コード例 #1
0
ファイル: ad_nfs_read.c プロジェクト: carsten-clauss/MP-MPICH
void ADIOI_NFS_ReadContig(ADIO_File fd, void *buf, int count, 
                     MPI_Datatype datatype, int file_ptr_type,
		     ADIO_Offset offset, ADIO_Status *status, int *error_code)
{
    int err=-1, datatype_size, len;
#ifndef PRINT_ERR_MSG
    static char myname[] = "ADIOI_NFS_READCONTIG";
#endif

    MPI_Type_size(datatype, &datatype_size);
    len = datatype_size * count;

    if (file_ptr_type == ADIO_EXPLICIT_OFFSET) {
	if (fd->fp_sys_posn != offset)
	    lseek(fd->fd_sys, offset, SEEK_SET);
	if (fd->atomicity)
	    ADIOI_WRITE_LOCK(fd, offset, SEEK_SET, len);
	else ADIOI_READ_LOCK(fd, offset, SEEK_SET, len);
	err = read(fd->fd_sys, buf, len);
	ADIOI_UNLOCK(fd, offset, SEEK_SET, len);
	fd->fp_sys_posn = offset + err;
	/* individual file pointer not updated */        
    }
    else {  /* read from curr. location of ind. file pointer */
	offset = fd->fp_ind;
	if (fd->fp_sys_posn != fd->fp_ind)
	    lseek(fd->fd_sys, fd->fp_ind, SEEK_SET);
	if (fd->atomicity)
	    ADIOI_WRITE_LOCK(fd, offset, SEEK_SET, len);
	else ADIOI_READ_LOCK(fd, offset, SEEK_SET, len);
	err = read(fd->fd_sys, buf, len);
	ADIOI_UNLOCK(fd, offset, SEEK_SET, len);
	fd->fp_ind += err;
	fd->fp_sys_posn = fd->fp_ind;
    }

#ifdef HAVE_STATUS_SET_BYTES
    if (err != -1) MPIR_Status_set_bytes(status, datatype, err);
#endif

#ifdef PRINT_ERR_MSG
    *error_code = (err == -1) ? MPI_ERR_UNKNOWN : MPI_SUCCESS;
#else
    if (err == -1) {
	*error_code = MPIR_Err_setmsg(MPI_ERR_IO, MPIR_ADIO_ERROR,
			      myname, "I/O Error", "%s", strerror(errno));
	ADIOI_Error(fd, *error_code, myname);	    
    }
    else *error_code = MPI_SUCCESS;
#endif
}
コード例 #2
0
ファイル: ad_nfs_read.c プロジェクト: hpc/mvapich-cce
void ADIOI_NFS_ReadContig(ADIO_File fd, void *buf, int count, 
                     MPI_Datatype datatype, int file_ptr_type,
		     ADIO_Offset offset, ADIO_Status *status, int *error_code)
{
    int err=-1, datatype_size, len;
    static char myname[] = "ADIOI_NFS_READCONTIG";

    MPI_Type_size(datatype, &datatype_size);
    len = datatype_size * count;

    if (file_ptr_type == ADIO_EXPLICIT_OFFSET) {
	if (fd->fp_sys_posn != offset)
	    lseek(fd->fd_sys, offset, SEEK_SET);
	if (fd->atomicity)
	    ADIOI_WRITE_LOCK(fd, offset, SEEK_SET, len);
	else ADIOI_READ_LOCK(fd, offset, SEEK_SET, len);
	err = read(fd->fd_sys, buf, len);
	ADIOI_UNLOCK(fd, offset, SEEK_SET, len);
	fd->fp_sys_posn = offset + err;
	/* individual file pointer not updated */        
    }
    else {  /* read from curr. location of ind. file pointer */
	offset = fd->fp_ind;
	if (fd->fp_sys_posn != fd->fp_ind)
	    lseek(fd->fd_sys, fd->fp_ind, SEEK_SET);
	if (fd->atomicity)
	    ADIOI_WRITE_LOCK(fd, offset, SEEK_SET, len);
	else ADIOI_READ_LOCK(fd, offset, SEEK_SET, len);
	err = read(fd->fd_sys, buf, len);
	ADIOI_UNLOCK(fd, offset, SEEK_SET, len);
	fd->fp_ind += err;
	fd->fp_sys_posn = fd->fp_ind;
    }

    /* --BEGIN ERROR HANDLING-- */
    if (err == -1) {
	*error_code = MPIO_Err_create_code(MPI_SUCCESS, MPIR_ERR_RECOVERABLE,
					   myname, __LINE__, MPI_ERR_IO,
					   "**io", "**io %s", strerror(errno));
	return;
    }
    /* --END ERROR HANDLING-- */

#ifdef HAVE_STATUS_SET_BYTES
    MPIR_Status_set_bytes(status, datatype, err);
#endif

    *error_code = MPI_SUCCESS;
}
コード例 #3
0
ファイル: ad_nfs_read.c プロジェクト: anandhis/ompi
void ADIOI_NFS_ReadStrided(ADIO_File fd, void *buf, int count,
                       MPI_Datatype datatype, int file_ptr_type,
                       ADIO_Offset offset, ADIO_Status *status, int
                       *error_code)
{
/* offset is in units of etype relative to the filetype. */

    ADIOI_Flatlist_node *flat_buf, *flat_file;
    int i, j, k, err=-1, brd_size, st_index=0;
    int num, size, sum, n_etypes_in_filetype, size_in_filetype;
    MPI_Count bufsize;
    int n_filetypes, etype_in_filetype;
    ADIO_Offset abs_off_in_filetype=0;
    int req_len, partial_read;
    MPI_Count filetype_size, etype_size, buftype_size;
    MPI_Aint filetype_extent, buftype_extent;
    int buf_count, buftype_is_contig, filetype_is_contig;
    ADIO_Offset userbuf_off;
    ADIO_Offset off, req_off, disp, end_offset=0, readbuf_off, start_off;
    char *readbuf, *tmp_buf, *value;
    int st_n_filetypes, readbuf_len;
    ADIO_Offset frd_size=0, new_frd_size, st_frd_size;
    int new_brd_size, err_flag=0, info_flag, max_bufsize;

    static char myname[] = "ADIOI_NFS_READSTRIDED";

    ADIOI_Datatype_iscontig(datatype, &buftype_is_contig);
    ADIOI_Datatype_iscontig(fd->filetype, &filetype_is_contig);

    MPI_Type_size_x(fd->filetype, &filetype_size);
    if ( ! filetype_size ) {
#ifdef HAVE_STATUS_SET_BYTES
	MPIR_Status_set_bytes(status, datatype, 0);
#endif
	*error_code = MPI_SUCCESS;
	return;
    }

    MPI_Type_extent(fd->filetype, &filetype_extent);
    MPI_Type_size_x(datatype, &buftype_size);
    MPI_Type_extent(datatype, &buftype_extent);
    etype_size = fd->etype_size;

    bufsize = buftype_size * count;

/* get max_bufsize from the info object. */

    value = (char *) ADIOI_Malloc((MPI_MAX_INFO_VAL+1)*sizeof(char));
    ADIOI_Info_get(fd->info, "ind_rd_buffer_size", MPI_MAX_INFO_VAL, value,
                 &info_flag);
    max_bufsize = atoi(value);
    ADIOI_Free(value);

    if (!buftype_is_contig && filetype_is_contig) {

/* noncontiguous in memory, contiguous in file. */

	ADIOI_Flatten_datatype(datatype);
	flat_buf = ADIOI_Flatlist;
	while (flat_buf->type != datatype) flat_buf = flat_buf->next;

        off = (file_ptr_type == ADIO_INDIVIDUAL) ? fd->fp_ind :
                 fd->disp + etype_size * offset;

	start_off = off;
	end_offset = off + bufsize - 1;
        readbuf_off = off;
        readbuf = (char *) ADIOI_Malloc(max_bufsize);
        readbuf_len = (int) (ADIOI_MIN(max_bufsize, end_offset-readbuf_off+1));

/* if atomicity is true, lock (exclusive) the region to be accessed */
        if (fd->atomicity)
            ADIOI_WRITE_LOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_lseek_a, 0, NULL );
#endif
	lseek(fd->fd_sys, readbuf_off, SEEK_SET);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_lseek_b, 0, NULL );
#endif
        if (!(fd->atomicity)) ADIOI_READ_LOCK(fd, readbuf_off, SEEK_SET, readbuf_len);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_a, 0, NULL );
#endif
        err = read(fd->fd_sys, readbuf, readbuf_len);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_b, 0, NULL );
#endif
        if (!(fd->atomicity)) ADIOI_UNLOCK(fd, readbuf_off, SEEK_SET, readbuf_len);
        if (err == -1) err_flag = 1;

        for (j=0; j<count; j++)
            for (i=0; i<flat_buf->count; i++) {
                userbuf_off = j*buftype_extent + flat_buf->indices[i];
		req_off = off;
		req_len = flat_buf->blocklens[i];
		ADIOI_BUFFERED_READ
                off += flat_buf->blocklens[i];
            }

        if (fd->atomicity)
            ADIOI_UNLOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

        if (file_ptr_type == ADIO_INDIVIDUAL) fd->fp_ind = off;

	ADIOI_Free(readbuf); /* malloced in the buffered_read macro */

	if (err_flag) {
	    *error_code = MPIO_Err_create_code(MPI_SUCCESS,
					       MPIR_ERR_RECOVERABLE, myname,
					       __LINE__, MPI_ERR_IO, "**io",
					       "**io %s", strerror(errno));
	}
	else *error_code = MPI_SUCCESS;
    }

    else {  /* noncontiguous in file */

/* filetype already flattened in ADIO_Open */
	flat_file = ADIOI_Flatlist;
	while (flat_file->type != fd->filetype) flat_file = flat_file->next;
	disp = fd->disp;

	if (file_ptr_type == ADIO_INDIVIDUAL) {
          /* Wei-keng reworked type processing to be a bit more efficient */
           offset       = fd->fp_ind - disp;
           n_filetypes  = (offset - flat_file->indices[0]) / filetype_extent;
          offset -= (ADIO_Offset)n_filetypes * filetype_extent;
          /* now offset is local to this extent */

           /* find the block where offset is located, skip blocklens[i]==0 */
           for (i=0; i<flat_file->count; i++) {
               ADIO_Offset dist;
               if (flat_file->blocklens[i] == 0) continue;
               dist = flat_file->indices[i] + flat_file->blocklens[i] - offset;
               /* frd_size is from offset to the end of block i */
              if (dist == 0) {
                  i++;
                  offset   = flat_file->indices[i];
                  frd_size = flat_file->blocklens[i];
                  break;
              }
              if (dist > 0 ) {
                   frd_size = dist;
		   break;
              }
          }
           st_index = i;  /* starting index in flat_file->indices[] */
           offset += disp + (ADIO_Offset)n_filetypes*filetype_extent;
       }
	else {
	    n_etypes_in_filetype = filetype_size/etype_size;
	    n_filetypes = (int) (offset / n_etypes_in_filetype);
	    etype_in_filetype = (int) (offset % n_etypes_in_filetype);
	    size_in_filetype = etype_in_filetype * etype_size;

	    sum = 0;
	    for (i=0; i<flat_file->count; i++) {
		sum += flat_file->blocklens[i];
		if (sum > size_in_filetype) {
		    st_index = i;
		    frd_size = sum - size_in_filetype;
		    abs_off_in_filetype = flat_file->indices[i] +
			size_in_filetype - (sum - flat_file->blocklens[i]);
		    break;
		}
	    }

	    /* abs. offset in bytes in the file */
	    offset = disp + (ADIO_Offset) n_filetypes*filetype_extent +
		    abs_off_in_filetype;
	}

        start_off = offset;

       /* Wei-keng Liao: read request is within a single flat_file contig
        * block e.g. with subarray types that actually describe the whole
        * array */
       if (buftype_is_contig && bufsize <= frd_size) {
            ADIO_ReadContig(fd, buf, bufsize, MPI_BYTE, ADIO_EXPLICIT_OFFSET,
                             offset, status, error_code);

           if (file_ptr_type == ADIO_INDIVIDUAL) {
                /* update MPI-IO file pointer to point to the first byte that
                * can be accessed in the fileview. */
               fd->fp_ind = offset + bufsize;
               if (bufsize == frd_size) {
                   do {
                       st_index++;
                       if (st_index == flat_file->count) {
                           st_index = 0;
                           n_filetypes++;
                       }
                    } while (flat_file->blocklens[st_index] == 0);
                   fd->fp_ind = disp + flat_file->indices[st_index]
                               + n_filetypes*filetype_extent;
               }
           }
           fd->fp_sys_posn = -1;   /* set it to null. */
#ifdef HAVE_STATUS_SET_BYTES
           MPIR_Status_set_bytes(status, datatype, bufsize);
#endif
            return;
       }

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

	st_frd_size = frd_size;
	st_n_filetypes = n_filetypes;
	i = 0;
	j = st_index;
	off = offset;
	frd_size = ADIOI_MIN(st_frd_size, bufsize);
	while (i < bufsize) {
	    i += frd_size;
	    end_offset = off + frd_size - 1;
            j = (j+1) % flat_file->count;
            n_filetypes += (j == 0) ? 1 : 0;
            while (flat_file->blocklens[j]==0) {
               j = (j+1) % flat_file->count;
               n_filetypes += (j == 0) ? 1 : 0;
	    }

	    off = disp + flat_file->indices[j] + (ADIO_Offset) n_filetypes*filetype_extent;
	    frd_size = ADIOI_MIN(flat_file->blocklens[j], bufsize-i);
	}

/* if atomicity is true, lock (exclusive) the region to be accessed */
        if (fd->atomicity)
            ADIOI_WRITE_LOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

        /* initial read into readbuf */
	readbuf_off = offset;
	readbuf = (char *) ADIOI_Malloc(max_bufsize);
	readbuf_len = (int) (ADIOI_MIN(max_bufsize, end_offset-readbuf_off+1));

#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_lseek_a, 0, NULL );
#endif
	lseek(fd->fd_sys, offset, SEEK_SET);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_lseek_b, 0, NULL );
#endif
        if (!(fd->atomicity)) ADIOI_READ_LOCK(fd, offset, SEEK_SET, readbuf_len);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_a, 0, NULL );
#endif
        err = read(fd->fd_sys, readbuf, readbuf_len);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_b, 0, NULL );
#endif
        if (!(fd->atomicity)) ADIOI_UNLOCK(fd, offset, SEEK_SET, readbuf_len);

        if (err == -1) err_flag = 1;

	if (buftype_is_contig && !filetype_is_contig) {

/* contiguous in memory, noncontiguous in file. should be the most
   common case. */

	    i = 0;
	    j = st_index;
	    off = offset;
	    n_filetypes = st_n_filetypes;
	    frd_size = ADIOI_MIN(st_frd_size, bufsize);
	    while (i < bufsize) {
                if (frd_size) {
                    /* TYPE_UB and TYPE_LB can result in
                       frd_size = 0. save system call in such cases */
		    /* lseek(fd->fd_sys, off, SEEK_SET);
		    err = read(fd->fd_sys, ((char *) buf) + i, frd_size);*/

		    req_off = off;
		    req_len = frd_size;
		    userbuf_off = i;
		    ADIOI_BUFFERED_READ
		}
		i += frd_size;

                if (off + frd_size < disp + flat_file->indices[j] +
                   flat_file->blocklens[j] + (ADIO_Offset) n_filetypes*filetype_extent)
                       off += frd_size;
                /* did not reach end of contiguous block in filetype.
                   no more I/O needed. off is incremented by frd_size. */
                else {
                    j = (j+1) % flat_file->count;
                    n_filetypes += (j == 0) ? 1 : 0;
                    while (flat_file->blocklens[j]==0) {
                        j = (j+1) % flat_file->count;
                        n_filetypes += (j == 0) ? 1 : 0;
                    }
		    off = disp + flat_file->indices[j] +
                                        (ADIO_Offset) n_filetypes*filetype_extent;
		    frd_size = ADIOI_MIN(flat_file->blocklens[j], bufsize-i);
		}
	    }
	}
	else {
コード例 #4
0
ファイル: ad_nfs_read.c プロジェクト: anandhis/ompi
void ADIOI_NFS_ReadContig(ADIO_File fd, void *buf, int count,
                     MPI_Datatype datatype, int file_ptr_type,
		     ADIO_Offset offset, ADIO_Status *status, int *error_code)
{
    ssize_t err=-1;
    MPI_Count datatype_size, len;
    ADIO_Offset bytes_xfered=0;
    size_t rd_count;
    static char myname[] = "ADIOI_NFS_READCONTIG";
    char *p;

    MPI_Type_size_x(datatype, &datatype_size);
    len = datatype_size * count;

    if (file_ptr_type == ADIO_INDIVIDUAL) {
	offset = fd->fp_ind;
    }

    p = buf;
    while (bytes_xfered < len ) {
        rd_count = len - bytes_xfered;
        /* FreeBSD and Darwin workaround: bigger than INT_MAX is an error */
        if (rd_count > INT_MAX)
            rd_count = INT_MAX;
	if (fd->atomicity)
            ADIOI_WRITE_LOCK(fd, offset+bytes_xfered, SEEK_SET, rd_count);
        else ADIOI_READ_LOCK(fd, offset+bytes_xfered, SEEK_SET, rd_count);
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_a, 0, NULL );
#endif
        err = pread(fd->fd_sys, p, rd_count, offset+bytes_xfered);
        /* --BEGIN ERROR HANDLING-- */
        if (err == -1) {
            *error_code = MPIO_Err_create_code(MPI_SUCCESS,
        	    MPIR_ERR_RECOVERABLE, myname, __LINE__, MPI_ERR_IO,
        	    "**io", "**io %s", strerror(errno));
        }
        /* --END ERROR HANDLING-- */
#ifdef ADIOI_MPE_LOGGING
        MPE_Log_event( ADIOI_MPE_read_b, 0, NULL );
#endif
        ADIOI_UNLOCK(fd, offset+bytes_xfered, SEEK_SET, rd_count);
        if (err == 0) {
            /* end of file */
            break;
        }
        bytes_xfered += err;
        p += err;
    }

    fd->fp_sys_posn = offset + bytes_xfered;
    if (file_ptr_type == ADIO_INDIVIDUAL) {
        fd->fp_ind += bytes_xfered;
    }
    /* --END ERROR HANDLING-- */

#ifdef HAVE_STATUS_SET_BYTES
    if (err != -1) MPIR_Status_set_bytes(status, datatype, bytes_xfered);
#endif

    *error_code = MPI_SUCCESS;
}
コード例 #5
0
int ADIOI_NFS_aio(ADIO_File fd, void *buf, int len, ADIO_Offset offset,
		  int wr, MPI_Request *request)
{
    int err=-1, fd_sys;
    int error_code, this_errno;

    struct aiocb *aiocbp;
    ADIOI_AIO_Request *aio_req;

    fd_sys = fd->fd_sys;

    aio_req = (ADIOI_AIO_Request*)ADIOI_Calloc(sizeof(ADIOI_AIO_Request), 1);
    aiocbp = (struct aiocb *) ADIOI_Calloc(sizeof(struct aiocb), 1);
    aiocbp->aio_offset = offset;
    aiocbp->aio_buf    = buf;
    aiocbp->aio_nbytes = len;

#ifdef ROMIO_HAVE_STRUCT_AIOCB_WITH_AIO_WHENCE
    aiocbp->aio_whence = SEEK_SET;
#endif
#ifdef ROMIO_HAVE_STRUCT_AIOCB_WITH_AIO_FILDES
    aiocbp->aio_fildes = fd_sys;
#endif
#ifdef ROMIO_HAVE_STRUCT_AIOCB_WITH_AIO_SIGEVENT
# ifdef AIO_SIGNOTIFY_NONE
    aiocbp->aio_sigevent.sigev_notify = SIGEV_NONE;
# endif
    aiocbp->aio_sigevent.sigev_signo = 0;
#endif
#ifdef ROMIO_HAVE_STRUCT_AIOCB_WITH_AIO_REQPRIO
# ifdef AIO_PRIO_DFL
    aiocbp->aio_reqprio = AIO_PRIO_DFL;   /* not needed in DEC Unix 4.0 */
# else
    aiocbp->aio_reqprio = 0;
# endif
#endif

    if (wr) ADIOI_WRITE_LOCK(fd, offset, SEEK_SET, len);
    else ADIOI_READ_LOCK(fd, offset, SEEK_SET, len);

#ifndef ROMIO_HAVE_AIO_CALLS_NEED_FILEDES
    if (wr) err = aio_write(aiocbp);
    else err = aio_read(aiocbp);
#else
    /* Broken IBM interface */
    if (wr) err = aio_write(fd_sys, aiocbp);
    else err = aio_read(fd_sys, aiocbp);
#endif

    this_errno = errno;
    ADIOI_UNLOCK(fd, offset, SEEK_SET, len);

    if (err == -1) {
	if (this_errno == EAGAIN) {
        /* exceeded the max. no. of outstanding requests.
           complete all previous async. requests and try again. */
	    ADIO_WriteContig(fd, buf, len, MPI_BYTE, ADIO_EXPLICIT_OFFSET,
			    offset, NULL, &error_code);
	    MPIO_Completed_request_create(&fd, len, &error_code, request);
	    return 0;
	} else {
	    return -this_errno;
	}
    }
    aio_req->aiocbp = aiocbp;
    if (ADIOI_GEN_greq_class == 0) {
	    MPIX_Grequest_class_create(ADIOI_GEN_aio_query_fn, 
			    ADIOI_GEN_aio_free_fn, MPIU_Greq_cancel_fn, 
			    ADIOI_GEN_aio_poll_fn, ADIOI_GEN_aio_wait_fn, 
			    &ADIOI_GEN_greq_class);
    }
    MPIX_Grequest_class_allocate(ADIOI_GEN_greq_class, aio_req, request);
    memcpy(&(aio_req->req), request, sizeof(MPI_Request));
    return 0;
}
コード例 #6
0
ファイル: ad_nfs_read.c プロジェクト: carsten-clauss/MP-MPICH
void ADIOI_NFS_ReadStrided(ADIO_File fd, void *buf, int count,
                       MPI_Datatype datatype, int file_ptr_type,
                       ADIO_Offset offset, ADIO_Status *status, int
                       *error_code)
{
/* offset is in units of etype relative to the filetype. */

    ADIOI_Flatlist_node *flat_buf, *flat_file;
    int i, j, k, err=-1, brd_size, frd_size=0, st_index=0;
    int bufsize, num, size, sum, n_etypes_in_filetype, size_in_filetype;
    int n_filetypes, etype_in_filetype;
    ADIO_Offset abs_off_in_filetype=0;
    int filetype_size, etype_size, buftype_size, req_len, partial_read;
    MPI_Aint filetype_extent, buftype_extent; 
    int buf_count, buftype_is_contig, filetype_is_contig;
    ADIO_Offset userbuf_off;
    ADIO_Offset off, req_off, disp, end_offset, readbuf_off, start_off;
    char *readbuf, *tmp_buf, *value;
    int flag, st_frd_size, st_n_filetypes, readbuf_len;
    int new_brd_size, new_frd_size, err_flag=0, info_flag, max_bufsize;
#ifndef PRINT_ERR_MSG
    static char myname[] = "ADIOI_NFS_READSTRIDED";
#endif

    ADIOI_Datatype_iscontig(datatype, &buftype_is_contig);
    ADIOI_Datatype_iscontig(fd->filetype, &filetype_is_contig);

    MPI_Type_size(fd->filetype, &filetype_size);
    MPI_Type_extent(fd->filetype, &filetype_extent);
    MPI_Type_size(datatype, &buftype_size);
    MPI_Type_extent(datatype, &buftype_extent);
    etype_size = fd->etype_size;

    bufsize = buftype_size * count;

/* get max_bufsize from the info object. */

    value = (char *) ADIOI_Malloc((MPI_MAX_INFO_VAL+1)*sizeof(char));
    MPI_Info_get(fd->info, "ind_rd_buffer_size", MPI_MAX_INFO_VAL, value, 
                 &info_flag);
    max_bufsize = atoi(value);
    ADIOI_Free(value);

    if (!buftype_is_contig && filetype_is_contig) {

/* noncontiguous in memory, contiguous in file. */

	ADIOI_Flatten_datatype(datatype);
	flat_buf = ADIOI_Flatlist;
	while (flat_buf->type != datatype) flat_buf = flat_buf->next;

        off = (file_ptr_type == ADIO_INDIVIDUAL) ? fd->fp_ind : 
                 fd->disp + etype_size * offset;

	start_off = off;
	end_offset = off + bufsize - 1;
        readbuf_off = off;
        readbuf = (char *) ADIOI_Malloc(max_bufsize);
        readbuf_len = (int) (ADIOI_MIN(max_bufsize, end_offset-readbuf_off+1));

/* if atomicity is true, lock (exclusive) the region to be accessed */
        if (fd->atomicity)
            ADIOI_WRITE_LOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

	lseek(fd->fd_sys, readbuf_off, SEEK_SET);
        if (!(fd->atomicity)) ADIOI_READ_LOCK(fd, readbuf_off, SEEK_SET, readbuf_len);
        err = read(fd->fd_sys, readbuf, readbuf_len);
        if (!(fd->atomicity)) ADIOI_UNLOCK(fd, readbuf_off, SEEK_SET, readbuf_len);
        if (err == -1) err_flag = 1;

        for (j=0; j<count; j++) 
            for (i=0; i<flat_buf->count; i++) {
                userbuf_off = j*buftype_extent + flat_buf->indices[i];
		req_off = off;
		req_len = flat_buf->blocklens[i];
		ADIOI_BUFFERED_READ
                off += flat_buf->blocklens[i];
            }

        if (fd->atomicity)
            ADIOI_UNLOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

        if (file_ptr_type == ADIO_INDIVIDUAL) fd->fp_ind = off;

	ADIOI_Free(readbuf); /* malloced in the buffered_read macro */

#ifdef PRINT_ERR_MSG
        *error_code = (err_flag) ? MPI_ERR_UNKNOWN : MPI_SUCCESS;
#else
	if (err_flag) {
	    *error_code = MPIR_Err_setmsg(MPI_ERR_IO, MPIR_ADIO_ERROR,
			      myname, "I/O Error", "%s", strerror(errno));
	    ADIOI_Error(fd, *error_code, myname);	    
	}
	else *error_code = MPI_SUCCESS;
#endif
    }

    else {  /* noncontiguous in file */

/* filetype already flattened in ADIO_Open */
	flat_file = ADIOI_Flatlist;
	while (flat_file->type != fd->filetype) flat_file = flat_file->next;
	disp = fd->disp;

	if (file_ptr_type == ADIO_INDIVIDUAL) {
	    offset = fd->fp_ind; /* in bytes */
	    n_filetypes = -1;
	    flag = 0;
	    while (!flag) {
                n_filetypes++;
		for (i=0; i<flat_file->count; i++) {
		    if (disp + flat_file->indices[i] + 
                        (ADIO_Offset) n_filetypes*filetype_extent + flat_file->blocklens[i] 
                            >= offset) {
			st_index = i;
			frd_size = (int) (disp + flat_file->indices[i] + 
			        (ADIO_Offset) n_filetypes*filetype_extent
			         + flat_file->blocklens[i] - offset);
			flag = 1;
			break;
		    }
		}
	    }
	}
	else {
	    n_etypes_in_filetype = filetype_size/etype_size;
	    n_filetypes = (int) (offset / n_etypes_in_filetype);
	    etype_in_filetype = (int) (offset % n_etypes_in_filetype);
	    size_in_filetype = etype_in_filetype * etype_size;
 
	    sum = 0;
	    for (i=0; i<flat_file->count; i++) {
		sum += flat_file->blocklens[i];
		if (sum > size_in_filetype) {
		    st_index = i;
		    frd_size = sum - size_in_filetype;
		    abs_off_in_filetype = flat_file->indices[i] +
			size_in_filetype - (sum - flat_file->blocklens[i]);
		    break;
		}
	    }

	    /* abs. offset in bytes in the file */
	    offset = disp + (ADIO_Offset) n_filetypes*filetype_extent + abs_off_in_filetype;
	}

        start_off = offset;

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

	st_frd_size = frd_size;
	st_n_filetypes = n_filetypes;
	i = 0;
	j = st_index;
	off = offset;
	frd_size = ADIOI_MIN(st_frd_size, bufsize);
	while (i < bufsize) {
	    i += frd_size;
	    end_offset = off + frd_size - 1;

	    if (j < (flat_file->count - 1)) j++;
	    else {
		j = 0;
		n_filetypes++;
	    }

	    off = disp + flat_file->indices[j] + (ADIO_Offset) n_filetypes*filetype_extent;
	    frd_size = ADIOI_MIN(flat_file->blocklens[j], bufsize-i);
	}

/* if atomicity is true, lock (exclusive) the region to be accessed */
        if (fd->atomicity)
            ADIOI_WRITE_LOCK(fd, start_off, SEEK_SET, end_offset-start_off+1);

        /* initial read into readbuf */
	readbuf_off = offset;
	readbuf = (char *) ADIOI_Malloc(max_bufsize);
	readbuf_len = (int) (ADIOI_MIN(max_bufsize, end_offset-readbuf_off+1));

	lseek(fd->fd_sys, offset, SEEK_SET);
        if (!(fd->atomicity)) ADIOI_READ_LOCK(fd, offset, SEEK_SET, readbuf_len);
        err = read(fd->fd_sys, readbuf, readbuf_len);
        if (!(fd->atomicity)) ADIOI_UNLOCK(fd, offset, SEEK_SET, readbuf_len);

        if (err == -1) err_flag = 1;

	if (buftype_is_contig && !filetype_is_contig) {

/* contiguous in memory, noncontiguous in file. should be the most
   common case. */

	    i = 0;
	    j = st_index;
	    off = offset;
	    n_filetypes = st_n_filetypes;
	    frd_size = ADIOI_MIN(st_frd_size, bufsize);
	    while (i < bufsize) {
                if (frd_size) { 
                    /* TYPE_UB and TYPE_LB can result in 
                       frd_size = 0. save system call in such cases */ 
		    /* lseek(fd->fd_sys, off, SEEK_SET);
		    err = read(fd->fd_sys, ((char *) buf) + i, frd_size);*/

		    req_off = off;
		    req_len = frd_size;
		    userbuf_off = i;
		    ADIOI_BUFFERED_READ
		}
		i += frd_size;

                if (off + frd_size < disp + flat_file->indices[j] +
                   flat_file->blocklens[j] + (ADIO_Offset) n_filetypes*filetype_extent)
                       off += frd_size;
                /* did not reach end of contiguous block in filetype.
                   no more I/O needed. off is incremented by frd_size. */
                else {
		    if (j < (flat_file->count - 1)) j++;
		    else {
			j = 0;
			n_filetypes++;
		    }
		    off = disp + flat_file->indices[j] + 
                                        (ADIO_Offset) n_filetypes*filetype_extent;
		    frd_size = ADIOI_MIN(flat_file->blocklens[j], bufsize-i);
		}
	    }
	}
	else {