/
pm_periodic.c
1284 lines (1060 loc) · 38.6 KB
/
pm_periodic.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
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
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
317
318
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
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
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
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
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
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
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
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
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
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
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
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
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
993
994
995
996
997
998
999
1000
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <float.h>
#include <mpi.h>
/*! \file pm_periodic.c
* \brief routines for periodic PM-force computation
*/
#ifdef PMGRID
#ifdef PERIODIC
#ifdef NOTYPEPREFIX_FFTW
#include <rfftw_mpi.h>
#else
#ifdef DOUBLEPRECISION_FFTW
#include <drfftw_mpi.h> /* double precision FFTW */
#else
#include <srfftw_mpi.h>
#endif
#endif
#include "allvars.h"
#include "proto.h"
#define PMGRID2 (2*(PMGRID/2 + 1))
static rfftwnd_mpi_plan fft_forward_plan, fft_inverse_plan;
static int slab_to_task[PMGRID];
static int *slabs_per_task;
static int *first_slab_of_task;
static int *meshmin_list, *meshmax_list;
static int slabstart_x, nslab_x, slabstart_y, nslab_y, smallest_slab;
static int fftsize, maxfftsize;
static fftw_real *rhogrid, *forcegrid, *workspace;
static fftw_complex *fft_of_rhogrid;
static FLOAT to_slab_fac;
/*! This routines generates the FFTW-plans to carry out the parallel FFTs
* later on. Some auxiliary variables are also initialized.
*/
void pm_init_periodic(void)
{
int i;
int slab_to_task_local[PMGRID];
All.Asmth[0] = ASMTH * All.BoxSize / PMGRID;
All.Rcut[0] = RCUT * All.Asmth[0];
/* Set up the FFTW plan files. */
fft_forward_plan = rfftw3d_mpi_create_plan(MPI_COMM_WORLD, PMGRID, PMGRID, PMGRID,
FFTW_REAL_TO_COMPLEX, FFTW_ESTIMATE | FFTW_IN_PLACE);
fft_inverse_plan = rfftw3d_mpi_create_plan(MPI_COMM_WORLD, PMGRID, PMGRID, PMGRID,
FFTW_COMPLEX_TO_REAL, FFTW_ESTIMATE | FFTW_IN_PLACE);
/* Workspace out the ranges on each processor. */
// KC 10/3/14
// Some notes on the MPI routines: data on the local processor represents a chunk sized nslab_x * (PMGRID**2)
// Indexes will represent: slabstart_x -> slabstart_x + nslab_x - 1 on the actual PMGRID**3
rfftwnd_mpi_local_sizes(fft_forward_plan, &nslab_x, &slabstart_x, &nslab_y, &slabstart_y, &fftsize);
// KC 10/3/14
// Here we initialize a mapping of slabs to tasks
for(i = 0; i < PMGRID; i++)
slab_to_task_local[i] = 0;
// KC 10/3/14
// Here we fill into this array that we (our task) is responsible for the range
// indicated (as provided by the FFTW implementation)
for(i = 0; i < nslab_x; i++)
slab_to_task_local[slabstart_x + i] = ThisTask;
// KC 10/3/14
// This MPI routine takes all the slab_to_task_local from each process and combines them into a global
// slab_to_task that contains which tasks are responsible for which slabs in the Fourier transform
// Sum works because things were initialized to zero, and if task 0 is responsible, it will stay zero
MPI_Allreduce(slab_to_task_local, slab_to_task, PMGRID, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
// KC 10/3/14
// Looks like this finds the smallest number of slabs allocated
MPI_Allreduce(&nslab_x, &smallest_slab, 1, MPI_INT, MPI_MIN, MPI_COMM_WORLD);
// KC 10/3/14
// This populates a list of how many slabs each process is responsible for
slabs_per_task = malloc(NTask * sizeof(int));
MPI_Allgather(&nslab_x, 1, MPI_INT, slabs_per_task, 1, MPI_INT, MPI_COMM_WORLD);
// KC 10/3/14
// Output some fun stuff on the root process
if(ThisTask == 0)
{
for(i = 0; i < NTask; i++)
printf("Task=%d FFT-Slabs=%d\n", i, slabs_per_task[i]);
}
// KC 10/3/14
// Gets the first slab that each task is responsible for
// With this data, we can figure out all the mappings correctly for our particles
first_slab_of_task = malloc(NTask * sizeof(int));
MPI_Allgather(&slabstart_x, 1, MPI_INT, first_slab_of_task, 1, MPI_INT, MPI_COMM_WORLD);
meshmin_list = malloc(3 * NTask * sizeof(int));
meshmax_list = malloc(3 * NTask * sizeof(int));
to_slab_fac = PMGRID / All.BoxSize;
// KC 10/3/14
// This figures out the maximum FFT that is being performed by the cluster
MPI_Allreduce(&fftsize, &maxfftsize, 1, MPI_INT, MPI_MAX, MPI_COMM_WORLD);
}
/*! This function allocates the memory neeed to compute the long-range PM
* force. Three fields are used, one to hold the density (and its FFT, and
* then the real-space potential), one to hold the force field obtained by
* finite differencing, and finally a workspace field, which is used both as
* workspace for the parallel FFT, and as buffer for the communication
* algorithm used in the force computation.
*/
void pm_init_periodic_allocate(int dimprod)
{
static int first_alloc = 1;
int dimprodmax;
double bytes_tot = 0;
size_t bytes;
MPI_Allreduce(&dimprod, &dimprodmax, 1, MPI_INT, MPI_MAX, MPI_COMM_WORLD);
/* allocate the memory to hold the FFT fields */
if(!(rhogrid = (fftw_real *) malloc(bytes = fftsize * sizeof(fftw_real))))
{
printf("failed to allocate memory for `FFT-rhogrid' (%g MB).\n", bytes / (1024.0 * 1024.0));
endrun(1);
}
bytes_tot += bytes;
if(!(forcegrid = (fftw_real *) malloc(bytes = imax(fftsize, dimprodmax) * sizeof(fftw_real))))
{
printf("failed to allocate memory for `FFT-forcegrid' (%g MB).\n", bytes / (1024.0 * 1024.0));
endrun(1);
}
bytes_tot += bytes;
if(!(workspace = (fftw_real *) malloc(bytes = imax(maxfftsize, dimprodmax) * sizeof(fftw_real))))
{
printf("failed to allocate memory for `FFT-workspace' (%g MB).\n", bytes / (1024.0 * 1024.0));
endrun(1);
}
bytes_tot += bytes;
if(first_alloc == 1)
{
first_alloc = 0;
if(ThisTask == 0)
printf("\nAllocated %g MByte for FFT data.\n\n", bytes_tot / (1024.0 * 1024.0));
}
fft_of_rhogrid = (fftw_complex *) & rhogrid[0];
}
/*! This routine frees the space allocated for the parallel FFT algorithm.
*/
void pm_init_periodic_free(void)
{
/* allocate the memory to hold the FFT fields */
free(workspace);
free(forcegrid);
free(rhogrid);
}
/*! Calculates the long-range periodic force given the particle positions
* using the PM method. The force is Gaussian filtered with Asmth, given in
* mesh-cell units. We carry out a CIC charge assignment, and compute the
* potenial by Fourier transform methods. The potential is finite differenced
* using a 4-point finite differencing formula, and the forces are
* interpolated tri-linearly to the particle positions. The CIC kernel is
* deconvolved. Note that the particle distribution is not in the slab
* decomposition that is used for the FFT. Instead, overlapping patches
* between local domains and FFT slabs are communicated as needed.
*/
void pmforce_periodic(void)
{
double k2, kx, ky, kz, smth;
double dx, dy, dz;
double fx, fy, fz, ff;
double asmth2, fac, acc_dim;
int i, j, slab, level, sendTask, recvTask;
int x, y, z, xl, yl, zl, xr, yr, zr, xll, yll, zll, xrr, yrr, zrr, ip, dim;
int slab_x, slab_y, slab_z;
int slab_xx, slab_yy, slab_zz;
int meshmin[3], meshmax[3], sendmin, sendmax, recvmin, recvmax;
int rep, ncont, cont_sendmin[2], cont_sendmax[2], cont_recvmin[2], cont_recvmax[2];
int dimx, dimy, dimz, recv_dimx, recv_dimy, recv_dimz;
MPI_Status status;
// KC 10/5/14
// Additional N_GRAVS variables
int nA, nB;
int k;
int offsets[N_GRAVS+1];
if(ThisTask == 0)
{
printf("Starting periodic PM calculation.\n");
fflush(stdout);
}
force_treefree();
asmth2 = (2 * M_PI) * All.Asmth[0] / All.BoxSize;
asmth2 *= asmth2;
fac = All.G / (M_PI * All.BoxSize); /* to get potential */
fac *= 1 / (2 * All.BoxSize / PMGRID); /* for finite differencing */
// KC 10/20/14
// Determine the ranges outside the main loop!? (yes)
/* first, establish the extension of the local patch in the PMGRID */
for(j = 0; j < 3; j++)
{
meshmin[j] = PMGRID;
meshmax[j] = 0;
}
// KC 10/20/14
// Compute the gravitational offsets here
offsets[0] = 0;
for(k = 1; k < N_GRAVS; ++k)
offsets[k] = offsets[k-1] + NgravLocal[k-1];
offsets[N_GRAVS] = NumPart;
for(i = 0; i < NumPart; ++i) {
for(j = 0; j < 3; j++)
{
// Clear this out before we start accumulating
// (Prev Gadget2 just overwrote this because it performed one assignment.
// we need to accumulate over all types)
P[i].GravPM[j] = 0;
slab = to_slab_fac * P[i].Pos[j];
if(slab >= PMGRID)
slab = PMGRID - 1;
if(slab < meshmin[j])
meshmin[j] = slab;
if(slab > meshmax[j])
meshmax[j] = slab;
}
}
MPI_Allgather(meshmin, 3, MPI_INT, meshmin_list, 3, MPI_INT, MPI_COMM_WORLD);
MPI_Allgather(meshmax, 3, MPI_INT, meshmax_list, 3, MPI_INT, MPI_COMM_WORLD);
dimx = meshmax[0] - meshmin[0] + 2;
dimy = meshmax[1] - meshmin[1] + 2;
dimz = meshmax[2] - meshmin[2] + 2;
pm_init_periodic_allocate((dimx + 4) * (dimy + 4) * (dimz + 4));
// Sources
for(nA = 0; nA < N_GRAVS; ++nA) {
// Receivers
for(nB = 0; nB < N_GRAVS; ++nB) {
// Clear the workspace for this round of computation
for(i = 0; i < dimx * dimy * dimz; i++)
workspace[i] = 0;
for(i = offsets[nA]; i < offsets[nA+1]; ++i) {
slab_x = to_slab_fac * P[i].Pos[0];
if(slab_x >= PMGRID)
slab_x = PMGRID - 1;
dx = to_slab_fac * P[i].Pos[0] - slab_x;
slab_x -= meshmin[0];
slab_xx = slab_x + 1;
slab_y = to_slab_fac * P[i].Pos[1];
if(slab_y >= PMGRID)
slab_y = PMGRID - 1;
dy = to_slab_fac * P[i].Pos[1] - slab_y;
slab_y -= meshmin[1];
slab_yy = slab_y + 1;
slab_z = to_slab_fac * P[i].Pos[2];
if(slab_z >= PMGRID)
slab_z = PMGRID - 1;
dz = to_slab_fac * P[i].Pos[2] - slab_z;
slab_z -= meshmin[2];
slab_zz = slab_z + 1;
// KC 10/3/14 This is some clouds-in-cells stuff...
workspace[(slab_x * dimy + slab_y) * dimz + slab_z] += P[i].Mass * (1.0 - dx) * (1.0 - dy) * (1.0 - dz);
workspace[(slab_x * dimy + slab_yy) * dimz + slab_z] += P[i].Mass * (1.0 - dx) * dy * (1.0 - dz);
workspace[(slab_x * dimy + slab_y) * dimz + slab_zz] += P[i].Mass * (1.0 - dx) * (1.0 - dy) * dz;
workspace[(slab_x * dimy + slab_yy) * dimz + slab_zz] += P[i].Mass * (1.0 - dx) * dy * dz;
workspace[(slab_xx * dimy + slab_y) * dimz + slab_z] += P[i].Mass * (dx) * (1.0 - dy) * (1.0 - dz);
workspace[(slab_xx * dimy + slab_yy) * dimz + slab_z] += P[i].Mass * (dx) * dy * (1.0 - dz);
workspace[(slab_xx * dimy + slab_y) * dimz + slab_zz] += P[i].Mass * (dx) * (1.0 - dy) * dz;
workspace[(slab_xx * dimy + slab_yy) * dimz + slab_zz] += P[i].Mass * (dx) * dy * dz;
}
for(i = 0; i < fftsize; i++) /* clear local density field */
rhogrid[i] = 0;
for(level = 0; level < (1 << PTask); level++) /* note: for level=0, target is the same task */
{
sendTask = ThisTask;
recvTask = ThisTask ^ level;
if(recvTask < NTask)
{
/* check how much we have to send */
sendmin = 2 * PMGRID;
sendmax = -1;
for(slab_x = meshmin[0]; slab_x < meshmax[0] + 2; slab_x++)
if(slab_to_task[slab_x % PMGRID] == recvTask)
{
if(slab_x < sendmin)
sendmin = slab_x;
if(slab_x > sendmax)
sendmax = slab_x;
}
if(sendmax == -1)
sendmin = 0;
/* check how much we have to receive */
recvmin = 2 * PMGRID;
recvmax = -1;
for(slab_x = meshmin_list[3 * recvTask]; slab_x < meshmax_list[3 * recvTask] + 2; slab_x++)
if(slab_to_task[slab_x % PMGRID] == sendTask)
{
if(slab_x < recvmin)
recvmin = slab_x;
if(slab_x > recvmax)
recvmax = slab_x;
}
if(recvmax == -1)
recvmin = 0;
if((recvmax - recvmin) >= 0 || (sendmax - sendmin) >= 0) /* ok, we have a contribution to the slab */
{
recv_dimx = meshmax_list[3 * recvTask + 0] - meshmin_list[3 * recvTask + 0] + 2;
recv_dimy = meshmax_list[3 * recvTask + 1] - meshmin_list[3 * recvTask + 1] + 2;
recv_dimz = meshmax_list[3 * recvTask + 2] - meshmin_list[3 * recvTask + 2] + 2;
if(level > 0)
{
// KC 10/4/14
// Here, we alter the TAG_PERIODIC_A
// It becomes:
// TAG_PERIODIC_A | nA << 6 | nB << 9
// In this way, we keep all the processors syncd on the interaction being computed
MPI_Sendrecv(workspace + (sendmin - meshmin[0]) * dimy * dimz,
(sendmax - sendmin + 1) * dimy * dimz * sizeof(fftw_real), MPI_BYTE, recvTask,
TAG_PERIODIC_A | (nA << 6) | (nB << 9), forcegrid,
(recvmax - recvmin + 1) * recv_dimy * recv_dimz * sizeof(fftw_real), MPI_BYTE,
recvTask, TAG_PERIODIC_A | (nA << 6) | (nB << 9), MPI_COMM_WORLD, &status);
}
else
{
memcpy(forcegrid, workspace + (sendmin - meshmin[0]) * dimy * dimz,
(sendmax - sendmin + 1) * dimy * dimz * sizeof(fftw_real));
}
for(slab_x = recvmin; slab_x <= recvmax; slab_x++)
{
slab_xx = (slab_x % PMGRID) - first_slab_of_task[ThisTask];
if(slab_xx >= 0 && slab_xx < slabs_per_task[ThisTask])
{
for(slab_y = meshmin_list[3 * recvTask + 1];
slab_y <= meshmax_list[3 * recvTask + 1] + 1; slab_y++)
{
slab_yy = slab_y;
if(slab_yy >= PMGRID)
slab_yy -= PMGRID;
for(slab_z = meshmin_list[3 * recvTask + 2];
slab_z <= meshmax_list[3 * recvTask + 2] + 1; slab_z++)
{
slab_zz = slab_z;
if(slab_zz >= PMGRID)
slab_zz -= PMGRID;
rhogrid[PMGRID * PMGRID2 * slab_xx + PMGRID2 * slab_yy + slab_zz] +=
forcegrid[((slab_x - recvmin) * recv_dimy +
(slab_y - meshmin_list[3 * recvTask + 1])) * recv_dimz +
(slab_z - meshmin_list[3 * recvTask + 2])];
}
}
}
}
}
}
}
// KC 11/30/14
// Workspace contains the real-space density grid
/* Do the FFT of the density field */
rfftwnd_mpi(fft_forward_plan, 1, rhogrid, workspace, FFTW_TRANSPOSED_ORDER);
/* multiply with Green's function for the potential */
for(y = slabstart_y; y < slabstart_y + nslab_y; y++)
for(x = 0; x < PMGRID; x++)
for(z = 0; z < PMGRID / 2 + 1; z++)
{
if(x > PMGRID / 2)
kx = x - PMGRID;
else
kx = x;
if(y > PMGRID / 2)
ky = y - PMGRID;
else
ky = y;
if(z > PMGRID / 2)
kz = z - PMGRID;
else
kz = z;
k2 = kx * kx + ky * ky + kz * kz;
/* do deconvolution */
if(k2 > 0) {
fx = fy = fz = 1;
if(kx != 0)
{
fx = (M_PI * kx) / PMGRID;
fx = sin(fx) / fx;
}
if(ky != 0)
{
fy = (M_PI * ky) / PMGRID;
fy = sin(fy) / fy;
}
if(kz != 0)
{
fz = (M_PI * kz) / PMGRID;
fz = sin(fz) / fz;
}
ff = 1 / (fx * fy * fz);
// KC 10/5/14
// The CIC charge assignment (sampling along a mesh of a CIC continuous charge distribution
// implied by the actual point charges). The implied continuous charge distribution is
// given by the space density convolved with the CIC kernel W(x-x') (c.f. Hockney 5-165)
//
// The Fourier space CIC kernel is (1/ff)**2
//
// We divide by the CIC kernel because we are deconvolving
//
// We divide by the CIC kernel twice: once comes from the charge assignment procedure above
// a second time comes from the force interpolation procedure
//
// We also apply the short range truncation, here in k-space
//
// This is the only place asmth has entered anywhere!
smth = (*GreensFxns[nA][nB])(All.MassTable[nA], All.MassTable[nB], k2, sqrt(k2), 1);
smth *= -exp(-k2*asmth2) * ff * ff * ff * ff;
// NOTE: Transposed order of indicies due to FFTW in k-space
ip = PMGRID * (PMGRID / 2 + 1) * (y - slabstart_y) + (PMGRID / 2 + 1) * x + z;
// KC 27.9.15
//
// CONSTRAINT:
// We send the MassTable masses as passive and active masses so that the
// scale of the force law can be appropriately set. If the source mass appears
// then it must be the same for all within the species. If the target mass appears,
// then ALL species must have uniform masses across their species.
//
// Note that the greens function can only ever depend on |k| since the force is
// isotropic!
//
/* end deconvolution */
// KC 12/4/14
// Note that smth contains the exponential factor and the deconvolution stuff
fft_of_rhogrid[ip].re *= smth;
fft_of_rhogrid[ip].im *= smth;
}
}
// Don't explicily kill 0 power?
if(slabstart_y == 0)
fft_of_rhogrid[0].re = fft_of_rhogrid[0].im = 0.0;
// KC 10/5/14
// This will produce the potential in real space
/* Do the FFT to get the potential */
rfftwnd_mpi(fft_inverse_plan, 1, rhogrid, workspace, FFTW_TRANSPOSED_ORDER);
/* Now rhogrid holds the potential */
/* construct the potential for the local patch */
dimx = meshmax[0] - meshmin[0] + 6;
dimy = meshmax[1] - meshmin[1] + 6;
dimz = meshmax[2] - meshmin[2] + 6;
for(level = 0; level < (1 << PTask); level++) /* note: for level=0, target is the same task */
{
sendTask = ThisTask;
recvTask = ThisTask ^ level;
if(recvTask < NTask)
{
/* check how much we have to send */
sendmin = 2 * PMGRID;
sendmax = -PMGRID;
for(slab_x = meshmin_list[3 * recvTask] - 2; slab_x < meshmax_list[3 * recvTask] + 4; slab_x++)
if(slab_to_task[(slab_x + PMGRID) % PMGRID] == sendTask)
{
if(slab_x < sendmin)
sendmin = slab_x;
if(slab_x > sendmax)
sendmax = slab_x;
}
if(sendmax == -PMGRID)
sendmin = sendmax + 1;
/* check how much we have to receive */
recvmin = 2 * PMGRID;
recvmax = -PMGRID;
for(slab_x = meshmin[0] - 2; slab_x < meshmax[0] + 4; slab_x++)
if(slab_to_task[(slab_x + PMGRID) % PMGRID] == recvTask)
{
if(slab_x < recvmin)
recvmin = slab_x;
if(slab_x > recvmax)
recvmax = slab_x;
}
if(recvmax == -PMGRID)
recvmin = recvmax + 1;
if((recvmax - recvmin) >= 0 || (sendmax - sendmin) >= 0) /* ok, we have a contribution to the slab */
{
recv_dimx = meshmax_list[3 * recvTask + 0] - meshmin_list[3 * recvTask + 0] + 6;
recv_dimy = meshmax_list[3 * recvTask + 1] - meshmin_list[3 * recvTask + 1] + 6;
recv_dimz = meshmax_list[3 * recvTask + 2] - meshmin_list[3 * recvTask + 2] + 6;
ncont = 1;
cont_sendmin[0] = sendmin;
cont_sendmax[0] = sendmax;
cont_sendmin[1] = sendmax + 1;
cont_sendmax[1] = sendmax;
cont_recvmin[0] = recvmin;
cont_recvmax[0] = recvmax;
cont_recvmin[1] = recvmax + 1;
cont_recvmax[1] = recvmax;
for(slab_x = sendmin; slab_x <= sendmax; slab_x++)
{
if(slab_to_task[(slab_x + PMGRID) % PMGRID] != ThisTask)
{
/* non-contiguous */
cont_sendmax[0] = slab_x - 1;
while(slab_to_task[(slab_x + PMGRID) % PMGRID] != ThisTask)
slab_x++;
cont_sendmin[1] = slab_x;
ncont++;
}
}
for(slab_x = recvmin; slab_x <= recvmax; slab_x++)
{
if(slab_to_task[(slab_x + PMGRID) % PMGRID] != recvTask)
{
/* non-contiguous */
cont_recvmax[0] = slab_x - 1;
while(slab_to_task[(slab_x + PMGRID) % PMGRID] != recvTask)
slab_x++;
cont_recvmin[1] = slab_x;
if(ncont == 1)
ncont++;
}
}
for(rep = 0; rep < ncont; rep++)
{
sendmin = cont_sendmin[rep];
sendmax = cont_sendmax[rep];
recvmin = cont_recvmin[rep];
recvmax = cont_recvmax[rep];
/* prepare what we want to send */
if(sendmax - sendmin >= 0)
{
for(slab_x = sendmin; slab_x <= sendmax; slab_x++)
{
slab_xx = ((slab_x + PMGRID) % PMGRID) - first_slab_of_task[ThisTask];
for(slab_y = meshmin_list[3 * recvTask + 1] - 2;
slab_y < meshmax_list[3 * recvTask + 1] + 4; slab_y++)
{
slab_yy = (slab_y + PMGRID) % PMGRID;
for(slab_z = meshmin_list[3 * recvTask + 2] - 2;
slab_z < meshmax_list[3 * recvTask + 2] + 4; slab_z++)
{
slab_zz = (slab_z + PMGRID) % PMGRID;
forcegrid[((slab_x - sendmin) * recv_dimy +
(slab_y - (meshmin_list[3 * recvTask + 1] - 2))) * recv_dimz +
slab_z - (meshmin_list[3 * recvTask + 2] - 2)] =
rhogrid[PMGRID * PMGRID2 * slab_xx + PMGRID2 * slab_yy + slab_zz];
}
}
}
}
if(level > 0)
{
// KC 10/4/14
// Here, we alter the TAG_PERIODIC_B
// It becomes:
// TAG_PERIODIC_B | nA << 6 | nB << 9
// In this way, we keep all the processors syncd on the interaction being computed
MPI_Sendrecv(forcegrid,
(sendmax - sendmin + 1) * recv_dimy * recv_dimz * sizeof(fftw_real),
MPI_BYTE, recvTask, TAG_PERIODIC_B | (nA << 6) | (nB << 9),
workspace + (recvmin - (meshmin[0] - 2)) * dimy * dimz,
(recvmax - recvmin + 1) * dimy * dimz * sizeof(fftw_real), MPI_BYTE,
recvTask, TAG_PERIODIC_B | (nA << 6) | (nB << 9), MPI_COMM_WORLD, &status);
}
else
{
memcpy(workspace + (recvmin - (meshmin[0] - 2)) * dimy * dimz,
forcegrid, (recvmax - recvmin + 1) * dimy * dimz * sizeof(fftw_real));
}
}
}
}
}
dimx = meshmax[0] - meshmin[0] + 2;
dimy = meshmax[1] - meshmin[1] + 2;
dimz = meshmax[2] - meshmin[2] + 2;
recv_dimx = meshmax[0] - meshmin[0] + 6;
recv_dimy = meshmax[1] - meshmin[1] + 6;
recv_dimz = meshmax[2] - meshmin[2] + 6;
for(dim = 0; dim < 3; dim++) /* Calculate each component of the force. */
{
/* get the force component by finite differencing the potential */
/* note: "workspace" now contains the potential for the local patch, plus a suffiently large buffer region */
for(x = 0; x < meshmax[0] - meshmin[0] + 2; x++) {
for(y = 0; y < meshmax[1] - meshmin[1] + 2; y++) {
for(z = 0; z < meshmax[2] - meshmin[2] + 2; z++)
{
xrr = xll = xr = xl = x;
yrr = yll = yr = yl = y;
zrr = zll = zr = zl = z;
switch (dim)
{
case 0:
xr = x + 1;
xrr = x + 2;
xl = x - 1;
xll = x - 2;
break;
case 1:
yr = y + 1;
yl = y - 1;
yrr = y + 2;
yll = y - 2;
break;
case 2:
zr = z + 1;
zl = z - 1;
zrr = z + 2;
zll = z - 2;
break;
}
forcegrid[(x * dimy + y) * dimz + z]
=
fac * ((4.0 / 3) *
(workspace[((xl + 2) * recv_dimy + (yl + 2)) * recv_dimz + (zl + 2)]
- workspace[((xr + 2) * recv_dimy + (yr + 2)) * recv_dimz + (zr + 2)]) -
(1.0 / 6) *
(workspace[((xll + 2) * recv_dimy + (yll + 2)) * recv_dimz + (zll + 2)] -
workspace[((xrr + 2) * recv_dimy + (yrr + 2)) * recv_dimz + (zrr + 2)]));
}
}
}
for(i = offsets[nB]; i < offsets[nB+1]; ++i) {
slab_x = to_slab_fac * P[i].Pos[0];
if(slab_x >= PMGRID)
slab_x = PMGRID - 1;
dx = to_slab_fac * P[i].Pos[0] - slab_x;
slab_x -= meshmin[0];
slab_xx = slab_x + 1;
slab_y = to_slab_fac * P[i].Pos[1];
if(slab_y >= PMGRID)
slab_y = PMGRID - 1;
dy = to_slab_fac * P[i].Pos[1] - slab_y;
slab_y -= meshmin[1];
slab_yy = slab_y + 1;
slab_z = to_slab_fac * P[i].Pos[2];
if(slab_z >= PMGRID)
slab_z = PMGRID - 1;
dz = to_slab_fac * P[i].Pos[2] - slab_z;
slab_z -= meshmin[2];
slab_zz = slab_z + 1;
acc_dim =
forcegrid[(slab_x * dimy + slab_y) * dimz + slab_z] * (1.0 - dx) * (1.0 - dy) * (1.0 - dz);
acc_dim += forcegrid[(slab_x * dimy + slab_yy) * dimz + slab_z] * (1.0 - dx) * dy * (1.0 - dz);
acc_dim += forcegrid[(slab_x * dimy + slab_y) * dimz + slab_zz] * (1.0 - dx) * (1.0 - dy) * dz;
acc_dim += forcegrid[(slab_x * dimy + slab_yy) * dimz + slab_zz] * (1.0 - dx) * dy * dz;
acc_dim += forcegrid[(slab_xx * dimy + slab_y) * dimz + slab_z] * (dx) * (1.0 - dy) * (1.0 - dz);
acc_dim += forcegrid[(slab_xx * dimy + slab_yy) * dimz + slab_z] * (dx) * dy * (1.0 - dz);
acc_dim += forcegrid[(slab_xx * dimy + slab_y) * dimz + slab_zz] * (dx) * (1.0 - dy) * dz;
acc_dim += forcegrid[(slab_xx * dimy + slab_yy) * dimz + slab_zz] * (dx) * dy * dz;
P[i].GravPM[dim] += acc_dim;
}
}
// KC 10/4/14
// Close out N_GRAVS extended loops
}
}
// KC 10/19/14
// DEBUG
// Output all the GravPM and then die
#if defined DEBUG_N_GRAVS_PMPERIODIC
for(i = 0; i < NumPart; ++i)
fprintf(stderr, "%d %g %g %g %g %g %g %d\n", P[i].ID, P[i].Pos[0], P[i].Pos[1], P[i].Pos[2], P[i].GravPM[0], P[i].GravPM[1], P[i].GravPM[2], P[i].Type);
endrun(6789);
#endif
pm_init_periodic_free();
force_treeallocate(All.TreeAllocFactor * All.MaxPart, All.MaxPart);
All.NumForcesSinceLastDomainDecomp = 1 + All.TotNumPart * All.TreeDomainUpdateFrequency;
if(ThisTask == 0)
{
printf("done PM.\n");
fflush(stdout);
}
}
/*! Calculates the long-range potential using the PM method. The potential is
* Gaussian filtered with Asmth, given in mesh-cell units. We carry out a CIC
* charge assignment, and compute the potenial by Fourier transform
* methods. The CIC kernel is deconvolved.
*/
void pmpotential_periodic(void)
{
double k2, kx, ky, kz, smth;
double dx, dy, dz;
double fx, fy, fz, ff;
double asmth2, fac;
int i, j, slab, level, sendTask, recvTask;
int x, y, z, ip;
int slab_x, slab_y, slab_z;
int slab_xx, slab_yy, slab_zz;
int meshmin[3], meshmax[3], sendmin, sendmax, recvmin, recvmax;
int rep, ncont, cont_sendmin[2], cont_sendmax[2], cont_recvmin[2], cont_recvmax[2];
int dimx, dimy, dimz, recv_dimx, recv_dimy, recv_dimz;
MPI_Status status;
// KC 10/5/14
// Additional N_GRAVS variables
int nA, nB;
int offsets[N_GRAVS+1];
int k;
if(ThisTask == 0)
{
printf("Starting periodic PM calculation.\n");
fflush(stdout);
}
asmth2 = (2 * M_PI) * All.Asmth[0] / All.BoxSize;
asmth2 *= asmth2;
fac = All.G / (M_PI * All.BoxSize); /* to get potential */
force_treefree();
for(j = 0; j < 3; j++)
{
meshmin[j] = PMGRID;
meshmax[j] = 0;
}
// KC 10/20/14
// Compute the gravitational offsets here
offsets[0] = 0;
for(k = 1; k < N_GRAVS; ++k)
offsets[k] = offsets[k-1] + NgravLocal[k-1];
offsets[N_GRAVS] = NumPart;
for(i = 0; i < NumPart; ++i) {
for(j = 0; j < 3; j++)
{
slab = to_slab_fac * P[i].Pos[j];
if(slab >= PMGRID)
slab = PMGRID - 1;
if(slab < meshmin[j])
meshmin[j] = slab;
if(slab > meshmax[j])
meshmax[j] = slab;
}
}
MPI_Allgather(meshmin, 3, MPI_INT, meshmin_list, 3, MPI_INT, MPI_COMM_WORLD);
MPI_Allgather(meshmax, 3, MPI_INT, meshmax_list, 3, MPI_INT, MPI_COMM_WORLD);
dimx = meshmax[0] - meshmin[0] + 2;
dimy = meshmax[1] - meshmin[1] + 2;
dimz = meshmax[2] - meshmin[2] + 2;
pm_init_periodic_allocate((dimx + 4) * (dimy + 4) * (dimz + 4));
// Sources
for(nA = 0; nA < N_GRAVS; ++nA) {
// Receivers
for(nB = 0; nB < N_GRAVS; ++nB) {
for(i = 0; i < dimx * dimy * dimz; i++)
workspace[i] = 0;
// So this is a range that we need to process
for(i = offsets[nA]; i < offsets[nA + 1]; ++i) {
slab_x = to_slab_fac * P[i].Pos[0];
if(slab_x >= PMGRID)
slab_x = PMGRID - 1;
dx = to_slab_fac * P[i].Pos[0] - slab_x;
slab_x -= meshmin[0];
slab_xx = slab_x + 1;
slab_y = to_slab_fac * P[i].Pos[1];
if(slab_y >= PMGRID)
slab_y = PMGRID - 1;
dy = to_slab_fac * P[i].Pos[1] - slab_y;
slab_y -= meshmin[1];
slab_yy = slab_y + 1;
slab_z = to_slab_fac * P[i].Pos[2];
if(slab_z >= PMGRID)
slab_z = PMGRID - 1;
dz = to_slab_fac * P[i].Pos[2] - slab_z;
slab_z -= meshmin[2];
slab_zz = slab_z + 1;
workspace[(slab_x * dimy + slab_y) * dimz + slab_z] += P[i].Mass * (1.0 - dx) * (1.0 - dy) * (1.0 - dz);
workspace[(slab_x * dimy + slab_yy) * dimz + slab_z] += P[i].Mass * (1.0 - dx) * dy * (1.0 - dz);
workspace[(slab_x * dimy + slab_y) * dimz + slab_zz] += P[i].Mass * (1.0 - dx) * (1.0 - dy) * dz;
workspace[(slab_x * dimy + slab_yy) * dimz + slab_zz] += P[i].Mass * (1.0 - dx) * dy * dz;
workspace[(slab_xx * dimy + slab_y) * dimz + slab_z] += P[i].Mass * (dx) * (1.0 - dy) * (1.0 - dz);
workspace[(slab_xx * dimy + slab_yy) * dimz + slab_z] += P[i].Mass * (dx) * dy * (1.0 - dz);
workspace[(slab_xx * dimy + slab_y) * dimz + slab_zz] += P[i].Mass * (dx) * (1.0 - dy) * dz;
workspace[(slab_xx * dimy + slab_yy) * dimz + slab_zz] += P[i].Mass * (dx) * dy * dz;
}
for(i = 0; i < fftsize; i++) /* clear local density field */
rhogrid[i] = 0;
for(level = 0; level < (1 << PTask); level++) /* note: for level=0, target is the same task */
{
sendTask = ThisTask;
recvTask = ThisTask ^ level;
if(recvTask < NTask)
{
/* check how much we have to send */
sendmin = 2 * PMGRID;
sendmax = -1;
for(slab_x = meshmin[0]; slab_x < meshmax[0] + 2; slab_x++)
if(slab_to_task[slab_x % PMGRID] == recvTask)
{
if(slab_x < sendmin)
sendmin = slab_x;
if(slab_x > sendmax)
sendmax = slab_x;
}
if(sendmax == -1)
sendmin = 0;
/* check how much we have to receive */
recvmin = 2 * PMGRID;
recvmax = -1;
for(slab_x = meshmin_list[3 * recvTask]; slab_x < meshmax_list[3 * recvTask] + 2; slab_x++)
if(slab_to_task[slab_x % PMGRID] == sendTask)
{
if(slab_x < recvmin)
recvmin = slab_x;
if(slab_x > recvmax)
recvmax = slab_x;
}
if(recvmax == -1)
recvmin = 0;
if((recvmax - recvmin) >= 0 || (sendmax - sendmin) >= 0) /* ok, we have a contribution to the slab */
{
recv_dimx = meshmax_list[3 * recvTask + 0] - meshmin_list[3 * recvTask + 0] + 2;
recv_dimy = meshmax_list[3 * recvTask + 1] - meshmin_list[3 * recvTask + 1] + 2;
recv_dimz = meshmax_list[3 * recvTask + 2] - meshmin_list[3 * recvTask + 2] + 2;
if(level > 0)
{
MPI_Sendrecv(workspace + (sendmin - meshmin[0]) * dimy * dimz,
(sendmax - sendmin + 1) * dimy * dimz * sizeof(fftw_real), MPI_BYTE, recvTask,
TAG_PERIODIC_C | nA << 6 | nB << 9, forcegrid,
(recvmax - recvmin + 1) * recv_dimy * recv_dimz * sizeof(fftw_real), MPI_BYTE,
recvTask, TAG_PERIODIC_C | nA << 6 | nB << 9, MPI_COMM_WORLD, &status);
}
else
{
memcpy(forcegrid, workspace + (sendmin - meshmin[0]) * dimy * dimz,
(sendmax - sendmin + 1) * dimy * dimz * sizeof(fftw_real));
}
for(slab_x = recvmin; slab_x <= recvmax; slab_x++)
{
slab_xx = (slab_x % PMGRID) - first_slab_of_task[ThisTask];
if(slab_xx >= 0 && slab_xx < slabs_per_task[ThisTask])
{
for(slab_y = meshmin_list[3 * recvTask + 1];
slab_y <= meshmax_list[3 * recvTask + 1] + 1; slab_y++)
{
slab_yy = slab_y;
if(slab_yy >= PMGRID)
slab_yy -= PMGRID;
for(slab_z = meshmin_list[3 * recvTask + 2];
slab_z <= meshmax_list[3 * recvTask + 2] + 1; slab_z++)
{
slab_zz = slab_z;
if(slab_zz >= PMGRID)
slab_zz -= PMGRID;
rhogrid[PMGRID * PMGRID2 * slab_xx + PMGRID2 * slab_yy + slab_zz] +=
forcegrid[((slab_x - recvmin) * recv_dimy +
(slab_y - meshmin_list[3 * recvTask + 1])) * recv_dimz +
(slab_z - meshmin_list[3 * recvTask + 2])];
}
}
}