summaryrefslogtreecommitdiffstats
path: root/src/H5Smpio.c
blob: b04faebc265b40fe60fc0ddd8aaf43239ac6eb60 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
 * Copyright by the Board of Trustees of the University of Illinois.         *
 * All rights reserved.                                                      *
 *                                                                           *
 * This file is part of HDF5.  The full HDF5 copyright notice, including     *
 * terms governing use, modification, and redistribution, is contained in    *
 * the files COPYING and Copyright.html.  COPYING can be found at the root   *
 * of the source code distribution tree; Copyright.html can be found at the  *
 * root level of an installed copy of the electronic HDF5 document set and   *
 * is linked from the top-level documents page.  It can also be found at     *
 * http://hdf.ncsa.uiuc.edu/HDF5/doc/Copyright.html.  If you do not have     *
 * access to either file, you may request a copy from hdfhelp@ncsa.uiuc.edu. *
 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

/*
 * Programmer:  rky 980813
 *
 * Purpose:	Functions to read/write directly between app buffer and file.
 *
 * 		Beware of the ifdef'ed print statements.
 *		I didn't make them portable.
 */

#define H5S_PACKAGE		/*suppress error about including H5Spkg	  */

/* Pablo information */
/* (Put before include files to avoid problems with inline functions) */
#define PABLO_MASK	H5S_mpio_mask

#include "H5private.h"		/* Generic Functions			*/
#include "H5Dprivate.h"		/* Datasets				*/
#include "H5Eprivate.h"		/* Error handling		  	*/
#include "H5Fprivate.h"		/* File access				*/
#include "H5FDprivate.h"	/* File drivers				*/
#include "H5Iprivate.h"		/* IDs			  		*/
#include "H5Oprivate.h"		/* Object headers		  	*/
#include "H5Pprivate.h"         /* Property lists                       */
#include "H5Spkg.h"		/* Dataspaces 				*/

#ifdef H5_HAVE_PARALLEL

static herr_t
H5S_mpio_all_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type );
static herr_t
H5S_mpio_none_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type );
static herr_t
H5S_mpio_hyper_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type );


/*-------------------------------------------------------------------------
 * Function:	H5S_mpio_all_type
 *
 * Purpose:	Translate an HDF5 "all" selection into an MPI type.
 *
 * Return:	non-negative on success, negative on failure.
 *
 * Outputs:	*new_type	  the MPI type corresponding to the selection
 *		*count		  how many objects of the new_type in selection
 *				  (useful if this is the buffer type for xfer)
 *		*extra_offset     Number of bytes of offset within dataset
 *		*is_derived_type  0 if MPI primitive type, 1 if derived
 *
 * Programmer:	rky 980813
 *
 * Modifications:
 *
 *      Quincey Koziol, June 18, 2002
 *      Added 'extra_offset' parameter
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5S_mpio_all_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type )
{
    hsize_t	total_bytes;
    hssize_t	snelmts;                /*total number of elmts	(signed) */
    hsize_t	nelmts;                 /*total number of elmts	*/
    herr_t		ret_value = SUCCEED;

    FUNC_ENTER_NOAPI_NOINIT(H5S_mpio_all_type);

    /* Check args */
    assert (space);

    /* Just treat the entire extent as a block of bytes */
    if((snelmts = H5S_GET_EXTENT_NPOINTS(space))<0)
	HGOTO_ERROR (H5E_ARGS, H5E_BADVALUE, FAIL, "src dataspace has invalid selection")
    H5_ASSIGN_OVERFLOW(nelmts,snelmts,hssize_t,hsize_t);

    total_bytes = (hsize_t)elmt_size*nelmts;

    /* fill in the return values */
    *new_type = MPI_BYTE;
    H5_ASSIGN_OVERFLOW(*count, total_bytes, hsize_t, size_t);
    *extra_offset = 0;
    *is_derived_type = 0;

done:
    FUNC_LEAVE_NOAPI(ret_value);
} /* H5S_mpio_all_type() */


/*-------------------------------------------------------------------------
 * Function:	H5S_mpio_none_type
 *
 * Purpose:	Translate an HDF5 "none" selection into an MPI type.
 *
 * Return:	non-negative on success, negative on failure.
 *
 * Outputs:	*new_type	  the MPI type corresponding to the selection
 *		*count		  how many objects of the new_type in selection
 *				  (useful if this is the buffer type for xfer)
 *		*extra_offset     Number of bytes of offset within dataset
 *		*is_derived_type  0 if MPI primitive type, 1 if derived
 *
 * Programmer:	Quincey Koziol, October 29, 2002
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5S_mpio_none_type( const H5S_t UNUSED *space, size_t UNUSED elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type )
{
    FUNC_ENTER_NOAPI_NOINIT_NOFUNC(H5S_mpio_none_type);

    /* fill in the return values */
    *new_type = MPI_BYTE;
    *count = 0;
    *extra_offset = 0;
    *is_derived_type = 0;

    FUNC_LEAVE_NOAPI(SUCCEED);
} /* H5S_mpio_none_type() */


/*-------------------------------------------------------------------------
 * Function:	H5S_mpio_hyper_type
 *
 * Purpose:	Translate an HDF5 hyperslab selection into an MPI type.
 *
 * Return:	non-negative on success, negative on failure.
 *
 * Outputs:	*new_type	  the MPI type corresponding to the selection
 *		*count		  how many objects of the new_type in selection
 *				  (useful if this is the buffer type for xfer)
 *		*extra_offset     Number of bytes of offset within dataset
 *		*is_derived_type  0 if MPI primitive type, 1 if derived
 *
 * Programmer:	rky 980813
 *
 * Modifications:  ppw 990401
 *		rky, ppw 2000-09-26 Freed old type after creating struct type.
 *		rky 2000-10-05 Changed displacements to be MPI_Aint.
 *		rky 2000-10-06 Added code for cases of empty hyperslab.
 *		akc, rky 2000-11-16 Replaced hard coded dimension size with
 *		    H5S_MAX_RANK.
 *
 *      Quincey Koziol, June 18, 2002
 *      Added 'extra_offset' parameter.  Also accomodate
 *      selection offset in MPI type built.
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5S_mpio_hyper_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type )
{
    H5S_sel_iter_t sel_iter;    /* Selection iteration info */
    hbool_t sel_iter_init=0;    /* Selection iteration info has been initialized */

    struct dim {	/* less hassle than malloc/free & ilk */
        hssize_t start;
        hsize_t strid;
        hsize_t block;
        hsize_t xtent;
        hsize_t count;
    } d[H5S_MAX_RANK];

    int			i;
    int			offset[H5S_MAX_RANK];
    int			max_xtent[H5S_MAX_RANK];
    H5S_hyper_dim_t	*diminfo;		/* [rank] */
    int		rank;
    int			block_length[2];
    MPI_Datatype	inner_type, outer_type, old_type[2];
    MPI_Aint            extent_len, displacement[2];
    int                 mpi_code;               /* MPI return code */
    herr_t		ret_value = SUCCEED;

    FUNC_ENTER_NOAPI_NOINIT(H5S_mpio_hyper_type);

    /* Check args */
    assert (space);
    assert(sizeof(MPI_Aint) >= sizeof(elmt_size));
    if (0==elmt_size)
        goto empty;

    /* Initialize selection iterator */
    if (H5S_select_iter_init(&sel_iter, space, elmt_size)<0)
        HGOTO_ERROR (H5E_DATASPACE, H5E_CANTINIT, FAIL, "unable to initialize selection iterator");
    sel_iter_init=1;	/* Selection iteration info has been initialized */

    /* Abbreviate args */
    diminfo=sel_iter.u.hyp.diminfo;
    assert (diminfo);

    /* make a local copy of the dimension info so we can operate with them */

    /* Check if this is a "flattened" regular hyperslab selection */
    if(sel_iter.u.hyp.iter_rank!=0 && sel_iter.u.hyp.iter_rank<space->extent.rank) {
        /* Flattened selection */
        rank=sel_iter.u.hyp.iter_rank;
        assert (rank >= 0 && rank<=H5S_MAX_RANK);	/* within array bounds */
        if (0==rank)
            goto empty;

#ifdef H5Smpi_DEBUG
            HDfprintf(stderr, "%s: Flattened selection\n",FUNC);
#endif
        for ( i=0; i<rank; ++i) {
            d[i].start = diminfo[i].start+sel_iter.u.hyp.sel_off[i];
            d[i].strid = diminfo[i].stride;
            d[i].block = diminfo[i].block;
            d[i].count = diminfo[i].count;
            d[i].xtent = sel_iter.u.hyp.size[i];
#ifdef H5Smpi_DEBUG
            HDfprintf(stderr, "%s: start=%Hd  stride=%Hu  count=%Hu  block=%Hu  xtent=%Hu",
                FUNC, d[i].start, d[i].strid, d[i].count, d[i].block, d[i].xtent );
            if (i==0)
                HDfprintf(stderr, "  rank=%d\n", rank );
            else
                HDfprintf(stderr, "\n" );
#endif
            if (0==d[i].block)
                goto empty;
            if (0==d[i].count)
                goto empty;
            if (0==d[i].xtent)
                goto empty;
        }
    } /* end if */
    else {
        /* Non-flattened selection */
        rank = space->extent.rank;
        assert (rank >= 0 && rank<=H5S_MAX_RANK);	/* within array bounds */
        if (0==rank)
            goto empty;

#ifdef H5Smpi_DEBUG
            HDfprintf(stderr, "%s: Non-flattened selection\n",FUNC);
#endif
        for ( i=0; i<rank; ++i) {
            d[i].start = diminfo[i].start+space->select.offset[i];
            d[i].strid = diminfo[i].stride;
            d[i].block = diminfo[i].block;
            d[i].count = diminfo[i].count;
            d[i].xtent = space->extent.size[i];
#ifdef H5Smpi_DEBUG
            HDfprintf(stderr, "%s: start=%Hd  stride=%Hu  count=%Hu  block=%Hu  xtent=%Hu",
                FUNC, d[i].start, d[i].strid, d[i].count, d[i].block, d[i].xtent );
            if (i==0)
                HDfprintf(stderr, "  rank=%d\n", rank );
            else
                HDfprintf(stderr, "\n" );
#endif
            if (0==d[i].block)
                goto empty;
            if (0==d[i].count)
                goto empty;
            if (0==d[i].xtent)
                goto empty;
        }
    } /* end else */
    
/**********************************************************************
    Compute array "offset[rank]" which gives the offsets for a multi-
    dimensional array with dimensions "d[i].xtent" (i=0,1,...,rank-1). 
**********************************************************************/
    offset[rank-1] = 1;
    max_xtent[rank-1] = d[rank-1].xtent;
#ifdef H5Smpi_DEBUG
    i=rank-1;
    HDfprintf(stderr, " offset[%2d]=%d; max_xtent[%2d]=%d\n",
                          i, offset[i], i, max_xtent[i]);
#endif
    for (i=rank-2; i>=0; --i) {
        offset[i] = offset[i+1]*d[i+1].xtent;
        max_xtent[i] = max_xtent[i+1]*d[i].xtent;
#ifdef H5Smpi_DEBUG
        HDfprintf(stderr, " offset[%2d]=%d; max_xtent[%2d]=%d\n",
                          i, offset[i], i, max_xtent[i]);
#endif
    }

    /*  Create a type covering the selected hyperslab.
     *  Multidimensional dataspaces are stored in row-major order.
     *  The type is built from the inside out, going from the
     *  fastest-changing (i.e., inner) dimension * to the slowest (outer). */

/*******************************************************
*  Construct contig type for inner contig dims:
*******************************************************/
#ifdef H5Smpi_DEBUG
    HDfprintf(stderr, "%s: Making contig type %d MPI_BYTEs\n", FUNC,elmt_size );
    for (i=rank-1; i>=0; --i)
        HDfprintf(stderr, "d[%d].xtent=%Hu \n", i, d[i].xtent);
#endif
    if (MPI_SUCCESS != (mpi_code= MPI_Type_contiguous( (int)elmt_size, MPI_BYTE, &inner_type )))
        HMPI_GOTO_ERROR(FAIL, "MPI_Type_contiguous failed", mpi_code);

/*******************************************************
*  Construct the type by walking the hyperslab dims
*  from the inside out:
*******************************************************/
    for ( i=rank-1; i>=0; --i) {
#ifdef H5Smpi_DEBUG
        HDfprintf(stderr, "%s: Dimension i=%d \n"
            "count=%Hu block=%Hu stride=%Hu\n",
            FUNC, i, d[i].count, d[i].block, d[i].strid );
#endif

#ifdef H5Smpi_DEBUG
        HDfprintf(stderr, "%s: i=%d  Making vector-type \n", FUNC,i);
#endif
       /****************************************
       *  Build vector in current dimension:
       ****************************************/
	mpi_code =MPI_Type_vector((int)(d[i].count),        /* count */
				  (int)(d[i].block),        /* blocklength */
				  (int)(d[i].strid),   	    /* stride */
				  inner_type,	            /* old type */
				  &outer_type );            /* new type */

        MPI_Type_free( &inner_type );
        if (mpi_code!=MPI_SUCCESS)
            HMPI_GOTO_ERROR(FAIL, "couldn't create MPI vector type", mpi_code);

        displacement[1] = (MPI_Aint)elmt_size * max_xtent[i];
        if(MPI_SUCCESS != (mpi_code = MPI_Type_extent(outer_type, &extent_len)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Type_extent failed", mpi_code);

       /*************************************************
       *  Restructure this datatype ("outer_type")
       *  so that it still starts at 0, but its extent
       *  is the full extent in this dimension.
       *************************************************/
        if ((int)extent_len < displacement[1]) {

#ifdef H5Smpi_DEBUG
            HDfprintf(stderr, "%s: i=%d Extending struct type\n"
                "***displacements: 0, %d\n", FUNC, i, displacement[1]);
#endif

#ifdef H5_HAVE_MPI2  /*  have MPI-2  (this function is not included in MPICH) */
            mpi_code = MPI_Type_create_resized
                                  ( outer_type,        /* old type  */
                                    0,                 /* blocklengths */
                                    displacement[1],   /* displacements */
                                    &inner_type);      /* new type */
#else        /*  do not have MPI-2  */
            block_length[0] = 1;
            block_length[1] = 1;
  
            displacement[0] = 0;
  
            old_type[0] = outer_type;
            old_type[1] = MPI_UB;
            mpi_code = MPI_Type_struct ( 2,               /* count */
                                    block_length,    /* blocklengths */
                                    displacement,    /* displacements */
                                    old_type,        /* old types */
                                    &inner_type);    /* new type */
#endif
  
            MPI_Type_free (&outer_type);
    	    if (mpi_code!=MPI_SUCCESS)
                HMPI_GOTO_ERROR(FAIL, "couldn't resize MPI vector type", mpi_code);
        }
        else {
            inner_type = outer_type;
        }
    } /* end for */
/***************************
*  End of loop, walking 
*  thru dimensions.
***************************/


    /* At this point inner_type is actually the outermost type, even for 0-trip loop */

/***************************************************************
*  Final task: create a struct which is a "clone" of the
*  current struct, but displaced according to the d[i].start
*  values given in the hyperslab description:
***************************************************************/
    displacement[0] = 0;
    for (i=rank-1; i>=0; i--)
        displacement[0] += d[i].start * offset[i];

    if (displacement[0] > 0) {
        displacement[0] *= elmt_size;
        block_length[0] = 1;
        old_type[0] = inner_type;

#ifdef H5Smpi_DEBUG
        HDfprintf(stderr, "%s:  Making final struct\n***count=1:\n", FUNC);
        HDfprintf(stderr, "\tblocklength[0]=%d; displacement[0]=%d\n",
                     block_length[0], displacement[0]);
#endif

  
        if (MPI_SUCCESS != (mpi_code= MPI_Type_struct( 1,                  /* count */
                     block_length,       /* blocklengths */
                     displacement,       /* displacements */
                     old_type,	         /* old type */
                     new_type ))         /* new type */
                )
            HMPI_GOTO_ERROR(FAIL, "couldn't create MPI struct type", mpi_code);

        if (MPI_SUCCESS != (mpi_code= MPI_Type_free (&old_type[0])))
            HMPI_GOTO_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
    }
    else {
        *new_type = inner_type;
    }

    if (MPI_SUCCESS != (mpi_code= MPI_Type_commit( new_type )))
        HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);

    /* fill in the remaining return values */
    *count = 1;			/* only have to move one of these suckers! */
    *extra_offset = 0;
    *is_derived_type = 1;
    HGOTO_DONE(SUCCEED);

empty:
    /* special case: empty hyperslab */
    *new_type = MPI_BYTE;
    *count = 0;
    *extra_offset = 0;
    *is_derived_type = 0;

done:
    /* Release selection iterator */
    if(sel_iter_init) {
        if (H5S_SELECT_ITER_RELEASE(&sel_iter)<0)
            HDONE_ERROR (H5E_DATASPACE, H5E_CANTRELEASE, FAIL, "unable to release selection iterator");
    } /* end if */

#ifdef H5Smpi_DEBUG
    HDfprintf(stderr, "Leave %s, count=%Hu  is_derived_type=%d\n",
		FUNC, *count, *is_derived_type );
#endif
    FUNC_LEAVE_NOAPI(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5S_mpio_space_type
 *
 * Purpose:	Translate an HDF5 dataspace selection into an MPI type.
 *		Currently handle only hyperslab and "all" selections.
 *
 * Return:	non-negative on success, negative on failure.
 *
 * Outputs:	*new_type	  the MPI type corresponding to the selection
 *		*count		  how many objects of the new_type in selection
 *				  (useful if this is the buffer type for xfer)
 *		*extra_offset     Number of bytes of offset within dataset
 *		*is_derived_type  0 if MPI primitive type, 1 if derived
 *
 * Programmer:	rky 980813
 *
 * Modifications:
 *
 *      Quincey Koziol, June 18, 2002
 *      Added 'extra_offset' parameter
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5S_mpio_space_type( const H5S_t *space, size_t elmt_size,
		     /* out: */
		     MPI_Datatype *new_type,
		     size_t *count,
		     hsize_t *extra_offset,
		     hbool_t *is_derived_type )
{
    herr_t	ret_value = SUCCEED;

    FUNC_ENTER_NOAPI_NOINIT(H5S_mpio_space_type);

    /* Check args */
    assert (space);

    /* Creat MPI type based on the kind of selection */
    switch (H5S_GET_EXTENT_TYPE(space)) {
        case H5S_NULL:
        case H5S_SCALAR:
        case H5S_SIMPLE:
            switch(H5S_GET_SELECT_TYPE(space)) {
                case H5S_SEL_NONE:
                    if ( H5S_mpio_none_type( space, elmt_size,
                        /* out: */ new_type, count, extra_offset, is_derived_type ) <0)
                        HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't convert \"all\" selection to MPI type");
                    break;

                case H5S_SEL_ALL:
                    if ( H5S_mpio_all_type( space, elmt_size,
                        /* out: */ new_type, count, extra_offset, is_derived_type ) <0)
                        HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't convert \"all\" selection to MPI type");
                    break;

                case H5S_SEL_POINTS:
                    /* not yet implemented */
                    ret_value = FAIL;
                    break;

                case H5S_SEL_HYPERSLABS:
                    if(H5S_mpio_hyper_type( space, elmt_size,
                            /* out: */ new_type, count, extra_offset, is_derived_type )<0)
                        HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't convert \"all\" selection to MPI type");
                    break;

                default:
                    assert("unknown selection type" && 0);
                    break;
            } /* end switch */
            break;

        case H5S_COMPLEX:
            /* not yet implemented */
            HGOTO_ERROR(H5E_DATASPACE, H5E_UNSUPPORTED, FAIL, "complex data spaces are not supported yet");

        default:
            assert("unknown data space type" && 0);
            break;
    }

done:
    FUNC_LEAVE_NOAPI(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5S_mpio_opt_possible
 *
 * Purpose:	Checks if an direct I/O transfer is possible between memory and
 *                  the file.
 *
 * Return:	Success:        Non-negative: TRUE or FALSE
 *		Failure:	Negative
 *
 * Programmer:	Quincey Koziol
 *              Wednesday, April 3, 2002
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
htri_t
H5S_mpio_opt_possible( const H5F_t *file, const H5S_t *mem_space, const H5S_t *file_space, const unsigned flags,const H5O_layout_t *layout)
{
    htri_t c1,c2;               /* Flags whether a selection is optimizable */
    htri_t ret_value=TRUE;

    FUNC_ENTER_NOAPI(H5S_mpio_opt_possible, FAIL);

    /* Check args */
    assert(mem_space);
    assert(file_space);

    /* Parallel I/O conversion flag must be set, if it is not collective IO, go to false. */
    if(!(flags&H5S_CONV_PAR_IO_POSSIBLE))
        HGOTO_DONE(FALSE);

    /* Check whether these are both simple or scalar dataspaces */
    if (!((H5S_SIMPLE==H5S_GET_EXTENT_TYPE(mem_space) || H5S_SCALAR==H5S_GET_EXTENT_TYPE(mem_space))
            && (H5S_SIMPLE==H5S_GET_EXTENT_TYPE(file_space) || H5S_SCALAR==H5S_GET_EXTENT_TYPE(file_space))))
        HGOTO_DONE(FALSE);

    /* Check whether both selections are "regular" */
    c1=H5S_SELECT_IS_REGULAR(file_space);
    c2=H5S_SELECT_IS_REGULAR(mem_space);
    if(c1==FAIL || c2==FAIL)
        HGOTO_ERROR(H5E_DATASPACE, H5E_BADRANGE, FAIL, "invalid check for single selection blocks");
    if(c1==FALSE || c2==FALSE)
        HGOTO_DONE(FALSE);

    /* Can't currently handle point selections */
    if (H5S_SEL_POINTS==H5S_GET_SELECT_TYPE(mem_space) || H5S_SEL_POINTS==H5S_GET_SELECT_TYPE(file_space))
        HGOTO_DONE(FALSE);

    /* Dataset storage must be contiguous or chunked */
    if ((flags&H5S_CONV_STORAGE_MASK)!=H5S_CONV_STORAGE_CONTIGUOUS && 
            (flags&H5S_CONV_STORAGE_MASK)!=H5S_CONV_STORAGE_CHUNKED)
        HGOTO_DONE(FALSE);

    if ((flags&H5S_CONV_STORAGE_MASK)==H5S_CONV_STORAGE_CHUNKED) {
        hsize_t chunk_dim[H5O_LAYOUT_NDIMS];        /* Chunk dimensions */
        hssize_t startf[H5S_MAX_RANK],      /* Selection start bounds */
            endf[H5S_MAX_RANK];     /* Selection end bounds */
        unsigned dim_rankf;         /* Number of dimensions of file dataspace */
        int pcheck_hyper,check_hyper,   /* Flags for checking if selection is in one chunk */
            tnum_chunkf,            /* Number of chunks selection overlaps */
            max_chunkf,             /* Maximum number of chunks selection overlaps */
            min_chunkf,             /* Minimum number of chunks selection overlaps */
            num_chunks_same;        /* Flag indicating whether all processes have the same # of chunks to operate on */
        unsigned dim_chunks;        /* Temporary number of chunks in a dimension */
        MPI_Comm comm;              /* MPI communicator for file */
        int mpi_rank;               /* Rank in MPI communicator */
        int mpi_code;               /* MPI return code */
        unsigned u;                 /* Local index variable */

        /* Getting MPI communicator and rank */
        if((comm = H5F_mpi_get_comm(file))==MPI_COMM_NULL)
            HGOTO_ERROR(H5E_DATASPACE, H5E_CANTGET, FAIL, "can't retrieve MPI communicator")
        if((mpi_rank = H5F_mpi_get_rank(file))<0)
            HGOTO_ERROR(H5E_DATASPACE, H5E_CANTGET, FAIL, "can't retrieve MPI rank")

      /* Currently collective chunking storage 
	 inside HDF5 is supported for either one of the following two cases:
	 1. All the hyperslabs for one process is inside one chunk.
	 2. For single hyperslab selection, the number of chunks that covered 
	    the single selection for all processes should be equal. 
	    KY, 2004/7/14
      */

      /* Quincey, please read.
	 This is maybe redundant, I think only when both memory and file space be SCALAR
	 space, the collective IO can work. Otherwise, SELECT_POINT will be reached,collective
	 IO shouldn't work.
	 Please clarify and correct the code on the following,
         Quincey said that it was probably okay if only one data space is SCALAR, 
         Still keep the code here until we added more tests later.
	 Kent */
        if(H5S_SCALAR==mem_space->extent.type || H5S_SCALAR ==file_space->extent.type) {
            if(!(H5S_SCALAR==mem_space->extent.type && H5S_SCALAR ==file_space->extent.type))
                HGOTO_DONE(FALSE)
            else
                HGOTO_DONE(TRUE)
        } /* end if */

        dim_rankf = file_space->extent.rank;

        if(H5S_SELECT_BOUNDS(file_space,startf,endf)==FAIL)
            HGOTO_ERROR(H5E_DATASPACE, H5E_BADRANGE,FAIL, "invalid check for single selection blocks");

        for(u=0; u < layout->u.chunk.ndims; u++) 
            chunk_dim[u] = layout->u.chunk.dim[u];

        /* Case 1: check whether all hyperslab in this process is inside one chunk.
           Note: we don't handle when starting point is less than zero since that may cover
           two chunks. */

        /*for file space checking*/
        pcheck_hyper = 1;
        for (u=0; u<dim_rankf; u++)
            if(endf[u]/chunk_dim[u]!=startf[u]/chunk_dim[u]) {
                pcheck_hyper = 0;
                break;
            }
      
        if (MPI_SUCCESS != (mpi_code= MPI_Reduce(&pcheck_hyper,&check_hyper,1,MPI_INT,MPI_LAND,0,comm)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Reduce failed", mpi_code)
        if (MPI_SUCCESS != (mpi_code= MPI_Bcast(&check_hyper,1,MPI_INT,0,comm)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code)

        /*if check_hyper is true, condition for collective IO case is fulfilled, no
         need to do further test. */
        if(check_hyper)
            HGOTO_DONE(TRUE); 
    
      /* Case 2:Check whether the number of chunks that covered the single hyperslab is the same.
	 If not,no collective chunk IO. 
	 KY, 2004/7/14
      */
	 
        c1 = H5S_SELECT_IS_SINGLE(file_space);
        c2 = H5S_SELECT_IS_SINGLE(mem_space);

        if(c1==FAIL || c2 ==FAIL)
            HGOTO_ERROR(H5E_DATASPACE, H5E_BADRANGE, FAIL, "invalid check for single selection blocks");
        if(c1==FALSE || c2 ==FALSE)
            HGOTO_DONE(FALSE);

        /* Compute the number of chunks covered by the selection on this process */
        tnum_chunkf = 1;
        for (u=0; u<dim_rankf; u++) {
            dim_chunks = (endf[u]/chunk_dim[u]-startf[u]/chunk_dim[u])+1;
            tnum_chunkf = dim_chunks*tnum_chunkf;
        }

        /* Determine the minimum and maximum # of chunks for all processes */
        if (MPI_SUCCESS != (mpi_code= MPI_Reduce(&tnum_chunkf,&max_chunkf,1,MPI_INT,MPI_MAX,0,comm)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Reduce failed", mpi_code)
        if (MPI_SUCCESS != (mpi_code= MPI_Reduce(&tnum_chunkf,&min_chunkf,1,MPI_INT,MPI_MIN,0,comm)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Reduce failed", mpi_code)
  
        /* Let the rank==0 process determine if the same number of chunks will be operated on by all processes */
        if(mpi_rank == 0)
            num_chunks_same = (max_chunkf==min_chunkf);
                    
        /* Broadcast the flag indicating the number of chunks are the same */
        if (MPI_SUCCESS != (mpi_code= MPI_Bcast(&num_chunks_same,1,MPI_INT,0,comm)))
            HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code)

        /* Can't handle case when number of chunks is different (yet) */
        if(!num_chunks_same)
            HGOTO_DONE(FALSE);
    }

done:
    FUNC_LEAVE_NOAPI(ret_value);
} /* H5S_mpio_opt_possible() */
#endif  /* H5_HAVE_PARALLEL */