codes-darshan3-io-wrkld.c 50.4 KB
Newer Older
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
/*
 * Copyright (C) 2013 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */
#include <assert.h>
#include <math.h>

#include "codes/codes-workload.h"
#include "codes/quickhash.h"

#include "darshan-logutils.h"

#define DARSHAN_NEGLIGIBLE_DELAY 0.00001

#define RANK_HASH_TABLE_SIZE 397

#define MIN(a, b) (((a) < (b)) ? (a) : (b))
#define MAX(a, b) (((a) > (b)) ? (a) : (b))

#define ALIGN_BY_8(x) ((x) + ((x) % 8))

/* structure for storing a darshan workload operation (a codes op with 2 timestamps) */
struct darshan_io_op
{
    struct codes_workload_op codes_op;
    double start_time;
    double end_time;
};

/* I/O context structure managed by each rank in the darshan workload */
struct rank_io_context
{
    int64_t my_rank;
    double last_op_time;
    void *io_op_dat;

    off_t next_off;

    struct qhash_head hash_link;
};

struct darshan_unified_record
{
    struct darshan_posix_file psx_file_rec;
    struct darshan_mpiio_file mpiio_file_rec;
    struct darshan_unified_record *next;
};

static void * darshan_io_workload_read_config(
        ConfigHandle * handle,
        char const * section_name,
        char const * annotation,
        int num_ranks);
/* Darshan workload generator's implementation of the CODES workload API */
static int darshan_psx_io_workload_load(const char *params, int app_id, int rank);
static void darshan_psx_io_workload_get_next(int app_id, int rank, struct codes_workload_op *op);
static int darshan_psx_io_workload_get_rank_cnt(const char *params, int app_id);
static int darshan_rank_hash_compare(void *key, struct qhash_head *link);

/* Darshan I/O op data structure access (insert, remove) abstraction */
static void *darshan_init_io_op_dat(void);
static void darshan_insert_next_io_op(void *io_op_dat, struct darshan_io_op *io_op);
static void darshan_remove_next_io_op(void *io_op_dat, struct darshan_io_op *io_op,
                                      double last_op_time);
static void darshan_finalize_io_op_dat(void *io_op_dat);
static int darshan_io_op_compare(const void *p1, const void *p2);

/* Helper functions for implementing the Darshan workload generator */
static void generate_psx_file_events(struct darshan_posix_file *file,
                                         struct rank_io_context *io_context);
static void generate_mpiio_file_events(
    struct darshan_mpiio_file *mfile, struct rank_io_context *io_context);
static double generate_open_event(darshan_record_id id, enum codes_workload_op_type type, int create_flag,
                                      double meta_op_time, double cur_time,
                                      struct rank_io_context *io_context, int insert_flag);
static double generate_close_event(darshan_record_id id, enum codes_workload_op_type type, double meta_op_time,
                                       double cur_time, struct rank_io_context *io_context,
                                       int insert_flag);
static double generate_psx_io_events(struct darshan_posix_file *file,
                                         double inter_io_delay, double cur_time,
                                         struct rank_io_context *io_context);
static double generate_mpiio_io_events(
    struct darshan_mpiio_file *mfile, double inter_io_delay,
    double cur_time, struct rank_io_context *io_context);
static void determine_mpiio_io_params(
    struct darshan_mpiio_file *mfile, int write_flag, size_t *io_sz, off_t *io_off,
    struct rank_io_context *io_context);
static void determine_psx_io_params(struct darshan_posix_file *file, int write_flag, size_t *io_sz,
                                    off_t *io_off, struct rank_io_context *io_context);
static void psx_calc_io_delays(struct darshan_posix_file *file, int64_t num_io_ops,
                           double total_delay, double *first_io_delay, double *close_delay,
                           double *inter_io_delay);
static void mpiio_calc_io_delays(
    struct darshan_mpiio_file *mfile, int64_t num_io_ops, double total_delay,
    double *first_io_delay, double *close_delay, double *inter_io_delay);
static void file_sanity_check(
    struct darshan_posix_file *file, struct darshan_mpiio_file *mfile,
    struct darshan_job *job, darshan_fd fd);

static int darshan_psx_io_workload_get_time(const char *params, int app_id, int rank, double *read_time, double *write_time,
												int64_t *read_bytes, int64_t *written_bytes);
/* workload method name and function pointers for the CODES workload API */
struct codes_workload_method darshan_posix_io_workload_method =
{
    .method_name = "darshan_io_workload",
    .codes_workload_read_config = darshan_io_workload_read_config,
    .codes_workload_load = darshan_psx_io_workload_load,
    .codes_workload_get_next = darshan_psx_io_workload_get_next,
    .codes_workload_get_rank_cnt = darshan_psx_io_workload_get_rank_cnt,
	.codes_workload_get_time = darshan_psx_io_workload_get_time,
};

/* posix_logutils functions */

extern struct darshan_mod_logutil_funcs posix_logutils;
static struct darshan_mod_logutil_funcs *psx_utils = &posix_logutils;
static struct darshan_mod_logutil_funcs *mpiio_utils = &mpiio_logutils;
static int total_rank_cnt = 0;

/* hash table to store per-rank workload contexts */
static struct qhash_table *rank_tbl = NULL;
static int rank_tbl_pop = 0;

static void * darshan_io_workload_read_config(
        ConfigHandle * handle,
        char const * section_name,
        char const * annotation,
        int num_ranks)
{
    darshan_params *d = malloc(sizeof(*d));
    assert(d);
    d->log_file_path[0] = '\0';
    
    /* silence warning */
    (void)num_ranks;

    int rc = configuration_get_value_relpath(handle, section_name,
            "darshan_log_file", annotation, d->log_file_path,
            MAX_NAME_LENGTH_WKLD);
    assert(rc > 0);
   
    return d;
}

static int darshan_psx_io_workload_get_time(const char *params, int app_id, int rank, double *read_time, double *write_time,
		int64_t *read_bytes, int64_t *written_bytes)
{
	struct darshan_posix_file * psx_file_rec;
	darshan_fd logfile_fd = NULL;
	psx_file_rec = (struct darshan_posix_file *)calloc(1, sizeof(struct darshan_posix_file));
	darshan_params *d_params = (darshan_params *)params;

        /* silence warning */
        (void)app_id;

	int ret;
	if (!d_params)
		return -1;

	/* open the darshan log to begin reading in file i/o info */
	//logfile_fd = darshan_log_open(d_params->log_file_path, "r");
	logfile_fd = darshan_log_open(d_params->log_file_path);
	if (!logfile_fd)
		return -1;
	while ((ret = psx_utils->log_get_record(logfile_fd, (void **)&psx_file_rec)) > 0)
	{
		 /* generate all i/o events contained in this independent file */
		if (psx_file_rec->base_rec.rank == rank)
		{
			*read_time += psx_file_rec->fcounters[POSIX_F_READ_TIME];
			*write_time += psx_file_rec->fcounters[POSIX_F_WRITE_TIME];
			*read_bytes += psx_file_rec->counters[POSIX_BYTES_READ];
			*written_bytes += psx_file_rec->counters[POSIX_BYTES_WRITTEN];
		}
	}
	darshan_log_close(logfile_fd);
	return 0;
}

/* load the workload generator for this rank, given input params */
static int darshan_psx_io_workload_load(const char *params, int app_id, int rank)
{
    darshan_params *d_params = (darshan_params *)params;
    darshan_fd logfile_fd = NULL;
    struct darshan_job job;
    //struct darshan_file next_file;
    /* open posix log file */
    struct darshan_posix_file *psx_file_rec;
    struct darshan_mpiio_file *mpiio_file_rec;
    struct rank_io_context *my_ctx;
    int ret;
    struct darshan_unified_record *dur_new = NULL;
    struct darshan_unified_record *dur_head = NULL;
    struct darshan_unified_record *dur_cur = NULL;
    struct darshan_unified_record *dur_tmp = NULL;

    psx_file_rec = (struct darshan_posix_file *) calloc(1, sizeof(struct darshan_posix_file));
    assert(psx_file_rec);
    mpiio_file_rec = (struct darshan_mpiio_file *) calloc(1, sizeof(struct darshan_mpiio_file));
    assert(mpiio_file_rec);

    APP_ID_UNSUPPORTED(app_id, "darshan")

    if (!d_params)
        return -1;

    /* open the darshan log to begin reading in file i/o info */
    //logfile_fd = darshan_log_open(d_params->log_file_path, "r");
    logfile_fd = darshan_log_open(d_params->log_file_path);
    if (!logfile_fd)
        return -1;

    /* get the per-job stats from the log */
    ret = darshan_log_get_job(logfile_fd, &job);
    if (ret < 0)
    {
        darshan_log_close(logfile_fd);
        return -1;
    }
    if (!total_rank_cnt)
    {
        total_rank_cnt = job.nprocs;
    }
    //printf("rank = %d, total_rank_cnt = %d\n", rank, total_rank_cnt);
    //printf("job: (%ld, %ld, %ld, %ld, %s)\n", job.jobid, job.uid, job.start_time, job.end_time, job.metadata);
    assert(rank < total_rank_cnt);

    /* allocate the i/o context needed by this rank */
    my_ctx = malloc(sizeof(struct rank_io_context));
    if (!my_ctx)
    {
        darshan_log_close(logfile_fd);
        return -1;
    }
    my_ctx->my_rank = (int64_t)rank;
    my_ctx->last_op_time = 0.0;
    my_ctx->io_op_dat = darshan_init_io_op_dat();
    my_ctx->next_off = 0;

    /* loop over all files contained in the log file */
    /* build a linked list to preserve order, we will make another pass over
     * mpiio records and combine them
     */
    while ((ret = psx_utils->log_get_record(logfile_fd, (void **)&psx_file_rec)) > 0)
    {
        dur_new = calloc(1, sizeof(*dur_new));
        assert(dur_new);

        dur_new->psx_file_rec = *psx_file_rec;

        if(!dur_head)
            dur_head = dur_new;
        if(dur_cur)
            dur_cur->next = dur_new;
        dur_cur = dur_new;
    }

    /* now loop over mpiio records (if present) and match them up with the
     * posix records
     */
    while ((ret = mpiio_utils->log_get_record(logfile_fd, (void **)&mpiio_file_rec)) > 0)
    {
        for(dur_cur = dur_head; dur_cur; dur_cur = dur_cur->next)
        {
            if((dur_cur->psx_file_rec.base_rec.rank ==
                mpiio_file_rec->base_rec.rank) &&
               (dur_cur->psx_file_rec.base_rec.id ==
                mpiio_file_rec->base_rec.id))
            {
                dur_cur->mpiio_file_rec = *mpiio_file_rec;
                break;
            }
274
275
276
277
278
279
280
281
282
283

            if((dur_cur->psx_file_rec.base_rec.id == mpiio_file_rec->base_rec.id)
                && (dur_cur->psx_file_rec.base_rec.rank == -1)
                && (mpiio_file_rec->base_rec.rank != -1))
            {
                fprintf(stderr, "WARNING: id %" PRIu64 " has non-shared MPI record and shared POSIX record.  Skipping POSIX record which may have been generated by stat() calls.\n", mpiio_file_rec->base_rec.id);

                dur_cur->psx_file_rec.counters[POSIX_OPENS] = 0;
            }
                
284
        }
285
        
Philip Carns's avatar
bug fix    
Philip Carns committed
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
        if(!dur_cur)
        {
            /* if we fall through to here, that means that an mpiio record is present
             * for which there is no exact match in the posix records.  This
             * could (for example) happen if mpiio was using deferred opens,
             * producing a shared record in mpi and unique records in posix.  Or
             * if mpiio is using a non-posix back end. Or if we skip the posix
             * records because the app issued a stat() on every rank but only
             * did I/O on a subset.
             */
            dur_new = calloc(1, sizeof(*dur_new));
            assert(dur_new);

            dur_new->mpiio_file_rec = *mpiio_file_rec;
            dur_new->next = dur_head;
            dur_head = dur_new;
        }
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
    }

    /* file records have all been retrieved from darshan log.  Now we loop
     * through them and generate workload events from each record
     */
    for(dur_cur = dur_head; dur_cur; dur_cur = dur_cur->next)
    {
        /* skip the file and emit a warning if it is RW */
        if(dur_cur->psx_file_rec.counters[POSIX_BYTES_READ] &&
            dur_cur->psx_file_rec.counters[POSIX_BYTES_WRITTEN])
        {
            if(rank == 0)
            {
                printf("WARNING: skipping R/W file record %lu with %ld bytes read and %ld bytes written\n", dur_cur->psx_file_rec.base_rec.id,
                    dur_cur->psx_file_rec.counters[POSIX_BYTES_READ],
                    dur_cur->psx_file_rec.counters[POSIX_BYTES_WRITTEN]);
            }
            continue;
        }

        /* MPI-IO */
        if(dur_cur->mpiio_file_rec.counters[MPIIO_COLL_OPENS] ||
            dur_cur->mpiio_file_rec.counters[MPIIO_INDEP_OPENS])
        {
            /* don't parse unless this file record belongs to this rank, or
             * this is a globally shared file record.
             */
            if(dur_cur->psx_file_rec.base_rec.rank == rank ||
                dur_cur->psx_file_rec.base_rec.rank == -1)
            {
                /* make sure the file i/o counters are valid */
                file_sanity_check(&dur_cur->psx_file_rec, 
                    &dur_cur->mpiio_file_rec, &job, logfile_fd);

                /* generate i/o events and store them in this rank's 
                 * workload context 
                 */
                generate_mpiio_file_events(&dur_cur->mpiio_file_rec, my_ctx);
            }
        }
        /* POSIX */
        else if(dur_cur->psx_file_rec.counters[POSIX_OPENS])
        {
            /* don't parse unless this file record belongs to this rank, or
             * this is a globally shared file record.
             */
            if(dur_cur->psx_file_rec.base_rec.rank == rank ||
                dur_cur->psx_file_rec.base_rec.rank == -1)
            {
                /* make sure the file i/o counters are valid */
                file_sanity_check(&dur_cur->psx_file_rec, 
                    &dur_cur->mpiio_file_rec, &job, logfile_fd);

                /* generate i/o events and store them in this rank's 
                 * workload context 
                 */
                generate_psx_file_events(&dur_cur->psx_file_rec, my_ctx);
            }
        }
        else
        {
            /* no I/O here that we can generate events for; continue */
            continue;
        }
    }
    if (ret < 0)
        return -1;

    darshan_log_close(logfile_fd);

    /* finalize the rank's i/o context so i/o ops may be retrieved later (in order) */
    darshan_finalize_io_op_dat(my_ctx->io_op_dat);
    if (!rank_tbl)
    {
        rank_tbl = qhash_init(darshan_rank_hash_compare, quickhash_64bit_hash, RANK_HASH_TABLE_SIZE);
        if (!rank_tbl)
            return -1;
    }

    /* add this rank context to the hash table */
    qhash_add(rank_tbl, &(my_ctx->my_rank), &(my_ctx->hash_link));
    rank_tbl_pop++;

    /* free linked list */
    for(dur_cur = dur_head; dur_cur;)
    {
        dur_tmp = dur_cur;
        dur_cur = dur_cur->next;
        free(dur_tmp);
    }
    free(psx_file_rec);
    free(mpiio_file_rec);
    return 0;
}

/* pull the next event (independent or collective) for this rank from its event context */
static void darshan_psx_io_workload_get_next(int app_id, int rank, struct codes_workload_op *op)
{
    int64_t my_rank = (int64_t)rank;
    struct qhash_head *hash_link = NULL;
    struct rank_io_context *tmp = NULL;
    struct darshan_io_op next_io_op;

    assert(rank < total_rank_cnt);

    /* find i/o context for this rank in the rank hash table */
    hash_link = qhash_search(rank_tbl, &my_rank);
    /* terminate the workload if there is no valid rank context */
    if (!hash_link)
    {
        op->op_type = CODES_WK_END;
        return;
    }

    /* get access to the rank's io_context data */
    tmp = qhash_entry(hash_link, struct rank_io_context, hash_link);
    assert(tmp->my_rank == my_rank);

    /* get the next darshan i/o op out of this rank's context */
    darshan_remove_next_io_op(tmp->io_op_dat, &next_io_op, tmp->last_op_time);

    /* free the rank's i/o context if this is the last i/o op */
    if (next_io_op.codes_op.op_type == CODES_WK_END)
    {
        qhash_del(hash_link);
        free(tmp);
        rank_tbl_pop--;
        if (!rank_tbl_pop)
        {
            qhash_finalize(rank_tbl);
            rank_tbl = NULL;
        }
    }
    else
    {
        /* else, set the last op time to be the end of the returned op */
        tmp->last_op_time = next_io_op.end_time;
    }

    /* return the codes op contained in the darshan i/o op */
    *op = next_io_op.codes_op;

    return;
}

static int darshan_psx_io_workload_get_rank_cnt(const char *params, int app_id)
{
    darshan_params *d_params = (darshan_params *)params;
    darshan_fd logfile_fd = NULL;
    struct darshan_job job;
    int ret;

    /* silence warning */
    (void)app_id;

    if (!d_params)
        return -1;
    //printf("opening log ... \n");
    /* open the darshan log to begin reading in file i/o info */
    logfile_fd = darshan_log_open(d_params->log_file_path);
    if (!logfile_fd)
        return -1;
    //printf("log open done, start getting job \n");
    /* get the per-job stats from the log */
    ret = darshan_log_get_job(logfile_fd, &job);
    if (ret < 0)
    {
        darshan_log_close(logfile_fd);
        return -1;
    }
    //printf("get_job done\n");
    darshan_log_close(logfile_fd);

    return job.nprocs;
}

/* comparison function for comparing two hash keys (used for storing multiple io contexts) */
static int darshan_rank_hash_compare(
    void *key, struct qhash_head *link)
{
    int64_t *in_rank = (int64_t *)key;
    struct rank_io_context *tmp;

    tmp = qhash_entry(link, struct rank_io_context, hash_link);
    if (tmp->my_rank == *in_rank)
        return 1;

    return 0;
}

/*****************************************/
/*                                       */
/*   Darshan I/O op storage abstraction  */
/*                                       */
/*****************************************/

#define DARSHAN_IO_OP_INC_CNT 100000

/* dynamically allocated array data structure for storing darshan i/o events */
struct darshan_io_dat_array
{
    struct darshan_io_op *op_array;
    int64_t op_arr_ndx;
    int64_t op_arr_cnt;
};

/* initialize the dynamic array data structure */
static void *darshan_init_io_op_dat()
{
    struct darshan_io_dat_array *tmp;

    /* initialize the array data structure */
    tmp = malloc(sizeof(struct darshan_io_dat_array));
    assert(tmp);
    tmp->op_array = malloc(DARSHAN_IO_OP_INC_CNT * sizeof(struct darshan_io_op));
    assert(tmp->op_array);
    tmp->op_arr_ndx = 0;
    tmp->op_arr_cnt = DARSHAN_IO_OP_INC_CNT;

    /* return the array info for this rank's i/o context */
    return (void *)tmp;
}

/* store the i/o event in this rank's i/o context */
static void darshan_insert_next_io_op(
    void *io_op_dat, struct darshan_io_op *io_op)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;
    struct darshan_io_op *tmp;

    assert(io_op->start_time >= 0);

    /* realloc array if it is already full */
    if (array->op_arr_ndx == array->op_arr_cnt)
    {
        tmp = malloc((array->op_arr_cnt + DARSHAN_IO_OP_INC_CNT) * sizeof(struct darshan_io_op));
        assert(tmp);
        memcpy(tmp, array->op_array, array->op_arr_cnt * sizeof(struct darshan_io_op));
        free(array->op_array);
        array->op_array = tmp;
        array->op_arr_cnt += DARSHAN_IO_OP_INC_CNT;
    }

    /* add the darshan i/o op to the array */
    array->op_array[array->op_arr_ndx++] = *io_op;

    return;
}

/* pull the next i/o event out of this rank's i/o context */
static void darshan_remove_next_io_op(
    void *io_op_dat, struct darshan_io_op *io_op, double last_op_time)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;

    /* if the array has been scanned completely already */
    if (array->op_arr_ndx == array->op_arr_cnt)
    {
        /* no more events just end the workload */
        io_op->codes_op.op_type = CODES_WK_END;
    }
    else
    {
        struct darshan_io_op *tmp = &(array->op_array[array->op_arr_ndx]);

        if ((tmp->start_time - last_op_time) <= DARSHAN_NEGLIGIBLE_DELAY)
        {
            /* there is no delay, just return the next op in the array */
            *io_op = *tmp;
            array->op_arr_ndx++;
        }
        else
        {
            /* there is a nonnegligible delay, so generate and return a delay event */
            io_op->codes_op.op_type = CODES_WK_DELAY;
            io_op->codes_op.u.delay.seconds = tmp->start_time - last_op_time;
            io_op->start_time = last_op_time;
            io_op->end_time = tmp->start_time;
        }
    }

    /* if this is the end op, free data structures */
    if (io_op->codes_op.op_type == CODES_WK_END)
    {
        free(array->op_array);
        free(array);
    }

    return;
}

/* sort the dynamic array in order of i/o op start time */
static void darshan_finalize_io_op_dat(
    void *io_op_dat)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;

    /* sort this rank's i/o op list */
    qsort(array->op_array, array->op_arr_ndx, sizeof(struct darshan_io_op), darshan_io_op_compare);
    array->op_arr_cnt = array->op_arr_ndx;
    array->op_arr_ndx = 0;

    return;
}

/* comparison function for sorting darshan_io_ops in order of start timestamps */
static int darshan_io_op_compare(
    const void *p1, const void *p2)
{
    struct darshan_io_op *a = (struct darshan_io_op *)p1;
    struct darshan_io_op *b = (struct darshan_io_op *)p2;
    
    assert(a->start_time != b->start_time);

    if (a->start_time < b->start_time)
        return -1;
    else if (a->start_time > b->start_time)
        return 1;
    else
        return 0;
}

/*****************************************/
/*                                       */
/* Darshan workload generation functions */
/*                                       */
/*****************************************/

static void generate_mpiio_file_events(
    struct darshan_mpiio_file *mfile, struct rank_io_context *io_context)
{
    double cur_time = mfile->fcounters[MPIIO_F_OPEN_TIMESTAMP];
    double total_delay;
    double first_io_delay = 0.0;
    double close_delay = 0.0;
    double inter_io_delay = 0.0;
    int num_io_ops = 0;
    double meta_op_time;
    int create_flag = 0;
    enum codes_workload_op_type open_type;

    /* determine delay available between first open and last close */
    if(mfile->base_rec.rank == -1)
    {
        /* shared file */
        total_delay = mfile->fcounters[MPIIO_F_CLOSE_TIMESTAMP] - mfile->fcounters[MPIIO_F_OPEN_TIMESTAMP] -
                      ((mfile->fcounters[MPIIO_F_READ_TIME] + mfile->fcounters[MPIIO_F_WRITE_TIME] +
                      mfile->fcounters[MPIIO_F_META_TIME])/total_rank_cnt);
    }
    else
    {
        /* uniq file */
        total_delay = mfile->fcounters[MPIIO_F_CLOSE_TIMESTAMP] - mfile->fcounters[MPIIO_F_OPEN_TIMESTAMP] -
                      mfile->fcounters[MPIIO_F_READ_TIME] - mfile->fcounters[MPIIO_F_WRITE_TIME] -
                      mfile->fcounters[MPIIO_F_META_TIME];
    }
    if(total_delay < 0)
        total_delay = 0;

    /* how many io operations on this file per rank (rounded up) ? */
    int all_io_ops = mfile->counters[MPIIO_INDEP_READS] + mfile->counters[MPIIO_INDEP_WRITES] + mfile->counters[MPIIO_COLL_READS] + mfile->counters[MPIIO_COLL_WRITES];
    if(mfile->base_rec.rank == -1)
    {
        num_io_ops = all_io_ops / total_rank_cnt;
        if(all_io_ops % total_rank_cnt)
            num_io_ops++;
    }
    else
        num_io_ops = all_io_ops;

    /* calculate synthetic delay values */
    mpiio_calc_io_delays(mfile, num_io_ops, total_delay,
        &first_io_delay, &close_delay, &inter_io_delay);

    /* calculate average meta op time, divide among open and close for now */
    meta_op_time = mfile->fcounters[MPIIO_F_META_TIME] / 2.0;
    if(mfile->base_rec.rank == -1)
        meta_op_time /= total_rank_cnt;

    /* set the create flag if the file was written to */
    if (mfile->counters[MPIIO_BYTES_WRITTEN])
    {
        create_flag = 1;
    }

    /* generate an open event */
    if(mfile->counters[MPIIO_COLL_OPENS] > 0)
        open_type = CODES_WK_MPI_COLL_OPEN;
    else
        open_type = CODES_WK_MPI_OPEN;
    cur_time = generate_open_event(mfile->base_rec.id, open_type, create_flag, meta_op_time, 
        cur_time, io_context, 1);

    /* account for potential delay from first open to first io */
    cur_time += first_io_delay;

    /* perform the calculated number of i/o operations for this file open */
    cur_time = generate_mpiio_io_events(mfile, inter_io_delay,
                                          cur_time, io_context);
    
    /* account for potential delay from last io to close */
    cur_time += close_delay;

    /* generate a close for the open event at the start of the loop */
    cur_time = generate_close_event(mfile->base_rec.id, CODES_WK_MPI_CLOSE, meta_op_time, cur_time, 
        io_context, 1);

    return;
}

/* generate events for a POSIX file */
static void generate_psx_file_events(
    struct darshan_posix_file *file, struct rank_io_context *io_context)
{
    double cur_time = file->fcounters[POSIX_F_OPEN_START_TIMESTAMP];
    double total_delay;
    double first_io_delay = 0.0;
    double close_delay = 0.0;
    double inter_io_delay = 0.0;
    int num_io_ops = 0;
    double meta_op_time;
    int create_flag = 0;

    /* determine delay available between first open and last close */
    if(file->base_rec.rank == -1)
    {
        /* shared file */
        total_delay = file->fcounters[POSIX_F_CLOSE_END_TIMESTAMP] - file->fcounters[POSIX_F_OPEN_START_TIMESTAMP] -
                      ((file->fcounters[POSIX_F_READ_TIME] + file->fcounters[POSIX_F_WRITE_TIME] +
                      file->fcounters[POSIX_F_META_TIME])/total_rank_cnt);
    }
    else
    {
        /* uniq file */
        total_delay = file->fcounters[POSIX_F_CLOSE_END_TIMESTAMP] - file->fcounters[POSIX_F_OPEN_START_TIMESTAMP] -
                      file->fcounters[POSIX_F_READ_TIME] - file->fcounters[POSIX_F_WRITE_TIME] -
                      file->fcounters[POSIX_F_META_TIME];
    }
    if(total_delay < 0)
        total_delay = 0;

    /* how many io operations on this file per rank (rounded up) ? */
    if(file->base_rec.rank == -1)
    {
        num_io_ops = (file->counters[POSIX_READS] + file->counters[POSIX_WRITES]) / total_rank_cnt;
        if((file->counters[POSIX_READS] + file->counters[POSIX_WRITES]) % total_rank_cnt)
            num_io_ops++;
    }
    else
        num_io_ops = file->counters[POSIX_READS] + file->counters[POSIX_WRITES];

    /* calculate synthetic delay values */
    psx_calc_io_delays(file, num_io_ops, total_delay,
        &first_io_delay, &close_delay, &inter_io_delay);

    assert(first_io_delay >= 0);
    assert(close_delay >= 0);
    assert(meta_op_time >= 0);
    assert(inter_io_delay >= 0);

    /* calculate average meta op time, divide among open and close for now */
    meta_op_time = file->fcounters[POSIX_F_META_TIME] / 2.0;
    if(file->base_rec.rank == -1)
        meta_op_time /= total_rank_cnt;

768
769
    if(meta_op_time < 0)
        meta_op_time = DARSHAN_NEGLIGIBLE_DELAY;
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

    /* set the create flag if the file was written to */
    if (file->counters[POSIX_BYTES_WRITTEN])
    {
        create_flag = 1;
    }

    /* generate an open event */
    cur_time = generate_open_event(file->base_rec.id, CODES_WK_OPEN, create_flag, meta_op_time, 
        cur_time, io_context, 1);

    /* account for potential delay from first open to first io */
    cur_time += first_io_delay;

    /* perform the calculated number of i/o operations for this file open */
    cur_time = generate_psx_io_events(file, inter_io_delay,
                                          cur_time, io_context);

    /* account for potential delay from last io to close */
    cur_time += close_delay;

    /* generate a close for the open event at the start of the loop */
    cur_time = generate_close_event(file->base_rec.id, CODES_WK_CLOSE, meta_op_time, cur_time, 
        io_context, 1);

    return;
}

/* fill in an open event structure and store it with the rank context */
static double generate_open_event(
    darshan_record_id id, enum codes_workload_op_type type, int create_flag, double meta_op_time,
    double cur_time, struct rank_io_context *io_context, int insert_flag)
{
    struct darshan_io_op next_io_op = 
    {
        .codes_op.op_type = type,
        .codes_op.u.open.file_id = id,
        .codes_op.u.open.create_flag = create_flag,
        .start_time = cur_time
    };
    //printf("OPEN event\n");
    /* set the end time of the event based on time spent in POSIX meta operations */
    cur_time += meta_op_time;
    next_io_op.end_time = cur_time;

    /* store the open event (if this rank performed it) */
    if (insert_flag)
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);

    return cur_time;
}

/* fill in a close event structure and store it with the rank context */
static double generate_close_event(
    darshan_record_id id, enum codes_workload_op_type type, double meta_op_time, double cur_time,
    struct rank_io_context *io_context, int insert_flag)
{
    struct darshan_io_op next_io_op =
    {
        .codes_op.op_type = type,
        .codes_op.u.close.file_id = id,
        .start_time = cur_time
    };

    /* set the end time of the event based on time spent in POSIX meta operations */
    cur_time += meta_op_time;
    next_io_op.end_time = cur_time;

    //printf("CLOSE event\n");
    /* store the close event (if this rank performed it) */
    if (insert_flag)
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);

    io_context->next_off = 0;

    return cur_time;
}

/* generate all i/o events for one file open and store them with the rank context */
static double generate_mpiio_io_events(
    struct darshan_mpiio_file *mfile, double inter_io_delay,
    double cur_time, struct rank_io_context *io_context)
{
    static double rd_bw = 0.0, wr_bw = 0.0;
    double io_op_time;
    size_t io_sz;
    off_t io_off;
    int64_t i;
    struct darshan_io_op next_io_op;

    /* initialize the rd and wr bandwidth values using total io size and time */
    if (mfile->fcounters[MPIIO_F_READ_TIME])
        rd_bw = mfile->counters[MPIIO_BYTES_READ] / mfile->fcounters[MPIIO_F_READ_TIME];
    if (mfile->fcounters[MPIIO_F_WRITE_TIME])
        wr_bw = mfile->counters[MPIIO_BYTES_WRITTEN] / mfile->fcounters[MPIIO_F_WRITE_TIME]; 

    /* generator only understands if all ranks participate in collective for
     * now
     */
    assert(mfile->counters[MPIIO_COLL_WRITES]%total_rank_cnt == 0);
    assert(mfile->counters[MPIIO_COLL_READS]%total_rank_cnt == 0);

    /* note: go through all writes even if this is a shared file so that we
     * can correctly track offsets and sizes in aggregate.  We'll just emit
     * events for this rank.
     */
    /* loop to generate all writes */
    for (i = 0; i < (mfile->counters[MPIIO_COLL_WRITES] + mfile->counters[MPIIO_INDEP_WRITES]); i++)
    {
        /* calculate what value to use for i/o size and offset */
        determine_mpiio_io_params(mfile, 1, &io_sz, &io_off, io_context);

882
883
        assert(io_sz > 0);

884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
        if(mfile->base_rec.rank == io_context->my_rank ||
            (mfile->base_rec.rank == -1 &&
            i%total_rank_cnt == io_context->my_rank))
        {
            /* generate a write event */
            /* first do collectives, then non-collectives */
            if(i<mfile->counters[MPIIO_COLL_WRITES])
                next_io_op.codes_op.op_type = CODES_WK_MPI_COLL_WRITE;
            else
                next_io_op.codes_op.op_type = CODES_WK_MPI_WRITE;
            next_io_op.codes_op.u.write.file_id = mfile->base_rec.id;
            next_io_op.codes_op.u.write.size = io_sz;
            next_io_op.codes_op.u.write.offset = io_off;
            next_io_op.start_time = cur_time;
            next_io_op.codes_op.start_time = cur_time;

            /* set the end time based on observed bandwidth and io size */
            if (wr_bw == 0.0)
902
                io_op_time = DARSHAN_NEGLIGIBLE_DELAY;
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
            else
                io_op_time = (io_sz / wr_bw);

            /* update time */
            cur_time += io_op_time;
            next_io_op.end_time = cur_time;
            next_io_op.codes_op.end_time = cur_time;

            /* store the i/o event */
            darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);

            /* update current time to account for possible delay between i/o operations */
            cur_time += inter_io_delay;
        }
    }

    /* loop to generate all reads */
    for (i = 0; i < (mfile->counters[MPIIO_COLL_READS] + mfile->counters[MPIIO_INDEP_READS]); i++)
    {
        /* calculate what value to use for i/o size and offset */
        determine_mpiio_io_params(mfile, 0, &io_sz, &io_off, io_context);
924
        assert(io_sz > 0);
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942

        if(mfile->base_rec.rank == io_context->my_rank ||
            (mfile->base_rec.rank == -1 &&
            i%total_rank_cnt == io_context->my_rank))
        {
            /* generate a read event */
            if(i<mfile->counters[MPIIO_COLL_READS])
                next_io_op.codes_op.op_type = CODES_WK_MPI_COLL_READ;
            else
                next_io_op.codes_op.op_type = CODES_WK_MPI_READ;
            next_io_op.codes_op.u.read.file_id = mfile->base_rec.id;
            next_io_op.codes_op.u.read.size = io_sz;
            next_io_op.codes_op.u.read.offset = io_off;
            next_io_op.start_time = cur_time;
            next_io_op.codes_op.start_time = cur_time;

            /* set the end time based on observed bandwidth and io size */
            if (rd_bw == 0.0)
943
                io_op_time = DARSHAN_NEGLIGIBLE_DELAY;
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
            else
                io_op_time = (io_sz / rd_bw);

            /* update time */
            cur_time += io_op_time;
            next_io_op.end_time = cur_time;
            next_io_op.codes_op.end_time = cur_time;

            /* store the i/o event */
            darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);

            /* update current time to account for possible delay between i/o operations */
            cur_time += inter_io_delay;
        }
    }

    /* the last op should not have incremented cur_time; we'll handle as a
     * close delay
     */
    cur_time -= inter_io_delay;

    return cur_time;
}


/* generate all i/o events for one file open and store them with the rank context */
static double generate_psx_io_events(
    struct darshan_posix_file *file, double inter_io_delay,
    double cur_time, struct rank_io_context *io_context)
{
    static double rd_bw = 0.0, wr_bw = 0.0;
    double io_op_time;
    size_t io_sz;
    off_t io_off;
    int64_t i;
    struct darshan_io_op next_io_op;

    /* initialize the rd and wr bandwidth values using total io size and time */
    if (file->fcounters[POSIX_F_READ_TIME])
        rd_bw = file->counters[POSIX_BYTES_READ] / file->fcounters[POSIX_F_READ_TIME];
    if (file->fcounters[POSIX_F_WRITE_TIME])
        wr_bw = file->counters[POSIX_BYTES_WRITTEN] / file->fcounters[POSIX_F_WRITE_TIME]; 

    /* note: go through all writes even if this is a shared file so that we
     * can correctly track offsets and sizes in aggregate.  We'll just emit
     * events for this rank.
     */
    /* loop to generate all writes */
    for (i = 0; i < file->counters[POSIX_WRITES]; i++)
    {
        /* calculate what value to use for i/o size and offset */
        determine_psx_io_params(file, 1, &io_sz, &io_off, io_context);

        if(file->base_rec.rank == io_context->my_rank ||
            (file->base_rec.rank == -1 &&
            i%total_rank_cnt == io_context->my_rank))
        {
For faster browsing, not all history is shown. View entire blame