summaryrefslogtreecommitdiffstats
path: root/perform/pio_engine.c
blob: a9499d32c498591b5ad55156320bb6901d2f27f8 (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
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
 * 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. *
 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

/*
 * Author: Albert Cheng of NCSA, Oct 24, 2001.
 */

#include <sys/types.h>
#include <sys/stat.h>
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>

#include "hdf5.h"

#ifdef H5_HAVE_PARALLEL

#include <mpi.h>

#ifndef MPI_FILE_NULL           /*MPIO may be defined in mpi.h already       */
#   include <mpio.h>
#endif  /* !MPI_FILE_NULL */

#ifdef H5_HAVE_GPFS
#   include <gpfs_fcntl.h>
#endif  /* H5_HAVE_GPFS */

#include "pio_perf.h"
#include "pio_timer.h"

/* Macro definitions */

/* sizes of various items. these sizes won't change during program execution */
/* The following three must have the same type */
#define ELMT_SIZE           (sizeof(int))          /* we're doing ints */
#define ELMT_MPI_TYPE       MPI_INT
#define ELMT_H5_TYPE        H5T_NATIVE_INT

#define GOTOERROR(errcode)	{ ret_code = errcode; goto done; }
#define GOTODONE		{ goto done; }
#define ERRMSG(mesg) {                                                  \
    fprintf(stderr, "Proc %d: ", pio_mpi_rank_g);                       \
    fprintf(stderr, "*** Assertion failed (%s) at line %4d in %s\n",    \
            mesg, (int)__LINE__, __FILE__);                             \
}

#define MSG(mesg) {                                     \
    fprintf(stderr, "Proc %d: ", pio_mpi_rank_g);       \
    fprintf(stderr, "(%s) at line %4d in %s\n",         \
            mesg, (int)__LINE__, __FILE__);             \
}

/* verify: if val is false (0), print mesg. */
#define VRFY(val, mesg) do {                            \
    if (!val) {                                         \
        ERRMSG(mesg);                                   \
        GOTOERROR(FAIL);                                \
    }                                                   \
} while(0)


/* POSIX I/O macros */
#define POSIXCREATE(fn)           HDopen(fn, O_CREAT|O_TRUNC|O_RDWR, 0600)
#define POSIXOPEN(fn, F)          HDopen(fn, F, 0600)
#define POSIXCLOSE(F)             HDclose(F)
#define POSIXSEEK(F,L)            HDlseek(F, L, SEEK_SET)
#define POSIXWRITE(F,B,S)         HDwrite(F,B,S)
#define POSIXREAD(F,B,S)          HDread(F,B,S)

enum {
    PIO_CREATE = 1,
    PIO_WRITE = 2,
    PIO_READ = 4
};

/* Global variables */
static int	clean_file_g = -1;	/*whether to cleanup temporary test     */
                                        /*files. -1 is not defined;             */
                                        /*0 is no cleanup; 1 is do cleanup      */

/*
 * In a parallel machine, the filesystem suitable for compiling is
 * unlikely a parallel file system that is suitable for parallel I/O.
 * There is no standard pathname for the parallel file system.  /tmp
 * is about the best guess.
 */
#ifndef HDF5_PARAPREFIX
#  ifdef __PUMAGON__
     /* For the PFS of TFLOPS */
#    define HDF5_PARAPREFIX     "pfs:/pfs_grande/multi/tmp_1"
#  else
#    define HDF5_PARAPREFIX     "/tmp"
#  endif    /* __PUMAGON__ */
#endif  /* !HDF5_PARAPREFIX */

#ifndef MIN
#   define MIN(a,b) ((a) < (b) ? (a) : (b))
#endif  /* !MIN */

/* the different types of file descriptors we can expect */
typedef union _file_descr {
    int         posixfd;    /* POSIX file handle*/
    MPI_File    mpifd;      /* MPI file         */
    hid_t       h5fd;       /* HDF5 file        */
} file_descr;

/* local functions */
static char  *pio_create_filename(iotype iot, const char *base_name,
                                  char *fullname, size_t size);
static herr_t do_write(results *res, file_descr *fd, parameters *parms,
    long ndsets, off_t nelmts, size_t buf_size, void *buffer);
static herr_t do_read(results *res, file_descr *fd, parameters *parms,
    long ndsets, off_t nelmts, size_t buf_size, void *buffer /*out*/);
static herr_t do_fopen(parameters *param, char *fname, file_descr *fd /*out*/,
                       int flags);
static herr_t do_fclose(iotype iot, file_descr *fd);
static void do_cleanupfile(iotype iot, char *fname);

/* GPFS-specific functions */
static void access_range(int handle, off_t start, off_t length, int is_write);
static void free_range(int handle, off_t start, off_t length);
static void clear_file_cache(int handle);
static void cancel_hints(int handle);
static void start_data_shipping(int handle, int num_insts);
static void stop_data_shipping(int handle);
static void invalidate_file_cache(const char *filename);

/*
 * Function:    do_pio
 * Purpose:     PIO Engine where Parallel IO are executed.
 * Return:      results
 * Programmer:  Albert Cheng, Bill Wendling 2001/12/12
 * Modifications:
 */
results
do_pio(parameters param)
{
    /* return codes */
    herr_t      ret_code = 0;   /*return code                           */
    results     res;

    file_descr	fd;
    iotype      iot;

    char        fname[FILENAME_MAX];
    int         maxprocs;
    long	nfiles, nf;
    long        ndsets;
    off_t       nelmts;
    char        *buffer = NULL;         /*data buffer pointer           */
    size_t      buf_size;               /*data buffer size in bytes     */

    /* HDF5 variables */
    herr_t          hrc;                /*HDF5 return code              */

    /* Sanity check parameters */

    /* IO type */
    iot = param.io_type;

    switch (iot) {
    case MPIO:
        fd.mpifd = MPI_FILE_NULL;
        res.timers = pio_time_new(MPI_TIMER);
        break;
    case POSIXIO:
        fd.posixfd = -1;
        res.timers = pio_time_new(MPI_TIMER);
        break;
    case PHDF5:
        fd.h5fd = -1;
        res.timers = pio_time_new(MPI_TIMER);
        break;
    default:
        /* unknown request */
        fprintf(stderr, "Unknown IO type request (%d)\n", iot);
        GOTOERROR(FAIL);
    }

    nfiles = param.num_files;       /* number of files                      */
    ndsets = param.num_dsets;       /* number of datasets per file          */
    nelmts = param.num_elmts;       /* number of elements per dataset       */
    maxprocs = param.num_procs;     /* max number of mpi-processes to use   */
    buf_size = param.buf_size;

    if (nfiles < 0 ) {
        fprintf(stderr,
                "number of files must be >= 0 (%ld)\n",
                nfiles);
        GOTOERROR(FAIL);
    }

    if (ndsets < 0 ) {
        fprintf(stderr,
                "number of datasets per file must be >= 0 (%ld)\n",
                ndsets);
        GOTOERROR(FAIL);
    }

    if (maxprocs <= 0 ) {
        fprintf(stderr,
                "maximum number of process to use must be > 0 (%d)\n",
                maxprocs);
        GOTOERROR(FAIL);
    }

    /* allocate transfer buffer */
    if(buf_size<=0) {
        HDfprintf(stderr,
	    "Transfer buffer size (%Hd) must be > 0\n", (long_long)buf_size);
	GOTOERROR(FAIL);
    }else{
        buffer = malloc(buf_size);

        if (buffer == NULL){
            HDfprintf(stderr, "malloc for transfer buffer size (%Hd) failed\n",
                (long_long)buf_size);
            GOTOERROR(FAIL);
        }
    }

    if (pio_debug_level >= 4) {
        int myrank;

        MPI_Comm_rank(pio_comm_g, &myrank);

        /* output all of the times for all iterations */
        if (myrank == 0)
            fprintf(output, "Timer details:\n");
    }

    for (nf = 1; nf <= nfiles; nf++) {
	/*
	 * Write performance measurement
	 */
        /* Open file for write */
        char base_name[256];

        MPI_Barrier(pio_comm_g);

        sprintf(base_name, "#pio_tmp_%lu", nf);
        pio_create_filename(iot, base_name, fname, sizeof(fname));

        set_time(res.timers, HDF5_GROSS_WRITE_FIXED_DIMS, START);
        hrc = do_fopen(&param, fname, &fd, PIO_CREATE | PIO_WRITE);

        VRFY((hrc == SUCCESS), "do_fopen failed");

        set_time(res.timers, HDF5_FINE_WRITE_FIXED_DIMS, START);
        hrc = do_write(&res, &fd, &param, ndsets, nelmts, buf_size, buffer);
        set_time(res.timers, HDF5_FINE_WRITE_FIXED_DIMS, STOP);

        VRFY((hrc == SUCCESS), "do_write failed");

        /* Close file for write */
        hrc = do_fclose(iot, &fd);

        set_time(res.timers, HDF5_GROSS_WRITE_FIXED_DIMS, STOP);
        VRFY((hrc == SUCCESS), "do_fclose failed");

        if (!param.h5_write_only) {
            /*
             * Read performance measurement
             */
            MPI_Barrier(pio_comm_g);

            /* Open file for read */
            set_time(res.timers, HDF5_GROSS_READ_FIXED_DIMS, START);
            hrc = do_fopen(&param, fname, &fd, PIO_READ);

            VRFY((hrc == SUCCESS), "do_fopen failed");

            set_time(res.timers, HDF5_FINE_READ_FIXED_DIMS, START);
            hrc = do_read(&res, &fd, &param, ndsets, nelmts, buf_size, buffer);
            set_time(res.timers, HDF5_FINE_READ_FIXED_DIMS, STOP);
            VRFY((hrc == SUCCESS), "do_read failed");

            /* Close file for read */
            hrc = do_fclose(iot, &fd);

            set_time(res.timers, HDF5_GROSS_READ_FIXED_DIMS, STOP);
            VRFY((hrc == SUCCESS), "do_fclose failed");
        }

        MPI_Barrier(pio_comm_g);
        do_cleanupfile(iot, fname);
    }

done:
    /* clean up */
    /* release HDF5 objects */

    /* close any opened files */
    /* no remove(fname) because that should have happened normally. */
    switch (iot) {
    case POSIXIO:
        if (fd.posixfd != -1)
            hrc = do_fclose(iot, &fd);
        break;
    case MPIO:
        if (fd.mpifd != MPI_FILE_NULL)
            hrc = do_fclose(iot, &fd);
        break;
    case PHDF5:
        if (fd.h5fd != -1)
            hrc = do_fclose(iot, &fd);
        break;
    }

    /* release generic resources */
    if(buffer)
        free(buffer);
    res.ret_code = ret_code;
    return res;
}

/*
 * Function:    pio_create_filename
 * Purpose:     Create a new filename to write to. Determine the correct
 *              suffix to append to the filename by the type of I/O we're
 *              doing. Also, place in the /tmp/{$USER,$LOGIN} directory if
 *              USER or LOGIN are specified in the environment.
 * Return:      Pointer to filename or NULL
 * Programmer:  Bill Wendling, 21. November 2001
 * Modifications:
 */
static char *
pio_create_filename(iotype iot, const char *base_name, char *fullname, size_t size)
{
    const char *prefix, *suffix="";
    char *ptr, last = '\0';
    size_t i, j;

    if (!base_name || !fullname || size < 1)
        return NULL;

    memset(fullname, 0, size);

    switch (iot) {
    case POSIXIO:
        suffix = ".posix";
        break;
    case MPIO:
        suffix = ".mpio";
        break;
    case PHDF5:
        suffix = ".h5";
        break;
    }

    /* First use the environment variable and then try the constant */
    prefix = getenv("HDF5_PARAPREFIX");

#ifdef HDF5_PARAPREFIX
    if (!prefix)
        prefix = HDF5_PARAPREFIX;
#endif  /* HDF5_PARAPREFIX */

    /* Prepend the prefix value to the base name */
    if (prefix && *prefix) {
        /* If the prefix specifies the HDF5_PARAPREFIX directory, then
         * default to using the "/tmp/$USER" or "/tmp/$LOGIN"
         * directory instead. */
        register char *user, *login, *subdir;

        user = getenv("USER");
        login = getenv("LOGIN");
        subdir = (user ? user : login);

        if (subdir) {
            for (i = 0; i < size && prefix[i]; i++)
                fullname[i] = prefix[i];

            fullname[i++] = '/';

            for (j = 0; i < size && subdir[j]; i++, j++)
                fullname[i] = subdir[j];
        } else {
            /* We didn't append the prefix yet */
            strncpy(fullname, prefix, MIN(strlen(prefix), size));
        }

        if ((strlen(fullname) + strlen(base_name) + 1) < size) {
            /* Append the base_name with a slash first. Multiple slashes are
             * handled below. */
            h5_stat_t buf;

            if (HDstat(fullname, &buf) < 0)
                /* The directory doesn't exist just yet */
                if (mkdir(fullname, (mode_t)0755) < 0 && errno != EEXIST) {
                    /* We couldn't make the "/tmp/${USER,LOGIN}" subdirectory.
                     * Default to PREFIX's original prefix value. */
                    strcpy(fullname, prefix);
                }

            strcat(fullname, "/");
            strcat(fullname, base_name);
        } else {
            /* Buffer is too small */
            return NULL;
        }
    } else if (strlen(base_name) >= size) {
        /* Buffer is too small */
        return NULL;
    } else {
        strcpy(fullname, base_name);
    }

    /* Append a suffix */
    if (suffix) {
        if (strlen(fullname) + strlen(suffix) >= size)
            return NULL;

        strcat(fullname, suffix);
    }

    /* Remove any double slashes in the filename */
    for (ptr = fullname, i = j = 0; ptr && i < size; i++, ptr++) {
        if (*ptr != '/' || last != '/')
            fullname[j++] = *ptr;

        last = *ptr;
    }

    return fullname;
}

/*
 * Function:    do_write
 * Purpose:     Write the required amount of data to the file.
 * Return:      SUCCESS or FAIL
 * Programmer:  Albert Cheng, Bill Wendling, 2001/12/13
 * Modifications:
 */
static herr_t
do_write(results *res, file_descr *fd, parameters *parms, long ndsets,
         off_t nelmts, size_t buf_size, void *buffer)
{
    int         ret_code = SUCCESS;
    int         rc;             /*routine return code                   */
    int         mrc;            /*MPI return code                       */
    MPI_Offset	mpi_offset;
    MPI_Status	mpi_status;
    long        ndset;
    off_t       nelmts_xfer;
    size_t      nelmts_toxfer;
    char        dname[64];
    off_t       dset_offset=0;    /*dataset offset in a file              */
    off_t       file_offset;    /*file offset of the next transfer      */
    off_t       dset_size;      /*one dataset size in bytes             */
    size_t      nelmts_in_buf;  /*how many element the buffer holds     */
    off_t       elmts_begin;    /*first elmt this process transfer      */
    off_t       elmts_count;    /*number of elmts this process transfer */

    /* HDF5 variables */
    herr_t      hrc;                    /*HDF5 return code              */
    hsize_t     h5dims[1];              /*dataset dim sizes             */
    hid_t       h5dset_space_id = -1;   /*dataset space ID              */
    hid_t       h5mem_space_id = -1;    /*memory dataspace ID           */
    hid_t       h5ds_id = -1;           /*dataset handle                */
    hsize_t	h5block[1];		/*dataspace selection        */
    hsize_t	h5stride[1];
    hsize_t	h5count[1];
    hssize_t	h5start[1];
    hid_t       h5dcpl = -1;            /* Dataset creation property list */
    hid_t       h5dxpl = -1;            /* Dataset transfer property list */

    /* calculate dataset parameters. data type is always native C int */
    dset_size = nelmts * (off_t)ELMT_SIZE;
    nelmts_in_buf = buf_size/ELMT_SIZE;

    /* hdf5 data space setup */
    if (parms->io_type == PHDF5){
        if(nelmts>0) {
            /* define a contiquous dataset of nelmts native ints */
            h5dims[0] = nelmts;
            h5dset_space_id = H5Screate_simple(1, h5dims, NULL);
            VRFY((h5dset_space_id >= 0), "H5Screate_simple");
        } /* end if */
        else {
            h5dset_space_id = H5Screate(H5S_SCALAR);
            VRFY((h5dset_space_id >= 0), "H5Screate");
        } /* end else */

        /* Create the memory dataspace that corresponds to the xfer buffer */
        if(nelmts_in_buf>0) {
            h5dims[0] = nelmts_in_buf;
            h5mem_space_id = H5Screate_simple(1, h5dims, NULL);
            VRFY((h5mem_space_id >= 0), "H5Screate_simple");
        } /* end if */
        else {
            h5mem_space_id = H5Screate(H5S_SCALAR);
            VRFY((h5mem_space_id >= 0), "H5Screate");
        } /* end else */

        /* Create the dataset transfer property list */
        h5dxpl = H5Pcreate(H5P_DATASET_XFER);
        if (h5dxpl < 0) {
            fprintf(stderr, "HDF5 Property List Create failed\n");
            GOTOERROR(FAIL);
        }

        /* Change to collective I/O, if asked */
        if(parms->collective) {
            hrc = H5Pset_dxpl_mpio(h5dxpl, H5FD_MPIO_COLLECTIVE);
            if (hrc < 0) {
                fprintf(stderr, "HDF5 Property List Set failed\n");
                GOTOERROR(FAIL);
            } /* end if */
        } /* end if */
    }

    for (ndset = 1; ndset <= ndsets; ++ndset) {

        /* Calculate dataset offset within a file */

        /* create dataset */
        switch (parms->io_type) {
        case POSIXIO:
        case MPIO:
            /* both posix and mpi io just need dataset offset in file*/
            dset_offset = (ndset - 1) * dset_size;
            break;

        case PHDF5:
            h5dcpl = H5Pcreate(H5P_DATASET_CREATE);
            if (h5dcpl < 0) {
                fprintf(stderr, "HDF5 Property List Create failed\n");
                GOTOERROR(FAIL);
            }

            /* Make the dataset chunked if asked */
            if(parms->h5_use_chunks) {
                /* Set the chunk size to be the same as the buffer size */
                h5dims[0] = nelmts_in_buf;
                hrc = H5Pset_chunk(h5dcpl, 1, h5dims);
                if (hrc < 0) {
                    fprintf(stderr, "HDF5 Property List Set failed\n");
                    GOTOERROR(FAIL);
                } /* end if */
            } /* end if */

#ifdef H5_HAVE_NOFILL
            /* Disable writing fill values if asked */
            if(parms->h5_no_fill) {
                hrc = H5Pset_fill_time(h5dcpl, H5D_FILL_TIME_NEVER);
                if (hrc < 0) {
                    fprintf(stderr, "HDF5 Property List Set failed\n");
                    GOTOERROR(FAIL);
                } /* end if */
            } /* end if */
#endif

            sprintf(dname, "Dataset_%ld", ndset);
            h5ds_id = H5Dcreate(fd->h5fd, dname, ELMT_H5_TYPE,
                                h5dset_space_id, h5dcpl);

            if (h5ds_id < 0) {
                fprintf(stderr, "HDF5 Dataset Create failed\n");
                GOTOERROR(FAIL);
            }

            hrc = H5Pclose(h5dcpl);
            /* verifying the close of the dcpl */
            if (hrc < 0) {
                fprintf(stderr, "HDF5 Property List Close failed\n");
                GOTOERROR(FAIL);
            }

            break;
        }

	/* There are two kinds of transfer patterns, contiguous and interleaved.
	 * Let 0,1,2,...,n be data accessed by process 0,1,2,...,n
	 *     where n is rank of the last process.
	 * In contiguous pattern, data are accessed as
	 *    000...111...222...nnn...
	 * In interleaved pattern, data are accessed as
	 *    012...n012...n...
	 * These are all in the scope of one dataset.
	 */
	/* Calculate the total number of elements (elmts_count) to be
	 * transferred by this process. It may be different for different
	 * transfer pattern due to rounding to integral values.
	 */
	if (parms->interleaved==0) {
	    /* Contiguous Pattern:
	     * Calculate the beginning element of this process and the next.
	     * elmts_count is the difference between these two beginnings.
	     * This way, it eliminates any rounding errors.
	     */
	    elmts_begin = (off_t)(((double)nelmts)/pio_mpi_nprocs_g*pio_mpi_rank_g);

	    /* Do not cast elmt_begin to other types, especially non-integral
	     * types, else it may introduce rounding discrepency. */
	    if (pio_mpi_rank_g < (pio_mpi_nprocs_g - 1))
		elmts_count = (off_t)(((double)nelmts) / pio_mpi_nprocs_g * (pio_mpi_rank_g + 1))
				- elmts_begin;
	    else
		/* last process.  Take whatever are left */
		elmts_count = nelmts - elmts_begin;
	} /* end if */
        else {
	    /* Interleaved Pattern:
	     * Each process takes buf_size of elements, starting with the first
	     * process.  So, the last process may have fewer or even none.
	     * Calculate the beginning element of this process.
	     * The elmnts_begin here marks only the beginning of the first
	     * block accessed by this process.
	     */
	    /* Algorithm:
	     * First allocate equal blocks per process, i.e. one block each
	     * process for every block_size*nprocs.
	     * If there is remaining unallocated, give a block each to process
	     * starting at proc 0.  The last process may get a partial block.
	     */ 
	    off_t remain_nelmts, remain_begin;	/* unallocated remaining*/

	    elmts_begin = (off_t)(nelmts_in_buf*pio_mpi_rank_g);

	    /* must use integer calculation next */
	    /* allocate equal blocks per process */
	    elmts_count = (nelmts / (off_t)(nelmts_in_buf*pio_mpi_nprocs_g)) *
			    (off_t)nelmts_in_buf;
	    remain_nelmts = nelmts % (off_t)(nelmts_in_buf*pio_mpi_nprocs_g);

	    /* allocate any remaining */
	    remain_begin = (off_t)(nelmts_in_buf*pio_mpi_rank_g);
	    if (remain_nelmts > remain_begin){
		/* it gets something */
		if (remain_nelmts > (remain_begin+(off_t)nelmts_in_buf)){
		    /* one full block */
		    elmts_count += nelmts_in_buf;
		}else{
		    /* only a partial block */
		    elmts_count += remain_nelmts - remain_begin;
		}
	    }
	}
	/* debug */
	if (pio_debug_level >= 4) {
	    HDprint_rank(output);
	    HDfprintf(output, "Debug(do_write): "
		"nelmts_in_buf=%Hd, elmts_begin=%Hd, elmts_count=%Hd\n",
		(long_long)nelmts_in_buf, (long_long)elmts_begin,
		(long_long)elmts_count);
	}


	/* The task is to transfer elmts_count elements, starting at
	 * elmts_begin position, using transfer buffer of buf_size bytes.
	 * If interleaved, select buf_size at a time, in round robin
	 * fashion, according to number of process. Otherwise, select
	 * all elmt_count in contiguous.
	 */
        nelmts_xfer = 0 ;

        /* Start "raw data" write timer */
        set_time(res->timers, HDF5_RAW_WRITE_FIXED_DIMS, START);

        while (nelmts_xfer < elmts_count){
	    /* transfer one buffer of data each round */
	    /* Note: because size_t is unsigned, avoid expressions that */
	    /* can be negative. */
            if ((nelmts_xfer + (off_t)nelmts_in_buf) <= elmts_count) {
                nelmts_toxfer = nelmts_in_buf;
            } else {
                /* last transfer of a partial buffer */
                nelmts_toxfer = elmts_count - nelmts_xfer;
            }

	    if (parms->verify) {
		/*Prepare write data for verify later*/
                int *intptr = (int *)buffer;
                size_t i;

                for (i = 0; i < nelmts_toxfer; ++i)
                    *intptr++ = pio_mpi_rank_g;
            }

            /* Write */
            /* Calculate offset of write within a dataset/file */
            switch (parms->io_type) {
            case POSIXIO:
		if (parms->interleaved==0) {
		    /* Contiguous pattern */
		    /* need to (off_t) the elmnts_begin expression because they */
		    /* may be of smaller sized integer types */
		    file_offset = dset_offset + (off_t)(elmts_begin + nelmts_xfer)*(off_t)ELMT_SIZE;
		} /* end if */
                else {
		    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    file_offset = dset_offset +
                        (off_t)(elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g))*(off_t)ELMT_SIZE;
		} /* end else */

                /* only care if seek returns error */
                rc = POSIXSEEK(fd->posixfd, file_offset) < 0 ? -1 : 0;
                VRFY((rc==0), "POSIXSEEK");
                /* check if all bytes are transferred */
                rc = ((ssize_t)(nelmts_toxfer*ELMT_SIZE) ==
                    POSIXWRITE(fd->posixfd, buffer, nelmts_toxfer*ELMT_SIZE));
                VRFY((rc != 0), "POSIXWRITE");
                break;

            case MPIO:
		if (parms->interleaved==0){
		    /* Contiguous pattern */
                    mpi_offset = dset_offset + (elmts_begin + nelmts_xfer)*(off_t)ELMT_SIZE;
                } /* end if */
                else {
		    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    mpi_offset = dset_offset + (elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g))*(off_t)ELMT_SIZE;
                } /* end else */

                if(parms->collective==0) {
                    mrc = MPI_File_write_at(fd->mpifd, mpi_offset, buffer,
                                            (int)nelmts_toxfer, ELMT_MPI_TYPE,
                                            &mpi_status);
                    VRFY((mrc==MPI_SUCCESS), "MPIO_WRITE");
                } /* end if */
                else {
                    mrc = MPI_File_write_at_all(fd->mpifd, mpi_offset, buffer,
                                            (int)nelmts_toxfer, ELMT_MPI_TYPE,
                                            &mpi_status);
                    VRFY((mrc==MPI_SUCCESS), "MPIO_WRITE");
                } /* end else */
                break;

            case PHDF5:
                /* Set up the file dset space id to select the segment to process */
                if (parms->interleaved==0){
                    /* Contiguous pattern */
                    h5start[0] = elmts_begin + nelmts_xfer;
                } /* end if */
                else {
                    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    h5start[0] = elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g);
                } /* end else */
                h5stride[0] = h5block[0] = nelmts_toxfer;
                h5count[0] = 1;
                hrc = H5Sselect_hyperslab(h5dset_space_id, H5S_SELECT_SET,
                          h5start, h5stride, h5count, h5block); 
                VRFY((hrc >= 0), "H5Sset_hyperslab");

                /* Only need selection in memory dataset if it is smaller than the whole buffer */
                if(nelmts_toxfer<nelmts_in_buf) {
                    /* Setup the memory space id too.  Only start is different */
                    h5start[0] = 0;
                    hrc = H5Sselect_hyperslab(h5mem_space_id, H5S_SELECT_SET,
			      h5start, h5stride, h5count, h5block); 
                    VRFY((hrc >= 0), "H5Sset_hyperslab");
                } /* end if */

                /* set write time here */
                hrc = H5Dwrite(h5ds_id, ELMT_H5_TYPE, h5mem_space_id,
                               h5dset_space_id, h5dxpl, buffer);
                VRFY((hrc >= 0), "H5Dwrite");
                break;
            } /* switch (parms->io_type) */

            /* Increment number of elements transferred */
            nelmts_xfer += nelmts_toxfer;
        }

        /* Stop "raw data" write timer */
        set_time(res->timers, HDF5_RAW_WRITE_FIXED_DIMS, STOP);

        /* Calculate write time */

        /* Close dataset. Only HDF5 needs to do an explicit close. */
        if (parms->io_type == PHDF5){
            hrc = H5Dclose(h5ds_id);

            if (hrc < 0) {
                fprintf(stderr, "HDF5 Dataset Close failed\n");
                GOTOERROR(FAIL);
            }

            h5ds_id = -1;
        }
    }

done:
    /* release HDF5 objects */
    if (h5dset_space_id != -1) {
        hrc = H5Sclose(h5dset_space_id);
        if (hrc < 0){
            fprintf(stderr, "HDF5 Dataset Space Close failed\n");
            ret_code = FAIL;
        } else {
            h5dset_space_id = -1;
        }
    }

    if (h5mem_space_id != -1) {
        hrc = H5Sclose(h5mem_space_id);
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Memory Space Close failed\n");
            ret_code = FAIL;
        } else {
            h5mem_space_id = -1;
        }
    }

    if (h5dxpl != -1) {
        hrc = H5Pclose(h5dxpl);
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Dataset Transfer Property List Close failed\n");
            ret_code = FAIL;
        } else {
            h5dxpl = -1;
        }
    }

    return ret_code;
}

/*
 * Function:    do_read
 * Purpose:     read the required amount of data from the file.
 * Return:      SUCCESS or FAIL
 * Programmer:  Albert Cheng 2001/12/13
 * Modifications:
 */
static herr_t
do_read(results *res, file_descr *fd, parameters *parms, long ndsets,
        off_t nelmts, size_t buf_size, void *buffer /*out*/)
{
    int         ret_code = SUCCESS;
    int         rc;             /*routine return code                   */
    int         mrc;            /*MPI return code                       */
    MPI_Offset  mpi_offset;
    MPI_Status  mpi_status;
    long        ndset;
    off_t       nelmts_xfer;
    size_t      nelmts_toxfer;
    char        dname[64];
    off_t       dset_offset=0;    /*dataset offset in a file              */
    off_t       file_offset;	/*file offset of the next transfer      */
    off_t       dset_size;      /*one dataset size in bytes             */
    size_t      nelmts_in_buf;  /*how many element the buffer holds     */
    off_t       elmts_begin;    /*first elmt this process transfer      */
    off_t       elmts_count;    /*number of elmts this process transfer */

    /* HDF5 variables */
    herr_t      hrc;            	/*HDF5 return code              */
    hsize_t     h5dims[1];      	/*dataset dim sizes             */
    hid_t       h5dset_space_id = -1;   /*dataset space ID              */
    hid_t       h5mem_space_id = -1;    /*memory dataspace ID           */
    hid_t       h5ds_id = -1;   	/*dataset handle                */
    hsize_t	h5block[1];		/*dataspace selection        */
    hsize_t	h5stride[1];
    hsize_t	h5count[1];
    hssize_t	h5start[1];
    hid_t       h5dxpl = -1;            /* Dataset transfer property list */

    /* calculate dataset parameters. data type is always native C int */
    dset_size = nelmts * (off_t)ELMT_SIZE;
    nelmts_in_buf = buf_size/ELMT_SIZE;

    /* hdf5 data space setup */
    if (parms->io_type == PHDF5){
        if(nelmts>0) {
            /* define a contiquous dataset of nelmts native ints */
            h5dims[0] = nelmts;
            h5dset_space_id = H5Screate_simple(1, h5dims, NULL);
            VRFY((h5dset_space_id >= 0), "H5Screate_simple");
        } /* end if */
        else {
            h5dset_space_id = H5Screate(H5S_SCALAR);
            VRFY((h5dset_space_id >= 0), "H5Screate");
        } /* end else */

        /* Create the memory dataspace that corresponds to the xfer buffer */
        if(nelmts_in_buf>0) {
            h5dims[0] = nelmts_in_buf;
            h5mem_space_id = H5Screate_simple(1, h5dims, NULL);
            VRFY((h5mem_space_id >= 0), "H5Screate_simple");
        } /* end if */
        else {
            h5mem_space_id = H5Screate(H5S_SCALAR);
            VRFY((h5mem_space_id >= 0), "H5Screate");
        } /* end else */

        /* Create the dataset transfer property list */
        h5dxpl = H5Pcreate(H5P_DATASET_XFER);
        if (h5dxpl < 0) {
            fprintf(stderr, "HDF5 Property List Create failed\n");
            GOTOERROR(FAIL);
        }

        /* Change to collective I/O, if asked */
        if(parms->collective) {
            hrc = H5Pset_dxpl_mpio(h5dxpl, H5FD_MPIO_COLLECTIVE);
            if (hrc < 0) {
                fprintf(stderr, "HDF5 Property List Set failed\n");
                GOTOERROR(FAIL);
            } /* end if */
        } /* end if */
    } /* end if */

    for (ndset = 1; ndset <= ndsets; ++ndset) {
        /* Calculate dataset offset within a file */

        /* create dataset */
        switch (parms->io_type) {
        case POSIXIO:
        case MPIO:
            /* both posix and mpi io just need dataset offset in file*/
            dset_offset = (ndset - 1) * dset_size;
            break;

        case PHDF5:
            sprintf(dname, "Dataset_%ld", ndset);
            h5ds_id = H5Dopen(fd->h5fd, dname);
            if (h5ds_id < 0) {
                fprintf(stderr, "HDF5 Dataset open failed\n");
                GOTOERROR(FAIL);
            }

            break;
        }

	/* There are two kinds of transfer patterns, contiguous and interleaved.
	 * Let 0,1,2,...,n be data accessed by process 0,1,2,...,n
	 *     where n is rank of the last process.
	 * In contiguous pattern, data are accessed as
	 *    000...111...222...nnn...
	 * In interleaved pattern, data are accessed as
	 *    012...n012...n...
	 * These are all in the scope of one dataset.
	 */
	/* Calculate the total number of elements (elmts_count) to be
	 * transferred by this process. It may be different for different
	 * transfer pattern due to rounding to integral values.
	 */
	if (parms->interleaved==0){
	    /* Contiguous Pattern:
	     * Calculate the beginning element of this process and the next.
	     * elmts_count is the difference between these two beginnings.
	     * This way, it eliminates any rounding errors.
	     */
	    elmts_begin = (off_t)(((double)nelmts)/pio_mpi_nprocs_g*pio_mpi_rank_g);

	    /* Do not cast elmt_begin to other types, especially non-integral
	     * types, else it may introduce rounding discrepency. */
	    if (pio_mpi_rank_g < (pio_mpi_nprocs_g - 1))
		elmts_count = (off_t)(((double)nelmts) / pio_mpi_nprocs_g * (pio_mpi_rank_g + 1))
				- elmts_begin;
	    else
		/* last process.  Take whatever are left */
		elmts_count = nelmts - elmts_begin;
	} /* end if */
        else {
	    /* Interleaved Pattern:
	     * Each process takes buf_size of elements, starting with the first
	     * process.  So, the last process may have fewer or even none.
	     * Calculate the beginning element of this process.
	     * The elmnts_begin here marks only the beginning of the first
	     * block accessed by this process.
	     */
	    /* Algorithm:
	     * First allocate equal blocks per process, i.e. one block each
	     * process for every block_size*nprocs.
	     * If there is remaining unallocated, give a block each to process
	     * starting at proc 0.  The last process may get a partial block.
	     */ 
	    off_t remain_nelmts, remain_begin;	/* unallocated remaining*/

	    elmts_begin = (off_t)(nelmts_in_buf*pio_mpi_rank_g);

	    /* must use integer calculation next */
	    /* allocate equal blocks per process */
	    elmts_count = (nelmts / (off_t)(nelmts_in_buf*pio_mpi_nprocs_g)) *
			    (off_t)nelmts_in_buf;
	    remain_nelmts = nelmts % ((off_t)(nelmts_in_buf*pio_mpi_nprocs_g));

	    /* allocate any remaining */
	    remain_begin = (off_t)(nelmts_in_buf*pio_mpi_rank_g);
	    if (remain_nelmts > remain_begin) {
		/* it gets something */
		if (remain_nelmts > (remain_begin+(off_t)nelmts_in_buf)) {
		    /* one full block */
		    elmts_count += nelmts_in_buf;
		} /* end if */
                else {
		    /* only a partial block */
		    elmts_count += remain_nelmts - remain_begin;
		} /* end else */
	    } /* end if */
	} /* end else */
	/* debug */
	if (pio_debug_level >= 4) {
	    HDprint_rank(output);
	    HDfprintf(output, "Debug(do_read): "
		"nelmts_in_buf=%Hd, elmts_begin=%Hd, elmts_count=%Hd\n",
		(long_long)nelmts_in_buf, (long_long)elmts_begin,
		(long_long)elmts_count);
	}


	/* The task is to transfer elmts_count elements, starting at
	 * elmts_begin position, using transfer buffer of buf_size bytes.
	 * If interleaved, select buf_size at a time, in round robin
	 * fashion, according to number of process. Otherwise, select
	 * all elmt_count in contiguous.
	 */
        nelmts_xfer = 0 ;

        /* Start "raw data" read timer */
        set_time(res->timers, HDF5_RAW_READ_FIXED_DIMS, START);

        while (nelmts_xfer < elmts_count){
	    /* transfer one buffer of data each round */
	    /* Note: because size_t is unsigned, avoid expressions that */
	    /* can be negative. */
            if ((nelmts_xfer + (off_t)nelmts_in_buf) <= elmts_count) {
                nelmts_toxfer = nelmts_in_buf;
            } else {
                /* last transfer of a partial buffer */
                nelmts_toxfer = elmts_count - nelmts_xfer;
            }

            /* read */
            /* Calculate offset of read within a dataset/file */
            switch (parms->io_type){
            case POSIXIO:
		if (parms->interleaved==0){
		    /* Contiguous pattern */
		    /* need to (off_t) the elmnts_begin expression because they */
		    /* may be of smaller sized integer types */
		    file_offset = dset_offset + (off_t)(elmts_begin + nelmts_xfer)*(off_t)ELMT_SIZE;
		} /* end if */
                else {
		    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    file_offset = dset_offset +
                        (off_t)(elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g))*(off_t)ELMT_SIZE;
		} /* end else */

                /* only care if seek returns error */
                rc = POSIXSEEK(fd->posixfd, file_offset) < 0 ? -1 : 0;
                VRFY((rc==0), "POSIXSEEK");
                /* check if all bytes are transferred */
                rc = ((ssize_t)(nelmts_toxfer*ELMT_SIZE) ==
                    POSIXREAD(fd->posixfd, buffer, nelmts_toxfer*ELMT_SIZE));
                VRFY((rc != 0), "POSIXREAD");
                break;

            case MPIO:
		if (parms->interleaved==0){
		    /* Contiguous pattern */
                    mpi_offset = dset_offset + (elmts_begin + nelmts_xfer)*(off_t)ELMT_SIZE;
                } /* end if */
                else {
		    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    mpi_offset = dset_offset + (elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g))*(off_t)ELMT_SIZE;
                } /* end else */

                if(parms->collective==0) {
                    mrc = MPI_File_read_at(fd->mpifd, mpi_offset, buffer,
                                           (int)nelmts_toxfer, ELMT_MPI_TYPE,
                                           &mpi_status);
                    VRFY((mrc==MPI_SUCCESS), "MPIO_read");
                } /* end if */
                else {
                    mrc = MPI_File_read_at_all(fd->mpifd, mpi_offset, buffer,
                                           (int)nelmts_toxfer, ELMT_MPI_TYPE,
                                           &mpi_status);
                    VRFY((mrc==MPI_SUCCESS), "MPIO_read");
                } /* end else */
                break;

            case PHDF5:
                /* Set up the dset space id to select the segment to process */
                if (parms->interleaved==0){
                    /* Contiguous pattern */
                    h5start[0] = elmts_begin + nelmts_xfer;
                } /* end if */
                else {
                    /* Interleaved access pattern */
                    /* Skip offset over blocks of other processes */
                    h5start[0] = elmts_begin + (nelmts_xfer*pio_mpi_nprocs_g);
                } /* end else */
                h5stride[0] = h5block[0] = nelmts_toxfer;
                h5count[0] = 1;
                hrc = H5Sselect_hyperslab(h5dset_space_id, H5S_SELECT_SET,
                          h5start, h5stride, h5count, h5block); 
                VRFY((hrc >= 0), "H5Sset_hyperslab");

                /* Only need selection in memory dataset if it is smaller than the whole buffer */
                if(nelmts_toxfer<nelmts_in_buf) {
                    /* Setup the memory space id too.  Only start is different */
                    h5start[0] = 0;
                    hrc = H5Sselect_hyperslab(h5mem_space_id, H5S_SELECT_SET,
			      h5start, h5stride, h5count, h5block); 
                    VRFY((hrc >= 0), "H5Sset_hyperslab");
                } /* end if */

                /* set read time here */
                hrc = H5Dread(h5ds_id, ELMT_H5_TYPE, h5mem_space_id,
                              h5dset_space_id, h5dxpl, buffer);
                VRFY((hrc >= 0), "H5Dread");
                break;
            } /* switch (parms->io_type) */

	    if (parms->verify) {
		/*verify read data*/
                int *intptr = (int *)buffer;
                size_t i;
		int nerror=0;

                for (i = 0; i < nelmts_toxfer; ++i){
                    if (*intptr++ != pio_mpi_rank_g){
			if (++nerror < 20){
			    /* report at most 20 errors */
			    HDprint_rank(output);
			    HDfprintf(output, "read data error, expected (%Hd), "
				     "got (%Hd)\n",
				     (long_long)pio_mpi_rank_g,
				     (long_long)*(intptr-1));
			}
		    }
		}
		if (nerror >= 20) {
		    HDprint_rank(output);
		    HDfprintf(output, "...");
		    HDfprintf(output, "total read data errors=%Hd\n",
			    nerror);
		}
            }	/* if (parms->verify) */

            /* Increment number of elements transferred */
            nelmts_xfer += nelmts_toxfer;
        }

        /* Stop "raw data" read timer */
        set_time(res->timers, HDF5_RAW_READ_FIXED_DIMS, STOP);

        /* Calculate read time */

        /* Close dataset. Only HDF5 needs to do an explicit close. */
        if (parms->io_type == PHDF5){
            hrc = H5Dclose(h5ds_id);

            if (hrc < 0) {
                fprintf(stderr, "HDF5 Dataset Close failed\n");
                GOTOERROR(FAIL);
            }

            h5ds_id = -1;
        }
    }

done:
    /* release HDF5 objects */
    if (h5dset_space_id != -1) {
        hrc = H5Sclose(h5dset_space_id);
        if (hrc < 0){
            fprintf(stderr, "HDF5 Dataset Space Close failed\n");
            ret_code = FAIL;
        } else {
            h5dset_space_id = -1;
        }
    }

    if (h5mem_space_id != -1) {
        hrc = H5Sclose(h5mem_space_id);
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Memory Space Close failed\n");
            ret_code = FAIL;
        } else {
            h5mem_space_id = -1;
        }
    }

    if (h5dxpl != -1) {
        hrc = H5Pclose(h5dxpl);
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Dataset Transfer Property List Close failed\n");
            ret_code = FAIL;
        } else {
            h5dxpl = -1;
        }
    }

    return ret_code;
}

/*
 * Function:    do_fopen
 * Purpose:     Open the specified file.
 * Return:      SUCCESS or FAIL
 * Programmer:  Albert Cheng, Bill Wendling, 2001/12/13
 * Modifications:
 */
static herr_t
do_fopen(parameters *param, char *fname, file_descr *fd /*out*/, int flags)
{
    int ret_code = SUCCESS, mrc;
    herr_t hrc;
    hid_t acc_tpl = -1;     /* file access templates */

    switch (param->io_type) {
    case POSIXIO:
        if (flags & (PIO_CREATE | PIO_WRITE))
            fd->posixfd = POSIXCREATE(fname);
        else
            fd->posixfd = POSIXOPEN(fname, O_RDONLY);

        if (fd->posixfd < 0 ) {
            fprintf(stderr, "POSIX File Open failed(%s)\n", fname);
            GOTOERROR(FAIL);
        }

         
        /* The perils of POSIX I/O in a parallel environment. The problem is:
         *
         *      - Process n opens a file with truncation and then starts
         *        writing to the file.
         *      - Process m also opens the file with truncation, but after
         *        process n has already started to write to the file. Thus,
         *        all of the stuff process n wrote is now lost.
         */
        MPI_Barrier(pio_comm_g);

        break;

    case MPIO:
        if (flags & (PIO_CREATE | PIO_WRITE)) {
            MPI_File_delete(fname, h5_io_info_g);
            mrc = MPI_File_open(pio_comm_g, fname, MPI_MODE_CREATE | MPI_MODE_RDWR,
                                h5_io_info_g, &fd->mpifd);

            if (mrc != MPI_SUCCESS) {
                fprintf(stderr, "MPI File Open failed(%s)\n", fname);
                GOTOERROR(FAIL);
            }

            /*since MPI_File_open with MPI_MODE_CREATE does not truncate  */
            /*filesize , set size to 0 explicitedly.	*/
            mrc = MPI_File_set_size(fd->mpifd, (MPI_Offset)0);

            if (mrc != MPI_SUCCESS) {
                fprintf(stderr, "MPI_File_set_size failed\n");
                GOTOERROR(FAIL);
            }
        } else {
            mrc = MPI_File_open(pio_comm_g, fname, MPI_MODE_RDONLY,
                                h5_io_info_g, &fd->mpifd);

            if (mrc != MPI_SUCCESS) {
                fprintf(stderr, "MPI File Open failed(%s)\n", fname);
                GOTOERROR(FAIL);
            }
        }

        break;

    case PHDF5:
        acc_tpl = H5Pcreate(H5P_FILE_ACCESS);
        if (acc_tpl < 0) {
            fprintf(stderr, "HDF5 Property List Create failed\n");
            GOTOERROR(FAIL);
        }

        /* Set the file driver to the MPI-I/O driver */
        hrc = H5Pset_fapl_mpio(acc_tpl, pio_comm_g, h5_io_info_g);     
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Property List Set failed\n");
            GOTOERROR(FAIL);
        }

        /* Set the alignment of objects in HDF5 file */
        hrc = H5Pset_alignment(acc_tpl, param->h5_thresh, param->h5_align);
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Property List Set failed\n");
            GOTOERROR(FAIL);
        }

        /* create the parallel file */
        if (flags & (PIO_CREATE | PIO_WRITE)) {
            fd->h5fd = H5Fcreate(fname, H5F_ACC_TRUNC, H5P_DEFAULT, acc_tpl);
        } else {
            fd->h5fd = H5Fopen(fname, H5F_ACC_RDONLY, acc_tpl);
        }

        hrc = H5Pclose(acc_tpl);

        if (fd->h5fd < 0) {
            fprintf(stderr, "HDF5 File Create failed(%s)\n", fname);
            GOTOERROR(FAIL);
        }

        /* verifying the close of the acc_tpl */
        if (hrc < 0) {
            fprintf(stderr, "HDF5 Property List Close failed\n");
            GOTOERROR(FAIL);
        }

        break;
    } 

done:
    return ret_code;
}

/*
 * Function:    do_fclose
 * Purpose:     Close the specified file descriptor.
 * Return:      SUCCESS or FAIL
 * Programmer:  Albert Cheng, Bill Wendling, 2001/12/13
 * Modifications:
 */
static herr_t
do_fclose(iotype iot, file_descr *fd /*out*/)
{
    herr_t ret_code = SUCCESS, hrc;
    int mrc = 0, rc = 0;

    switch (iot) {
    case POSIXIO:
        rc = POSIXCLOSE(fd->posixfd);

        if (rc != 0){
            fprintf(stderr, "POSIX File Close failed\n");
            GOTOERROR(FAIL);
        }

        fd->posixfd = -1;
        break;

    case MPIO:
        mrc = MPI_File_close(&fd->mpifd);

        if (mrc != MPI_SUCCESS){
            fprintf(stderr, "MPI File close failed\n");
            GOTOERROR(FAIL);
        }

        fd->mpifd = MPI_FILE_NULL;
        break;

    case PHDF5:
        hrc = H5Fclose(fd->h5fd);

        if (hrc < 0) {
            fprintf(stderr, "HDF5 File Close failed\n");
            GOTOERROR(FAIL);
        }

        fd->h5fd = -1;
        break;
    }

done:
    return ret_code;
}


/*
 * Function:    do_fclose
 * Purpose:     Cleanup temporary file unless HDF5_NOCLEANUP is set.
 * 		Only Proc 0 of the PIO communicator will do the cleanup.
 *		Other processes just return.
 * Return:      void
 * Programmer:  Albert Cheng 2001/12/12
 * Modifications:
 */
static void
do_cleanupfile(iotype iot, char *fname)
{
    if (pio_mpi_rank_g != 0)
        return;
    
    if (clean_file_g == -1)
        clean_file_g = (getenv("HDF5_NOCLEANUP")==NULL) ? 1 : 0;
    
    if (clean_file_g){
        switch (iot){
        case POSIXIO:
            remove(fname);
            break;
        case MPIO:
        case PHDF5:
            MPI_File_delete(fname, h5_io_info_g);
            break;
        }
    }
}

#ifdef H5_HAVE_GPFS

    /* Descriptions here come from the IBM GPFS Manual */

/*
 * Function:    access_range
 * Purpose:     Declares an access range within a file for an
 *              application.
 *
 *              The application will access file offsets within the given
 *              range, and will not access offsets outside the range.
 *              Violating this hint may produce worse performance than if
 *              no hint was specified.
 *
 *              This hint is useful in situations where a file is
 *              partitioned coarsely among several nodes. If the ranges
 *              do not overlap, each node can specify which range of the
 *              file it will access, with a performance improvement in
 *              some cases, such as for sequential writing within a
 *              range.
 *
 *              Subsequent GPFS_ACCESS_RANGE hints will replace a hint
 *              passed earlier.
 *
 *                  START  - The start of the access range offset, in
 *                           bytes, from the beginning of the file.
 *                  LENGTH - Length of the access range. 0 indicates to
 *                           the end of the file.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 03. June 2002
 * Modifications:
 */
static void
access_range(int handle, off_t start, off_t length, int is_write)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsAccessRange_t access;
    } access_range;

    access_range.hdr.totalLength = sizeof(access_range);
    access_range.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    access_range.hdr.fcntlReserved = 0;
    access_range.start.structLen = sizeof(gpfsAccessRange_t);
    access_range.start.structType = GPFS_ACCESS_RANGE;
    access_range.start.start = start;
    access_range.start.length = length;
    access_range.start.isWrite = is_write;

    if (gpfs_fcntl(handle, &access_range) != 0) {
        fprintf(stderr,
                "gpfs_fcntl DS start directive failed. errno=%d errorOffset=%d\n",
                errno, ds_start.hdr.errorOffset);
        exit(EXIT_FAILURE);
    }
}

/*
 * Function:    free_range
 * Purpose:     Undeclares an access range within a file for an
 *              application.
 *
 *              The application will no longer access file offsets within
 *              the given range. GPFS flushes the data at the file
 *              offsets and removes it from the cache.
 *
 *              Multi-node applications that have finished one phase of
 *              their computation may wish to use this hint before the
 *              file is accessed in a conflicting mode from another node
 *              in a later phase. The potential performance benefit is
 *              that GPFS can avoid later synchronous cache consistency
 *              operations.
 *
 *                  START  - The start of the access range offset, in
 *                           bytes from the beginning of the file.
 *                  LENGTH - Length of the access range. 0 indicates to
 *                           the end of the file.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 03. June 2002
 * Modifications:
 */
static void
free_range(int handle, off_t start, off_t length)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsFreeRange_t range;
    } free_range;

    /* Issue the invalidate hint */
    free_range.hdr.totalLength = sizeof(free_range);
    free_range.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    free_range.hdr.fcntlReserved = 0;
    free_range.range.structLen = sizeof(gpfsFreeRange_t);
    free_range.range.structType = GPFS_FREE_RANGE;
    free_range.range.start = start;
    free_range.range.length = length;

    if (gpfs_fcntl(handle, &free_range) != 0) {
        fprintf(stderr,
                "gpfs_fcntl free range failed for range %d:%d. errno=%d errorOffset=%d\n",
                start, length, errno, free_range.hdr.errorOffset);
        exit(EXIT_FAILURE);
    }
}

/*
 * Function:    clear_file_cache
 * Purpose:     Indicates file access in the near future is not expected.
 *
 *              The application does not expect to make any further
 *              accesses to the file in the near future, so GPFS removes
 *              any data or metadata pertaining to the file from its
 *              cache.
 *
 *              Multi-node applications that have finished one phase of
 *              their computation may wish to use this hint before the
 *              file is accessed in a conflicting mode from another node
 *              in a later phase. The potential performance benefit is
 *              that GPFS can avoid later synchronous cache consistency
 *              operations.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 03. June 2002
 * Modifications:
 */
static void
clear_file_cache(int handle)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsClearFileCache_t clear;
    } clear_cache;

    clear_cache.hdr.totalLength = sizeof(clear_cache);
    clear_cache.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    clear_cache.hdr.fcntlReserved = 0;
    clear_cache.start.structLen = sizeof(gpfsClearFileCache_t);
    clear_cache.start.structType = GPFS_CLEAR_FILE_CACHE;

    if (gpfs_fcntl(handle, &clear_cache) != 0) {
        fprintf(stderr,
                "gpfs_fcntl clear file cache directive failed. errno=%d errorOffset=%d\n",
                errno, clear_cache.hdr.errorOffset);
        exit(EXIT_FAILURE);
    }
}

/*
 * Function:    cancel_hints
 * Purpose:     Indicates to remove any hints against the open file
 *              handle.
 *
 *              GPFS removes any hints that may have been issued against
 *              this open file handle:
 *
 *                  - The hint status of the file is restored ot what it
 *                    would have been immediately after being opened, but
 *                    does not affect the contents of the GPFS file
 *                    cache. Cancelling an earlier hint that resulted in
 *                    data being removed from the GPFS file cache does
 *                    not bring that data back int othe cache; data
 *                    re-enters the cache only pon access by the
 *                    application or by user-driven or automatic
 *                    prefetching.
 *                  - Only the GPFS_MULTIPLE_ACCESS_RANGE hint has a
 *                    state that might be removed by the
 *                    GPFS_CANCEL_HINTS directive.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 03. June 2002
 * Modifications:
 */
static void
cancel_hints(int handle)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsCancelHints_t cancel;
    } cancel_hints;

    cancel_hints.hdr.totalLength = sizeof(cancel_hints);
    cancel_hints.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    cancel_hints.hdr.fcntlReserved = 0;
    cancel_hints.start.structLen = sizeof(gpfsCancelHints_t);
    cancel_hints.start.structType = GPFS_CANCEL_HINTS;

    if (gpfs_fcntl(handle, &cancel_hints) != 0) {
        fprintf(stderr,
                "gpfs_fcntl cancel hints directive failed. errno=%d errorOffset=%d\n",
                errno, ds_start.hdr.errorOffset);
        exit(EXIT_FAILURE);
    }
}

/*
 * Function:    start_data_shipping
 * Purpose:     Start up data shipping. The second parameter is the total
 *              number of open instances on all nodes that will be
 *              operating on the file. Must be called for every such
 *              instance with the same value of NUM_INSTS.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 28. May 2002
 * Modifications:
 */
static void
start_data_shipping(int handle, int num_insts)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsDataShipStart_t start;
    } ds_start;

    ds_start.hdr.totalLength = sizeof(ds_start);
    ds_start.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    ds_start.hdr.fcntlReserved = 0;
    ds_start.start.structLen = sizeof(gpfsDataShipStart_t);
    ds_start.start.structType = GPFS_DATA_SHIP_START;
    ds_start.start.numInstances = num_insts;
    ds_start.start.reserved = 0;

    if (gpfs_fcntl(handle, &ds_start) != 0) {
        fprintf(stderr,
                "gpfs_fcntl DS start directive failed. errno=%d errorOffset=%d\n",
                errno, ds_start.hdr.errorOffset);
        exit(EXIT_FAILURE);
    }
}

/*
 * Function:    stop_data_shipping
 * Purpose:     Shut down data shipping. Must be called for every handle
 *              for which start_data_shipping was called.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 28. May 2002
 * Modifications:
 */
static void
stop_data_shipping(int handle)
{
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsDataShipStop_t stop;
    } ds_stop;

    ds_stop.hdr.totalLength = sizeof(ds_stop);
    ds_stop.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    ds_stop.hdr.fcntlReserved = 0;
    ds_stop.stop.structLen = sizeof(ds_stop.stop);
    ds_stop.stop.structType = GPFS_DATA_SHIP_STOP;

    if (gpfs_fcntl(handle, &ds_stop) != 0)
        fprintf(stderr,
                "gpfs_fcntl DS stop directive failed. errno=%d errorOffset=%d\n",
                errno, ds_stop.hdr.errorOffset);
}

/*
 * Function:    invalidate_file_cache
 * Purpose:     Invalidate all cached data held on behalf of a file on
 *              this node.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 03. June 2002
 * Modifications:
 */
static void
invalidate_file_cache(const char *filename)
{
    int handle;
    struct {
        gpfsFcntlHeader_t hdr;
        gpfsClearFileCache_t inv;
    } inv_cache_hint;

    /* Open the file.  If the open fails, the file cannot be cached. */
    handle = open(filename, O_RDONLY, 0);

    if (handle == -1)
        return;

    /* Issue the invalidate hint */
    inv_cache_hint.hdr.totalLength = sizeof(inv_cache_hint);
    inv_cache_hint.hdr.fcntlVersion = GPFS_FCNTL_CURRENT_VERSION;
    inv_cache_hint.hdr.fcntlReserved = 0;
    inv_cache_hint.inv.structLen = sizeof(gpfsClearFileCache_t);
    inv_cache_hint.inv.structType = GPFS_CLEAR_FILE_CACHE;

    if (gpfs_fcntl(handle, &inv_cache_hint) != 0) {
        fprintf(stderr,
                "gpfs_fcntl clear cache hint failed for file '%s'.",
                filename);
        fprintf(stderr, " errno=%d errorOffset=%d\n",
                errno, inv_cache_hint.hdr.errorOffset);
        exit(1);
    }

    /* Close the file */
    if (close(handle) == -1) {
        fprintf(stderr,
                "could not close file '%s' after flushing file cache,",
                filename);
        fprintf(stderr, "errno=%d\n", errno);
        exit(1);
    }
}

#else

/* H5_HAVE_GPFS isn't defined...stub functions */

static void
access_range(int UNUSED handle, off_t UNUSED start, off_t UNUSED length, int UNUSED is_write)
{
    return;
}

static void
free_range(int UNUSED handle, off_t UNUSED start, off_t UNUSED length)
{
    return;
}

static void
clear_file_cache(int UNUSED handle)
{
    return;
}

static void
cancel_hints(int UNUSED handle)
{
    return;
}

static void
start_data_shipping(int UNUSED handle, int UNUSED num_insts)
{
    return;
}

static void
stop_data_shipping(int UNUSED handle)
{
    return;
}

static void
invalidate_file_cache(const char UNUSED *filename)
{
    return;
}

#endif  /* H5_HAVE_GPFS */

#ifdef TIME_MPI
/* instrument the MPI_File_wrirte_xxx and read_xxx calls to measure
 * pure time spent in MPI_File code.
 */
int MPI_File_read_at(MPI_File fh, MPI_Offset offset, void *buf,
	  int count, MPI_Datatype datatype, MPI_Status *status)
{
    int err;
    set_time(timer_g, HDF5_MPI_READ, START);
    err=PMPI_File_read_at(fh, offset, buf, count, datatype, status);
    set_time(timer_g, HDF5_MPI_READ, STOP);
    return err;
}


int MPI_File_read_at_all(MPI_File fh, MPI_Offset offset, void *buf,
	int count, MPI_Datatype datatype, MPI_Status *status)
{
    int err;
    set_time(timer_g, HDF5_MPI_READ, START);
    err=PMPI_File_read_at_all(fh, offset, buf, count, datatype, status);
    set_time(timer_g, HDF5_MPI_READ, STOP);
    return err;
}

int MPI_File_write_at(MPI_File fh, MPI_Offset offset, void *buf,
      int count, MPI_Datatype datatype, MPI_Status *status)
{
    int err;
    set_time(timer_g, HDF5_MPI_WRITE, START);
    err=PMPI_File_write_at(fh, offset, buf, count, datatype, status);
    set_time(timer_g, HDF5_MPI_WRITE, STOP);
    return err;
}

int MPI_File_write_at_all(MPI_File fh, MPI_Offset offset, void *buf,
    int count, MPI_Datatype datatype, MPI_Status *status)
{
    int err;
    set_time(timer_g, HDF5_MPI_WRITE, START);
    err=PMPI_File_write_at_all(fh, offset, buf, count, datatype, status);
    set_time(timer_g, HDF5_MPI_WRITE, STOP);
    return err;
}

#endif	/* TIME_MPI */
#endif /* H5_HAVE_PARALLEL */