                        // Calculate initial y+ value
           
           if(!PreloadFlag)
              SetInitBoundaryLayer(J,delta_bl);               // Set Initial boundary layer profile
       }
       UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >* TmpSubDomain=NULL;
       UMatrix2D< FlowNodeCore2D<FP,NUM_COMPONENTS> >* TmpCoreSubDomain=NULL;
       FlowNode2D<FP,NUM_COMPONENTS>* TmpMatrixPtr=J->GetMatrixPtr();
       int SubStartIndex, Overlap, SubMaxX;
       *o_stream << "Allocate SubDomain:\n";
       
       for (unsigned int i=0;i<GlobalSubDomain->GetNumElements();i++) {
            SubStartIndex = GlobalSubDomain->GetElementPtr(i)->GetX();  
            SubMaxX = GlobalSubDomain->GetElementPtr(i)->GetY();
            if(i == GlobalSubDomain->GetNumElements()-1)
               Overlap = 0;
            else
               Overlap = 1;
            TmpMatrixPtr=(FlowNode2D<FP,NUM_COMPONENTS>*)((ulong)J->GetMatrixPtr()+(ulong)(sizeof(FlowNode2D<FP,NUM_COMPONENTS>)*SubStartIndex*MaxY));
            TmpSubDomain     = new UMatrix2D< FlowNode2D<FP,NUM_COMPONENTS> >(TmpMatrixPtr,(SubMaxX-SubStartIndex)+Overlap,MaxY);
            TmpCoreSubDomain = new UMatrix2D< FlowNodeCore2D<FP,NUM_COMPONENTS> >((SubMaxX-SubStartIndex)+Overlap,MaxY);
            SubDomainArray->AddElement(&TmpSubDomain);
            CoreSubDomainArray->AddElement(&TmpCoreSubDomain);
           }
       SubStartIndex = 0;
       for(unsigned int i=0;i<SubDomainArray->GetNumElements();i++) {
          *o_stream << "SubDomain(" << i << ")[" << SubDomainArray->GetElement(i)->GetX() << "x" << \
          SubDomainArray->GetElement(i)->GetY() << "] Size= " << SubDomainArray->GetElement(i)->GetMatrixSize()/(1024*1024) << "+" \
          << CoreSubDomainArray->GetElement(i)->GetMatrixSize()/(1024*1024) << " Mb\n"; 
           o_stream->flush();
          }
       *o_stream << "\nStart computation...\n" << flush;
       o_stream->flush();
       DEEPS2D_Run((ofstream*)o_stream);
//---------------------- OpenMP version ------------------------------------
#endif // _MPI

#ifdef _DEBUG_0
    }__except(SysException e) {{
            const char ConstErrorMessage[]=" handled by <LibExcept> module in <InputData>.\n";
            *o_stream << "SIG";
            if (e == SIGSEGV)
                *o_stream << "SEGV" << ConstErrorMessage ;
            else if (e == SIGBUS)
                *o_stream << "BUS" << ConstErrorMessage ;
            else if (e == SIGFPE)
                *o_stream << "FPE" << ConstErrorMessage ;
            else if (e == SIGINT)
                *o_stream << "INT" << ConstErrorMessage ;
            else if (e == SIGABRT)
                *o_stream << "ABRT" << ConstErrorMessage ;
            else if (e == SIGIO)
                *o_stream << "IO" << ConstErrorMessage ;
            else if (e == SIGTRAP)
                *o_stream << "TRAP" << ConstErrorMessage ;
            else
                *o_stream << " No: "<< e << ConstErrorMessage ;
        }
    } __end_except;