space.c 31.9 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
 * Coypright (c) 2012 Pedro Gonnet (pedro.gonnet@durham.ac.uk)
4
 *
Pedro Gonnet's avatar
Pedro Gonnet committed
5
6
7
8
 * 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.
9
 *
Pedro Gonnet's avatar
Pedro Gonnet committed
10
11
12
13
 * 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.
14
 *
Pedro Gonnet's avatar
Pedro Gonnet committed
15
16
 * 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/>.
17
 *
Pedro Gonnet's avatar
Pedro Gonnet committed
18
19
20
21
22
23
24
25
26
 ******************************************************************************/

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

/* Some standard headers. */
#include <float.h>
#include <limits.h>
#include <math.h>
27
#include <string.h>
Pedro Gonnet's avatar
Pedro Gonnet committed
28

29
30
/* MPI headers. */
#ifdef WITH_MPI
31
#include <mpi.h>
32
33
#endif

34
35
36
/* This object's header. */
#include "space.h"

Pedro Gonnet's avatar
Pedro Gonnet committed
37
/* Local headers. */
38
#include "atomic.h"
39
#include "engine.h"
40
#include "error.h"
41
42
43
#include "kernel.h"
#include "lock.h"
#include "runner.h"
Pedro Gonnet's avatar
Pedro Gonnet committed
44
45
46

/* Split size. */
int space_splitsize = space_splitsize_default;
47
int space_subsize = space_subsize_default;
48
int space_maxsize = space_maxsize_default;
Pedro Gonnet's avatar
Pedro Gonnet committed
49
50
51

/* Map shift vector to sortlist. */
const int sortlistID[27] = {
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
    /* ( -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};

80
81
82
83
84
85
86
87
88
89
90
91
/**
 * @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.
 */

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
int space_getsid(struct space *s, struct cell **ci, struct cell **cj,
                 double *shift) {

  int k, sid = 0, periodic = s->periodic;
  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];
    if (periodic && dx[k] < -s->dim[k] / 2)
      shift[k] = s->dim[k];
    else if (periodic && dx[k] > s->dim[k] / 2)
      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;
}
127

128
/**
129
 * @brief Recursively dismantle a cell tree.
130
131
 *
 */
132
133
134
135
136
137
138
139
140
141
142
143
144
145

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;
      }
}

146
/**
147
 * @brief Re-build the cell grid.
148
 *
149
150
 * @param s The #space.
 * @param cell_max Maximum cell edge length.
151
 */
152
153
154
155
156
157
158
159
160
161
162
163
164

void space_regrid(struct space *s, double cell_max) {

  float h_max = s->cell_min / kernel_gamma / space_stretch, dmin;
  int i, j, k, cdim[3], nr_parts = s->nr_parts;
  struct cell *restrict c;
  // ticks tic;

  /* Run through the parts and get the current h_max. */
  // tic = getticks();
  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;
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
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
  } 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;
  }

/* If we are running in parallel, make sure everybody agrees on
   how large the largest cell should be. */
#ifdef WITH_MPI
  {
    float buff;
    if (MPI_Allreduce(&h_max, &buff, 1, MPI_FLOAT, MPI_MAX, MPI_COMM_WORLD) !=
        MPI_SUCCESS)
      error("Failed to aggreggate the rebuild flag accross nodes.");
    h_max = buff;
  }
#endif
  message("h_max is %.3e (cell_max=%.3e).", h_max, cell_max);

  /* Get the new putative cell dimensions. */
  for (k = 0; k < 3; k++)
    cdim[k] =
        floor(s->dim[k] / fmax(h_max * kernel_gamma * space_stretch, cell_max));

  /* Check if we have enough cells for periodicity. */
  if (s->periodic && (cdim[0] < 3 || cdim[1] < 3 || cdim[2] < 3))
    error(
        "Must have at least 3 cells in each spatial dimension when periodicity "
        "is switched on.");

/* 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

  /* Do we need to re-build the upper-level cells? */
  // tic = getticks();
  if (s->cells == NULL || cdim[0] < s->cdim[0] || cdim[1] < s->cdim[1] ||
      cdim[2] < s->cdim[2]) {

    /* Free the old cells, if they were allocated. */
    if (s->cells != NULL) {
      for (k = 0; k < s->nr_cells; k++) {
        space_rebuild_recycle(s, &s->cells[k]);
        if (s->cells[k].sort != NULL) free(s->cells[k].sort);
      }
      free(s->cells);
      s->maxdepth = 0;
    }

    /* Set the new cell dimensions only if smaller. */
    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];
    }
    dmin = fminf(s->h[0], fminf(s->h[1], s->h[2]));

    /* Allocate the highest level of cells. */
    s->tot_cells = s->nr_cells = cdim[0] * cdim[1] * cdim[2];
    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];
          c->dmin = dmin;
          c->depth = 0;
          c->count = 0;
          c->gcount = 0;
          c->super = c;
          lock_init(&c->lock);
Pedro Gonnet's avatar
Pedro Gonnet committed
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

    /* Be verbose about the change. */
    message("set cell dimensions to [ %i %i %i ].", cdim[0], cdim[1], cdim[2]);
    fflush(stdout);

  } /* re-build upper-level cells? */
  // message( "rebuilding upper-level cells took %.3f ms." , (double)(getticks()
  // - tic) / CPU_TPS * 1000 );

  /* 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;
      s->cells[k].nr_force = 0;
      s->cells[k].density = NULL;
      s->cells[k].force = NULL;
      s->cells[k].dx_max = 0.0f;
      s->cells[k].sorted = 0;
      s->cells[k].count = 0;
      s->cells[k].gcount = 0;
      s->cells[k].kick1 = NULL;
      s->cells[k].kick2 = NULL;
      s->cells[k].super = &s->cells[k];
281
    }
282
283
284
    s->maxdepth = 0;
  }
}
285
286
287
288
289
290
291
292

/**
 * @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.
 *
 */
293
294
295

void space_rebuild(struct space *s, double cell_max) {

296
  int j, k, cdim[3], nr_parts = s->nr_parts, nr_gparts = s->nr_gparts;
297
298
299
300
  struct cell *restrict c, *restrict cells;
  struct part *restrict finger, *restrict p, *parts = s->parts;
  struct xpart *xfinger, *xparts = s->xparts;
  struct gpart *gp, *gparts = s->gparts, *gfinger;
301
  int *ind;
302
303
304
305
306
307
308
309
310
311
312
313
  double ih[3], dim[3];
  // ticks tic;

  /* Be verbose about this. */
  // message( "re)building space..." ); fflush(stdout);

  /* Re-grid if necessary, or just re-set the cell data. */
  space_regrid(s, cell_max);
  cells = s->cells;

  /* Run through the particles and get their cell index. */
  // tic = getticks();
314
315
  const int ind_size = s->size_parts;
  if ((ind = (int *)malloc(sizeof(int) * ind_size)) == NULL)
316
317
318
319
320
321
322
323
324
325
    error("Failed to allocate temporary particle indices.");
  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];
  cdim[0] = s->cdim[0];
  cdim[1] = s->cdim[1];
  cdim[2] = s->cdim[2];
326
  for (k = 0; k < nr_parts; k++) {
327
328
329
330
331
332
    p = &parts[k];
    for (j = 0; j < 3; j++)
      if (p->x[j] < 0.0)
        p->x[j] += dim[j];
      else if (p->x[j] >= dim[j])
        p->x[j] -= dim[j];
333
    ind[k] =
334
        cell_getid(cdim, p->x[0] * ih[0], p->x[1] * ih[1], p->x[2] * ih[2]);
335
    cells[ind[k]].count++;
336
337
338
339
340
341
342
  }
// message( "getting particle indices took %.3f ms." , (double)(getticks() -
// tic) / CPU_TPS * 1000 );

#ifdef WITH_MPI
  /* Move non-local parts to the end of the list. */
  int nodeID = s->e->nodeID;
343
344
345
346
  for (k = 0; k < nr_parts; k++)
    if (cells[ind[k]].nodeID != nodeID) {
      cells[ind[k]].count -= 1;
      nr_parts -= 1;
347
      struct part tp = parts[k];
348
349
      parts[k] = parts[nr_parts];
      parts[nr_parts] = tp;
350
      struct xpart txp = xparts[k];
351
352
353
354
355
      xparts[k] = xparts[nr_parts];
      xparts[nr_parts] = txp;
      int t = ind[k];
      ind[k] = ind[nr_parts];
      ind[nr_parts] = t;
356
357
    }

358
359
360
  /* Exchange the strays, note that this potentially re-allocates
     the parts arrays. */
  s->nr_parts =
361
362
      nr_parts + engine_exchange_strays(s->e, nr_parts, &ind[nr_parts],
                                        s->nr_parts - nr_parts);
363
364
365
366
  parts = s->parts;
  xparts = s->xparts;

  /* Re-allocate the index array if needed.. */
367
368
369
  if (s->nr_parts > ind_size) {
    int *ind_new;
    if ((ind_new = (int *)malloc(sizeof(int) * s->nr_parts)) == NULL)
370
      error("Failed to allocate temporary particle indices.");
371
372
373
    memcpy(ind_new, ind, sizeof(int) * nr_parts);
    free(ind);
    ind = ind_new;
374
375
376
  }

  /* Assign each particle to its cell. */
377
  for (k = nr_parts; k < s->nr_parts; k++) {
378
    p = &parts[k];
379
    ind[k] =
380
        cell_getid(cdim, p->x[0] * ih[0], p->x[1] * ih[1], p->x[2] * ih[2]);
381
382
383
384
    cells[ind[k]].count += 1;
    /* if ( cells[ ind[k] ].nodeID != nodeID )
        error( "Received part that does not belong to me (nodeID=%i)." , cells[
       ind[k] ].nodeID ); */
385
  }
386
  nr_parts = s->nr_parts;
387
388
389
390
#endif

  /* Sort the parts according to their cells. */
  // tic = getticks();
391
  parts_sort(parts, xparts, ind, nr_parts, 0, s->nr_cells - 1);
392
393
394
395
  // message( "parts_sort took %.3f ms." , (double)(getticks() - tic) / CPU_TPS
  // * 1000 );

  /* Re-link the gparts. */
396
  for (k = 0; k < nr_parts; k++)
397
398
399
400
    if (parts[k].gpart != NULL) parts[k].gpart->part = &parts[k];

  /* Verify sort. */
  /* for ( k = 1 ; k < nr_parts ; k++ ) {
401
      if ( ind[k-1] > ind[k] ) {
402
403
          error( "Sort failed!" );
          }
404
      else if ( ind[k] != cell_getid( cdim , parts[k].x[0]*ih[0] ,
405
406
407
408
409
     parts[k].x[1]*ih[1] , parts[k].x[2]*ih[2] ) )
          error( "Incorrect indices!" );
      } */

  /* We no longer need the indices as of here. */
410
  free(ind);
411
412
413

  /* Run through the gravity particles and get their cell index. */
  // tic = getticks();
414
  if ((ind = (int *)malloc(sizeof(int) * s->size_gparts)) == NULL)
415
416
417
418
419
420
421
422
    error("Failed to allocate temporary particle indices.");
  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];
423
    ind[k] =
424
        cell_getid(cdim, gp->x[0] * ih[0], gp->x[1] * ih[1], gp->x[2] * ih[2]);
425
    cells[ind[k]].gcount++;
426
427
428
429
430
431
432
433
  }
  // 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();
434
  gparts_sort(gparts, ind, nr_gparts, 0, s->nr_cells - 1);
435
436
437
438
439
440
441
442
  // 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. */
443
  free(ind);
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464

  /* Hook the cells up to the parts. */
  // tic = getticks();
  finger = parts;
  xfinger = xparts;
  gfinger = gparts;
  for (k = 0; k < s->nr_cells; k++) {
    c = &cells[k];
    c->parts = finger;
    c->xparts = xfinger;
    c->gparts = gfinger;
    finger = &finger[c->count];
    xfinger = &xfinger[c->count];
    gfinger = &gfinger[c->gcount];
  }
  // message( "hooking up cells took %.3f ms." , (double)(getticks() - tic) /
  // CPU_TPS * 1000 );

  /* At this point, we have the upper-level cells, old or new. Now make
     sure that the parts in each cell are ok. */
  // tic = getticks();
465
466
  for (k = 0; k < s->nr_cells; k++) space_split(s, &cells[k]);

467
468
469
  // message( "space_split took %.3f ms." , (double)(getticks() - tic) / CPU_TPS
  // * 1000 );
}
470

471
/**
472
473
 * @brief Sort the particles and condensed particles according to the given
 *indices.
474
475
 *
 * @param parts The list of #part
476
 * @param xparts The list of reduced particles
477
478
479
480
481
 * @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.
 */
482

483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
void parts_sort(struct part *parts, struct xpart *xparts, 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 part temp_p;
  struct xpart temp_xp;

  /* 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;

518
519
  /* Main loop. */
  while (waiting > 0) {
520

521
522
    /* Grab an interval off the queue. */
    qid = (first++) % qstack_size;
523

524
525
526
527
528
529
    /* Get the stack entry. */
    i = qstack[qid].i;
    j = qstack[qid].j;
    min = qstack[qid].min;
    max = qstack[qid].max;
    qstack[qid].ready = 0;
530

531
532
    /* Loop over sub-intervals. */
    while (1) {
533

534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
      /* 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;
          temp_xp = xparts[ii];
          xparts[ii] = xparts[jj];
          xparts[jj] = temp_xp;
        }
      }
555

556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
      /* 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 = (last++) % qstack_size;
          qstack[qid].i = i;
          qstack[qid].j = jj;
          qstack[qid].min = min;
          qstack[qid].max = pivot;
          qstack[qid].ready = 1;
          if (waiting++ >= qstack_size) error("Qstack overflow.");
        }
583

584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
        /* 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 = (last++) % qstack_size;
          qstack[qid].i = jj + 1;
          qstack[qid].j = j;
          qstack[qid].min = pivot + 1;
          qstack[qid].max = max;
          qstack[qid].ready = 1;
          if ((waiting++) >= qstack_size) error("Qstack overflow.");
        }
603

604
605
606
607
608
609
610
        /* Recurse on the left? */
        if (jj > i && pivot > min) {
          j = jj;
          max = pivot;
        } else
          break;
      }
611

612
613
614
615
616
    } /* loop over sub-intervals. */

    waiting--;

  } /* main loop. */
617
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

  /* 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);
}

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;

661
662
  /* Main loop. */
  while (waiting > 0) {
663

664
665
    /* Grab an interval off the queue. */
    qid = (first++) % qstack_size;
666

667
668
669
670
671
672
673
674
675
    /* Get the stack entry. */
    i = qstack[qid].i;
    j = qstack[qid].j;
    min = qstack[qid].min;
    max = qstack[qid].max;
    qstack[qid].ready = 0;

    /* Loop over sub-intervals. */
    while (1) {
676

677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
      /* 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;
        }
      }
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
      /* 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 = (last++) % qstack_size;
          qstack[qid].i = i;
          qstack[qid].j = jj;
          qstack[qid].min = min;
          qstack[qid].max = pivot;
          qstack[qid].ready = 1;
          if ((waiting++) >= qstack_size) error("Qstack overflow.");
        }
723

724
725
726
727
728
729
        /* Recurse on the right? */
        if (jj + 1 < j && pivot + 1 < max) {
          i = jj + 1;
          min = pivot + 1;
        } else
          break;
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
      } else {

        /* Recurse on the right? */
        if (jj + 1 < j && pivot + 1 < max) {
          qid = (last++) % qstack_size;
          qstack[qid].i = jj + 1;
          qstack[qid].j = j;
          qstack[qid].min = pivot + 1;
          qstack[qid].max = max;
          qstack[qid].ready = 1;
          if ((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. */

    waiting--;

  } /* main loop. */
757
758
759
760
761
762
763
764
765
766

  /* 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);
}
767

Pedro Gonnet's avatar
Pedro Gonnet committed
768
/**
769
 * @brief Mapping function to free the sorted indices buffers.
Pedro Gonnet's avatar
Pedro Gonnet committed
770
771
 */

772
void space_map_clearsort(struct cell *c, void *data) {
Pedro Gonnet's avatar
Pedro Gonnet committed
773

774
775
776
777
778
  if (c->sort != NULL) {
    free(c->sort);
    c->sort = NULL;
  }
}
Pedro Gonnet's avatar
Pedro Gonnet committed
779

780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805

/**
 * @brief Map a function to all particles in a cell recursively.
 *
 * @param s The #space we are working in.
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
 */

static void rec_map_parts(struct cell * c,
		   void (*fun)(struct part *p, struct cell *c, void *data),
		   void *data) {

  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_parts(c->progeny[k], fun, data);
}


Pedro Gonnet's avatar
Pedro Gonnet committed
806
/**
807
 * @brief Map a function to all particles in a space.
Pedro Gonnet's avatar
Pedro Gonnet committed
808
809
 *
 * @param s The #space we are working in.
810
811
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
Pedro Gonnet's avatar
Pedro Gonnet committed
812
813
 */

814
815
816
void space_map_parts(struct space *s,
                     void (*fun)(struct part *p, struct cell *c, void *data),
                     void *data) {
Pedro Gonnet's avatar
Pedro Gonnet committed
817

818
819
  int cid = 0;

820
821
822
  /* Call the recursive function on all higher-level cells. */
  for (cid = 0; cid < s->nr_cells; cid++) rec_map_parts(&s->cells[cid], fun, data);
}
823
824


825
826
827
828
829
830
831
832
/**
 * @brief Map a function to all particles in a cell recursively.
 *
 * @param s The #space we are working in.
 * @param full Map to all cells, including cells with sub-cells.
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
 */
833

834
835
836
static void rec_map_cells_post(struct cell * c, int full,
                          void (*fun)(struct cell *c, void *data), 
			  void *data) {
837

838
839
840
841
842
843
844
845
846
  int k;

  /* Recurse. */
  if (c->split)
    for (k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) rec_map_cells_post(c->progeny[k], full, fun, data);
  
  /* No progeny? */
  if (full || !c->split) fun(c, data);
847
}
Pedro Gonnet's avatar
Pedro Gonnet committed
848

849

Pedro Gonnet's avatar
Pedro Gonnet committed
850
/**
851
 * @brief Map a function to all particles in a space.
Pedro Gonnet's avatar
Pedro Gonnet committed
852
853
 *
 * @param s The #space we are working in.
854
 * @param full Map to all cells, including cells with sub-cells.
855
856
 * @param fun Function pointer to apply on the cells.
 * @param data Data passed to the function fun.
Pedro Gonnet's avatar
Pedro Gonnet committed
857
 */
858

859
void space_map_cells_post(struct space *s, int full,
860
861
                          void (*fun)(struct cell *c, void *data), 
			  void *data) {
862

863
  int cid = 0;
864

865
866
867
  /* Call the recursive function on all higher-level cells. */
  for (cid = 0; cid < s->nr_cells; cid++) rec_map_cells_post(&s->cells[cid], full, fun, data);
}
868

Pedro Gonnet's avatar
Pedro Gonnet committed
869

870

871
872
873
static void rec_map_cells_pre(struct cell * c, int full,
                         void (*fun)(struct cell *c, void *data), 
			 void *data) {
874

875
876
877
878
879
880
881
882
883
  int k;
  
  /* No progeny? */
  if (full || !c->split) fun(c, data);
  
  /* Recurse. */
  if (c->split)
    for (k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) rec_map_cells_pre(c->progeny[k], full, fun, data);
884
}
Pedro Gonnet's avatar
Pedro Gonnet committed
885

886
887


888
889
890
void space_map_cells_pre(struct space *s, int full,
                         void (*fun)(struct cell *c, void *data), 
			 void *data) {
891

892
  int cid = 0;
893
894

  /* Call the recursive function on all higher-level cells. */
895
  for (cid = 0; cid < s->nr_cells; cid++) rec_map_cells_pre(&s->cells[cid], full, fun, data);
896
}
Pedro Gonnet's avatar
Pedro Gonnet committed
897

898

Pedro Gonnet's avatar
Pedro Gonnet committed
899
900
901
902
903
904
/**
 * @brief Split cells that contain too many particles.
 *
 * @param s The #space we are working in.
 * @param c The #cell under consideration.
 */
905

906
907
908
void space_split(struct space *s, struct cell *c) {

  int k, count = c->count, gcount = c->gcount, maxdepth = 0;
909
  float h, h_max = 0.0f, dt, dt_min = c->parts[0].dt, dt_max = dt_min;
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
  struct cell *temp;
  struct part *p, *parts = c->parts;
  struct xpart *xp, *xparts = c->xparts;

  /* Check the depth. */
  if (c->depth > s->maxdepth) s->maxdepth = c->depth;

  /* Split or let it be? */
  if (count > space_splitsize || gcount > space_splitsize) {

    /* 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;
      temp->gcount = 0;
      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;
      temp->nodeID = c->nodeID;
      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++)
      if (c->progeny[k]->count == 0 && c->progeny[k]->gcount == 0) {
        space_recycle(s, c->progeny[k]);
        c->progeny[k] = NULL;
      } else {
        space_split(s, c->progeny[k]);
        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;
      }

    /* Set the values for this cell. */
    c->h_max = h_max;
    c->dt_min = dt_min;
    c->dt_max = dt_max;
    c->maxdepth = maxdepth;

  }

  /* Otherwise, collect the data for this cell. */
  else {

    /* Clear the progeny. */
    bzero(c->progeny, sizeof(struct cell *) * 8);
    c->split = 0;
    c->maxdepth = c->depth;

    /* Get dt_min/dt_max. */

    for (k = 0; k < count; k++) {
      p = &parts[k];
      xp = &xparts[k];
      xp->x_old[0] = p->x[0];
      xp->x_old[1] = p->x[1];
      xp->x_old[2] = p->x[2];
      dt = p->dt;
      h = p->h;
      if (h > h_max) h_max = h;
      if (dt < dt_min) dt_min = dt;
      if (dt > dt_max) dt_max = dt;
993
    }
994
995
996
997
    c->h_max = h_max;
    c->dt_min = dt_min;
    c->dt_max = dt_max;
  }
998

999
1000
  /* Set ownership accorind to the start of the parts array. */
  c->owner = ((c->parts - s->parts) % s->nr_parts) * s->nr_queues / s->nr_parts;
For faster browsing, not all history is shown. View entire blame