adio/ad_bgl/ad_bgl_rdcoll.c

/* ---------------------------------------------------------------- */
/* (C)Copyright IBM Corp.  2007, 2008                               */
/* ---------------------------------------------------------------- */
/**
 * \file ad_bgl_rdcoll.c
 * \brief ???
 */

/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
 *
 *   Copyright (C) 1997 University of Chicago.
 *   See COPYRIGHT notice in top-level directory.
 */

#include "adio.h"
#include "adio_extern.h"
#include "ad_bgl.h"
#include "ad_bgl_pset.h"
#include "ad_bgl_aggrs.h"

#ifdef PROFILE
#include "mpe.h"
#endif

#ifdef USE_DBG_LOGGING
  #define RDCOLL_DEBUG 1
#endif
#ifdef AGGREGATION_PROFILE
#include "mpe.h"
#endif

/* prototypes of functions used for collective reads only. */
static void ADIOI_Read_and_exch(ADIO_File fd, void *buf, MPI_Datatype
				datatype, int nprocs,
				int myrank, ADIOI_Access
				*others_req, ADIO_Offset *offset_list,
				ADIO_Offset *len_list, int contig_access_count,
				ADIO_Offset
				min_st_offset, ADIO_Offset fd_size,
				ADIO_Offset *fd_start, ADIO_Offset *fd_end,
				int *buf_idx, int *error_code);
static void ADIOI_R_Exchange_data(ADIO_File fd, void *buf, ADIOI_Flatlist_node
				  *flat_buf, ADIO_Offset *offset_list, ADIO_Offset
				  *len_list, int *send_size, int *recv_size,
				  int *count, int *start_pos,
				  int *partial_send,
				  int *recd_from_proc, int nprocs,
				  int myrank, int
				  buftype_is_contig, int contig_access_count,
				  ADIO_Offset min_st_offset,
				  ADIO_Offset fd_size,
				  ADIO_Offset *fd_start, ADIO_Offset *fd_end,
				  ADIOI_Access *others_req,
				  int iter,
				  MPI_Aint buftype_extent, int *buf_idx);
static void ADIOI_R_Exchange_data_alltoallv(ADIO_File fd, void *buf, ADIOI_Flatlist_node
                                  *flat_buf, ADIO_Offset *offset_list, ADIO_Offset
                                  *len_list, int *send_size, int *recv_size,
                                  int *count, int *start_pos,
                                  int *partial_send,
                                  int *recd_from_proc, int nprocs,
                                  int myrank, int
                                  buftype_is_contig, int contig_access_count,
                                  ADIO_Offset min_st_offset,
                                  ADIO_Offset fd_size,
                                  ADIO_Offset *fd_start, ADIO_Offset *fd_end,
                                  ADIOI_Access *others_req,
                                  int iter,
                                  MPI_Aint buftype_extent, int *buf_idx);
static void ADIOI_Fill_user_buffer(ADIO_File fd, void *buf, ADIOI_Flatlist_node
				   *flat_buf, char **recv_buf, ADIO_Offset
				   *offset_list, ADIO_Offset *len_list,
				   unsigned *recv_size,
				   MPI_Request *requests, MPI_Status *statuses,
				   int *recd_from_proc, int nprocs,
				   int contig_access_count,
				   ADIO_Offset min_st_offset,
				   ADIO_Offset fd_size, ADIO_Offset *fd_start,
				   ADIO_Offset *fd_end,
				   MPI_Aint buftype_extent);

extern void ADIOI_Calc_my_off_len(ADIO_File fd, int bufcount, MPI_Datatype
			    datatype, int file_ptr_type, ADIO_Offset
			    offset, ADIO_Offset **offset_list_ptr, ADIO_Offset
			    **len_list_ptr, ADIO_Offset *start_offset_ptr,
			    ADIO_Offset *end_offset_ptr, int
			   *contig_access_count_ptr);

void ADIOI_BGL_ReadStridedColl(ADIO_File fd, void *buf, int count,
			       MPI_Datatype datatype, int file_ptr_type,
			       ADIO_Offset offset, ADIO_Status *status, int
			       *error_code)
{
/* Uses a generalized version of the extended two-phase method described
   in "An Extended Two-Phase Method for Accessing Sections of
   Out-of-Core Arrays", Rajeev Thakur and Alok Choudhary,
   Scientific Programming, (5)4:301--317, Winter 1996.
   http://www.mcs.anl.gov/home/thakur/ext2ph.ps */

    ADIOI_Access *my_req;
    /* array of nprocs structures, one for each other process in
       whose file domain this process's request lies */

    ADIOI_Access *others_req;
    /* array of nprocs structures, one for each other process
       whose request lies in this process's file domain. */

    int i, filetype_is_contig, nprocs, nprocs_for_coll, myrank;
    int contig_access_count=0, interleave_count = 0, buftype_is_contig;
    int *count_my_req_per_proc, count_my_req_procs, count_others_req_procs;
    ADIO_Offset start_offset, end_offset, orig_fp, fd_size, min_st_offset, off;
    ADIO_Offset *offset_list = NULL, *st_offsets = NULL, *fd_start = NULL,
	*fd_end = NULL, *end_offsets = NULL;
    ADIO_Offset *bgl_offsets0 = NULL, *bgl_offsets = NULL;
    int  ii;
    ADIO_Offset *len_list = NULL;
    int *buf_idx = NULL;
#if BGL_PROFILE
    BGLMPIO_T_CIO_RESET( 0, r )
#endif

#ifdef HAVE_STATUS_SET_BYTES
    int bufsize, size;
#endif

#if 0
/*   From common code - not implemented for bgl. */
    if (fd->hints->cb_pfr != ADIOI_HINT_DISABLE) {
        ADIOI_IOStridedColl (fd, buf, count, ADIOI_READ, datatype,
			file_ptr_type, offset, status, error_code);
        return;
    } */
#endif
#ifdef PROFILE
        MPE_Log_event(13, 0, "start computation");
#endif

    MPI_Comm_size(fd->comm, &nprocs);
    MPI_Comm_rank(fd->comm, &myrank);

    /* number of aggregators, cb_nodes, is stored in the hints */
    nprocs_for_coll = fd->hints->cb_nodes;
    orig_fp = fd->fp_ind;

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 0, 1, 0, BGLMPIO_CIO_LCOMP, BGLMPIO_CIO_LAST )
#endif

    /* only check for interleaving if cb_read isn't disabled */
    if (fd->hints->cb_read != ADIOI_HINT_DISABLE) {
	/* For this process's request, calculate the list of offsets and
	   lengths in the file and determine the start and end offsets. */

	/* Note: end_offset points to the last byte-offset that will be accessed.
	   e.g., if start_offset=0 and 100 bytes to be read, end_offset=99*/

	ADIOI_Calc_my_off_len(fd, count, datatype, file_ptr_type, offset,
			      &offset_list, &len_list, &start_offset,
			      &end_offset, &contig_access_count);

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 1, 1, 1, BGLMPIO_CIO_GATHER, BGLMPIO_CIO_LCOMP )
#endif

#ifdef RDCOLL_DEBUG
    for (i=0; i<contig_access_count; i++) {
	      DBG_FPRINTF(stderr, "rank %d  off %lld  len %lld\n",
			      myrank, offset_list[i], len_list[i]);
    }
#endif

	/* each process communicates its start and end offsets to other
	   processes. The result is an array each of start and end offsets
	   stored in order of process rank. */

	st_offsets   = (ADIO_Offset *) ADIOI_Malloc(nprocs*sizeof(ADIO_Offset));
	end_offsets  = (ADIO_Offset *) ADIOI_Malloc(nprocs*sizeof(ADIO_Offset));

    if (bglmpio_tunegather) {
	    bgl_offsets0 = (ADIO_Offset *) ADIOI_Malloc(2*nprocs*sizeof(ADIO_Offset));
	    bgl_offsets  = (ADIO_Offset *) ADIOI_Malloc(2*nprocs*sizeof(ADIO_Offset));
	    for (ii=0; ii<nprocs; ii++)  {
		bgl_offsets0[ii*2]   = 0;
		bgl_offsets0[ii*2+1] = 0;
	    }
	    bgl_offsets0[myrank*2]   = start_offset;
	    bgl_offsets0[myrank*2+1] =   end_offset;

	MPI_Allreduce( bgl_offsets0, bgl_offsets, nprocs*2, ADIO_OFFSET, MPI_MAX, fd->comm );

	    for (ii=0; ii<nprocs; ii++)  {
		st_offsets [ii] = bgl_offsets[ii*2]  ;
		end_offsets[ii] = bgl_offsets[ii*2+1];
	    }
	    ADIOI_Free( bgl_offsets0 );
	    ADIOI_Free( bgl_offsets  );
    } else {
        MPI_Allgather(&start_offset, 1, ADIO_OFFSET, st_offsets, 1,
                      ADIO_OFFSET, fd->comm);
        MPI_Allgather(&end_offset, 1, ADIO_OFFSET, end_offsets, 1,
                      ADIO_OFFSET, fd->comm);
    }

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 0, 1, 1, BGLMPIO_CIO_PATANA, BGLMPIO_CIO_GATHER )
#endif

	/* are the accesses of different processes interleaved? */
	for (i=1; i<nprocs; i++)
	    if ((st_offsets[i] < end_offsets[i-1]) &&
                (st_offsets[i] <= end_offsets[i]))
                interleave_count++;
	/* This is a rudimentary check for interleaving, but should suffice
	   for the moment. */
    }

    ADIOI_Datatype_iscontig(datatype, &buftype_is_contig);

    if (fd->hints->cb_read == ADIOI_HINT_DISABLE
	|| (!interleave_count && (fd->hints->cb_read == ADIOI_HINT_AUTO)))
    {
	/* don't do aggregation */
	if (fd->hints->cb_read != ADIOI_HINT_DISABLE) {
	    ADIOI_Free(offset_list);
	    ADIOI_Free(len_list);
	    ADIOI_Free(st_offsets);
	    ADIOI_Free(end_offsets);
	}

	fd->fp_ind = orig_fp;
	ADIOI_Datatype_iscontig(fd->filetype, &filetype_is_contig);

	if (buftype_is_contig && filetype_is_contig) {
	    if (file_ptr_type == ADIO_EXPLICIT_OFFSET) {
		off = fd->disp + (ADIO_Offset)(fd->etype_size) * offset;
		ADIO_ReadContig(fd, buf, count, datatype, ADIO_EXPLICIT_OFFSET,
                       off, status, error_code);
	    }
	    else ADIO_ReadContig(fd, buf, count, datatype, ADIO_INDIVIDUAL,
                       0, status, error_code);
	}
	else ADIO_ReadStrided(fd, buf, count, datatype, file_ptr_type,
                       offset, status, error_code);

	return;
    }

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 1, 1, 1, BGLMPIO_CIO_FD_PART, BGLMPIO_CIO_PATANA )
#endif

    /* We're going to perform aggregation of I/O.  Here we call
     * ADIOI_Calc_file_domains() to determine what processes will handle I/O
     * to what regions.  We pass nprocs_for_coll into this function; it is
     * used to determine how many processes will perform I/O, which is also
     * the number of regions into which the range of bytes must be divided.
     * These regions are called "file domains", or FDs.
     *
     * When this function returns, fd_start, fd_end, fd_size, and
     * min_st_offset will be filled in.  fd_start holds the starting byte
     * location for each file domain.  fd_end holds the ending byte location.
     * min_st_offset holds the minimum byte location that will be accessed.
     *
     * Both fd_start[] and fd_end[] are indexed by an aggregator number; this
     * needs to be mapped to an actual rank in the communicator later.
     *
     */
    if (bglmpio_tuneblocking)
    ADIOI_BGL_GPFS_Calc_file_domains(st_offsets, end_offsets, nprocs,
			    nprocs_for_coll, &min_st_offset,
			    &fd_start, &fd_end, &fd_size, fd->fs_ptr);
    else
    ADIOI_Calc_file_domains(st_offsets, end_offsets, nprocs,
			    nprocs_for_coll, &min_st_offset,
			    &fd_start, &fd_end,
			    fd->hints->min_fdomain_size, &fd_size,
			    fd->hints->striping_unit);

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 0, 1, 1, BGLMPIO_CIO_MYREQ, BGLMPIO_CIO_FD_PART )
#endif

    /* calculate where the portions of the access requests of this process
     * are located in terms of the file domains.  this could be on the same
     * process or on other processes.  this function fills in:
     * count_my_req_procs - number of processes (including this one) for which
     *     this process has requests in their file domain
     * count_my_req_per_proc - count of requests for each process, indexed
     *     by rank of the process
     * my_req[] - array of data structures describing the requests to be
     *     performed by each process (including self).  indexed by rank.
     * buf_idx[] - array of locations into which data can be directly moved;
     *     this is only valid for contiguous buffer case
     */
    if (bglmpio_tuneblocking)
    ADIOI_BGL_Calc_my_req(fd, offset_list, len_list, contig_access_count,
		      min_st_offset, fd_start, fd_end, fd_size,
		      nprocs, &count_my_req_procs,
		      &count_my_req_per_proc, &my_req,
		      &buf_idx);
    else
    ADIOI_Calc_my_req(fd, offset_list, len_list, contig_access_count,
		      min_st_offset, fd_start, fd_end, fd_size,
		      nprocs, &count_my_req_procs,
		      &count_my_req_per_proc, &my_req,
		      &buf_idx);

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 1, 1, 1, BGLMPIO_CIO_OTHREQ, BGLMPIO_CIO_MYREQ )
#endif

    /* perform a collective communication in order to distribute the
     * data calculated above.  fills in the following:
     * count_others_req_procs - number of processes (including this
     *     one) which have requests in this process's file domain.
     * count_others_req_per_proc[] - number of separate contiguous
     *     requests from proc i lie in this process's file domain.
     */
    if (bglmpio_tuneblocking)
    ADIOI_BGL_Calc_others_req(fd, count_my_req_procs,
			  count_my_req_per_proc, my_req,
			  nprocs, myrank, &count_others_req_procs,
			  &others_req);

    else
    ADIOI_Calc_others_req(fd, count_my_req_procs,
			  count_my_req_per_proc, my_req,
			  nprocs, myrank, &count_others_req_procs,
			  &others_req);

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 1, 1, 1, BGLMPIO_CIO_DEXCH, BGLMPIO_CIO_OTHREQ )
#endif

    /* my_req[] and count_my_req_per_proc aren't needed at this point, so
     * let's free the memory
     */
    ADIOI_Free(count_my_req_per_proc);
    for (i=0; i<nprocs; i++) {
	if (my_req[i].count) {
	    ADIOI_Free(my_req[i].offsets);
	    ADIOI_Free(my_req[i].lens);
	}
    }
    ADIOI_Free(my_req);


    /* read data in sizes of no more than ADIOI_Coll_bufsize,
     * communicate, and fill user buf.
     */
    ADIOI_Read_and_exch(fd, buf, datatype, nprocs, myrank,
                        others_req, offset_list,
			len_list, contig_access_count, min_st_offset,
			fd_size, fd_start, fd_end, buf_idx, error_code);

#if BGL_PROFILE
    BGLMPIO_T_CIO_SET_GET( 0, r, 1, 0, 1, BGLMPIO_CIO_LAST, BGLMPIO_CIO_T_DEXCH )
    BGLMPIO_T_CIO_SET_GET( 0, r, 0, 0, 1, BGLMPIO_CIO_LAST, BGLMPIO_CIO_T_MPIO_CRW )

    BGLMPIO_T_CIO_REPORT( 0, r, fd, myrank )
#endif

    if (!buftype_is_contig) ADIOI_Delete_flattened(datatype);

    /* free all memory allocated for collective I/O */
    for (i=0; i<nprocs; i++) {
	if (others_req[i].count) {
	    ADIOI_Free(others_req[i].offsets);
	    ADIOI_Free(others_req[i].lens);
	    ADIOI_Free(others_req[i].mem_ptrs);
	}
    }
    ADIOI_Free(others_req);

    ADIOI_Free(buf_idx);
    ADIOI_Free(offset_list);
    ADIOI_Free(len_list);
    ADIOI_Free(st_offsets);
    ADIOI_Free(end_offsets);
    ADIOI_Free(fd_start);
    ADIOI_Free(fd_end);

#ifdef HAVE_STATUS_SET_BYTES
    MPI_Type_size(datatype, &size);
    bufsize = size * count;
    MPIR_Status_set_bytes(status, datatype, bufsize);
/* This is a temporary way of filling in status. The right way is to
   keep track of how much data was actually read and placed in buf
   during collective I/O. */
#endif

    fd->fp_sys_posn = -1;   /* set it to null. */
}

static void ADIOI_Read_and_exch(ADIO_File fd, void *buf, MPI_Datatype
			 datatype, int nprocs,
			 int myrank, ADIOI_Access
			 *others_req, ADIO_Offset *offset_list,
			 ADIO_Offset *len_list, int contig_access_count, ADIO_Offset
                         min_st_offset, ADIO_Offset fd_size,
			 ADIO_Offset *fd_start, ADIO_Offset *fd_end,
                         int *buf_idx, int *error_code)
{
/* Read in sizes of no more than coll_bufsize, an info parameter.
   Send data to appropriate processes.
   Place recd. data in user buf.
   The idea is to reduce the amount of extra memory required for
   collective I/O. If all data were read all at once, which is much
   easier, it would require temp space more than the size of user_buf,
   which is often unacceptable. For example, to read a distributed
   array from a file, where each local array is 8Mbytes, requiring
   at least another 8Mbytes of temp space is unacceptable. */

    int i, j, m, ntimes, max_ntimes, buftype_is_contig;
    ADIO_Offset st_loc=-1, end_loc=-1, off, done, real_off, req_off;
    char *read_buf = NULL, *tmp_buf;
    int *curr_offlen_ptr, *count, *send_size, *recv_size;
    int *partial_send, *recd_from_proc, *start_pos;
    /* Not convinced end_loc-st_loc couldn't be > int, so make these offsets*/
    ADIO_Offset real_size, size, for_curr_iter, for_next_iter;
    int req_len, flag, rank;
    MPI_Status status;
    ADIOI_Flatlist_node *flat_buf=NULL;
    MPI_Aint buftype_extent;
    int coll_bufsize;
#ifdef RDCOLL_DEBUG
    int iii;
#endif
    *error_code = MPI_SUCCESS;  /* changed below if error */
    /* only I/O errors are currently reported */

/* calculate the number of reads of size coll_bufsize
   to be done by each process and the max among all processes.
   That gives the no. of communication phases as well.
   coll_bufsize is obtained from the hints object. */

    coll_bufsize = fd->hints->cb_buffer_size;

    /* grab some initial values for st_loc and end_loc */
    for (i=0; i < nprocs; i++) {
	if (others_req[i].count) {
	    st_loc = others_req[i].offsets[0];
	    end_loc = others_req[i].offsets[0];
	    break;
	}
    }

    /* now find the real values */
    for (i=0; i < nprocs; i++)
	for (j=0; j<others_req[i].count; j++) {
	    st_loc = ADIOI_MIN(st_loc, others_req[i].offsets[j]);
	    end_loc = ADIOI_MAX(end_loc, (others_req[i].offsets[j]
					  + others_req[i].lens[j] - 1));
	}

    /* calculate ntimes, the number of times this process must perform I/O
     * operations in order to complete all the requests it has received.
     * the need for multiple I/O operations comes from the restriction that
     * we only use coll_bufsize bytes of memory for internal buffering.
     */
    if ((st_loc==-1) && (end_loc==-1)) {
	/* this process does no I/O. */
	ntimes = 0;
    }
    else {
	/* ntimes=ceiling_div(end_loc - st_loc + 1, coll_bufsize)*/
	ntimes = (int) ((end_loc - st_loc + coll_bufsize)/coll_bufsize);
    }

    MPI_Allreduce(&ntimes, &max_ntimes, 1, MPI_INT, MPI_MAX, fd->comm);

    if (ntimes) read_buf = (char *) ADIOI_Malloc(coll_bufsize);

    curr_offlen_ptr = (int *) ADIOI_Calloc(nprocs, sizeof(int));
    /* its use is explained below. calloc initializes to 0. */

    count = (int *) ADIOI_Malloc(nprocs * sizeof(int));
    /* to store count of how many off-len pairs per proc are satisfied
       in an iteration. */

    partial_send = (int *) ADIOI_Calloc(nprocs, sizeof(int));
    /* if only a portion of the last off-len pair is sent to a process
       in a particular iteration, the length sent is stored here.
       calloc initializes to 0. */

    send_size = (int *) ADIOI_Malloc(nprocs * sizeof(int));
    /* total size of data to be sent to each proc. in an iteration */

    recv_size = (int *) ADIOI_Malloc(nprocs * sizeof(int));
    /* total size of data to be recd. from each proc. in an iteration.
       Of size nprocs so that I can use MPI_Alltoall later. */

    recd_from_proc = (int *) ADIOI_Calloc(nprocs, sizeof(int));
    /* amount of data recd. so far from each proc. Used in
       ADIOI_Fill_user_buffer. initialized to 0 here. */

    start_pos = (int *) ADIOI_Malloc(nprocs*sizeof(int));
    /* used to store the starting value of curr_offlen_ptr[i] in
       this iteration */

    ADIOI_Datatype_iscontig(datatype, &buftype_is_contig);
    if (!buftype_is_contig) {
	ADIOI_Flatten_datatype(datatype);
	flat_buf = ADIOI_Flatlist;
        while (flat_buf->type != datatype) flat_buf = flat_buf->next;
    }
    MPI_Type_extent(datatype, &buftype_extent);

    done = 0;
    off = st_loc;
    for_curr_iter = for_next_iter = 0;

    MPI_Comm_rank(fd->comm, &rank);

#ifdef PROFILE
        MPE_Log_event(14, 0, "end computation");
#endif

    for (m=0; m<ntimes; m++) {
       /* read buf of size coll_bufsize (or less) */
       /* go through all others_req and check if any are satisfied
          by the current read */

       /* since MPI guarantees that displacements in filetypes are in
          monotonically nondecreasing order, I can maintain a pointer
	  (curr_offlen_ptr) to
          current off-len pair for each process in others_req and scan
          further only from there. There is still a problem of filetypes
          such as:  (1, 2, 3 are not process nos. They are just numbers for
          three chunks of data, specified by a filetype.)

                   1  -------!--
                   2    -----!----
                   3       --!-----

          where ! indicates where the current read_size limitation cuts
          through the filetype.  I resolve this by reading up to !, but
          filling the communication buffer only for 1. I copy the portion
          left over for 2 into a tmp_buf for use in the next
	  iteration. i.e., 2 and 3 will be satisfied in the next
	  iteration. This simplifies filling in the user's buf at the
	  other end, as only one off-len pair with incomplete data
	  will be sent. I also don't need to send the individual
	  offsets and lens along with the data, as the data is being
	  sent in a particular order. */

          /* off = start offset in the file for the data actually read in
                   this iteration
             size = size of data read corresponding to off
             real_off = off minus whatever data was retained in memory from
                  previous iteration for cases like 2, 3 illustrated above
             real_size = size plus the extra corresponding to real_off
             req_off = off in file for a particular contiguous request
                       minus what was satisfied in previous iteration
             req_size = size corresponding to req_off */

#ifdef PROFILE
        MPE_Log_event(13, 0, "start computation");
#endif
	size = ADIOI_MIN((unsigned)coll_bufsize, end_loc-st_loc+1-done);
	real_off = off - for_curr_iter;
	real_size = size + for_curr_iter;

	for (i=0; i<nprocs; i++) count[i] = send_size[i] = 0;
	for_next_iter = 0;

	for (i=0; i<nprocs; i++) {
#ifdef RDCOLL_DEBUG
	    DBG_FPRINTF(stderr, "rank %d, i %d, others_count %d\n", rank, i, others_req[i].count);
#endif
	    if (others_req[i].count) {
		start_pos[i] = curr_offlen_ptr[i];
		for (j=curr_offlen_ptr[i]; j<others_req[i].count;
		     j++) {
		    if (partial_send[i]) {
			/* this request may have been partially
			   satisfied in the previous iteration. */
			req_off = others_req[i].offsets[j] +
			    partial_send[i];
                        req_len = others_req[i].lens[j] -
			    partial_send[i];
			partial_send[i] = 0;
			/* modify the off-len pair to reflect this change */
			others_req[i].offsets[j] = req_off;
			others_req[i].lens[j] = req_len;
		    }
		    else {
			req_off = others_req[i].offsets[j];
                        req_len = others_req[i].lens[j];
		    }
		    if (req_off < real_off + real_size) {
			count[i]++;
      ADIOI_Assert((((ADIO_Offset)(MPIR_Upint)read_buf)+req_off-real_off) == (ADIO_Offset)(MPIR_Upint)(read_buf+req_off-real_off));
			MPI_Address(read_buf+req_off-real_off,
                               &(others_req[i].mem_ptrs[j]));
      ADIOI_Assert((real_off + real_size - req_off) == (int)(real_off + real_size - req_off));
			send_size[i] += (int)(ADIOI_MIN(real_off + real_size - req_off,
                                      (ADIO_Offset)(unsigned)req_len));

			if (real_off+real_size-req_off < (ADIO_Offset)(unsigned)req_len) {
			    partial_send[i] = (int) (real_off + real_size - req_off);
			    if ((j+1 < others_req[i].count) &&
                                 (others_req[i].offsets[j+1] <
                                     real_off+real_size)) {
				/* this is the case illustrated in the
				   figure above. */
				for_next_iter = ADIOI_MAX(for_next_iter,
					  real_off + real_size - others_req[i].offsets[j+1]);
				/* max because it must cover requests
				   from different processes */
			    }
			    break;
			}
		    }
		    else break;
		}
		curr_offlen_ptr[i] = j;
	    }
	}

	flag = 0;
	for (i=0; i<nprocs; i++)
	    if (count[i]) flag = 1;

#ifdef PROFILE
        MPE_Log_event(14, 0, "end computation");
#endif
	if (flag) {
      ADIOI_Assert(size == (int)size);
	    ADIO_ReadContig(fd, read_buf+for_curr_iter, (int)size, MPI_BYTE,
			    ADIO_EXPLICIT_OFFSET, off, &status, error_code);
#ifdef RDCOLL_DEBUG
	    DBG_FPRINTF(stderr, "\tread_coll: 700, data read [%lld] = ", size );
	    for (iii=0; iii<size && iii<80; iii++) { DBGV_FPRINTF(stderr, "%3d,", *((unsigned char *)read_buf + for_curr_iter + iii) ); }
	    DBG_FPRINTF(stderr, "\n" );
#endif

	    if (*error_code != MPI_SUCCESS) return;
	}

	for_curr_iter = for_next_iter;

#ifdef PROFILE
        MPE_Log_event(7, 0, "start communication");
#endif
	if (bglmpio_comm == 1)
	ADIOI_R_Exchange_data(fd, buf, flat_buf, offset_list, len_list,
			    send_size, recv_size, count,
       			    start_pos, partial_send, recd_from_proc, nprocs,
			    myrank,
			    buftype_is_contig, contig_access_count,
			    min_st_offset, fd_size, fd_start, fd_end,
			    others_req,
                            m, buftype_extent, buf_idx);
        else
	if (bglmpio_comm == 0) {
        ADIOI_R_Exchange_data_alltoallv(fd, buf, flat_buf, offset_list, len_list,
                            send_size, recv_size, count,
                            start_pos, partial_send, recd_from_proc, nprocs,
                            myrank,
                            buftype_is_contig, contig_access_count,
                            min_st_offset, fd_size, fd_start, fd_end,
                            others_req,
                            m, buftype_extent, buf_idx);
	}


#ifdef PROFILE
        MPE_Log_event(8, 0, "end communication");
#endif

	if (for_next_iter) {
	    tmp_buf = (char *) ADIOI_Malloc(for_next_iter);
      ADIOI_Assert((((ADIO_Offset)(MPIR_Upint)read_buf)+real_size-for_next_iter) == (ADIO_Offset)(MPIR_Upint)(read_buf+real_size-for_next_iter));
      ADIOI_Assert((for_next_iter+coll_bufsize) == (size_t)(for_next_iter+coll_bufsize));
	    memcpy(tmp_buf, read_buf+real_size-for_next_iter, for_next_iter);
	    ADIOI_Free(read_buf);
	    read_buf = (char *) ADIOI_Malloc(for_next_iter+coll_bufsize);
	    memcpy(read_buf, tmp_buf, for_next_iter);
	    ADIOI_Free(tmp_buf);
	}

	off += size;
	done += size;
    }

    for (i=0; i<nprocs; i++) count[i] = send_size[i] = 0;
#ifdef PROFILE
        MPE_Log_event(7, 0, "start communication");
#endif
    for (m=ntimes; m<max_ntimes; m++)
/* nothing to send, but check for recv. */

	if (bglmpio_comm == 1)
	ADIOI_R_Exchange_data(fd, buf, flat_buf, offset_list, len_list,
			    send_size, recv_size, count,
			    start_pos, partial_send, recd_from_proc, nprocs,
			    myrank,
			    buftype_is_contig, contig_access_count,
			    min_st_offset, fd_size, fd_start, fd_end,
			    others_req, m,
                            buftype_extent, buf_idx);
        else    /* strncmp( env_switch, "alltoall", 8 ) == 0 */
	if (bglmpio_comm == 0)
        ADIOI_R_Exchange_data_alltoallv(fd, buf, flat_buf, offset_list, len_list,
                            send_size, recv_size, count,
                            start_pos, partial_send, recd_from_proc, nprocs,
                            myrank,
                            buftype_is_contig, contig_access_count,
                            min_st_offset, fd_size, fd_start, fd_end,
                            others_req,
                            m, buftype_extent, buf_idx);

#ifdef PROFILE
        MPE_Log_event(8, 0, "end communication");
#endif

    if (ntimes) ADIOI_Free(read_buf);
    ADIOI_Free(curr_offlen_ptr);
    ADIOI_Free(count);
    ADIOI_Free(partial_send);
    ADIOI_Free(send_size);
    ADIOI_Free(recv_size);
    ADIOI_Free(recd_from_proc);
    ADIOI_Free(start_pos);
}

static void ADIOI_R_Exchange_data(ADIO_File fd, void *buf, ADIOI_Flatlist_node
			 *flat_buf, ADIO_Offset *offset_list, ADIO_Offset
                         *len_list, int *send_size, int *recv_size,
			 int *count, int *start_pos, int *partial_send,
			 int *recd_from_proc, int nprocs,
			 int myrank, int
			 buftype_is_contig, int contig_access_count,
			 ADIO_Offset min_st_offset, ADIO_Offset fd_size,
			 ADIO_Offset *fd_start, ADIO_Offset *fd_end,
			 ADIOI_Access *others_req,
                         int iter, MPI_Aint buftype_extent, int *buf_idx)
{
    int i, j, k=0, tmp=0, nprocs_recv, nprocs_send;
    char **recv_buf = NULL;
    MPI_Request *requests;
    MPI_Datatype send_type;
    MPI_Status *statuses;

/* exchange send_size info so that each process knows how much to
   receive from whom and how much memory to allocate. */

    MPI_Alltoall(send_size, 1, MPI_INT, recv_size, 1, MPI_INT, fd->comm);

    nprocs_recv = 0;
    for (i=0; i < nprocs; i++) if (recv_size[i]) nprocs_recv++;

    nprocs_send = 0;
    for (i=0; i<nprocs; i++) if (send_size[i]) nprocs_send++;

    requests = (MPI_Request *)
	ADIOI_Malloc((nprocs_send+nprocs_recv+1)*sizeof(MPI_Request));
/* +1 to avoid a 0-size malloc */

/* post recvs. if buftype_is_contig, data can be directly recd. into
   user buf at location given by buf_idx. else use recv_buf. */

#ifdef AGGREGATION_PROFILE
    MPE_Log_event (5032, 0, NULL);
#endif

    if (buftype_is_contig) {
	j = 0;
	for (i=0; i < nprocs; i++)
	    if (recv_size[i]) {
		MPI_Irecv(((char *) buf) + buf_idx[i], recv_size[i],
		  MPI_BYTE, i, myrank+i+100*iter, fd->comm, requests+j);
		j++;
		buf_idx[i] += recv_size[i];
	    }
    }
    else {
/* allocate memory for recv_buf and post receives */
	recv_buf = (char **) ADIOI_Malloc(nprocs * sizeof(char*));
	for (i=0; i < nprocs; i++)
	    if (recv_size[i]) recv_buf[i] =
                                  (char *) ADIOI_Malloc(recv_size[i]);

	    j = 0;
	    for (i=0; i < nprocs; i++)
		if (recv_size[i]) {
		    MPI_Irecv(recv_buf[i], recv_size[i], MPI_BYTE, i,
			      myrank+i+100*iter, fd->comm, requests+j);
		    j++;
#ifdef RDCOLL_DEBUG
		    DBG_FPRINTF(stderr, "node %d, recv_size %d, tag %d \n",
		       myrank, recv_size[i], myrank+i+100*iter);
#endif
		}
    }

/* create derived datatypes and send data */

    j = 0;
    for (i=0; i<nprocs; i++) {
	if (send_size[i]) {
/* take care if the last off-len pair is a partial send */
	    if (partial_send[i]) {
		k = start_pos[i] + count[i] - 1;
		tmp = others_req[i].lens[k];
		others_req[i].lens[k] = partial_send[i];
	    }
	    MPI_Type_hindexed(count[i],
                 &(others_req[i].lens[start_pos[i]]),
	            &(others_req[i].mem_ptrs[start_pos[i]]),
			 MPI_BYTE, &send_type);
	    /* absolute displacement; use MPI_BOTTOM in send */
	    MPI_Type_commit(&send_type);
	    MPI_Isend(MPI_BOTTOM, 1, send_type, i, myrank+i+100*iter,
		      fd->comm, requests+nprocs_recv+j);
	    MPI_Type_free(&send_type);
	    if (partial_send[i]) others_req[i].lens[k] = tmp;
	    j++;
	}
    }

    statuses = (MPI_Status *) ADIOI_Malloc((nprocs_send+nprocs_recv+1) * \
                                     sizeof(MPI_Status));
     /* +1 to avoid a 0-size malloc */

    /* wait on the receives */
    if (nprocs_recv) {
#ifdef NEEDS_MPI_TEST
	j = 0;
	while (!j) MPI_Testall(nprocs_recv, requests, &j, statuses);
#else
	MPI_Waitall(nprocs_recv, requests, statuses);
#endif

	/* if noncontiguous, to the copies from the recv buffers */
	if (!buftype_is_contig)
	    ADIOI_Fill_user_buffer(fd, buf, flat_buf, recv_buf,
				   offset_list, len_list, (unsigned*)recv_size,
				   requests, statuses, recd_from_proc,
				   nprocs, contig_access_count,
				   min_st_offset, fd_size, fd_start, fd_end,
				   buftype_extent);
    }

    /* wait on the sends*/
    MPI_Waitall(nprocs_send, requests+nprocs_recv, statuses+nprocs_recv);

    ADIOI_Free(statuses);
    ADIOI_Free(requests);

    if (!buftype_is_contig) {
	for (i=0; i < nprocs; i++)
	    if (recv_size[i]) ADIOI_Free(recv_buf[i]);
	ADIOI_Free(recv_buf);
    }
#ifdef AGGREGATION_PROFILE
    MPE_Log_event (5033, 0, NULL);
#endif
}

#define ADIOI_BUF_INCR \
{ \
    while (buf_incr) { \
	size_in_buf = ADIOI_MIN(buf_incr, flat_buf_sz); \
	user_buf_idx += size_in_buf; \
	flat_buf_sz -= size_in_buf; \
	if (!flat_buf_sz) { \
            if (flat_buf_idx < (flat_buf->count - 1)) flat_buf_idx++; \
            else { \
                flat_buf_idx = 0; \
                n_buftypes++; \
            } \
            user_buf_idx = flat_buf->indices[flat_buf_idx] + \
                              (ADIO_Offset)n_buftypes*(ADIO_Offset)buftype_extent; \
	    flat_buf_sz = flat_buf->blocklens[flat_buf_idx]; \
	} \
	buf_incr -= size_in_buf; \
    } \
}


#define ADIOI_BUF_COPY \
{ \
    while (size) { \
	size_in_buf = ADIOI_MIN(size, flat_buf_sz); \
  ADIOI_Assert((((ADIO_Offset)(MPIR_Upint)buf) + user_buf_idx) == (ADIO_Offset)(MPIR_Upint)(buf + user_buf_idx)); \
  ADIOI_Assert(size_in_buf == (size_t)size_in_buf); \
	memcpy(((char *) buf) + user_buf_idx, \
	       &(recv_buf[p][recv_buf_idx[p]]), size_in_buf); \
	recv_buf_idx[p] += size_in_buf; /* already tested (size_t)size_in_buf*/ \
	user_buf_idx += size_in_buf; \
	flat_buf_sz -= size_in_buf; \
	if (!flat_buf_sz) { \
            if (flat_buf_idx < (flat_buf->count - 1)) flat_buf_idx++; \
            else { \
                flat_buf_idx = 0; \
                n_buftypes++; \
            } \
            user_buf_idx = flat_buf->indices[flat_buf_idx] + \
                              (ADIO_Offset)n_buftypes*(ADIO_Offset)buftype_extent; \
	    flat_buf_sz = flat_buf->blocklens[flat_buf_idx]; \
	} \
	size -= size_in_buf; \
	buf_incr -= size_in_buf; \
    } \
    ADIOI_BUF_INCR \
}

static void ADIOI_Fill_user_buffer(ADIO_File fd, void *buf, ADIOI_Flatlist_node
				   *flat_buf, char **recv_buf, ADIO_Offset
				   *offset_list, ADIO_Offset *len_list,
				   unsigned *recv_size,
				   MPI_Request *requests, MPI_Status *statuses,
				   int *recd_from_proc, int nprocs,
				   int contig_access_count,
				   ADIO_Offset min_st_offset,
				   ADIO_Offset fd_size, ADIO_Offset *fd_start,
				   ADIO_Offset *fd_end,
				   MPI_Aint buftype_extent)
{

/* this function is only called if buftype is not contig */

    int i, p, flat_buf_idx;
    ADIO_Offset flat_buf_sz, size_in_buf, buf_incr, size;
    int n_buftypes;
    ADIO_Offset off, len, rem_len, user_buf_idx;
    /* Not sure unsigned is necessary, but it makes the math safer */
    unsigned *curr_from_proc, *done_from_proc, *recv_buf_idx;

    ADIOI_UNREFERENCED_ARG(requests);
    ADIOI_UNREFERENCED_ARG(statuses);

/*  curr_from_proc[p] = amount of data recd from proc. p that has already
                        been accounted for so far
    done_from_proc[p] = amount of data already recd from proc. p and
                        filled into user buffer in previous iterations
    user_buf_idx = current location in user buffer
    recv_buf_idx[p] = current location in recv_buf of proc. p  */
    curr_from_proc = (unsigned *) ADIOI_Malloc(nprocs * sizeof(unsigned));
    done_from_proc = (unsigned *) ADIOI_Malloc(nprocs * sizeof(unsigned));
    recv_buf_idx   = (unsigned *) ADIOI_Malloc(nprocs * sizeof(unsigned));

    for (i=0; i < nprocs; i++) {
	recv_buf_idx[i] = curr_from_proc[i] = 0;
	done_from_proc[i] = recd_from_proc[i];
    }

    user_buf_idx = flat_buf->indices[0];
    flat_buf_idx = 0;
    n_buftypes = 0;
    flat_buf_sz = flat_buf->blocklens[0];

    /* flat_buf_idx = current index into flattened buftype
       flat_buf_sz = size of current contiguous component in
                flattened buf */

    for (i=0; i<contig_access_count; i++) {
	off     = offset_list[i];
	rem_len = len_list[i];

	/* this request may span the file domains of more than one process */
	while (rem_len > 0) {
	    len = rem_len;
	    /* NOTE: len value is modified by ADIOI_Calc_aggregator() to be no
	     * longer than the single region that processor "p" is responsible
	     * for.
	     */
	    p = ADIOI_BGL_Calc_aggregator(fd,
				      off,
				      min_st_offset,
				      &len,
				      fd_size,
				      fd_start,
				      fd_end);

	    if (recv_buf_idx[p] < recv_size[p]) {
		if (curr_from_proc[p]+len > done_from_proc[p]) {
		    if (done_from_proc[p] > curr_from_proc[p]) {
			size = ADIOI_MIN(curr_from_proc[p] + len -
			      done_from_proc[p], recv_size[p]-recv_buf_idx[p]);
			buf_incr = done_from_proc[p] - curr_from_proc[p];
			ADIOI_BUF_INCR
			buf_incr = curr_from_proc[p]+len-done_from_proc[p];
      ADIOI_Assert((done_from_proc[p] + size) == (unsigned)((ADIO_Offset)done_from_proc[p] + size));
			curr_from_proc[p] = done_from_proc[p] + size;
			ADIOI_BUF_COPY
		    }
		    else {
			size = ADIOI_MIN(len,recv_size[p]-recv_buf_idx[p]);
			buf_incr = len;
      ADIOI_Assert((curr_from_proc[p] + size) == (unsigned)((ADIO_Offset)curr_from_proc[p] + size));
			curr_from_proc[p] += (unsigned) size;
			ADIOI_BUF_COPY
		    }
		}
		else {
        ADIOI_Assert((curr_from_proc[p] + len) == (unsigned)((ADIO_Offset)curr_from_proc[p] + len));
		    curr_from_proc[p] += (unsigned) len;
		    buf_incr = len;
		    ADIOI_BUF_INCR
		}
	    }
	    else {
		buf_incr = len;
		ADIOI_BUF_INCR
	    }
	    off     += len;
	    rem_len -= len;
	}
    }
    for (i=0; i < nprocs; i++)
	if (recv_size[i]) recd_from_proc[i] = curr_from_proc[i];

    ADIOI_Free(curr_from_proc);
    ADIOI_Free(done_from_proc);
    ADIOI_Free(recv_buf_idx);
}

static void ADIOI_R_Exchange_data_alltoallv(
                ADIO_File fd, void *buf, ADIOI_Flatlist_node
                *flat_buf, ADIO_Offset *offset_list, ADIO_Offset
                *len_list, int *send_size, int *recv_size,
                int *count, int *start_pos, int *partial_send,
                int *recd_from_proc, int nprocs,
                int myrank, int
                buftype_is_contig, int contig_access_count,
                ADIO_Offset min_st_offset, ADIO_Offset fd_size,
                ADIO_Offset *fd_start, ADIO_Offset *fd_end,
                ADIOI_Access *others_req,
                int iter, MPI_Aint buftype_extent, int *buf_idx)
{
    int i, j, k=0, tmp=0, nprocs_recv, nprocs_send;
    char **recv_buf = NULL;
    MPI_Request *requests=NULL;
    MPI_Status *statuses=NULL;
    int rtail, stail;
    char *sbuf_ptr, *from_ptr;
    int  len;
    int  *sdispls, *rdispls;
    char *all_recv_buf, *all_send_buf;

  /* exchange send_size info so that each process knows how much to
     receive from whom and how much memory to allocate. */
    MPI_Alltoall(send_size, 1, MPI_INT, recv_size, 1, MPI_INT, fd->comm);

    nprocs_recv = 0;
    for (i=0; i<nprocs; i++) if (recv_size[i]) { nprocs_recv++; break; }

    nprocs_send = 0;
    for (i=0; i<nprocs; i++) if (send_size[i]) { nprocs_send++; break; }

  /* receiver side data structures */
    rdispls = (int *) ADIOI_Malloc( nprocs * sizeof(int) );
    rtail = 0;
    for (i=0; i<nprocs; i++) { rdispls[i] = rtail; rtail += recv_size[i]; }

        /* data buffer */
    all_recv_buf = (char *) ADIOI_Malloc( rtail );
    recv_buf = (char **) ADIOI_Malloc(nprocs * sizeof(char *));
    for (i=0; i<nprocs; i++) { recv_buf[i] = all_recv_buf + rdispls[i]; }

  /* sender side data structures */
    sdispls = (int *) ADIOI_Malloc( nprocs * sizeof(int) );
    stail = 0;
    for (i=0; i<nprocs; i++) { sdispls[i] = stail; stail += send_size[i]; }

        /* data buffer */
    all_send_buf = (char *) ADIOI_Malloc( stail );
    for (i=0; i<nprocs; i++)
    {
        if (send_size[i]) {
	    if (partial_send[i]) {
		k = start_pos[i] + count[i] - 1;
		tmp = others_req[i].lens[k];
		others_req[i].lens[k] = partial_send[i];
	    }
            sbuf_ptr = all_send_buf + sdispls[i];
            for (j=0; j<count[i]; j++) {
                ADIOI_ENSURE_AINT_FITS_IN_PTR( others_req[i].mem_ptrs[ start_pos[i]+j ]);
                from_ptr = (char *) ADIOI_AINT_CAST_TO_VOID_PTR ( others_req[i].mem_ptrs[ start_pos[i]+j ] );
                len      =           others_req[i].lens[     start_pos[i]+j ]  ;
                memcpy( sbuf_ptr, from_ptr, len );
                sbuf_ptr += len;
            }
	    if (partial_send[i]) others_req[i].lens[k] = tmp;
        }
    }

#if RDCOLL_DEBUG
    DBG_FPRINTF(stderr, "\tsend_size = [%d]%2d,",0,send_size[0]);
    for (i=1; i<nprocs; i++) if(send_size[i-1]!=send_size[i]){ DBG_FPRINTF(stderr, "\t\t[%d]%2d,", i,send_size[i] ); }
    DBG_FPRINTF(stderr, "\trecv_size =  [%d]%2d,",0,recv_size[0]);
    for (i=1; i<nprocs; i++) if(recv_size[i-1]!=recv_size[i]){ DBG_FPRINTF(stderr, "\t\t[%d]%2d,", i,recv_size[i] ); }
    DBG_FPRINTF(stderr, "\tsdispls   =  [%d]%2d,",0,sdispls[0]);
    for (i=1; i<nprocs; i++) if(sdispls[i-1]!=sdispls[i]){ DBG_FPRINTF(stderr, "\t\t[%d]%2d,", i,sdispls  [i] ); }
    DBG_FPRINTF(stderr, "\trdispls   =  [%d]%2d,",0,rdispls[0]);
    for (i=1; i<nprocs; i++) if(rdispls[i-1]!=rdispls[i]){ DBG_FPRINTF(stderr, "\t\t[%d]%2d,", i,rdispls  [i] ); }
    DBG_FPRINTF(stderr, "\ttails = %4d, %4d\n", stail, rtail );
    if (nprocs_send) {
    DBG_FPRINTF(stderr, "\tall_send_buf =  [%d]%2d,",0,all_send_buf[0]);
    for (i=1; i<nprocs; i++) if(all_send_buf[(i-1)*131072]!=all_send_buf[i*131072]){ DBG_FPRINTF(stderr, "\t\t[%d]%2d,", i, all_send_buf  [i*131072] ); }
    }
#endif

  /* alltoallv */
    MPI_Alltoallv(
            all_send_buf, send_size, sdispls, MPI_BYTE,
            all_recv_buf, recv_size, rdispls, MPI_BYTE,
            fd->comm );

#if 0
    DBG_FPRINTF(stderr, "\tall_recv_buf = " );
    for (i=131072; i<131073; i++) { DBG_FPRINTF(stderr, "%2d,", all_recv_buf  [i] ); }
    DBG_FPRINTF(stderr, "\n" );
#endif

  /* unpack at the receiver side */
    if (nprocs_recv) {
        if (!buftype_is_contig)
            ADIOI_Fill_user_buffer(fd, buf, flat_buf, recv_buf,
                                   offset_list, len_list, (unsigned*)recv_size,
                                   requests, statuses,          /* never used inside */
                                   recd_from_proc,
                                   nprocs, contig_access_count,
                                   min_st_offset, fd_size, fd_start, fd_end,
                               buftype_extent);
        else {
	    rtail = 0;
            for (i=0; i < nprocs; i++)
                if (recv_size[i]) {
                    memcpy( (char *)buf + buf_idx[i], all_recv_buf + rtail, recv_size[i] );
		    buf_idx[i] += recv_size[i];
		    rtail += recv_size[i];
		}
        }
    }

    ADIOI_Free( all_send_buf );
    ADIOI_Free( all_recv_buf );
    ADIOI_Free( recv_buf  );
    ADIOI_Free( sdispls );
    ADIOI_Free( rdispls );
    return;
}