mw_addslave2.cpp 6.94 KB
Newer Older
Camille Coti's avatar
Camille Coti committed
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
#include <iostream>
#include <mpi.h>
#include <ginac/ginac.h>

#include "products.h"
#include "utils_parall.h"
#include "parall_constants.h"
#include "parall_internal.h"
#include "utils.h"
#include "profiling.h"

namespace gi = GiNaC;

/*******************************************************************************
 *         Parallel 1-level decomposition with addition on a slave             *
 *******************************************************************************/

gi::ex multiply_1level_master_addslave2( tensor3D_t& T, unsigned int size, MPI_Comm comm = MPI_COMM_WORLD ) { 
    gi::ex Tens = 0;
    unsigned int a4;
    gi::ex A;
    gi::lst symbols;

    MPI_Status status;
    char* expr_c;
    size_t expr_c_size = 0;
    int src, np, running = 0;
    unsigned int len;
    parameters_s_t pzero( 0 );

    MPI_Comm_size( comm, &np );

    expr_c = NULL;
    expr_c = (char*) malloc( 3279 );
    
    int receivedresults = 0;
    unsigned int N = size/2;

    std::vector<parameters_s_t> input;
    std::vector<std::string> results; /* length and char* */

    /* Build a list of argument sets */
    
    for( a4 = 0 ; a4 < N ; a4++ ){
        parameters_s_t p( a4 );
        input.push_back( p );
	}

    /* Compute the set of symbols */
    /* Could be done while the first slave is working */
    
    symbols = all_symbols_3D( size );

    /* Distribute the work */

    while( input.size() > 0 ) {
        MPI_Recv( &len, 1, MPI_UNSIGNED, MPI_ANY_SOURCE, MPI_ANY_TAG, comm, &status );
        
        if( status.MPI_TAG == TAG_PULL ) {
            /* Nothing else will come: just send wome work */
            src = status.MPI_SOURCE;
            send_work( input, src );
            
        } else {
            if( status.MPI_TAG == TAG_RES ){
                src = status.MPI_SOURCE;

                /* The first message contains the length of what is coming next */
                if( len != 0 ) {
                    if( len > expr_c_size ) {
                        expr_c_size = len;
                        if( NULL != expr_c ) free( expr_c );
                        expr_c = (char*)malloc( expr_c_size ); // The \0 was added by the slave
                    }
                    
                    /* Receive the result */
                    MPI_Recv( expr_c, len, MPI_CHAR, src, TAG_EXPR, comm, &status );

                    /* Put it in the result queue */
                    results.push_back( std::string( expr_c ) );
                }                    
                    
                /* Send more work  */
                send_work_addslave( input, results, src );
            } else {
                std::cerr << "Wrong tag received " << status.MPI_TAG << std::endl;
            }
            
        }
   }

   /* Wait until everyone is done */

    running = np - 1; // all the slaves are running 
    while( running > 0 ) {
        /* TODO: here all we should receive is either TAG_EXPR or TAG_PULL if the input is too small */
        MPI_Recv( &len, 1, MPI_UNSIGNED, MPI_ANY_SOURCE, MPI_ANY_TAG, comm, &status );
        src = status.MPI_SOURCE;
        
        if( len != 0 ) {
            if( len > expr_c_size ) {
                expr_c_size = len;
                if( NULL != expr_c ) free( expr_c );
                expr_c = (char*)malloc( expr_c_size ); // The \0 was added by the slave
            }

            /* Receive the result */
            MPI_Recv( expr_c, len, MPI_CHAR, src, TAG_EXPR, comm, &status );

            /* Put it in the result queue */
            results.push_back( std::string( expr_c ) );
        }
        send_add_or_end_addslave( results, src, &running, pzero );
    }

    /* Add whatever I have left */
    Tens = add_expressions( results, symbols );
    
#if DEBUG
    std::cout << "Received " << receivedresults << " results" << std::endl;

    std::cout << "Tpara=" << Tens << ";" << std::endl;
#endif
    
    if( NULL != expr_c) free( expr_c );
    return Tens;
}

void multiply_1level_slave_addslave2( tensor3D_t& T, unsigned int size, MPI_Comm comm = MPI_COMM_WORLD ) {
    gi::ex Tens;
    int  a4;
    unsigned int len = 0;
    
    parameters_s_t params;
    MPI_Status status;
    char* expr_c;

    int rank;
    MPI_Comm_rank( comm, &rank );

    /* Ask for some work */
    
    MPI_Send( &len, 1, MPI_UNSIGNED, ROOT, TAG_PULL, comm );

    /* Compute the set of symbols */
    
    gi::lst symbols = all_symbols_3D( size );

    while( true ){
        /* Receive a set of parameters */

        MPI_Recv( &params, 1, DT_PARAMETERS_S, ROOT, MPI_ANY_TAG, comm, &status );
        
        if( status.MPI_TAG == TAG_WORK ){
            a4 = params.a4;

            Tens = one_level1_product( &T, size, a4 );
            send_result( Tens );

        } else {
            if( status.MPI_TAG == TAG_ADD ) {
                /* Receive a set of expressions to add */

                /* Number of expressions received */
                int nb = params.a4;
                
                /* Length of each string */

                unsigned int* lengths = (unsigned int*) malloc( nb*sizeof( unsigned int ) );
                MPI_Recv( lengths, nb, MPI_INT, ROOT, TAG_ADD, comm, &status );
                std::vector<std::string> results_s;
                char* c_str;
                int i;
                int len;
                for( i = 0 ; i < nb ; i++ ) {
                    len = lengths[i] + 1;
                    c_str = (char*) malloc( len );
                    MPI_Recv( c_str, len, MPI_CHAR, ROOT, TAG_ADD, comm, &status );
                    c_str[len-1] = '\0';    // The master sends C++ strings, which do not contain the final '\0'
                    results_s.push_back( std::string( c_str ) );
                    free( c_str );
                }

                /* Delinearize all the expressions and add them */

Camille Coti's avatar
Camille Coti committed
186
                double t1 = getTime();
Camille Coti's avatar
Camille Coti committed
187
                Tens = add_expressions( results_s, symbols );
Camille Coti's avatar
Camille Coti committed
188
189
                std::cout << "Addition: " <<  getTime() - t1 << std::endl;

Camille Coti's avatar
Camille Coti committed
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
                /* Send the result */

                send_result( Tens );

            } else {
                if( status.MPI_TAG == TAG_END ){
                    return;
                } else {
                    std::cerr << "Wrong tag received on slave " << status.MPI_TAG << std::endl;
                }
            }
        }
    }
}

/* Communication protocol:
   M -> W: always the same size, therefore unique communication
   W -> M: send an unsigned int (size of the expression), then the expression (table of chars)
*/
        
gi::ex multiply_1level_mw_addslave2( tensor3D_t& T, int size ) {  // simpler: same dimension everywhere
    int rank;
    gi::ex Tens = 0;
    MPI_Comm_rank( MPI_COMM_WORLD, &rank );

    /* Create a new datatype for the parameters */
    
    create_parameters_datatype_s();

    /* Here we go */

    if( 0 == rank ) {
        Tens = multiply_1level_master_addslave2( T, size );
    } else {
        multiply_1level_slave_addslave2( T, size );
    }

    /* Finalize */
    
    free_parameters_s_dt();
    return Tens;
}