MPI_Finalize 中的总线错误

Question

我正在为并行计算类编写 MPI 程序。我已经让代码工作了，它输出了正确的结果，但是当我尝试使用多个进程调用 MPI_Finalize 时，我得到了一个 Buss 错误。我通过 Eclipse 中的 PTP 环境在 OS X 上运行它。错误如下：

[Fruity:49034] *** Process received signal ***
[Fruity:49034] Signal: Bus error (10)
[Fruity:49034] Signal code:  (2)
[Fruity:49034] Failing at address: 0x100336d7e
[Fruity:49034] [ 0] 2   libSystem.B.dylib                   0x00007fff865cc1ba _sigtramp + 26
[Fruity:49034] [ 1] 3   ???                                 0x0000000000000000 0x0 + 0
[Fruity:49034] [ 2] 4   libSystem.B.dylib                   0x00007fff86570c27 tiny_malloc_from_free_list + 1196
[Fruity:49034] [ 3] 5   libSystem.B.dylib                   0x00007fff8656fabd szone_malloc_should_clear + 242
[Fruity:49034] [ 4] 6   libopen-pal.0.dylib                 0x0000000100187b9f opal_memory_base_open + 527
[Fruity:49034] [ 5] 7   libSystem.B.dylib                   0x00007fff8656f98a malloc_zone_malloc + 82
[Fruity:49034] [ 6] 8   libSystem.B.dylib                   0x00007fff8656dc88 malloc + 44
[Fruity:49034] [ 7] 9   libSystem.B.dylib                   0x00007fff8657846d asprintf + 157
[Fruity:49034] [ 8] 10  libopen-rte.0.dylib                 0x000000010013aebc orte_schema_base_get_job_segment_name + 108
[Fruity:49034] [ 9] 11  libopen-rte.0.dylib                 0x000000010013d899 orte_smr_base_set_proc_state + 57
[Fruity:49034] [10] 12  libmpi.0.dylib                      0x0000000100063758 ompi_mpi_finalize + 312
[Fruity:49034] [11] 13  Assignment31                        0x0000000100002642 main + 491
[Fruity:49034] [12] 14  Assignment31                        0x0000000100001688 start + 52
[Fruity:49034] *** End of error message ***
mpirun noticed that job rank 0 with PID 49033 on node Fruity.local exited on signal 15 (Terminated).
1 additional process aborted (not shown)

这是我的代码的主要功能。我敢肯定这里有一些不好的 C++ 实践（我已经多年没有使用它并且它是自学的），但它确实输出了正确的值。如果我需要发布文件的其余部分，我可以这样做。如果有明显的问题，我只是不想把这个问题变成一个大问题。

int main(int argc, char* argv[]){
    /* start up MPI */
    MPI_Init(&argc, &argv);

    /* find out process rank */
    MPI_Comm_rank(MPI_COMM_WORLD, &myRank);

    /* find out number of processes */
    MPI_Comm_size(MPI_COMM_WORLD, &numProcs);


    /* find which nodes this processor is responsible for */
    findStartAndEndPositions();

    /*Intitialize the array to its starting values. */
    initializeArray();

    /*Find the elements that are dependent on outside processors */
    findDependentElements();

    MPI_Barrier(MPI_COMM_WORLD);
    if(myRank == 0){
        startTime = MPI_Wtime();
        printArray();
    }

    int iter;
    for(iter = 0; iter < NUM_ITERATIONS; iter++){
        doCommunication();
        MPI_Barrier(MPI_COMM_WORLD);
        doIteration();
    }


    double check = computeCheck();
    double receive = 0;

    if(myRank == 0){
        MPI_Reduce(&check, &receive, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
        std::cout << "The total time was: " << MPI_Wtime() - startTime << " \n";
        std::cout << "The checksum was: " << receive << " \n";
        printArray();
    }

    else{
        MPI_Reduce(&check, &receive, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
    }

    /* shut down MPI */
    MPI_Barrier(MPI_COMM_WORLD);
    MPI_Finalize();
    return 0;
}

编辑：我已将问题范围缩小到我的 doIteration 函数中的某个位置。我只有在调用该函数时才会收到错误消息，并且只有在我运行多个进程时才会出现错误。这是我的 doIteration 函数。它应该用自身及其四个邻居的最大值替换不在矩阵边缘的矩阵的每个值。值应该在整个更新完成后更新（因此使用数组 temp）。

void doIteration(){
    int pos;
    double* temp = new double[end - start + 1];
    for(pos = start; pos <= end; pos++){
        int i, row, col;
        double max;

        convertToRowCol(pos, &row, &col);

        if(isEdgeNode(row, col))
            continue;

        int dependents[4];
        getDependentsOfPosition(pos, dependents);
        max = a[row][col];

        for(i = 0; i < 4; i++){
            if(isInvalidPos(dependents[i]))
                continue;

            int dRow, dCol;
            convertToRowCol(dependents[i], &dRow, &dCol);
            max = std::max(max, a[dRow][dCol]);
        }

        temp[pos] = max;
    }

    for(pos = start; pos <= end; pos++){
        int row, col;
        convertToRowCol(pos, &row, &col);
        if(! isEdgeNode(row, col))
            a[row][col] = temp[pos];
    }

    delete [] temp;
}

Answer 1

不知道是不是这个原因，但是MPI_Reduce一般都是一行，没必要写两行。试试这个看看是否有帮助。

MPI_Reduce(&check, &receive, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
if(myRank == 0){
     std::cout << "The total time was: " << MPI_Wtime() - startTime << " \n";
     std::cout << "The checksum was: " << receive << " \n";
     printArray();
}