通过MPI传递Armadillo C++矩阵

我需要通过MPI传递Armadillo C++矩阵库定义的矩阵或复杂矩阵类型。对此有什么好方法？我想尝试：

1. 将矩阵写入某种数组，然后发送
   其中的行/列，以及对MPI_send/recv 两侧的阵列进行分解/重建的方法

2. 使用类似MPI_BYTE类型的东西？

感谢

更新

因此，我试图通过在一个节点上发送和接收来实现另一种方案。

translate.cpp

    #include <mpi.h>
    #include <armadillo>
    #include <vector> 
    #include <cstdlib>
    using namespace std; 
    using namespace arma; 
    using std::vector; 
    class ArmadilloMPI
    {
        public:
            ArmadilloMPI(int nRows, int nCols)
            {
                this->nRows = nRows;
                this->nCols = nCols; 
                realArray = (double **)malloc(nCols * nRows * sizeof(double*));
                imArray = (double **)malloc(nCols * nRows * sizeof(double*));
            }
            ~ArmadilloMPI()
            {
                free(realArray[0]);
                free(realArray);
                free(imArray[0]);
                free(imArray);
            }
            double **realArray; 
            double **imArray; 
            int nCols; 
            int nRows; 
            cx_mat matConstructRecv(int src, int tag)
            {
                cx_mat A(nRows, nCols); 
                MPI_Recv(&(imArray[0][0]),  nRows * nCols, MPI_DOUBLE, src, tag, MPI_COMM_WORLD,0);
                MPI_Recv(&(realArray[0][0]),nRows * nCols, MPI_DOUBLE, src, tag, MPI_COMM_WORLD,0);
                for(int  i = 0; i < nRows; ++i )
                {
                    for(int j = 0; i < nCols; ++j)
                    {
                        real(A(i,j)) = *realArray[i * nRows + j]; 
                        imag(A(i,j)) = *imArray[i * nRows + j];
                    }
                }
                return A; 
            }
            void matDestroySend(cx_mat &A, int dest, int tag)
            {
                for(int  i = 0; i < nRows; ++i )
                {
                    for(int j = 0; i < nCols; ++j)
                    {
                        realArray[i * nRows + j]  = &real(A(i,j)); 
                        imArray[i * nRows + j] = &imag(A(i,j)); 
                    }
                }
                MPI_Send(&(realArray[0][0]), nRows * nCols, MPI_DOUBLE, dest, tag, MPI_COMM_WORLD);
                MPI_Send(&(imArray[0][0]), nRows * nCols, MPI_DOUBLE, dest, tag, MPI_COMM_WORLD);
            }
    };
    int main(int argc, char** argv)
    {
        MPI::Init(argc, argv);
        int size = MPI::COMM_WORLD.Get_size();
        int rank = MPI::COMM_WORLD.Get_rank();
        cout << "test"<<endl; 
        vector<cx_mat> world; 
        for(int i = 0; i < size; ++i )
        {
            world.push_back(randu<cx_mat>(4,4));
        }
        cx_mat A;
        A = randu<cx_mat>(4,4);
        ArmadilloMPI* armaMPI = new ArmadilloMPI(4,4); 
        if(rank==0)
        {
            for(int i = 1; i < size; i++)
            {   
                cout << "A is now " << A << endl; 
                A += armaMPI->matConstructRecv(i, 0);
            }
        }
        else
        {
            armaMPI->matDestroySend(world[rank], 1, 0);
        }
        cout << A << endl; 
        delete armaMPI;
        MPI::Finalize();
    }

但我们有一个seg错误。

*** Process received signal *** 
Signal: Segmentation fault: 11 (11) 
Signal     code:     (0) 
Failing at address: 0x0 translate(1032,0x7fff747ad310) malloc: ***   error for object     0x41434d5f49504d4f: pointer being freed was not allocated

想法？