alltoall.c

fortran并行计算包

/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
 *
 *  (C) 2001 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

#include "mpiimpl.h"

/* -- Begin Profiling Symbol Block for routine MPI_Alltoall */
#if defined(HAVE_PRAGMA_WEAK)
#pragma weak MPI_Alltoall = PMPI_Alltoall
#elif defined(HAVE_PRAGMA_HP_SEC_DEF)
#pragma _HP_SECONDARY_DEF PMPI_Alltoall  MPI_Alltoall
#elif defined(HAVE_PRAGMA_CRI_DUP)
#pragma _CRI duplicate MPI_Alltoall as PMPI_Alltoall
#endif
/* -- End Profiling Symbol Block */

/* Define MPICH_MPI_FROM_PMPI if weak symbols are not supported to build
   the MPI routines */
#ifndef MPICH_MPI_FROM_PMPI
#undef MPI_Alltoall
#define MPI_Alltoall PMPI_Alltoall

/* This is the default implementation of alltoall. The algorithm is:
   
   Algorithm: MPI_Alltoall

   We use four algorithms for alltoall. For short messages and
   (comm_size >= 8), we use the algorithm by Jehoshua Bruck et al,
   IEEE TPDS, Nov. 1997. It is a store-and-forward algorithm that
   takes lgp steps. Because of the extra communication, the bandwidth
   requirement is (n/2).lgp.beta.

   Cost = lgp.alpha + (n/2).lgp.beta

   where n is the total amount of data a process needs to send to all
   other processes.

   For medium size messages and (short messages for comm_size < 8), we
   use an algorithm that posts all irecvs and isends and then does a
   waitall. We scatter the order of sources and destinations among the
   processes, so that all processes don't try to send/recv to/from the
   same process at the same time.

   For long messages and power-of-two number of processes, we use a
   pairwise exchange algorithm, which takes p-1 steps. We
   calculate the pairs by using an exclusive-or algorithm:
           for (i=1; i<comm_size; i++)
               dest = rank ^ i;
   This algorithm doesn't work if the number of processes is not a power of
   two. For a non-power-of-two number of processes, we use an
   algorithm in which, in step i, each process  receives from (rank-i)
   and sends to (rank+i). 

   Cost = (p-1).alpha + n.beta

   where n is the total amount of data a process needs to send to all
   other processes.

   Possible improvements: 

   End Algorithm: MPI_Alltoall
*/

/* begin:nested */
/* not declared static because a machine-specific function may call this one in some cases */
int MPIR_Alltoall( 
    void *sendbuf, 
    int sendcount, 
    MPI_Datatype sendtype, 
    void *recvbuf, 
    int recvcount, 
    MPI_Datatype recvtype, 
    MPID_Comm *comm_ptr )
{
    static const char FCNAME[] = "MPIR_Alltoall";
    int          comm_size, i, j, pof2;
    MPI_Aint     sendtype_extent, recvtype_extent;
    MPI_Aint recvtype_true_extent, recvbuf_extent, recvtype_true_lb;
    int mpi_errno=MPI_SUCCESS, src, dst, rank, nbytes;
    MPI_Status status;
    int sendtype_size, pack_size, block, position, *displs, count;
    MPI_Datatype newtype;
    void *tmp_buf;
    MPI_Comm comm;
    MPI_Request *reqarray;
    MPI_Status *starray;
#ifdef MPIR_OLD_SHORT_ALLTOALL_ALG
    MPI_Aint sendtype_true_extent, sendbuf_extent, sendtype_true_lb;
    int k, p, curr_cnt, dst_tree_root, my_tree_root;
    int last_recv_cnt, mask, tmp_mask, tree_root, nprocs_completed;
#endif

    if (sendcount == 0) return MPI_SUCCESS;
    
    comm = comm_ptr->handle;
    comm_size = comm_ptr->local_size;
    rank = comm_ptr->rank;
    
    /* Get extent of send and recv types */
    MPID_Datatype_get_extent_macro(recvtype, recvtype_extent);
    MPID_Datatype_get_extent_macro(sendtype, sendtype_extent);

    MPID_Datatype_get_size_macro(sendtype, sendtype_size);
    nbytes = sendtype_size * sendcount;
    
    /* check if multiple threads are calling this collective function */
    MPIDU_ERR_CHECK_MULTIPLE_THREADS_ENTER( comm_ptr );
    
    if ((nbytes <= MPIR_ALLTOALL_SHORT_MSG) && (comm_size >= 8)) {

        /* use the indexing algorithm by Jehoshua Bruck et al,
         * IEEE TPDS, Nov. 97 */ 

        /* allocate temporary buffer */
        NMPI_Pack_size(recvcount*comm_size, recvtype, comm, &pack_size);
        tmp_buf = MPIU_Malloc(pack_size);
	/* --BEGIN ERROR HANDLING-- */
        if (!tmp_buf) {
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
            return mpi_errno;
        }
	/* --END ERROR HANDLING-- */

        /* Do Phase 1 of the algorithim. Shift the data blocks on process i
         * upwards by a distance of i blocks. Store the result in recvbuf. */
        mpi_errno = MPIR_Localcopy((char *) sendbuf + 
			   rank*sendcount*sendtype_extent, 
                           (comm_size - rank)*sendcount, sendtype, recvbuf, 
                           (comm_size - rank)*recvcount, recvtype);
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
        mpi_errno = MPIR_Localcopy(sendbuf, rank*sendcount, sendtype, 
                        (char *) recvbuf + 
				   (comm_size-rank)*recvcount*recvtype_extent, 
                                   rank*recvcount, recvtype);
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
        /* Input data is now stored in recvbuf with datatype recvtype */

        /* Now do Phase 2, the communication phase. It takes
           ceiling(lg p) steps. In each step i, each process sends to rank+2^i
           and receives from rank-2^i, and exchanges all data blocks
           whose ith bit is 1. */

        /* allocate displacements array for indexed datatype used in
           communication */

        displs = MPIU_Malloc(comm_size * sizeof(int));
	/* --BEGIN ERROR HANDLING-- */
        if (!displs) {
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
            return mpi_errno;
        }
	/* --END ERROR HANDLING-- */

        pof2 = 1;
        while (pof2 < comm_size) {
            dst = (rank + pof2) % comm_size;
            src = (rank - pof2 + comm_size) % comm_size;

            /* Exchange all data blocks whose ith bit is 1 */
            /* Create an indexed datatype for the purpose */

            count = 0;
            for (block=1; block<comm_size; block++) {
                if (block & pof2) {
                    displs[count] = block * recvcount;
                    count++;
                }
            }

            mpi_errno = NMPI_Type_create_indexed_block(count, recvcount, 
                                               displs, recvtype, &newtype);
	    if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

            mpi_errno = NMPI_Type_commit(&newtype);
	    if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

            position = 0;
            mpi_errno = NMPI_Pack(recvbuf, 1, newtype, tmp_buf, pack_size, 
                                  &position, comm);

            mpi_errno = MPIC_Sendrecv(tmp_buf, position, MPI_PACKED, dst,
                                      MPIR_ALLTOALL_TAG, recvbuf, 1, newtype,
                                      src, MPIR_ALLTOALL_TAG, comm,
                                      MPI_STATUS_IGNORE);
	    if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

            mpi_errno = NMPI_Type_free(&newtype);
	    if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

            pof2 *= 2;
        }

        MPIU_Free(displs);
        MPIU_Free(tmp_buf);

        /* Rotate blocks in recvbuf upwards by (rank + 1) blocks. Need
         * a temporary buffer of the same size as recvbuf. */
        
        /* get true extent of recvtype */
        mpi_errno = NMPI_Type_get_true_extent(recvtype, &recvtype_true_lb,
                                              &recvtype_true_extent);  
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

        recvbuf_extent = recvcount * comm_size *
            (MPIR_MAX(recvtype_true_extent, recvtype_extent));
        tmp_buf = MPIU_Malloc(recvbuf_extent);
	/* --BEGIN ERROR HANDLING-- */
        if (!tmp_buf) {
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
            return mpi_errno;
        }
	/* --END ERROR HANDLING-- */
        /* adjust for potential negative lower bound in datatype */
        tmp_buf = (void *)((char*)tmp_buf - recvtype_true_lb);

        mpi_errno = MPIR_Localcopy((char *) recvbuf + (rank+1)*recvcount*recvtype_extent, 
                       (comm_size - rank - 1)*recvcount, recvtype, tmp_buf, 
                       (comm_size - rank - 1)*recvcount, recvtype);
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
        mpi_errno = MPIR_Localcopy(recvbuf, (rank+1)*recvcount, recvtype, 
                       (char *) tmp_buf + (comm_size-rank-1)*recvcount*recvtype_extent, 
                       (rank+1)*recvcount, recvtype);
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

        /* Blocks are in the reverse order now (comm_size-1 to 0). 
         * Reorder them to (0 to comm_size-1) and store them in recvbuf. */

        for (i=0; i<comm_size; i++) 
            MPIR_Localcopy((char *) tmp_buf + i*recvcount*recvtype_extent,
                           recvcount, recvtype, 
                           (char *) recvbuf + (comm_size-i-1)*recvcount*recvtype_extent, 
                           recvcount, recvtype); 

        MPIU_Free((char*)tmp_buf + recvtype_true_lb);



#ifdef MPIR_OLD_SHORT_ALLTOALL_ALG
        /* Short message. Use recursive doubling. Each process sends all
           its data at each step along with all data it received in
           previous steps. */
        
        /* need to allocate temporary buffer of size
           sendbuf_extent*comm_size */
        
        /* get true extent of sendtype */
        mpi_errno = NMPI_Type_get_true_extent(sendtype, &sendtype_true_lb,
                                              &sendtype_true_extent);  
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }

        sendbuf_extent = sendcount * comm_size *
            (MPIR_MAX(sendtype_true_extent, sendtype_extent));
        tmp_buf = MPIU_Malloc(sendbuf_extent*comm_size);
	/* --BEGIN ERROR HANDLING-- */
        if (!tmp_buf) {
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
            return mpi_errno;
        }
	/* --END ERROR HANDLING-- */
        
        /* adjust for potential negative lower bound in datatype */
        tmp_buf = (void *)((char*)tmp_buf - sendtype_true_lb);
        
        /* copy local sendbuf into tmp_buf at location indexed by rank */
        curr_cnt = sendcount*comm_size;
        mpi_errno = MPIR_Localcopy(sendbuf, curr_cnt, sendtype,
                                   ((char *)tmp_buf + rank*sendbuf_extent),
                                   curr_cnt, sendtype);
	if (mpi_errno) { MPIU_ERR_POP(mpi_errno);}
        
        mask = 0x1;
        i = 0;
        while (mask < comm_size) {
            dst = rank ^ mask;
            
            dst_tree_root = dst >> i;
            dst_tree_root <<= i;
            
            my_tree_root = rank >> i;
            my_tree_root <<= i;
            
            if (dst < comm_size) {
                mpi_errno = MPIC_Sendrecv(((char *)tmp_buf +
                                           my_tree_root*sendbuf_extent),
                                          curr_cnt, sendtype,
                                          dst, MPIR_ALLTOALL_TAG, 
                                          ((char *)tmp_buf +
                                           dst_tree_root*sendbuf_extent),
					  sendbuf_extent*(comm_size-dst_tree_root),
                                          sendtype, dst, MPIR_ALLTOALL_TAG, 
                                          comm, &status);
		if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
                
                /* in case of non-power-of-two nodes, less data may be
                   received than specified */
                NMPI_Get_count(&status, sendtype, &last_recv_cnt);
                curr_cnt += last_recv_cnt;
            }
            
            /* if some processes in this process's subtree in this step
               did not have any destination process to communicate with
               because of non-power-of-two, we need to send them the
               result. We use a logarithmic recursive-halfing algorithm
               for this. */
            
            if (dst_tree_root + mask > comm_size) {
                nprocs_completed = comm_size - my_tree_root - mask;
                /* nprocs_completed is the number of processes in this
                   subtree that have all the data. Send data to others
                   in a tree fashion. First find root of current tree
                   that is being divided into two. k is the number of
                   least-significant bits in this process's rank that
                   must be zeroed out to find the rank of the root */ 
                j = mask;
                k = 0;
                while (j) {
                    j >>= 1;
                    k++;
                }
                k--;
                
                tmp_mask = mask >> 1;
                while (tmp_mask) {
                    dst = rank ^ tmp_mask;
                    
                    tree_root = rank >> k;
                    tree_root <<= k;
                    
                    /* send only if this proc has data and destination
                       doesn't have data. at any step, multiple processes
                       can send if they have the data */
                    if ((dst > rank) && 
                        (rank < tree_root + nprocs_completed)
                        && (dst >= tree_root + nprocs_completed)) {
                        /* send the data received in this step above */
                        mpi_errno = MPIC_Send(((char *)tmp_buf +
                                               dst_tree_root*sendbuf_extent),
                                              last_recv_cnt, sendtype,
                                              dst, MPIR_ALLTOALL_TAG,
                                              comm);  
			if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
                    }
                    /* recv only if this proc. doesn't have data and sender
                       has data */
                    else if ((dst < rank) && 
                             (dst < tree_root + nprocs_completed) &&
                             (rank >= tree_root + nprocs_completed)) {
                        mpi_errno = MPIC_Recv(((char *)tmp_buf +
                                               dst_tree_root*sendbuf_extent),
					      sendbuf_extent*(comm_size-dst_tree_root),
                                              sendtype,   
                                              dst, MPIR_ALLTOALL_TAG,
                                              comm, &status); 
			if (mpi_errno) { MPIU_ERR_POP(mpi_errno); }
                        NMPI_Get_count(&status, sendtype, &last_recv_cnt);
                        curr_cnt += last_recv_cnt;
                    }