如何在C中使用MPI查找給定數字的總和？

Question

我試圖找到一個數組中所有給定數字的總和。 我必須將數組拆分成相等大小，然后發送給每個進程並計算總和。 稍后將計算得出的總和從每個過程發送到根過程以得到最終答案。 實際上，我知道我可以使用MPI_Scatter 。 但是我的問題是，如果我的清單是奇數 ，該怎么辦？ 例如，我有一個包含13元素的數組，那么我有3過程。 因此，默認情況下， MPI_Scatter會將數組除以3並保留最后一個元素。 基本上，它將只計算12元素的總和。 我只使用MPI_Scatter時的輸出：

myid = 0 total = 6
myid = 1 total = 22
myid = 2 total = 38
results from all processors_= 66 
size= 13 

因此，我計划使用MPI_Scatter和MPI_Send 。 因此，我可以獲取最后一個元素，並通過MPI_Send發送該元素並MPI_Send進行計算，然后在根進程中進行接收。 但是我遇到了問題..我的代碼：

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <mpi.h>

/*  globals */
int numnodes, myid, mpi_err;
int last_core;
int n;
int last_elements[];

#define mpi_root 0
/* end globals  */

void init_it(int  *argc, char ***argv);

void init_it(int  *argc, char ***argv) {
    mpi_err = MPI_Init(argc, argv);
    mpi_err = MPI_Comm_size( MPI_COMM_WORLD, &numnodes );
    mpi_err = MPI_Comm_rank(MPI_COMM_WORLD, &myid);
}

int main(int argc, char *argv[]) {
    int *myray, *send_ray, *back_ray;
    int count;
    int size, mysize, i, k, j, total;

    MPI_Status status;

    init_it(&argc, &argv);

    /* each processor will get count elements from the root */
    count = 4;
    myray = (int*)malloc(count * sizeof(int));
    size = (count * numnodes) + 1;
    send_ray = (int*)malloc(size * sizeof(int));
    back_ray = (int*)malloc(numnodes * sizeof(int));
    last_core = numnodes - 1;

    /* create the data to be sent on the root */
    if(myid == mpi_root){
        for(i = 0; i < size; i++)
        {
            send_ray[i] = i;
        }
    }

    /* send different data to each processor */
    mpi_err = MPI_Scatter( send_ray, count, MPI_INT,
                           myray, count, MPI_INT,
                           mpi_root, MPI_COMM_WORLD);

    if(myid == mpi_root) {
        n = 1;
        memcpy(last_elements, &send_ray[size-n], n * sizeof(int));

        //Send the last numbers to the last core through send command
        MPI_Send(last_elements, n, MPI_INT, last_core, 99, MPI_COMM_WORLD);
    }

    /* each processor does a local sum */
    total = 0;
    for(i = 0; i < count; i++)
        total = total + myray[i];
        //total = total + send_ray[size-1];
    printf("myid= %d total= %d\n", myid, total);

    if(myid == last_core)
    {
        printf("Last core\n");
        MPI_Recv(last_elements, n, MPI_INT, 0, 99, MPI_COMM_WORLD, &status);
    }

    /* send the local sums back to the root */
    mpi_err = MPI_Gather(&total, 1, MPI_INT,
                        back_ray, 1, MPI_INT,
                        mpi_root, MPI_COMM_WORLD);

    /* the root prints the global sum */
    if(myid == mpi_root){
        total=0;
        for(i = 0; i < numnodes; i++)
            total = total + back_ray[i];
        printf("results from all processors_= %d \n", total);
        printf("size= %d \n ", size);
    }

    mpi_err = MPI_Finalize();
}

輸出：

myid = 0 total = 6
myid = 1 total = 22
myid = 2 total = 38
Last core
[ubuntu:11884] *** An error occurred in MPI_Recv
[ubuntu:11884] *** on communicator MPI_COMM_WORLD
[ubuntu:11884] *** MPI_ERR_TRUNCATE: message truncated
[ubuntu:11884] *** MPI_ERRORS_ARE_FATAL: your MPI job will now abort
--------------------------------------------------------------------------
mpiexec has exited due to process rank 2 with PID 11884 on
node ubuntu exiting improperly. There are two reasons this could occur:

1. this process did not call "init" before exiting, but others in
the job did. This can cause a job to hang indefinitely while it waits
for all processes to call "init". By rule, if one process calls "init",
then ALL processes must call "init" prior to termination.

2. this process called "init", but exited without calling "finalize".
By rule, all processes that call "init" MUST call "finalize" prior to
exiting or it will be considered an "abnormal termination"

This may have caused other processes in the application to be
terminated by signals sent by mpiexec (as reported here).

我知道我做錯了。 如果您能指出我，我將不勝感激。

Answer 1

您的last_elements數組沒有指定大小。 MPI_Recv出錯，因為沒有空間放置正在發送的項目。 您的代碼缺少last_elements的malloc。

Answer 2

我可能回答得很晚，但可能其他人可以得到幫助。

請檢查以下代碼

# include <cstdlib>
# include <iostream>
# include <iomanip>
# include <ctime>
# include <mpi.h>

using namespace std;

int main ( int argc, char *argv[] );
void timestamp ( );

//****************************************************************************80

int main ( int argc, char *argv[] )

//****************************************************************************80

{
  int *a;
  int dest;
  float factor;
  int global;
  int i;
  int id;
  int ierr;
  int n;
  int npart;
  int p;
  int source;
  int start;
  MPI_Status status;
  int tag;
  int tag_target = 1;
  int tag_size = 2;
  int tag_data = 3;
  int tag_found = 4;
  int tag_done = 5;
  int target;
  int workers_done;
  int x;
//
//  Initialize MPI.
//
  ierr = MPI_Init ( &argc, &argv );
//
//  Get this processes's rank.
//
  ierr = MPI_Comm_rank ( MPI_COMM_WORLD, &id );
//
//  Find out how many processes are available.
//
  ierr = MPI_Comm_size ( MPI_COMM_WORLD, &p );

  if ( id == 0 )
  {
    timestamp ( );
    cout << "\n";
    cout << "SEARCH - Master process:\n";
    cout << "  C++ version\n";
    cout << "  An example MPI program to search an array.\n";
    cout << "\n";
    cout << "  Compiled on " << __DATE__ << " at " << __TIME__ << ".\n";
    cout << "\n";
    cout << "  The number of processes is " << p << "\n";
  }

  cout << "\n";
  cout << "Process " << id << " is active.\n";
//
//  Have the master process generate the target and data.  In a more 
//  realistic application, the data might be in a file which the master 
//  process would read.  Here, the master process decides.
//
  if ( id == 0 )
  {
//
//  Pick the number of data items per process, and set the total.
//
    factor = ( float ) rand ( ) / ( float ) RAND_MAX;
    npart = 50 + ( int ) ( factor * 100.0E+00 );
    n = npart * p;

    cout << "\n";
    cout << "SEARCH - Master process:\n";
    cout << "  The number of data items per process is " << npart << "\n";
    cout << "  The total number of data items is       " << n << ".\n";
//
//  Now allocate the master copy of A, fill it with values, and pick 
//  a value for the target.
//
    a = new int[n];

    factor = ( float ) n / 10.0E+00 ;

    for ( i = 0; i < n; i++ ) 
    {
      a[i] = ( int ) ( factor * ( float ) rand ( ) / ( float ) RAND_MAX );
    }
    target = a[n/2];

    cout << "  The target value is " << target << ".\n";
//
//  The worker processes need to have the target value, the number of data items,
//  and their individual chunk of the data vector.
//
    for ( i = 1; i <= p-1; i++ )
    {
      dest = i;
      tag = tag_target;

      ierr = MPI_Send ( &target, 1, MPI_INT, dest, tag, MPI_COMM_WORLD );

      tag = tag_size;

      ierr = MPI_Send ( &npart, 1, MPI_INT, dest, tag, MPI_COMM_WORLD );

      start = ( i - 1 ) * npart;
      tag = tag_data;

      ierr = MPI_Send ( a+start, npart, MPI_INT, dest, tag,
    MPI_COMM_WORLD );
    }
//
//  Now the master process simply waits for each worker process to report that 
//  it is done.
//
    workers_done = 0;

    while ( workers_done < p-1 )
    {
      ierr = MPI_Recv ( &x, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG,
    MPI_COMM_WORLD, &status );

      source = status.MPI_SOURCE;
      tag = status.MPI_TAG;

      if ( tag == tag_done )
      {
    workers_done = workers_done + 1;
      }
      else if ( tag == tag_found )
      {
    cout << "P" << source << "  " << x << "  " << a[x] << "\n";
      }
      else
      {
    cout << "  Master process received message with unknown tag = "
         << tag << ".\n";
      }

    }
//
//  The master process can throw away A now.
//
    delete [] a;
  }
//
//  Each worker process expects to receive the target value, the number of data
//  items, and the data vector.
//
  else 
  {
    source = 0;
    tag = tag_target;

    ierr = MPI_Recv ( &target, 1, MPI_INT, source, tag, MPI_COMM_WORLD,
      &status );

    source = master;
    tag = tag_size;

    ierr = MPI_Recv ( &npart, 1, MPI_INT, source, tag, MPI_COMM_WORLD, 
      &status );

    a = new int[npart];

    source = 0;
    tag = tag_data;

    ierr = MPI_Recv ( a, npart, MPI_INT, source, tag, MPI_COMM_WORLD,
      &status );
//
//  The worker simply checks each entry to see if it is equal to the target
//  value.
//
    for ( i = 0; i < npart; i++ )
    {
      if ( a[i] == target )
      {
    global = ( id - 1 ) * npart + i;
    dest = 0;
    tag = tag_found;

    ierr = MPI_Send ( &global, 1, MPI_INT, dest, tag, MPI_COMM_WORLD );
      }
    }
//
//  When the worker is finished with the loop, it sends a dummy data value with
//  the tag "TAG_DONE" indicating that it is done.
//
    dest = 0;
    tag = tag_done;

    ierr = MPI_Send ( &target, 1, MPI_INT, dest, tag, MPI_COMM_WORLD );

    delete [] ( a );
  }
//
//  Terminate MPI.
//
  MPI_Finalize ( );
//
//  Terminate.
//
  if ( id == 0 )
  {
    cout << "\n";
    cout << "SEARCH - Master process:\n";
    cout << "  Normal end of execution.\n";
    cout << "\n";
    timestamp ( );
  } 
  return 0;
}
//****************************************************************************80

void timestamp ( )

//****************************************************************************80

{
# define TIME_SIZE 40

  static char time_buffer[TIME_SIZE];
  const struct std::tm *tm_ptr;
  size_t len;
  std::time_t now;

  now = std::time ( NULL );
  tm_ptr = std::localtime ( &now );

  len = std::strftime ( time_buffer, TIME_SIZE, "%d %B %Y %I:%M:%S %p", tm_ptr );

  std::cout << time_buffer << "\n";

  return;
# undef TIME_SIZE
}

輸出為：

SEARCH - Master process:
A program using MPI, to search an array.
Compiled on jan  14 2018 at 11:21:45.

The number of processes is 4

Process 0 is active.

SEARCH - Master process:
The number of data items per process is 101
The total number of data items is       404.
The target value is 14.
P3  202  14
P2  145  14
P2  178  14
P2  180  14
P3  211  14
P3  240  14
P3  266  14
P3  295  14
P1  12  14
P1  23  14
P1  36  14
P1  71  14

SEARCH - Master process:
  Normal end of execution.

Process 1 is active.

Process 2 is active.

Process 3 is active.