MPI + FORTRAN 的子例行子程序测试

最新推荐文章于 2024-09-12 16:19:29 发布

zhuliting

最新推荐文章于 2024-09-12 16:19:29 发布

阅读量1.2w

点赞数 1

分类专栏： FORTRAN MPI 文章标签： fortran integer subroutine include signal c

本文链接：https://blog.csdn.net/zhuliting/article/details/6552648

版权

MPI 同时被 2 个专栏收录

29 篇文章 1 订阅

订阅专栏

FORTRAN

21 篇文章 2 订阅

订阅专栏

用MPI实现一个简单的标准发送接收操作，如下，0进程将一个数据发送给1号进程，1号进程接收从0进程发送的数据。

program main implicit none include 'mpif.h' integer myid, ierr,status(MPI_STATUS_SIZE) real(kind=8) i, j i = 3.14159 j = 4.22222 call MPI_INIT(ierr) call MPI_COMM_RANK(MPI_COMM_WORLD, myid, ierr) if(myid.eq.0) then call MPI_Send(i, 1, MPI_DOUBLE_PRECISION, 1, 99, & MPI_COMM_WORLD,IERR) print *,'sending',i else call MPI_Recv(j, 1, MPI_DOUBLE_PRECISION, 0, 99, & MPI_COMM_WORLD,STATUS,IERR) print *, 'receiving',j endif call MPI_FINALIZE(ierr) end

result:

[root@c0108 zlt]# mpif90 abc.F -o a [root@c0108 zlt]# mpiexec -n 2 ./a sending 3.14159011840820 receiving 3.14159011840820

现在，我把发送和接收操作分别放到子例行子程序里，如下

program main implicit none include 'mpif.h' integer myid, ierr,status(MPI_STATUS_SIZE) real(kind=8) i, j i = 3.14159 j = 4.22222 call MPI_INIT(ierr) call MPI_COMM_RANK(MPI_COMM_WORLD, myid, ierr) if(myid.eq.0) then call mySend() else call myRecv() endif call MPI_FINALIZE(ierr) end subroutine mySend() integer ierr real(kind=8) i i = 2.22222222 call MPI_Send(i, 1, MPI_DOUBLE_PRECISION, 1, 99, & MPI_COMM_WORLD,IERR) print *,'sending',i end subroutine myRecv() include 'mpif.h' integer ierr, status(MPI_STATUS_SIZE) real(kind=8) j j = 3.3333333 call MPI_Recv(j, 1, MPI_DOUBLE_PRECISION, 0, 99, & MPI_COMM_WORLD,STATUS,IERR) print *, 'receiving',j end

于是，出错了。。。

[root@c0108 zlt]# mpif90 abc.F -o a [root@c0108 zlt]# mpiexec -n 2 ./a rank 0 in job 10 c0108_43628 caused collective abort of all ranks exit status of rank 0: killed by signal 11 [root@c0108 zlt]#

感谢hahe 的指正，终于找到了错误原因：

子例行子程序中，也要加头文件包含

#include 'mpif.h'

附一个修改后类似的代码，将i,j作为参数传递给子程序，执行发送接收操作。

program main implicit none include 'mpif.h' integer myid, ierr,status(MPI_STATUS_SIZE) real(kind=8) i, j i = 3.14159 j = 4.22222 call MPI_INIT(ierr) call MPI_COMM_RANK(MPI_COMM_WORLD, myid, ierr) if(myid.eq.0) then call mySend(i) else call myRecv(j) endif print *,'i,j:',i,j, 'of process',myid call MPI_FINALIZE(ierr) end subroutine mySend(i) implicit none include 'mpif.h' integer ierr real(kind=8) i call MPI_Send(i, 1, MPI_DOUBLE_PRECISION, 1, 99, & MPI_COMM_WORLD,IERR) print *,'sending',i end subroutine myRecv(j) implicit none include 'mpif.h' integer ierr, status(MPI_STATUS_SIZE) real(kind=8) j j = 3.3333333 call MPI_Recv(j, 1, MPI_DOUBLE_PRECISION, 0, 99, & MPI_COMM_WORLD,STATUS,IERR) print *, 'receiving',j end

result:

[root@c0109 zlt]# mpif90 abc.F -o abc [root@c0109 zlt]# mpiexec -n 2 ./abc receiving 3.14159011840820 i,j: 3.14159011840820 3.14159011840820 of process 1 sending 3.14159011840820 i,j: 3.14159011840820 4.22221994400024 of process 0

并且可以将子例行程序单独放进一个文件，然后分开编译：

[root@c0109 zlt]# mpif90 -o abc abc.F myfun.F [root@c0109 zlt]# mpiexec -n 2 ./abc receiving 3.14159011840820 i,j: 3.14159011840820 3.14159011840820 of process 1 sending 3.14159011840820 i,j: 3.14159011840820 4.22221994400024 of process 0 [root@c0109 zlt]#