codes-workload-dump.c 15.1 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
/*
 * Copyright (C) 2014 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */

#include <assert.h>
#include <getopt.h>
#include <stdio.h>
#include "codes/codes-workload.h"
11
#include <inttypes.h>
12
13

static char type[128] = {'\0'};
14
static darshan_params d_params = {"", 0}; 
15
static iolang_params i_params = {0, 0, "", ""};
16
static recorder_params r_params = {"", 0};
17
18
static dumpi_trace_params du_params = {"", 0};
static checkpoint_wrkld_params c_params = {0, 0, 0, 0, 0};
19
static int n = -1;
20
static int start_rank = 0;
21
22
23
24
25

static struct option long_opts[] = 
{
    {"type", required_argument, NULL, 't'},
    {"num-ranks", required_argument, NULL, 'n'},
26
    {"start-rank", required_argument, NULL, 'r'},
27
28
    {"d-log", required_argument, NULL, 'l'},
    {"d-aggregator-cnt", required_argument, NULL, 'a'},
29
    {"i-meta", required_argument, NULL, 'm'},
30
    {"i-use-relpath", no_argument, NULL, 'p'},
31
    {"r-trace-dir", required_argument, NULL, 'd'},
32
    {"r-nprocs", required_argument, NULL, 'x'},
33
    {"dumpi-log", required_argument, NULL, 'w'},
34
35
36
37
    {"chkpoint-size", required_argument, NULL, 'S'},
    {"chkpoint-bw", required_argument, NULL, 'B'},
    {"chkpoint-runtime", required_argument, NULL, 'R'},
    {"chkpoint-mtti", required_argument, NULL, 'M'},
38
39
40
    {NULL, 0, NULL, 0}
};

41
42
void usage(){
    fprintf(stderr,
43
            "Usage: codes-workload-dump --type TYPE --num-ranks N [OPTION...]\n"
44
            "--type: type of workload (\"darshan_io_workload\", \"iolang_workload\", dumpi-trace-workload\" etc.)\n"
45
            "--num-ranks: number of ranks to process (if not set, it is set by the workload)\n"
46
            "-s: print final workload stats\n"
47
            "DARSHAN OPTIONS (darshan_io_workload)\n"
48
            "--d-log: darshan log file\n"
49
            "--d-aggregator-cnt: number of aggregators for collective I/O in darshan\n"
50
            "IOLANG OPTIONS (iolang_workload)\n"
51
            "--i-meta: i/o language kernel meta file path\n"
52
            "--i-use-relpath: use i/o kernel path relative meta file path\n"
53
            "RECORDER OPTIONS (recorder_io_workload)\n"
54
            "--r-trace-dir: directory containing recorder trace files\n"
55
            "--r-nprocs: number of ranks in original recorder workload\n"
56
57
58
59
60
61
62
            "DUMPI TRACE OPTIONS (dumpi-trace-workload) \n"
            "--dumpi-log: dumpi log file \n"
            "CHECKPOINT OPTIONS (checkpoint_io_workload)\n"
            "--chkpoint-size: size of aggregate checkpoint to write\n"
            "--chkpoint-bw: checkpointing bandwidth\n"
            "--chkpoint-runtime: desired application runtime\n"
            "--chkpoint-mtti: mean time to interrupt\n");
63
}
64
65
66

int main(int argc, char *argv[])
{
67
68
69
70
    int print_stats = 0;
    double total_delay = 0.0;
    int64_t num_barriers = 0;
    int64_t num_opens = 0;
71
    int64_t num_closes = 0;
72
73
74
75
    int64_t num_reads = 0;
    int64_t read_size = 0;
    int64_t num_writes = 0;
    int64_t write_size = 0;
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
    int64_t num_sends = 0;
    int64_t send_size = 0;
    int64_t num_recvs = 0;
    int64_t recv_size = 0;
    int64_t num_isends = 0;
    int64_t isend_size = 0;
    int64_t num_irecvs = 0;
    int64_t irecv_size = 0;
    int64_t num_bcasts = 0;
    int64_t bcast_size = 0;
    int64_t num_allgathers = 0;
    int64_t allgather_size = 0;
    int64_t num_allgathervs = 0;
    int64_t allgatherv_size = 0;
    int64_t num_alltoalls = 0;
    int64_t alltoall_size = 0;
    int64_t num_alltoallvs = 0;
    int64_t alltoallv_size = 0;
    int64_t num_reduces = 0;
    int64_t reduce_size = 0;
    int64_t num_allreduces = 0;
    int64_t allreduce_size = 0;
    int64_t num_collectives = 0;
    int64_t collective_size = 0;
    int64_t num_waitalls = 0;
    int64_t num_waits = 0;
    int64_t num_waitsomes = 0;
    int64_t num_waitanys = 0;
    int64_t num_testalls = 0;
105

106
    char ch;
107
    while ((ch = getopt_long(argc, argv, "t:n:l:a:m:sp:wr:", long_opts, NULL)) != -1){
108
109
110
111
112
113
114
115
116
117
118
119
120
121
        switch (ch){
            case 't':
                strcpy(type, optarg);
                break;
            case 'n':
                n = atoi(optarg);
                assert(n>0);
                break;
            case 'l':
                strcpy(d_params.log_file_path, optarg);
                break;
            case 'a':
                d_params.aggregator_cnt = atol(optarg);
                break;
122
            case 'm':
123
                strcpy(i_params.io_kernel_meta_path, optarg);
124
                break;
125
            case 'p':
126
                i_params.use_relpath = 1;
127
                break;
128
129
130
            case 'd':
                strcpy(r_params.trace_dir_path, optarg);
                break;
131
132
133
            case 'x':
                r_params.nprocs = atol(optarg);
                break;
134
135
136
            case 'w':
                strcpy(du_params.file_name, optarg);
                break;
137
138
139
            case 's':
                print_stats = 1;
                break;
140
141
142
143
            case 'r':
                start_rank = atoi(optarg);
                assert(n>0);
                break;
144
145
146
147
148
149
150
151
152
153
154
155
            case 'S':
                c_params.checkpoint_sz = atof(optarg);
                break;
            case 'B':
                c_params.checkpoint_wr_bw = atof(optarg);
                break;
            case 'R':
                c_params.app_runtime = atof(optarg);
                break;
            case 'M':
                c_params.mtti = atof(optarg);
                break;
156
157
158
        }
    }

159
160
161
162
163
164
    if (type[0] == '\0'){
        fprintf(stderr, "Expected \"--type\" argument\n");
        usage();
        return 1;
    }

165
166
167
    int i;
    char *wparams;
    if (strcmp(type, "darshan_io_workload") == 0){
168
169
170
171
172
173
174
175
176
177
178
179
180
        if (d_params.log_file_path[0] == '\0'){
            fprintf(stderr, "Expected \"--d-log\" argument for darshan workload\n");
            usage();
            return 1;
        }
        else if (d_params.aggregator_cnt == 0){
            fprintf(stderr, "Expected \"--d-aggregator-cnt\" argument for darshan workload\n");
            usage();
            return 1;
        }
        else{
            wparams = (char*)&d_params;
        }
181
    }
182
    else if (strcmp(type, "iolang_workload") == 0){
183
184
185
186
187
188
189
190
        if (n == -1){
            fprintf(stderr,
                    "Expected \"--num-ranks\" argument for iolang workload\n");
            usage();
            return 1;
        }
        else{
            i_params.num_cns = n;
191
        }
192
193
194
        if (i_params.io_kernel_meta_path[0] == '\0'){
            fprintf(stderr,
                    "Expected \"--i-meta\" argument for iolang workload\n");
195
196
197
            usage();
            return 1;
        }
198

199
        wparams = (char *)&i_params;
200
    }
201
202
203
204
205
206
    else if (strcmp(type, "recorder_io_workload") == 0){
        if (r_params.trace_dir_path[0] == '\0'){
            fprintf(stderr, "Expected \"--r-trace-dir\" argument for recorder workload\n");
            usage();
            return 1;
        }
207
208
209
210
211
        if (r_params.nprocs == 0){
            fprintf(stderr, "Expected \"--r-nprocs\" argument for recorder workload\n");
            usage();
            return 1;
        }
212
213
214
215
        else{
            wparams = (char *)&r_params;
        }
    }
216
217
218
219
   else if(strcmp(type, "dumpi-trace-workload") == 0)
	{
	if(n == -1){
            fprintf(stderr,
Jonathan Jenkins's avatar
typo    
Jonathan Jenkins committed
220
                    "Expected \"--num-ranks\" argument for dumpi workload\n");
221
222
223
224
225
226
227
228
            usage();
            return 1;		
	}
	else{
	    du_params.num_net_traces = n;	
	}

	if(du_params.file_name[0] == '\0' ){
229
            fprintf(stderr, "Expected \"--dumpi-log\" argument for dumpi workload\n");
230
231
232
233
234
235
236
237
            usage();
            return 1;
	}
	else
	{
	  wparams = (char*)&du_params;
	}
	}
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
    else if(strcmp(type, "checkpoint_io_workload") == 0)
    {
        if(c_params.checkpoint_sz == 0 || c_params.checkpoint_wr_bw == 0 ||
           c_params.app_runtime == 0 || c_params.mtti == 0)
        {
            fprintf(stderr, "All checkpoint workload arguments are required\n");
            usage();
            return 1;
        }
        else
        {
            c_params.nprocs = n;
            wparams = (char *)&c_params;
        }
    }
253
    else {
254
255
256
        fprintf(stderr, "Invalid type argument\n");
        usage();
        return 1;
257
    }
258
259
260

    /* if num_ranks not set, pull it from the workload */
    if (n == -1){
261
        n = codes_workload_get_rank_cnt(type, wparams, 0);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
262
263
264
265
266
267
        if (n == -1) {
            fprintf(stderr,
                    "Unable to get rank count from workload. "
                    "Specify option --num-ranks\n");
            return 1;
        }
268
269
    }

270
    for (i = start_rank ; i < start_rank+n; i++){
271
272
        struct codes_workload_op op;
        printf("loading %s, %d\n", type, i);
273
        int id = codes_workload_load(type, wparams, 0, i);
274
275
        assert(id != -1);
        do {
276
277
            codes_workload_get_next(id, 0, i, &op);
            codes_workload_print_op(stdout, &op, 0, i);
278
279
280

            switch(op.op_type)
            {
281
282
                case CODES_WK_DELAY:
                    total_delay += op.u.delay.seconds;
283
284
285
286
                    break;
                case CODES_WK_BARRIER:
                    num_barriers++;
                    break;
287
288
                case CODES_WK_OPEN:
                    num_opens++;
289
                    break;
290
291
                case CODES_WK_CLOSE:
                    num_closes++;
292
293
294
295
296
                    break;
                case CODES_WK_WRITE:
                    num_writes++;
                    write_size += op.u.write.size;
                    break;
297
298
299
300
301
302
303
                case CODES_WK_READ:
                    num_reads++;
                    read_size += op.u.write.size;
                    break;
                case CODES_WK_SEND:
                    num_sends++;
                    send_size += op.u.send.num_bytes;
304
                    break;
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
                case CODES_WK_RECV:
                    num_recvs++;
                    recv_size += op.u.recv.num_bytes;
                    break;
                case CODES_WK_ISEND:
                    num_irecvs++;
                    irecv_size += op.u.send.num_bytes;
                    break;
                case CODES_WK_IRECV:
                    num_irecvs++;
                    irecv_size += op.u.recv.num_bytes;
                    break;
                /* NOTE: all collectives are currently represented as the
                 * generic "collective" type */
                case CODES_WK_BCAST:
                    num_bcasts++;
                    bcast_size += op.u.collective.num_bytes;
322
                    break;
323
                case CODES_WK_ALLGATHER:
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
                    num_allgathers++;
                    allgather_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_ALLGATHERV:
                    num_allgathervs++;
                    allgatherv_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_ALLTOALL:
                    num_alltoalls++;
                    alltoall_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_ALLTOALLV:
                    num_alltoallvs++;
                    alltoallv_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_REDUCE:
                    num_reduces++;
                    reduce_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_ALLREDUCE:
                    num_allreduces++;
                    allreduce_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_COL:
                    num_collectives++;
                    collective_size += op.u.collective.num_bytes;
                    break;
                case CODES_WK_WAITALL:
                    num_waitalls++;
                    break;
                case CODES_WK_WAIT:
                    num_waits++;
                    break;
                case CODES_WK_WAITSOME:
                    num_waitsomes++;
                    break;
                case CODES_WK_WAITANY:
                    num_waitanys++;
                    break;
                case CODES_WK_TESTALL:
                    num_testalls++;
                    break;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
366
367
368
369
                case CODES_WK_END:
                    break;
                case CODES_WK_IGNORE:
                    break;
370
371
372
373
                default:
                    fprintf(stderr,
                            "WARNING: unknown workload op type (code %d)\n",
                            op.op_type);
374
            }
375
376
377
        } while (op.op_type != CODES_WK_END);
    }

378
379
380
    if (print_stats)
    {
        fprintf(stderr, "\n* * * * * FINAL STATS * * * * * *\n");
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
        fprintf(stderr, "NUM_OPENS:       %"PRId64"\n", num_opens);
        fprintf(stderr, "NUM_CLOSES:      %"PRId64"\n", num_closes);
        fprintf(stderr, "NUM_BARRIERS:    %"PRId64"\n", num_barriers);
        fprintf(stderr, "TOTAL_DELAY:     %.4lf\n", total_delay);
        fprintf(stderr, "NUM_READS:       %"PRId64"\n", num_reads);
        fprintf(stderr, "READ_SIZE:       %"PRId64"\n", read_size);
        fprintf(stderr, "NUM_WRITES:      %"PRId64"\n", num_writes);
        fprintf(stderr, "WRITE_SIZE:      %"PRId64"\n", write_size);
        fprintf(stderr, "NUM_SENDS:       %"PRId64"\n", num_sends);
        fprintf(stderr, "SEND_SIZE:       %"PRId64"\n", send_size);
        fprintf(stderr, "NUM_RECVS:       %"PRId64"\n", num_recvs);
        fprintf(stderr, "RECV_SIZE:       %"PRId64"\n", recv_size);
        fprintf(stderr, "NUM_ISENDS:      %"PRId64"\n", num_isends);
        fprintf(stderr, "ISEND_SIZE:      %"PRId64"\n", isend_size);
        fprintf(stderr, "NUM_IRECVS:      %"PRId64"\n", num_irecvs);
        fprintf(stderr, "IRECV_SIZE:      %"PRId64"\n", irecv_size);
        fprintf(stderr, "NUM_BCASTS:      %"PRId64"\n", num_bcasts);
        fprintf(stderr, "BCAST_SIZE:      %"PRId64"\n", bcast_size);
        fprintf(stderr, "NUM_ALLGATHERS:  %"PRId64"\n", num_allgathers);
        fprintf(stderr, "ALLGATHER_SIZE:  %"PRId64"\n", allgather_size);
        fprintf(stderr, "NUM_ALLGATHERVS: %"PRId64"\n", num_allgathervs);
        fprintf(stderr, "ALLGATHERV_SIZE: %"PRId64"\n", allgatherv_size);
        fprintf(stderr, "NUM_ALLTOALLS:   %"PRId64"\n", num_alltoalls);
        fprintf(stderr, "ALLTOALL_SIZE:   %"PRId64"\n", alltoall_size);
        fprintf(stderr, "NUM_ALLTOALLVS:  %"PRId64"\n", num_alltoallvs);
        fprintf(stderr, "ALLTOALLV_SIZE:  %"PRId64"\n", alltoallv_size);
        fprintf(stderr, "NUM_REDUCES:     %"PRId64"\n", num_reduces);
        fprintf(stderr, "REDUCE_SIZE:     %"PRId64"\n", reduce_size);
        fprintf(stderr, "NUM_ALLREDUCE:   %"PRId64"\n", num_allreduces);
        fprintf(stderr, "ALLREDUCE_SIZE:  %"PRId64"\n", allreduce_size);
        fprintf(stderr, "NUM_COLLECTIVE:  %"PRId64"\n", num_collectives);
        fprintf(stderr, "COLLECTIVE_SIZE: %"PRId64"\n", collective_size);
        fprintf(stderr, "NUM_WAITALLS:    %"PRId64"\n", num_waitalls);
        fprintf(stderr, "NUM_WAITS:       %"PRId64"\n", num_waits);
        fprintf(stderr, "NUM_WAITSOMES:   %"PRId64"\n", num_waitsomes);
        fprintf(stderr, "NUM_WAITANYS:    %"PRId64"\n", num_waitanys);
        fprintf(stderr, "NUM_TESTALLS:    %"PRId64"\n", num_testalls);
418
    }
419
420
421
422
423
424
425
426
427
428
429
430

    return 0;
}

/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */