engine.c 11.6 KB
Newer Older
Pedro Gonnet's avatar
Pedro Gonnet 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
/*******************************************************************************
 * This file is part of GadgetSMP.
 * Coypright (c) 2012 Pedro Gonnet (pedro.gonnet@durham.ac.uk)
 * 
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published
 * by the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 * 
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 * 
 * You should have received a copy of the GNU Lesser General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 * 
 ******************************************************************************/

/* Config parameters. */
#include "../config.h"

/* Some standard headers. */
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <pthread.h>
#include <math.h>
#include <float.h>
#include <limits.h>
#include <omp.h>
#include <sched.h>

/* Local headers. */
#include "cycle.h"
#include "const.h"
#include "lock.h"
#include "task.h"
#include "part.h"
#include "cell.h"
#include "space.h"
#include "queue.h"
#include "engine.h"
#include "runner.h"
#include "runner_iact.h"

/* Error macro. */
48
#define error(s) { fprintf( stderr , "%s:%s:%i: %s\n" , __FILE__ , __FUNCTION__ , __LINE__ , s ); abort(); }
Pedro Gonnet's avatar
Pedro Gonnet committed
49
50
51
52
53

/* Convert cell location to ID. */
#define cell_getid( cdim , i , j , k ) ( (int)(k) + (cdim)[2]*( (int)(j) + (cdim)[1]*(int)(i) ) )


54
55
56
57
58
59
60
61
62
/**
 * @brief Prepare the #engine by re-building the cells and tasks.
 *
 * @param e The #engine to prepare.
 * @param force Flag to force re-building the cell and task structure.
 */
 
void engine_prepare ( struct engine *e , int force ) {

63
    int j, k, qid, changes, count;
64
    struct space *s = e->s;
65
    // ticks tic;
66
67

    /* Rebuild the space. */
68
    // tic = getticks();
Pedro Gonnet's avatar
Pedro Gonnet committed
69
    changes = space_rebuild( e->s , force , 0 );
70
    // printf( "engine_prepare: space_rebuild with %i changes took %.3f ms.\n" , changes , (double)(getticks() - tic) / CPU_TPS * 1000 );
71
72
    
    /* Has anything changed? */
73
    // tic = getticks();
74
75
76
77
78
79
    if ( changes ) {
    
        /* Rank the tasks in topological order. */
        engine_ranktasks( e );
    
        /* Fill the queues (round-robin). */
80
81
82
83
84
85
86
87
88
89
90
        #pragma omp parallel for schedule(static) private(count,k)
        for ( qid = 0 ; qid < e->nr_queues ; qid++ ) {
            queue_init( &e->queues[qid] , s->nr_tasks , s->tasks );
            for ( count = 0 , k = qid ; k < s->nr_tasks ; k += e->nr_queues ) {
                if ( s->tasks[ s->tasks_ind[k] ].type == task_type_none )
                    continue;
                e->queues[qid].tid[ count ] = s->tasks_ind[k];
                count += 1;
                }
            e->queues[qid].count = count;
            e->queues[qid].next = 0;
91
92
93
            }
            
        }
94
    // printf( "engine_prepare: re-filling queues took %.3f ms.\n" , (double)(getticks() - tic) / CPU_TPS * 1000 );
95

96
    /* Re-set the particle data. */
97
98
    // tic = getticks();
    #pragma omp parallel for schedule(static) 
99
100
101
102
103
104
    for ( k = 0 ; k < s->nr_parts ; k++ ) {
        s->parts[k].wcount = 0.0f;
        s->parts[k].wcount_dh = 0.0f;
        s->parts[k].rho = 0.0f;
        s->parts[k].rho_dh = 0.0f;
        }
105
    // printf( "engine_prepare: re-setting particle data took %.3f ms.\n" , (double)(getticks() - tic) / CPU_TPS * 1000 );
106
107
    
    /* Run throught the tasks and get all the waits right. */
108
109
    // tic = getticks();
    #pragma omp parallel for schedule(static) private(j)
110
111
112
113
114
115
    for ( k = 0 ; k < s->nr_tasks ; k++ ) {
        for ( j = 0 ; j < s->tasks[k].nr_unlock_tasks ; j++ )
            __sync_add_and_fetch( &s->tasks[k].unlock_tasks[j]->wait , 1 );
        for ( j = 0 ; j < s->tasks[k].nr_unlock_cells ; j++ )
            __sync_add_and_fetch( &s->tasks[k].unlock_cells[j]->wait , 1 );
        }
116
    // printf( "engine_prepare: preparing task dependencies took %.3f ms.\n" , (double)(getticks() - tic) / CPU_TPS * 1000 );
117
118
119
120
121
    
    /* Re-set the queues.*/
    for ( k = 0 ; k < e->nr_queues ; k++ )
        e->queues[k].next = 0;
    
122
123
124
    }


Pedro Gonnet's avatar
Pedro Gonnet committed
125
126
127
128
129
130
131
132
133
134
135
/** 
 * @brief Sort the tasks in topological order over all queues.
 *
 * @param e The #engine.
 */
 
void engine_ranktasks ( struct engine *e ) {

    int i, j = 0, k, temp, left = 0, rank;
    struct task *t;
    struct space *s = e->s;
136
    int *tid = s->tasks_ind;
137
    
Pedro Gonnet's avatar
Pedro Gonnet committed
138
139
    /* Run throught the tasks and get all the waits right. */
    for ( k = 0 ; k < s->nr_tasks ; k++ ) {
140
        tid[k] = k;
Pedro Gonnet's avatar
Pedro Gonnet committed
141
142
143
144
145
        for ( j = 0 ; j < s->tasks[k].nr_unlock_tasks ; j++ )
            s->tasks[k].unlock_tasks[j]->wait += 1;
        }
        
    /* Main loop. */
146
    for ( j = 0 , rank = 0 ; left < s->nr_tasks ; rank++ ) {
Pedro Gonnet's avatar
Pedro Gonnet committed
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
225
        
        /* Load the tids of tasks with no waits. */
        for ( k = left ; k < s->nr_tasks ; k++ )
            if ( s->tasks[ tid[k] ].wait == 0 ) {
                temp = tid[j]; tid[j] = tid[k]; tid[k] = temp;
                j += 1;
                }

        /* Traverse the task tree and add tasks with no weight. */
        for ( i = left ; i < j ; i++ ) {
            t = &s->tasks[ tid[i] ];
            t->rank = rank;
            s->tasks_ind[i] = t - s->tasks;
            /* printf( "engine_ranktasks: task %i of type %s has rank %i.\n" , i , 
                (t->type == task_type_self) ? "self" : (t->type == task_type_pair) ? "pair" : "sort" , rank ); */
            for ( k = 0 ; k < t->nr_unlock_tasks ; k++ )
                t->unlock_tasks[k]->wait -= 1;
            }
            
        /* The new left (no, not tony). */
        left = j;
            
        }
        
    }


/**
 * @brief Implements a barrier for the #runner threads.
 *
 * @param e The #engine.
 */
 
void engine_barrier( struct engine *e ) {

    /* First, get the barrier mutex. */
    if ( pthread_mutex_lock( &e->barrier_mutex ) != 0 )
        error( "Failed to get barrier mutex." );
        
    /* Wait for the barrier to close. */
    while ( e->barrier_count < 0 )
        if ( pthread_cond_wait( &e->barrier_cond , &e->barrier_mutex ) != 0 )
            error( "Eror waiting for barrier to close." );
        
    /* Once I'm in, increase the barrier count. */
    e->barrier_count += 1;
    
    /* If all threads are in, send a signal... */
    if ( e->barrier_count == e->nr_threads )
        if ( pthread_cond_broadcast( &e->barrier_cond ) != 0 )
            error( "Failed to broadcast barrier full condition." );
        
    /* Wait for barrier to be released. */
    while ( e->barrier_count > 0 )
        if ( pthread_cond_wait( &e->barrier_cond , &e->barrier_mutex ) != 0 )
            error( "Error waiting for barrier to be released." );
            
    /* Decrease the counter before leaving... */
    e->barrier_count += 1;
    
    /* If I'm the last one out, signal the condition again. */
    if ( e->barrier_count == 0 )
        if ( pthread_cond_broadcast( &e->barrier_cond ) != 0 )
            error( "Failed to broadcast empty barrier condition." );
            
    /* Last but not least, release the mutex. */
    if ( pthread_mutex_unlock( &e->barrier_mutex ) != 0 )
        error( "Failed to get unlock the barrier mutex." );

    }
    
    
/**
 * @brief Let the #engine loose to compute the forces.
 *
 * @param e The #engine.
 * @param sort_queues Flag to try to sort the queues topologically.
 */
 
226
void engine_run ( struct engine *e , int sort_queues , float dt_max ) {
Pedro Gonnet's avatar
Pedro Gonnet committed
227

228
229
    int k;

Pedro Gonnet's avatar
Pedro Gonnet committed
230
231
232
233
234
235
236
237
    /* Re-set the queues.*/
    if ( sort_queues ) {
        #pragma omp parallel for default(none), shared(e)
        for ( k = 0 ; k < e->nr_queues ; k++ ) {
            queue_sort( &e->queues[k] );
            e->queues[k].next = 0;
            }
        }
238
239
240
241
        
    /* Set the maximum dt. */
    e->dt_max = dt_max;
    e->s->dt_max = dt_max;
Pedro Gonnet's avatar
Pedro Gonnet committed
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
    
    /* Cry havoc and let loose the dogs of war. */
    e->barrier_count = -e->barrier_count;
    if ( pthread_cond_broadcast( &e->barrier_cond ) != 0 )
        error( "Failed to broadcast barrier open condition." );
        
    /* Sit back and wait for the runners to come home. */
    while ( e->barrier_count < e->nr_threads )
        if ( pthread_cond_wait( &e->barrier_cond , &e->barrier_mutex ) != 0 )
            error( "Error while waiting for barrier." );
    
    }
    
    
/**
 * @brief init an engine with the given number of threads, queues, and
 *      the given policy.
 *
 * @param e The #engine.
 * @param s The #space in which this #runner will run.
 * @param nr_threads The number of threads to spawn.
 * @param nr_queues The number of task queues to create.
 * @param policy The queueing policy to use.
 */
 
void engine_init ( struct engine *e , struct space *s , int nr_threads , int nr_queues , int policy ) {

    #if defined(HAVE_SETAFFINITY)
        cpu_set_t cpuset;
    #endif
    int k, qid, nrq;
    
    /* Store the values. */
    e->s = s;
    e->nr_threads = nr_threads;
    e->nr_queues = nr_queues;
    e->policy = policy;
    
    /* First of all, init the barrier and lock it. */
    if ( pthread_mutex_init( &e->barrier_mutex , NULL ) != 0 )
        error( "Failed to initialize barrier mutex." );
    if ( pthread_cond_init( &e->barrier_cond , NULL ) != 0 )
        error( "Failed to initialize barrier condition variable." );
    if ( pthread_mutex_lock( &e->barrier_mutex ) != 0 )
        error( "Failed to lock barrier mutex." );
    e->barrier_count = 0;
    
    /* Allocate the queues. */
    if ( posix_memalign( (void *)(&e->queues) , 64 , nr_queues * sizeof(struct queue) ) != 0 )
        error( "Failed to allocate queues." );
    bzero( e->queues , nr_queues * sizeof(struct queue) );
        
    /* Init the queues. */
    for ( k = 0 ; k < nr_queues ; k++ )
        queue_init( &e->queues[k] , s->nr_tasks , s->tasks );
        
    /* Rank the tasks in topological order. */
    engine_ranktasks( e );
    
    /* How many queues to fill initially? */
    for ( nrq = 0 , k = nr_queues ; k > 0 ; k = k / 2 )
        nrq += 1;
        
    /* Fill the queues (round-robin). */
    for ( k = 0 ; k < s->nr_tasks ; k++ ) {
        if ( s->tasks[ s->tasks_ind[k] ].type == task_type_none )
            continue;
        // qid = 0;
        // qid = k % nrq;
        qid = k % nr_queues;
        e->queues[qid].tid[ e->queues[qid].count ] = s->tasks_ind[k];
        e->queues[qid].count += 1;
        }
        
    /* Sort the queues topologically. */
317
318
    // for ( k = 0 ; k < nr_queues ; k++ )
    //     queue_sort( &e->queues[k] );
Pedro Gonnet's avatar
Pedro Gonnet committed
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
        
    /* Allocate and init the threads. */
    if ( ( e->runners = (struct runner *)malloc( sizeof(struct runner) * nr_threads ) ) == NULL )
        error( "Failed to allocate threads array." );
    for ( k = 0 ; k < nr_threads ; k++ ) {
        e->runners[k].id = k;
        e->runners[k].e = e;
        if ( pthread_create( &e->runners[k].thread , NULL , &runner_main , &e->runners[k] ) != 0 )
            error( "Failed to create runner thread." );
        #if defined(HAVE_SETAFFINITY)
            /* Set the cpu mask to zero | e->id. */
            CPU_ZERO( &cpuset );
            CPU_SET( e->runners[k].id , &cpuset );

            /* Apply this mask to the runner's pthread. */
            if ( pthread_setaffinity_np( e->runners[k].thread , sizeof(cpu_set_t) , &cpuset ) != 0 )
                error( "Failed to set thread affinity." );
        #endif
        }
        
    /* Wait for the runner threads to be in place. */
    while ( e->barrier_count != e->nr_threads )
        if ( pthread_cond_wait( &e->barrier_cond , &e->barrier_mutex ) != 0 )
            error( "Error while waiting for runner threads to get in place." );
    
    }