space.c 38.1 KB
Newer Older
Pedro Gonnet's avatar
Pedro Gonnet committed
1
/*******************************************************************************
2
 * This file is part of SWIFT.
Pedro Gonnet's avatar
Pedro Gonnet committed
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
 * 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 <float.h>
#include <limits.h>
#include <math.h>
31
#include <omp.h>
Pedro Gonnet's avatar
Pedro Gonnet committed
32

33
34
35
36
37
/* MPI headers. */
#ifdef WITH_MPI
    #include <mpi.h>
#endif

Pedro Gonnet's avatar
Pedro Gonnet committed
38
/* Local headers. */
39
#include "const.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
40
#include "cycle.h"
41
#include "atomic.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
42
#include "lock.h"
43
#include "task.h"
44
#include "kernel.h"
45
#include "part.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
46
#include "space.h"
47
#include "multipole.h"
48
49
50
#include "cell.h"
#include "scheduler.h"
#include "engine.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
51
#include "runner.h"
52
#include "error.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
53
54
55

/* Split size. */
int space_splitsize = space_splitsize_default;
56
int space_subsize = space_subsize_default;
57
int space_maxsize = space_maxsize_default;
Pedro Gonnet's avatar
Pedro Gonnet committed
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

/* Map shift vector to sortlist. */
const int sortlistID[27] = {
    /* ( -1 , -1 , -1 ) */   0 ,
    /* ( -1 , -1 ,  0 ) */   1 , 
    /* ( -1 , -1 ,  1 ) */   2 ,
    /* ( -1 ,  0 , -1 ) */   3 ,
    /* ( -1 ,  0 ,  0 ) */   4 , 
    /* ( -1 ,  0 ,  1 ) */   5 ,
    /* ( -1 ,  1 , -1 ) */   6 ,
    /* ( -1 ,  1 ,  0 ) */   7 , 
    /* ( -1 ,  1 ,  1 ) */   8 ,
    /* (  0 , -1 , -1 ) */   9 ,
    /* (  0 , -1 ,  0 ) */   10 , 
    /* (  0 , -1 ,  1 ) */   11 ,
    /* (  0 ,  0 , -1 ) */   12 ,
    /* (  0 ,  0 ,  0 ) */   0 , 
    /* (  0 ,  0 ,  1 ) */   12 ,
    /* (  0 ,  1 , -1 ) */   11 ,
    /* (  0 ,  1 ,  0 ) */   10 , 
    /* (  0 ,  1 ,  1 ) */   9 ,
    /* (  1 , -1 , -1 ) */   8 ,
    /* (  1 , -1 ,  0 ) */   7 , 
    /* (  1 , -1 ,  1 ) */   6 ,
    /* (  1 ,  0 , -1 ) */   5 ,
    /* (  1 ,  0 ,  0 ) */   4 , 
    /* (  1 ,  0 ,  1 ) */   3 ,
    /* (  1 ,  1 , -1 ) */   2 ,
    /* (  1 ,  1 ,  0 ) */   1 , 
    /* (  1 ,  1 ,  1 ) */   0 
    };
89
90
    
    
91
92
93
94
95
96
97
98
99
100
101
102
103
104
/**
 * @brief Get the shift-id of the given pair of cells, swapping them
 *      if need be.
 *
 * @param s The space
 * @param ci Pointer to first #cell.
 * @param cj Pointer second #cell.
 * @param shift Vector from ci to cj.
 *
 * @return The shift ID and set shift, may or may not swap ci and cj.
 */
 
int space_getsid ( struct space *s , struct cell **ci , struct cell **cj , double *shift ) {

105
    int k, sid = 0, periodic = s->periodic;
106
107
108
109
110
111
    struct cell *temp;
    double dx[3];

    /* Get the relative distance between the pairs, wrapping. */
    for ( k = 0 ; k < 3 ; k++ ) {
        dx[k] = (*cj)->loc[k] - (*ci)->loc[k];
112
        if ( periodic && dx[k] < -s->dim[k]/2 )
113
            shift[k] = s->dim[k];
114
        else if ( periodic && dx[k] > s->dim[k]/2 )
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
            shift[k] = -s->dim[k];
        else
            shift[k] = 0.0;
        dx[k] += shift[k];
        }
        
    /* Get the sorting index. */
    for ( k = 0 ; k < 3 ; k++ )
        sid = 3*sid + ( (dx[k] < 0.0) ? 0 : ( (dx[k] > 0.0) ? 2 : 1 ) );

    /* Switch the cells around? */
    if ( runner_flip[sid] ) {
        temp = *ci; *ci = *cj; *cj = temp;
        for ( k = 0 ; k < 3 ; k++ )
            shift[k] = -shift[k];
        }
    sid = sortlistID[sid];
    
    /* Return the sort ID. */
    return sid;

    }


139
/**
140
 * @brief Recursively dismantle a cell tree.
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
 *
 */
 
void space_rebuild_recycle ( struct space *s , struct cell *c ) {
    
    int k;
    
    if ( c->split )
        for ( k = 0 ; k < 8 ; k++ )
            if ( c->progeny[k] != NULL ) {
                space_rebuild_recycle( s , c->progeny[k] );
                space_recycle( s , c->progeny[k] );
                c->progeny[k] = NULL;
                }
    
    }
157
158
    
    
159
/**
160
 * @brief Re-build the cell grid.
161
 *
162
163
 * @param s The #space.
 * @param cell_max Maximum cell edge length.
164
165
 */
 
166
void space_regrid ( struct space *s , double cell_max ) {
167

168
    float h_max = s->cell_min / kernel_gamma / space_stretch, dmin;
169
    int i, j, k, cdim[3], nr_parts = s->nr_parts;
170
    struct cell *restrict c;
171
    // ticks tic;
172
173
    
    /* Run through the parts and get the current h_max. */
174
    // tic = getticks();
175
176
177
178
179
180
181
182
183
184
185
186
    if ( s->cells != NULL ) {
        for ( k = 0 ; k < s->nr_cells ; k++ ) {
            if ( s->cells[k].h_max > h_max )
                h_max = s->cells[k].h_max;
            }
        }
    else {
        for ( k = 0 ; k < nr_parts ; k++ ) {
            if ( s->parts[k].h > h_max )
                h_max = s->parts[k].h;
            }
        s->h_max = h_max;
187
        }
188
189
    // message( "h_max is %.3e (cell_max=%.3e)." , h_max , cell_max );
    // message( "getting h_min and h_max took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
190
191
192
    
    /* Get the new putative cell dimensions. */
    for ( k = 0 ; k < 3 ; k++ )
193
        cdim[k] = floor( s->dim[k] / fmax( h_max*kernel_gamma*space_stretch , cell_max ) );
194
        
195
196
197
198
199
200
    /* In MPI-Land, we're not allowed to change the top-level cell size. */
    #ifdef WITH_MPI
        if ( cdim[0] < s->cdim[0] || cdim[1] < s->cdim[1] || cdim[2] < s->cdim[2] )
            error( "Root-level change of cell size not allowed." );
    #endif
        
201
    /* Do we need to re-build the upper-level cells? */
202
    // tic = getticks();
Pedro Gonnet's avatar
Pedro Gonnet committed
203
    if ( s->cells == NULL ||
204
         cdim[0] < s->cdim[0] || cdim[1] < s->cdim[1] || cdim[2] < s->cdim[2] ) {
205
206
207
    
        /* Free the old cells, if they were allocated. */
        if ( s->cells != NULL ) {
208
            for ( k = 0 ; k < s->nr_cells ; k++ ) {
209
                space_rebuild_recycle( s , &s->cells[k] );
210
211
212
                if ( s->cells[k].sort != NULL )
                    free( s->cells[k].sort );
                }
213
214
215
216
            free( s->cells );
            s->maxdepth = 0;
            }
            
217
        /* Set the new cell dimensions only if smaller. */
218
219
220
221
222
        for ( k = 0 ; k < 3 ; k++ ) {
            s->cdim[k] = cdim[k];
            s->h[k] = s->dim[k] / cdim[k];
            s->ih[k] = 1.0 / s->h[k];
            }
223
        dmin = fminf( s->h[0] , fminf( s->h[1] , s->h[2] ) );
224

225
        /* Allocate the highest level of cells. */
Pedro Gonnet's avatar
Pedro Gonnet committed
226
        s->tot_cells = s->nr_cells = cdim[0] * cdim[1] * cdim[2];
227
228
229
230
231
232
233
234
235
236
237
238
239
240
        if ( posix_memalign( (void *)&s->cells , 64 , s->nr_cells * sizeof(struct cell) ) != 0 )
            error( "Failed to allocate cells." );
        bzero( s->cells , s->nr_cells * sizeof(struct cell) );
        for ( k = 0 ; k < s->nr_cells ; k++ )
            if ( lock_init( &s->cells[k].lock ) != 0 )
                error( "Failed to init spinlock." );

        /* Set the cell location and sizes. */
        for ( i = 0 ; i < cdim[0] ; i++ )
            for ( j = 0 ; j < cdim[1] ; j++ )
                for ( k = 0 ; k < cdim[2] ; k++ ) {
                    c = &s->cells[ cell_getid( cdim , i , j , k ) ];
                    c->loc[0] = i*s->h[0]; c->loc[1] = j*s->h[1]; c->loc[2] = k*s->h[2];
                    c->h[0] = s->h[0]; c->h[1] = s->h[1]; c->h[2] = s->h[2];
241
                    c->dmin = dmin;
242
                    c->depth = 0;
Pedro Gonnet's avatar
Pedro Gonnet committed
243
                    c->count = 0;
244
                    c->gcount = 0;
245
                    c->super = c;
246
                    lock_init( &c->lock );
247
                    }
248
249
           
        /* Be verbose about the change. */         
250
        message( "set cell dimensions to [ %i %i %i ]." , cdim[0] , cdim[1] , cdim[2] ); fflush(stdout);
251
252
                    
        } /* re-build upper-level cells? */
253
    // message( "rebuilding upper-level cells took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
254
        
Pedro Gonnet's avatar
Pedro Gonnet committed
255
256
257
258
259
260
261
262
263
    /* Otherwise, just clean up the cells. */
    else {
    
        /* Free the old cells, if they were allocated. */
        for ( k = 0 ; k < s->nr_cells ; k++ ) {
            space_rebuild_recycle( s , &s->cells[k] );
            s->cells[k].sorts = NULL;
            s->cells[k].nr_tasks = 0;
            s->cells[k].nr_density = 0;
264
            s->cells[k].nr_force = 0;
265
266
            s->cells[k].density = NULL;
            s->cells[k].force = NULL;
Pedro Gonnet's avatar
Pedro Gonnet committed
267
            s->cells[k].dx_max = 0.0f;
268
            s->cells[k].sorted = 0;
Pedro Gonnet's avatar
Pedro Gonnet committed
269
            s->cells[k].count = 0;
270
            s->cells[k].gcount = 0;
271
            s->cells[k].kick1 = NULL;
Pedro Gonnet's avatar
Pedro Gonnet committed
272
            s->cells[k].kick2 = NULL;
273
            s->cells[k].super = &s->cells[k];
Pedro Gonnet's avatar
Pedro Gonnet committed
274
275
276
277
            }
        s->maxdepth = 0;
    
        }
278
        
279
280
281
282
283
284
285
286
287
288
289
290
291
    }
    

/**
 * @brief Re-build the cells as well as the tasks.
 *
 * @param s The #space in which to update the cells.
 * @param cell_max Maximal cell size.
 *
 */
 
void space_rebuild ( struct space *s , double cell_max ) {

292
    int j, k, cdim[3], nr_parts = s->nr_parts, nr_gparts = s->nr_gparts;
293
294
295
    struct cell *restrict c, *restrict cells = s->cells;
    struct part *restrict finger, *restrict p, *parts = s->parts;
    struct xpart *xfinger, *xparts = s->xparts;
296
    struct gpart *gp, *gparts = s->gparts, *gfinger;
297
298
299
300
301
302
303
    int *ind;
    double ih[3], dim[3];
    // ticks tic;
    
    /* Be verbose about this. */
    // message( "re)building space..." ); fflush(stdout);
    
304
305
    /* Re-grid if necessary, or just re-set the cell data. */
    space_regrid( s , cell_max );
306
        
307
    /* Run through the particles and get their cell index. */
308
    // tic = getticks();
309
    if ( ( ind = (int *)malloc( sizeof(int) * s->size_parts ) ) == NULL )
310
        error( "Failed to allocate temporary particle indices." );
Pedro Gonnet's avatar
Pedro Gonnet committed
311
312
    ih[0] = s->ih[0]; ih[1] = s->ih[1]; ih[2] = s->ih[2];
    dim[0] = s->dim[0]; dim[1] = s->dim[1]; dim[2] = s->dim[2];
Pedro Gonnet's avatar
Pedro Gonnet committed
313
    cdim[0] = s->cdim[0]; cdim[1] = s->cdim[1]; cdim[2] = s->cdim[2];
314
    #pragma omp parallel for private(p,j)
315
316
    for ( k = 0 ; k < nr_parts ; k++ )  {
        p = &parts[k];
317
318
        for ( j = 0 ; j < 3 ; j++ )
            if ( p->x[j] < 0.0 )
Pedro Gonnet's avatar
Pedro Gonnet committed
319
320
321
322
                p->x[j] += dim[j];
            else if ( p->x[j] >= dim[j] )
                p->x[j] -= dim[j];
        ind[k] = cell_getid( cdim , p->x[0]*ih[0] , p->x[1]*ih[1] , p->x[2]*ih[2] );
323
        atomic_inc( &cells[ ind[k] ].count );
324
        }
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
    // message( "getting particle indices took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );


    /* Move non-local parts to the end of the list. */
    #ifdef WITH_MPI
        int nodeID = s->e->nodeID;
        for ( k = 0 ; k < nr_parts ; k++ )
            if ( cells[ ind[k] ].nodeID != nodeID ) {
                cells[ ind[k] ].count -= 1;
                nr_parts -= 1;
                struct part tp = parts[k];
                parts[k] = parts[ nr_parts ];
                parts[ nr_parts ] = tp;
                struct xpart txp = xparts[k];
                xparts[k] = xparts[ nr_parts ];
                xparts[ nr_parts ] = txp;
                int t = ind[k];
                ind[k] = ind[ nr_parts ];
                ind[ nr_parts ] = t;
                }
        s->nr_parts = nr_parts + engine_exchange_strays( s->e , &parts[nr_parts] , &xparts[nr_parts] , &ind[nr_parts] , s->nr_parts - nr_parts );
        for ( k = nr_parts ; k < s->nr_parts ; k++ ) {
            p = &parts[k];
            ind[k] = cell_getid( cdim , p->x[0]*ih[0] , p->x[1]*ih[1] , p->x[2]*ih[2] );
            cells[ ind[k] ].count += 1;
350
351
            /* if ( cells[ ind[k] ].nodeID != nodeID )
                error( "Received part that does not belong to me (nodeID=%i)." , cells[ ind[k] ].nodeID ); */
352
353
354
355
            }
        nr_parts = s->nr_parts;
    #endif
    
356
357

    /* Sort the parts according to their cells. */
358
    // tic = getticks();
359
360
    parts_sort( parts , xparts , ind , nr_parts , 0 , s->nr_cells-1 );
    // message( "parts_sort took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
361
    
362
363
364
365
366
    /* Re-link the gparts. */
    for ( k = 0 ; k < nr_parts ; k++ )
        if ( parts[k].gpart != NULL )
            parts[k].gpart->part = &parts[k];
    
Pedro Gonnet's avatar
Pedro Gonnet committed
367
368
369
370
371
372
373
374
375
    /* Verify sort. */
    /* for ( k = 1 ; k < nr_parts ; k++ ) {
        if ( ind[k-1] > ind[k] ) {
            error( "Sort failed!" );
            }
        else if ( ind[k] != cell_getid( cdim , parts[k].x[0]*ih[0] , parts[k].x[1]*ih[1] , parts[k].x[2]*ih[2] ) )
            error( "Incorrect indices!" );
        } */
    
376
377
    /* We no longer need the indices as of here. */
    free( ind );    
378

379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414


    /* Run through the gravity particles and get their cell index. */
    // tic = getticks();
    if ( ( ind = (int *)malloc( sizeof(int) * s->size_gparts ) ) == NULL )
        error( "Failed to allocate temporary particle indices." );
    #pragma omp parallel for private(gp,j)
    for ( k = 0 ; k < nr_gparts ; k++ )  {
        gp = &gparts[k];
        for ( j = 0 ; j < 3 ; j++ )
            if ( gp->x[j] < 0.0 )
                gp->x[j] += dim[j];
            else if ( gp->x[j] >= dim[j] )
                gp->x[j] -= dim[j];
        ind[k] = cell_getid( cdim , gp->x[0]*ih[0] , gp->x[1]*ih[1] , gp->x[2]*ih[2] );
        atomic_inc( &cells[ ind[k] ].gcount );
        }
    // message( "getting particle indices took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );

    /* TODO: Here we should exchange the gparts as well! */

    /* Sort the parts according to their cells. */
    // tic = getticks();
    gparts_sort( gparts ,ind , nr_parts , 0 , s->nr_cells-1 );
    // message( "gparts_sort took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
    
    /* Re-link the parts. */
    for ( k = 0 ; k < nr_gparts ; k++ )
        if ( gparts[k].id > 0 )
            gparts[k].part->gpart = &gparts[k];

    /* We no longer need the indices as of here. */
    free( ind );    



415
    /* Hook the cells up to the parts. */
416
    // tic = getticks();
417
418
    finger = parts;
    xfinger = xparts;
419
    gfinger = gparts;
420
    for ( k = 0 ; k < s->nr_cells ; k++ ) {
421
        c = &cells[ k ];
422
        c->parts = finger;
423
        c->xparts = xfinger;
424
        c->gparts = gfinger;
425
        finger = &finger[ c->count ];
426
        xfinger = &xfinger[ c->count ];
427
        gfinger = &gfinger[ c->gcount ];
428
        }
429
    // message( "hooking up cells took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
430
        
431
432
    /* At this point, we have the upper-level cells, old or new. Now make
       sure that the parts in each cell are ok. */
433
    // tic = getticks();
Pedro Gonnet's avatar
Pedro Gonnet committed
434
    k = 0;
435
    #pragma omp parallel shared(s,k)
Pedro Gonnet's avatar
Pedro Gonnet committed
436
    {
437
438
439
440
        if ( omp_get_thread_num() < 8 )
            while ( 1 ) {
                int myk = atomic_inc( &k );
                if ( myk < s->nr_cells )
441
                    space_split( s , &cells[myk] );
442
443
444
                else
                    break;
                }
Pedro Gonnet's avatar
Pedro Gonnet committed
445
        }
446
447
    // message( "space_split took %.3f ms." , (double)(getticks() - tic) / CPU_TPS * 1000 );
    
448
449
450
    }


451
/**
452
 * @brief Sort the particles and condensed particles according to the given indices.
453
454
455
456
457
458
459
460
 *
 * @param parts The list of #part
 * @param ind The indices with respect to which the parts are sorted.
 * @param N The number of parts
 * @param min Lowest index.
 * @param max highest index.
 */
 
461
void parts_sort ( struct part *parts , struct xpart *xparts , int *ind , int N , int min , int max ) {
462

463
    struct qstack {
464
        volatile int i, j, min, max;
465
        volatile int ready;
466
467
        };
    struct qstack *qstack;
468
    int qstack_size = 2*(max-min) + 10;
469
    volatile unsigned int first, last, waiting;
470
471
472
    
    int pivot;
    int i, ii, j, jj, temp_i, qid;
473
    struct part temp_p;
474
    struct xpart temp_xp;
475
476
477
478

    /* for ( int k = 0 ; k < N ; k++ )
        if ( ind[k] > max || ind[k] < min )
	    error( "ind[%i]=%i is not in [%i,%i]." , k , ind[k] , min , max ); */
479
    
480
481
482
483
    /* Allocate the stack. */
    if ( ( qstack = malloc( sizeof(struct qstack) * qstack_size ) ) == NULL )
        error( "Failed to allocate qstack." );
    
484
485
486
487
488
    /* Init the interval stack. */
    qstack[0].i = 0;
    qstack[0].j = N-1;
    qstack[0].min = min;
    qstack[0].max = max;
489
    qstack[0].ready = 1;
490
    for ( i = 1 ; i < qstack_size ; i++ )
491
492
        qstack[i].ready = 0;
    first = 0; last = 1; waiting = 1;
493
494
    
    /* Parallel bit. */
495
    #pragma omp parallel default(none) shared(N,first,last,waiting,qstack,parts,xparts,ind,qstack_size,stderr,engine_rank) private(pivot,i,ii,j,jj,min,max,temp_i,qid,temp_xp,temp_p)
496
    {
497
    
498
        /* Main loop. */
499
        if ( omp_get_thread_num() < 8 )
500
501
        while ( waiting > 0 ) {
        
502
            /* Grab an interval off the queue. */
503
            qid = atomic_inc( &first ) % qstack_size;
504
            
505
            /* Wait for the interval to be ready. */
506
            while ( waiting > 0 && atomic_cas( &qstack[qid].ready , 1 , 1 ) != 1 );
507
508
509
510
511
            
            /* Broke loop for all the wrong reasons? */
            if ( waiting == 0 )
                break;
        
512
513
514
515
516
            /* Get the stack entry. */
            i = qstack[qid].i;
            j = qstack[qid].j;
            min = qstack[qid].min;
            max = qstack[qid].max;
517
            qstack[qid].ready = 0;
518
            // message( "thread %i got interval [%i,%i] with values in [%i,%i]." , omp_get_thread_num() , i , j , min , max );
519
            
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
            /* Loop over sub-intervals. */
            while ( 1 ) {
            
                /* Bring beer. */
                pivot = (min + max) / 2;
                
                /* One pass of QuickSort's partitioning. */
                ii = i; jj = j;
                while ( ii < jj ) {
                    while ( ii <= j && ind[ii] <= pivot )
                        ii++;
                    while ( jj >= i && ind[jj] > pivot )
                        jj--;
                    if ( ii < jj ) {
                        temp_i = ind[ii]; ind[ii] = ind[jj]; ind[jj] = temp_i;
                        temp_p = parts[ii]; parts[ii] = parts[jj]; parts[jj] = temp_p;
536
                        temp_xp = xparts[ii]; xparts[ii] = xparts[jj]; xparts[jj] = temp_xp;
537
                        }
538
                    }
539

540
541
542
                /* Verify sort. */
                /* for ( int k = i ; k <= jj ; k++ )
                    if ( ind[k] > pivot ) {
543
                        message( "sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i, N=%i." , k , ind[k] , pivot , i , j , N );
544
545
546
547
                        error( "Partition failed (<=pivot)." );
                        }
                for ( int k = jj+1 ; k <= j ; k++ )
                    if ( ind[k] <= pivot ) {
548
                        message( "sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i, N=%i." , k , ind[k] , pivot , i , j , N );
549
550
551
552
553
554
555
556
                        error( "Partition failed (>pivot)." );
                        } */
                        
                /* Split-off largest interval. */
                if ( jj - i > j - jj+1 ) {

                    /* Recurse on the left? */
                    if ( jj > i  && pivot > min ) {
557
                        qid = atomic_inc( &last ) % qstack_size;
558
                        while ( atomic_cas( &qstack[qid].ready , 0 , 0 ) != 0 );
559
560
561
562
563
                        qstack[qid].i = i;
                        qstack[qid].j = jj;
                        qstack[qid].min = min;
                        qstack[qid].max = pivot;
                        qstack[qid].ready = 1;
564
565
                        if ( atomic_inc( &waiting ) >= qstack_size )
                            error( "Qstack overflow." );
566
567
568
569
570
571
572
573
574
575
                        }

                    /* Recurse on the right? */
                    if ( jj+1 < j && pivot+1 < max ) {
                        i = jj+1;
                        min = pivot+1;
                        }
                    else
                        break;
                        
576
                    }
577
578
579
580
581
                    
                else {
                
                    /* Recurse on the right? */
                    if ( jj+1 < j && pivot+1 < max ) {
582
                        qid = atomic_inc( &last ) % qstack_size;
583
                        while ( atomic_cas( &qstack[qid].ready , 0 , 0 ) != 0 );
584
585
586
587
588
                        qstack[qid].i = jj+1;
                        qstack[qid].j = j;
                        qstack[qid].min = pivot+1;
                        qstack[qid].max = max;
                        qstack[qid].ready = 1;
589
590
                        if ( atomic_inc( &waiting ) >= qstack_size )
                            error( "Qstack overflow." );
591
592
593
594
595
596
597
598
599
                        }
                        
                    /* Recurse on the left? */
                    if ( jj > i  && pivot > min ) {
                        j = jj;
                        max = pivot;
                        }
                    else
                        break;
600

601
602
603
                    }
                    
                } /* loop over sub-intervals. */
604
    
605
            atomic_dec( &waiting );
606

607
608
609
            } /* main loop. */
    
        } /* parallel bit. */
610
    
611
    /* Verify sort. */
612
    /* for ( i = 1 ; i < N ; i++ )
613
        if ( ind[i-1] > ind[i] )
614
            error( "Sorting failed (ind[%i]=%i,ind[%i]=%i)." , i-1 , ind[i-1] , i , ind[i] ); */
615
616
617
            
    /* Clean up. */
    free( qstack );
618

619
620
621
    }


622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
void gparts_sort ( struct gpart *gparts , int *ind , int N , int min , int max ) {

    struct qstack {
        volatile int i, j, min, max;
        volatile int ready;
        };
    struct qstack *qstack;
    int qstack_size = 2*(max-min) + 10;
    volatile unsigned int first, last, waiting;
    
    int pivot;
    int i, ii, j, jj, temp_i, qid;
    struct gpart temp_p;

    /* for ( int k = 0 ; k < N ; k++ )
        if ( ind[k] > max || ind[k] < min )
	    error( "ind[%i]=%i is not in [%i,%i]." , k , ind[k] , min , max ); */
    
    /* Allocate the stack. */
    if ( ( qstack = malloc( sizeof(struct qstack) * qstack_size ) ) == NULL )
        error( "Failed to allocate qstack." );
    
    /* Init the interval stack. */
    qstack[0].i = 0;
    qstack[0].j = N-1;
    qstack[0].min = min;
    qstack[0].max = max;
    qstack[0].ready = 1;
    for ( i = 1 ; i < qstack_size ; i++ )
        qstack[i].ready = 0;
    first = 0; last = 1; waiting = 1;
    
    /* Parallel bit. */
    #pragma omp parallel default(none) shared(N,first,last,waiting,qstack,gparts,ind,qstack_size,stderr,engine_rank) private(pivot,i,ii,j,jj,min,max,temp_i,qid,temp_p)
    {
    
        /* Main loop. */
        if ( omp_get_thread_num() < 8 )
        while ( waiting > 0 ) {
        
            /* Grab an interval off the queue. */
            qid = atomic_inc( &first ) % qstack_size;
            
            /* Wait for the interval to be ready. */
            while ( waiting > 0 && atomic_cas( &qstack[qid].ready , 1 , 1 ) != 1 );
            
            /* Broke loop for all the wrong reasons? */
            if ( waiting == 0 )
                break;
        
            /* Get the stack entry. */
            i = qstack[qid].i;
            j = qstack[qid].j;
            min = qstack[qid].min;
            max = qstack[qid].max;
            qstack[qid].ready = 0;
            // message( "thread %i got interval [%i,%i] with values in [%i,%i]." , omp_get_thread_num() , i , j , min , max );
            
            /* Loop over sub-intervals. */
            while ( 1 ) {
            
                /* Bring beer. */
                pivot = (min + max) / 2;
                
                /* One pass of QuickSort's partitioning. */
                ii = i; jj = j;
                while ( ii < jj ) {
                    while ( ii <= j && ind[ii] <= pivot )
                        ii++;
                    while ( jj >= i && ind[jj] > pivot )
                        jj--;
                    if ( ii < jj ) {
                        temp_i = ind[ii]; ind[ii] = ind[jj]; ind[jj] = temp_i;
                        temp_p = gparts[ii]; gparts[ii] = gparts[jj]; gparts[jj] = temp_p;
                        }
                    }

                /* Verify sort. */
                /* for ( int k = i ; k <= jj ; k++ )
                    if ( ind[k] > pivot ) {
                        message( "sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i, N=%i." , k , ind[k] , pivot , i , j , N );
                        error( "Partition failed (<=pivot)." );
                        }
                for ( int k = jj+1 ; k <= j ; k++ )
                    if ( ind[k] <= pivot ) {
                        message( "sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i, N=%i." , k , ind[k] , pivot , i , j , N );
                        error( "Partition failed (>pivot)." );
                        } */
                        
                /* Split-off largest interval. */
                if ( jj - i > j - jj+1 ) {

                    /* Recurse on the left? */
                    if ( jj > i  && pivot > min ) {
                        qid = atomic_inc( &last ) % qstack_size;
                        while ( atomic_cas( &qstack[qid].ready , 0 , 0 ) != 0 );
                        qstack[qid].i = i;
                        qstack[qid].j = jj;
                        qstack[qid].min = min;
                        qstack[qid].max = pivot;
                        qstack[qid].ready = 1;
                        if ( atomic_inc( &waiting ) >= qstack_size )
                            error( "Qstack overflow." );
                        }

                    /* Recurse on the right? */
                    if ( jj+1 < j && pivot+1 < max ) {
                        i = jj+1;
                        min = pivot+1;
                        }
                    else
                        break;
                        
                    }
                    
                else {
                
                    /* Recurse on the right? */
                    if ( jj+1 < j && pivot+1 < max ) {
                        qid = atomic_inc( &last ) % qstack_size;
                        while ( atomic_cas( &qstack[qid].ready , 0 , 0 ) != 0 );
                        qstack[qid].i = jj+1;
                        qstack[qid].j = j;
                        qstack[qid].min = pivot+1;
                        qstack[qid].max = max;
                        qstack[qid].ready = 1;
                        if ( atomic_inc( &waiting ) >= qstack_size )
                            error( "Qstack overflow." );
                        }
                        
                    /* Recurse on the left? */
                    if ( jj > i  && pivot > min ) {
                        j = jj;
                        max = pivot;
                        }
                    else
                        break;

                    }
                    
                } /* loop over sub-intervals. */
    
            atomic_dec( &waiting );

            } /* main loop. */
    
        } /* parallel bit. */
    
    /* Verify sort. */
    /* for ( i = 1 ; i < N ; i++ )
        if ( ind[i-1] > ind[i] )
            error( "Sorting failed (ind[%i]=%i,ind[%i]=%i)." , i-1 , ind[i-1] , i , ind[i] ); */
            
    /* Clean up. */
    free( qstack );

    }


Pedro Gonnet's avatar
Pedro Gonnet committed
781
/**
782
 * @brief Mapping function to free the sorted indices buffers.
Pedro Gonnet's avatar
Pedro Gonnet committed
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
 */

void space_map_clearsort ( struct cell *c , void *data ) {

    if ( c->sort != NULL ) {
        free( c->sort );
        c->sort = NULL;
        }

    }


/**
 * @brief Map a function to all particles in a aspace.
 *
 * @param s The #space we are working in.
799
800
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
Pedro Gonnet's avatar
Pedro Gonnet committed
801
802
803
804
 */
 
void space_map_parts ( struct space *s , void (*fun)( struct part *p , struct cell *c , void *data ) , void *data ) {

Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
805
    int cid = 0;
Pedro Gonnet's avatar
Pedro Gonnet committed
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824

    void rec_map ( struct cell *c ) {
    
        int k;
        
        /* No progeny? */
        if ( !c->split )
            for ( k = 0 ; k < c->count ; k++ )
                fun( &c->parts[k] , c , data );
                
        /* Otherwise, recurse. */
        else
            for ( k = 0 ; k < 8 ; k++ )
                if ( c->progeny[k] != NULL )
                    rec_map( c->progeny[k] );
                
        }
        
    /* Call the recursive function on all higher-level cells. */
825
826
827
828
829
830
831
    #pragma omp parallel shared(cid)
    {
        int mycid;
        while ( 1 ) {
            #pragma omp critical
            mycid = cid++;
            if ( mycid < s->nr_cells )
Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
832
                rec_map( &s->cells[mycid] );
833
834
835
836
            else
                break;
            }
        }
Pedro Gonnet's avatar
Pedro Gonnet committed
837
838
839
840
841
842
843
844

    }


/**
 * @brief Map a function to all particles in a aspace.
 *
 * @param s The #space we are working in.
845
 * @param full Map to all cells, including cells with sub-cells.
846
847
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
Pedro Gonnet's avatar
Pedro Gonnet committed
848
849
 */
 
850
851
void space_map_cells_post ( struct space *s , int full , void (*fun)( struct cell *c , void *data ) , void *data ) {

Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
852
    int cid = 0;
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870

    void rec_map ( struct cell *c ) {
    
        int k;
        
        /* Recurse. */
        if ( c->split )
            for ( k = 0 ; k < 8 ; k++ )
                if ( c->progeny[k] != NULL )
                    rec_map( c->progeny[k] );
                
        /* No progeny? */
        if ( full || !c->split )
            fun( c , data );
                
        }
        
    /* Call the recursive function on all higher-level cells. */
871
    // #pragma omp parallel shared(s,cid)
872
873
874
    {
        int mycid;
        while ( 1 ) {
875
            // #pragma omp critical
876
877
            mycid = cid++;
            if ( mycid < s->nr_cells )
Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
878
                rec_map( &s->cells[mycid] );
879
880
881
882
            else
                break;
            }
        }
883
884
885
886
887

    }


void space_map_cells_pre ( struct space *s , int full , void (*fun)( struct cell *c , void *data ) , void *data ) {
Pedro Gonnet's avatar
Pedro Gonnet committed
888

Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
889
    int cid = 0;
Pedro Gonnet's avatar
Pedro Gonnet committed
890
891
892
893
894
895

    void rec_map ( struct cell *c ) {
    
        int k;
        
        /* No progeny? */
896
        if ( full || !c->split )
Pedro Gonnet's avatar
Pedro Gonnet committed
897
898
            fun( c , data );
                
899
900
        /* Recurse. */
        if ( c->split )
Pedro Gonnet's avatar
Pedro Gonnet committed
901
902
903
904
905
906
907
            for ( k = 0 ; k < 8 ; k++ )
                if ( c->progeny[k] != NULL )
                    rec_map( c->progeny[k] );
                
        }
        
    /* Call the recursive function on all higher-level cells. */
908
    // #pragma omp parallel shared(s,cid)
Pedro Gonnet's avatar
Pedro Gonnet committed
909
910
911
    {
        int mycid;
        while ( 1 ) {
912
            // #pragma omp critical
Pedro Gonnet's avatar
Pedro Gonnet committed
913
914
            mycid = cid++;
            if ( mycid < s->nr_cells )
Pedro Gonnet's avatar
bug.    
Pedro Gonnet committed
915
                rec_map( &s->cells[mycid] );
Pedro Gonnet's avatar
Pedro Gonnet committed
916
917
918
919
            else
                break;
            }
        }
Pedro Gonnet's avatar
Pedro Gonnet committed
920
921
922
923
924
925
926
927
928
929
930
931
932

    }


/**
 * @brief Split cells that contain too many particles.
 *
 * @param s The #space we are working in.
 * @param c The #cell under consideration.
 */
 
void space_split ( struct space *s , struct cell *c ) {

933
    int k, count = c->count, gcount = c->gcount, maxdepth = 0;
934
935
936
    float h, h_max = 0.0f, dt, dt_min = c->parts[0].dt, dt_max = dt_min;
    struct cell *temp;
    struct part *p, *parts = c->parts;
937
    struct xpart *xp, *xparts = c->xparts;
938
939
940
941
942
943
    
    /* Check the depth. */
    if ( c->depth > s->maxdepth )
        s->maxdepth = c->depth;
    
    /* Split or let it be? */
944
    if ( count > space_splitsize || gcount > space_splitsize ) {
945
946
947
948
949
950
951
952
    
        /* No longer just a leaf. */
        c->split = 1;
        
        /* Create the cell's progeny. */
        for ( k = 0 ; k < 8 ; k++ ) {
            temp = space_getcell( s );
            temp->count = 0;
953
            temp->gcount = 0;
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
            temp->loc[0] = c->loc[0];
            temp->loc[1] = c->loc[1];
            temp->loc[2] = c->loc[2];
            temp->h[0] = c->h[0]/2;
            temp->h[1] = c->h[1]/2;
            temp->h[2] = c->h[2]/2;
            temp->dmin = c->dmin/2;
            if ( k & 4 )
                temp->loc[0] += temp->h[0];
            if ( k & 2 )
                temp->loc[1] += temp->h[1];
            if ( k & 1 )
                temp->loc[2] += temp->h[2];
            temp->depth = c->depth + 1;
            temp->split = 0;
            temp->h_max = 0.0;
            temp->dx_max = 0.0;
971
            temp->nodeID = c->nodeID;
972
973
974
975
976
977
978
979
980
            temp->parent = c;
            c->progeny[k] = temp;
            }
            
        /* Split the cell data. */
        cell_split( c );
            
        /* Remove any progeny with zero parts. */
        for ( k = 0 ; k < 8 ; k++ )
981
            if ( c->progeny[k]->count == 0 && c->progeny[k]->gcount == 0 ) {
982
983
984
985
986
                space_recycle( s , c->progeny[k] );
                c->progeny[k] = NULL;
                }
            else {
                space_split( s , c->progeny[k] );
Pedro Gonnet's avatar
Pedro Gonnet committed
987
988
989
990
991
                h_max = fmaxf( h_max , c->progeny[k]->h_max );
                dt_min = fminf( dt_min , c->progeny[k]->dt_min );
                dt_max = fmaxf( dt_max , c->progeny[k]->dt_max );
                if ( c->progeny[k]->maxdepth > maxdepth )
                    maxdepth = c->progeny[k]->maxdepth;
992
                }
993
994
995
996
997
998
                
        /* Set the values for this cell. */
        c->h_max = h_max;
        c->dt_min = dt_min;
        c->dt_max = dt_max;
        c->maxdepth = maxdepth;
999
1000
1001
1002
1003
1004
1005
1006
1007
                
        }
        
    /* Otherwise, collect the data for this cell. */
    else {
    
        /* Clear the progeny. */
        bzero( c->progeny , sizeof(struct cell *) * 8 );
        c->split = 0;
Pedro Gonnet's avatar
Pedro Gonnet committed
1008
        c->maxdepth = c->depth;
1009
1010
1011
1012
1013
        
        /* Get dt_min/dt_max. */
        
        for ( k = 0 ; k < count ; k++ ) {
            p = &parts[k];
1014
            xp = &xparts[k];
Pedro Gonnet's avatar
Pedro Gonnet committed
1015
1016
1017
            xp->x_old[0] = p->x[0];
            xp->x_old[1] = p->x[1];
            xp->x_old[2] = p->x[2];
Pedro Gonnet's avatar
Pedro Gonnet committed
1018
            dt = p->dt;
1019
            h = p->h;
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
            if ( h > h_max )
                h_max = h;
            if ( dt < dt_min )
                dt_min = dt;
            if ( dt > dt_max )
                dt_max = dt;
            }
        c->h_max = h_max;
        c->dt_min = dt_min;
        c->dt_max = dt_max;
            
        }
1032
1033
        
    /* Set ownership accorind to the start of the parts array. */
1034
    c->owner = ( ( c->parts - s->parts ) % s->nr_parts ) * s->nr_queues / s->nr_parts;
1035
1036
1037
1038

    }


Pedro Gonnet's avatar
Pedro Gonnet committed
1039
1040
1041
1042
1043
1044
1045
1046
1047
/**
 * @brief Return a used cell to the cell buffer.
 *
 * @param s The #space.
 * @param c The #cell.
 */
 
void space_recycle ( struct space *s , struct cell *c ) {

1048
1049
1050
    /* Lock the space. */
    lock_lock( &s->lock );
    
Pedro Gonnet's avatar
Pedro Gonnet committed
1051
1052
1053
    /* Clear the cell. */
    if ( lock_destroy( &c->lock ) != 0 )
        error( "Failed to destroy spinlock." );
1054
1055
1056
1057
        
    /* Clear this cell's sort arrays. */
    if ( c->sort != NULL )
        free( c->sort );
1058
1059
1060
        
    /* Clear the cell data. */
    bzero( c , sizeof(struct cell) );
Pedro Gonnet's avatar
Pedro Gonnet committed
1061
1062
1063
1064
1065
1066
    
    /* Hook this cell into the buffer. */
    c->next = s->cells_new;
    s->cells_new = c;
    s->tot_cells -= 1;
    
1067
1068
1069
    /* Unlock the space. */
    lock_unlock_blind( &s->lock );
    
Pedro Gonnet's avatar
Pedro Gonnet committed
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
    }


/**
 * @brief Get a new empty cell.
 *
 * @param s The #space.
 */
 
struct cell *space_getcell ( struct space *s ) {

    struct cell *c;
    int k;
    
1084
1085
1086
    /* Lock the space. */
    lock_lock( &s->lock );
    
Pedro Gonnet's avatar
Pedro Gonnet committed
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
    /* Is the buffer empty? */
    if ( s->cells_new == NULL ) {
        if ( posix_memalign( (void *)&s->cells_new , 64 , space_cellallocchunk * sizeof(struct cell) ) != 0 )
            error( "Failed to allocate more cells." );
        bzero( s->cells_new , space_cellallocchunk * sizeof(struct cell) );
        for ( k = 0 ; k < space_cellallocchunk-1 ; k++ )
            s->cells_new[k].next = &s->cells_new[k+1];
        s->cells_new[ space_cellallocchunk-1 ].next = NULL;
        }

    /* Pick off the next cell. */
    c = s->cells_new;
    s->cells_new = c->next;
    s->tot_cells += 1;
    
1102
1103
1104
    /* Unlock the space. */
    lock_unlock_blind( &s->lock );
    
Pedro Gonnet's avatar
Pedro Gonnet committed
1105
    /* Init some things in the cell. */
1106
1107
    bzero( c , sizeof(struct cell) );
    c->nodeID = -1;
1108
1109
1110
    if ( lock_init( &c->lock ) != 0 ||
         lock_init( &c->glock ) != 0 )
        error( "Failed to initialize cell spinlocks." );
Pedro Gonnet's avatar
Pedro Gonnet committed
1111
1112
1113
1114
1115
1116
1117
1118
1119
        
    return c;

    }


/**
 * @brief Split the space into cells given the array of particles.
 *
1120
 * @param s The #space to initialize.
Pedro Gonnet's avatar
Pedro Gonnet committed
1121
1122
1123
1124
 * @param dim Spatial dimensions of the domain.
 * @param parts Pointer to an array of #part.
 * @param N The number of parts in the space.
 * @param periodic flag whether the domain is periodic or not.
1125
 * @param h_max The maximal interaction radius.
Pedro Gonnet's avatar
Pedro Gonnet committed
1126
1127
1128
1129
1130
1131
1132
1133
 *
 * Makes a grid of edge length > r_max and fills the particles
 * into the respective cells. Cells containing more than #space_maxppc
 * parts with a cutoff below half the cell width are then split
 * recursively.
 */


Pedro Gonnet's avatar
Pedro Gonnet committed
1134
void space_init ( struct space *s , double dim[3] , struct part *parts , int N , int periodic , double h_max ) {
Pedro Gonnet's avatar
Pedro Gonnet committed
1135
1136
1137
1138
1139

    /* Store eveything in the space. */
    s->dim[0] = dim[0]; s->dim[1] = dim[1]; s->dim[2] = dim[2];
    s->periodic = periodic;
    s->nr_parts = N;
1140
    s->size_parts = N;
1141
    s->parts = parts;
1142
    s->cell_min = h_max;
1143
    s->nr_queues = 1;
1144
    s->size_parts_foreign = 0;
1145
    
1146
    /* Allocate the xtra parts array. */
1147
1148
    if ( posix_memalign( (void *)&s->xparts , 32 , N * sizeof(struct xpart) ) != 0 )
        error( "Failed to allocate xparts." );
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
    bzero( s->xparts , N * sizeof(struct xpart) );
    
    /* Initialize the velocities and internal energies. */
    for ( int k = 0 ; k < N ; k++ ) {
        struct part *p = &parts[k];
        struct xpart *xp = &s->xparts[k];
        xp->v_hdt[0] = p->v[0];
        xp->v_hdt[1] = p->v[1];
        xp->v_hdt[2] = p->v[2];
        xp->u_hdt = p->u;
        }
1160
        
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
        
    /* For now, clone the parts to make gparts. */
    if ( posix_memalign( (void *)&s->gparts , part_align , N * sizeof(struct gpart) ) != 0 )
        error( "Failed to allocate gparts." );
    bzero( s->gparts , N * sizeof(struct gpart) );
    for ( int k = 0 ; k < N ; k++ ) {
        s->gparts[k].x[0] = s->parts[k].x[0];
        s->gparts[k].x[1] = s->parts[k].x[1];
        s->gparts[k].x[2] = s->parts[k].x[2];
        s->gparts[k].v[0] = s->parts[k].v[0];
        s->gparts[k].v[1] = s->parts[k].v[1];
        s->gparts[k].v[2] = s->parts[k].v[2];
        s->gparts[k].mass = s->parts[k].mass;
        s->gparts[k].dt = s->parts[k].dt;
        s->gparts[k].id = s->parts[k].id;
        s->gparts[k].part = &s->parts[k];
        s->parts[k].gpart = &s->gparts[k];
        }
    s->nr_gparts = s->nr_parts;
    s->size_gparts = s->size_parts;
    
        
1183
1184
1185
    /* Init the space lock. */
    if ( lock_init( &s->lock ) != 0 )
        error( "Failed to create space spin-lock." );
Pedro Gonnet's avatar
Pedro Gonnet committed
1186
    
1187
    /* Build the cells and the tasks. */
1188
    space_regrid( s , h_max );
Pedro Gonnet's avatar
Pedro Gonnet committed
1189
1190
1191
        
    }