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
|
/*
* Copyright (C) by Argonne National Laboratory
* See COPYRIGHT in top-level directory
*/
#include "mpi.h"
#include "stdio.h"
#include "stdlib.h"
#include "mpitest.h"
/* This is the tree-based scalable version of the fetch-and-add
example from Using MPI-2, pg 206-207. The code in the book (Fig
6.16) has bugs that are fixed below. */
#define NTIMES 20 /* no of times each process calls the counter
* routine */
int localvalue = 0; /* contribution of this process to the counter. We
* define it as a global variable because attribute
* caching on the window is not enabled yet. */
void Get_nextval_tree(MPI_Win win, int *get_array, MPI_Datatype get_type,
MPI_Datatype acc_type, int nlevels, int *value);
int compar(const void *a, const void *b);
int main(int argc, char *argv[])
{
int rank, nprocs, i, *counter_mem, *get_array, *get_idx, *acc_idx,
mask, nlevels, level, idx, tmp_rank, pof2;
MPI_Datatype get_type, acc_type;
MPI_Win win;
int errs = 0, *results, *counter_vals;
MTest_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if (rank == 0) {
/* allocate counter memory and initialize to 0 */
/* find the next power-of-two >= nprocs */
pof2 = 1;
while (pof2 < nprocs)
pof2 *= 2;
counter_mem = (int *) calloc(pof2 * 2, sizeof(int));
MPI_Win_create(counter_mem, pof2 * 2 * sizeof(int), sizeof(int),
MPI_INFO_NULL, MPI_COMM_WORLD, &win);
MPI_Win_free(&win);
free(counter_mem);
/* gather the results from other processes, sort them, and check
* whether they represent a counter being incremented by 1 */
results = (int *) malloc(NTIMES * nprocs * sizeof(int));
for (i = 0; i < NTIMES * nprocs; i++)
results[i] = -1;
MPI_Gather(MPI_IN_PLACE, 0, MPI_DATATYPE_NULL, results, NTIMES, MPI_INT, 0, MPI_COMM_WORLD);
qsort(results + NTIMES, NTIMES * (nprocs - 1), sizeof(int), compar);
for (i = NTIMES + 1; i < (NTIMES * nprocs); i++)
if (results[i] != results[i - 1] + 1)
errs++;
free(results);
} else {
/* Get the largest power of two smaller than nprocs */
mask = 1;
nlevels = 0;
while (mask < nprocs) {
mask <<= 1;
nlevels++;
}
mask >>= 1;
get_array = (int *) malloc(nlevels * sizeof(int));
get_idx = (int *) malloc(nlevels * sizeof(int));
acc_idx = (int *) malloc(nlevels * sizeof(int));
level = 0;
idx = 0;
tmp_rank = rank;
while (mask >= 1) {
if (tmp_rank < mask) {
/* go to left for acc_idx, go to right for
* get_idx. set idx=acc_idx for next iteration */
acc_idx[level] = idx + 1;
get_idx[level] = idx + mask * 2;
idx = idx + 1;
} else {
/* go to right for acc_idx, go to left for
* get_idx. set idx=acc_idx for next iteration */
acc_idx[level] = idx + mask * 2;
get_idx[level] = idx + 1;
idx = idx + mask * 2;
}
level++;
tmp_rank = tmp_rank % mask;
mask >>= 1;
}
/* for (i=0; i<nlevels; i++)
printf("Rank %d, acc_idx[%d]=%d, get_idx[%d]=%d\n", rank,
i, acc_idx[i], i, get_idx[i]);
*/
MPI_Type_create_indexed_block(nlevels, 1, get_idx, MPI_INT, &get_type);
MPI_Type_create_indexed_block(nlevels, 1, acc_idx, MPI_INT, &acc_type);
MPI_Type_commit(&get_type);
MPI_Type_commit(&acc_type);
/* allocate array to store the values obtained from the
* fetch-and-add counter */
counter_vals = (int *) malloc(NTIMES * sizeof(int));
MPI_Win_create(NULL, 0, 1, MPI_INFO_NULL, MPI_COMM_WORLD, &win);
for (i = 0; i < NTIMES; i++) {
Get_nextval_tree(win, get_array, get_type, acc_type, nlevels, counter_vals + i);
/* printf("Rank %d, counter %d\n", rank, value); */
}
MPI_Win_free(&win);
free(get_array);
free(get_idx);
free(acc_idx);
MPI_Type_free(&get_type);
MPI_Type_free(&acc_type);
/* gather the results to the root */
MPI_Gather(counter_vals, NTIMES, MPI_INT, NULL, 0, MPI_DATATYPE_NULL, 0, MPI_COMM_WORLD);
free(counter_vals);
}
MTest_Finalize(errs);
return MTestReturnValue(errs);
}
void Get_nextval_tree(MPI_Win win, int *get_array, MPI_Datatype get_type,
MPI_Datatype acc_type, int nlevels, int *value)
{
int *one, i;
one = (int *) malloc(nlevels * sizeof(int));
for (i = 0; i < nlevels; i++)
one[i] = 1;
MPI_Win_lock(MPI_LOCK_EXCLUSIVE, 0, 0, win);
MPI_Accumulate(one, nlevels, MPI_INT, 0, 0, 1, acc_type, MPI_SUM, win);
MPI_Get(get_array, nlevels, MPI_INT, 0, 0, 1, get_type, win);
MPI_Win_unlock(0, win);
*value = localvalue;
for (i = 0; i < nlevels; i++)
*value = *value + get_array[i];
localvalue++;
free(one);
}
int compar(const void *a, const void *b)
{
return (*((int *) a) - *((int *) b));
}
|
