1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
|
/*
* Copyright (c) 2013-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2013-2017 Inria. All rights reserved.
* Copyright (c) 2015 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2016 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
pml monitoring tester.
Designed by George Bosilca <bosilca@icl.utk.edu> Emmanuel Jeannot <emmanuel.jeannot@inria.fr> and Clément Foyer <clement.foyer@inria.fr>
Contact the authors for questions.
To options are available for this test, with/without MPI_Tools, and with/without RMA operations. The default mode is without MPI_Tools, and with RMA operations.
To enable the MPI_Tools use, add "--with-mpit" as an application parameter.
To disable the RMA operations testing, add "--without-rma" as an application parameter.
To be run as (without using MPI_Tool):
mpirun -n 4 --mca pml_monitoring_enable 2 --mca pml_monitoring_enable_output 3 --mca pml_monitoring_filename prof/output ./monitoring_test
with the results being, as an example:
output.1.prof
# POINT TO POINT
E 1 2 104 bytes 26 msgs sent 0,0,0,26,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
E 1 3 208 bytes 52 msgs sent 8,0,0,65,1,5,2,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
I 1 0 140 bytes 27 msgs sent
I 1 2 2068 bytes 1 msgs sent
I 1 3 2256 bytes 31 msgs sent
# OSC
S 1 0 0 bytes 1 msgs sent
R 1 0 40960 bytes 1 msgs sent
S 1 2 40960 bytes 1 msgs sent
# COLLECTIVES
C 1 0 140 bytes 27 msgs sent
C 1 2 140 bytes 27 msgs sent
C 1 3 140 bytes 27 msgs sent
D MPI COMMUNICATOR 4 DUP FROM 0 procs: 0,1,2,3
O2A 1 0 bytes 0 msgs sent
A2O 1 0 bytes 0 msgs sent
A2A 1 276 bytes 15 msgs sent
D MPI_COMM_WORLD procs: 0,1,2,3
O2A 1 0 bytes 0 msgs sent
A2O 1 0 bytes 0 msgs sent
A2A 1 96 bytes 9 msgs sent
D MPI COMMUNICATOR 5 SPLIT_TYPE FROM 4 procs: 0,1,2,3
O2A 1 0 bytes 0 msgs sent
A2O 1 0 bytes 0 msgs sent
A2A 1 48 bytes 3 msgs sent
D MPI COMMUNICATOR 3 SPLIT FROM 0 procs: 1,3
O2A 1 0 bytes 0 msgs sent
A2O 1 0 bytes 0 msgs sent
A2A 1 0 bytes 0 msgs sent
*/
#include "mpi.h"
#include <stdio.h>
#include <string.h>
static MPI_T_pvar_handle flush_handle;
static const char flush_pvar_name[] = "pml_monitoring_flush";
static const void*nullbuf = NULL;
static int flush_pvar_idx;
static int with_mpit = 0;
static int with_rma = 1;
int main(int argc, char* argv[])
{
int rank, size, n, to, from, tagno, MPIT_result, provided, count, world_rank;
MPI_T_pvar_session session;
MPI_Comm newcomm;
MPI_Comm comm = MPI_COMM_WORLD;
char filename[1024];
for ( int arg_it = 1; argc > 1 && arg_it < argc; ++arg_it ) {
if( 0 == strcmp(argv[arg_it], "--with-mpit") ) {
with_mpit = 1;
printf("enable MPIT support\n");
} else if( 0 == strcmp(argv[arg_it], "--without-rma") ) {
with_rma = 0;
printf("disable RMA testing\n");
}
}
/* first phase : make a token circulated in MPI_COMM_WORLD */
n = -1;
MPI_Init(NULL, NULL);
MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
rank = world_rank;
to = (rank + 1) % size;
from = (rank - 1) % size;
tagno = 201;
if( with_mpit ) {
MPIT_result = MPI_T_init_thread(MPI_THREAD_SINGLE, &provided);
if (MPIT_result != MPI_SUCCESS)
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
MPIT_result = MPI_T_pvar_get_index(flush_pvar_name, MPI_T_PVAR_CLASS_GENERIC, &flush_pvar_idx);
if (MPIT_result != MPI_SUCCESS) {
printf("cannot find monitoring MPI_T \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_session_create(&session);
if (MPIT_result != MPI_SUCCESS) {
printf("cannot create a session for \"%s\" pvar\n", flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
/* Allocating a new PVAR in a session will reset the counters */
MPIT_result = MPI_T_pvar_handle_alloc(session, flush_pvar_idx,
&comm, &flush_handle, &count);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to allocate handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_start(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to start handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
}
if (rank == 0) {
n = 25;
MPI_Send(&n,1,MPI_INT,to,tagno,MPI_COMM_WORLD);
}
while (1) {
MPI_Recv(&n, 1, MPI_INT, from, tagno, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
if (rank == 0) {n--;tagno++;}
MPI_Send(&n, 1, MPI_INT, to, tagno, MPI_COMM_WORLD);
if (rank != 0) {n--;tagno++;}
if (n<0){
break;
}
}
if( with_mpit ) {
/* Build one file per processes
Every thing that has been monitored by each
process since the last flush will be output in filename */
/*
Requires directory prof to be created.
Filename format should display the phase number
and the process rank for ease of parsing with
aggregate_profile.pl script
*/
sprintf(filename, "prof/phase_1");
if( MPI_SUCCESS != MPI_T_pvar_write(session, flush_handle, filename) ) {
fprintf(stderr, "Process %d cannot save monitoring in %s.%d.prof\n",
world_rank, filename, world_rank);
}
/* Force the writing of the monitoring data */
MPIT_result = MPI_T_pvar_stop(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to stop handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_start(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to start handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
/* Don't set a filename. If we stop the session before setting it, then no output file
* will be generated.
*/
if( MPI_SUCCESS != MPI_T_pvar_write(session, flush_handle, (void*)&nullbuf) ) {
fprintf(stderr, "Process %d cannot save monitoring in %s\n", world_rank, filename);
}
}
/*
Second phase. Work with different communicators.
even ranks will circulate a token
while odd ranks will perform a all_to_all
*/
MPI_Comm_split(MPI_COMM_WORLD, rank%2, rank, &newcomm);
if(rank%2){ /*odd ranks (in COMM_WORD) circulate a token*/
MPI_Comm_rank(newcomm, &rank);
MPI_Comm_size(newcomm, &size);
if( size > 1 ) {
to = (rank + 1) % size;
from = (rank - 1) % size;
tagno = 201;
if (rank == 0){
n = 50;
MPI_Send(&n, 1, MPI_INT, to, tagno, newcomm);
}
while (1){
MPI_Recv(&n, 1, MPI_INT, from, tagno, newcomm, MPI_STATUS_IGNORE);
if (rank == 0) {n--; tagno++;}
MPI_Send(&n, 1, MPI_INT, to, tagno, newcomm);
if (rank != 0) {n--; tagno++;}
if (n<0){
break;
}
}
}
} else { /*even ranks (in COMM_WORD) will perform a all_to_all and a barrier*/
int send_buff[10240];
int recv_buff[10240];
MPI_Comm newcomm2;
MPI_Comm_rank(newcomm, &rank);
MPI_Comm_size(newcomm, &size);
MPI_Alltoall(send_buff, 10240/size, MPI_INT, recv_buff, 10240/size, MPI_INT, newcomm);
MPI_Comm_split(newcomm, rank%2, rank, &newcomm2);
MPI_Barrier(newcomm2);
MPI_Comm_free(&newcomm2);
}
if( with_mpit ) {
/* Build one file per processes
Every thing that has been monitored by each
process since the last flush will be output in filename */
/*
Requires directory prof to be created.
Filename format should display the phase number
and the process rank for ease of parsing with
aggregate_profile.pl script
*/
sprintf(filename, "prof/phase_2");
if( MPI_SUCCESS != MPI_T_pvar_write(session, flush_handle, filename) ) {
fprintf(stderr, "Process %d cannot save monitoring in %s.%d.prof\n",
world_rank, filename, world_rank);
}
/* Force the writing of the monitoring data */
MPIT_result = MPI_T_pvar_stop(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to stop handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_start(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to start handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
/* Don't set a filename. If we stop the session before setting it, then no output
* will be generated.
*/
if( MPI_SUCCESS != MPI_T_pvar_write(session, flush_handle, (void*)&nullbuf ) ) {
fprintf(stderr, "Process %d cannot save monitoring in %s\n", world_rank, filename);
}
}
if( with_rma ) {
MPI_Win win;
int rs_buff[10240];
int win_buff[10240];
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
to = (rank + 1) % size;
from = (rank + size - 1) % size;
for( int v = 0; v < 10240; ++v )
rs_buff[v] = win_buff[v] = rank;
MPI_Win_create(win_buff, 10240 * sizeof(int), sizeof(int), MPI_INFO_NULL, MPI_COMM_WORLD, &win);
MPI_Win_fence(MPI_MODE_NOPRECEDE, win);
if( rank%2 ) {
MPI_Win_fence(MPI_MODE_NOSTORE | MPI_MODE_NOPUT, win);
MPI_Get(rs_buff, 10240, MPI_INT, from, 0, 10240, MPI_INT, win);
} else {
MPI_Put(rs_buff, 10240, MPI_INT, to, 0, 10240, MPI_INT, win);
MPI_Win_fence(MPI_MODE_NOSTORE | MPI_MODE_NOPUT, win);
}
MPI_Win_fence(MPI_MODE_NOSUCCEED, win);
for( int v = 0; v < 10240; ++v )
if( rs_buff[v] != win_buff[v] && ((rank%2 && rs_buff[v] != from) || (!(rank%2) && rs_buff[v] != rank)) ) {
printf("Error on checking exchanged values: %s_buff[%d] == %d instead of %d\n",
rank%2 ? "rs" : "win", v, rs_buff[v], rank%2 ? from : rank);
MPI_Abort(MPI_COMM_WORLD, -1);
}
MPI_Group world_group, newcomm_group, distant_group;
MPI_Comm_group(MPI_COMM_WORLD, &world_group);
MPI_Comm_group(newcomm, &newcomm_group);
MPI_Group_difference(world_group, newcomm_group, &distant_group);
if( rank%2 ) {
MPI_Win_post(distant_group, 0, win);
MPI_Win_wait(win);
/* Check received values */
for( int v = 0; v < 10240; ++v )
if( from != win_buff[v] ) {
printf("Error on checking exchanged values: win_buff[%d] == %d instead of %d\n",
v, win_buff[v], from);
MPI_Abort(MPI_COMM_WORLD, -1);
}
} else {
MPI_Win_start(distant_group, 0, win);
MPI_Put(rs_buff, 10240, MPI_INT, to, 0, 10240, MPI_INT, win);
MPI_Win_complete(win);
}
MPI_Group_free(&world_group);
MPI_Group_free(&newcomm_group);
MPI_Group_free(&distant_group);
MPI_Barrier(MPI_COMM_WORLD);
for( int v = 0; v < 10240; ++v ) rs_buff[v] = rank;
MPI_Win_lock(MPI_LOCK_EXCLUSIVE, to, 0, win);
MPI_Put(rs_buff, 10240, MPI_INT, to, 0, 10240, MPI_INT, win);
MPI_Win_unlock(to, win);
MPI_Barrier(MPI_COMM_WORLD);
/* Check received values */
for( int v = 0; v < 10240; ++v )
if( from != win_buff[v] ) {
printf("Error on checking exchanged values: win_buff[%d] == %d instead of %d\n",
v, win_buff[v], from);
MPI_Abort(MPI_COMM_WORLD, -1);
}
MPI_Win_free(&win);
}
if( with_mpit ) {
/* the filename for flushing monitoring now uses 3 as phase number! */
sprintf(filename, "prof/phase_3");
if( MPI_SUCCESS != MPI_T_pvar_write(session, flush_handle, filename) ) {
fprintf(stderr, "Process %d cannot save monitoring in %s.%d.prof\n",
world_rank, filename, world_rank);
}
MPIT_result = MPI_T_pvar_stop(session, flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to stop handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_handle_free(session, &flush_handle);
if (MPIT_result != MPI_SUCCESS) {
printf("failed to free handle on \"%s\" pvar, check that you have monitoring pml\n",
flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
MPIT_result = MPI_T_pvar_session_free(&session);
if (MPIT_result != MPI_SUCCESS) {
printf("cannot close a session for \"%s\" pvar\n", flush_pvar_name);
MPI_Abort(MPI_COMM_WORLD, MPIT_result);
}
(void)MPI_T_finalize();
}
MPI_Comm_free(&newcomm);
/* Now, in MPI_Finalize(), the pml_monitoring library outputs, in
STDERR, the aggregated recorded monitoring of all the phases*/
MPI_Finalize();
return 0;
}
|
