您好,我是使用 MPI 编程的新手。我正在尝试将两个矩阵相乘(一个 NxN 矩阵(A)和一个 Nx1(B)矩阵)来得到一个结果 C 矩阵(Nx1)。每个进程都应该计算矩阵 C 中的一行(元素),但是只有进程 0(我的主进程)计算正确,因为它似乎没有等待其他进程完成计算。我也不确定非主进程是否正确地发回了结果(或者他们是否甚至需要?)。这是我的代码:
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include "mpi.h"
#define PRINT_VECS 1
#define MAX_RAND 100
#define MASTER 0
#define COLUMNS_B 1
#define N 4
void init_vec(int *vec, int len);
void print_vec(const char *label, int *vec, int len);
void init_vec(int *vec, int len)
{
int i;
for (i = 0; i < len; i++)
{
vec[i] = rand() % MAX_RAND;
}
}
void print_vec(const char *label, int *vec, int len)
{
#if PRINT_VECS
printf("%s", label);
int i;
for (i = 0; i < len; i++)
{
printf("%d ", vec[i]);
}
printf("\n\n");
#endif
}
void init_matrix(int** matrix, int rows, int cols)
{
int i,j;
for (i = 0; i < rows; i++)
{
for (j = 0; j < cols; j++)
{
matrix[i][j] = rand() % MAX_RAND;
}
}
}
void print_matrix(int** matrix, int rows, int cols)
{
int i;
for (i = 0; i < rows; i++)
{
printf("|");
int j;
for (j = 0; j < cols; j++)
{
printf("%d ", matrix[i][j]);
}
printf("|\n");
}
}
int main(int argc, char *argv[])
{
int my_rank;
int num_procs;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); //grab this process's rank
MPI_Comm_size(MPI_COMM_WORLD, &num_procs); //grab the total num of processes
int results[num_procs]; // used to store the partial sum computed
int rows, cols, colsB;
rows = N;
cols = N;
colsB = COLUMNS_B;
int **A; // N x N Matrix
int B[N]; // N x 1 Matrix
int **C; // N x 1 Matrix
double start_time; // use these for timing
double stop_time;
if (my_rank == MASTER)
{
printf("Number of processes: %d\n", num_procs);
printf("N: %d\n", N);
srand(time(NULL));
// init A
int i;
A = malloc(rows * sizeof *A);
for (i = 0; i < rows; i++)
{
A[i] = malloc(cols * sizeof *A[i]);
}
init_matrix(A, rows, cols);
printf("Matrix A:\n");
print_matrix(A, rows, cols);
// init B
init_vec(B, N);
print_vec("Matrix B:\n", B, N);
// init C
C = malloc(rows * sizeof *C);
for (i = 0; i < rows; i++)
{
C[i] = malloc(colsB * sizeof *C[i]);
}
start_time = MPI_Wtime();
}
MPI_Bcast(B, N, MPI_INT, 0, MPI_COMM_WORLD);
//MPI_Bcast(A, N, MPI_INT, 0, MPI_COMM_WORLD);
int row = my_rank;
int my_sum = 0;
int i;
if (my_rank < N)
{
for (i = 0; i < N; i++)
{
int num = A[row][i] * B[i];
my_sum = my_sum + num;
}
C[row] = &my_sum;
printf("HAI FROM PROCESS %d! I will calculate row %d. My calculation: %d\n", my_rank, row, my_sum);
}
//MPI_Gather(&C, 1, MPI_INT, results, 1, MPI_INT, 0, MPI_COMM_WORLD);
if (my_rank == MASTER)
{
stop_time = MPI_Wtime();
printf("\nMatrix C:\n");
print_matrix(C, rows, colsB);
printf("Total time (sec): %f\n", stop_time - start_time);
}
MPI_Finalize();
return EXIT_SUCCESS;;
}
我很确定我很接近,但我只是错过了一些东西。我已经尝试通过广播 A 矩阵和/或调用 MPI_GATHER 添加一些已注释掉的语句,但似乎除了主 proc 之外的任何 proc 都没有给出结果,所以很明显我仍然做错了什么。这是一些示例输出:
Number of processes: 28
N: 4
Matrix A:
|11 30 69 24 |
|83 38 66 71 |
|68 71 27 33 |
|58 5 50 10 |
Matrix B:
1 58 81 44
HAI FROM PROCESS 0! I will calculate row 0. My calculation: 8396
Matrix C:
|8396 |
|-2107258888 |
|-2107258920 |
|-2107258888 |
Total time (sec): 0.000078
所以 proc 0 计算正确,但是我的错误消息是 proc 1 出现段错误,我不知道为什么。我得到的错误是:
mpirun noticed that process rank 1 with PID 0 exited on signal 11 (Segmentation fault).
任何帮助将不胜感激!