在 C 中的 pthread 中使用 MPI_Send答案

【问题标题】：Using MPI_Send inside a pthread in C在 C 中的 pthread 中使用 MPI_Send
【发布时间】：2014-11-27 01:54:56
【问题描述】：

我正在尝试创建一个 mpi 进程环，其中每个 MPI 进程将启动一个 pthread 并且线程将执行该环，我使用了 pthread，因此我可以使用 MPI 进程来执行另一个任务。似乎我不能在 pthread 中使用 MPI_send 或 MPI_Recv，我没有编译错误，但我确实有运行时错误。

我使用这个命令编译 mpicc -lpthread threaded_ring.c

这是运行时错误

a.out:28372 terminated with signal 11 at PC=2aaaaaae312d SP=2aaab0771860.  Backtrace:
/usr/lib64/libpsm_infinipath.so.1(psmi_mpool_get+0xd)[0x2aaaaaae312d]

a.out:28366 terminated with signal 11 at PC=333c00c110 SP=2aaab02d9698.  Backtrace:
/lib64/libpthread.so.0(pthread_spin_lock+0x0)[0x333c00c110]
/usr/lib64/libpsm_infinipath.so.1(psmi_amsh_short_request+0x180)[0x2aaaaaad31b0]
/usr/lib64/libpsm_infinipath.so.1(+0xd9f6)[0x2aaaaaad49f6]
/usr/lib64/libpsm_infinipath.so.1(psm_mq_send+0x41)[0x2aaaaaaf5d51]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(psm_send_pkt+0xb1)[0x2aaaaae0af21]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(psm_istartmsgv+0x130)[0x2aaaaae0a010]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(MPIDI_CH3_iStartMsgv+0x6)[0x2aaaaaddf1e6]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(MPIDI_CH3_EagerContigSend+0x89)[0x2aaaaada6e39]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(MPID_Send+0x116)[0x2aaaaade3136]
/usr/local/mpi/mvapich2/intel12/1.8.1/lib/libmpich.so.3(MPI_Send+0xf8)[0x2aaaaae2a408]
./a.out[0x4022ba]
/lib64/libpthread.so.0[0x333c0077f1]
/lib64/libc.so.6(clone+0x6d)[0x333bce570d]

a.out:28373 terminated with signal 11 at PC=333bf9d428 SP=2aaab0771838.  Backtrace:

a.out:28370 terminated with signal 11 at PC=2aaaaaae312d SP=2aaab0771860.  Backtrace:

这是我的代码

#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <time.h>

void *ring_func(void *p)
{
  int token=1;
  int world_rank;
  MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
  int world_size;
  MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  if (world_rank==0){
   MPI_Send(&token, 1, MPI_INT, (world_rank + 1) % world_size, 0,
         MPI_COMM_WORLD);
   }

  if (world_rank != 0) {
    MPI_Recv(&token, 1, MPI_INT, world_rank - 1, 0, MPI_COMM_WORLD,
             MPI_STATUS_IGNORE);
    printf("Process %d received token %d from process %d\n", world_rank, token,
           world_rank - 1);
}
pthread_exit(NULL);
}

    int main(int argc, char** argv) {
  // Initialize the MPI threaded environment

  int provided;
  MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE , &provided);
 if (provided < MPI_THREAD_MULTIPLE)
 {
    printf("Error: the MPI library doesn't provide the required thread level\n");
       MPI_Abort(MPI_COMM_WORLD, 0);
       }
  pthread_t ring ;
  pthread_create (&ring, NULL, ring_func, NULL) ;

  MPI_Barrier(MPI_COMM_WORLD);
  MPI_Finalize();
}

【问题讨论】：

MPI+pthreads. Program stuck on MPI_Ssend and MPI_Recv 的可能重复项
我已经添加了 MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);但我仍然有同样的错误！！
在调用MPI_Init_thread 后检查provided 是否等于MPI_THREAD_MULTIPLE。此外，您永远不会加入线程，也无法保证主线程在第二个线程开始之前不会成功通过障碍并调用MPI_Finalize()。
嗨 lliev 谢谢你的帮助，我确实添加了提供的并确保它等于 MPI_THREAD_MULTIPLE 但它似乎没有解决问题，我认为你是对的有时主线程只会调用 MPI_Finlize 但我不知道如何解决这个问题。在我的 pthread 完成之前，你能想出一些方法来阻止主线程终止吗？
pthread_join(ring, NULL); 屏障前。

标签： pthreads mpi

【解决方案1】：

感谢 Hristo lliev 我能够解决问题，问题是主线程在我的 pthread 之前完成，但是当我添加 pthread_join 时，主线程在调用之前等待 pthread 加入MPI_Finalize()。这是新代码

#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <time.h>

void *ring_func(void *p)
{
  int token;
  // Receive from the lower process and send to the higher process. Take care
  // of the special case when you are the first process to prevent deadlock.

  int world_rank;
  MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
  int world_size;
  MPI_Comm_size(MPI_COMM_WORLD, &world_size);

  if (world_rank != 0) {
    MPI_Recv(&token, 1, MPI_INT, world_rank - 1, 0, MPI_COMM_WORLD,
             MPI_STATUS_IGNORE);
    printf("Process %d received token %d from process %d\n", world_rank, token,
           world_rank - 1);
  } else {
// Set the token's value if you are process
 token = -1;
  }
MPI_Send(&token, 1, MPI_INT, (world_rank + 1) % world_size, 0,
           MPI_COMM_WORLD);

 if (world_rank == 0) {
//    sleep(20);
    MPI_Recv(&token, 1, MPI_INT, world_size - 1, 0, MPI_COMM_WORLD,
             MPI_STATUS_IGNORE);
    printf("Process %d received token %d from process %d\n", world_rank, token,
           world_size - 1);
  }

pthread_exit(NULL);
}

int main(int argc, char** argv) {
  // Initialize the MPI threaded environment

  int provided;
  MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE , &provided);
 if (provided != MPI_THREAD_MULTIPLE)
 {
    printf("Error: the MPI library doesn't provide the required thread level\n");
       MPI_Abort(MPI_COMM_WORLD, 0);
       }
  pthread_t ring ;
  pthread_create (&ring, NULL, ring_func, NULL) ;
  pthread_join(ring,NULL);
  MPI_Barrier(MPI_COMM_WORLD);
  MPI_Finalize();
}

【讨论】：