MPI_Comm_Split和MPI_Bcast问题

问：

几天前我提出了一个问题，多亏了一个好心人，我可以解决我的问题。现在，我尝试使用 MPI 在 C 语言中实现一个更复杂的程序，使用 MPI_Comm_Split 为某些进程组创建通信器。我有一个全局排名为 0 的进程，它不在任何组中，因为我想把它当作一个中央服务器，实际上任务是初始化一个由 4 个浮动 Noumbers 组成的数组并将其发送到属于组 0 的进程。这个组 0 由 3 个进程（全局排名为 1、2 和 3 的进程）组成，我将它们视为中间服务器，因此它们从中央服务器接收数组并将其转发给其他组。我还有其他 3 个包含其余进程的组（这 3 个组中的每一个都有相同的进程）。

代码如下：

#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>

int main(int argc, char *argv[]) {
    MPI_Init(&argc, &argv);

    int rank, size;
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
    MPI_Comm_size(MPI_COMM_WORLD, &size);

    if (size < 7) {
        if (rank == 0) {
            printf("This program requires at least 7 processes.\n");
        }
    } else {
        int group_number;

        if (rank == 0) {
            group_number = -1; // Process with rank 0 does not belong to any group
        } else if (rank >= 1 && rank <= 3) {
            group_number = 0; // New group formed by processes with rank 1, 2, and 3
        } else {
            group_number = (rank - 4) % 3 + 1; // Subdivision of the other processes into 3 groups with numbers 1, 2, and 3
        }

        MPI_Comm comm_group;
        MPI_Comm_split(MPI_COMM_WORLD, group_number, rank, &comm_group);

        int group_rank;
        int group_size;
        MPI_Comm_rank(comm_group, &group_rank);
        MPI_Comm_size(comm_group, &group_size);

        if (rank == 0) {
            printf("Process %d does not belong to any group because it is the central server.\n", rank);
        } else if(group_number != 0) {
            printf("Process %d in group %d. Rank in the group: %d\n", rank, group_number, group_rank);
        }else{
            printf("Process %d in intermediate server group %d. Rank in the group: %d\n", rank, group_number, group_rank);
        }

        float global_message[4];
        float intermediary_message[4];

        if (rank == 0)
        {
           // Process with global rank 0 initializes the message
            global_message[0] = 10.0;
            global_message[1] = 11.0;
            global_message[2] = 12.0;
            global_message[3] = 13.0;

            printf("Message to send: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);
        }

        // Send message only to processes in group 0
        if (group_number == 0 || rank == 0) {
            MPI_Bcast(&global_message, 4, MPI_FLOAT, 0,MPI_COMM_WORLD);
        }

        if (group_number == 0 ) {
            // Print the message received from process 0
            printf("Received message from central server: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);
            
        } 

        if (rank == 1) {
            
            printf("intermediate message 1: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);

            for (int i = 0; i < 4; i++) {
                intermediary_message[i] = global_message[i];
            }
        }

        if (rank == 2) {
            
            printf("intermediate message 2: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);

            for (int i = 0; i < 4; i++) {
                intermediary_message[i] = global_message[i];
            }
        }

        if (rank == 3) {
            
            printf("intermediate message 3: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);

            for (int i = 0; i < 4; i++) {
                intermediary_message[i] = global_message[i];
            }
        }

        if (group_number == 1 || rank == 1)
        {
            MPI_Bcast(&intermediary_message, 4, MPI_FLOAT, 0, comm_group);
        }

         if (group_number == 2 || rank == 2)
        {
            MPI_Bcast(&intermediary_message, 4, MPI_FLOAT, 1, comm_group);
        }

         if (group_number == 3 || rank == 3)
        {
            MPI_Bcast(&intermediary_message, 4, MPI_FLOAT, 2, comm_group);
        }

        if (group_number == 1 || group_number == 2 || group_number == 3 )
        {
            // Print the message received from each process in group 0
            printf("Received the message from the intermediate server: %f %f %f %f\n\n", global_message[0], global_message[1], global_message[2], global_message[3]);
        }

        int group_members[group_size];
        MPI_Gather(&rank, 1, MPI_INT, group_members, 1, MPI_INT, 0, comm_group);

        if (group_number != -1) {
            if (group_rank == 0) {
                printf("Group %d includes the following members: ", group_number);
                for (int i = 0; i < group_size; i++) {
                    printf("%d ", group_members[i]);
                }
                printf("\n\n");
            }
        }

        if (group_number >= 0 && group_number != 0) {
            MPI_Comm_free(&comm_group);
        }
    }

    MPI_Finalize();
    return 0;
}

这是输出：

mpiexec -np 13 third_example

Process 8 in group 2. Rank in the group: 1
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Process 0 does not belong to any group because it is the central server.
Message to send: 10.000000 11.000000 12.000000 13.000000

Process 1 in intermediate server group 0. Rank in the group: 0
Received message from central server: 10.000000 11.000000 12.000000 13.000000

intermediate message 1: 10.000000 11.000000 12.000000 13.000000

Group 0 includes the following members: 1 2 3

Process 2 in intermediate server group 0. Rank in the group: 1
Received message from central server: 10.000000 11.000000 12.000000 13.000000

intermediate message 2: 10.000000 11.000000 12.000000 13.000000

Process 5 in group 2. Rank in the group: 0
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Group 2 includes the following members: 5 8 11

Process 12 in group 3. Rank in the group: 2
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Process 7 in group 1. Rank in the group: 1
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Process 9 in group 3. Rank in the group: 1
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Process 3 in intermediate server group 0. Rank in the group: 2
Received message from central server: 10.000000 11.000000 12.000000 13.000000

intermediate message 3: 10.000000 11.000000 12.000000 13.000000

Process 6 in group 3. Rank in the group: 0
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Group 3 includes the following members: 6 9 12

Process 11 in group 2. Rank in the group: 2
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Process 4 in group 1. Rank in the group: 0
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

Group 1 includes the following members: 4 7 10

Process 10 in group 1. Rank in the group: 2
Received the message from the intermediate server: -0.000001 0.000000 -0.000001 0.000000

如您所见，组 0 的数组转发有效，但对其他组无效。我称之为“中间服务器”的进程和组 1、2、3 的通信器是相同的，但它不起作用。我认为问题在于用于将阵列从中央服务器转发到中间服务器组的通信器是全局通信器（MPI_COMM_WORLD），但是当我尝试使用组通信器更改它时，代码不会执行。 请有人能帮我吗？