summaryrefslogtreecommitdiffstats
path: root/perform/pio_perf.c
blob: e32f6750e101f79cc2f1eaccfed246b5d2de41db (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
/*
 * Copyright (C) 2001, 2002
 *     National Center for Supercomputing Applications
 *     All rights reserved.
 *
 */

/*
 * Parallel HDF5 Performance Testing Code
 * --------------------------------------
 *
 * Portable code to test performance on the different platforms we support.
 * This is what the report should look like:
 *
 *  nprocs = Max#Procs
 *      IO Type = Raw
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *      IO Type = MPIO
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *      IO Type = PHDF5
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *  nprocs = Max#Procs / 2
 *
 *      . . .
 *
 */

/* system header files */
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>

#include "hdf5.h"

#ifdef H5_HAVE_PARALLEL

/* library header files */
#include <mpi.h>

/* our header files */
#include "h5tools_utils.h"
#include "pio_perf.h"

/* useful macros */
#define TAB_SPACE           4

#define ONE_KB              1024
#define ONE_MB              (ONE_KB * ONE_KB)
#define ONE_GB              (ONE_MB * ONE_KB)

#define PIO_RAW             0x10
#define PIO_MPI             0x20
#define PIO_HDF5            0x40

#define MB_PER_SEC(bytes,t) (((bytes) / ONE_MB) / t)

/* global variables */
MPI_Comm    pio_comm_g;         /* Communicator to run the PIO          */
int         pio_mpi_rank_g;     /* MPI rank of pio_comm_g               */
int         pio_mpi_nprocs_g;   /* Number of processes of pio_comm_g    */
int         pio_debug_level = 0;/* The debug level:
                                 *   0 - Off
                                 *   1 - Minimal
                                 *   2 - Some more
                                 *   3 - Maximal
                                 */

/* local variables */
static const char  *progname = "pio_perf";

/*
 * Command-line options: The user can specify short or long-named
 * parameters. The long-named ones can be partially spelled. When
 * adding more, make sure that they don't clash with each other.
 */
#if 1
static const char *s_opts = "hD:f:HP:p:X:x:md:F:i:o:r";
#else
static const char *s_opts = "hbD:f:HP:p:X:x:md:F:i:o:r";
#endif  /* 1 */
static struct long_options l_opts[] = {
    { "help", no_arg, 'h' },
    { "hel", no_arg, 'h' },
    { "he", no_arg, 'h' },
#if 0
    /* a siting of the elusive binary option */
    { "binary", no_arg, 'b' },
    { "binar", no_arg, 'b' },
    { "bina", no_arg, 'b' },
    { "bin", no_arg, 'b' },
    { "bi", no_arg, 'b' },
#endif  /* 0 */
    { "debug", require_arg, 'D' },
    { "debu", require_arg, 'D' },
    { "deb", require_arg, 'D' },
    { "de", require_arg, 'D' },
    { "d", require_arg, 'D' },
    { "file-size", require_arg, 'f' },
    { "file-siz", require_arg, 'f' },
    { "file-si", require_arg, 'f' },
    { "file-s", require_arg, 'f' },
    { "file", require_arg, 'f' },
    { "fil", require_arg, 'f' },
    { "fi", require_arg, 'f' },
    { "hdf5", no_arg, 'H' },
    { "hdf", no_arg, 'H' },
    { "hd", no_arg, 'H' },
    { "max-num-processes", require_arg, 'P' },
    { "max-num-processe", require_arg, 'P' },
    { "max-num-process", require_arg, 'P' },
    { "max-num-proces", require_arg, 'P' },
    { "max-num-proce", require_arg, 'P' },
    { "max-num-proc", require_arg, 'P' },
    { "max-num-pro", require_arg, 'P' },
    { "max-num-pr", require_arg, 'P' },
    { "max-num-p", require_arg, 'P' },
    { "min-num-processes", require_arg, 'p' },
    { "min-num-processe", require_arg, 'p' },
    { "min-num-process", require_arg, 'p' },
    { "min-num-proces", require_arg, 'p' },
    { "min-num-proce", require_arg, 'p' },
    { "min-num-proc", require_arg, 'p' },
    { "min-num-pro", require_arg, 'p' },
    { "min-num-pr", require_arg, 'p' },
    { "min-num-p", require_arg, 'p' },
    { "max-xfer-size", require_arg, 'X' },
    { "max-xfer-siz", require_arg, 'X' },
    { "max-xfer-si", require_arg, 'X' },
    { "max-xfer-s", require_arg, 'X' },
    { "max-xfer", require_arg, 'X' },
    { "max-xfe", require_arg, 'X' },
    { "max-xf", require_arg, 'X' },
    { "max-x", require_arg, 'X' },
    { "min-xfer-size", require_arg, 'x' },
    { "min-xfer-siz", require_arg, 'x' },
    { "min-xfer-si", require_arg, 'x' },
    { "min-xfer-s", require_arg, 'x' },
    { "min-xfer", require_arg, 'x' },
    { "min-xfe", require_arg, 'x' },
    { "min-xf", require_arg, 'x' },
    { "min-x", require_arg, 'x' },
    { "mpiio", no_arg, 'm' },
    { "mpii", no_arg, 'm' },
    { "mpi", no_arg, 'm' },
    { "mp", no_arg, 'm' },
    { "num-dsets", require_arg, 'd' },
    { "num-dset", require_arg, 'd' },
    { "num-dse", require_arg, 'd' },
    { "num-ds", require_arg, 'd' },
    { "num-d", require_arg, 'd' },
    { "num-files", require_arg, 'F' },
    { "num-file", require_arg, 'F' },
    { "num-fil", require_arg, 'F' },
    { "num-fi", require_arg, 'F' },
    { "num-f", require_arg, 'F' },
    { "num-iterations", require_arg, 'i' },
    { "num-iteration", require_arg, 'i' },
    { "num-iteratio", require_arg, 'i' },
    { "num-iterati", require_arg, 'i' },
    { "num-iterat", require_arg, 'i' },
    { "num-itera", require_arg, 'i' },
    { "num-iter", require_arg, 'i' },
    { "num-ite", require_arg, 'i' },
    { "num-it", require_arg, 'i' },
    { "num-i", require_arg, 'i' },
    { "output", require_arg, 'o' },
    { "outpu", require_arg, 'o' },
    { "outp", require_arg, 'o' },
    { "out", require_arg, 'o' },
    { "ou", require_arg, 'o' },
    { "raw", no_arg, 'r' },
    { "ra", no_arg, 'r' },
    { NULL, 0, '\0' }
};

struct options {
    long io_types;              /* bitmask of which I/O types to test   */
    const char *output_file;    /* file to print report to              */
    long file_size;             /* size of file                         */
    long num_dsets;             /* number of datasets                   */
    long num_files;             /* number of files                      */
    long num_iters;             /* number of iterations                 */
    long max_num_procs;         /* maximum number of processes to use   */
    long min_num_procs;         /* minimum number of processes to use   */
    long max_xfer_size;         /* maximum transfer buffer size         */
    long min_xfer_size;         /* minimum transfer buffer size         */
};

typedef struct _minmax {
    double min;
    double max;
    double sum;
    int num;
} minmax;

/* local functions */
static long parse_size_directive(const char *size);
static struct options *parse_command_line(int argc, char *argv[]);
static void run_test_loop(FILE *output, struct options *options);
static int run_test(FILE *output, iotype iot, parameters parms);
static void output_all_info(FILE *output, minmax *mm, int count, int indent_level);
static void get_minmax(minmax *mm, double val);
static minmax accumulate_minmax_stuff(minmax *mm, long raw_size, int count);
static int create_comm_world(int num_procs, int *doing_pio);
static int destroy_comm_world(void);
static void output_report(FILE *output, const char *fmt, ...);
static void print_indent(register FILE *output, register int indent);
static void usage(const char *prog);

/*
 * Function:    main
 * Purpose:     Start things up. Initialize MPI and then call the test looping
 *              function.
 * Return:      EXIT_SUCCESS or EXIT_FAILURE
 * Programmer:  Bill Wendling, 30. October 2001
 * Modifications:
 */
int
main(int argc, char **argv)
{
    int world_size, ret;
    int exit_value = EXIT_SUCCESS;
    FILE *output = stdout;
    struct options *opts = NULL;

    /* initialize MPI and get the maximum num of processors we started with */
    MPI_Init(&argc, &argv);
    ret = MPI_Comm_size(MPI_COMM_WORLD, &world_size);

    if (ret != MPI_SUCCESS) {
        fprintf(stderr, "%s: MPI_Comm_size call failed\n", progname);

        if (ret == MPI_ERR_COMM)
            fprintf(stderr, "invalid MPI communicator\n");
        else
            fprintf(stderr, "invalid argument\n");

        exit_value = EXIT_FAILURE;
        goto finish;
    }

    pio_comm_g = MPI_COMM_WORLD;

    opts = parse_command_line(argc, argv);

    if (!opts) {
        exit_value = EXIT_FAILURE;
        goto finish;
    }

    if (opts->output_file) {
        if ((output = fopen(opts->output_file, "w")) == NULL) {
            fprintf(stderr, "%s: cannot open output file\n", progname);
            perror(opts->output_file);
            goto finish;
        }
    }

    run_test_loop(output, opts);

finish:
    MPI_Finalize();
    free(opts);
    return exit_value;
}

/*
 * Function:    run_test_loop
 * Purpose:     Run the I/O tests. Write the results to OUTPUT.
 *
 *            - The slowest changing part of the test is the number of
 *              processors to use. For each loop iteration, we divide that
 *              number by 2 and rerun the test.
 *
 *            - The second slowest is what type of IO to perform. We have
 *              three choices: RAW, MPI-IO, and PHDF5.
 *
 *            - Then we change the size of the buffer. This information is
 *              inferred from the number of datasets to create and the number
 *              of integers to put into each dataset. The backend code figures
 *              this out.
 *
 * Return:      Nothing
 * Programmer:  Bill Wendling, 30. October 2001
 * Modifications:
 */
static void
run_test_loop(FILE *output, struct options *opts)
{
    parameters parms;
    long num_procs;
    int		doing_pio;		/* if this process is doing PIO */
    int io_runs = PIO_HDF5 | PIO_MPI | PIO_RAW; /* default to run all tests */

    if (opts->io_types & ~0x7) {
        /* we want to run only a select subset of these tests */
        io_runs = 0;

        if (opts->io_types & PIO_HDF5)
            io_runs |= PIO_HDF5;

        if (opts->io_types & PIO_MPI)
            io_runs |= PIO_MPI;

        if (opts->io_types & PIO_RAW)
            io_runs |= PIO_RAW;
    }

    parms.num_files = opts->num_files;
    parms.num_dsets = opts->num_dsets;
    parms.num_iters = opts->num_iters;

    /* multiply the maximum number of processors by 2 for each loop iter */
    for (num_procs = opts->min_num_procs;
            num_procs <= opts->max_num_procs; num_procs <<= 1) {
        register long buf_size;

        parms.num_procs = num_procs;

        if (create_comm_world(parms.num_procs, &doing_pio) != SUCCESS) {
            /* do something harsh */
        }

	/* only processes doing PIO will run the tests */
	if (doing_pio){
            output_report(output, "Number of processors = %ld\n", parms.num_procs);

            /* multiply the xfer buffer size by 2 for each loop iteration */
            for (buf_size = opts->min_xfer_size;
                    buf_size <= opts->max_xfer_size; buf_size <<= 1) {
                parms.buf_size = buf_size;
                parms.num_elmts = opts->file_size / (parms.num_dsets * sizeof(int));

                print_indent(output, 1);
                output_report(output, "Transfer Buffer Size: %ld bytes, File size: %.2f MBs\n",
                              buf_size,
                              ((double)parms.num_dsets * parms.num_elmts * sizeof(int)) / ONE_MB);
                print_indent(output, 1);
                output_report(output,
                              "  # of files: %ld, # of dsets: %ld, # of elmts per dset: %ld\n",
                              parms.num_files, parms.num_dsets, parms.num_elmts);

                if (io_runs & PIO_RAW)
                    run_test(output, RAW, parms);

                if (io_runs & PIO_MPI)
                    run_test(output, MPIO, parms);

                if (io_runs & PIO_HDF5)
                    run_test(output, PHDF5, parms);
            }

            if (destroy_comm_world() != SUCCESS) {
                /* do something harsh */
            }
	}
    }
}

/*
 * Function:    run_test
 * Purpose:     Inner loop call to actually run the I/O test.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 18. December 2001
 * Modifications:
 */
static int
run_test(FILE *output, iotype iot, parameters parms)
{
    results         res;
    register int    i, ret_value = SUCCESS;
    int             comm_size;
    long            raw_size;
    minmax          total_mm;
    minmax         *write_mm_table;
    minmax         *write_gross_mm_table;
    minmax         *read_mm_table;
    minmax         *read_gross_mm_table;
    minmax          write_mm = {0.0, 0.0, 0.0, 0};
    minmax          write_gross_mm = {0.0, 0.0, 0.0, 0};
    minmax          read_mm = {0.0, 0.0, 0.0, 0};
    minmax          read_gross_mm = {0.0, 0.0, 0.0, 0};

    raw_size = parms.num_dsets * parms.num_elmts * sizeof(int);
    parms.io_type = iot;
    print_indent(output, 2);
    output_report(output, "Type of IO = ");

    switch (iot) {
    case RAW:
        output_report(output, "Raw\n");
        break;
    case MPIO:
        output_report(output, "MPIO\n");
        break;
    case PHDF5:
        output_report(output, "PHDF5\n");
        break;
    }

    MPI_Comm_size(pio_comm_g, &comm_size);

    write_mm_table = malloc(parms.num_iters * sizeof(minmax));
    write_gross_mm_table = malloc(parms.num_iters * sizeof(minmax));
    read_mm_table = malloc(parms.num_iters * sizeof(minmax));
    read_gross_mm_table = malloc(parms.num_iters * sizeof(minmax));

    for (i = 0; i < parms.num_iters; ++i) {
        write_mm_table[i].min = 0.0;
        write_mm_table[i].max = 0.0;
        write_mm_table[i].sum = 0.0;
        write_mm_table[i].num = 0;

        write_gross_mm_table[i].min = 0.0;
        write_gross_mm_table[i].max = 0.0;
        write_gross_mm_table[i].sum = 0.0;
        write_gross_mm_table[i].num = 0;

        read_mm_table[i].min = 0.0;
        read_mm_table[i].max = 0.0;
        read_mm_table[i].sum = 0.0;
        read_mm_table[i].num = 0;

        read_gross_mm_table[i].min = 0.0;
        read_gross_mm_table[i].max = 0.0;
        read_gross_mm_table[i].sum = 0.0;
        read_gross_mm_table[i].num = 0;
    }

    /* call Albert's testing here */
    for (i = 0; i < parms.num_iters; ++i) {
        double t;

        MPI_Barrier(pio_comm_g);
        res = do_pio(parms);

        /* gather all of the "write" times */
        t = get_time(res.timers, HDF5_FINE_WRITE_FIXED_DIMS);
	get_minmax(&write_mm, t);

        write_mm_table[i] = write_mm;

        /* gather all of the "write" times from open to close */
        t = get_time(res.timers, HDF5_GROSS_WRITE_FIXED_DIMS);
	get_minmax(&write_gross_mm, t);

        write_gross_mm_table[i] = write_gross_mm;

        /* gather all of the "read" times */
        t = get_time(res.timers, HDF5_FINE_READ_FIXED_DIMS);
	get_minmax(&read_mm, t);

        read_mm_table[i] = read_mm;

        /* gather all of the "read" times from open to close */
        t = get_time(res.timers, HDF5_GROSS_READ_FIXED_DIMS);
	get_minmax(&read_gross_mm, t);

        read_gross_mm_table[i] = read_gross_mm;
    }

    /* accumulate and output the max, min, and average "write" times */
    if (pio_debug_level == 3) {
        /* output all of the times for all iterations */
        print_indent(output, 3);
        output_report(output, "Write details:\n");
        output_all_info(output, write_mm_table, parms.num_iters, 4);
    }

    total_mm = accumulate_minmax_stuff(write_mm_table, raw_size, parms.num_iters);

    print_indent(output, 3);
    output_report(output, "Write (%d iteration(s)):\n", parms.num_iters);

    print_indent(output, 4);
    output_report(output, "Minimum Throughput: %.2f MB/s\n", total_mm.min);
    print_indent(output, 4);
    output_report(output, "Maximum Throughput: %.2f MB/s\n", total_mm.max);
    print_indent(output, 4);
    output_report(output, "Average Throughput: %.2f MB/s\n",
                  total_mm.sum / total_mm.num);

    /* accumulate and output the max, min, and average "gross write" times */
    if (pio_debug_level == 3) {
        /* output all of the times for all iterations */
        print_indent(output, 3);
        output_report(output, "Write Open-Close details:\n");
        output_all_info(output, write_gross_mm_table, parms.num_iters, 4);
    }

    total_mm = accumulate_minmax_stuff(write_gross_mm_table, raw_size, parms.num_iters);

    print_indent(output, 3);
    output_report(output, "Write Open-Close (%d iteration(s)):\n", parms.num_iters);

    print_indent(output, 4);
    output_report(output, "Minimum Throughput: %.2f MB/s\n", total_mm.min);
    print_indent(output, 4);
    output_report(output, "Maximum Throughput: %.2f MB/s\n", total_mm.max);
    print_indent(output, 4);
    output_report(output, "Average Throughput: %.2f MB/s\n",
                  total_mm.sum / total_mm.num);

    /* accumulate and output the max, min, and average "read" times */
    if (pio_debug_level == 3) {
        /* output all of the times for all iterations */
        print_indent(output, 3);
        output_report(output, "Read details:\n");
        output_all_info(output, read_mm_table, parms.num_iters, 4);
    }

    total_mm = accumulate_minmax_stuff(read_mm_table, raw_size, parms.num_iters);

    print_indent(output, 3);
    output_report(output, "Read (%d iteration(s)):\n", parms.num_iters);

    print_indent(output, 4);
    output_report(output, "Minimum Throughput: %.2f MB/s\n", total_mm.min);
    print_indent(output, 4);
    output_report(output, "Maximum Throughput: %.2f MB/s\n", total_mm.max);
    print_indent(output, 4);
    output_report(output, "Average Throughput: %.2f MB/s\n",
                  total_mm.sum / total_mm.num);

    /* accumulate and output the max, min, and average "gross read" times */
    if (pio_debug_level == 3) {
        /* output all of the times for all iterations */
        print_indent(output, 3);
        output_report(output, "Read Open-Close details:\n");
        output_all_info(output, read_gross_mm_table, parms.num_iters, 4);
    }

    total_mm = accumulate_minmax_stuff(read_gross_mm_table, raw_size, parms.num_iters);

    print_indent(output, 3);
    output_report(output, "Read Open-Close (%d iteration(s)):\n", parms.num_iters);

    print_indent(output, 4);
    output_report(output, "Minimum Throughput: %.2f MB/s\n", total_mm.min);
    print_indent(output, 4);
    output_report(output, "Maximum Throughput: %.2f MB/s\n", total_mm.max);
    print_indent(output, 4);
    output_report(output, "Average Throughput: %.2f MB/s\n",
                  total_mm.sum / total_mm.num);

    free(write_mm_table);
    free(read_mm_table);
    pio_time_destroy(res.timers);
    return ret_value;
}

/*
 * Function:    output_all_info
 * Purpose:     
 * Return:      Nothing
 * Programmer:  Bill Wendling, 29. January 2002
 * Modifications:
 */
static void
output_all_info(FILE *output, minmax *mm, int count, int indent_level)
{
    register int i;

    for (i = 0; i < count; ++i) {
        print_indent(output, indent_level);
        output_report(output, "Iteration %d:\n", i);
        print_indent(output, indent_level + 1);
        output_report(output, "Minimum Time: %.2fs\n", mm[i].min);
        print_indent(output, indent_level + 1);
        output_report(output, "Maximum Time: %.2fs\n", mm[i].max);
    }
}

/*
 * Function:    get_minmax_stuff
 * Purpose:     Gather all the min, max and total of val.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 21. December 2001
 * Modifications:
 *    Use MPI_Allreduce to do it. -akc, 2002/01/11
 */
static void
get_minmax(minmax *mm, double val)
{
    int myrank;

    MPI_Comm_rank(pio_comm_g, &myrank);
    MPI_Comm_size(pio_comm_g, &mm->num);

    MPI_Allreduce(&val, &mm->max, 1, MPI_DOUBLE, MPI_MAX, pio_comm_g);
    MPI_Allreduce(&val, &mm->min, 1, MPI_DOUBLE, MPI_MIN, pio_comm_g);
    MPI_Allreduce(&val, &mm->sum, 1, MPI_DOUBLE, MPI_SUM, pio_comm_g);
}

/*
 * Function:    accumulate_minmax_stuff
 * Purpose:     Accumulate the minimum, maximum, and average of the times
 *              across all processes.
 * Return:      TOTAL_MM - the total of all of these.
 * Programmer:  Bill Wendling, 21. December 2001
 * Modifications:
 */
static minmax
accumulate_minmax_stuff(minmax *mm, long raw_size, int count)
{
    register int i;
    minmax total_mm;
    
    total_mm.sum = total_mm.max = total_mm.min = MB_PER_SEC(raw_size, mm[0].max);
    total_mm.num = count;

    for (i = 1; i < count; ++i) {
        double m = MB_PER_SEC(raw_size, total_mm.max);

        total_mm.sum += m;

        if (m < total_mm.min)
            total_mm.min = m;

        if (m > total_mm.max)
            total_mm.max = m;
    }

    return total_mm;
}

/*
 * Function:    create_comm_world
 * Purpose:     Create an MPI Comm world and store it in pio_comm_g, which
 *              is a global variable.
 * Return:      SUCCESS on success.
 *              FAIL otherwise.
 * Programmer:  Bill Wendling, 19. December 2001
 * Modifications:
 */
static int
create_comm_world(int num_procs, int *doing_pio)
{
    /* MPI variables */
    int     mrc, ret_value;     /* return values                */
    int     color;              /* for communicator creation    */
    int     myrank, nprocs;

    pio_comm_g = MPI_COMM_NULL;

    /*
     * Create a sub communicator for this PIO run. Easier to use the first N
     * processes.
     */
    MPI_Comm_size(MPI_COMM_WORLD, &nprocs);

    if (num_procs > nprocs) {
        fprintf(stderr,
                "number of process(%d) must be <= number of processes in MPI_COMM_WORLD(%d)\n",
                num_procs, nprocs);
        goto error_done;
    }

    MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
    color = (myrank < num_procs);
    mrc = MPI_Comm_split(MPI_COMM_WORLD, color, myrank, &pio_comm_g);

    if (mrc != MPI_SUCCESS) {
        fprintf(stderr, "MPI_Comm_split failed\n");
        goto error_done;
    }

    if (!color) {
        /* not involved in this run */
        mrc = destroy_comm_world();
        goto done;
    }

    /* determine the MPI rank in the PIO communicator */
    MPI_Comm_size(pio_comm_g, &pio_mpi_nprocs_g);
    MPI_Comm_rank(pio_comm_g, &pio_mpi_rank_g);

done:
    *doing_pio = color;
    return ret_value;

error_done:
    destroy_comm_world();
    return FAIL;
}

/*
 * Function:    destroy_comm_world
 * Purpose:     Destroy the created MPI Comm world which is stored in the
 *              pio_comm_g global variable.
 * Return:      SUCCESS on success.
 *              FAIL otherwise.
 * Programmer:  Bill Wendling, 19. December 2001
 * Modifications:
 */
static int
destroy_comm_world(void)
{
    int     mrc = SUCCESS;      /* return code      */

    /* release MPI resources */
    if (pio_comm_g != MPI_COMM_NULL)
        mrc = (MPI_Comm_free(&pio_comm_g) == MPI_SUCCESS ? SUCCESS : FAIL);

    return mrc;
}

/*
 * Function:    output_report
 * Purpose:     Print a line of the report. Only do so if I'm the 0 process.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 19. December 2001
 * Modifications:
 */
static void
output_report(FILE *output, const char *fmt, ...)
{
    int myrank;

    MPI_Comm_rank(pio_comm_g, &myrank);

    if (myrank == 0) {
        va_list ap;

        va_start(ap, fmt);
        vfprintf(output, fmt, ap);
        va_end(ap);
    }
}

/*
 * Function:    print_indent
 * Purpose:     Print spaces to indent a new line of text for pretty printing
 *              things.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 29. October 2001
 * Modifications:
 */
static void
print_indent(register FILE *output, register int indent)
{
    indent *= TAB_SPACE;

    for (; indent > 0; --indent)
        fputc(' ', output);
}

/*
 * Function:    parse_command_line
 * Purpose:     Parse the command line options and return a STRUCT OPTIONS
 *              structure which will need to be freed by the calling function.
 * Return:      Pointer to an OPTIONS structure
 * Programmer:  Bill Wendling, 31. October 2001
 * Modifications:
 */
static struct options *
parse_command_line(int argc, char *argv[])
{
    register int opt;
    struct options *cl_opts;

    cl_opts = (struct options *)malloc(sizeof(struct options));

    cl_opts->output_file = NULL;
    cl_opts->file_size = 64 * ONE_MB;
    cl_opts->io_types = 0x7;    /* bottom bits indicate default type to run */
    cl_opts->num_dsets = 1;
    cl_opts->num_files = 1;
    cl_opts->num_iters = 1;
    cl_opts->max_num_procs = 1;
    cl_opts->min_num_procs = 1;
    cl_opts->max_xfer_size = 1 * ONE_MB;
    cl_opts->min_xfer_size = 128 * ONE_KB;

    while ((opt = get_option(argc, (const char **)argv, s_opts, l_opts)) != EOF) {
        switch ((char)opt) {
#if 0
        case 'b':
            /* the future "binary" option */
            break;
#endif  /* 0 */
        case 'd':
            cl_opts->num_dsets = strtol(opt_arg, NULL, 10);
            break;
        case 'D':
            pio_debug_level = strtol(opt_arg, NULL, 10);

            if (pio_debug_level > 3)
                pio_debug_level = 3;
            else if (pio_debug_level < 0)
                pio_debug_level = 0;

            break;
        case 'f':
            cl_opts->file_size = parse_size_directive(opt_arg);
            break;
        case 'F':
            cl_opts->num_files = strtol(opt_arg, NULL, 10);
            break;
        case 'H':
            cl_opts->io_types &= ~0x7;
            cl_opts->io_types |= PIO_HDF5;
            break;
        case 'i':
            cl_opts->num_iters = strtol(opt_arg, NULL, 10);
            break;
        case 'm':
            cl_opts->io_types &= ~0x7;
            cl_opts->io_types |= PIO_MPI;
            break;
        case 'o':
            cl_opts->output_file = opt_arg;
            break;
        case 'p':
            cl_opts->min_num_procs = strtol(opt_arg, NULL, 10);
            break;
        case 'P':
            cl_opts->max_num_procs = strtol(opt_arg, NULL, 10);
            break;
        case 'r':
            cl_opts->io_types &= ~0x7;
            cl_opts->io_types |= PIO_RAW;
            break;
        case 'x':
            cl_opts->min_xfer_size = parse_size_directive(opt_arg);
            break;
        case 'X':
            cl_opts->max_xfer_size = parse_size_directive(opt_arg);
            break;
        case 'h':
        case '?':
        default:
            usage(progname);
            free(cl_opts);
            return NULL;
        }
    }

    return cl_opts;
}

/*
 * Function:    parse_size_directive
 * Purpose:     Parse the size directive passed on the commandline. The size
 *              directive is an integer followed by a size indicator:
 *
 *                  K, k - Kilobyte
 *                  M, m - Megabyte
 *                  G, g - Gigabyte
 *
 * Return:      The size as a LONG. If an unknown size indicator is used, then
 *              the program will exit with EXIT_FAILURE as the return value.
 * Programmer:  Bill Wendling, 18. December 2001
 * Modifications:
 */
static long
parse_size_directive(const char *size)
{
    long s;
    char *endptr;

    s = strtol(size, &endptr, 10);

    if (endptr && *endptr) {
        while (*endptr != '\0' && (*endptr == ' ' || *endptr == '\t'))
            ++endptr;

        switch (*endptr) {
            case 'K':
            case 'k':
                s *= ONE_KB;
                break;
            case 'M':
            case 'm':
                s *= ONE_MB;
                break;
            case 'G':
            case 'g':
                s *= ONE_GB;
                break;
            default:
                fprintf(stderr, "Illegal size specifier '%c'\n", *endptr);
                exit(EXIT_FAILURE);
        }
    }

    return s;
}

/*
 * Function:    usage
 * Purpose:     Print a usage message and then exit.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 31. October 2001
 * Modifications:
 */
static void
usage(const char *prog)
{
    int myrank;

    MPI_Comm_rank(pio_comm_g, &myrank);

    if (myrank == 0) {
        fflush(stdout);
        fprintf(stdout, "usage: %s [OPTIONS]\n", prog);
        fprintf(stdout, "  OPTIONS\n");
        fprintf(stdout, "     -h, --help                  Print a usage message and exit\n");
        fprintf(stdout, "     -d N, --num-dsets=N         Number of datasets per file [default:1]\n");
        fprintf(stdout, "     -D N, --debug=N             Indicate the debugging level [default:0]\n");
        fprintf(stdout, "     -f S, --file-size=S         Size of a single file [default: 64M]\n");
        fprintf(stdout, "     -F N, --num-files=N         Number of files [default: 1]\n");
        fprintf(stdout, "     -H, --hdf5                  Run HDF5 performance test\n");
        fprintf(stdout, "     -i, --num-iterations        Number of iterations to perform [default: 1]\n");
        fprintf(stdout, "     -m, --mpiio                 Run MPI/IO performance test\n");
        fprintf(stdout, "     -o F, --output=F            Output raw data into file F [default: none]\n");
        fprintf(stdout, "     -P N, --max-num-processes=N Maximum number of processes to use [default: 1]\n");
        fprintf(stdout, "     -p N, --min-num-processes=N Minimum number of processes to use [default: 1]\n");
        fprintf(stdout, "     -r, --raw                   Run raw (UNIX) performance test\n");
        fprintf(stdout, "     -X S, --max-xfer-size=S     Maximum transfer buffer size [default: 1M]\n");
        fprintf(stdout, "     -x S, --min-xfer-size=S     Minimum transfer buffer size [default: 128K]\n");
        fprintf(stdout, "\n");
        fprintf(stdout, "  F - is a filename.\n");
        fprintf(stdout, "  N - is an integer >=0.\n");
        fprintf(stdout, "  S - is a size specifier, an integer >=0 followed by a size indicator:\n");
        fprintf(stdout, "\n");
        fprintf(stdout, "          K - Kilobyte\n");
        fprintf(stdout, "          M - Megabyte\n");
        fprintf(stdout, "          G - Gigabyte\n");
        fprintf(stdout, "\n");
        fprintf(stdout, "      Example: 37M = 37 Megabytes\n");
        fprintf(stdout, "\n");
        fprintf(stdout, "  Debugging levels are:\n");
        fprintf(stdout, "\n");
        fprintf(stdout, "    0 - None\n");
        fprintf(stdout, "    1 - Minimal\n");
        fprintf(stdout, "    2 - Not quite everything\n");
        fprintf(stdout, "    3 - Everything\n");
        fprintf(stdout, "\n");
        fflush(stdout);
    }
}

#else /* H5_HAVE_PARALLEL */

/*
 * Function:    main
 * Purpose:     Dummy main() function for if HDF5 was configured without
 *              parallel stuff.
 * Return:      EXIT_SUCCESS
 * Programmer:  Bill Wendling, 14. November 2001
 * Modifications:
 */
int
main(void)
{
    printf("No parallel IO performance because parallel is not configured\n");
    return EXIT_SUCCESS;
}

#endif /* !H5_HAVE_PARALLEL */