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
|
/*
* Copyright (C) by Argonne National Laboratory
* See COPYRIGHT in top-level directory
*/
#include "mpi.h"
#include <stdio.h>
#include <stdlib.h>
#include "mpitest.h"
#include <string.h>
/*
static char MTEST_Descrip[] = "Mix synchronization types";
*/
void delay(double time);
void delay(double time)
{
double t1;
t1 = MPI_Wtime();
while (MPI_Wtime() - t1 < time);
}
int main(int argc, char *argv[])
{
int errs = 0;
int crank, csize, source, dest, loop;
int *buf0, *buf1, *buf2, *inbuf2, count0, count1, count2, count, i;
MPI_Comm comm;
MPI_Win win;
int *winbuf;
MTest_Init(&argc, &argv);
comm = MPI_COMM_WORLD;
count0 = 1000;
count1 = 1;
count2 = 100;
count = count0 + count1 + count2 + 2;
/* Allocate and initialize the local buffers */
buf0 = (int *) malloc(count0 * sizeof(int));
buf1 = (int *) malloc(count1 * sizeof(int));
buf2 = (int *) malloc(count2 * sizeof(int));
inbuf2 = (int *) malloc(count2 * sizeof(int));
if (!buf0 || !buf1 || !buf2 || !inbuf2) {
fprintf(stderr, "Unable to allocated buf0-2\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
for (i = 0; i < count0; i++)
buf0[i] = i;
for (i = 0; i < count1; i++)
buf1[i] = i + count0;
for (i = 0; i < count2; i++)
buf2[i] = i + count0 + count1;
/* Allocate the window buffer and create the memory window. */
MPI_Alloc_mem(count * sizeof(int), MPI_INFO_NULL, &winbuf);
if (!winbuf) {
fprintf(stderr, "Unable to allocate %d words\n", count);
MPI_Abort(MPI_COMM_WORLD, 0);
}
MPI_Win_create(winbuf, count * sizeof(int), sizeof(int), MPI_INFO_NULL, comm, &win);
MPI_Comm_size(comm, &csize);
MPI_Comm_rank(comm, &crank);
dest = 0;
source = 1;
for (loop = 0; loop < 2; loop++) {
/* Perform several communication operations, mixing synchronization
* types. Use multiple communication to avoid the single-operation
* optimization that may be present. */
MTestPrintfMsg(3, "Beginning loop %d of mixed sync put operations\n", loop);
MPI_Barrier(comm);
if (crank == source) {
MTestPrintfMsg(3, "About to perform exclusive lock\n");
MPI_Win_lock(MPI_LOCK_EXCLUSIVE, dest, 0, win);
MPI_Put(buf0, count0, MPI_INT, dest, 0, count0, MPI_INT, win);
MPI_Put(buf1, count1, MPI_INT, dest, count0, count1, MPI_INT, win);
MPI_Put(buf2, count2, MPI_INT, dest, count0 + count1, count2, MPI_INT, win);
MPI_Win_unlock(dest, win);
MTestPrintfMsg(3, "Released exclusive lock\n");
} else if (crank == dest) {
/* Just delay a bit */
delay(0.0001);
}
/* The synchronization mode can only be changed when the process
* memory and public copy are guaranteed to have the same values
* (See 11.7, Semantics and Correctness). This barrier ensures that
* the lock/unlock completes before the fence call. */
MPI_Barrier(comm);
MTestPrintfMsg(3, "About to start fence\n");
MPI_Win_fence(0, win);
if (crank == source) {
MPI_Put(buf0, count0, MPI_INT, dest, 1, count0, MPI_INT, win);
MPI_Put(buf1, count1, MPI_INT, dest, 1 + count0, count1, MPI_INT, win);
MPI_Put(buf2, count2, MPI_INT, dest, 1 + count0 + count1, count2, MPI_INT, win);
}
MPI_Win_fence(0, win);
MTestPrintfMsg(3, "Finished with fence sync\n");
/* Check results */
if (crank == dest) {
for (i = 0; i < count0 + count1 + count2; i++) {
if (winbuf[1 + i] != i) {
errs++;
if (errs < 10) {
fprintf(stderr, "winbuf[%d] = %d, expected %d\n", 1 + i, winbuf[1 + i], i);
fflush(stderr);
}
}
}
}
/* End of test loop */
}
/* Use mixed put and accumulate */
for (loop = 0; loop < 2; loop++) {
/* Perform several communication operations, mixing synchronization
* types. Use multiple communication to avoid the single-operation
* optimization that may be present. */
MTestPrintfMsg(3, "Beginning loop %d of mixed sync put/acc operations\n", loop);
memset(winbuf, 0, count * sizeof(int));
MPI_Barrier(comm);
if (crank == source) {
MPI_Win_lock(MPI_LOCK_EXCLUSIVE, dest, 0, win);
MPI_Accumulate(buf0, count0, MPI_INT, dest, 0, count0, MPI_INT, MPI_SUM, win);
MPI_Accumulate(buf1, count1, MPI_INT, dest, count0, count1, MPI_INT, MPI_SUM, win);
MPI_Put(buf2, count2, MPI_INT, dest, count0 + count1, count2, MPI_INT, win);
MPI_Win_unlock(dest, win);
} else if (crank == dest) {
/* Just delay a bit */
delay(0.0001);
}
/* See above - the fence should not start until the unlock completes */
MPI_Barrier(comm);
MPI_Win_fence(0, win);
if (crank == source) {
MPI_Accumulate(buf0, count0, MPI_INT, dest, 1, count0, MPI_INT, MPI_REPLACE, win);
MPI_Accumulate(buf1, count1, MPI_INT, dest, 1 + count0, count1,
MPI_INT, MPI_REPLACE, win);
MPI_Put(buf2, count2, MPI_INT, dest, 1 + count0 + count1, count2, MPI_INT, win);
}
MPI_Win_fence(0, win);
/* Check results */
if (crank == dest) {
for (i = 0; i < count0 + count1 + count2; i++) {
if (winbuf[1 + i] != i) {
errs++;
if (errs < 10) {
fprintf(stderr, "winbuf[%d] = %d, expected %d\n", 1 + i, winbuf[1 + i], i);
fflush(stderr);
}
}
}
}
/* End of test loop */
}
/* Use mixed accumulate and get */
for (loop = 0; loop < 2; loop++) {
/* Perform several communication operations, mixing synchronization
* types. Use multiple communication to avoid the single-operation
* optimization that may be present. */
MTestPrintfMsg(3, "Beginning loop %d of mixed sync put/get/acc operations\n", loop);
MPI_Barrier(comm);
if (crank == source) {
MPI_Win_lock(MPI_LOCK_EXCLUSIVE, dest, 0, win);
MPI_Accumulate(buf0, count0, MPI_INT, dest, 0, count0, MPI_INT, MPI_REPLACE, win);
MPI_Put(buf1, count1, MPI_INT, dest, count0, count1, MPI_INT, win);
MPI_Get(inbuf2, count2, MPI_INT, dest, count0 + count1, count2, MPI_INT, win);
MPI_Win_unlock(dest, win);
} else if (crank == dest) {
/* Just delay a bit */
delay(0.0001);
}
/* See above - the fence should not start until the unlock completes */
MPI_Barrier(comm);
MPI_Win_fence(0, win);
if (crank == source) {
MPI_Accumulate(buf0, count0, MPI_INT, dest, 1, count0, MPI_INT, MPI_REPLACE, win);
MPI_Put(buf1, count1, MPI_INT, dest, 1 + count0, count1, MPI_INT, win);
MPI_Get(inbuf2, count2, MPI_INT, dest, 1 + count0 + count1, count2, MPI_INT, win);
}
MPI_Win_fence(0, win);
/* Check results */
if (crank == dest) {
/* Do the put/accumulate parts */
for (i = 0; i < count0 + count1; i++) {
if (winbuf[1 + i] != i) {
errs++;
if (errs < 10) {
fprintf(stderr, "winbuf[%d] = %d, expected %d\n", 1 + i, winbuf[1 + i], i);
fflush(stderr);
}
}
}
}
/* End of test loop */
}
MTestPrintfMsg(3, "Freeing the window\n");
MPI_Barrier(comm);
MPI_Win_free(&win);
MPI_Free_mem(winbuf);
free(buf0);
free(buf1);
free(buf2);
free(inbuf2);
MTest_Finalize(errs);
return MTestReturnValue(errs);
}
|
