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
|
/*
* Copyright (C) by Argonne National Laboratory
* See COPYRIGHT in top-level directory
*/
#include "mpi.h"
#include <stdio.h>
#include "mpitest.h"
#define MAX_PROCESSES 10
int main(int argc, char **argv)
{
int rank, size, i, j, iter;
int table[MAX_PROCESSES][MAX_PROCESSES];
int errors = 0;
int participants;
int displs[MAX_PROCESSES];
int recv_counts[MAX_PROCESSES];
MPI_Request *reqs;
MPI_Info info;
MTest_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Info_create(&info);
/* A maximum of MAX_PROCESSES processes can participate */
if (size > MAX_PROCESSES)
participants = MAX_PROCESSES;
else
participants = size;
/* while (MAX_PROCESSES % participants) participants--; */
if (MAX_PROCESSES % participants) {
fprintf(stderr, "Number of processors must divide %d\n", MAX_PROCESSES);
MPI_Abort(MPI_COMM_WORLD, 1);
}
reqs = (MPI_Request *) malloc(participants * sizeof(MPI_Request));
if ((rank < participants)) {
if (!reqs) {
fprintf(stderr, "Cannot allocate request handle array\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
/* Determine what rows are my responsibility */
int block_size = MAX_PROCESSES / participants;
int begin_row = rank * block_size;
int end_row = (rank + 1) * block_size;
int send_count = block_size * MAX_PROCESSES;
/* Fill in the displacements and recv_counts */
for (i = 0; i < participants; i++) {
displs[i] = i * block_size * MAX_PROCESSES;
recv_counts[i] = send_count;
}
for (i = 0; i < participants; i++) {
void *sendbuf = (i == rank ? MPI_IN_PLACE : &table[begin_row][0]);
MPI_Gatherv_init(sendbuf, send_count, MPI_INT, &table[0][0], recv_counts, displs,
MPI_INT, i, MPI_COMM_WORLD, info, &reqs[i]);
}
for (iter = 0; iter < 10; iter++) {
/* Paint my rows my color */
for (i = begin_row; i < end_row; i++)
for (j = 0; j < MAX_PROCESSES; j++)
table[i][j] = rank + 10;
/* Gather everybody's result together - sort of like an */
/* inefficient allgather */
for (i = 0; i < participants; i++) {
MPI_Start(&reqs[i]);
MPI_Wait(&reqs[i], MPI_STATUS_IGNORE);
}
/* Everybody should have the same table now.
*
* The entries are:
* Table[i][j] = (i/block_size) + 10;
*/
for (i = 0; i < MAX_PROCESSES; i++)
if ((table[i][0] - table[i][MAX_PROCESSES - 1] != 0))
errors++;
for (i = 0; i < MAX_PROCESSES; i++) {
for (j = 0; j < MAX_PROCESSES; j++) {
if (table[i][j] != (i / block_size) + 10)
errors++;
}
}
if (errors) {
/* Print out table if there are any errors */
for (i = 0; i < MAX_PROCESSES; i++) {
printf("\n");
for (j = 0; j < MAX_PROCESSES; j++)
printf(" %d", table[i][j]);
}
printf("\n");
}
}
for (i = 0; i < participants; i++) {
MPI_Request_free(&reqs[i]);
}
}
free(reqs);
MPI_Info_free(&info);
MTest_Finalize(errors);
return MTestReturnValue(errors);
}
|
