darshan-mpi-io.c 73.4 KB
Newer Older
1 2 3 4 5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6 7 8
#define _XOPEN_SOURCE 500
#define _GNU_SOURCE /* for tdestroy() */

9
#include "darshan-runtime-config.h"
10

11
#include <stdio.h>
12
#ifdef HAVE_MNTENT_H
13
#include <mntent.h>
14
#endif
15 16 17 18 19 20 21
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <limits.h>
#include <unistd.h>
#include <pthread.h>
#include <sys/types.h>
22
#include <sys/stat.h>
23
#include <sys/vfs.h>
24 25 26 27 28 29
#include <zlib.h>
#include <assert.h>
#include <search.h>

#include "mpi.h"
#include "darshan.h"
30
#include "darshan-dynamic.h"
31

32 33 34 35 36
extern char* __progname;

/* maximum number of memory segments each process will write to the log */
#define CP_MAX_MEM_SEGMENTS 8

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
/* Some old versions of MPI don't provide all of these COMBINER definitions.  
 * If any are missing then we define them to an arbitrary value just to 
 * prevent compile errors in DATATYPE_INC().
 */
#ifndef MPI_COMBINER_NAMED
    #define MPI_COMBINER_NAMED CP_COMBINER_NAMED
#endif
#ifndef MPI_COMBINER_DUP
    #define MPI_COMBINER_DUP CP_COMBINER_DUP
#endif
#ifndef MPI_COMBINER_CONTIGUOUS
    #define MPI_COMBINER_CONTIGUOUS CP_COMBINER_CONTIGUOUS
#endif
#ifndef MPI_COMBINER_VECTOR
    #define MPI_COMBINER_VECTOR CP_COMBINER_VECTOR
#endif
#ifndef MPI_COMBINER_HVECTOR_INTEGER
    #define MPI_COMBINER_HVECTOR_INTEGER CP_COMBINER_HVECTOR_INTEGER
#endif
#ifndef MPI_COMBINER_HVECTOR
    #define MPI_COMBINER_HVECTOR CP_COMBINER_HVECTOR
#endif
#ifndef MPI_COMBINER_INDEXED
    #define MPI_COMBINER_INDEXED CP_COMBINER_INDEXED
#endif
#ifndef MPI_COMBINER_HINDEXED_INTEGER
    #define MPI_COMBINER_HINDEXED_INTEGER CP_COMBINER_HINDEXED_INTEGER
#endif
#ifndef MPI_COMBINER_HINDEXED
    #define MPI_COMBINER_HINDEXED CP_COMBINER_HINDEXED
#endif
#ifndef MPI_COMBINER_INDEXED_BLOCK
    #define MPI_COMBINER_INDEXED_BLOCK CP_COMBINER_INDEXED_BLOCK
#endif
#ifndef MPI_COMBINER_STRUCT_INTEGER
    #define MPI_COMBINER_STRUCT_INTEGER CP_COMBINER_STRUCT_INTEGER
#endif
#ifndef MPI_COMBINER_STRUCT
    #define MPI_COMBINER_STRUCT CP_COMBINER_STRUCT
#endif
#ifndef MPI_COMBINER_SUBARRAY
    #define MPI_COMBINER_SUBARRAY CP_COMBINER_SUBARRAY
#endif
#ifndef MPI_COMBINER_DARRAY
    #define MPI_COMBINER_DARRAY CP_COMBINER_DARRAY
#endif
#ifndef MPI_COMBINER_F90_REAL
    #define MPI_COMBINER_F90_REAL CP_COMBINER_F90_REAL
#endif
#ifndef MPI_COMBINER_F90_COMPLEX
    #define MPI_COMBINER_F90_COMPLEX CP_COMBINER_F90_COMPLEX
#endif
#ifndef MPI_COMBINER_F90_INTEGER
    #define MPI_COMBINER_F90_INTEGER CP_COMBINER_F90_INTEGER
#endif
#ifndef MPI_COMBINER_RESIZED
    #define MPI_COMBINER_RESIZED CP_COMBINER_RESIZED
#endif

96 97
#define CP_DATATYPE_INC(__file, __datatype) do {\
    int num_integers, num_addresses, num_datatypes, combiner, ret; \
98 99
    ret = DARSHAN_MPI_CALL(PMPI_Type_get_envelope)(__datatype, &num_integers, \
        &num_addresses, &num_datatypes, &combiner); \
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
    if(ret == MPI_SUCCESS) { \
        switch(combiner) { \
            case MPI_COMBINER_NAMED:\
                CP_INC(__file,CP_COMBINER_NAMED,1); break; \
            case MPI_COMBINER_DUP:\
                CP_INC(__file,CP_COMBINER_DUP,1); break; \
            case MPI_COMBINER_CONTIGUOUS:\
                CP_INC(__file,CP_COMBINER_CONTIGUOUS,1); break; \
            case MPI_COMBINER_VECTOR:\
                CP_INC(__file,CP_COMBINER_VECTOR,1); break; \
            case MPI_COMBINER_HVECTOR_INTEGER:\
                CP_INC(__file,CP_COMBINER_HVECTOR_INTEGER,1); break; \
            case MPI_COMBINER_HVECTOR:\
                CP_INC(__file,CP_COMBINER_HVECTOR,1); break; \
            case MPI_COMBINER_INDEXED:\
                CP_INC(__file,CP_COMBINER_INDEXED,1); break; \
            case MPI_COMBINER_HINDEXED_INTEGER:\
                CP_INC(__file,CP_COMBINER_HINDEXED_INTEGER,1); break; \
            case MPI_COMBINER_HINDEXED:\
                CP_INC(__file,CP_COMBINER_HINDEXED,1); break; \
            case MPI_COMBINER_INDEXED_BLOCK:\
                CP_INC(__file,CP_COMBINER_INDEXED_BLOCK,1); break; \
            case MPI_COMBINER_STRUCT_INTEGER:\
                CP_INC(__file,CP_COMBINER_STRUCT_INTEGER,1); break; \
            case MPI_COMBINER_STRUCT:\
                CP_INC(__file,CP_COMBINER_STRUCT,1); break; \
            case MPI_COMBINER_SUBARRAY:\
                CP_INC(__file,CP_COMBINER_SUBARRAY,1); break; \
            case MPI_COMBINER_DARRAY:\
                CP_INC(__file,CP_COMBINER_DARRAY,1); break; \
            case MPI_COMBINER_F90_REAL:\
                CP_INC(__file,CP_COMBINER_F90_REAL,1); break; \
            case MPI_COMBINER_F90_COMPLEX:\
                CP_INC(__file,CP_COMBINER_F90_COMPLEX,1); break; \
            case MPI_COMBINER_F90_INTEGER:\
                CP_INC(__file,CP_COMBINER_F90_INTEGER,1); break; \
            case MPI_COMBINER_RESIZED:\
                CP_INC(__file,CP_COMBINER_RESIZED,1); break; \
        } \
    } \
} while(0)

#define CP_RECORD_MPI_WRITE(__ret, __fh, __count, __datatype, __counter, __tm1, __tm2) do { \
    struct darshan_file_runtime* file; \
    int size = 0; \
    MPI_Aint extent = 0; \
    if(__ret != MPI_SUCCESS) break; \
    file = darshan_file_by_fh(__fh); \
    if(!file) break; \
149
    DARSHAN_MPI_CALL(PMPI_Type_size)(__datatype, &size);  \
150
    size = size * __count; \
151
    DARSHAN_MPI_CALL(PMPI_Type_extent)(__datatype, &extent); \
152 153 154 155
    CP_BUCKET_INC(file, CP_SIZE_WRITE_AGG_0_100, size); \
    CP_BUCKET_INC(file, CP_EXTENT_WRITE_0_100, extent); \
    CP_INC(file, __counter, 1); \
    CP_DATATYPE_INC(file, __datatype); \
156
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_mpi_write_end, CP_F_MPI_WRITE_TIME); \
157 158 159 160 161 162 163 164 165 166 167 168
    if(CP_F_VALUE(file, CP_F_WRITE_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_WRITE_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_WRITE_END_TIMESTAMP, __tm2); \
} while(0)

#define CP_RECORD_MPI_READ(__ret, __fh, __count, __datatype, __counter, __tm1, __tm2) do { \
    struct darshan_file_runtime* file; \
    int size = 0; \
    MPI_Aint extent = 0; \
    if(__ret != MPI_SUCCESS) break; \
    file = darshan_file_by_fh(__fh); \
    if(!file) break; \
169
    DARSHAN_MPI_CALL(PMPI_Type_size)(__datatype, &size);  \
170
    size = size * __count; \
171
    DARSHAN_MPI_CALL(PMPI_Type_extent)(__datatype, &extent); \
172 173 174 175
    CP_BUCKET_INC(file, CP_SIZE_READ_AGG_0_100, size); \
    CP_BUCKET_INC(file, CP_EXTENT_READ_0_100, extent); \
    CP_INC(file, __counter, 1); \
    CP_DATATYPE_INC(file, __datatype); \
176
    CP_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_mpi_read_end, CP_F_MPI_READ_TIME); \
177 178 179 180 181
    if(CP_F_VALUE(file, CP_F_READ_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_READ_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_READ_END_TIMESTAMP, __tm2); \
} while(0)

182 183 184
static void cp_log_construct_indices(struct darshan_job_runtime* final_job,
    int rank, int* inout_count, int* lengths, void** pointers, char*
    trailing_data);
185
static int cp_log_write(struct darshan_job_runtime* final_job, int rank, 
186
    char* logfile_name, int count, int* lengths, void** pointers, double start_log_time);
187
static void cp_log_record_hints(struct darshan_job_runtime* final_job, int rank);
188
static int cp_log_reduction(struct darshan_job_runtime* final_job, int rank, 
189
    char* logfile_name, MPI_Offset* next_offset);
190 191 192 193 194 195
static void darshan_file_reduce(void* infile_v, 
    void* inoutfile_v, int *len, 
    MPI_Datatype *datatype);
static int cp_log_compress(struct darshan_job_runtime* final_job,
    int rank, int* inout_count, int* lengths, void** pointers);
static int file_compare(const void* a, const void* b);
196
       void darshan_mpi_initialize(int *argc, char ***argv);
197 198 199 200 201 202
static int darshan_file_variance(
    struct darshan_file *infile_array,
    struct darshan_file *outfile_array,
    int count, int rank);
static void pairwise_variance_reduce (
    void *invec, void *inoutvec, int *len, MPI_Datatype *dt);
203
static void debug_mounts(const char* mtab_file, const char* out_file);
204

205 206 207
static struct darshan_file_runtime* darshan_file_by_fh(MPI_File fh);
static void darshan_file_close_fh(MPI_File fh);
static struct darshan_file_runtime* darshan_file_by_name_setfh(const char* name, MPI_File fh);
208

209
#define CP_MAX_MNTS 32
210 211
static uint64_t mnt_hash_array[CP_MAX_MNTS] = {0};
static int64_t mnt_id_array[CP_MAX_MNTS] = {0};
212
static int64_t mnt_block_size_array[CP_MAX_MNTS] = {0};
213
static char* mnt_path_array[CP_MAX_MNTS] = {0};
214

215 216
static uint64_t mnt_hash_array_root[CP_MAX_MNTS] = {0};
static int64_t mnt_id_array_root[CP_MAX_MNTS] = {0};
217 218 219 220 221
struct
{
    int64_t mnt_id_local;
    int64_t mnt_id_root;
} mnt_mapping[CP_MAX_MNTS];
222

223 224 225 226 227 228 229
struct variance_dt
{
    double n;
    double T;
    double S;
};

230
void darshan_mpi_initialize(int *argc, char ***argv)
231 232 233 234
{
    int nprocs;
    int rank;

235 236
    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &rank);
237 238 239 240 241 242 243 244 245 246 247

    if(argc && argv)
    {
        darshan_initialize(*argc, *argv, nprocs, rank);
    }
    else
    {
        /* we don't see argc and argv here in fortran */
        darshan_initialize(0, NULL, nprocs, rank);
    }

248
    return;
249 250 251 252 253 254 255 256 257 258 259 260
}

void darshan_shutdown(int timing_flag)
{
    int rank;
    char* logfile_name;
    struct darshan_job_runtime* final_job;
    double start_log_time = 0;
    int all_ret = 0;
    int local_ret = 0;
    MPI_Offset next_offset = 0;
    char* jobid_str;
261 262
    char* envjobid;
    char* logpath;
263 264 265 266 267 268
    int jobid;
    int index_count = 0;
    int lengths[CP_MAX_MEM_SEGMENTS];
    void* pointers[CP_MAX_MEM_SEGMENTS];
    int ret;
    double red1=0, red2=0, gz1=0, gz2=0, write1=0, write2=0, tm_end=0;
269
    double bcst1=0, bcst2=0, bcst3=0;
270
    int nprocs;
271 272
    int i, j;
    int map_index = 0;
273
    time_t start_time_tmp = 0;
274 275
    uint64_t logmod;
    char hname[HOST_NAME_MAX];
276 277 278 279 280
    char* logpath_override = NULL;
#ifdef __CP_LOG_ENV
    char env_check[256];
    char* env_tok;
#endif
281
    uint64_t hlevel;
282 283 284 285 286

    CP_LOCK();
    if(!darshan_global_job)
    {
        CP_UNLOCK();
287
        return;
288 289 290 291 292 293 294 295
    }
    /* disable further tracing while hanging onto the data so that we can
     * write it out
     */
    final_job = darshan_global_job;
    darshan_global_job = NULL;
    CP_UNLOCK();

296
    start_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();
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

    /* figure out which access sizes to log */
    darshan_walk_file_accesses(final_job);

    /* if the records have been condensed, then zero out fields that are no
     * longer valid for safety 
     */
    if(final_job->flags & CP_FLAG_CONDENSED && final_job->file_count)
    {
        CP_SET(&final_job->file_runtime_array[0], CP_MODE, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_CONSEC_READS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_CONSEC_WRITES, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_SEQ_READS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_SEQ_WRITES, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE1_STRIDE, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE2_STRIDE, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE3_STRIDE, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE4_STRIDE, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE1_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE2_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE3_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_STRIDE4_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS1_ACCESS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS2_ACCESS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS3_ACCESS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS4_ACCESS, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS1_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS2_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS3_COUNT, 0);
        CP_SET(&final_job->file_runtime_array[0], CP_ACCESS4_COUNT, 0);
        
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_OPEN_TIMESTAMP, 0);
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_CLOSE_TIMESTAMP, 0);
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_READ_START_TIMESTAMP, 0);
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_READ_END_TIMESTAMP, 0);
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_WRITE_START_TIMESTAMP, 0);
        CP_F_SET(&final_job->file_runtime_array[0], CP_F_WRITE_END_TIMESTAMP, 0);
    }

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
        darshan_finalize(final_job);
        return;
    }

343
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &rank);
344

345 346
    /* broadcast mount point information from root */
    if(rank == 0)
347
    {
348 349 350 351
        memcpy(mnt_hash_array_root, mnt_hash_array,
            CP_MAX_MNTS*sizeof(uint64_t));
        memcpy(mnt_id_array_root, mnt_id_array,
            CP_MAX_MNTS*sizeof(int64_t));
352
    }
353

354
    bcst1=DARSHAN_MPI_CALL(PMPI_Wtime)();
355 356
    hlevel = bcst1*1000000;

357 358 359 360 361
    DARSHAN_MPI_CALL(PMPI_Bcast)(mnt_id_array_root,
        CP_MAX_MNTS*sizeof(int64_t), MPI_BYTE, 0, MPI_COMM_WORLD);
    DARSHAN_MPI_CALL(PMPI_Bcast)(mnt_hash_array_root,
        CP_MAX_MNTS*sizeof(uint64_t), MPI_BYTE, 0, MPI_COMM_WORLD);
    bcst2=DARSHAN_MPI_CALL(PMPI_Wtime)();
362

363 364 365 366 367 368 369 370 371 372 373 374
    /* identify any common mount points that have different device ids on
     * non-root processes
     */
    for(i=0; (i<CP_MAX_MNTS && mnt_hash_array_root[i] != 0); i++)
    {
        for(j=0; (j<CP_MAX_MNTS && mnt_hash_array[j] != 0); j++)
        {
            if(mnt_hash_array_root[i] == mnt_hash_array[j])
            {
                /* found a shared mount point */
                if(mnt_id_array_root[i] != mnt_id_array[j])
                {
375 376 377 378 379 380
                    /* mismatching ids; record correct mapping */
                    mnt_mapping[map_index].mnt_id_local =
                        mnt_id_array[j];
                    mnt_mapping[map_index].mnt_id_root = 
                        mnt_id_array_root[i];
                    map_index++;
381 382 383 384 385
                }
                break;
            }
        }
    }
386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401
 
    /* adjust affected file records */
    for(i=0; (i<final_job->file_count && map_index > 0); i++)
    {
        for(j=0; j<map_index; j++)
        {
            if(final_job->file_array[i].counters[CP_DEVICE] ==
                mnt_mapping[j].mnt_id_local)
            {
                final_job->file_array[i].counters[CP_DEVICE] =  
                    mnt_mapping[j].mnt_id_root;
                break;
            }
        }
    }
   
402 403 404 405 406 407
    /* construct log file name */
    if(rank == 0)
    {
        char cuser[L_cuserid] = {0};
        struct tm* my_tm;

408 409 410 411 412 413 414 415 416 417 418
        /* Use CP_JOBID_OVERRIDE for the env var or CP_JOBID */
        envjobid = getenv(CP_JOBID_OVERRIDE);
        if (!envjobid)
        {
            envjobid = CP_JOBID;
        }

        /* Use CP_LOG_PATH_OVERRIDE for the value or __CP_LOG_PATH */
        logpath = getenv(CP_LOG_PATH_OVERRIDE);
        if (!logpath)
        {
419
#ifdef __CP_LOG_PATH
420
            logpath = __CP_LOG_PATH;
421
#endif
422 423
        }

424
        /* find a job id */
425
        jobid_str = getenv(envjobid);
426 427 428 429 430 431 432 433 434 435 436 437
        if(jobid_str)
        {
            /* in cobalt we can find it in env var */
            ret = sscanf(jobid_str, "%d", &jobid);
        }
        if(!jobid_str || ret != 1)
        {
            /* use pid as fall back */
            jobid = getpid();
        }

        /* break out time into something human readable */
438 439
        start_time_tmp += final_job->log_job.start_time;
        my_tm = localtime(&start_time_tmp);
440

441 442 443 444 445 446 447 448 449 450 451
        /* get the username for this job.  In order we will try each of the
         * following until one of them succeeds:
         *
         * - cuserid()
         * - getenv("LOGNAME")
         * - snprintf(..., geteuid());
         *
         * Note that we do not use getpwuid() because it generally will not
         * work in statically compiled binaries.
         */

452
#ifndef DARSHAN_DISABLE_CUSERID
453
        cuserid(cuser);
454
#endif
455 456 457 458 459 460 461 462 463 464 465 466 467 468

        /* if cuserid() didn't work, then check the environment */
        if (strcmp(cuser, "") == 0)
        {
            char* logname_string;
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }

        }

        /* if cuserid() and environment both fail, then fall back to uid */
469 470 471 472 473
        if (strcmp(cuser, "") == 0)
        {
            uid_t uid = geteuid();
            snprintf(cuser, sizeof(cuser), "%u", uid);
        }
474

475 476
        /* generate a random number to help differentiate the log */
        (void) gethostname(hname, sizeof(hname));
477
        logmod = darshan_hash((void*)hname,strlen(hname),hlevel);
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
        /* see if darshan was configured using the --with-logpath-by-env
         * argument, which allows the user to specify an absolute path to
         * place logs via an env variable.
         */
#ifdef __CP_LOG_ENV
        /* just silently skip if the environment variable list is too big */
        if(strlen(__CP_LOG_ENV) < 256)
        {
            /* copy env variable list to a temporary buffer */
            strcpy(env_check, __CP_LOG_ENV);
            /* tokenize the comma-separated list */
            env_tok = strtok(env_check, ",");
            if(env_tok)
            {
                do
                {
                    /* check each env variable in order */
                    logpath_override = getenv(env_tok); 
                    if(logpath_override)
                    {
                        /* stop as soon as we find a match */
                        break;
                    }
                }while((env_tok = strtok(NULL, ",")));
            }
        }
#endif

507
       
508
        if(logpath_override)
509
        {
510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
            ret = snprintf(logfile_name, PATH_MAX, 
                "%s/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath_override, 
                cuser, __progname, jobid,
                (my_tm->tm_mon+1), 
                my_tm->tm_mday, 
                (my_tm->tm_hour*60*60 + my_tm->tm_min*60 + my_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath_override, jobid);
            }
        }
526
        else if(logpath)
527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
        {
            ret = snprintf(logfile_name, PATH_MAX, 
                "%s/%d/%d/%d/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath, (my_tm->tm_year+1900), 
                (my_tm->tm_mon+1), my_tm->tm_mday, 
                cuser, __progname, jobid,
                (my_tm->tm_mon+1), 
                my_tm->tm_mday, 
                (my_tm->tm_hour*60*60 + my_tm->tm_min*60 + my_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath, jobid);
            }
544
        }
545 546 547 548
        else
        {
            logfile_name[0] = '\0';
        }
549 550 551

        /* add jobid */
        final_job->log_job.jobid = (int64_t)jobid;
552 553 554
    }

    /* broadcast log file name */
555 556 557
    bcst3=DARSHAN_MPI_CALL(PMPI_Wtime)();
    DARSHAN_MPI_CALL(PMPI_Bcast)(logfile_name, PATH_MAX, MPI_CHAR, 0,
        MPI_COMM_WORLD);
558

559 560 561 562 563 564 565
    if(strlen(logfile_name) == 0)
    {
        /* failed to generate log file name */
        darshan_finalize(final_job);
	return;
    }

566 567 568 569
    final_job->log_job.end_time = time(NULL);

    /* reduce records for shared files */
    if(timing_flag)
570
        red1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
571 572
    local_ret = cp_log_reduction(final_job, rank, logfile_name, 
        &next_offset);
573
    if(timing_flag)
574 575
        red2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&local_ret, &all_ret, 1, MPI_INT, MPI_LOR, 
576 577
        MPI_COMM_WORLD);

578 579 580 581 582
    /* if we are using any hints to write the log file, then record those
     * hints in the log file header
     */
    cp_log_record_hints(final_job, rank);

583 584 585 586
    if(all_ret == 0)
    {
        /* collect data to write from local process */
        cp_log_construct_indices(final_job, rank, &index_count, lengths, 
587
            pointers, final_job->trailing_data);
588 589 590 591 592 593
    }

    if(all_ret == 0)
    {
        /* compress data */
        if(timing_flag)
594
            gz1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
595 596 597
        local_ret = cp_log_compress(final_job, rank, &index_count, 
            lengths, pointers);
        if(timing_flag)
598 599 600
            gz2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&local_ret, &all_ret, 1,
            MPI_INT, MPI_LOR, MPI_COMM_WORLD);
601 602 603 604 605 606
    }

    if(all_ret == 0)
    {
        /* actually write out log file */
        if(timing_flag)
607
            write1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
608
        local_ret = cp_log_write(final_job, rank, logfile_name, 
609 610
            index_count, lengths, pointers, start_log_time);
        if(timing_flag)
611 612 613
            write2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&local_ret, &all_ret, 1,
            MPI_INT, MPI_LOR, MPI_COMM_WORLD);
614 615
    }

616
    if(rank == 0)
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
        if(all_ret != 0)
        {
            fprintf(stderr, "darshan library warning: unable to write log file %s\n", logfile_name);
            /* if any process failed to write log, then delete the whole 
             * file so we don't leave corrupted results
             */
            unlink(logfile_name);
        }
        else
        {
            /* rename from *.darshan_partial to *-<logwritetime>.darshan.gz,
             * which indicates that this log file is complete and ready for
             * analysis
             */ 
            char* mod_index;
            double end_log_time;
            char* new_logfile_name;

            new_logfile_name = malloc(PATH_MAX);
            if(new_logfile_name)
            {
                new_logfile_name[0] = '\0';
                end_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();
                strcat(new_logfile_name, logfile_name);
                mod_index = strstr(new_logfile_name, ".darshan_partial");
                sprintf(mod_index, "_%d.darshan.gz", (int)(end_log_time-start_log_time+1));
                rename(logfile_name, new_logfile_name);
                /* set permissions on log file */
646 647 648
#ifdef __CP_GROUP_READABLE_LOGS
                chmod(new_logfile_name, (S_IRUSR|S_IRGRP)); 
#else
649
                chmod(new_logfile_name, (S_IRUSR)); 
650
#endif
651 652 653
                free(new_logfile_name);
            }
        }
654 655
    }

656 657
    if(final_job->trailing_data)
        free(final_job->trailing_data);
658 659 660
    for(i=0; i<CP_MAX_MNTS; i++)
    {
        if(mnt_path_array[i])
661
        {
662
            free(mnt_path_array[i]);
663 664
            mnt_path_array[i] = NULL;
        }
665
    }
666 667 668 669 670 671 672 673 674
    free(logfile_name);
    darshan_finalize(final_job);
    
    if(timing_flag)
    {
        double red_tm, red_slowest;
        double gz_tm, gz_slowest;
        double write_tm, write_slowest;
        double all_tm, all_slowest;
675
        double bcst_tm, bcst_slowest;
676
        
677
        tm_end = DARSHAN_MPI_CALL(PMPI_Wtime)();
678

679
        bcst_tm=(bcst2-bcst1)+(red1-bcst3);
680 681 682 683 684
        red_tm = red2-red1;
        gz_tm = gz2-gz1;
        write_tm = write2-write1;
        all_tm = tm_end-start_log_time;

685
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&red_tm, &red_slowest, 1,
686
            MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
687
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&gz_tm, &gz_slowest, 1,
688
            MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
689
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&write_tm, &write_slowest, 1,
690
            MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
691
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&all_tm, &all_slowest, 1,
692
            MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
693
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&bcst_tm, &bcst_slowest, 1,
694
            MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
695 696 697

        if(rank == 0)
        {
698
            DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
699
            printf("#<op>\t<nprocs>\t<time>\n");
700
            printf("bcst\t%d\t%f\n", nprocs, bcst_slowest);
701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719
            printf("reduce\t%d\t%f\n", nprocs, red_slowest);
            printf("gzip\t%d\t%f\n", nprocs, gz_slowest);
            printf("write\t%d\t%f\n", nprocs, write_slowest);
            printf("all\t%d\t%f\n", nprocs, all_slowest);
        }
    }

    return;
}

int MPI_File_open(MPI_Comm comm, char *filename, int amode, MPI_Info info, MPI_File *fh) 
{
    int ret;
    struct darshan_file_runtime* file;
    char* tmp;
    int comm_size;
    double tm1, tm2;

    tm1 = darshan_wtime();
720
    ret = DARSHAN_MPI_CALL(PMPI_File_open)(comm, filename, amode, info, fh);
721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
    tm2 = darshan_wtime();

    if(ret == MPI_SUCCESS)
    {
        CP_LOCK();

        /* use ROMIO approach to strip prefix if present */
        /* strip off prefix if there is one, but only skip prefixes
         * if they are greater than length one to allow for windows
         * drive specifications (e.g. c:\...) 
         */
        tmp = strchr(filename, ':');
        if (tmp > filename + 1) {
            filename = tmp + 1;
        }

737 738
        file = darshan_file_by_name_setfh(filename, (*fh));
        if(file)
739 740
        {
            CP_SET(file, CP_MODE, amode);
741
            CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_mpi_meta_end, CP_F_MPI_META_TIME);
742
            if(CP_F_VALUE(file, CP_F_OPEN_TIMESTAMP) == 0)
743 744 745
                CP_F_SET(file, CP_F_OPEN_TIMESTAMP,
                DARSHAN_MPI_CALL(PMPI_Wtime)());
            DARSHAN_MPI_CALL(PMPI_Comm_size)(comm, &comm_size);
746 747 748 749 750 751 752 753 754 755 756 757 758
            if(comm_size == 1)
            {
                CP_INC(file, CP_INDEP_OPENS, 1);
            }
            else
            {
                CP_INC(file, CP_COLL_OPENS, 1);
            }
            if(info != MPI_INFO_NULL)
            {
                CP_INC(file, CP_HINTS, 1);
            }
        }
759

760 761 762 763 764 765 766 767 768 769 770 771 772 773
        CP_UNLOCK();
    }

    return(ret);
}

int MPI_File_close(MPI_File *fh) 
{
    struct darshan_file_runtime* file;
    MPI_File tmp_fh = *fh;
    double tm1, tm2;
    int ret;
    
    tm1 = darshan_wtime();
774
    ret = DARSHAN_MPI_CALL(PMPI_File_close)(fh);
775 776 777 778 779 780
    tm2 = darshan_wtime();

    CP_LOCK();
    file = darshan_file_by_fh(tmp_fh);
    if(file)
    {
781
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, DARSHAN_MPI_CALL(PMPI_Wtime)());
782
        CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_mpi_meta_end, CP_F_MPI_META_TIME);
783
        darshan_file_close_fh(tmp_fh);
784 785 786 787 788 789 790 791 792 793 794 795 796
    }
    CP_UNLOCK();

    return(ret);
}

int MPI_File_sync(MPI_File fh)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    tm1 = darshan_wtime();
797
    ret = DARSHAN_MPI_CALL(PMPI_File_sync)(fh);
798 799 800 801 802 803 804
    tm2 = darshan_wtime();
    if(ret == MPI_SUCCESS)
    {
        CP_LOCK();
        file = darshan_file_by_fh(fh);
        if(file)
        {
805
            CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_mpi_write_end, CP_F_MPI_WRITE_TIME);
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
            CP_INC(file, CP_SYNCS, 1);
        }
        CP_UNLOCK();
    }

    return(ret);
}


int MPI_File_set_view(MPI_File fh, MPI_Offset disp, MPI_Datatype etype, 
    MPI_Datatype filetype, char *datarep, MPI_Info info)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    tm1 = darshan_wtime();
823 824
    ret = DARSHAN_MPI_CALL(PMPI_File_set_view)(fh, disp, etype,
        filetype, datarep, info);
825 826 827 828 829 830 831 832 833 834
    tm2 = darshan_wtime();
    if(ret == MPI_SUCCESS)
    {
        CP_LOCK();
        file = darshan_file_by_fh(fh);
        if(file)
        {
            CP_INC(file, CP_VIEWS, 1);
            if(info != MPI_INFO_NULL)
            {
835
                CP_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_mpi_meta_end, CP_F_MPI_META_TIME);
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
                CP_INC(file, CP_HINTS, 1);
            }
            CP_DATATYPE_INC(file, filetype);
        }
        CP_UNLOCK();
    }

    return(ret);
}

int MPI_File_read(MPI_File fh, void *buf, int count, 
    MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
853
    ret = DARSHAN_MPI_CALL(PMPI_File_read)(fh, buf, count, datatype, status);
854 855 856 857 858 859 860 861 862 863 864 865 866 867
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_INDEP_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_at(MPI_File fh, MPI_Offset offset, void *buf,
    int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
868 869
    ret = DARSHAN_MPI_CALL(PMPI_File_read_at)(fh, offset, buf,
        count, datatype, status);
870 871 872 873 874 875 876 877 878 879 880 881 882 883
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_INDEP_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_at_all(MPI_File fh, MPI_Offset offset, void * buf,
    int count, MPI_Datatype datatype, MPI_Status * status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
884 885
    ret = DARSHAN_MPI_CALL(PMPI_File_read_at_all)(fh, offset, buf,
        count, datatype, status);
886 887 888 889 890 891 892 893 894 895 896 897 898
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_COLL_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_all(MPI_File fh, void * buf, int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
899 900
    ret = DARSHAN_MPI_CALL(PMPI_File_read_all)(fh, buf, count,
        datatype, status);
901 902 903 904 905 906 907 908 909 910 911 912 913
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_COLL_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_shared(MPI_File fh, void * buf, int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
914 915
    ret = DARSHAN_MPI_CALL(PMPI_File_read_shared)(fh, buf, count,
        datatype, status);
916 917 918 919 920 921 922 923 924 925 926 927 928 929
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_INDEP_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_ordered(MPI_File fh, void * buf, int count, 
    MPI_Datatype datatype, MPI_Status * status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
930 931
    ret = DARSHAN_MPI_CALL(PMPI_File_read_ordered)(fh, buf, count,
        datatype, status);
932 933 934 935 936 937 938 939 940 941 942 943 944 945
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_COLL_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_at_all_begin(MPI_File fh, MPI_Offset offset, void * buf,
    int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
946 947
    ret = DARSHAN_MPI_CALL(PMPI_File_read_at_all_begin)(fh, offset, buf,
        count, datatype);
948 949 950 951 952 953 954 955 956 957 958 959 960
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_SPLIT_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_all_begin(MPI_File fh, void * buf, int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
961
    ret = DARSHAN_MPI_CALL(PMPI_File_read_all_begin)(fh, buf, count, datatype);
962 963 964 965 966 967 968 969 970 971 972 973 974
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_SPLIT_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_read_ordered_begin(MPI_File fh, void * buf, int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
975 976
    ret = DARSHAN_MPI_CALL(PMPI_File_read_ordered_begin)(fh, buf, count,
        datatype);
977 978 979 980 981 982 983 984
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_SPLIT_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_iread_at(MPI_File fh, MPI_Offset offset, void * buf,
985
    int count, MPI_Datatype datatype, __D_MPI_REQUEST *request)
986 987 988 989 990
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
991 992
    ret = DARSHAN_MPI_CALL(PMPI_File_iread_at)(fh, offset, buf, count,
        datatype, request);
993 994 995 996 997 998 999
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_NB_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

1000
int MPI_File_iread(MPI_File fh, void * buf, int count, MPI_Datatype datatype, __D_MPI_REQUEST * request)
1001 1002 1003 1004 1005
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1006
    ret = DARSHAN_MPI_CALL(PMPI_File_iread)(fh, buf, count, datatype, request);
1007 1008 1009 1010 1011 1012 1013 1014
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_NB_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_iread_shared(MPI_File fh, void * buf, int count,
1015
    MPI_Datatype datatype, __D_MPI_REQUEST * request)
1016 1017 1018 1019 1020
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1021 1022
    ret = DARSHAN_MPI_CALL(PMPI_File_iread_shared)(fh, buf, count,
        datatype, request);
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_READ(ret, fh, count, datatype, CP_NB_READS, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}


int MPI_File_write(MPI_File fh, void *buf, int count, 
    MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1038
    ret = DARSHAN_MPI_CALL(PMPI_File_write)(fh, buf, count, datatype, status);
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_INDEP_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_at(MPI_File fh, MPI_Offset offset, void *buf,
    int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1053 1054
    ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(fh, offset, buf,
        count, datatype, status);
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_INDEP_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_at_all(MPI_File fh, MPI_Offset offset, void * buf,
    int count, MPI_Datatype datatype, MPI_Status * status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1069 1070
    ret = DARSHAN_MPI_CALL(PMPI_File_write_at_all)(fh, offset, buf,
        count, datatype, status);
1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_COLL_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_all(MPI_File fh, void * buf, int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1084 1085
    ret = DARSHAN_MPI_CALL(PMPI_File_write_all)(fh, buf, count,
        datatype, status);
1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_COLL_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_shared(MPI_File fh, void * buf, int count, MPI_Datatype datatype, MPI_Status *status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1099 1100
    ret = DARSHAN_MPI_CALL(PMPI_File_write_shared)(fh, buf, count,
        datatype, status);
1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_INDEP_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_ordered(MPI_File fh, void * buf, int count, 
    MPI_Datatype datatype, MPI_Status * status)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1115 1116
    ret = DARSHAN_MPI_CALL(PMPI_File_write_ordered)(fh, buf, count,
         datatype, status);
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_COLL_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_at_all_begin(MPI_File fh, MPI_Offset offset, void * buf,
    int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1131 1132
    ret = DARSHAN_MPI_CALL(PMPI_File_write_at_all_begin)(fh, offset,
        buf, count, datatype);
1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_SPLIT_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_all_begin(MPI_File fh, void * buf, int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1146
    ret = DARSHAN_MPI_CALL(PMPI_File_write_all_begin)(fh, buf, count, datatype);
1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_SPLIT_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_write_ordered_begin(MPI_File fh, void * buf, int count, MPI_Datatype datatype)
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1160 1161
    ret = DARSHAN_MPI_CALL(PMPI_File_write_ordered_begin)(fh, buf, count,
        datatype);
1162 1163 1164 1165 1166 1167 1168 1169
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_SPLIT_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_iwrite_at(MPI_File fh, MPI_Offset offset, void * buf,
1170
    int count, MPI_Datatype datatype, __D_MPI_REQUEST *request)
1171 1172 1173 1174 1175
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1176 1177
    ret = DARSHAN_MPI_CALL(PMPI_File_iwrite_at)(fh, offset, buf,
        count, datatype, request);
1178 1179 1180 1181 1182 1183 1184
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_NB_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

1185
int MPI_File_iwrite(MPI_File fh, void * buf, int count, MPI_Datatype datatype, __D_MPI_REQUEST * request)
1186 1187 1188 1189 1190
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1191
    ret = DARSHAN_MPI_CALL(PMPI_File_iwrite)(fh, buf, count, datatype, request);
1192 1193 1194 1195 1196 1197 1198 1199
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_NB_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

int MPI_File_iwrite_shared(MPI_File fh, void * buf, int count,
1200
    MPI_Datatype datatype, __D_MPI_REQUEST * request)
1201 1202 1203 1204 1205
{
    int ret;
    double tm1, tm2;

    tm1 = darshan_wtime();
1206 1207
    ret = DARSHAN_MPI_CALL(PMPI_File_iwrite_shared)(fh, buf, count,
        datatype, request);
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_MPI_WRITE(ret, fh, count, datatype, CP_NB_WRITES, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

/* cp_log_reduction()
 *
 * Identify shared files and reduce them to one log entry
 *
 * returns 0 on success, -1 on failure
 */
static int cp_log_reduction(struct darshan_job_runtime* final_job, int rank, 
1222
    char* logfile_name, MPI_Offset* next_offset)
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
{
    /* TODO: these need to be allocated differently now, too big */
    uint64_t hash_array[CP_MAX_FILES] = {0};
    int mask_array[CP_MAX_FILES] = {0};
    int all_mask_array[CP_MAX_FILES] = {0};
    int ret;
    int i;
    int j;
    MPI_Op reduce_op;
    MPI_Datatype rtype;
    struct darshan_file* tmp_array = NULL;
    int shared_count = 0;

    /* register a reduction operation */
1237
    ret = DARSHAN_MPI_CALL(PMPI_Op_create)(darshan_file_reduce, 1, &reduce_op); 
1238 1239 1240 1241 1242 1243 1244 1245
    if(ret != 0)
    {
        return(-1);
    }

    /* construct a datatype for a file record.  This is serving no purpose
     * except to make sure we can do a reduction on proper boundaries
     */
1246 1247 1248
    DARSHAN_MPI_CALL(PMPI_Type_contiguous)(sizeof(struct darshan_file),
        MPI_BYTE, &rtype); 
    DARSHAN_MPI_CALL(PMPI_Type_commit)(&rtype); 
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259

    /* gather list of files that root process has opened */
    if(rank == 0)
    {
        for(i=0; i<final_job->file_count; i++)
        {
            hash_array[i] = final_job->file_array[i].hash;
        }
    }

    /* broadcast list of files to all other processes */
1260 1261
    ret = DARSHAN_MPI_CALL(PMPI_Bcast)(hash_array,
        (CP_MAX_FILES * sizeof(uint64_t)), 
1262 1263 1264
        MPI_BYTE, 0, MPI_COMM_WORLD);
    if(ret != 0)
    {
1265
        DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1266
        DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
        return(-1);
    }

    /* everyone looks to see if they have also opened that same file */
    for(i=0; (i<CP_MAX_FILES && hash_array[i] != 0); i++)
    {
        for(j=0; j<final_job->file_count; j++)
        {
            if(hash_array[i] && final_job->file_array[j].hash == hash_array[i])
            {
                /* we opened that file too */
                mask_array[i] = 1;
                break;
            }
        }
    }

    /* now allreduce so that everyone agrees on which files are shared */
1285 1286
    ret = DARSHAN_MPI_CALL(PMPI_Allreduce)(mask_array, all_mask_array,
        CP_MAX_FILES, MPI_INT, MPI_LAND, MPI_COMM_WORLD);
1287 1288
    if(ret != 0)
    {
1289
        DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1290
        DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
        return(-1);
    }

    /* walk through mask array counting entries and marking corresponding
     * files with a rank of -1
     */
    for(i=0; i<CP_MAX_FILES; i++)
    {
        if(all_mask_array[i])
        {
            shared_count++;
            for(j=0; j<final_job->file_count; j++)
            {
                if(final_job->file_array[j].hash == hash_array[i])
                {
1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330

                    /*
                     * Initialize fastest/slowest info prior
                     * to the reduction.
                     */
                    final_job->file_array[j].counters[CP_FASTEST_RANK] =
                      final_job->file_array[j].rank;
                    final_job->file_array[j].counters[CP_FASTEST_RANK_BYTES] =
                      final_job->file_array[j].counters[CP_BYTES_READ] +
                      final_job->file_array[j].counters[CP_BYTES_WRITTEN];
                    final_job->file_array[j].fcounters[CP_F_FASTEST_RANK_TIME] =
                      final_job->file_array[j].fcounters[CP_F_POSIX_META_TIME] +
                      final_job->file_array[j].fcounters[CP_F_POSIX_READ_TIME] +
                      final_job->file_array[j].fcounters[CP_F_POSIX_WRITE_TIME];

                    final_job->file_array[j].counters[CP_SLOWEST_RANK] =
                      final_job->file_array[j].rank;
                    final_job->file_array[j].counters[CP_SLOWEST_RANK_BYTES] =
                      final_job->file_array[j].counters[CP_BYTES_READ] +
                      final_job->file_array[j].counters[CP_BYTES_WRITTEN];
                    final_job->file_array[j].fcounters[CP_F_SLOWEST_RANK_TIME] =
                      final_job->file_array[j].fcounters[CP_F_POSIX_META_TIME] +
                      final_job->file_array[j].fcounters[CP_F_POSIX_READ_TIME] +
                      final_job->file_array[j].fcounters[CP_F_POSIX_WRITE_TIME];

1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346
                    final_job->file_array[j].rank = -1;
                    break;
                }
            }
        }
    }

    if(shared_count)
    {
        if(rank == 0)
        {
            /* root proc needs to allocate memory to store reduction */
            tmp_array = malloc(shared_count*sizeof(struct darshan_file));
            if(!tmp_array)
            {
                /* TODO: think more about how to handle errors like this */
1347
                DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1348
                DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359
                return(-1);
            }
        }

        /* sort the array of files descending by rank so that we get all of the 
         * shared files (marked by rank -1) in a contiguous portion at end 
         * of the array
         */
        qsort(final_job->file_array, final_job->file_count, 
            sizeof(struct darshan_file), file_compare);

1360
        ret = DARSHAN_MPI_CALL(PMPI_Reduce)(
1361 1362 1363 1364
            &final_job->file_array[final_job->file_count-shared_count], 
            tmp_array, shared_count, rtype, reduce_op, 0, MPI_COMM_WORLD);
        if(ret != 0)
        {
1365
            DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1366
            DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1367 1368 1369
            return(-1);
        }

1370 1371 1372 1373 1374
        ret = darshan_file_variance(
            &final_job->file_array[final_job->file_count-shared_count],
            tmp_array, shared_count, rank);
        if (ret)
        {
1375
            DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1376
            DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1377 1378 1379
            return(-1);
        }

1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394
        if(rank == 0)
        {
            /* root replaces local files with shared ones */
            memcpy(&final_job->file_array[final_job->file_count-shared_count],
                tmp_array, shared_count*sizeof(struct darshan_file));
            free(tmp_array);
            tmp_array = NULL;
        }
        else
        {
            /* everyone else simply discards those file records */
            final_job->file_count -= shared_count;
        }
    }
    
1395
    DARSHAN_MPI_CALL(PMPI_Op_free)(&reduce_op);
1396
    DARSHAN_MPI_CALL(PMPI_Type_free)(&rtype);
1397

1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427
    return(0);
}

/* TODO: should we use more of the CP macros here? */
static void darshan_file_reduce(void* infile_v, 
    void* inoutfile_v, int *len, 
    MPI_Datatype *datatype)
{
    struct darshan_file tmp_file;
    struct darshan_file* infile = infile_v;
    struct darshan_file* inoutfile = inoutfile_v;
    struct darshan_file_runtime tmp_runtime;
    int i;
    int j;
    int k;

    for(i=0; i<*len; i++)
    {
        memset(&tmp_file, 0, sizeof(tmp_file));

        tmp_file.hash = infile->hash;
        tmp_file.rank = -1; /* indicates shared across all procs */

        /* sum */
        for(j=CP_INDEP_OPENS; j<=CP_VIEWS; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + 
                inoutfile->counters[j];
        }

1428 1429 1430 1431 1432
        /* pick one, favoring complete records if available */
        if(CP_FILE_PARTIAL(infile))
            tmp_file.counters[CP_MODE] = inoutfile->counters[CP_MODE];
        else
            tmp_file.counters[CP_MODE] = infile->counters[CP_MODE];
1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457


        /* sum */
        for(j=CP_BYTES_READ; j<=CP_BYTES_WRITTEN; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + 
                inoutfile->counters[j];
        }

        /* max */
        for(j=CP_MAX_BYTE_READ; j<=CP_MAX_BYTE_WRITTEN; j++)
        {
            tmp_file.counters[j] = (
                (infile->counters[j] > inoutfile->counters[j]) ? 
                infile->counters[j] :
                inoutfile->counters[j]);
        }

        /* sum */
        for(j=CP_CONSEC_READS; j<=CP_MEM_NOT_ALIGNED; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + 
                inoutfile->counters[j];
        }

1458 1459 1460 1461 1462 1463
        /* pick one, favoring complete records if available */
        if(CP_FILE_PARTIAL(infile))
            tmp_file.counters[CP_MEM_ALIGNMENT] = inoutfile->counters[CP_MEM_ALIGNMENT];
        else
            tmp_file.counters[CP_MEM_ALIGNMENT] = infile->counters[CP_MEM_ALIGNMENT];

1464 1465 1466 1467 1468 1469 1470
        /* sum */
        for(j=CP_FILE_NOT_ALIGNED; j<=CP_FILE_NOT_ALIGNED; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + 
                inoutfile->counters[j];
        }

1471 1472 1473 1474 1475
        /* pick one, favoring complete records if available */
        if(CP_FILE_PARTIAL(infile))
            tmp_file.counters[CP_FILE_ALIGNMENT] = inoutfile->counters[CP_FILE_ALIGNMENT];
        else
            tmp_file.counters[CP_FILE_ALIGNMENT] = infile->counters[CP_FILE_ALIGNMENT];
1476
        
1477 1478
        /* skip CP_MAX_*_TIME_SIZE; handled in floating point section */

1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548
        /* sum */
        for(j=CP_SIZE_READ_0_100; j<=CP_EXTENT_WRITE_1G_PLUS; j++)
        {
            tmp_file.counters[j] = infile->counters[j] + 
                inoutfile->counters[j];
        }

        /* pick the 4 most common strides out of the 8 we have to chose from */

        /* first collapse any duplicates */
        for(j=CP_STRIDE1_STRIDE; j<=CP_STRIDE4_STRIDE; j++)
        {
            for(k=CP_STRIDE1_STRIDE; k<=CP_STRIDE4_STRIDE; k++)
            {
                if(infile->counters[j] == inoutfile->counters[k])
                {
                    infile->counters[j+4] += inoutfile->counters[k+4];
                    inoutfile->counters[k] = 0;
                    inoutfile->counters[k+4] = 0;
                }
            }
        }

        /* placeholder so we can re-use macros */
        tmp_runtime.log_file = &tmp_file;
        /* first set */
        for(j=CP_STRIDE1_STRIDE; j<=CP_STRIDE4_STRIDE; j++)
        {
            CP_COUNTER_INC(&tmp_runtime, infile->counters[j],
                infile->counters[j+4], 1, CP_STRIDE1_STRIDE, CP_STRIDE1_COUNT);
        }
        /* second set */
        for(j=CP_STRIDE1_STRIDE; j<=CP_STRIDE4_STRIDE; j++)
        {
            CP_COUNTER_INC(&tmp_runtime, inoutfile->counters[j],
                inoutfile->counters[j+4], 1, CP_STRIDE1_STRIDE, CP_STRIDE1_COUNT);
        }

        /* TODO: subroutine so we don't duplicate so much */
        /* same for access counts */

        /* first collapse any duplicates */
        for(j=CP_ACCESS1_ACCESS; j<=CP_ACCESS4_ACCESS; j++)
        {
            for(k=CP_ACCESS1_ACCESS; k<=CP_ACCESS4_ACCESS; k++)
            {
                if(infile->counters[j] == inoutfile->counters[k])
                {
                    infile->counters[j+4] += inoutfile->counters[k+4];
                    inoutfile->counters[k] = 0;
                    inoutfile->counters[k+4] = 0;
                }
            }
        }

        /* placeholder so we can re-use macros */
        tmp_runtime.log_file = &tmp_file;
        /* first set */
        for(j=CP_ACCESS1_ACCESS; j<=CP_ACCESS4_ACCESS; j++)
        {
            CP_COUNTER_INC(&tmp_runtime, infile->counters[j],
                infile->counters[j+4], 1, CP_ACCESS1_ACCESS, CP_ACCESS1_COUNT);
        }
        /* second set */
        for(j=CP_ACCESS1_ACCESS; j<=CP_ACCESS4_ACCESS; j++)
        {
            CP_COUNTER_INC(&tmp_runtime, inoutfile->counters[j],
                inoutfile->counters[j+4], 1, CP_ACCESS1_ACCESS, CP_ACCESS1_COUNT);
        }

1549
        /* min non-zero (if available) value */
1550 1551
        for(j=CP_F_OPEN_TIMESTAMP; j<=CP_F_WRITE_START_TIMESTAMP; j++)
        {
1552
            if(infile->fcounters[j] > inoutfile->fcounters[j] && inoutfile->fcounters[j] > 0)
1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573
                tmp_file.fcounters[j] = inoutfile->fcounters[j];
            else
                tmp_file.fcounters[j] = infile->fcounters[j];
        }

        /* max */
        for(j=CP_F_CLOSE_TIMESTAMP; j<=CP_F_WRITE_END_TIMESTAMP; j++)
        {
            if(infile->fcounters[j] > inoutfile->fcounters[j])
                tmp_file.fcounters[j] = infile->fcounters[j];
            else
                tmp_file.fcounters[j] = inoutfile->fcounters[j];
        }

        /* sum */
        for(j=CP_F_POSIX_READ_TIME; j<=CP_F_MPI_WRITE_TIME; j++)
        {
            tmp_file.fcounters[j] = infile->fcounters[j] + 
                inoutfile->fcounters[j];
        }

1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606
        /* max (special case) */
        if(infile->fcounters[CP_F_MAX_WRITE_TIME] > 
            inoutfile->fcounters[CP_F_MAX_WRITE_TIME])
        {
            tmp_file.fcounters[CP_F_MAX_WRITE_TIME] = 
                infile->fcounters[CP_F_MAX_WRITE_TIME];
            tmp_file.counters[CP_MAX_WRITE_TIME_SIZE] = 
                infile->counters[CP_MAX_WRITE_TIME_SIZE];
        }
        else
        {
            tmp_file.fcounters[CP_F_MAX_WRITE_TIME] = 
                inoutfile->fcounters[CP_F_MAX_WRITE_TIME];
            tmp_file.counters[CP_MAX_WRITE_TIME_SIZE] = 
                inoutfile->counters[CP_MAX_WRITE_TIME_SIZE];
        }

        if(infile->fcounters[CP_F_MAX_READ_TIME] > 
            inoutfile->fcounters[CP_F_MAX_READ_TIME])
        {
            tmp_file.fcounters[CP_F_MAX_READ_TIME] = 
                infile->fcounters[CP_F_MAX_READ_TIME];
            tmp_file.counters[CP_MAX_READ_TIME_SIZE] = 
                infile->counters[CP_MAX_READ_TIME_SIZE];
        }
        else
        {
            tmp_file.fcounters[CP_F_MAX_READ_TIME] = 
                inoutfile->fcounters[CP_F_MAX_READ_TIME];
            tmp_file.counters[CP_MAX_READ_TIME_SIZE] = 
                inoutfile->counters[CP_MAX_READ_TIME_SIZE];
        }

1607
        /* min (zeroes are ok here; some procs don't do I/O) */
1608 1609
        if(infile->fcounters[CP_F_FASTEST_RANK_TIME] <
           inoutfile->fcounters[CP_F_FASTEST_RANK_TIME])
1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648
        {
            tmp_file.counters[CP_FASTEST_RANK] =
                infile->counters[CP_FASTEST_RANK];
            tmp_file.counters[CP_FASTEST_RANK_BYTES] = 
                infile->counters[CP_FASTEST_RANK_BYTES];
            tmp_file.fcounters[CP_F_FASTEST_RANK_TIME] =
                infile->fcounters[CP_F_FASTEST_RANK_TIME];
        }
        else
        {
            tmp_file.counters[CP_FASTEST_RANK] =
                inoutfile->counters[CP_FASTEST_RANK];
            tmp_file.counters[CP_FASTEST_RANK_BYTES] =
                inoutfile->counters[CP_FASTEST_RANK_BYTES];
            tmp_file.fcounters[CP_F_FASTEST_RANK_TIME] = 
                inoutfile->fcounters[CP_F_FASTEST_RANK_TIME];
        }

        /* max */
        if(infile->fcounters[CP_F_SLOWEST_RANK_TIME] >
           inoutfile->fcounters[CP_F_SLOWEST_RANK_TIME])
        {
            tmp_file.counters[CP_SLOWEST_RANK] =
                infile->counters[CP_SLOWEST_RANK];
            tmp_file.counters[CP_SLOWEST_RANK_BYTES] =
                infile->counters[CP_SLOWEST_RANK_BYTES];
            tmp_file.fcounters[CP_F_SLOWEST_RANK_TIME] = 
                infile->fcounters[CP_F_SLOWEST_RANK_TIME];
        }
        else
        {
            tmp_file.counters[CP_SLOWEST_RANK] = 
                inoutfile->counters[CP_SLOWEST_RANK];
            tmp_file.counters[CP_SLOWEST_RANK_BYTES] = 
                inoutfile->counters[CP_SLOWEST_RANK_BYTES];
            tmp_file.fcounters[CP_F_SLOWEST_RANK_TIME] = 
                inoutfile->fcounters[CP_F_SLOWEST_RANK_TIME];
        }

1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
        /* pick one device id and file size, favoring complete records if
         * available
         */
        if(CP_FILE_PARTIAL(infile))
        {
            tmp_file.counters[CP_DEVICE] = inoutfile->counters[CP_DEVICE];
            tmp_file.counters[CP_SIZE_AT_OPEN] = inoutfile->counters[CP_SIZE_AT_OPEN];
        }
        else
        {
            tmp_file.counters[CP_DEVICE] = infile->counters[CP_DEVICE];
            tmp_file.counters[CP_SIZE_AT_OPEN] = infile->counters[CP_SIZE_AT_OPEN];
        }
1662

1663 1664 1665
        /* pick one name suffix (every file record should have this, whether
         * it is a partial record or not
         */
1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679
        strcpy(tmp_file.name_suffix, infile->name_suffix);

        *inoutfile = tmp_file;
        inoutfile++;
        infile++;
    }
    
    return;
}
/* cp_log_construct_indices()
 *
 * create memory datatypes to describe the log data to write out
 */
static void cp_log_construct_indices(struct darshan_job_runtime* final_job, 
1680 1681
    int rank, int* inout_count, int* lengths, void** pointers, char*
    trailing_data)
1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693
{
    *inout_count = 0;

    if(rank == 0)
    {
        /* root process is responsible for writing header */
        lengths[*inout_count] = sizeof(final_job->log_job);
        pointers[*inout_count] = &final_job->log_job;
        (*inout_count)++;

        /* also string containing exe command line */
        lengths[*inout_count] = CP_EXE_LEN + 1; 
1694
        pointers[*inout_count] = trailing_data;
1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713
        (*inout_count)++;
    }

    /* everyone adds their own file records, if present */
    if(final_job->file_count > 0)
    {
        lengths[*inout_count] = final_job->file_count*CP_FILE_RECORD_SIZE;
        pointers[*inout_count] = final_job->file_array;
        (*inout_count)++;
    }
    
    return;
}

/* cp_log_write()
 *
 * actually write log information to disk
 */
static int cp_log_write(struct darshan_job_runtime* final_job, int rank, 
1714
    char* logfile_name, int count, int* lengths, void** pointers, double start_log_time)
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726
{
    int ret;
    MPI_File fh;
    MPI_Status status;
    MPI_Datatype mtype;
    int my_total = 0;
    long my_total_long;
    long offset;
    int i;
    MPI_Aint displacements[CP_MAX_MEM_SEGMENTS];
    void* buf;
    int failed_write = 0;
1727
    char* hints;
1728 1729 1730 1731 1732
    char* key;
    char* value;
    char* tok_str;
    char* saveptr = NULL;
    MPI_Info info;
1733

1734 1735
    /* skip building a datatype if we don't have anything to write */
    if(count > 0)
1736
    {
1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752
        /* construct data type to describe everything we are writing */
        /* NOTE: there may be a bug in MPI-IO when using MPI_BOTTOM with an
         * hindexed data type.  We will instead use the first pointer as a base
         * and adjust the displacements relative to it.
         */
        buf = pointers[0];
        for(i=0; i<count; i++)
        {
            /* use this transform to be compiler safe */
            uintptr_t ptr  = (uintptr_t) pointers[i];
            uintptr_t base = (uintptr_t) buf;
            displacements[i] = (MPI_Aint)(ptr - base);
        }
        DARSHAN_MPI_CALL(PMPI_Type_hindexed)(count, lengths, displacements,
            MPI_BYTE, &mtype);
        DARSHAN_MPI_CALL(PMPI_Type_commit)(&mtype); 
1753
    }
1754 1755
    
    MPI_Info_create(&info);
1756

1757 1758 1759 1760 1761 1762 1763 1764
    /* check environment variable to see if the default MPI file hints have
     * been overridden
     */
    hints = getenv(CP_LOG_HINTS_OVERRIDE);
    if(!hints)
    {
        hints = __CP_LOG_HINTS;
    }
1765

1766
    if(hints && strlen(hints) > 0)
1767
    {
1768 1769
        tok_str = strdup(hints);
        if(tok_str)
1770
        {
1771
            do
1772
            {
1773 1774
                /* split string on semicolon */
                key = strtok_r(tok_str, ";", &saveptr);
1775
                if(key)
1776
                {
1777 1778 1779 1780 1781 1782 1783 1784
                    tok_str = NULL;
                    /* look for = sign splitting key/value pairs */
                    value = index(key, '=');
                    if(value)
                    {
                        /* break key and value into separate null terminated strings */
                        value[0] = '\0';
                        value++;
1785 1786
                        if(strlen(key) > 0)
                            MPI_Info_set(info, key, value);
1787
                    }
1788
                }
1789 1790 1791
            }while(key != NULL);
            free(tok_str);
        }
1792
    }
1793
    
1794 1795
    ret = DARSHAN_MPI_CALL(PMPI_File_open)(MPI_COMM_WORLD, logfile_name,
        MPI_MODE_CREATE | MPI_MODE_WRONLY | MPI_MODE_EXCL, MPI_INFO_NULL, &fh);
1796
    MPI_Info_free(&info);
1797 1798
    if(ret != MPI_SUCCESS)
    {
1799 1800
        if(rank == 0)
        {
1801 1802 1803 1804 1805
            int msg_len;
            char msg[MPI_MAX_ERROR_STRING] = {0};
            
            MPI_Error_string(ret, msg, &msg_len);
            fprintf(stderr, "darshan library warning: unable to open log file %s: %s\n", logfile_name, msg);
1806
        }
1807 1808
        if(count > 0)
            DARSHAN_MPI_CALL(PMPI_Type_free)(&mtype);
1809 1810
        return(-1);
    }
1811
  
1812
    /* figure out where everyone is writing */
1813 1814 1815 1816
    if(count > 0)
        DARSHAN_MPI_CALL(PMPI_Type_size)(mtype, &my_total);
    else
        my_total = 0;
1817
    my_total_long = my_total;
1818 1819
    DARSHAN_MPI_CALL(PMPI_Scan)(&my_total_long, &offset, 1,
        MPI_LONG, MPI_SUM, MPI_COMM_WORLD); 
1820 1821 1822
    /* scan is inclusive; subtract local size back out */
    offset -= my_total_long;

1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833
    if(count > 0)
    {
        /* collectively write out file records from all processes */
        ret = DARSHAN_MPI_CALL(PMPI_File_write_at_all)(fh, offset, buf, 
            1, mtype, &status);
        if(ret != MPI_SUCCESS)
        {
            failed_write = 1;
        }
    }
    else
1834
    {
1835 1836 1837 1838 1839 1840 1841 1842 1843
        /* nothing to write, but we need to participate in the 
         * collectivee anyway 
         */
        ret = DARSHAN_MPI_CALL(PMPI_File_write_at_all)(fh, offset, NULL, 
            0, MPI_BYTE, &status);
        if(ret != MPI_SUCCESS)
        {
            failed_write = 1;
        }