darshan-parser.c 29.4 KB
Newer Older
1 2 3 4 5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6 7 8 9 10 11 12 13
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <zlib.h>
#include <time.h>
14
#include <stdlib.h>
15
#include <getopt.h>
16
#include <assert.h>
17

18
#include "darshan-logutils.h"
19

20
#include "uthash-1.9.2/src/uthash.h"
21 22 23 24

/*
 * Options
 */
Philip Carns's avatar
Philip Carns committed
25 26 27 28
#define OPTION_BASE  (1 << 0)  /* darshan log fields */
#define OPTION_TOTAL (1 << 1)  /* aggregated fields */
#define OPTION_PERF  (1 << 2)  /* derived performance */
#define OPTION_FILE  (1 << 3)  /* file count totals */
29 30 31 32
#define OPTION_ALL (\
  OPTION_BASE|\
  OPTION_TOTAL|\
  OPTION_PERF|\
33
  OPTION_FILE)
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

#define FILETYPE_SHARED (1 << 0)
#define FILETYPE_UNIQUE (1 << 1)
#define FILETYPE_PARTSHARED (1 << 2)

#define max(a,b) (((a) > (b)) ? (a) : (b))
#define max3(a,b,c) (((a) > (b)) ? (((a) > (c)) ? (a) : (c)) : (((b) > (c)) ? (b) : (c)))

/*
 * Datatypes
 */
typedef struct hash_entry_s
{
    UT_hash_handle hlink;
    int64_t hash;
    int64_t type;
    int64_t procs;
    int64_t counters[CP_NUM_INDICES];
    double  fcounters[CP_F_NUM_INDICES];
    double cumul_time;
    double meta_time;
} hash_entry_t;

typedef struct perf_data_s
{
    int64_t total_bytes;
    double slowest_rank_time;
    double slowest_rank_meta_time;
    double shared_time_by_cumul;
    double shared_time_by_open;
    double shared_time_by_open_lastio;
    double shared_time_by_slowest;
    double shared_meta_time;
    double agg_perf_by_cumul;
    double agg_perf_by_open;
    double agg_perf_by_open_lastio;
    double agg_perf_by_slowest;
71 72
    double *rank_cumul_io_time;
    double *rank_cumul_md_time;
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
} perf_data_t;

typedef struct file_data_s
{
    int64_t total;
    int64_t total_size;
    int64_t total_max;
    int64_t read_only;
    int64_t read_only_size;
    int64_t read_only_max;
    int64_t write_only;
    int64_t write_only_size;
    int64_t write_only_max;
    int64_t read_write;
    int64_t read_write_size;
    int64_t read_write_max;
    int64_t unique;
    int64_t unique_size;
    int64_t unique_max;
    int64_t shared;
    int64_t shared_size;
    int64_t shared_max;
} file_data_t;

/*
 * Prototypes
 */
void accum_perf(struct darshan_file *, hash_entry_t *, perf_data_t *);
void calc_perf(struct darshan_job *, hash_entry_t *, perf_data_t *);

void accum_file(struct darshan_file *, hash_entry_t *, file_data_t *);
void calc_file(struct darshan_job *, hash_entry_t *, file_data_t *);

int usage (char *exename)
{
    fprintf(stderr, "Usage: %s [options] <filename>\n", exename);
    fprintf(stderr, "    --all   : all sub-options are enabled\n");
    fprintf(stderr, "    --base  : darshan log field data [default]\n");
    fprintf(stderr, "    --file  : total file counts\n");
    fprintf(stderr, "    --perf  : derived perf data\n");
    fprintf(stderr, "    --total : aggregated darshan field data\n");

    exit(1);
}

int parse_args (int argc, char **argv, char **filename)
{
    int index;
    int mask;
    static struct option long_opts[] =
    {
        {"all",   0, NULL, OPTION_ALL},
        {"base",  0, NULL, OPTION_BASE},
        {"file",  0, NULL, OPTION_FILE},
        {"perf",  0, NULL, OPTION_PERF},
        {"total", 0, NULL, OPTION_TOTAL},
129 130
        {"help",  0, NULL, 0},
        {0, 0, 0, 0}
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
    };

    mask = 0;

    while(1)
    {
        int c = getopt_long(argc, argv, "", long_opts, &index);

        if (c == -1) break;

        switch(c)
        {
            case OPTION_ALL:
            case OPTION_BASE:
            case OPTION_FILE:
            case OPTION_PERF:
            case OPTION_TOTAL:
                mask |= c;
                break;
            case 0:
151
            case '?':
152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
            default:
                usage(argv[0]);
                break;
        }
    }

    if (optind < argc)
    {
        *filename = argv[optind];
    }
    else
    {
        usage(argv[0]);
    }

    /* default mask value if none specified */
    if (mask == 0)
    {
        mask = OPTION_BASE;
    }

    return mask;
}

176 177 178
int main(int argc, char **argv)
{
    int ret;
179 180
    int mask;
    char *filename;
181 182 183 184
    struct darshan_job job;
    struct darshan_file cp_file;
    char tmp_string[1024];
    time_t tmp_time = 0;
185
    darshan_fd file;
186
    int i;
187
    int mount_count;
188
    int64_t* devs;
189 190
    char** mnt_pts;
    char** fs_types;
191
    int last_rank = 0;
192 193
    char *token;
    char *save;
194
    char buffer[DARSHAN_JOB_METADATA_LEN];
195

196 197 198 199 200 201 202 203 204 205 206
    hash_entry_t *file_hash = NULL;
    hash_entry_t *curr = NULL;
    hash_entry_t *tmp = NULL;
    hash_entry_t total;
    perf_data_t pdata;
    file_data_t fdata;

    memset(&pdata, 0, sizeof(pdata));
    memset(&total, 0, sizeof(total));

    mask = parse_args(argc, argv, &filename);
207

208
    file = darshan_log_open(filename, "r");
209 210
    if(!file)
    {
211
        fprintf(stderr, "darshan_log_open() failed to open %s\n.", filename);
212 213 214 215
        return(-1);
    }
   
    /* read job info */
216
    ret = darshan_log_getjob(file, &job);
217
    if(ret < 0)
218
    {
219
        fprintf(stderr, "Error: unable to read job information from log file.\n");
220
        darshan_log_close(file);
221 222 223
        return(-1);
    }

224 225 226
    /* warn user about any missing information in this log format */
    darshan_log_print_version_warnings(&job);

227
    ret = darshan_log_getexe(file, tmp_string);
228
    if(ret < 0)
229
    {
230
        fprintf(stderr, "Error: unable to read trailing job information.\n");
231
        darshan_log_close(file);
232 233 234
        return(-1);
    }

235
    /* print job summary */
236
    printf("# darshan log version: %s\n", job.version_string);
237 238
    printf("# size of file statistics: %zu bytes\n", sizeof(cp_file));
    printf("# size of job statistics: %zu bytes\n", sizeof(job));
239
    printf("# exe: %s\n", tmp_string);
Philip Carns's avatar
Philip Carns committed
240 241 242
    printf("# uid: %" PRId64 "\n", job.uid);
    printf("# jobid: %" PRId64 "\n", job.jobid);
    printf("# start_time: %" PRId64 "\n", job.start_time);
243
    tmp_time += job.start_time;
244
    printf("# start_time_asci: %s", ctime(&tmp_time));
Philip Carns's avatar
Philip Carns committed
245
    printf("# end_time: %" PRId64 "\n", job.end_time);
246 247
    tmp_time = 0;
    tmp_time += job.end_time;
248
    printf("# end_time_asci: %s", ctime(&tmp_time));
Philip Carns's avatar
Philip Carns committed
249 250
    printf("# nprocs: %" PRId64 "\n", job.nprocs);
    printf("# run time: %" PRId64 "\n", job.end_time - job.start_time + 1);
251 252 253 254 255
    for(token=strtok_r(job.metadata, "\n", &save);
        token != NULL;
        token=strtok_r(NULL, "\n", &save))
    {
        char *key;
256 257 258 259 260 261 262 263 264 265 266 267 268 269
        char *value;
        /* NOTE: we intentionally only split on the first = character.
         * There may be additional = characters in the value portion
         * (for example, when storing mpi-io hints).
         */
        strcpy(buffer, token);
        key = buffer;
        value = index(buffer, '=');
        if(!value)
            continue;
        /* convert = to a null terminator to split key and value */
        value[0] = '\0';
        value++;
        printf("# metadata: %s = %s\n", key, value);
270
    }
271 272
 
    /* print table of mounted file systems */
273
    ret = darshan_log_getmounts(file, &devs, &mnt_pts, &fs_types, &mount_count);
274
    printf("\n# mounted file systems (device, mount point, and fs type)\n");
275 276 277
    printf("# -------------------------------------------------------\n");
    for(i=0; i<mount_count; i++)
    {
Philip Carns's avatar
Philip Carns committed
278
        printf("# mount entry: %" PRId64 "\t%s\t%s\n", devs[i], mnt_pts[i], fs_types[i]);
279 280
    }
  
281 282 283 284 285 286 287 288 289
    /* try to retrieve first record (may not exist) */
    ret = darshan_log_getfile(file, &job, &cp_file);
    if(ret < 0)
    {
        fprintf(stderr, "Error: failed to parse log file.\n");
        fflush(stderr);
        return(-1);
    }
    if(ret == 0)
290 291 292
    {
        /* it looks like the app didn't open any files */
        printf("# no files opened.\n");
293
        darshan_log_close(file);
294 295 296
        return(0);
    }

297 298 299 300 301 302 303 304 305
    if ((mask & OPTION_BASE))
    {
        printf("\n# description of columns:\n");
        printf("#   <rank>: MPI rank.  -1 indicates that the file is shared\n");
        printf("#      across all processes and statistics are aggregated.\n");
        printf("#   <file>: hash of file path.  0 indicates that statistics\n");
        printf("#      are condensed to refer to all files opened at the given\n");
        printf("#      process.\n");
        printf("#   <counter> and <value>: statistical counters.\n");
306 307
        printf("#      A value of -1 indicates that Darshan could not monitor\n");
        printf("#      that counter, and its value should be ignored.\n");
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
        printf("#   <name suffix>: last %d characters of file name.\n", CP_NAME_SUFFIX_LEN);
        printf("#   <mount pt>: mount point that the file resides on.\n");
        printf("#   <fs type>: type of file system that the file resides on.\n");
        printf("\n# description of counters:\n");
        printf("#   CP_POSIX_*: posix operation counts.\n");
        printf("#   CP_COLL_*: MPI collective operation counts.\n");
        printf("#   CP_INDEP_*: MPI independent operation counts.\n");
        printf("#   CP_SPIT_*: MPI split collective operation counts.\n");
        printf("#   CP_NB_*: MPI non blocking operation counts.\n");
        printf("#   READS,WRITES,OPENS,SEEKS,STATS, and MMAPS are types of operations.\n");
        printf("#   CP_*_NC_OPENS: number of indep. and collective pnetcdf opens.\n");
        printf("#   CP_HDF5_OPENS: number of hdf5 opens.\n");
        printf("#   CP_COMBINER_*: combiner counts for MPI mem and file datatypes.\n");
        printf("#   CP_HINTS: number of times MPI hints were used.\n");
        printf("#   CP_VIEWS: number of times MPI file views were used.\n");
        printf("#   CP_MODE: mode that file was opened in.\n");
        printf("#   CP_BYTES_*: total bytes read and written.\n");
        printf("#   CP_MAX_BYTE_*: highest offset byte read and written.\n");
        printf("#   CP_CONSEC_*: number of exactly adjacent reads and writes.\n");
        printf("#   CP_SEQ_*: number of reads and writes from increasing offsets.\n");
        printf("#   CP_RW_SWITCHES: number of times access alternated between read and write.\n");
        printf("#   CP_*_ALIGNMENT: memory and file alignment.\n");
        printf("#   CP_*_NOT_ALIGNED: number of reads and writes that were not aligned.\n");
        printf("#   CP_MAX_*_TIME_SIZE: size of the slowest read and write operations.\n");
        printf("#   CP_SIZE_READ_*: histogram of read access sizes.\n");
        printf("#   CP_SIZE_READ_AGG_*: histogram of MPI datatype total sizes.\n");
        printf("#   CP_EXTENT_READ_*: histogram of MPI datatype extents.\n");
        printf("#   CP_STRIDE*_STRIDE: the four most common strides detected.\n");
        printf("#   CP_STRIDE*_COUNT: count of the four most common strides.\n");
        printf("#   CP_ACCESS*_ACCESS: the four most common access sizes.\n");
        printf("#   CP_ACCESS*_COUNT: count of the four most common access sizes.\n");
        printf("#   CP_DEVICE: device id reported by stat().\n");
        printf("#   CP_SIZE_AT_OPEN: size of file when first opened.\n");
        printf("#   CP_*_RANK_BYTES: fastest, slowest and variance of bytes transfer.\n");
        printf("#   CP_F_OPEN_TIMESTAMP: timestamp of first open (mpi or posix).\n");
        printf("#   CP_F_*_START_TIMESTAMP: timestamp of first read/write (mpi or posix).\n");
        printf("#   CP_F_*_END_TIMESTAMP: timestamp of last read/write (mpi or posix).\n");
        printf("#   CP_F_CLOSE_TIMESTAMP: timestamp of last close (mpi or posix).\n");
        printf("#   CP_F_POSIX_READ/WRITE_TIME: cumulative time spent in posix reads or writes.\n");
        printf("#   CP_F_MPI_READ/WRITE_TIME: cumulative time spent in mpi-io reads or writes.\n");
        printf("#   CP_F_POSIX_META_TIME: cumulative time spent in posix open, close, fsync, stat and seek, .\n");
        printf("#   CP_F_MPI_META_TIME: cumulative time spent in mpi-io open, close, set_view, and sync.\n");
        printf("#   CP_MAX_*_TIME: duration of the slowest read and write operations.\n");
        printf("#   CP_*_RANK_TIME: fastest, slowest variance of transfer time.\n");

        printf("\n");
        CP_PRINT_HEADER();
    }
356

357 358 359 360 361 362 363 364 365 366 367 368 369 370
    pdata.rank_cumul_io_time = malloc(sizeof(double)*job.nprocs);
    pdata.rank_cumul_md_time = malloc(sizeof(double)*job.nprocs);
    if (!pdata.rank_cumul_io_time || !pdata.rank_cumul_md_time)
    {
        perror("malloc failed");
        darshan_log_close(file);
        return(-1);
    }
    else
    {
        memset(pdata.rank_cumul_io_time, 0, sizeof(double)*job.nprocs);
        memset(pdata.rank_cumul_md_time, 0, sizeof(double)*job.nprocs);
    }

371
    do
372
    {
373 374
        char* mnt_pt = NULL;
        char* fs_type = NULL;
375
        hash_entry_t *hfile = NULL;
376 377 378 379

        if(cp_file.rank != -1 && cp_file.rank < last_rank)
        {
            fprintf(stderr, "Error: log file contains out of order rank data.\n");
380
            fflush(stderr);
381 382 383 384
            return(-1);
        }
        if(cp_file.rank != -1)
            last_rank = cp_file.rank;
385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
        
        for(i=0; i<mount_count; i++)
        {
            if(cp_file.counters[CP_DEVICE] == devs[i])
            {
                mnt_pt = mnt_pts[i];
                fs_type = fs_types[i];
                break;
            }
        }
        if(!mnt_pt)
            mnt_pt = "UNKNOWN";
        if(!fs_type)
            fs_type = "UNKNOWN";

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
        HASH_FIND(hlink,file_hash,&cp_file.hash,sizeof(int64_t),hfile);
        if (!hfile)
        {
            hfile = (hash_entry_t*) malloc(sizeof(*hfile));
            if (!hfile)
            {
                fprintf(stderr,"malloc failure");
                exit(1);
            }

            /* init */
            memset(hfile, 0, sizeof(*hfile));
            hfile->hash          = cp_file.hash;
            hfile->type          = 0;
            hfile->procs         = 0;
            hfile->cumul_time    = 0.0;
            hfile->meta_time     = 0.0;

            HASH_ADD(hlink,file_hash,hash,sizeof(int64_t),hfile);
        }

        accum_file(&cp_file, &total, NULL);
        accum_file(&cp_file, hfile, &fdata);
        accum_perf(&cp_file, hfile, &pdata);

        if ((mask & OPTION_BASE))
        {
            for(i=0; i<CP_NUM_INDICES; i++)
            {
                CP_PRINT(&job, &cp_file, i, mnt_pt, fs_type);
            }
            for(i=0; i<CP_F_NUM_INDICES; i++)
            {
                CP_F_PRINT(&job, &cp_file, i, mnt_pt, fs_type);
            }
        }
436
    }while((ret = darshan_log_getfile(file, &job, &cp_file)) == 1);
437 438 439 440

    /* Total Calc */
    if ((mask & OPTION_TOTAL))
    {
441 442
        for(i=0; i<CP_NUM_INDICES; i++)
        {
Philip Carns's avatar
Philip Carns committed
443
            printf("total_%s: %" PRId64 "\n",
444
                   darshan_names[i], total.counters[i]);
445 446 447
        }
        for(i=0; i<CP_F_NUM_INDICES; i++)
        {
448 449
            printf("total_%s: %lf\n",
                   darshan_f_names[i], total.fcounters[i]);
450
        }
451 452
    }

453 454 455 456 457 458
    /* Perf Calc */
    calc_perf(&job, file_hash, &pdata);
    if ((mask & OPTION_PERF))
    {
        printf("\n# performance\n");
        printf("# -----------\n");
Philip Carns's avatar
Philip Carns committed
459
        printf("# total_bytes: %" PRId64 "\n", pdata.total_bytes);
460 461 462 463 464 465 466 467 468 469 470
        printf("# slowest_rank_time: %lf\n", pdata.slowest_rank_time);
        printf("# slowest_rank_meta_time: %lf\n", pdata.slowest_rank_meta_time);
        printf("# shared_time_by_cumul: %lf\n", pdata.shared_time_by_cumul);
        printf("# shared_time_by_open: %lf\n", pdata.shared_time_by_open);
        printf("# shared_time_by_open_lastio: %lf\n", pdata.shared_time_by_open_lastio);
        printf("# shared_meta_time: %lf\n", pdata.shared_meta_time);
        printf("# agg_perf_by_cumul: %lf\n", pdata.agg_perf_by_cumul);
        printf("# agg_perf_by_open: %lf\n", pdata.agg_perf_by_open);
        printf("# agg_perf_by_open_lastio: %lf\n", pdata.agg_perf_by_open_lastio);
        printf("# agg_perf_by_slowest: %lf\n", pdata.agg_perf_by_slowest);
    }
471

472 473 474 475 476 477
    /* File Calc */
    calc_file(&job, file_hash, &fdata);
    if ((mask & OPTION_FILE))
    {
        printf("\n# files\n");
        printf("# -----\n");
Philip Carns's avatar
Philip Carns committed
478
        printf("# total: %" PRId64 " %" PRId64 " %" PRId64 "\n",
479 480 481
               fdata.total,
               fdata.total_size,
               fdata.total_max);
Philip Carns's avatar
Philip Carns committed
482
        printf("# read_only: %" PRId64 " %" PRId64 " %" PRId64 "\n",
483 484 485
               fdata.read_only,
               fdata.read_only_size,
               fdata.read_only_max);
Philip Carns's avatar
Philip Carns committed
486
        printf("# write_only: %" PRId64 " %" PRId64 " %" PRId64 "\n",
487 488 489
               fdata.write_only,
               fdata.write_only_size,
               fdata.write_only_max);
Philip Carns's avatar
Philip Carns committed
490
        printf("# read_write: %" PRId64 " %" PRId64 " %" PRId64 "\n",
491 492 493
               fdata.read_write,
               fdata.read_write_size,
               fdata.read_write_max);
Philip Carns's avatar
Philip Carns committed
494
        printf("# unique: %" PRId64 " %" PRId64 " %" PRId64 "\n",
495 496 497
               fdata.unique,
               fdata.unique_size,
               fdata.unique_max);
Philip Carns's avatar
Philip Carns committed
498
        printf("# shared: %" PRId64 " %" PRId64 " %" PRId64 "\n",
499 500 501
               fdata.shared,
               fdata.shared_size,
               fdata.shared_max);
502 503
    }

504
    if(ret < 0)
505
    {
506
        fprintf(stderr, "Error: failed to parse log file.\n");
507
        fflush(stderr);
508 509 510
        return(-1);
    }

511 512 513 514 515
    for(i=0; i<mount_count; i++)
    {
        free(mnt_pts[i]);
        free(fs_types[i]);
    }
516 517 518 519 520 521
    if(mount_count > 0)
    {
        free(devs);
        free(mnt_pts);
        free(fs_types);
    }
522
 
523
    darshan_log_close(file);
524 525 526 527 528 529 530

    HASH_ITER(hlink, file_hash, curr, tmp)
    {
        HASH_DELETE(hlink, file_hash, curr);
        free(curr);
    }

531 532
    return(0);
}
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

void accum_file(struct darshan_file *dfile,
                hash_entry_t *hfile, 
                file_data_t *fdata)
{
    int i;

    hfile->procs += 1;

    if (dfile->rank == -1)
    {
        hfile->type |= FILETYPE_SHARED;
    }
    else if (hfile->procs > 1)
    {
        hfile->type &= (~FILETYPE_UNIQUE);
        hfile->type |= FILETYPE_PARTSHARED;
    }
    else
    {
        hfile->type |= FILETYPE_UNIQUE;
    }

    for (i = 0; i < CP_NUM_INDICES; i++)
    {
        switch(i)
        {
        case CP_DEVICE:
        case CP_MODE:
        case CP_MEM_ALIGNMENT:
        case CP_FILE_ALIGNMENT:
564 565
            if(CP_FILE_PARTIAL(hfile))
                hfile->counters[i] = dfile->counters[i];
566 567 568 569 570 571
            break;
        case CP_SIZE_AT_OPEN:
            if (hfile->counters[i] == -1)
            {
                hfile->counters[i] = dfile->counters[i];
            }
572
            if (hfile->counters[i] > dfile->counters[i] && !CP_FILE_PARTIAL(dfile))
573 574 575 576 577 578 579 580 581 582 583
            {
                hfile->counters[i] = dfile->counters[i];
            }
            break;
        case CP_MAX_BYTE_READ:
        case CP_MAX_BYTE_WRITTEN:
            if (hfile->counters[i] < dfile->counters[i])
            {
                hfile->counters[i] = dfile->counters[i];
            }
            break;
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598

        case CP_STRIDE1_STRIDE:
        case CP_STRIDE2_STRIDE:
        case CP_STRIDE3_STRIDE:
        case CP_STRIDE4_STRIDE:
        case CP_ACCESS1_ACCESS:
        case CP_ACCESS2_ACCESS:
        case CP_ACCESS3_ACCESS:
        case CP_ACCESS4_ACCESS:
           /*
            * do nothing here because these will be stored
            * when the _COUNT is accessed.
            */
           break;
 
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
        case CP_STRIDE1_COUNT:
        case CP_STRIDE2_COUNT:
        case CP_STRIDE3_COUNT:
        case CP_STRIDE4_COUNT:
        case CP_ACCESS1_COUNT:
        case CP_ACCESS2_COUNT:
        case CP_ACCESS3_COUNT:
        case CP_ACCESS4_COUNT:
            if (hfile->counters[i] < dfile->counters[i])
            {
                hfile->counters[i]   = dfile->counters[i];
                hfile->counters[i-4] = dfile->counters[i-4];
            }
            break;
        case CP_FASTEST_RANK:
        case CP_SLOWEST_RANK:
        case CP_FASTEST_RANK_BYTES:
        case CP_SLOWEST_RANK_BYTES:
            hfile->counters[i] = 0;
            break;
        case CP_MAX_READ_TIME_SIZE:
        case CP_MAX_WRITE_TIME_SIZE:
            break;
        default:
            hfile->counters[i] += dfile->counters[i];
            break;
        }
    }

    for (i = 0; i < CP_F_NUM_INDICES; i++)
    {
        switch(i)
        {
            case CP_F_FASTEST_RANK_TIME:
            case CP_F_SLOWEST_RANK_TIME:
            case CP_F_VARIANCE_RANK_TIME:
            case CP_F_VARIANCE_RANK_BYTES:
                hfile->fcounters[i] = 0;
                break;
            case CP_F_MAX_READ_TIME:
                if (hfile->fcounters[i] > dfile->fcounters[i])
                {
                    hfile->fcounters[i] = dfile->fcounters[i];
                    hfile->counters[CP_MAX_READ_TIME_SIZE] =
                        dfile->counters[CP_MAX_READ_TIME_SIZE];
                }
                break;
            case CP_F_MAX_WRITE_TIME:
                if (hfile->fcounters[i] > dfile->fcounters[i])
                {
                    hfile->fcounters[i] = dfile->fcounters[i];
                    hfile->counters[CP_MAX_WRITE_TIME_SIZE] =
                        dfile->counters[CP_MAX_WRITE_TIME_SIZE];
                }
                break;
            default:
                hfile->fcounters[i] += dfile->fcounters[i];
                break;
        }
    }

    return;
}


void calc_file(struct darshan_job *djob,
               hash_entry_t *file_hash, 
               file_data_t *fdata)
{
    hash_entry_t *curr = NULL;
    hash_entry_t *tmp = NULL;

    memset(fdata, 0, sizeof(*fdata));

    HASH_ITER(hlink, file_hash, curr, tmp)
    {
        int64_t max;
        int64_t r;
        int64_t w;

        max = max3(curr->counters[CP_SIZE_AT_OPEN],
                   curr->counters[CP_MAX_BYTE_READ],
                   curr->counters[CP_MAX_BYTE_WRITTEN]);

        r = (curr->counters[CP_POSIX_READS]+
             curr->counters[CP_POSIX_FREADS]+
             curr->counters[CP_INDEP_READS]+
             curr->counters[CP_COLL_READS]+
             curr->counters[CP_SPLIT_READS]+
             curr->counters[CP_NB_READS]);

        w = (curr->counters[CP_POSIX_WRITES]+
             curr->counters[CP_POSIX_FWRITES]+
             curr->counters[CP_INDEP_WRITES]+
             curr->counters[CP_COLL_WRITES]+
             curr->counters[CP_SPLIT_WRITES]+
             curr->counters[CP_NB_WRITES]);

        fdata->total += 1;
        fdata->total_size += max;
        fdata->total_max = max(fdata->total_max, max);

        if (r && !w)
        {
            fdata->read_only += 1;
            fdata->read_only_size += max;
            fdata->read_only_max = max(fdata->read_only_max, max);
        }

        if (!r && w)
        {
            fdata->write_only += 1;
            fdata->write_only_size += max;
            fdata->write_only_max = max(fdata->write_only_max, max);
        }

        if (r && w)
        {
            fdata->read_write += 1;
            fdata->read_write_size += max;
            fdata->read_write_max = max(fdata->read_write_max, max);
        }

        if ((curr->type & (FILETYPE_SHARED|FILETYPE_PARTSHARED)))
        {
            fdata->shared += 1;
            fdata->shared_size += max;
            fdata->shared_max = max(fdata->shared_max, max);
        }

        if ((curr->type & (FILETYPE_UNIQUE)))
        {
            fdata->unique += 1;
            fdata->unique_size += max;
            fdata->unique_max = max(fdata->unique_max, max);
        }
    }

    return;
}

void accum_perf(struct darshan_file *dfile,
                hash_entry_t *hfile,
                perf_data_t *pdata)
{
    int64_t mpi_file;

    pdata->total_bytes += dfile->counters[CP_BYTES_READ] +
                          dfile->counters[CP_BYTES_WRITTEN];

    mpi_file = dfile->counters[CP_INDEP_OPENS] +
               dfile->counters[CP_COLL_OPENS];

    /*
     * Calculation of Shared File Time
     *   Four Methods!!!!
     *     by_cumul: sum time counters and divide by nprocs
     *               (inaccurate if lots of variance between procs)
     *     by_open: difference between timestamp of open and close
     *              (inaccurate if file is left open without i/o happening)
     *     by_open_lastio: difference between timestamp of open and the
     *                     timestamp of last i/o
     *                     (similar to above but fixes case where file is left
     *                      open after io is complete)
     *     by_slowest: use slowest rank time from log data
     *                 (most accurate but requires newer log version)
     */
    if (dfile->rank == -1)
    {
        /* by_open (same for MPI or POSIX) */
769 770 771 772 773 774 775
        if (dfile->fcounters[CP_F_CLOSE_TIMESTAMP] >
            dfile->fcounters[CP_F_OPEN_TIMESTAMP])
        {
            pdata->shared_time_by_open +=
                dfile->fcounters[CP_F_CLOSE_TIMESTAMP] -
                dfile->fcounters[CP_F_OPEN_TIMESTAMP];
        }
776 777 778 779 780

        /* by_open_lastio (same for MPI or POSIX) */
        if (dfile->fcounters[CP_F_READ_END_TIMESTAMP] >
            dfile->fcounters[CP_F_WRITE_END_TIMESTAMP])
        {
781 782 783 784 785 786 787
            /* be careful: file may have been opened but not read or written */
            if(dfile->fcounters[CP_F_READ_END_TIMESTAMP] > dfile->fcounters[CP_F_OPEN_TIMESTAMP])
            {
                pdata->shared_time_by_open_lastio += 
                    dfile->fcounters[CP_F_READ_END_TIMESTAMP] - 
                    dfile->fcounters[CP_F_OPEN_TIMESTAMP];
            }
788 789 790
        }
        else
        {
791 792 793 794 795 796 797
            /* be careful: file may have been opened but not read or written */
            if(dfile->fcounters[CP_F_WRITE_END_TIMESTAMP] > dfile->fcounters[CP_F_OPEN_TIMESTAMP])
            {
                pdata->shared_time_by_open_lastio += 
                    dfile->fcounters[CP_F_WRITE_END_TIMESTAMP] - 
                    dfile->fcounters[CP_F_OPEN_TIMESTAMP];
            }
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
        }

        /* by_cumul */
        if (mpi_file)
        {
            pdata->shared_time_by_cumul +=
                dfile->fcounters[CP_F_MPI_META_TIME] +
                dfile->fcounters[CP_F_MPI_READ_TIME] +
                dfile->fcounters[CP_F_MPI_WRITE_TIME];
            pdata->shared_meta_time += dfile->fcounters[CP_F_MPI_META_TIME];
        }
        else
        {
            pdata->shared_time_by_cumul +=
                dfile->fcounters[CP_F_POSIX_META_TIME] +
                dfile->fcounters[CP_F_POSIX_READ_TIME] +
                dfile->fcounters[CP_F_POSIX_WRITE_TIME];
            pdata->shared_meta_time += dfile->fcounters[CP_F_POSIX_META_TIME];
        }

        /* by_slowest (same for MPI or POSIX) */
        pdata->shared_time_by_slowest +=
            dfile->fcounters[CP_F_SLOWEST_RANK_TIME];
    }

    /*
     * Calculation of Unique File Time
     *   record the data for each file and sum it 
     */
    else
    {
        if (mpi_file)
        {
831
#if 0
832 833 834 835
            hfile->cumul_time += dfile->fcounters[CP_F_MPI_META_TIME] +
                                dfile->fcounters[CP_F_MPI_READ_TIME] +
                                dfile->fcounters[CP_F_MPI_WRITE_TIME];
            hfile->meta_time += dfile->fcounters[CP_F_MPI_META_TIME];
836 837 838 839 840 841
#else
            pdata->rank_cumul_io_time[dfile->rank] += dfile->fcounters[CP_F_MPI_META_TIME] +
                                dfile->fcounters[CP_F_MPI_READ_TIME] +
                                dfile->fcounters[CP_F_MPI_WRITE_TIME];
            pdata->rank_cumul_md_time[dfile->rank] += dfile->fcounters[CP_F_MPI_META_TIME];
#endif
842 843 844
        }
        else
        {
845
#if 0
846 847 848 849
             hfile->cumul_time += dfile->fcounters[CP_F_POSIX_META_TIME] +
                                 dfile->fcounters[CP_F_POSIX_READ_TIME] +
                                 dfile->fcounters[CP_F_POSIX_WRITE_TIME];
             hfile->meta_time += dfile->fcounters[CP_F_POSIX_META_TIME];
850 851 852 853 854 855 856
#else
            pdata->rank_cumul_io_time[dfile->rank] += dfile->fcounters[CP_F_POSIX_META_TIME] +
                                dfile->fcounters[CP_F_POSIX_READ_TIME] +
                                dfile->fcounters[CP_F_POSIX_WRITE_TIME];
            pdata->rank_cumul_md_time[dfile->rank] += dfile->fcounters[CP_F_POSIX_META_TIME];

#endif
857
        }
858

859
#if 0
860 861
        pdata->rank_cumul_io_time[dfile->rank] += hfile->cumul_time;
        pdata->rank_cumul_md_time[dfile->rank] += hfile->meta_time;
862
#endif
863 864 865 866 867 868 869 870 871
    }

    return;
}

void calc_perf(struct darshan_job *djob,
               hash_entry_t *hash_rank_uniq,
               perf_data_t *pdata)
{
872
    int64_t i;
873 874 875 876 877 878

    pdata->shared_time_by_cumul =
        pdata->shared_time_by_cumul / (double)djob->nprocs;

    pdata->shared_meta_time = pdata->shared_meta_time / (double)djob->nprocs;

879
    for (i=0; i<djob->nprocs; i++)
880
    {
881
        if (pdata->rank_cumul_io_time[i] > pdata->slowest_rank_time)
882
        {
883
            pdata->slowest_rank_time = pdata->rank_cumul_io_time[i];
884
            pdata->slowest_rank_meta_time = pdata->rank_cumul_md_time[i];
885 886 887
        }
    }

888
    if (pdata->slowest_rank_time + pdata->shared_time_by_cumul)
889
    pdata->agg_perf_by_cumul = ((double)pdata->total_bytes / 1048576.0) /
890 891 892 893
                                  (pdata->slowest_rank_time +
                                   pdata->shared_time_by_cumul);

    if (pdata->slowest_rank_time + pdata->shared_time_by_open)
894
    pdata->agg_perf_by_open  = ((double)pdata->total_bytes / 1048576.0) / 
895 896 897 898
                                   (pdata->slowest_rank_time +
                                    pdata->shared_time_by_open);

    if (pdata->slowest_rank_time + pdata->shared_time_by_open_lastio)
899 900 901
    pdata->agg_perf_by_open_lastio = ((double)pdata->total_bytes / 1048576.0) /
                                     (pdata->slowest_rank_time +
                                      pdata->shared_time_by_open_lastio);
902

903 904
    if (pdata->slowest_rank_time + pdata->shared_time_by_slowest)
    pdata->agg_perf_by_slowest = ((double)pdata->total_bytes / 1048576.0) /
905 906
                                     (pdata->slowest_rank_time +
                                      pdata->shared_time_by_slowest);
907 908 909

    return;
}