**
一、前言
**
关于Docker的安装见博文windows安装Docker
本文涉及的MPI编程为求两个矩阵的乘积，语言为C++
**
二、安装MPI
**
1.因为我们使用的语言是C++，需要gcc或g++支持C++编程，所以查看Docker内哪些镜像有g++
docker search g++
2.选择其中一个镜像拉取
docker pull eclipse/cpp_gcc
3.查看当前所有镜像
docker images
发现拉取成功。
4.用镜像生成容器，并进入命令行
docker run -it -u root --net=host eclipse/cpp_gcc /bin/bash
5.后续下载MPI编程所需要的配置文件时，需要访问国外仓库，我们可以把源换为国内的，速度更快。
sed -i 's/deb.debian.org/mirrors.ustc.edu.cn/g' /etc/apt/sources.list
cat /etx/apt/sources.list
rm -Rf /var/lib/apt/lists/*
apt-get update
6.安装mpich和libeigen3-dev
apt install mpich
apt install libeigen3-dev
7.退出容器，查看容器id
exit
docker ps -a
8.用容器id查找容器全名Id，并记住全名，在下一步要用到
docker inspect id
**
三、运行MPI
**
1.退出Docker后，将自己写好的mpi代码从本机复制到镜像文件夹内并查看。注：这里我的mpi代码文件为testmpi.cpp；注意区分容器名(id)和容器全名(Id)
docker cp 本机代码路径 容器全名:/home/user/代码文件名
docker start 容器名
docker exec -it -u root 容器名 bash
cd /home/user
ls
可以看到该文件
2.复制mpi编译所需的libmpi.so到当前目录下
cp /usr/lib/libmpi.so /home/user/
cd /home/user
ls
3.mpi编译
g++ testmpi.cpp libmpi.so -o mpi
ls
4.运行结果
mpiexec -n 4 ./mpi
其中，4为设置的进程数
5.testmpi.cpp代码如下
#include <stdio.h>
#include <mpich/mpi.h>
#include <string.h>
#include <stdlib.h>
#include <math.h>
#include <time.h>
#include<omp.h>
#include<iostream>
using namespace std;
//生成随机矩阵
int *generate_matrix(int size)
{
srand((unsigned)time(NULL) + (unsigned)rand());
int *matrix;
matrix = (int *)malloc(sizeof(int) * size*size);
for (int i = 0; i < size*size; i++)
{
matrix[i] = rand() % 10;
}
return matrix;
}
//输出矩阵
void print_matrx(int *a, int size)
{
for (int i = 0; i < size; i++)
{
for (int j = 0; j < size; j++)
{
printf("%d ", a[i*size + j]);
}
printf("\n");
}
printf("\n");
}
//矩阵相乘
int * Multiplication(int a[], int b[], int size, int line)
{
int *result;
int temp = 0;
result = (int *)malloc(sizeof(int) * size*size);
//#pragma omp parallel for num_threads(2)
for (int i = 0; i < line; i++)
{
for (int j = 0; j < size; j++)
{
temp = 0;
for (int k = 0; k < size; k++)
temp += a[i*size + k] * b[k*size + j];
result[i*size + j] = temp;
}
}
return result;
}
int main(int argc, char *argv[])
{
clock_t time1, time2;
int size = 16, rank, line, num;
time1 = clock();
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &num);
int *matrix1;
int *matrix2;
int *matrix3;
int *resultMg;
int *revMg;
int *resultMg0;
line = size / num ; //num为进程数，line为每个进程的行数
matrix1 = (int*)malloc(sizeof(int)*size*size);
matrix2 = (int*)malloc(sizeof(int)*size*size);
matrix3 = (int*)malloc(sizeof(int)*size*size);
resultMg = (int*)malloc(sizeof(int)*size*line);
resultMg0 = (int*)malloc(sizeof(int)*size*line);
revMg = (int*)malloc(sizeof(int)*size*line);
if (rank == 0)
{
matrix1 = generate_matrix(size);
matrix2 = generate_matrix(size);
printf("matrix1 is :\n");
print_matrx((int *)matrix1, size);
printf("matrix2 is :\n");
print_matrx((int *)matrix2, size);
resultMg0=Multiplication(matrix1,matrix2, size, line);
for (int m = 0; m < line; m++)
for (int n = 0; n < size; n++)
matrix3[m*size + n] = resultMg0[m*size + n];
for (int i = 1; i < num; i++)
MPI_Send(matrix2, size*size, MPI_INT, i, 0, MPI_COMM_WORLD);
for (int i = 1; i < num; i++)
MPI_Send(matrix1 + i*line*size, size*line, MPI_INT, i, 1, MPI_COMM_WORLD);
for (int i = 1; i < num; i++)
{
MPI_Recv(resultMg, line*size, MPI_INT, i, 3, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
for (int m = 0; m < line; m++)
for (int n = 0; n < size; n++)
matrix3[(i*line + m)*size + n] = resultMg[m*size + n];
}
time2 = clock();
print_matrx((int *)matrix3, size);
cout << time2 - time1 << endl;
free(matrix1);
free(matrix2);
free(matrix3);
free(revMg);
free(resultMg);
}
else {
MPI_Recv(matrix2, size*size, MPI_INT, 0, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(revMg, size*line, MPI_INT, 0, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
resultMg = Multiplication(revMg, matrix2, size, line);
MPI_Send(resultMg, line*size, MPI_INT, 0, 3, MPI_COMM_WORLD);
}
MPI_Finalize();
return 0;
}