Linux【8】-软件管理-9-软件安装8-在centos上安装MPI

二、安装

2.1 MPI(centos7.2)

查看MPI可安装的版本，我是直接全部安装

yum list mpich*

安装：

yum install -y mpich*

可以查看到设置的路径

which mpicc

如果找不到，就需要添加环境变量

添加环境变量

vim ~/.bashrc

#在文件中添加，具体路径要根据系统和具体环境，以下是我的情况：

export PATH=$PATH:/usr/lib64/mpich/bin/

然后使其生效

source .bashrc

yum 默认mpich安装路径是在/usr/lib64。如果在不到，就需要find / -name “mpicc” 查找安装路径

2.2 MPICH安装

1.获得MPI源码；

mpich-3.1.tar.gz（从http://www.mpich.org/官网下载）

2.在目录/home下创建目录mpi/mpich-3.1/src

3.将mpich-3.1.tar.gz拷贝到

/home/mpi/mpich-3.1/src下；

4.解压mpich-3.1.tar.gz,生成目录mpich-3.1；命令

tar –zxvf MPICH-3.1.tar.gz

5.进入目录mpich-3.1；

6.配置：

./configure –prefix=/home/mpi/mpich-3.1  (/home/mpi/mpich-3.1是配置安装目录)

7.编译：

make

8.安装：

make install

9.打开~/.bash_profile文件；命令：vim ~/.bash_profile

在文件最后添加以下三行：

export MPI_ROOT=/home/mpi/mpich-3.1
export PATH=$MPI_ROOT/bin:$PATH
export MANPATH=$MPI_ROOT/man:$MANPATH

10.测试环境变量设置；命令which mpicc 显示mpicc所在刚才安装的目录的bin目录下；例如：

[root@localhost mpich-3.1]# which mpicc
/home/mpi/mpich-3.1/bin/mpicc

[root@localhost mpich-3.1]# which mpirun
/home/mpi/mpich-3.1/bin/mpirun

11.测试MPI程序，在刚才解压的mpich-3.1目录下有个examples目录，在exapmle目录下cpi.c文件，此程序是并行计算pi值；

首先，编译cpi.c文件，生成cpi可执行文件（若目录已经有了cpi则此步骤可省略）

mpicc cpi.c –o cpi

然后，执行cpi：

mpirun -n <number> . /cpi   //<number> 是要启动的进程数，例如8；

执行结果如下：

【root@localhost exapmles】#mpirun –n 8 ./cpi
Process 1 of 8 is on localhost
Process 3 of 8 is on localhost
Process 4 of 8 is on localhost
Process 5 of 8 is on localhost
Process 0 of 8 is on localhost
Process 2 of 8 is on localhost
Process 6 of 8 is on localhost
Process 7 of 8 is on localhost
Pi is approximately 3.1415926544231265,Error is 0.00000000008333334
Wall clock time =0.000161

　　localhost是主机名，8是启动了8个进程，1是本进程号是1等；

到此完成安装；

二、测试

mpi代码示例（hello.c）：

#include <mpi.h>
#include <stdio.h>
#include <math.h>
int main(int argc,char** argv)
{
    int myid,numproces;
    int namelen;
    char processor_name[MPI_MAX_PROCESSOR_NAME];
    MPI_Init(&argc,&argv);
    MPI_Comm_rank(MPI_COMM_WORLD,&myid);
    MPI_Comm_size(MPI_COMM_WORLD,&numproces);
    MPI_Get_processor_name(processor_name,&namelen);
    fprintf(stdout,"hello world! Process %d of %d on %s\n",
            myid,numproces,processor_name);
    MPI_Finalize();

    return 0;
}

编译：

mpicc -o hello hello.c

运行：

mpirun -np 4 ./hello

输出:

hello world! Process 0 of 4 on node25
hello world! Process 1 of 4 on node25
hello world! Process 3 of 4 on node25
hello world! Process 2 of 4 on node25

三、报错

mpi报错

/data/software/amber/amber18//bin/pmemd.MPI: error while loading shared libraries: libmpifort.so.12: cannot open shared object file: No such file or directory
/data/software/amber/amber18//bin/pmemd.MPI: error while loading shared libraries: libmpifort.so.12: cannot open shared object file: No such file or directory
/data/software/amber/amber18//bin/pmemd.MPI: error while loading shared libraries: libmpifort.so.12: cannot open shared objec

天呐，直接重新安装就好。。

参考资料

• https://www.jianshu.com/p/57b1460c8afd