summaryrefslogtreecommitdiffstats
path: root/testpar/t_mpi.c
blob: 97bac9ceb9fdd55658bdff28577aeaa7f8f1f2d4 (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
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
 * 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. *
 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

/*
 * MPIO independent overlapping writes.
 *
 * First n-1 processes open 1 file.
 * Each of the n-1 process writes chunks of data to the file in round-robin
 * fashion, in a interleaved but not overlapped fashion.  Using increasing
 * chunk sizes for the benefits of testing different write sizes and also
 * reducing the numbers of writes.
 *
 * Last process (n-1) just waits.
 * First n-1 processes finish writing and cloose the file.
 * Last process opens the same file and verifies the data.
 */

#include "testpar.h"

/* FILENAME and filenames must have the same number of names */
const char *FILENAME[2]={
	    "MPItest",
	    NULL};
char	filenames[2][200];
int	nerrors = 0;
hid_t	fapl;				/* file access property list */

/* protocols */
static int errors_sum(int nerrs);

#define MPIO_TEST_WRITE_SIZE 1024*1024     /* 1 MB */

static int
test_mpio_overlap_writes(char *filename)
{
    int mpi_size, mpi_rank;
    MPI_Comm comm;
    MPI_Info info = MPI_INFO_NULL;
    int color, mrc;
    MPI_File	fh;
    int i;
    int vrfyerrs, nerrs;
    unsigned char  buf[4093];		/* use some prime number for size */
    int bufsize = sizeof(buf);
    MPI_Offset  stride;
    MPI_Offset  mpi_off;
    MPI_Status  mpi_stat;


    if (VERBOSE_MED)
	printf("MPIO independent overlapping writes test on file %s\n",
	    filename);

    nerrs = 0;
    /* set up MPI parameters */
    MPI_Comm_size(MPI_COMM_WORLD,&mpi_size);
    MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank);

    /* Need at least 2 processes */
    if (mpi_size < 2) {
	if (MAINPROCESS)
	    printf("Need at least 2 processes to run MPIO test.\n");
	    printf(" -SKIP- \n");
	return 0;
    }

    /* splits processes 0 to n-2 into one comm. and the last one into another */
    color = ((mpi_rank < (mpi_size - 1)) ? 0 : 1);
    mrc = MPI_Comm_split (MPI_COMM_WORLD, color, mpi_rank, &comm);
    VRFY((mrc==MPI_SUCCESS), "Comm_split succeeded");

    if (color==0){
	/* First n-1 processes (color==0) open a file and write it */
	mrc = MPI_File_open(comm, filename, MPI_MODE_CREATE|MPI_MODE_RDWR,
		info, &fh);
	VRFY((mrc==MPI_SUCCESS), "");

	stride = 1;
	mpi_off = mpi_rank*stride;
	while (mpi_off < MPIO_TEST_WRITE_SIZE){
	    /* make sure the write does not exceed the TEST_WRITE_SIZE */
	    if (mpi_off+stride > MPIO_TEST_WRITE_SIZE)
		stride = MPIO_TEST_WRITE_SIZE - mpi_off;

	    /* set data to some trivial pattern for easy verification */
	    for (i=0; i<stride; i++)
		buf[i] = (unsigned char)(mpi_off+i);
	    mrc = MPI_File_write_at(fh, mpi_off, buf, (int)stride, MPI_BYTE,
		    &mpi_stat);
	    VRFY((mrc==MPI_SUCCESS), "");
	    
	    /* move the offset pointer to last byte written by all processes */
	    mpi_off += (mpi_size - 1 - mpi_rank) * stride;

	    /* Increase chunk size without exceeding buffer size. */
	    /* Then move the starting offset for next write. */
	    stride *= 2;
	    if (stride > bufsize)
		stride = bufsize;
	    mpi_off += mpi_rank*stride;
	}

	/* close file and free the communicator */
	mrc = MPI_File_close(&fh);
	VRFY((mrc==MPI_SUCCESS), "MPI_FILE_CLOSE");
	mrc = MPI_Comm_free(&comm);
	VRFY((mrc==MPI_SUCCESS), "MPI_Comm_free");
    
	/* sync with the other waiting processes */
	mrc = MPI_Barrier(MPI_COMM_WORLD);
	VRFY((mrc==MPI_SUCCESS), "Sync after writes");
    }else{
	/* last process waits till writes are done,
	 * then opens file to verify data.
	 */
	mrc = MPI_Barrier(MPI_COMM_WORLD);
	VRFY((mrc==MPI_SUCCESS), "Sync after writes");

	mrc = MPI_File_open(comm, filename, MPI_MODE_RDONLY,
		info, &fh);
	VRFY((mrc==MPI_SUCCESS), "");

	stride = bufsize;
	for (mpi_off=0; mpi_off < MPIO_TEST_WRITE_SIZE; mpi_off += bufsize){
	    /* make sure it does not read beyond end of data */
	    if (mpi_off+stride > MPIO_TEST_WRITE_SIZE)
		stride = MPIO_TEST_WRITE_SIZE - mpi_off;
	    mrc = MPI_File_read_at(fh, mpi_off, buf, (int)stride, MPI_BYTE,
		    &mpi_stat);
	    VRFY((mrc==MPI_SUCCESS), "");
	    vrfyerrs=0;
	    for (i=0; i<stride; i++){
		unsigned char expected;
		expected = (unsigned char)(mpi_off+i);
		if ((expected != buf[i]) &&
		    (vrfyerrs++ < MAX_ERR_REPORT || VERBOSE_MED)) {
			printf("proc %d: found data error at [%ld], expect %u, got %u\n",
			    mpi_rank, (long)(mpi_off+i), expected, buf[i]);
		}
	    }
	    if (vrfyerrs > MAX_ERR_REPORT && !VERBOSE_MED)
		printf("proc %d: [more errors ...]\n", mpi_rank);

	    nerrs += vrfyerrs;
	}

	/* close file and free the communicator */
	mrc = MPI_File_close(&fh);
	VRFY((mrc==MPI_SUCCESS), "MPI_FILE_CLOSE");
	mrc = MPI_Comm_free(&comm);
	VRFY((mrc==MPI_SUCCESS), "MPI_Comm_free");
    }

    /*
     * one more sync to ensure all processes have done reading
     * before ending this test.
     */
    mrc = MPI_Barrier(MPI_COMM_WORLD);
    VRFY((mrc==MPI_SUCCESS), "Sync before leaving test");
    return (nerrs);
}


#define MB      1048576        /* 1024*1024 == 2**20 */
#define GB      1073741824     /* 1024**3 == 2**30 */
#define TWO_GB_LESS1    2147483647     /* 2**31 - 1 */
#define FOUR_GB_LESS1   4294967295L     /* 2**32 - 1 */
/*
 * Verify that MPI_Offset exceeding 2**31 can be computed correctly.
 * Print any failure as information only, not as an error so that this
 * won't abort the remaining test or other separated tests.
 *
 * Test if MPIO can write file from under 2GB to over 2GB and then 
 * from under 4GB to over 4GB.
 * Each process writes 1MB in round robin fashion.
 * Then reads the file back in by reverse order, that is process 0 
 * reads the data of process n-1 and vice versa.
 */
static int
test_mpio_gb_file(char *filename)
{
    int mpi_size, mpi_rank;
    MPI_Info info = MPI_INFO_NULL;
    int mrc;
    MPI_File	fh;
    int i, j, n;
    int vrfyerrs;
    int writerrs;		/* write errors */
    int nerrs;
    int ntimes;			/* how many times */
    char  *buf = NULL;
    char  expected;
    MPI_Offset  mpi_off;
    MPI_Offset  mpi_off_old;
    MPI_Status  mpi_stat;
    int is_signed, sizeof_mpi_offset;

    nerrs = 0;
    /* set up MPI parameters */
    MPI_Comm_size(MPI_COMM_WORLD,&mpi_size);
    MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank);

    if (VERBOSE_MED)
        printf("MPI_Offset range test\n");

    /* figure out the signness and sizeof MPI_Offset */
    mpi_off = 0;
    is_signed = ((MPI_Offset)(mpi_off - 1)) < 0;
    sizeof_mpi_offset = (int)(sizeof(MPI_Offset));

    /*
     * Verify the sizeof MPI_Offset and correctness of handling multiple GB
     * sizes.
     */
    if (MAINPROCESS){			/* only process 0 needs to check it*/
	printf("MPI_Offset is %s %d bytes integeral type\n",
	    is_signed ? "signed" : "unsigned", (int)sizeof(MPI_Offset));
	if (sizeof_mpi_offset <= 4 && is_signed){
	    printf("Skipped 2GB range test "
		    "because MPI_Offset cannot support it\n");
	}else {
	    /* verify correctness of assigning 2GB sizes */
	    mpi_off = 2 * 1024 * (MPI_Offset)MB;
	    INFO((mpi_off>0), "2GB OFFSET assignment no overflow");
	    INFO((mpi_off-1)==TWO_GB_LESS1, "2GB OFFSET assignment succeed");

	    /* verify correctness of increasing from below 2 GB to above 2GB */
	    mpi_off = TWO_GB_LESS1;
	    for (i=0; i < 3; i++){
		mpi_off_old = mpi_off;
		mpi_off = mpi_off + 1;
		/* no overflow */
		INFO((mpi_off>0), "2GB OFFSET increment no overflow");
		/* correct inc. */
		INFO((mpi_off-1)==mpi_off_old, "2GB OFFSET increment succeed");
	    }
	}

	if (sizeof_mpi_offset <= 4){
	    printf("Skipped 4GB range test "
		    "because MPI_Offset cannot support it\n");
	}else {
	    /* verify correctness of assigning 4GB sizes */
	    mpi_off = 4 * 1024 * (MPI_Offset)MB;
	    INFO((mpi_off>0), "4GB OFFSET assignment no overflow");
	    INFO((mpi_off-1)==FOUR_GB_LESS1, "4GB OFFSET assignment succeed");

	    /* verify correctness of increasing from below 4 GB to above 4 GB */
	    mpi_off = FOUR_GB_LESS1;
	    for (i=0; i < 3; i++){
		mpi_off_old = mpi_off;
		mpi_off = mpi_off + 1;
		/* no overflow */
		INFO((mpi_off>0), "4GB OFFSET increment no overflow");
		/* correct inc. */
		INFO((mpi_off-1)==mpi_off_old, "4GB OFFSET increment succeed");
	    }
	}
    }

    /*
     * Verify if we can write to a file of multiple GB sizes.
     */
    if (VERBOSE_MED)
	printf("MPIO GB file test %s\n", filename);

    if (sizeof_mpi_offset <= 4){
	printf("Skipped GB file range test "
		"because MPI_Offset cannot support it\n");
    }else{
	buf = malloc(MB);
	VRFY((buf!=NULL), "malloc succeed");

	/* open a new file. Remove it first in case it exists. */
	if (MAINPROCESS)
	    remove(filename);
	MPI_Barrier(MPI_COMM_WORLD);	/* prevent racing condition */

	mrc = MPI_File_open(MPI_COMM_WORLD, filename, MPI_MODE_CREATE|MPI_MODE_RDWR,
		    info, &fh);
	VRFY((mrc==MPI_SUCCESS), "MPI_FILE_OPEN");

	printf("MPIO GB file write test %s\n", filename);

	/* instead of writing every bytes of the file, we will just write
	 * some data around the 2 and 4 GB boundaries.  That should cover
	 * potential integer overflow and filesystem size limits.
	 */
	writerrs = 0;
	for (n=2; n <= 4; n+=2){
	    ntimes = GB/MB*n/mpi_size + 1;
	    for (i=ntimes-2; i <= ntimes; i++){
		mpi_off = (i*mpi_size + mpi_rank)*(MPI_Offset)MB;
		if (VERBOSE_MED)
		    HDfprintf(stdout,"proc %d: write to mpi_off=%016llx, %lld\n",
			mpi_rank, mpi_off, mpi_off);
		/* set data to some trivial pattern for easy verification */
		for (j=0; j<MB; j++)
		    *(buf+j) = i*mpi_size + mpi_rank;
		if (VERBOSE_MED)
		    HDfprintf(stdout,"proc %d: writing %d bytes at offset %lld\n",
			mpi_rank, MB, mpi_off);
		mrc = MPI_File_write_at(fh, mpi_off, buf, MB, MPI_BYTE, &mpi_stat);
		INFO((mrc==MPI_SUCCESS), "GB size file write");
		if (mrc!=MPI_SUCCESS)
		    writerrs++;
	    }
	}

	/* close file and free the communicator */
	mrc = MPI_File_close(&fh);
	VRFY((mrc==MPI_SUCCESS), "MPI_FILE_CLOSE");
	
	mrc = MPI_Barrier(MPI_COMM_WORLD);
	VRFY((mrc==MPI_SUCCESS), "Sync after writes");

	/*
	 * Verify if we can read the multiple GB file just created.
	 */
	/* open it again to verify the data written */
	/* but only if there was no write errors */
	printf("MPIO GB file read test %s\n", filename);
	if (errors_sum(writerrs)>0){
	    printf("proc %d: Skip read test due to previous write errors\n",
		mpi_rank);
	    goto finish;
	}
	mrc = MPI_File_open(MPI_COMM_WORLD, filename, MPI_MODE_RDONLY, info, &fh);
	VRFY((mrc==MPI_SUCCESS), "");

	/* Only read back parts of the file that have been written. */
	for (n=2; n <= 4; n+=2){
	    ntimes = GB/MB*n/mpi_size + 1;
	    for (i=ntimes-2; i <= ntimes; i++){
		mpi_off = (i*mpi_size + (mpi_size - mpi_rank - 1))*(MPI_Offset)MB;
		if (VERBOSE_MED)
		    HDfprintf(stdout,"proc %d: read from mpi_off=%016llx, %lld\n",
			mpi_rank, mpi_off, mpi_off);
		mrc = MPI_File_read_at(fh, mpi_off, buf, MB, MPI_BYTE, &mpi_stat);
		INFO((mrc==MPI_SUCCESS), "GB size file read");
		expected = i*mpi_size + (mpi_size - mpi_rank - 1);
		vrfyerrs=0;
		for (j=0; j<MB; j++){
		    if ((*(buf+j) != expected) &&
			(vrfyerrs++ < MAX_ERR_REPORT || VERBOSE_MED)){
			    printf("proc %d: found data error at [%ld+%d], expect %d, got %d\n",
				mpi_rank, (long)mpi_off, j, expected, *(buf+j));
		    }
		}
		if (vrfyerrs > MAX_ERR_REPORT && !VERBOSE_MED)
		    printf("proc %d: [more errors ...]\n", mpi_rank);

		nerrs += vrfyerrs;
	    }
	}

	/* close file and free the communicator */
	mrc = MPI_File_close(&fh);
	VRFY((mrc==MPI_SUCCESS), "MPI_FILE_CLOSE");

	/*
	 * one more sync to ensure all processes have done reading
	 * before ending this test.
	 */
	mrc = MPI_Barrier(MPI_COMM_WORLD);
	VRFY((mrc==MPI_SUCCESS), "Sync before leaving test");
    }

finish:
    if (buf)
	HDfree(buf);
    return (nerrs);
}


/*
 * MPI-IO Test: One writes, Many reads.
 * Verify if only one process writes some data and then all other
 * processes can read them back correctly. This tests if the
 * underlaying parallel I/O and file system supports parallel I/O
 * correctly.
 *
 * Algorithm: Only one process (e.g., process 0) writes some data.
 * Then all processes, including the writing process, read the data
 * back and verify them against the original values.
 */

/*
 * Default filename can be specified via first program argument.
 * Each process writes something, then reads all data back.
 */

#define DIMSIZE	32		/* Dimension size. */
#define PRINTID printf("Proc %d: ", mpi_rank)
#define USENONE 0
#define USEATOM 1		/* request atomic I/O */
#define USEFSYNC 2		/* request file_sync */


static int
test_mpio_1wMr(char *filename, int special_request)
{
    char hostname[128];
    int  mpi_size, mpi_rank;
    MPI_File fh;
    char mpi_err_str[MPI_MAX_ERROR_STRING];
    int  mpi_err_strlen;
    int  mpi_err;
    unsigned char writedata[DIMSIZE], readdata[DIMSIZE];
    unsigned char expect_val;
    int  i, irank; 
    int  nerrs = 0;		/* number of errors */
    int  atomicity;
    MPI_Offset  mpi_off;
    MPI_Status  mpi_stat;

    MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
    MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);

    if (MAINPROCESS && VERBOSE_MED){
        printf("Testing one process writes, all processes read.\n");
	printf("Using %d processes accessing file %s\n", mpi_size, filename);
        printf("    (Filename can be specified via program argument)\n");
    }

    /* show the hostname so that we can tell where the processes are running */
    if (VERBOSE_DEF){
	if (gethostname(hostname, 128) < 0){
	    PRINTID;
	    printf("gethostname failed\n");
	    return 1;
	}
	PRINTID;
	printf("hostname=%s\n", hostname);
    }

    /* Delete any old file in order to start anew. */
    /* Must delete because MPI_File_open does not have a Truncate mode. */
    /* Don't care if it has error. */
    MPI_File_delete(filename, MPI_INFO_NULL);

    if ((mpi_err = MPI_File_open(MPI_COMM_WORLD, filename,
	    MPI_MODE_RDWR | MPI_MODE_CREATE ,
	    MPI_INFO_NULL, &fh))
	    != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_open failed (%s)\n", mpi_err_str);
	return 1;
    }

if (special_request & USEATOM){
    /* ==================================================
     * Set atomcity to true (1).  A POSIX compliant filesystem
     * should not need this.
     * ==================================================*/
    if ((mpi_err = MPI_File_get_atomicity(fh, &atomicity)) != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_get_atomicity failed (%s)\n", mpi_err_str);
    }
    if (VERBOSE_HI)
	printf("Initial atomicity = %d\n", atomicity);
    if ((mpi_err = MPI_File_set_atomicity(fh, 1)) != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_set_atomicity failed (%s)\n", mpi_err_str);
    }
    if ((mpi_err = MPI_File_get_atomicity(fh, &atomicity)) != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_get_atomicity failed (%s)\n", mpi_err_str);
    }
    if (VERBOSE_HI)
	printf("After set_atomicity atomicity = %d\n", atomicity);
}

    /* This barrier is not necessary but do it anyway. */
    MPI_Barrier(MPI_COMM_WORLD);
    if (VERBOSE_HI){
	PRINTID;
	printf("between MPI_Barrier and MPI_File_write_at\n");
    }

    /* ==================================================
     * Each process calculates what to write but
     * only process irank(0) writes.
     * ==================================================*/
    irank=0;
    for (i=0; i < DIMSIZE; i++)
	writedata[i] = irank*DIMSIZE + i;
    mpi_off = irank*DIMSIZE;

    /* Only one process writes */
    if (mpi_rank==irank){
	if (VERBOSE_HI){
	    PRINTID; printf("wrote %d bytes at %ld\n", DIMSIZE, (long)mpi_off);
	}
	if ((mpi_err = MPI_File_write_at(fh, mpi_off, writedata, DIMSIZE,
			MPI_BYTE, &mpi_stat))
		!= MPI_SUCCESS){
	    MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	    PRINTID;
	    printf("MPI_File_write_at offset(%ld), bytes (%d), failed (%s)\n",
		    (long) mpi_off, DIMSIZE, mpi_err_str);
	    return 1;
	};
    };

    /* Bcast the return code and */
    /* make sure all writing are done before reading. */
    MPI_Bcast(&mpi_err, 1, MPI_INT, irank, MPI_COMM_WORLD);
    if (VERBOSE_HI){
	PRINTID;
	printf("MPI_Bcast: mpi_err = %d\n", mpi_err);
    }

if (special_request & USEFSYNC){
    /* ==================================================
     * Do a file sync.  A POSIX compliant filesystem
     * should not need this.
     * ==================================================*/
    if (VERBOSE_HI)
	printf("Apply MPI_File_sync\n");
    /* call file_sync to force the write out */
    if ((mpi_err = MPI_File_sync(fh)) != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_sync failed (%s)\n", mpi_err_str);
    }
    MPI_Barrier(MPI_COMM_WORLD);
    /* call file_sync to force the write out */
    if ((mpi_err = MPI_File_sync(fh)) != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_sync failed (%s)\n", mpi_err_str);
    }
}

    /* This barrier is not necessary because the Bcase or File_sync above */
    /* should take care of it.  Do it anyway. */
    MPI_Barrier(MPI_COMM_WORLD);
    if (VERBOSE_HI){
	PRINTID;
	printf("after MPI_Barrier\n");
    }

    /* ==================================================
     * Each process reads what process 0 wrote and verify.
     * ==================================================*/
    irank=0;
    mpi_off = irank*DIMSIZE;
    if ((mpi_err = MPI_File_read_at(fh, mpi_off, readdata, DIMSIZE, MPI_BYTE,
	    &mpi_stat))
	    != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	PRINTID;
	printf("MPI_File_read_at offset(%ld), bytes (%d), failed (%s)\n",
		(long) mpi_off, DIMSIZE, mpi_err_str);
	return 1;
    };
    for (i=0; i < DIMSIZE; i++){
	expect_val = irank*DIMSIZE + i;
	if (readdata[i] != expect_val){
	    PRINTID;
	    printf("read data[%d:%d] got %02x, expect %02x\n", irank, i,
		    readdata[i], expect_val);
	    nerrs++;
	}
    }

    MPI_File_close(&fh);

    if (VERBOSE_HI){
	PRINTID;
	printf("%d data errors detected\n", nerrs);
    }

    mpi_err = MPI_Barrier(MPI_COMM_WORLD);
    return nerrs;
}

/* 

Function: test_mpio_derived_dtype

Test Whether the Displacement of MPI derived datatype 
(+ File_set_view + MPI_write)works or not on this MPI-IO package 
and this platform.

1. Details for the test: 
1) Create two derived datatypes with MPI_Type_hindexed:
        datatype1: 
	count = 1, blocklens = 1, offsets = 0, 
	base type = MPI_BYTE(essentially a char)
        datatype2: 
	count = 1, blocklens = 1, offsets = 1(byte), 
	base type = MPI_BYTE
        
2) Using these two derived datatypes,
   Build another derived datatype with MPI_Type_struct:
        advtype: derived from datatype1 and datatype2
        advtype: 
	count = 2, blocklens[0] = 1, blocklens[1]=1, 
	offsets[0] = 0, offsets[1] = 1(byte), 
        bas_type[0]=datatype1, 
        bas_type[1] = datatype2;

3) Setting MPI file view with advtype
4) Writing 2 bytes 1 to 2 using MPI_File_write to a file
5) File content:
Supposed the file value of the file is 0(most machines indeed do so) 
and Fill value is embraced with "() in the following output:
Expected output should be:
1,0,2



However, at some platforms, for example, IBM AIX(at March 23rd, 2005):
the following values were obtained:
1,2,0 

The problem is that the displacement of the second derived datatype(datatype2) which formed the final derived datatype(advtype)
 has been put after the basic datatype(MPI_BYTE) of datatype2. This is a bug.


2. This test will verify whether the complicated derived datatype is working on
the current platform. 

If this bug has been fixed in the previous not-working package, this test will issue a printf message to tell the developer to change
the configuration specific file of HDF5 so that we can change our configurationsetting to support collective IO for irregular selections.

If it turns out that the previous working MPI-IO package no longer works, this test will also issue a message to inform the corresponding failure so that
we can turn off collective IO support for irregular selections.
*/

static int test_mpio_derived_dtype(char *filename) {

    MPI_File fh;
    char mpi_err_str[MPI_MAX_ERROR_STRING];
    int  mpi_err_strlen;
    int  mpi_err;
    int  i; 
    int  nerrors = 0;		/* number of errors */
    MPI_Datatype  etype,filetype;
    MPI_Datatype  adv_filetype,bas_filetype[2];
    MPI_Datatype  etypenew, filetypenew;
    MPI_Offset    disp,dispnew;
    MPI_Aint      adv_disp[2];
    MPI_Aint      offsets[1],adv_offsets[2];
    int           blocklens[1],adv_blocklens[2];
    int           count,outcount;
    int           ret;

    int mpi_rank,mpi_size;

    char          buf[2],outbuf[2];

    MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
    MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
    ret = 0;
    for(i=0;i<2;i++)
      buf[i] = i+1;

  
    if ((mpi_err = MPI_File_open(MPI_COMM_WORLD, filename,
				 MPI_MODE_RDWR | MPI_MODE_CREATE, 
				 MPI_INFO_NULL, &fh))
	    != MPI_SUCCESS){
	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_open failed (%s)\n", mpi_err_str);
	return 1;
    }

    disp  = 0;
    etype = MPI_BYTE;

    count = 1;
    blocklens[0] = 1;
    offsets[0]   = 0;

    if((mpi_err= MPI_Type_hindexed(count,blocklens,offsets,MPI_BYTE,&filetype))
       != MPI_SUCCESS){
      	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_contiguous failed (%s)\n", mpi_err_str);
	return 1;
    }

    if((mpi_err=MPI_Type_commit(&filetype))!=MPI_SUCCESS){
        MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_commit failed (%s)\n", mpi_err_str);
	return 1;
    }

    count = 1;
    blocklens[0]=1;
    offsets[0] = 1;
    if((mpi_err= MPI_Type_hindexed(count,blocklens,offsets,MPI_BYTE,&filetypenew))
       != MPI_SUCCESS){
      	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_contiguous failed (%s)\n", mpi_err_str);
	return 1;
    }

    if((mpi_err=MPI_Type_commit(&filetypenew))!=MPI_SUCCESS){
        MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_commit failed (%s)\n", mpi_err_str);
	return 1;
    }

    outcount         = 2;
    adv_blocklens[0] = 1;
    adv_blocklens[1] = 1;
    adv_disp[0]      = 0;
    adv_disp[1]      = 1;
    bas_filetype[0]  = filetype;
    bas_filetype[1]  = filetypenew;

    if((mpi_err= MPI_Type_struct(outcount,adv_blocklens,adv_disp,bas_filetype,&adv_filetype))
       != MPI_SUCCESS){
      	MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_struct failed (%s)\n", mpi_err_str);
	return 1;
    }
    if((mpi_err=MPI_Type_commit(&adv_filetype))!=MPI_SUCCESS){
        MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_Type_commit failed (%s)\n", mpi_err_str);
	return 1;
    }


    if((mpi_err = MPI_File_set_view(fh,disp,etype,adv_filetype,"native",MPI_INFO_NULL))!= MPI_SUCCESS){
      MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_set_view failed (%s)\n", mpi_err_str);
	return 1;
    }

    if((mpi_err = MPI_File_write(fh,buf,2,MPI_BYTE,MPI_STATUS_IGNORE))!= MPI_SUCCESS){
        MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_write failed (%s)\n", mpi_err_str);
	return 1;
      ;
    }


    if((mpi_err = MPI_File_close(&fh)) != MPI_SUCCESS){
       MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_close failed (%s)\n", mpi_err_str);
	return 1;
    }


    if((mpi_err = MPI_File_open(MPI_COMM_WORLD,filename,MPI_MODE_RDONLY,MPI_INFO_NULL,&fh)) != MPI_SUCCESS){
       MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_open failed (%s)\n", mpi_err_str);
	return 1;
    }
    
    if((mpi_err = MPI_File_set_view(fh,0,MPI_BYTE,MPI_BYTE,"native",MPI_INFO_NULL))!= MPI_SUCCESS){
        MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_set_view failed (%s)\n", mpi_err_str);
	return 1;
    }
    if((mpi_err = MPI_File_read(fh,outbuf,2,MPI_BYTE,MPI_STATUS_IGNORE))!=MPI_SUCCESS){
      MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
      printf("MPI_File_read failed (%s)\n", mpi_err_str);
      return 1;
    }

    if(outbuf[1]==0) {
       ret = 0;
    }
    if(outbuf[1]==2) {
/*      if(mpi_rank == 0) {
       printf("complicated derived datatype is NOT working at this platform\n");
       printf("go back to hdf5/config and find the corresponding\n");
       printf("configure-specific file and change ?????\n");
      }
*/
       ret = -1;
   }

    if((mpi_err = MPI_File_close(&fh)) != MPI_SUCCESS){
       MPI_Error_string(mpi_err, mpi_err_str, &mpi_err_strlen);
	printf("MPI_File_close failed (%s)\n", mpi_err_str);
	return 1;
    }

    
    mpi_err = MPI_Barrier(MPI_COMM_WORLD);
    return ret;
}
/*
 * parse the command line options
 */
static int
parse_options(int argc, char **argv)
{
    while (--argc){
	if (**(++argv) != '-'){
	    break;
	}else{
	    switch(*(*argv+1)){
		case 'v':   if (*((*argv+1)+1))
				ParseTestVerbosity((*argv+1)+1);
			    else
				SetTestVerbosity(VERBO_MED);
			    break;
		case 'f':   if (--argc < 1) {
				nerrors++;
				return(1);
			    }
			    if (**(++argv) == '-') {
				nerrors++;
				return(1);
			    }
			    paraprefix = *argv;
			    break;
		case 'h':   /* print help message--return with nerrors set */
			    return(1);
		default:    nerrors++;
			    return(1);
	    }
	}
    } /*while*/

    /* compose the test filenames */
    {
	int i, n;
	hid_t plist;

	plist = H5Pcreate (H5P_FILE_ACCESS);
	H5Pset_fapl_mpio(plist, MPI_COMM_WORLD, MPI_INFO_NULL);
	n = sizeof(FILENAME)/sizeof(FILENAME[0]) - 1;	/* exclude the NULL */

	for (i=0; i < n; i++)
	    if (h5_fixname(FILENAME[i],plist,filenames[i],sizeof(filenames[i]))
		== NULL){
		printf("h5_fixname failed\n");
		nerrors++;
		return(1);
	    }
	H5Pclose(plist);
	if (VERBOSE_MED){
	    printf("Test filenames are:\n");
	    for (i=0; i < n; i++)
		printf("    %s\n", filenames[i]);
	}
    }

    return(0);
}


/*
 * Show command usage
 */
static void
usage(void)
{
    printf("Usage: t_mpi [-v<verbosity>] [-f <prefix>]\n");
    printf("\t-v<verbosity>\tset verbose level (0-9,l,m,h)\n");
    printf("\t-f <prefix>\tfilename prefix\n");
    printf("\n");
}

/*
 * return the sum of all errors.
 */
static int
errors_sum(int nerrs)
{
    int temp;
    MPI_Allreduce(&nerrs, &temp, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
    return(temp);
}


int
main(int argc, char **argv)
{
    int mpi_size, mpi_rank;				/* mpi variables */
    int ret_code;

    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
    MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);

    H5open();
    if (parse_options(argc, argv) != 0){
	if (MAINPROCESS)
	    usage();
	goto finish;
    }

    if (MAINPROCESS){
	printf("===================================\n");
	printf("MPI functionality tests\n");
	printf("===================================\n");
    }

    if (VERBOSE_MED)
	h5_show_hostname();

    fapl = H5Pcreate (H5P_FILE_ACCESS);
    H5Pset_fapl_mpio(fapl, MPI_COMM_WORLD, MPI_INFO_NULL);

    MPI_BANNER("MPIO complicated derived datatype test...");
    ret_code = test_mpio_derived_dtype(filenames[0]);
#ifdef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS 
    if(ret_code == -1) {
      if(mpi_rank == 0) {
        printf("Complicated derived datatype is NOT working at this platform\n"); 
        printf("Go back to hdf5/config and find the corresponding\n");  
        printf("configure-specific file (for example, powerpc-ibm-aix5.x) and add\n");
        printf("hdf5_mpi_complex_derived_datatype_works=${hdf5_mpi_complex_derived_datatype-works='no'}\n");
        printf(" at the end of the file.\n");
        printf(" Please report to hdfhelp@ncsa.uiuc.edu about this problem.\n");
      }
        ret_code = 1;
    }
#else
    if(ret_code == 0) {
      if(mpi_rank == 0) {
        printf(" This is NOT an error, What it really says is\n");
        printf("Complicated derived datatype is WORKING at this platform\n");
        printf(" Go back to hdf5/config and find the corresponding \n");
        printf(" configure-specific file (for example, powerpc-ibm-aix5.x) and delete the line\n");
        printf("hdf5_mpi_complex_derived_datatype_works=${hdf5_mpi_complex_derived_datatype-works='no'}\n");
        printf(" at the end of the file.\n");
        printf("Please report to hdfhelp@ncsa.uiuc.edu about this problem.\n");
      }
      ret_code = 1;
    }
    if(ret_code == -1) ret_code = 0;

    ret_code = errors_sum(ret_code);
    if (mpi_rank==0 && ret_code > 0){
	printf("***FAILED with %d total errors\n", ret_code);
	nerrors += ret_code;
    }
#endif

    MPI_BANNER("MPIO 1 write Many read test...");
    ret_code = test_mpio_1wMr(filenames[0], USENONE);
    ret_code = errors_sum(ret_code);
    if (mpi_rank==0 && ret_code > 0){
	printf("***FAILED with %d total errors\n", ret_code);
	nerrors += ret_code;
    }

    /* test atomicity and file sync in high verbose mode only         */
    /* since they often hang when broken and PHDF5 does not use them. */
    if (VERBOSE_HI){
	MPI_BANNER("MPIO 1 write Many read test with atomicity...");
	ret_code = test_mpio_1wMr(filenames[0], USEATOM);
	ret_code = errors_sum(ret_code);
	if (mpi_rank==0 && ret_code > 0){
	    printf("***FAILED with %d total errors\n", ret_code);
	    nerrors += ret_code;
	}

	MPI_BANNER("MPIO 1 write Many read test with file sync...");
	ret_code = test_mpio_1wMr(filenames[0], USEFSYNC);
	ret_code = errors_sum(ret_code);
	if (mpi_rank==0 && ret_code > 0){
	    printf("***FAILED with %d total errors\n", ret_code);
	    nerrors += ret_code;
	}
    }

    MPI_BANNER("MPIO File size range test...");
    ret_code = test_mpio_gb_file(filenames[0]);
    ret_code = errors_sum(ret_code);
    if (mpi_rank==0 && ret_code > 0){
	printf("***FAILED with %d total errors\n", ret_code);
	nerrors += ret_code;
    }

    MPI_BANNER("MPIO independent overlapping writes...");
    ret_code = test_mpio_overlap_writes(filenames[0]);
    ret_code = errors_sum(ret_code);
    if (mpi_rank==0 && ret_code > 0){
	printf("***FAILED with %d total errors\n", ret_code);
	nerrors += ret_code;
    }

finish:
    /* make sure all processes are finished before final report, cleanup
     * and exit.
     */
    MPI_Barrier(MPI_COMM_WORLD);
    if (MAINPROCESS){		/* only process 0 reports */
	printf("===================================\n");
	if (nerrors){
	    printf("***MPI tests detected %d errors***\n", nerrors);
	}
	else{
	    printf("MPI tests finished with no errors\n");
	}
	printf("===================================\n");
    }

    h5_cleanup(FILENAME, fapl);
    H5close();

    /* MPI_Finalize must be called AFTER H5close which may use MPI calls */
    MPI_Finalize();

    /* cannot just return (nerrors) because exit code is limited to 1byte */
    return(nerrors!=0);
}