Question

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

int main()
{
        int a_r = 0, a_c = 0, v_s = 0, i = 0, rank = 0, size = 0;
        int local_row = 0, partial_sum = 0, sum = 0, j = 0;
        int my_first_ele = 0, my_last_ele = 0;
        int a[10][10], v[10], partial_mul[10] = {0}, mul[10] = {0};

        MPI_Init(NULL, NULL);
        MPI_Comm_rank(MPI_COMM_WORLD, &rank);
        MPI_Comm_size(MPI_COMM_WORLD, &size);

        if(rank == 0)
        {
                printf("Enter the row of array A: ");
                scanf("%d", &a_r);
                printf("Enter the column of array A: ");
                scanf("%d", &a_c);
                printf("Enter the array A: ");

                for(i = 0; i < a_r; i++)
                {
                        for(j = 0; j < a_c; j++)
                                scanf("%d", &a[i][j]);
                }

                printf("Enter the size of vector array: ");
                scanf("%d", &v_s);
                printf("Enter the vector array: ");
                for(i = 0; i < v_s; i++)
                {
                        scanf("%d", &v[i]);
                }

                MPI_Bcast(&a_r, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(&a_c, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(&v_s, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(a, a_r*a_c, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(v, v_s, MPI_INT, 0, MPI_COMM_WORLD);

                local_row = a_r / size;
                my_first_ele = rank * local_row;
                my_last_ele = (rank+1) * local_row;

                if(a_c == v_s)
                {      
                        for(i = my_first_ele; i < my_last_ele; i++)
                        {
                                for(j = 0; j < a_c; j++)
                                {
                                        partial_mul[i] = partial_mul[i] + (a[i][j]*v[j]);
                                }
                        }
                        printf("\nPartial multiplication in Rank 0: \n");
                        for(i = my_first_ele; i < my_last_ele; i++)
                                printf("%d \n", partial_mul[i]);

                        MPI_Gather(partial_mul, local_row, MPI_INT, mul, local_row, MPI_INT, 0, MPI_COMM_WORLD);

                        printf("\n \nGlobal Multiplication: \n");
                        for(i = 0; i < a_r; i++)
                        {
                                printf("%d \n", mul[i]);
                        }
                }
                else
                        printf("\nCan't multiply. \n");
        }

        else
        {
                MPI_Bcast(&a_r, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(&a_c, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(&v_s, 1, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(a, a_r*a_c, MPI_INT, 0, MPI_COMM_WORLD);
                MPI_Bcast(v, v_s, MPI_INT, 0, MPI_COMM_WORLD);

                local_row = a_r / size;
                my_first_ele = rank * local_row;
                my_last_ele = (rank+1) * local_row;
                if(a_c == v_s)
                {     
                        for(i = my_first_ele; i < my_last_ele; i++)
                        {
                                for(j = 0; j < a_c; j++)
                                {
                                        partial_mul[i] = partial_mul[i] + (a[i][j]*v[j]);
                                }
                        }
                        printf("\nPartial multiplication in Rank %d: \n", rank);
                        for(i = my_first_ele; i < my_last_ele; i++)
                                printf("%d \n", partial_mul[i]);

                        MPI_Gather(partial_mul, local_row, MPI_INT, mul, local_row, MPI_INT, 0, MPI_COMM_WORLD);

                }
                else
                        printf("\nCan't multiply. \n");
        }
        MPI_FINALIZE();
}

我上面的代码有问题。我的部分乘法值是正确的。但是在我的整体乘法中，我只能收集等级0的元素，其余的值被打印为0.有什么问题可以解释。

Answer 1

查看您的数据布局我认为您误解了MPI中的数据结构：所有数据在每个等级中保持独立，没有任何共享或分发。您的向量2 4 6 8 10 custom 8 9 10 3 8 13 18 custom 1 2 3 4 5在每个等级上是独立的，每个等级都包含完整的10个元素。假设partial_sum，size=2和零初始化，计算后内容将如下所示：

排名0：a_r=10
排名1：{x0,x1,x2,x3,x4,0,0,0,0,0}

其中x是正确的计算值。然后，Gather将收集每个排名中的第一个{0,0,0,0,0,x5,x6,x7,x8,x9}元素，结果为local_row=5。

你可以通过添加正确的偏移来解决这个问题：

{x0,x1,x2,x3,x4,0,0,0,0,0}

但请不要这样做。相反，您应该重新考虑数据结构以真正分配数据，为矢量/数组的每个部分保留正确的大小。要将部分数据发送到每个排名，请使用MPI_Gather(&partial_mul[my_first_ele], local_row, MPI_INT, mul, local_row, MPI_INT, 0, MPI_COMM_WORLD);（与MPI_Scatter相反）。最困难的是使矩阵正确。 this excellent answer详细解释了这一点。

MPI矢量乘法

1 个答案: