使用 MPI_Bcast 用于

Question

我想在为 MPI 执行编写的代码中执行以下操作：

每个子进程都会创建一个包含矩阵
只有属于第一个子过程的矩阵已填充
矩阵是交换给所有子流程，这样所有子流程现在又一次拥有相同的矩阵，被第一个进程修改

因此我尝试在我的代码中使用 MPI_Bcast：

#include <iostream>
#include <mpi.h>
#include <stdio.h>
#include <armadillo>

class arma_matrix_container{
public:
    arma::mat test_matrix;

    arma_matrix_container(const int size, const bool first_matrix)
    {
        if(first_matrix)
            test_matrix = arma::mat(size, size, arma::fill::ones) * 1000;
        else
            test_matrix = arma::mat(size, size, arma::fill::zeros);
    }

    void update_matrix(const arma::mat &update_matrix)
    {
        this->test_matrix = update_matrix;
    }

    double get_first_matrix_value(void)
    {
        double first_element = this->test_matrix(0, 0);
        return first_element;
    }
};


int main(int argc, char* argv[])
{
    MPI_Init(&argc, &argv);
    const int matrix_size = 4;
    // Get the number of processes
    int world_size;
    MPI_Comm_size(MPI_COMM_WORLD, &world_size);

    // Get the rank of the process
    int world_rank;
    MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);

    arma_matrix_container local_matrix(matrix_size, (world_rank == 0)?true:false);
    double *matrix_pointer = local_matrix.test_matrix.memptr();
    double first_element = local_matrix.get_first_matrix_value();
    // Get the name of the processor
    char processor_name[MPI_MAX_PROCESSOR_NAME];
    int name_len;
    MPI_Get_processor_name(processor_name, &name_len);

    // Print off a hello world message
    printf("Hello world from processor %s, rank %d"
           " out of %d processors,\nthe first element of the matrix is %f\n",
           processor_name, world_rank, world_size, first_element);

    if(world_rank == 0)
        std::cout << "\nBroadcasting data: \n";
    MPI_Barrier(MPI_COMM_WORLD);
    MPI_Bcast(matrix_pointer, matrix_size * matrix_size * sizeof(double), MPI_DOUBLE, 0, MPI_COMM_WORLD);
    if(world_rank == 0)
        std::cout << "Data was broadcasted\n\n";

    printf("Hello world from processor %s, rank %d"
           " out of %d processors,\nthe first element of the matrix is %f\n",
           processor_name, world_rank, world_size, first_element);

    // Finalize the MPI environment.
    MPI_Finalize();
}

现在，如果我删除 MPI_Bcast-命令，我会得到（如预期的那样）：

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000

Broadcasting data: 
Hello world from processor MPI-PC, rank 3 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 1 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 2 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 3 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 1 out of 4 processors,
the first element of the matrix is 0.000000
Data was broadcasted

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000
Hello world from processor MPI-PC, rank 2 out of 4 processors,
the first element of the matrix is 0.000000

包括 MPI_Bcast-命令，我得到一个段错误和以下输出：

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000

Broadcasting data: 
Hello world from processor MPI-PC, rank 1 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 2 out of 4 processors,
the first element of the matrix is 0.000000
Hello world from processor MPI-PC, rank 3 out of 4 processors,
the first element of the matrix is 0.000000
Data was broadcasted

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000
Data was broadcasted

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000
Data was broadcasted

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000
Data was broadcasted

Hello world from processor MPI-PC, rank 0 out of 4 processors,
the first element of the matrix is 1000.000000
-------------------------------------------------------
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
-------------------------------------------------------
--------------------------------------------------------------------------
mpirun noticed that process rank 3 with PID 0 on node MPI-PC exited on signal 11 (Segmentation fault).

我哪里忘记正确初始化数据了？

Answer 1

MPI_Bcast 的第二个参数 (int count) 不应包括数据元素大小。这是从第三个参数 (MPI_Datatype datatype) 获得的。

所以你应该这样称呼它：

MPI_Bcast(matrix_pointer, matrix_size * matrix_size, MPI_DOUBLE, 0, MPI_COMM_WORLD);

使用 MPI_Bcast 用于

Using MPI_Bcast for

c++

mpi