libdune-common-dev 2.2.1-2 / usr / include / dune / common / mpicollectivecommunication.hh

// -*- tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
// vi: set et ts=4 sw=2 sts=2:
#ifndef DUNE_MPICOLLECTIVECOMMUNICATION_HH
#define DUNE_MPICOLLECTIVECOMMUNICATION_HH

/*!
  \file
  \ingroup ParallelCommunication
 */

#include<iostream>
#include<complex>
#include<algorithm>
#include<functional>

#include"exceptions.hh"
#include"collectivecommunication.hh"
#include"binaryfunctions.hh"
#include"shared_ptr.hh"
#include"mpitraits.hh"

#if HAVE_MPI
// MPI header
#include<mpi.h>

namespace Dune
{

  //=======================================================
  // use singleton pattern and template specialization to 
  // generate MPI operations
  //=======================================================

  template<typename Type, typename BinaryFunction>
  class Generic_MPI_Op
  {
    
  public:
	static MPI_Op get ()
	{
	  if (!op)
		{
		  op = shared_ptr<MPI_Op>(new MPI_Op);
		  MPI_Op_create((void (*)(void*, void*, int*, MPI_Datatype*))&operation,true,op.get());
		}
	  return *op;
	}
  private:
	static void operation (Type *in, Type *inout, int *len, MPI_Datatype *dptr)
	{
	  BinaryFunction func;
	  
	  for (int i=0; i< *len; ++i, ++in, ++inout){
	    Type temp;
	    temp = func(*in, *inout);
	    *inout = temp;
	  }
	}
	Generic_MPI_Op () {}
	Generic_MPI_Op (const Generic_MPI_Op& ) {}
	static shared_ptr<MPI_Op> op;
  };

  
  template<typename Type, typename BinaryFunction>
  shared_ptr<MPI_Op> Generic_MPI_Op<Type,BinaryFunction>::op = shared_ptr<MPI_Op>(static_cast<MPI_Op*>(0));
  
#define ComposeMPIOp(type,func,op) \
  template<> \
  class Generic_MPI_Op<type, func<type> >{ \
  public:\
    static MPI_Op get(){ \
      return op; \
    } \
  private:\
    Generic_MPI_Op () {}\
    Generic_MPI_Op (const Generic_MPI_Op& ) {}\
  }

    
  ComposeMPIOp(char, std::plus, MPI_SUM);
  ComposeMPIOp(unsigned char, std::plus, MPI_SUM);
  ComposeMPIOp(short, std::plus, MPI_SUM);
  ComposeMPIOp(unsigned short, std::plus, MPI_SUM);
  ComposeMPIOp(int, std::plus, MPI_SUM);
  ComposeMPIOp(unsigned int, std::plus, MPI_SUM);
  ComposeMPIOp(long, std::plus, MPI_SUM);
  ComposeMPIOp(unsigned long, std::plus, MPI_SUM);
  ComposeMPIOp(float, std::plus, MPI_SUM);
  ComposeMPIOp(double, std::plus, MPI_SUM);
  ComposeMPIOp(long double, std::plus, MPI_SUM);

  ComposeMPIOp(char, std::multiplies, MPI_PROD);
  ComposeMPIOp(unsigned char, std::multiplies, MPI_PROD);
  ComposeMPIOp(short, std::multiplies, MPI_PROD);
  ComposeMPIOp(unsigned short, std::multiplies, MPI_PROD);
  ComposeMPIOp(int, std::multiplies, MPI_PROD);
  ComposeMPIOp(unsigned int, std::multiplies, MPI_PROD);
  ComposeMPIOp(long, std::multiplies, MPI_PROD);
  ComposeMPIOp(unsigned long, std::multiplies, MPI_PROD);
  ComposeMPIOp(float, std::multiplies, MPI_PROD);
  ComposeMPIOp(double, std::multiplies, MPI_PROD);
  ComposeMPIOp(long double, std::multiplies, MPI_PROD);

  ComposeMPIOp(char, Min, MPI_MIN);
  ComposeMPIOp(unsigned char, Min, MPI_MIN);
  ComposeMPIOp(short, Min, MPI_MIN);
  ComposeMPIOp(unsigned short, Min, MPI_MIN);
  ComposeMPIOp(int, Min, MPI_MIN);
  ComposeMPIOp(unsigned int, Min, MPI_MIN);
  ComposeMPIOp(long, Min, MPI_MIN);
  ComposeMPIOp(unsigned long, Min, MPI_MIN);
  ComposeMPIOp(float, Min, MPI_MIN);
  ComposeMPIOp(double, Min, MPI_MIN);
  ComposeMPIOp(long double, Min, MPI_MIN);

  ComposeMPIOp(char, Max, MPI_MAX);
  ComposeMPIOp(unsigned char, Max, MPI_MAX);
  ComposeMPIOp(short, Max, MPI_MAX);
  ComposeMPIOp(unsigned short, Max, MPI_MAX);
  ComposeMPIOp(int, Max, MPI_MAX);
  ComposeMPIOp(unsigned int, Max, MPI_MAX);
  ComposeMPIOp(long, Max, MPI_MAX);
  ComposeMPIOp(unsigned long, Max, MPI_MAX);
  ComposeMPIOp(float, Max, MPI_MAX);
  ComposeMPIOp(double, Max, MPI_MAX);
  ComposeMPIOp(long double, Max, MPI_MAX);

#undef ComposeMPIOp


  //=======================================================
  // use singleton pattern and template specialization to 
  // generate MPI operations
  //=======================================================

  /*! \brief Specialization of CollectiveCommunication for MPI
	\ingroup ParallelCommunication
  */
  template<>
  class CollectiveCommunication<MPI_Comm>
  {
  public:
	//! Instantiation using a MPI communicator
	CollectiveCommunication (const MPI_Comm& c)
	  : communicator(c)
	{
	  if(communicator!=MPI_COMM_NULL){
	    MPI_Comm_rank(communicator,&me);
	    MPI_Comm_size(communicator,&procs);
	  }else{
	    procs=0;
	    me=-1;
	  }
	}

	//! @copydoc CollectiveCommunication::rank
	int rank () const
	{
	  return me;
	}

	//! @copydoc CollectiveCommunication::size
	int size () const
	{
	  return procs;
	}

	//! @copydoc CollectiveCommunication::sum
	template<typename T>
	T sum (T& in) const // MPI does not know about const :-(
	{
	  T out;
	  allreduce<std::plus<T> >(&in,&out,1);
	  return out;
	}

	//! @copydoc CollectiveCommunication::sum
	template<typename T>
	int sum (T* inout, int len) const
	{
	  return allreduce<std::plus<T> >(inout,len);
	}

	//! @copydoc CollectiveCommunication::prod
	template<typename T>
	T prod (T& in) const // MPI does not know about const :-(
	{
	  T out;
	  allreduce<std::multiplies<T> >(&in,&out,1);
	  return out;
	}

	//! @copydoc CollectiveCommunication::prod
	template<typename T>
	int prod (T* inout, int len) const
	{
	  return allreduce<std::plus<T> >(inout,len);
	}

	//! @copydoc CollectiveCommunication::min
	template<typename T>
	T min (T& in) const // MPI does not know about const :-(
	{
	  T out;
	  allreduce<Min<T> >(&in,&out,1);
	  return out;
	}

	//! @copydoc CollectiveCommunication::min
	template<typename T>
	int min (T* inout, int len) const
	{
	  return allreduce<Min<T> >(inout,len);
	}

    
	//! @copydoc CollectiveCommunication::max
	template<typename T>
	T max (T& in) const // MPI does not know about const :-(
	{
	  T out;
	  allreduce<Max<T> >(&in,&out,1);
	  return out;
	}

	//! @copydoc CollectiveCommunication::max
	template<typename T>
	int max (T* inout, int len) const
	{
	  return allreduce<Max<T> >(inout,len);
	}

	//! @copydoc CollectiveCommunication::barrier
	int barrier () const
	{
	  return MPI_Barrier(communicator);
	}

	//! @copydoc CollectiveCommunication::broadcast
	template<typename T>
	int broadcast (T* inout, int len, int root) const
	{
	  return MPI_Bcast(inout,len,MPITraits<T>::getType(),root,communicator);
	}
	
	//! @copydoc CollectiveCommunication::gather()
	template<typename T>
	int gather (T* in, T* out, int len, int root) const // note out must have space for P*len elements
	{
	  return MPI_Gather(in,len,MPITraits<T>::getType(),
						out,len,MPITraits<T>::getType(),
						root,communicator);
	}

    //! @copydoc CollectiveCommunication::scatter()
    template<typename T>
    int scatter (T* send, T* recv, int len, int root) const // note out must have space for P*len elements
    {
      return MPI_Scatter(send,len,MPITraits<T>::getType(),
                         recv,len,MPITraits<T>::getType(),
                         root,communicator);
    }

	operator MPI_Comm () const
	{
	  return communicator;
	}

    //! @copydoc CollectiveCommunication::allgather()
    template<typename T, typename T1>
    int allgather(T* sbuf, int count, T1* rbuf) const
    {
      return MPI_Allgather(sbuf, count, MPITraits<T>::getType(),
                           rbuf, count, MPITraits<T1>::getType(),
                           communicator);
    }

    //! @copydoc CollectiveCommunication::allreduce(Type* inout,int len) const
    template<typename BinaryFunction, typename Type>
    int allreduce(Type* inout, int len) const
    {
      Type* out = new Type[len];
      int ret = allreduce<BinaryFunction>(inout,out,len);
      std::copy(out, out+len, inout);
      delete[] out;
      return ret;
    }
    
    //! @copydoc CollectiveCommunication::allreduce(Type* in,Type* out,int len) const
    template<typename BinaryFunction, typename Type>
    int allreduce(Type* in, Type* out, int len) const
    {
      return MPI_Allreduce(in, out, len, MPITraits<Type>::getType(),
		    (Generic_MPI_Op<Type, BinaryFunction>::get()),communicator);
    }
    
  private:
	MPI_Comm communicator;
	int me;
	int procs;
  };
} // namespace dune

#endif
#endif