summaryrefslogtreecommitdiffstats
path: root/src/H5FDmpio.c
blob: 69e2b49cb7807f1c5903923b69a4eb6e57ab05e9 (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
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
/*
 * Copyright (c) 1999-2002 NCSA
 *                       All rights reserved.
 *
 * Programmer:  Robb Matzke <matzke@llnl.gov>
 *              Thursday, July 29, 1999
 *
 * Purpose:	This is the MPI-2 I/O driver.
 *
 * Limitations:
 *     	H5FD_mpio_read & H5FD_mpio_write
 *		Eventually these should choose collective or independent i/o
 *		based on a parameter that is passed down to it from H5Dwrite,
 *		rather than the access_parms (which are fixed at the open).
 *
 *	H5FD_mpio_read
 *		One implementation of MPI/MPI-IO causes MPI_Get_count
 *		to return (incorrectly) a negative count. I (who?) added code
 *		to detect this, and a kludge to pretend that the number of
 *		bytes read is always equal to the number requested.  This
 *		kluge is activated by #ifdef MPI_KLUGE0202.
 */
#include "H5private.h"		/*library functions			*/
#include "H5Eprivate.h"		/*error handling			*/
#include "H5Fprivate.h"		/*files					*/
#include "H5FDprivate.h"	/*file driver				  */
#include "H5FDmpio.h"           /* MPI I/O file driver */
#include "H5Iprivate.h"		/*object IDs				  */
#include "H5MMprivate.h"        /* Memory allocation */
#include "H5Pprivate.h"		/*property lists			*/

/*
 * The driver identification number, initialized at runtime if H5_HAVE_PARALLEL
 * is defined. This allows applications to still have the H5FD_MPIO
 * "constants" in their source code (it also makes this file strictly ANSI
 * compliant when H5_HAVE_PARALLEL isn't defined)
 */
static hid_t H5FD_MPIO_g = 0;

#ifdef H5_HAVE_PARALLEL

/*
 * The description of a file belonging to this driver.
 * The EOF value is only used just after the file is opened in order for the
 * library to determine whether the file is empty, truncated, or okay. The MPIO
 * driver doesn't bother to keep it updated since it's an expensive operation.
 */
typedef struct H5FD_mpio_t {
    H5FD_t	pub;		/*public stuff, must be first		*/
    MPI_File	f;		/*MPIO file handle			*/
    MPI_Comm	comm;		/*communicator				*/
    MPI_Info	info;		/*file information			*/
    int         mpi_rank;       /* This process's rank                  */
    int         mpi_size;       /* Total number of processes            */
    int         mpi_round;      /* Current round robin process (for metadata I/O) */
    haddr_t	eof;		/*end-of-file marker			*/
    haddr_t	eoa;		/*end-of-address marker			*/
    haddr_t	last_eoa;	/* Last known end-of-address marker	*/
    MPI_Datatype btype;		/*buffer type for xfers			*/
    MPI_Datatype ftype;		/*file type for xfers			*/
    haddr_t	 disp;		/*displacement for set_view in xfers	*/
    int		 use_types;	/*if !0, use btype, ftype, disp.else do
				 * simple byteblk xfer		
				 */
    int		old_use_types; /*remember value of use_types		*/
} H5FD_mpio_t;

/* Prototypes */
static haddr_t H5FD_mpio_MPIOff_to_haddr(MPI_Offset mpi_off);
static herr_t H5FD_mpio_haddr_to_MPIOff(haddr_t addr, MPI_Offset *mpi_off/*out*/);

/* Callbacks */
static void *H5FD_mpio_fapl_get(H5FD_t *_file);
static H5FD_t *H5FD_mpio_open(const char *name, unsigned flags, hid_t fapl_id,
			      haddr_t maxaddr);
static herr_t H5FD_mpio_close(H5FD_t *_file);
static herr_t H5FD_mpio_query(const H5FD_t *_f1, unsigned long *flags);
static haddr_t H5FD_mpio_get_eoa(H5FD_t *_file);
static herr_t H5FD_mpio_set_eoa(H5FD_t *_file, haddr_t addr);
static haddr_t H5FD_mpio_get_eof(H5FD_t *_file);
static herr_t H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t type, hid_t fapl_id, haddr_t addr,
			     size_t size, void *buf);
static herr_t H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t fapl_id, haddr_t addr,
			      size_t size, const void *buf);
static herr_t H5FD_mpio_flush(H5FD_t *_file, unsigned closing);

/* MPIO-specific file access properties */
typedef struct H5FD_mpio_fapl_t {
    MPI_Comm		comm;		/*communicator			*/
    MPI_Info		info;		/*file information		*/
} H5FD_mpio_fapl_t;

/* The MPIO file driver information */
static const H5FD_class_t H5FD_mpio_g = {
    "mpio",					/*name			*/
    HADDR_MAX,					/*maxaddr		*/
    H5F_CLOSE_SEMI,				/* fc_degree		*/
    NULL,					/*sb_size		*/
    NULL,					/*sb_encode		*/
    NULL,					/*sb_decode		*/
    sizeof(H5FD_mpio_fapl_t),			/*fapl_size		*/
    H5FD_mpio_fapl_get,				/*fapl_get		*/
    NULL,					/*fapl_copy		*/
    NULL, 					/*fapl_free		*/
    sizeof(H5FD_mpio_dxpl_t),			/*dxpl_size		*/
    NULL,					/*dxpl_copy		*/
    NULL,					/*dxpl_free		*/
    H5FD_mpio_open,				/*open			*/
    H5FD_mpio_close,				/*close			*/
    NULL,					/*cmp			*/
    H5FD_mpio_query,		                /*query			*/
    NULL,					/*alloc			*/
    NULL,					/*free			*/
    H5FD_mpio_get_eoa,				/*get_eoa		*/
    H5FD_mpio_set_eoa, 				/*set_eoa		*/
    H5FD_mpio_get_eof,				/*get_eof		*/
    H5FD_mpio_read,				/*read			*/
    H5FD_mpio_write,				/*write			*/
    H5FD_mpio_flush,				/*flush			*/
    H5FD_FLMAP_SINGLE,				/*fl_map		*/
};

#ifdef H5FDmpio_DEBUG
/* Flags to control debug actions in H5Fmpio.
 * Meant to be indexed by characters.
 *
 * 'c' show result of MPI_Get_count after read
 * 'r' show read offset and size
 * 't' trace function entry and exit
 * 'w' show write offset and size
 */
static int H5FD_mpio_Debug[256] =
        { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
          0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };
#endif

/* Global var to allow elimination of redundant metadata writes
 * to be controlled by the value of an environment variable. */
/* Use the elimination by default unless this is the Intel Red machine */
#ifndef __PUMAGON__
hbool_t	H5_mpi_1_metawrite_g = TRUE;
#else
hbool_t	H5_mpi_1_metawrite_g = FALSE;
#endif

/* Interface initialization */
#define PABLO_MASK	H5FD_mpio_mask
#define INTERFACE_INIT	H5FD_mpio_init
static int interface_initialize_g = 0;


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_init
 *
 * Purpose:	Initialize this driver by registering the driver with the
 *		library.
 *
 * Return:	Success:	The driver ID for the mpio driver.
 *
 *		Failure:	Negative.
 *
 * Programmer:	Robb Matzke
 *              Thursday, August 5, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
hid_t
H5FD_mpio_init(void)
{
    FUNC_ENTER_NOAPI(H5FD_mpio_init, FAIL);

    if (H5I_VFL!=H5Iget_type(H5FD_MPIO_g))
        H5FD_MPIO_g = H5FDregister(&H5FD_mpio_g);

    FUNC_LEAVE(H5FD_MPIO_g);
}


/*-------------------------------------------------------------------------
 * Function:	H5Pset_fapl_mpio
 *
 * Purpose:	Store the user supplied MPIO communicator COMM and INFO in
 *		the file access property list FAPL_ID which can then be used
 *		to create and/or open the file.  This function is available
 *		only in the parallel HDF5 library and is not collective.
 *
 *		COMM is the MPI communicator to be used for file open as
 *		defined in MPI_FILE_OPEN of MPI-2. This function does not
 *		make a duplicated communicator. Any modification to COMM
 *		after this function call returns may have undetermined effect
 *		on the access property list. Users should not modify the
 *		communicator while it is defined in a property list.
 *
 *		INFO is the MPI info object to be used for file open as
 *		defined in MPI_FILE_OPEN of MPI-2. This function does not
 *		make a duplicated info. Any modification to info after this
 *		function call returns may have undetermined effect on the
 *		access property list. Users should not modify the info while
 *		it is defined in a property list.
 *
 * Return:	Success:	Non-negative
 *
 * 		Failure:	Negative
 *
 * Programmer:	Albert Cheng
 *		Feb 3, 1998
 *
 * Modifications:
 *		Robb Matzke, 1998-02-18
 *		Check all arguments before the property list is updated so we
 *		don't leave the property list in a bad state if something
 *		goes wrong.  Also, the property list data type changed to
 *		allow more generality so all the mpi-related stuff is in the
 *		`u.mpi' member.  The `access_mode' will contain only
 * 		mpi-related flags defined in H5Fpublic.h.
 *
 *		Albert Cheng, 1998-04-16
 *		Removed the ACCESS_MODE argument.  The access mode is changed
 *		to be controlled by data transfer property list during data
 *		read/write calls.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *
 *		Raymond Lu 
 * 		Tuesday, Oct 23, 2001
 *		Changed the file access list to the new generic property 
 *		list.
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5Pset_fapl_mpio(hid_t fapl_id, MPI_Comm comm, MPI_Info info)
{
    H5FD_mpio_fapl_t	fa;
    H5P_genplist_t *plist;      /* Property list pointer */
    herr_t ret_value=FAIL;
    
    FUNC_ENTER_API(H5Pset_fapl_mpio, FAIL);
    H5TRACE3("e","iMcMi",fapl_id,comm,info);

    /* Check arguments */
    if(TRUE!=H5P_isa_class(fapl_id,H5P_FILE_ACCESS) || NULL == (plist = H5I_object(fapl_id)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a file access list");


#ifdef LATER
#warning "We need to verify that COMM and INFO contain sensible information."
#endif

    /* Initialize driver specific properties */
    fa.comm = comm;
    fa.info = info;

    ret_value= H5P_set_driver(plist, H5FD_MPIO, &fa);

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5Pget_fapl_mpio
 *
 * Purpose:	If the file access property list is set to the H5FD_MPIO
 *		driver then this function returns the MPI communicator and
 *		information through the COMM and INFO pointers.
 *
 * Return:	Success:	Non-negative with the communicator and
 *				information returned through the COMM and
 *				INFO arguments if non-null. Neither piece of
 *				information is copied and they are therefore
 *				valid only until the file access property
 *				list is modified or closed.
 *
 * 		Failure:	Negative
 *
 * Programmer:	Robb Matzke
 *		Thursday, February 26, 1998
 *
 * Modifications:
 *
 *	        Albert Cheng, Apr 16, 1998
 *	        Removed the access_mode argument.  The access_mode is changed
 *	        to be controlled by data transfer property list during data
 *	        read/write calls.
 *
 *		Raymond Lu 
 * 		Tuesday, Oct 23, 2001
 *		Changed the file access list to the new generic property 
 *		list.
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5Pget_fapl_mpio(hid_t fapl_id, MPI_Comm *comm/*out*/, MPI_Info *info/*out*/)
{
    H5FD_mpio_fapl_t	*fa;
    H5P_genplist_t *plist;      /* Property list pointer */
    
    FUNC_ENTER_API(H5Pget_fapl_mpio, FAIL);
    H5TRACE3("e","ixx",fapl_id,comm,info);

    if(TRUE!=H5P_isa_class(fapl_id,H5P_FILE_ACCESS) || NULL == (plist = H5I_object(fapl_id)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a file access list");
    if (H5FD_MPIO!=H5P_get_driver(plist))
        HRETURN_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "incorrect VFL driver");
    if (NULL==(fa=H5P_get_driver_info(plist)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "bad VFL driver info");

    if (comm)
        *comm = fa->comm;
    if (info)
        *info = fa->info;

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5Pset_dxpl_mpio
 *
 * Purpose:	Set the data transfer property list DXPL_ID to use transfer
 *		mode XFER_MODE. The property list can then be used to control
 *		the I/O transfer mode during data I/O operations. The valid
 *		transfer modes are:
 *
 * 		H5FD_MPIO_INDEPENDENT:
 *			Use independent I/O access (the default).
 *
 * 		H5FD_MPIO_COLLECTIVE:
 *			Use collective I/O access.
 *			
 * Return:	Success:	Non-negative
 *
 * 		Failure:	Negative
 *
 * Programmer:	Albert Cheng
 *		April 2, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
herr_t
H5Pset_dxpl_mpio(hid_t dxpl_id, H5FD_mpio_xfer_t xfer_mode)
{
    H5FD_mpio_dxpl_t	dx;
    H5P_genplist_t *plist;      /* Property list pointer */
    herr_t ret_value=FAIL;

    FUNC_ENTER_API(H5Pset_dxpl_mpio, FAIL);
    H5TRACE2("e","iDt",dxpl_id,xfer_mode);
    
    /* Check arguments */
    if(TRUE!=H5P_isa_class(dxpl_id,H5P_DATASET_XFER) || NULL == (plist = H5I_object(dxpl_id)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl");
    if (H5FD_MPIO_INDEPENDENT!=xfer_mode && H5FD_MPIO_COLLECTIVE!=xfer_mode)
        HRETURN_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "incorrect xfer_mode");

    /* Initialize driver-specific properties */
    dx.xfer_mode = xfer_mode;

    ret_value= H5P_set_driver(plist, H5FD_MPIO, &dx);

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5Pget_dxpl_mpio
 *
 * Purpose:	Queries the transfer mode current set in the data transfer
 *		property list DXPL_ID. This is not collective.
 *
 * Return:	Success:	Non-negative, with the transfer mode returned
 *				through the XFER_MODE argument if it is
 *				non-null.
 *
 * 		Failure:	Negative
 *
 * Programmer:	Albert Cheng
 *		April 2, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
herr_t
H5Pget_dxpl_mpio(hid_t dxpl_id, H5FD_mpio_xfer_t *xfer_mode/*out*/)
{
    H5FD_mpio_dxpl_t	*dx;
    H5P_genplist_t *plist;      /* Property list pointer */

    FUNC_ENTER_API(H5Pget_dxpl_mpio, FAIL);
    H5TRACE2("e","ix",dxpl_id,xfer_mode);

    if(TRUE!=H5P_isa_class(dxpl_id,H5P_DATASET_XFER) || NULL == (plist = H5I_object(dxpl_id)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl");
    if (H5FD_MPIO!=H5P_get_driver(plist))
        HRETURN_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "incorrect VFL driver");
    if (NULL==(dx=H5P_get_driver_info(plist)))
        HRETURN_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "bad VFL driver info");

    if (xfer_mode)
        *xfer_mode = dx->xfer_mode;

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_communicator
 *
 * Purpose:	Returns the MPI communicator for the file.
 *
 * Return:	Success:	The communicator
 *
 *		Failure:	NULL
 *
 * Programmer:	Robb Matzke
 *              Monday, August  9, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
MPI_Comm
H5FD_mpio_communicator(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_communicator, MPI_COMM_NULL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    FUNC_LEAVE(file->comm);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_mpi_rank
 *
 * Purpose:	Returns the MPI rank for a process
 *
 * Return:	Success: non-negative
 *		Failure: negative
 *
 * Programmer:	Quincey Koziol
 *              Thursday, May 16, 2002
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
int
H5FD_mpio_mpi_rank(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_mpi_rank, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    FUNC_LEAVE(file->mpi_rank);
} /* end H5FD_mpio_mpi_rank() */


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_mpi_size
 *
 * Purpose:	Returns the number of MPI processes
 *
 * Return:	Success: non-negative
 *		Failure: negative
 *
 * Programmer:	Quincey Koziol
 *              Thursday, May 16, 2002
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
int
H5FD_mpio_mpi_size(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_mpi_rank, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    FUNC_LEAVE(file->mpi_size);
} /* end H5FD_mpio_mpi_size() */


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_setup
 *
 * Purpose:	Set the buffer type BTYPE, file type FTYPE, and absolute base
 *		address DISP (i.e., the file view displacement) for a data
 *		transfer. Also request a dataspace transfer or an elementary
 *		byteblock transfer depending on whether USE_TYPES is non-zero
 *		or zero, respectively.
 *
 * Return:	Success:	0
 *
 *		Failure:	-1
 *
 * Programmer:	Robb Matzke
 *              Monday, August  9, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5FD_mpio_setup(H5FD_t *_file, MPI_Datatype btype, MPI_Datatype ftype,
		haddr_t disp, hbool_t use_types)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_setup, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    file->btype = btype;
    file->ftype = ftype;
    file->disp = disp;
    file->use_types = use_types;

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_wait_for_left_neighbor
 *
 * Purpose:	Blocks until (empty) msg is received from immediately
 *		lower-rank neighbor. In conjunction with
 *		H5FD_mpio_signal_right_neighbor, useful for enforcing
 *		1-process-at-at-time access to critical regions to avoid race
 *		conditions (though it is overkill to require that the
 *		processes be allowed to proceed strictly in order of their
 *		rank).
 *
 * Note:	This routine doesn't read or write any file, just performs
 *		interprocess coordination. It really should reside in a
 *		separate package of such routines.
 *
 * Return:	Success:	0
 *		Failure:	-1
 *
 * Programmer:	rky
 *              19981207
 *
 * Modifications:
 *		Robb Matzke, 1999-08-09
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
herr_t
H5FD_mpio_wait_for_left_neighbor(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;
    char msgbuf[1];
    MPI_Status rcvstat;

    FUNC_ENTER_NOAPI(H5FD_mpio_wait_for_left_neighbor, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    /* p0 has no left neighbor; all other procs wait for msg */
    if (file->mpi_rank != 0) {
        if (MPI_SUCCESS!= MPI_Recv( &msgbuf, 1, MPI_CHAR, file->mpi_rank-1, MPI_ANY_TAG, file->comm, &rcvstat ))
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Recv failed");
    }
    
    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_signal_right_neighbor
 *
 * Purpose:	Blocks until (empty) msg is received from immediately
 *		lower-rank neighbor. In conjunction with
 *		H5FD_mpio_wait_for_left_neighbor, useful for enforcing
 *		1-process-at-at-time access to critical regions to avoid race
 *		conditions (though it is overkill to require that the
 *		processes be allowed to proceed strictly in order of their
 *		rank).
 *
 * Note: 	This routine doesn't read or write any file, just performs
 *		interprocess coordination. It really should reside in a
 *		separate package of such routines.
 *
 * Return:	Success:	0
 *		Failure:	-1
 *
 * Programmer:	rky
 *              19981207
 *
 * Modifications:
 *		Robb Matzke, 1999-08-09
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
herr_t
H5FD_mpio_signal_right_neighbor(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;
    char msgbuf[1];

    FUNC_ENTER_NOAPI(H5FD_mpio_signal_right_neighbor, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    if (file->mpi_rank != (file->mpi_size-1)) {
        if (MPI_SUCCESS!= MPI_Send(&msgbuf, 0/*empty msg*/, MPI_CHAR, file->mpi_rank+1, 0, file->comm))
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Send failed");
    }
    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_fapl_get
 *
 * Purpose:	Returns a file access property list which could be used to
 *		create another file the same as this one.
 *
 * Return:	Success:	Ptr to new file access property list with all
 *				fields copied from the file pointer.
 *
 *		Failure:	NULL
 *
 * Programmer:	Robb Matzke
 *              Friday, August 13, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static void *
H5FD_mpio_fapl_get(H5FD_t *_file)
{
    H5FD_mpio_t		*file = (H5FD_mpio_t*)_file;
    H5FD_mpio_fapl_t	*fa = NULL;

    FUNC_ENTER_NOAPI(H5FD_mpio_fapl_get, NULL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    if (NULL==(fa=H5MM_calloc(sizeof(H5FD_mpio_fapl_t))))
        HRETURN_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed");

    /* These should both be copied. --rpm, 1999-08-13 */
    fa->comm = file->comm;
    fa->info = file->info;

    FUNC_LEAVE(fa);
}


/*-------------------------------------------------------------------------
 * Function:    H5FD_mpio_open
 *
 * Purpose:     Opens a file with name NAME.  The FLAGS are a bit field with
 *		purpose similar to the second argument of open(2) and which
 *		are defined in H5Fpublic.h. The file access property list
 *		FAPL_ID contains the properties driver properties and MAXADDR
 *		is the largest address which this file will be expected to
 *		access.  This is collective.
 *
 * Return:      Success:        A new file pointer.
 *
 *              Failure:        NULL
 *
 * Programmer:  
 *              January 30, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1998-02-18
 *		Added the ACCESS_PARMS argument.  Moved some error checking
 *		here from elsewhere.
 *
 *      	rky, 1998-01-11
 *      	Added H5FD_mpio_Debug debug flags controlled by MPI_Info.
 *
 *      	rky, 1998-08-28
 *		Init flag controlling redundant metadata writes to disk.
 *
 *      	rky, 1998-12-07
 *		Added barrier after MPI_File_set_size to prevent race
 *		condition -- subsequent writes were being truncated, causing
 *		holes in file.
 *		
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *
 *      	rky & ppw, 1999-11-07
 *		Modified "H5FD_mpio_open" so that file-truncation is
 *              avoided for brand-new files (with zero filesize).
 *-------------------------------------------------------------------------
 */
static H5FD_t *
H5FD_mpio_open(const char *name, unsigned flags, hid_t fapl_id,
	       haddr_t maxaddr/*unused*/)
{
    H5FD_mpio_t			*file=NULL;
    MPI_File			fh;
    int				mpi_amode;
    int				mpi_rank;       /* MPI rank of this process */
    int				mpi_size;       /* Total number of MPI processes */
    int				mpi_code;	/* mpi return code */
    MPI_Offset			size;
    const H5FD_mpio_fapl_t	*fa=NULL;
    H5FD_mpio_fapl_t		_fa;
    H5P_genplist_t *plist;      /* Property list pointer */

    FUNC_ENTER_NOAPI(H5FD_mpio_open, NULL);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t']) {
    	fprintf(stdout, "Entering H5FD_mpio_open(name=\"%s\", flags=0x%x, "
		"fapl_id=%d, maxaddr=%lu)\n", name, flags, (int)fapl_id, (unsigned long)maxaddr);
    }
#endif

    /* Obtain a pointer to mpio-specific file access properties */
    if(TRUE!=H5P_isa_class(fapl_id,H5P_FILE_ACCESS) || NULL == (plist = H5I_object(fapl_id)))
        HRETURN_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list");
    if (H5P_DEFAULT==fapl_id || H5FD_MPIO!=H5P_get_driver(plist)) {
	_fa.comm = MPI_COMM_SELF; /*default*/
	_fa.info = MPI_INFO_NULL; /*default*/
	fa = &_fa;
    } else {
	fa = H5P_get_driver_info(plist);
	assert(fa);
    }

    /* convert HDF5 flags to MPI-IO flags */
    /* some combinations are illegal; let MPI-IO figure it out */
    mpi_amode  = (flags&H5F_ACC_RDWR) ? MPI_MODE_RDWR : MPI_MODE_RDONLY;
    if (flags&H5F_ACC_CREAT)	mpi_amode |= MPI_MODE_CREATE;
    if (flags&H5F_ACC_EXCL)	mpi_amode |= MPI_MODE_EXCL;

#ifdef H5FDmpio_DEBUG
    {
	/* set debug mask */
	/* Should this be done in H5F global initialization instead of here? */
        const char *s = HDgetenv ("H5FD_mpio_Debug");
        if (s) {
	    while (*s){
		H5FD_mpio_Debug[(int)*s]++;
		s++;
	    }
        }
    }
    
    /* Check for debug commands in the info parameter */
#if 0
    /* Temporary KLUGE rky 2000-06-29, because fa->info is invalid (-1)*/
    {
	char debug_str[128];
        int infoerr, flag, i;
        if (fa->info) {
            infoerr = MPI_Info_get(fa->info, H5F_MPIO_DEBUG_KEY, 127,
				   debug_str, &flag);
            if (flag) {
                fprintf(stdout, "H5FD_mpio debug flags=%s\n", debug_str );
                for (i=0;
                     debug_str[i]/*end of string*/ && i<128/*just in case*/;
                     ++i) {
                    H5FD_mpio_Debug[(int)debug_str[i]] = 1;
                }
            }
        }
    }
    /* END Temporary KLUGE rky 2000-06-29, because fa->info is invalid (-1) */
#endif
#endif

    /*OKAY: CAST DISCARDS CONST*/
    if (MPI_SUCCESS !=
	(mpi_code=MPI_File_open(fa->comm, (char*)name, mpi_amode, fa->info, &fh)))
	    HMPI_RETURN_ERROR(NULL, "MPI_File_open failed", mpi_code);

    /* Get the MPI rank of this process and the total number of processes */
    if (MPI_SUCCESS != MPI_Comm_rank (fa->comm, &mpi_rank))
          HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_Comm_rank failed");
    if (MPI_SUCCESS != MPI_Comm_size (fa->comm, &mpi_size))
          HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_Comm_size failed");

/*  Following changes in handling file-truncation made be rkyates and ppweidhaas, sep 99  */

    /* Only processor p0 will get the filesize and broadcast it. */
    if (mpi_rank == 0) {
      /* Get current file size */
      if (MPI_SUCCESS != MPI_File_get_size(fh, &size)) {
          MPI_File_close(&fh);
          HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_File_get_size failed");
      }
    }

    /* Broadcast file-size */
    if (MPI_SUCCESS != MPI_Bcast(&size, sizeof(MPI_Offset), MPI_BYTE, 0, fa->comm))
          HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_Bcast failed");

    /* Only if size > 0, truncate the file - if requested */
    if (size && (flags & H5F_ACC_TRUNC)) {
        if (MPI_SUCCESS != MPI_File_set_size(fh, (MPI_Offset)0)) {
            MPI_File_close(&fh);
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_File_set_size failed");
        }

	/* Don't let any proc return until all have truncated the file. */
        if (MPI_SUCCESS!= MPI_Barrier(fa->comm)) {
            MPI_File_close(&fh);
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, NULL, "MPI_Barrier failed");
        }
        size = 0;
    }

    /* Build the return value and initialize it */
    if (NULL==(file=H5MM_calloc(sizeof(H5FD_mpio_t))))
        HRETURN_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed");

    file->f = fh;
    file->comm = fa->comm;
    file->info = fa->info;
    file->mpi_rank = mpi_rank;
    file->mpi_size = mpi_size;
    file->eof = H5FD_mpio_MPIOff_to_haddr(size);
    file->btype = MPI_DATATYPE_NULL;
    file->ftype = MPI_DATATYPE_NULL;

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t']) {
	fprintf(stdout, "Leaving H5FD_mpio_open\n" );
    }
#endif

    FUNC_LEAVE((H5FD_t*)file);
}


/*-------------------------------------------------------------------------
 * Function:    H5FD_mpio_close
 *
 * Purpose:     Closes a file.  This is collective.
 *
 * Return:      Success:	Non-negative
 *
 * 		Failure:	Negative
 *
 * Programmer:  Unknown
 *              January 30, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1998-02-18
 *		Added the ACCESS_PARMS argument.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_close(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_close, FAIL);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Entering H5FD_mpio_close\n");
#endif
    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    /* MPI_File_close sets argument to MPI_FILE_NULL */
    if (MPI_SUCCESS != MPI_File_close(&(file->f)/*in,out*/))
        HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_close failed");

    /* Clean up other stuff */
    H5MM_xfree(file);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Leaving H5FD_mpio_close\n");
#endif

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_query
 *
 * Purpose:	Set the flags that this VFL driver is capable of supporting.
 *              (listed in H5FDpublic.h)
 *
 * Return:	Success:	non-negative
 *
 *		Failure:	negative
 *
 * Programmer:	Quincey Koziol
 *              Friday, August 25, 2000
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_query(const H5FD_t *_file, unsigned long *flags /* out */)
{
    const H5FD_mpio_t	*file = (const H5FD_mpio_t*)_file;
    herr_t ret_value=SUCCEED;

    FUNC_ENTER_NOAPI(H5FD_mpio_query, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    /* Set the VFL feature flags that this driver supports */
    if(flags) {
        *flags=0;
        *flags|=H5FD_FEAT_AGGREGATE_METADATA; /* OK to aggregate metadata allocations */
    } /* end if */

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_get_eoa
 *
 * Purpose:	Gets the end-of-address marker for the file. The EOA marker
 *		is the first address past the last byte allocated in the
 *		format address space.
 *
 * Return:	Success:	The end-of-address marker.
 *
 *		Failure:	HADDR_UNDEF
 *
 * Programmer:	Robb Matzke
 *              Friday, August  6, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static haddr_t
H5FD_mpio_get_eoa(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_get_eoa, HADDR_UNDEF);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    FUNC_LEAVE(file->eoa);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_set_eoa
 *
 * Purpose:	Set the end-of-address marker for the file. This function is
 *		called shortly after an existing HDF5 file is opened in order
 *		to tell the driver where the end of the HDF5 data is located.
 *
 * Return:	Success:	0
 *
 *		Failure:	-1
 *
 * Programmer:	Robb Matzke
 *              Friday, August 6, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_set_eoa(H5FD_t *_file, haddr_t addr)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_set_eoa, FAIL);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    file->eoa = addr;

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_get_eof
 *
 * Purpose:	Gets the end-of-file marker for the file. The EOF marker
 *		is the real size of the file.
 *
 *		The MPIO driver doesn't bother keeping this field updated
 *		since that's a relatively expensive operation. Fortunately
 *		the library only needs the EOF just after the file is opened
 *		in order to determine whether the file is empty, truncated,
 *		or okay.  Therefore, any MPIO I/O function will set its value
 *		to HADDR_UNDEF which is the error return value of this
 *		function.
 *
 * Return:	Success:	The end-of-address marker.
 *
 *		Failure:	HADDR_UNDEF
 *
 * Programmer:	Robb Matzke
 *              Friday, August  6, 1999
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static haddr_t
H5FD_mpio_get_eof(H5FD_t *_file)
{
    H5FD_mpio_t	*file = (H5FD_mpio_t*)_file;

    FUNC_ENTER_NOAPI(H5FD_mpio_get_eof, HADDR_UNDEF);

    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    FUNC_LEAVE(file->eof);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_read
 *
 * Purpose:	Reads SIZE bytes of data from FILE beginning at address ADDR
 *		into buffer BUF according to data transfer properties in
 *		DXPL_ID using potentially complex file and buffer types to
 *		effect the transfer.
 *
 *		Reading past the end of the MPI file returns zeros instead of
 *		failing.  MPI is able to coalesce requests from different
 *		processes (collective or independent).
 *
 * Return:	Success:	Zero. Result is stored in caller-supplied
 *				buffer BUF.
 *
 *		Failure:	-1, Contents of buffer BUF are undefined.
 *
 * Programmer:	rky, 1998-01-30
 *
 * Modifications:
 * 		Robb Matzke, 1998-02-18
 *		Added the ACCESS_PARMS argument.
 *
 * 		rky, 1998-04-10
 *		Call independent or collective MPI read, based on
 *		ACCESS_PARMS.
 *
 * 		Albert Cheng, 1998-06-01
 *		Added XFER_MODE to control independent or collective MPI
 *		read.
 *
 * 		rky, 1998-08-16
 *		Use BTYPE, FTYPE, and DISP from access parms. The guts of
 *		H5FD_mpio_read and H5FD_mpio_write should be replaced by a
 *		single dual-purpose routine.
 *
 * 		Robb Matzke, 1999-04-21
 *		Changed XFER_MODE to XFER_PARMS for all H5F_*_read()
 *		callbacks.
 *
 * 		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *		
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *
 *		Quincey Koziol,  2002-05-14
 *		Only call MPI_Get_count if we can use MPI_BYTE for the MPI type
 *              for the I/O transfer.  Someday we might include code to decode
 *              the MPI type used for more complicated transfers and call
 *              MPI_Get_count all the time.
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t dxpl_id, haddr_t addr, size_t size,
	       void *buf/*out*/)
{
    H5FD_mpio_t			*file = (H5FD_mpio_t*)_file;
    const H5FD_mpio_dxpl_t	*dx=NULL;
    H5FD_mpio_dxpl_t		_dx;
    MPI_Offset			mpi_off, mpi_disp;
    MPI_Status  		mpi_stat = {0};
    MPI_Datatype		buf_type, file_type;
    int         		size_i, bytes_read, n;
    int				use_types_this_time, used_types_last_time;
    H5P_genplist_t *plist;      /* Property list pointer */
    herr_t              	ret_value=SUCCEED;

    FUNC_ENTER_NOAPI(H5FD_mpio_read, FAIL);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Entering H5FD_mpio_read\n" );
#endif
    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);
    /* Make certain we have the correct type of property list */
    assert(H5I_GENPROP_LST==H5I_get_type(dxpl_id));
    assert(TRUE==H5P_isa_class(dxpl_id,H5P_DATASET_XFER));

    /* some numeric conversions */
    if (H5FD_mpio_haddr_to_MPIOff(addr, &mpi_off/*out*/)<0)
        HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off");
    size_i = (int)size;
    if ((hsize_t)size_i != size)
        HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from size to size_i");

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'r'])
        fprintf(stdout, "in H5FD_mpio_read  mpi_off=%ld  size_i=%d\n",
		(long)mpi_off, size_i );
#endif

    /* Obtain the data transfer properties */
    if(NULL == (plist = H5I_object(dxpl_id)))
        HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list");
    if (H5FD_MPIO!=H5P_get_driver(plist)) {
        _dx.xfer_mode = H5FD_MPIO_INDEPENDENT; /*the default*/
        dx = &_dx;
    } else {
        dx = H5P_get_driver_info(plist);
        assert(dx);
    }
    
    /*
     * Set up for a fancy xfer using complex types, or single byte block. We
     * wouldn't need to rely on the use_types field if MPI semantics allowed
     * us to test that btype=ftype=MPI_BYTE (or even MPI_TYPE_NULL, which
     * could mean "use MPI_BYTE" by convention).
     */
    use_types_this_time = file->use_types;
    if (use_types_this_time) {
        /* prepare for a full-blown xfer using btype, ftype, and disp */
        buf_type = file->btype;
        file_type = file->ftype;
        if (H5FD_mpio_haddr_to_MPIOff(file->disp, &mpi_disp)<0)
            HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off");
    } else {
        /*
         * Prepare for a simple xfer of a contiguous block of bytes. The
         * btype, ftype, and disp fields are not used.
         */
        buf_type = MPI_BYTE;
        file_type = MPI_BYTE;
        mpi_disp = 0;		/* mpi_off is sufficient */
    }

    /*
     * Don't bother to reset the view if we're not using the types this time,
     * and did we didn't use them last time either.
     */
    used_types_last_time = file->old_use_types;
    if (used_types_last_time || /* change to new ftype or MPI_BYTE */
            use_types_this_time) { 	/* almost certainly a different ftype */
        /*OKAY: CAST DISCARDS CONST QUALIFIER*/
        if (MPI_SUCCESS != MPI_File_set_view(file->f, mpi_disp, MPI_BYTE, file_type, (char*)"native",  file->info))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
    }
    
    /*
     * We always set the use_types flag to 0 because the default is not to
     * use types next time, unless someone explicitly requests it by setting
     * this flag to !=0.
     */
    file->old_use_types = use_types_this_time;
    file->use_types = 0;

    /* Read the data. */
    assert(H5FD_MPIO_INDEPENDENT==dx->xfer_mode || H5FD_MPIO_COLLECTIVE==dx->xfer_mode);
    if (H5FD_MPIO_INDEPENDENT==dx->xfer_mode) {
        if (MPI_SUCCESS!= MPI_File_read_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_read_at failed");
    } else {
#ifdef H5FDmpio_DEBUG
	if (H5FD_mpio_Debug[(int)'t'])
	    fprintf(stdout, "H5FD_mpio_read: using MPIO collective mode\n");
#endif
        if (MPI_SUCCESS!= MPI_File_read_at_all(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat ))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_read_at_all failed");
    }

    /* KLUDGE, Robb Matzke, 2000-12-29
     * The LAM implementation of MPI_Get_count() says
     *    MPI_Get_count: invalid argument (rank 0, MPI_COMM_WORLD)
     * So I'm commenting this out until it can be investigated. The
     * returned `bytes_written' isn't used anyway because of Kim's
     * kludge to avoid bytes_written<0. Likewise in H5FD_mpio_write(). */

#ifdef H5_HAVE_MPI_GET_COUNT /* Bill and Albert's kludge*/
    /* Yet Another KLUDGE, Albert Cheng & Bill Wendling, 2001-05-11.
     * Many systems don't support MPI_Get_count so we need to do a
     * configure thingy to fix this. */

    /* Calling MPI_Get_count with "MPI_BYTE" is only valid when we actually
     * had the 'buf_type' set to MPI_BYTE -QAK
     */
    if(use_types_this_time) {
        /* Figure out the mapping from the MPI 'buf_type' to bytes, someday...
         * If this gets fixed (and MPI_Get_count() is reliable), the
         * kludge below where the 'bytes_read' value from MPI_Get_count() is
         * overwritten with the 'size_i' parameter can be removed. -QAK
         */
    } /* end if */
    else {
        /* How many bytes were actually read? */
        if (MPI_SUCCESS != MPI_Get_count(&mpi_stat, MPI_BYTE, &bytes_read))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Get_count failed");
    } /* end else */
#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'c'])
    	fprintf(stdout,
	    "In H5FD_mpio_read after Get_count size_i=%d bytes_read=%d\n",
	    size_i, bytes_read );
#endif
#endif /* H5_HAVE_MPI_GET_COUNT */

    /*
     * KLUGE rky 1998-02-02
     * MPI_Get_count incorrectly returns negative count; fake a complete
     * read.
     */
    bytes_read = size_i;

    /* Check for read failure */
    if (bytes_read<0 || bytes_read>size_i)
        HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed");

    /*
     * This gives us zeroes beyond end of physical MPI file.  What about
     * reading past logical end of HDF5 file???
     */
    if ((n=(size_i-bytes_read)) > 0) {
        if (use_types_this_time) {
            /*
             * INCOMPLETE rky 1998-09-18
             * Haven't implemented reading zeros beyond EOF. What to do???
             */
            HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "eof file read failed");
        } else {
            memset((char*)buf+bytes_read, 0, (size_t)n);
        }
    }

done:
#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Leaving H5FD_mpio_read\n" );
#endif

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5FD_mpio_write
 *
 * Purpose:	Writes SIZE bytes of data to FILE beginning at address ADDR
 *		from buffer BUF according to data transfer properties in
 *		DXPL_ID using potentially complex file and buffer types to
 *		effect the transfer.
 *
 *		MPI is able to coalesce requests from different processes
 *		(collective and independent).
 *
 * Return:	Success:	Zero. USE_TYPES and OLD_USE_TYPES in the
 *				access params are altered.
 *
 *		Failure:	-1, USE_TYPES and OLD_USE_TYPES in the
 *				access params may be altered.
 *
 * Programmer:	Unknown
 *              January 30, 1998
 *
 * Modifications:
 *		rky, 1998-08-28
 *		If the file->allsame flag is set, we assume that all the
 *		procs in the relevant MPI communicator will write identical
 *		data at identical offsets in the file, so only proc 0 will
 *		write, and all other procs will wait for p0 to finish. This
 *		is useful for writing metadata, for example. Note that we
 *		don't _check_ that the data is identical. Also, the mechanism
 *		we use to eliminate the redundant writes is by requiring a
 *		call to H5FD_mpio_tas_allsame before the write, which is
 *		rather klugey. Would it be better to pass a parameter to
 *		low-level writes like H5F_block_write and H5F_low_write,
 *		instead?  Or...??? Also, when I created this mechanism I
 *		wanted to minimize the difference in behavior between the old
 *		way of doing things (i.e., all procs write) and the new way,
 *		so the writes are eliminated at the very lowest level, here
 *		in H5FD_mpio_write. It may be better to rethink that, and
 *		short-circuit the writes at a higher level (e.g., at the
 *		points in the code where H5FD_mpio_tas_allsame is called).
 *
 *
 * 		Robb Matzke, 1998-02-18
 *		Added the ACCESS_PARMS argument.
 *
 * 		rky, 1998-04-10
 *		Call independent or collective MPI write, based on
 *		ACCESS_PARMS.
 *
 * 		rky, 1998-04-24
 *		Removed redundant write from H5FD_mpio_write.
 *
 * 		Albert Cheng, 1998-06-01
 *		Added XFER_MODE to control independent or collective MPI
 *		write.
 *
 * 		rky, 1998-08-16
 *		Use BTYPE, FTYPE, and DISP from access parms. The guts of
 *		H5FD_mpio_read and H5FD_mpio_write should be replaced by a
 *		single dual-purpose routine.
 *
 * 		rky, 1998-08-28
 *		Added ALLSAME parameter to make all but proc 0 skip the
 *		actual write.
 *
 * 		Robb Matzke, 1999-04-21
 *		Changed XFER_MODE to XFER_PARMS for all H5FD_*_write()
 *		callbacks.
 *
 * 		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *
 *		Albert Cheng, 1999-12-19
 *		When only-p0-write-allsame-data, p0 Bcasts the
 *		ret_value to other processes.  This prevents
 *		a racing condition (that other processes try to
 *		read the file before p0 finishes writing) and also
 *		allows all processes to report the same ret_value.
 *
 *		Kim Yates, Pat Weidhaas,  2000-09-26
 *		Move block of coding where only p0 writes after the
 *              MPI_File_set_view call. 
 *
 *		Quincey Koziol,  2002-05-10
 *		Instead of always writing metadata from process 0, spread the
 *              burden among all the processes by using a round-robin rotation
 *              scheme.
 *
 *		Quincey Koziol,  2002-05-10
 *		Removed allsame code, keying off the type parameter instead.
 *
 *		Quincey Koziol,  2002-05-14
 *		Only call MPI_Get_count if we can use MPI_BYTE for the MPI type
 *              for the I/O transfer.  Someday we might include code to decode
 *              the MPI type used for more complicated transfers and call
 *              MPI_Get_count all the time.
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr,
		size_t size, const void *buf)
{
    H5FD_mpio_t			*file = (H5FD_mpio_t*)_file;
    const H5FD_mpio_dxpl_t	*dx=NULL;
    H5FD_mpio_dxpl_t		_dx;
    MPI_Offset 		 	mpi_off, mpi_disp;
    MPI_Status			mpi_stat = {0};
    MPI_Datatype		buf_type, file_type;
    int         		size_i, bytes_written;
    int				use_types_this_time, used_types_last_time;
    H5P_genplist_t *plist;      /* Property list pointer */
    herr_t              	ret_value=SUCCEED;

    FUNC_ENTER_NOAPI(H5FD_mpio_write, FAIL);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Entering H5FD_mpio_write\n" );
#endif
    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);
    /* Make certain we have the correct type of property list */
    assert(H5I_GENPROP_LST==H5I_get_type(dxpl_id));
    assert(TRUE==H5P_isa_class(dxpl_id,H5P_DATASET_XFER));

    /* some numeric conversions */
    if (H5FD_mpio_haddr_to_MPIOff(addr, &mpi_off)<0)
        HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off");
    if (H5FD_mpio_haddr_to_MPIOff(file->disp, &mpi_disp)<0)
        HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off");
    size_i = (int)size;
    if ((hsize_t)size_i != size)
        HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from size to size_i");

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'w'])
        fprintf(stdout, "in H5FD_mpio_write  mpi_off=%ld  size_i=%d\n",
                (long)mpi_off, size_i);
#endif

    /* Obtain the data transfer properties */
    if(NULL == (plist = H5I_object(dxpl_id)))
        HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list");
    if (H5FD_MPIO!=H5P_get_driver(plist)) {
        _dx.xfer_mode = H5FD_MPIO_INDEPENDENT; /*the default*/
        dx = &_dx;
    } else {
        dx = H5P_get_driver_info(plist);
        assert(dx);
    }
    
    /*
     * Set up for a fancy xfer using complex types, or single byte block. We
     * wouldn't need to rely on the use_types field if MPI semantics allowed
     * us to test that btype=ftype=MPI_BYTE (or even MPI_TYPE_NULL, which
     * could mean "use MPI_BYTE" by convention).
     */
    use_types_this_time = file->use_types;
    if (use_types_this_time) {
        /* prepare for a full-blown xfer using btype, ftype, and disp */
        buf_type = file->btype;
        file_type = file->ftype;
        if (H5FD_mpio_haddr_to_MPIOff(file->disp, &mpi_disp)<0)
            HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off");
    } else {
        /*
         * Prepare for a simple xfer of a contiguous block of bytes.
         * The btype, ftype, and disp fields are not used.
         */
        buf_type = MPI_BYTE;
        file_type = MPI_BYTE;
        mpi_disp = 0;		/* mpi_off is sufficient */
    }

    /*
     * Don't bother to reset the view if we're not using the types this time,
     * and did we didn't use them last time either.
     */
    used_types_last_time = file->old_use_types;
    if (used_types_last_time ||	/* change to new ftype or MPI_BYTE */
            use_types_this_time) {	/* almost certainly a different ftype */
        /*OKAY: CAST DISCARDS CONST QUALIFIER*/
        if (MPI_SUCCESS != MPI_File_set_view(file->f, mpi_disp, MPI_BYTE, file_type, (char*)"native", file->info))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
    }
    
    /*
     * We always set the use_types flag to 0 because the default is not to
     * use types next time, unless someone explicitly requests it by setting
     * this flag to !=0.
     */
    file->old_use_types = use_types_this_time;
    file->use_types = 0;

    /* Only p<round> will do the actual write if all procs in comm write same data */
    if ((type!=H5FD_MEM_DRAW) && H5_mpi_1_metawrite_g) {
        if (file->mpi_rank != file->mpi_round) {
#ifdef H5FDmpio_DEBUG
            if (H5FD_mpio_Debug[(int)'w']) {
                fprintf(stdout,
		    "  proc %d: in H5FD_mpio_write (write omitted)\n",
		    file->mpi_rank );
            }
#endif
            HGOTO_DONE(SUCCEED) /* skip the actual write */
        }
    }

    /* Write the data. */
    assert(H5FD_MPIO_INDEPENDENT==dx->xfer_mode || H5FD_MPIO_COLLECTIVE==dx->xfer_mode);
    if (H5FD_MPIO_INDEPENDENT==dx->xfer_mode) {
        /*OKAY: CAST DISCARDS CONST QUALIFIER*/
        if (MPI_SUCCESS != MPI_File_write_at(file->f, mpi_off, (void*)buf, size_i, buf_type, &mpi_stat))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_write_at failed");
    } else {
#ifdef H5FDmpio_DEBUG
        if (H5FD_mpio_Debug[(int)'t'])
            fprintf(stdout, "H5FD_mpio_write: using MPIO collective mode\n");
#endif
        /*OKAY: CAST DISCARDS CONST QUALIFIER*/
        if (MPI_SUCCESS != MPI_File_write_at_all(file->f, mpi_off, (void*)buf, size_i, buf_type, &mpi_stat))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_write_at_all failed");
    }

    /* KLUDGE, Robb Matzke, 2000-12-29
     * The LAM implementation of MPI_Get_count() says
     *    MPI_Get_count: invalid argument (rank 0, MPI_COMM_WORLD)
     * So I'm commenting this out until it can be investigated. The
     * returned `bytes_written' isn't used anyway because of Kim's
     * kludge to avoid bytes_written<0. Likewise in H5FD_mpio_read(). */

#ifdef H5_HAVE_MPI_GET_COUNT /* Bill and Albert's kludge*/
    /* Yet Another KLUDGE, Albert Cheng & Bill Wendling, 2001-05-11.
     * Many systems don't support MPI_Get_count so we need to do a
     * configure thingy to fix this. */

    /* Calling MPI_Get_count with "MPI_BYTE" is only valid when we actually
     * had the 'buf_type' set to MPI_BYTE -QAK
     */
    if(use_types_this_time) {
        /* Figure out the mapping from the MPI 'buf_type' to bytes, someday...
         * If this gets fixed (and MPI_Get_count() is reliable), the
         * kludge below where the 'bytes_written' value from MPI_Get_count() is
         * overwritten with the 'size_i' parameter can be removed. -QAK
         */
    } /* end if */
    else {
        /* How many bytes were actually written? */
        if (MPI_SUCCESS!= MPI_Get_count(&mpi_stat, MPI_BYTE, &bytes_written))
            HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Get_count failed");
    } /* end else */
#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'c'])
    	fprintf(stdout,
	    "In H5FD_mpio_write after Get_count size_i=%d bytes_written=%d\n",
	    size_i, bytes_written );
#endif
#endif /* H5_HAVE_MPI_GET_COUNT */

    /*
     * KLUGE rky, 1998-02-02
     * MPI_Get_count incorrectly returns negative count; fake a complete
     * write.
     */
    bytes_written = size_i;

    /* Check for write failure */
    if (bytes_written<0 || bytes_written>size_i)
        HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file write failed");

    /* Forget the EOF value (see H5FD_mpio_get_eof()) --rpm 1999-08-06 */
    file->eof = HADDR_UNDEF;
    
done:
    /* Guard against getting into metadate broadcast in failure cases */
    if(ret_value!=FAIL) {
        /* if only p<round> writes, need to broadcast the ret_value to other processes */
        if ((type!=H5FD_MEM_DRAW) && H5_mpi_1_metawrite_g) {
            if (MPI_SUCCESS !=
                MPI_Bcast(&ret_value, sizeof(ret_value), MPI_BYTE, file->mpi_round, file->comm))
              HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Bcast failed");

            /* Round-robin rotate to the next process */
            file->mpi_round = (++file->mpi_round)%file->mpi_size;
#ifdef QAK
    {
        int max,min;

        MPI_Allreduce(&file->mpi_round, &max, 1, MPI_INT, MPI_MAX, file->comm);
        MPI_Allreduce(&file->mpi_round, &min, 1, MPI_INT, MPI_MIN, file->comm);
        if(max!=file->mpi_round)
            printf("%s: rank=%d, round=%d, max=%d\n",FUNC,file->mpi_rank,file->mpi_round,max);
        if(min!=file->mpi_round)
            printf("%s: rank=%d, round=%d, min=%d\n",FUNC,file->mpi_rank,file->mpi_round,min);
    }
#endif /* QAK */
        } /* end if */
    } /* end if */

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "proc %d: Leaving H5FD_mpio_write with ret_value=%d\n",
	    file->mpi_rank, ret_value );
#endif
    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:    H5FD_mpio_flush
 *
 * Purpose:     Makes sure that all data is on disk.  This is collective.
 *
 * Return:      Success:	Non-negative
 *
 * 		Failure:	Negative
 *
 * Programmer:  Unknown
 *              January 30, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1998-02-18
 *		Added the ACCESS_PARMS argument.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *
 *              Robb Matzke, 2000-12-29
 *              Make sure file size is at least as large as the last
 *              allocated byte.
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_flush(H5FD_t *_file, unsigned closing)
{
    H5FD_mpio_t		*file = (H5FD_mpio_t*)_file;
    int			mpi_code;	/* mpi return code */
    MPI_Offset          mpi_off;
    herr_t              ret_value=SUCCEED;
#ifdef OLD_WAY
    uint8_t             byte=0;
    MPI_Status          mpi_stat = {0};
#endif  /* OLD_WAY */

    FUNC_ENTER_NOAPI(H5FD_mpio_flush, FAIL);

#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Entering H5FD_mpio_flush\n" );
#endif
    assert(file);
    assert(H5FD_MPIO==file->pub.driver_id);

    /* Extend the file to make sure it's large enough, then sync.
     * Unfortunately, keeping track of EOF is an expensive operation, so
     * we can't just check whether EOF<EOA like with other drivers.
     * Therefore we'll just read the byte at EOA-1 and then write it back. */
    if(file->eoa>file->last_eoa) {
#ifdef OLD_WAY
        if (0==file->mpi_rank) {
            if (H5FD_mpio_haddr_to_MPIOff(file->eoa-1, &mpi_off)<0)
                HRETURN_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "cannot convert from haddr_t to MPI_Offset");
            if (MPI_SUCCESS != MPI_File_read_at(file->f, mpi_off, &byte, 1, MPI_BYTE, &mpi_stat))
                HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_read_at() failed");
            if (MPI_SUCCESS != MPI_File_write_at(file->f, mpi_off, &byte, 1, MPI_BYTE, &mpi_stat))
                HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_write_at() failed");
        } /* end if */
#else /* OLD_WAY */
        if (H5FD_mpio_haddr_to_MPIOff(file->eoa, &mpi_off)<0)
            HRETURN_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "cannot convert from haddr_t to MPI_Offset");

        /* Extend the file's size */
        if (MPI_SUCCESS != MPI_File_set_size(file->f, mpi_off))
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_size failed");

	/* Don't let any proc return until all have extended the file.
         * (Prevents race condition where some processes go ahead and write
         * more data to the file before all the processes have finished making
         * it the shorter length, potentially truncating the file and dropping
         * the new data written)
         */
        if (MPI_SUCCESS!= MPI_Barrier(file->comm))
            HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Barrier failed");
#endif /* OLD_WAY */

        /* Update the 'last' eoa value */
        file->last_eoa=file->eoa;
    } /* end if */

    /* Only sync the file if we are not going to immediately close it */
    if(!closing) {
        if (MPI_SUCCESS != (mpi_code=MPI_File_sync(file->f)))
            HMPI_GOTO_ERROR(FAIL, "MPI_File_sync failed", mpi_code);
    } /* end if */

done:
#ifdef H5FDmpio_DEBUG
    if (H5FD_mpio_Debug[(int)'t'])
    	fprintf(stdout, "Leaving H5FD_mpio_flush\n" );
#endif

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:    H5FD_mpio_MPIOff_to_haddr
 *
 * Purpose:     Convert an MPI_Offset value to haddr_t.
 *
 * Return:      Success:	The haddr_t equivalent of the MPI_OFF
 *				argument.
 *				
 *              Failure:	HADDR_UNDEF
 *
 * Programmer:  Unknown
 *              January 30, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1999-04-23
 *		An error is reported for address overflows. The ADDR output
 *		argument is optional.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *------------------------------------------------------------------------- 
 */
static haddr_t
H5FD_mpio_MPIOff_to_haddr(MPI_Offset mpi_off)
{
    haddr_t ret_value=HADDR_UNDEF;

    FUNC_ENTER_NOINIT(H5FD_mpio_MPIOff_to_haddr);

    if (mpi_off != (MPI_Offset)(haddr_t)mpi_off)
        ret_value=HADDR_UNDEF;
    else
        ret_value=(haddr_t)mpi_off;

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:    H5FD_mpio_haddr_to_MPIOff
 *
 * Purpose:     Convert an haddr_t value to MPI_Offset.
 *
 * Return:      Success:	Non-negative, the MPI_OFF argument contains
 *				the converted value.
 *
 * 		Failure:	Negative, MPI_OFF is undefined.
 *
 * Programmer:  Unknown
 *              January 30, 1998
 *
 * Modifications:
 * 		Robb Matzke, 1999-04-23
 *		An error is reported for address overflows. The ADDR output
 *		argument is optional.
 *
 * 		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *
 * 		Robb Matzke, 1999-08-06
 *		Modified to work with the virtual file layer.
 *-------------------------------------------------------------------------
 */
static herr_t
H5FD_mpio_haddr_to_MPIOff(haddr_t addr, MPI_Offset *mpi_off/*out*/)
{
    herr_t ret_value=FAIL;

    FUNC_ENTER_NOINIT(H5FD_mpio_haddr_to_MPIOff);

    if (mpi_off)
        *mpi_off = (MPI_Offset)addr;
    if (addr != (haddr_t)(MPI_Offset)addr)
        ret_value=FAIL;
    else
        ret_value=SUCCEED;

    FUNC_LEAVE(ret_value);
}
#endif /*H5_HAVE_PARALLEL*/