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
|
/*
* Copyright (C) by Argonne National Laboratory
* See COPYRIGHT in top-level directory
*/
/* This test is going to test the atomicity for "read-modify-write" in GACC
* operations */
/* This test is similar with atomic_rmw_fop.c.
* There are three processes involved in this test: P0 (origin_shm), P1 (origin_am),
* and P2 (dest). P0 and P1 issues multiple GACC with MPI_SUM and OP_COUNT integers
* (value 1) to P2 via SHM and AM respectively. The correct results should be that the
* results on P0 and P1 never be the same for integers on the corresponding index
* in [0...OP_COUNT-1].
*/
#include "mpi.h"
#include <stdio.h>
#include "mpitest.h"
#define OP_COUNT 10
#define AM_BUF_NUM 10
#define SHM_BUF_NUM 10000
#define WIN_BUF_NUM 1
#define LOOP_SIZE 15
#define CHECK_TAG 123
int rank, size;
int dest, origin_shm, origin_am;
int *orig_buf = NULL, *result_buf = NULL, *target_buf = NULL, *check_buf = NULL;
MPI_Win win;
static void checkResults(int loop_k, int *errors)
{
int i, j, m;
MPI_Status status;
if (rank != dest) {
/* check results on P0 and P2 (origin) */
if (rank == origin_am) {
MPI_Send(result_buf, AM_BUF_NUM * OP_COUNT, MPI_INT, origin_shm, CHECK_TAG,
MPI_COMM_WORLD);
} else if (rank == origin_shm) {
MPI_Alloc_mem(sizeof(int) * AM_BUF_NUM * OP_COUNT, MPI_INFO_NULL, &check_buf);
MPI_Recv(check_buf, AM_BUF_NUM * OP_COUNT, MPI_INT, origin_am, CHECK_TAG,
MPI_COMM_WORLD, &status);
for (i = 0; i < AM_BUF_NUM; i++) {
for (j = 0; j < SHM_BUF_NUM; j++) {
for (m = 0; m < OP_COUNT; m++) {
if (check_buf[i * OP_COUNT + m] == result_buf[j * OP_COUNT + m]) {
printf
("LOOP=%d, rank=%d, FOP, both check_buf[%d] and result_buf[%d] equal to %d, expected to be different. \n",
loop_k, rank, i * OP_COUNT + m, j * OP_COUNT + m,
check_buf[i * OP_COUNT + m]);
(*errors)++;
}
}
}
}
MPI_Free_mem(check_buf);
}
} else {
MPI_Win_lock(MPI_LOCK_SHARED, rank, 0, win);
/* check results on P1 */
for (i = 0; i < OP_COUNT; i++) {
if (target_buf[i] != AM_BUF_NUM + SHM_BUF_NUM) {
printf("LOOP=%d, rank=%d, FOP, target_buf[%d] = %d, expected %d. \n",
loop_k, rank, i, target_buf[i], AM_BUF_NUM + SHM_BUF_NUM);
(*errors)++;
}
}
MPI_Win_unlock(rank, win);
}
}
int main(int argc, char *argv[])
{
int i, k;
int errors = 0;
int my_buf_num = 0; /* to suppress warning */
MPI_Datatype origin_dtp, target_dtp;
MTest_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
if (size != 3) {
/* run this test with three processes */
goto exit_test;
}
MPI_Type_contiguous(OP_COUNT, MPI_INT, &origin_dtp);
MPI_Type_commit(&origin_dtp);
MPI_Type_contiguous(OP_COUNT, MPI_INT, &target_dtp);
MPI_Type_commit(&target_dtp);
/* this works when MPIR_PARAM_CH3_ODD_EVEN_CLIQUES is set */
dest = 2;
origin_shm = 0;
origin_am = 1;
if (rank == origin_am)
my_buf_num = AM_BUF_NUM;
else if (rank == origin_shm)
my_buf_num = SHM_BUF_NUM;
if (rank != dest) {
MPI_Alloc_mem(sizeof(int) * my_buf_num * OP_COUNT, MPI_INFO_NULL, &orig_buf);
MPI_Alloc_mem(sizeof(int) * my_buf_num * OP_COUNT, MPI_INFO_NULL, &result_buf);
}
MPI_Win_allocate(sizeof(int) * WIN_BUF_NUM * OP_COUNT, sizeof(int), MPI_INFO_NULL,
MPI_COMM_WORLD, &target_buf, &win);
for (k = 0; k < LOOP_SIZE; k++) {
/* ====== Part 1: test basic datatypes ======== */
/* init buffers */
if (rank != dest) {
for (i = 0; i < my_buf_num * OP_COUNT; i++) {
orig_buf[i] = 1;
result_buf[i] = 0;
}
} else {
MPI_Win_lock(MPI_LOCK_SHARED, rank, 0, win);
for (i = 0; i < WIN_BUF_NUM * OP_COUNT; i++) {
target_buf[i] = 0;
}
MPI_Win_unlock(rank, win);
}
MPI_Barrier(MPI_COMM_WORLD);
MPI_Win_lock_all(0, win);
if (rank != dest) {
for (i = 0; i < my_buf_num; i++) {
MPI_Get_accumulate(&(orig_buf[i * OP_COUNT]), OP_COUNT, MPI_INT,
&(result_buf[i * OP_COUNT]), OP_COUNT, MPI_INT,
dest, 0, OP_COUNT, MPI_INT, MPI_SUM, win);
MPI_Win_flush(dest, win);
}
}
MPI_Win_unlock_all(win);
MPI_Barrier(MPI_COMM_WORLD);
checkResults(k, &errors);
/* ====== Part 2: test derived datatypes (origin derived, target derived) ======== */
/* init buffers */
if (rank != dest) {
for (i = 0; i < my_buf_num * OP_COUNT; i++) {
orig_buf[i] = 1;
result_buf[i] = 0;
}
} else {
MPI_Win_lock(MPI_LOCK_SHARED, rank, 0, win);
for (i = 0; i < WIN_BUF_NUM * OP_COUNT; i++) {
target_buf[i] = 0;
}
MPI_Win_unlock(rank, win);
}
MPI_Barrier(MPI_COMM_WORLD);
MPI_Win_lock_all(0, win);
if (rank != dest) {
for (i = 0; i < my_buf_num; i++) {
MPI_Get_accumulate(&(orig_buf[i * OP_COUNT]), 1, origin_dtp,
&(result_buf[i * OP_COUNT]), 1, origin_dtp,
dest, 0, 1, target_dtp, MPI_SUM, win);
MPI_Win_flush(dest, win);
}
}
MPI_Win_unlock_all(win);
MPI_Barrier(MPI_COMM_WORLD);
checkResults(k, &errors);
/* ====== Part 3: test derived datatypes (origin basic, target derived) ======== */
/* init buffers */
if (rank != dest) {
for (i = 0; i < my_buf_num * OP_COUNT; i++) {
orig_buf[i] = 1;
result_buf[i] = 0;
}
} else {
MPI_Win_lock(MPI_LOCK_SHARED, rank, 0, win);
for (i = 0; i < WIN_BUF_NUM * OP_COUNT; i++) {
target_buf[i] = 0;
}
MPI_Win_unlock(rank, win);
}
MPI_Barrier(MPI_COMM_WORLD);
MPI_Win_lock_all(0, win);
if (rank != dest) {
for (i = 0; i < my_buf_num; i++) {
MPI_Get_accumulate(&(orig_buf[i * OP_COUNT]), OP_COUNT, MPI_INT,
&(result_buf[i * OP_COUNT]), OP_COUNT, MPI_INT,
dest, 0, 1, target_dtp, MPI_SUM, win);
MPI_Win_flush(dest, win);
}
}
MPI_Win_unlock_all(win);
MPI_Barrier(MPI_COMM_WORLD);
checkResults(k, &errors);
/* ====== Part 4: test derived datatypes (origin derived target basic) ======== */
/* init buffers */
if (rank != dest) {
for (i = 0; i < my_buf_num * OP_COUNT; i++) {
orig_buf[i] = 1;
result_buf[i] = 0;
}
} else {
MPI_Win_lock(MPI_LOCK_SHARED, rank, 0, win);
for (i = 0; i < WIN_BUF_NUM * OP_COUNT; i++) {
target_buf[i] = 0;
}
MPI_Win_unlock(rank, win);
}
MPI_Barrier(MPI_COMM_WORLD);
MPI_Win_lock_all(0, win);
if (rank != dest) {
for (i = 0; i < my_buf_num; i++) {
MPI_Get_accumulate(&(orig_buf[i * OP_COUNT]), 1, origin_dtp,
&(result_buf[i * OP_COUNT]), 1, origin_dtp,
dest, 0, OP_COUNT, MPI_INT, MPI_SUM, win);
MPI_Win_flush(dest, win);
}
}
MPI_Win_unlock_all(win);
MPI_Barrier(MPI_COMM_WORLD);
checkResults(k, &errors);
}
MPI_Win_free(&win);
if (rank == origin_am || rank == origin_shm) {
MPI_Free_mem(orig_buf);
MPI_Free_mem(result_buf);
}
MPI_Type_free(&origin_dtp);
MPI_Type_free(&target_dtp);
exit_test:
MTest_Finalize(errors);
return MTestReturnValue(errors);
}
|
