-
Notifications
You must be signed in to change notification settings - Fork 0
/
mpi_2snb_strassen1.hpp
202 lines (168 loc) · 7.84 KB
/
mpi_2snb_strassen1.hpp
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
#include <mpi.h>
#include "omp_strassen.hpp"
void mpi_2snb_strassen(int **A, int **B, int **C, int n, int m, int p, int rank) {
int n1 = (n + 1) / 2, m1 = (m + 1) / 2, p1 = (p + 1) / 2;
//divide matrix A into 4 submatrix
int **A11 = new_matrix(n1,m1);
int **A12 = new_matrix(n1,m1);
int **A21 = new_matrix(n1,m1);
int **A22 = new_matrix(n1,m1);
int **B11 = new_matrix(m1,p1);
int **B12 = new_matrix(m1,p1);
int **B21 = new_matrix(m1,p1);
int **B22 = new_matrix(m1,p1);
if (rank == 0) {
init_sub_matrix(A,A11,n,m,0,0);
init_sub_matrix(A,A12,n,m,0,m1);
init_sub_matrix(A,A21,n,m,n1,0);
init_sub_matrix(A,A22,n,m,n1,m1);
init_sub_matrix(B, B11, m, p, 0, 0);
init_sub_matrix(B, B12, m, p, 0, p1);
init_sub_matrix(B, B21, m, p, m1, 0);
init_sub_matrix(B, B22, m, p, m1, p1);
}
// submatrix results of strassen
int **M1 = new_matrix(n1,p1);
int **M2 = new_matrix(n1, p1);
int **M3 = new_matrix(n1, p1);
int **M4 = new_matrix(n1, p1);
int **M5 = new_matrix(n1, p1);
int **M6 = new_matrix(n1, p1);
int **M7 = new_matrix(n1, p1);
int a_sub_size = n1 * p1;
int b_sub_size = m1 * p1;
int c_sub_size = n1 * p1;
//target process
if (rank == 1) {
MPI_Recv(&A21[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&A22[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B11[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B21[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
int **M21 = new_matrix(n1, m1);
int **M42 = new_matrix(m1, p1);
// calculate M2
matrix_add(A21, A22, M21, n1, m1);
omp_strassen(M21, B11, M2, n1, m1, p1);
// calculate M4
matrix_sub(B21, B11, M42, m1, p1);
omp_strassen(A22, M42, M4, n1, m1, p1);
MPI_Send(&M2[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
MPI_Send(&M4[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
delete_matrix(M21);
delete_matrix(M42);
}
if (rank == 2) {
MPI_Recv(&A11[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&A21[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B11[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B12[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B22[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
int **M32 = new_matrix(m1, p1);
int **M61 = new_matrix(n1, m1);
int **M62 = new_matrix(m1, p1);
// calculate M3
matrix_sub(B12, B22, M32, m1, p1);
omp_strassen(A11, M32, M3, n1, m1, p1);
// calcualte M6
matrix_sub(A21, A11, M61, n1, m1);
matrix_add(B11, B12, M62, m1, p1);
omp_strassen(M61, M62, M6, n1, m1, p1);
MPI_Send(&M3[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
MPI_Send(&M6[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
delete_matrix(M32);
delete_matrix(M61);
delete_matrix(M62);
}
if (rank == 3) {
MPI_Recv(&A11[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&A12[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&A22[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B21[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(&B22[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD, MPI_STATUS_IGNORE);
int **M51 = new_matrix(n1, m1);
int **M71 = new_matrix(n1, m1);
int **M72 = new_matrix(m1, p1);
// calculate M5
matrix_add(A11, A12, M51, n1, m1);
omp_strassen(M51, B22, M5, n1, m1, p1);
//calculate M7
matrix_sub(A12, A22, M71, n1, m1);
matrix_add(B21, B22, M72, m1, p1);
omp_strassen(M71, M72, M7, n1, m1, p1);
MPI_Send(&M5[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
MPI_Send(&M7[0][0], c_sub_size, MPI_INT, 0, 0, MPI_COMM_WORLD);
delete_matrix(M51);
delete_matrix(M72);
delete_matrix(M71);
}
//sending process
if (rank == 0) {
MPI_Request recv_request[6];
MPI_Irecv(&M2[0][0], c_sub_size, MPI_INT, 1, 0, MPI_COMM_WORLD, &recv_request[0]);
MPI_Irecv(&M4[0][0], c_sub_size, MPI_INT, 1, 0, MPI_COMM_WORLD, &recv_request[1]);
MPI_Irecv(&M3[0][0], c_sub_size, MPI_INT, 2, 0, MPI_COMM_WORLD, &recv_request[2]);
MPI_Irecv(&M6[0][0], c_sub_size, MPI_INT, 2, 0, MPI_COMM_WORLD, &recv_request[3]);
MPI_Irecv(&M5[0][0], c_sub_size, MPI_INT, 3, 0, MPI_COMM_WORLD, &recv_request[4]);
MPI_Irecv(&M7[0][0], c_sub_size, MPI_INT, 3, 0, MPI_COMM_WORLD, &recv_request[5]);
// //send to machine 0 to process
// MPI_Send(&A11[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD);
// MPI_Send(&A22[0][0],a_sub_size,MPI_INT,0,0,MPI_COMM_WORLD);
// MPI_Send(&B11[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD);
// MPI_Send(&B22[0][0],b_sub_size,MPI_INT,0,0,MPI_COMM_WORLD);
MPI_Request send_request[14];
//send to machine 1 to process
MPI_Isend(&A21[0][0],a_sub_size,MPI_INT,1,0,MPI_COMM_WORLD, &send_request[0]);
MPI_Isend(&A22[0][0],a_sub_size,MPI_INT,1,0,MPI_COMM_WORLD, &send_request[1]);
MPI_Isend(&B11[0][0],b_sub_size,MPI_INT,1,0,MPI_COMM_WORLD, &send_request[2]);
MPI_Isend(&B21[0][0],b_sub_size,MPI_INT,1,0,MPI_COMM_WORLD, &send_request[3]);
//send to machine 2 to process
MPI_Isend(&A11[0][0],a_sub_size,MPI_INT,2,0,MPI_COMM_WORLD, &send_request[4]);
MPI_Isend(&A21[0][0],a_sub_size,MPI_INT,2,0,MPI_COMM_WORLD, &send_request[5]);
MPI_Isend(&B11[0][0],b_sub_size,MPI_INT,2,0,MPI_COMM_WORLD, &send_request[6]);
MPI_Isend(&B12[0][0],b_sub_size,MPI_INT,2,0,MPI_COMM_WORLD, &send_request[7]);
MPI_Isend(&B22[0][0],b_sub_size,MPI_INT,2,0,MPI_COMM_WORLD, &send_request[8]);
//send to machine 3 to process
MPI_Isend(&A11[0][0],a_sub_size,MPI_INT,3,0,MPI_COMM_WORLD, &send_request[9]);
MPI_Isend(&A12[0][0],a_sub_size,MPI_INT,3,0,MPI_COMM_WORLD, &send_request[10]);
MPI_Isend(&A22[0][0],a_sub_size,MPI_INT,3,0,MPI_COMM_WORLD, &send_request[11]);
MPI_Isend(&B21[0][0],b_sub_size,MPI_INT,3,0,MPI_COMM_WORLD, &send_request[12]);
MPI_Isend(&B22[0][0],b_sub_size,MPI_INT,3,0,MPI_COMM_WORLD, &send_request[13]);
for (int i = 0; i < 14; ++i) MPI_Wait(&send_request[i], MPI_STATUS_IGNORE);
int **M11 = new_matrix(n1, m1);
int **M12 = new_matrix(m1, p1);
// calculate M1
matrix_add(A11, A22, M11, n1, m1);
matrix_add(B11, B22, M12, m1, p1);
omp_strassen(M11, M12, M1, n1, m1, p1);
delete_matrix(M11);
delete_matrix(M12);
for (int i = 0; i < 6; ++i) MPI_Wait(&recv_request[i], MPI_STATUS_IGNORE);
for (int i = 0; i < n1; ++i) {
for (int j = 0; j < p1; ++j) {
C[i][j] = M1[i][j] + M4[i][j] - M5[i][j] + M7[i][j];
if (p1 + j < p) C[i][p1 + j] = M3[i][j] + M5[i][j];
if (n1 + i < n) {
C[n1 + i][j] = M2[i][j] + M4[i][j];
if (p1 + j < p) {
C[n1 + i][p1 + j] = M1[i][j] - M2[i][j] + M3[i][j] + M6[i][j];
}
}
}
}
}
delete_matrix(A11);
delete_matrix(A12);
delete_matrix(A21);
delete_matrix(A22);
delete_matrix(B11);
delete_matrix(B12);
delete_matrix(B21);
delete_matrix(B22);
delete_matrix(M1);
delete_matrix(M2);
delete_matrix(M3);
delete_matrix(M4);
delete_matrix(M5);
delete_matrix(M6);
delete_matrix(M7);
}