当前位置: 首页>>代码示例>>C++>>正文

C++ AtomVec::pack_comm方法代码示例

本文整理汇总了C++中AtomVec::pack_comm方法的典型用法代码示例。如果您正苦于以下问题：C++ AtomVec::pack_comm方法的具体用法？C++ AtomVec::pack_comm怎么用？C++ AtomVec::pack_comm使用的例子？那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类AtomVec的用法示例。


在下文中一共展示了AtomVec::pack_comm方法的5个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的C++代码示例。

示例1: forward_comm_unpack_cuda

void CommCuda::forward_comm_unpack_cuda()
{
	static int count=0;
	static double kerneltime=0.0;
	static double copytime=0.0;
    timespec time1,time2,time3;
  int n;
  MPI_Request request;
  MPI_Status status;
  AtomVec *avec = atom->avec;
  double **x = atom->x;

  cuda->shared_data.domain.xy=domain->xy;
  cuda->shared_data.domain.xz=domain->xz;
  cuda->shared_data.domain.yz=domain->yz;
  cuda->shared_data.domain.prd[0]=domain->prd[0];
  cuda->shared_data.domain.prd[1]=domain->prd[1];
  cuda->shared_data.domain.prd[2]=domain->prd[2];
  cuda->shared_data.domain.triclinic=domain->triclinic;
  if(not comm_x_only && not avec->cudable) cuda->downloadAll();  //if not comm_x_only the communication routine of the atom_vec style class is used 

  // exchange data with another proc
  // if other proc is self, just copy
  // if comm_x_only set, exchange or copy directly to x, don't unpack

  for (int iswap = 0; iswap < nswap; iswap++) {
    if (sendproc[iswap] != me) 
    {   
      if (comm_x_only) 
      {   
 
        //Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],cuda->shared_data.comm.buf_recv[iswap],iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
        Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],buf_recv,iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
 	
      } 
      else if (ghost_velocity) 
      { 
        //Cuda_CommCuda_UnpackComm_Vel(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],(void*)&buf_recv[iswap*maxrecv]); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
      } 
      else 
      {
	    MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
		  recvproc[iswap],0,world,&request);
        
        if(avec->cudable)
          n = avec->pack_comm(sendnum[iswap],&iswap,
                           buf_send,pbc_flag[iswap],pbc[iswap]);
        else
	      n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
	
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
	    avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
      }

    }
    else  //sendproc == me
    {
      if (comm_x_only) 
      {     	
	    if (sendnum[iswap])
		{
		}
      } 
      else if (ghost_velocity) 
      {
      } 
      else 
      {
		n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
		avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
      }
    }
  }
  if(not comm_x_only && not avec->cudable) cuda->uploadAll(); 
}
开发者ID:naadtum，项目名称:lammps-sph，代码行数:78，代码来源:comm_cuda.cpp

示例2: forward_comm_transfer_cuda


//.........这里部分代码省略.........
        //printf("A: %i \n",size_forward_recv_now/1024*4);
        //MPI_Irecv(cuda->shared_data.comm.buf_recv[iswap],size_forward_recv_now,MPI_DOUBLE,
        //         recvproc[iswap],0,world,&request);
        MPI_Irecv(buf_recv,size_forward_recv_now,MPI_DOUBLE,
                 recvproc[iswap],0,world,&request);
		//printf("%p %p %i\n",buf_send, cuda->shared_data.comm.buf_send_dev[iswap], cuda->shared_data.comm.send_size[iswap]*sizeof(double));
        //memcpy(buf_send,cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap]*sizeof(double));
	//	CudaWrapper_SyncStream(1);
        //printf("B: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
		CudaWrapper_DownloadCudaDataAsync((void*) buf_send, cuda->shared_data.comm.buf_send_dev[iswap], cuda->shared_data.comm.send_size[iswap]*sizeof(double),2);
	    //MPI_Send(cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
clock_gettime(CLOCK_REALTIME,&time1);
        CudaWrapper_SyncStream(2);
        //printf("C: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
clock_gettime(CLOCK_REALTIME,&time2);
cuda->shared_data.cuda_timings.comm_forward_download+=
      time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
	    MPI_Send(buf_send,cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
        //printf("D: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
		CudaWrapper_UploadCudaDataAsync((void*) buf_recv,cuda->shared_data.comm.buf_recv_dev[iswap], size_forward_recv_now*sizeof(double),2);
clock_gettime(CLOCK_REALTIME,&time1);
        CudaWrapper_SyncStream(2);
        //printf("E: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
        //memcpy(cuda->shared_data.comm.buf_recv[iswap],buf_recv,size_forward_recv_now*sizeof(double));
 		//printf("RecvSize: %i SendSize: %i\n",size_forward_recv_now*sizeof(double),cuda->shared_data.comm.send_size[iswap]*sizeof(double));      
clock_gettime(CLOCK_REALTIME,&time3);
cuda->shared_data.cuda_timings.comm_forward_upload+=
      time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
      time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;
clock_gettime(CLOCK_REALTIME,&time3);
cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
      time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
      } 
      else if (ghost_velocity) 
      {
 /*       int size_forward_recv_now=0;
        
        if((sizeof(X_FLOAT)!=sizeof(double)) && size_forward_recv[iswap]) //some complicated way to safe some transfer size if single precision is used
          size_forward_recv_now=(size_forward_recv[iswap]+1)*sizeof(X_FLOAT)/sizeof(double);
        else
          size_forward_recv_now=size_forward_recv[iswap];
		
clock_gettime(CLOCK_REALTIME,&time1);
        
        MPI_Irecv(cuda->shared_data.comm.buf_recv[iswap],size_forward_recv_now,MPI_DOUBLE,
                 recvproc[iswap],0,world,&request);
		
clock_gettime(CLOCK_REALTIME,&time2);
		
	    MPI_Send(cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
       
clock_gettime(CLOCK_REALTIME,&time3);
cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
      time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
      time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;*/
 
       } 
      else 
      {
	    MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
		  recvproc[iswap],0,world,&request);
        
        if(avec->cudable)
          n = avec->pack_comm(sendnum[iswap],&iswap,
                           buf_send,pbc_flag[iswap],pbc[iswap]);
        else
	      n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
	
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
	    avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
      }

    }
    else  //sendproc == me
    {
      if (comm_x_only) 
      {     	
	    if (sendnum[iswap])
		{
		}
      } 
      else if (ghost_velocity) 
      {
      } 
      else 
      {
		n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
		avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
      }
    }
  }
  if(not comm_x_only && not avec->cudable) cuda->uploadAll(); 
}
开发者ID:naadtum，项目名称:lammps-sph，代码行数:101，代码来源:comm_cuda.cpp

示例3: forward_comm_cuda

void CommCuda::forward_comm_cuda()
{
  static int count=0;
  static double kerneltime=0.0;
  static double copytime=0.0;
  timespec time1,time2,time3;
  
  int n;
  MPI_Request request;
  MPI_Status status;
  AtomVec *avec = atom->avec;
  double **x = atom->x;

  cuda->shared_data.domain.xy=domain->xy;
  cuda->shared_data.domain.xz=domain->xz;
  cuda->shared_data.domain.yz=domain->yz;
  cuda->shared_data.domain.prd[0]=domain->prd[0];
  cuda->shared_data.domain.prd[1]=domain->prd[1];
  cuda->shared_data.domain.prd[2]=domain->prd[2];
  cuda->shared_data.domain.triclinic=domain->triclinic;
  if(not comm_x_only && not avec->cudable) 
  {
  	cuda->downloadAll(); 
    Comm::forward_comm();
    cuda->uploadAll();
    return;
  } 
  
  // exchange data with another proc
  // if other proc is self, just copy
  // if comm_x_only set, exchange or copy directly to x, don't unpack

  for (int iswap = 0; iswap < nswap; iswap++) {
    if (sendproc[iswap] != me) 
    {   
      if (comm_x_only) 
      {

        int size_forward_recv_now=0;
        
        if((sizeof(X_FLOAT)!=sizeof(double)) && size_forward_recv[iswap]) //some complicated way to safe some transfer size if single precision is used
          size_forward_recv_now=(size_forward_recv[iswap]+1)*sizeof(X_FLOAT)/sizeof(double);
        else
          size_forward_recv_now=size_forward_recv[iswap];
clock_gettime(CLOCK_REALTIME,&time1);
        
        MPI_Irecv(buf_recv,size_forward_recv_now,MPI_DOUBLE,
                 recvproc[iswap],0,world,&request);
        n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*) buf_send,pbc[iswap],pbc_flag[iswap]);
		
clock_gettime(CLOCK_REALTIME,&time2);
		
        if((sizeof(X_FLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
          n=(n+1)*sizeof(X_FLOAT)/sizeof(double);

		//printf("RecvSize: %i SendSize: %i\n",size_forward_recv_now,n);
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
       
clock_gettime(CLOCK_REALTIME,&time3);
cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
      time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
      time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;
 
        Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],(void*)buf_recv,iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
 	
      } 
      else if (ghost_velocity) 
      {
	    MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
		  recvproc[iswap],0,world,&request);
        
        if(avec->cudable)
          n = avec->pack_comm_vel(sendnum[iswap],&iswap,
                           buf_send,pbc_flag[iswap],pbc[iswap]);
        else
	      n = avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
				buf_send,pbc_flag[iswap],pbc[iswap]);
	    
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
	    avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_recv);
      } 
      else 
      {
	    MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
		  recvproc[iswap],0,world,&request);
        
        if(avec->cudable)
          n = avec->pack_comm(sendnum[iswap],&iswap,
                           buf_send,pbc_flag[iswap],pbc[iswap]);
        else
	      n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
	
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
	    avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
      }
//.........这里部分代码省略.........
开发者ID:naadtum，项目名称:lammps-sph，代码行数:101，代码来源:comm_cuda.cpp

示例4: forward_comm_pack_cuda

void CommCuda::forward_comm_pack_cuda()
{
	static int count=0;
	static double kerneltime=0.0;
	static double copytime=0.0;
    timespec time1,time2,time3;
  int n;  // initialize comm buffers & exchange memory

  MPI_Request request;
  MPI_Status status;
  AtomVec *avec = atom->avec;
  double **x = atom->x;

  cuda->shared_data.domain.xy=domain->xy;
  cuda->shared_data.domain.xz=domain->xz;
  cuda->shared_data.domain.yz=domain->yz;
  cuda->shared_data.domain.prd[0]=domain->prd[0];
  cuda->shared_data.domain.prd[1]=domain->prd[1];
  cuda->shared_data.domain.prd[2]=domain->prd[2];
  cuda->shared_data.domain.triclinic=domain->triclinic;
  if(not comm_x_only && not avec->cudable) cuda->downloadAll();  //if not comm_x_only the communication routine of the atom_vec style class is used 

  // exchange data with another proc
  // if other proc is self, just copy
  // if comm_x_only set, exchange or copy directly to x, don't unpack

  for (int iswap = 0; iswap < nswap; iswap++) {
    if (sendproc[iswap] != me) 
    {   
      if (comm_x_only) 
      {

		
clock_gettime(CLOCK_REALTIME,&time1);
        
      //  n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*) cuda->shared_data.comm.buf_send[iswap],pbc[iswap],pbc_flag[iswap]);
		  n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*)buf_send,pbc[iswap],pbc_flag[iswap]);
		
clock_gettime(CLOCK_REALTIME,&time2);
		
        if((sizeof(X_FLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
          n=(n+1)*sizeof(X_FLOAT)/sizeof(double);
		cuda->shared_data.comm.send_size[iswap]=n;
      } 
      else if (ghost_velocity) 
      {
clock_gettime(CLOCK_REALTIME,&time1);

       // n = Cuda_CommCuda_PackComm_Vel(&cuda->shared_data,sendnum[iswap],iswap,(void*) &buf_send[iswap*maxsend],pbc[iswap],pbc_flag[iswap]);
		
clock_gettime(CLOCK_REALTIME,&time2);
		
        if((sizeof(X_FLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
          n=(n+1)*sizeof(X_FLOAT)/sizeof(double);
		cuda->shared_data.comm.send_size[iswap]=n;
       } 
      else 
      {
	    MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
		  recvproc[iswap],0,world,&request);
        
        if(avec->cudable)
          n = avec->pack_comm(sendnum[iswap],&iswap,
                           cuda->shared_data.comm.buf_send[iswap],pbc_flag[iswap],pbc[iswap]);
        else
	      n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    cuda->shared_data.comm.buf_send[iswap],pbc_flag[iswap],pbc[iswap]);
	
	    MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
	    MPI_Wait(&request,&status);
	    avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
      }

    }
    else  //sendproc == me
    {
      if (comm_x_only) 
      {     	
	    if (sendnum[iswap])
		{
          n = Cuda_CommCuda_PackComm_Self(&cuda->shared_data,sendnum[iswap],iswap,firstrecv[iswap],pbc[iswap],pbc_flag[iswap]);
          if(n<0) error->all(FLERR," # CUDA ERRROR on PackComm_Self");
          if((sizeof(X_FLOAT)!=sizeof(double)) && n)
            n=(n+1)*sizeof(X_FLOAT)/sizeof(double);
		}
      } 
      else if (ghost_velocity) 
      {
		n = avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
				buf_send,pbc_flag[iswap],pbc[iswap]);
	    avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_send);
      } 
      else 
      {
		n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
			    buf_send,pbc_flag[iswap],pbc[iswap]);
		avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
      }
    }
  }
//.........这里部分代码省略.........
开发者ID:naadtum，项目名称:lammps-sph，代码行数:101，代码来源:comm_cuda.cpp

示例5: forward_comm

void CommBrick::forward_comm(int dummy)
{
  int n;
  MPI_Request request;
  AtomVec *avec = atom->avec;
  double **x = atom->x;
  double *buf;

  // exchange data with another proc
  // if other proc is self, just copy
  // if comm_x_only set, exchange or copy directly to x, don't unpack

  for (int iswap = 0; iswap < nswap; iswap++) {
    if (sendproc[iswap] != me) {
      if (comm_x_only) {
        if (size_forward_recv[iswap]) {
          if (size_forward_recv[iswap]) buf = x[firstrecv[iswap]];
          else buf = NULL;
          MPI_Irecv(buf,size_forward_recv[iswap],MPI_DOUBLE,
                    recvproc[iswap],0,world,&request);
        }
        n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
                            buf_send,pbc_flag[iswap],pbc[iswap]);
        if (n) MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
        if (size_forward_recv[iswap]) MPI_Wait(&request,MPI_STATUS_IGNORE);
      } else if (ghost_velocity) {
        if (size_forward_recv[iswap])
          MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
                    recvproc[iswap],0,world,&request);
        n = avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
                                buf_send,pbc_flag[iswap],pbc[iswap]);
        if (n) MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
        if (size_forward_recv[iswap]) MPI_Wait(&request,MPI_STATUS_IGNORE);
        avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_recv);
      } else {
        if (size_forward_recv[iswap])
          MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
                    recvproc[iswap],0,world,&request);
        n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
                            buf_send,pbc_flag[iswap],pbc[iswap]);
        if (n) MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
        if (size_forward_recv[iswap]) MPI_Wait(&request,MPI_STATUS_IGNORE);
        avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
      }

    } else {
      if (comm_x_only) {
        if (sendnum[iswap])
          avec->pack_comm(sendnum[iswap],sendlist[iswap],
                          x[firstrecv[iswap]],pbc_flag[iswap],pbc[iswap]);
      } else if (ghost_velocity) {
        avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
                            buf_send,pbc_flag[iswap],pbc[iswap]);
        avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_send);
      } else {
        avec->pack_comm(sendnum[iswap],sendlist[iswap],
                        buf_send,pbc_flag[iswap],pbc[iswap]);
        avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
      }
    }
  }
}
开发者ID:vanceeasleaf，项目名称:lammps，代码行数:62，代码来源:comm_brick.cpp


注：本文中的AtomVec::pack_comm方法示例由纯净天空整理自Github/MSDocs等开源代码及文档管理平台，相关代码片段筛选自各路编程大神贡献的开源项目，源码版权归原作者所有，传播和使用请参考对应项目的License；未经允许，请勿转载。