-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathnested.c
1325 lines (1079 loc) · 44.7 KB
/
nested.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
/* Nested Sampling procedure for Crankite
*
* Skeleton of code from J.Skilling p188 of
* Data Analysis A Bayesian Tutorial D.S.Sivia
*
* Copyright (c) 2007 - 2013 Nikolas Burkoff, Csilla Varnai and David Wild
*
*/
#include<stdlib.h>
#include<stdio.h>
#include<math.h>
#include<string.h>
#include<float.h>
#include<time.h>
#include"error.h"
#include"params.h"
#include"vector.h"
#include"rotation.h"
#include"peptide.h"
#include"vdw.h"
#include"energy.h"
#include"metropolis.h"
#include"probe.h"
#include"nested.h"
#ifdef PARALLEL
#include<mpi.h>
#include"random16.h"
#include"flex.h"
#endif
#include"checkpoint_io.h"
int lowtemp = 0;
#define PLUS(x,y) (x>y ? x + log(1+exp(y-x)) : y + log(1+exp(x-y)))
/* A bit of ordering of the chains wrt. the logL in ascending order.
It takes the first one, and moves it towards the end a bit, or all
the way to the end, if it has the largest logL.
If the first one has the largest logL, it will be moved to the end. */
void heapifyhashout(ChainHash* chainhash ,int length){
int booly = 0;
int k = 1,i;
ChainHash v;
copyhash(&v,&(chainhash[k]));
while((booly == 0) && 2 * k <= length){
i = 2 * k;
if(i < length) //2 children
if(chainhash[i].ll > chainhash[i+1].ll) i += 1;
if(v.ll < chainhash[i].ll) booly = 1;
else{copyhash(&(chainhash[k]),&(chainhash[i])); k = i;}
}
copyhash(&(chainhash[k]),&v);
}
/* A bit of ordering of the chains wrt. the logL in ascending order.
It takes the last one that has probably just been added to the heap,
and moves it down the line a bit, or to the front, if it has the
smallest logL.
If the last one has the smallest logL, it will be moved to the front. */
void heapifyhashin(ChainHash *chainhash, int length){
int current = length;
int booly = 0;
ChainHash v;
while((booly == 0) && current > 1){
int next = current / 2;
if(chainhash[current].ll < chainhash[next].ll){
copyhash(&v,&(chainhash[next]));
copyhash(&(chainhash[next]),&(chainhash[current]));
copyhash(&(chainhash[current]),&v);
current /= 2;
}
else{
booly = 1;
}
}
}
void repairheap(ChainHash* chainhash,int N){
constructhashheap(chainhash, N);
}
void new_amplitude(Chain **cpoints, Biasmap *biasmap, int current_stored, simulation_params *sim_params, void *mpi_comm){
int rank = 0;
int P = 1;
double amptot = 0;
double oldamp = sim_params->amplitude;
Chaint *chaint;
chaint = (Chaint*)malloc(sizeof(Chaint));
chaint->aat = NULL; chaint->ergt = NULL; chaint->xaat = NULL; chaint->xaat_prev = NULL;
double currE;
int copies;
Chain temporary;
temporary.aa = NULL; temporary.xaa = NULL; temporary.erg = NULL; temporary.xaa_prev = NULL;
allocmem_chain(&temporary,(*cpoints)[0].NAA,(*cpoints)[0].Nchains);
//calculate amplitude
#ifdef PARALLEL
MPI_Comm *MPI_COMM = mpi_comm;
MPI_Status status;
MPI_Comm_size(*MPI_COMM, &P);
MPI_Comm_rank(*MPI_COMM, &rank);
if(P == 1){
for(int k = 0; k < 5; k++){
copies = (int)(rand()/(double)RAND_MAX * current_stored) % (current_stored);
sim_params->amplitude = oldamp;
copybetween(&temporary,&((*cpoints)[copies]));
aat_init(&temporary,chaint);
currE = -temporary.ll;
move(&temporary,chaint,biasmap,sim_params->logLstar,&currE,-1, sim_params);
for(int i = 1; (i < sim_params->iter_max || i < 1024); i++){
move(&temporary,chaint,biasmap,sim_params->logLstar,&currE,1, sim_params);
}
amptot += sim_params->amplitude;
}
sim_params->amplitude = amptot / 5;
return;
}
if (rank != 0) { // one chain per processor
#else
for(int k = 0; k < 5; k++){ //5 chains only
#endif
copies = (int)(rand()/(double)RAND_MAX * current_stored) % (current_stored);
//recalculate amplitude on temporary, not changing the actual NS points
copybetween(&temporary,&((*cpoints)[copies]));
aat_init(&temporary,chaint);
#ifndef PARALLEL
sim_params->amplitude = oldamp;
#endif
currE = -temporary.ll;
move(&temporary,chaint,biasmap,sim_params->logLstar,&currE,-1, sim_params);
for(int i = 1; (i < sim_params->iter_max || i < 1024); i++){
move(&temporary,chaint,biasmap,sim_params->logLstar,&currE,1, sim_params);
}
//collect and add up amplitude
#ifdef PARALLEL
//send my new amplitude
MPI_Send(&(sim_params->amplitude),1,MPI_DOUBLE,0,sim_params->iter*6+5,*MPI_COMM);
} else {
for(int k = 1; k < P;k++){
MPI_Recv(&(sim_params->amplitude),1,MPI_DOUBLE,k,sim_params->iter*6+5,*MPI_COMM,&status);
amptot += sim_params->amplitude;
}
sim_params->amplitude = amptot / (P-1);
}
#else
amptot += sim_params->amplitude;
}
sim_params->amplitude = amptot / 5;
//fprintf(stderr,"Amplitude: %f\n",-sim_params->amplitude);
#endif
//broadcast amplitude
#ifdef PARALLEL
//receive the new amplitude from rank 0
MPI_Bcast(&(sim_params->amplitude),1,MPI_DOUBLE,0,*MPI_COMM);
#endif
freemem_chain(&temporary);
freemem_chaint(chaint);
free(chaint);
}
#ifdef PARALLEL
/*set up 2 separate communicators, FLEX_WORLD for
* processors 0,1,...nma_params.number_of_processors >> the master processor and the
* (for now) 1 processor involved in FLEX
* and NS_WORLD for processors
* 0,nma_params.number_of_processors+1,FLEX_params.number_of_processors+2,...P-1
* for the standard NS part
*/
void setup_communicators(MPI_Comm * FLEX_WORLD, MPI_Comm * NS_WORLD, int S, int *P, int *rank,int *SP, int *Srank, int *NSgrouprank, int *FLEXgrouprank, int *in_NS){
MPI_Comm_size(MPI_COMM_WORLD, P);
MPI_Comm_rank(MPI_COMM_WORLD, rank);
if(*P-S < 1){
stop("You must have at least 1 processor running the NS simulation\n");
}
int setup_groups;
MPI_Group NS_GROUP, FLEX_GROUP, ALL_GROUP;
MPI_Comm_group( MPI_COMM_WORLD, &ALL_GROUP );
if(S > 0){
int *FLEX_group_ranks = malloc(sizeof(int)*(S+1));
for(setup_groups = 0; setup_groups <= S; setup_groups++ ){
FLEX_group_ranks[setup_groups] = setup_groups;
}
MPI_Group_incl(ALL_GROUP,S+1,FLEX_group_ranks,&FLEX_GROUP);
MPI_Comm_create(MPI_COMM_WORLD,FLEX_GROUP,FLEX_WORLD);
free(FLEX_group_ranks);
}
int *NS_group_ranks = malloc(sizeof(int)*(*P-S));
NS_group_ranks[0] = 0;
for(setup_groups = 1; setup_groups < *P-S; setup_groups++ ){
NS_group_ranks[setup_groups] = setup_groups+S;
}
MPI_Group_incl(ALL_GROUP,*P-S,NS_group_ranks,&NS_GROUP);
MPI_Comm_create(MPI_COMM_WORLD,NS_GROUP,NS_WORLD);
free(NS_group_ranks);
if(S>0 && *rank <= S){
MPI_Group_rank(FLEX_GROUP,FLEXgrouprank);
MPI_Comm_size(*FLEX_WORLD, SP);
MPI_Comm_rank(*FLEX_WORLD, Srank);
if(*Srank != 0) *in_NS = -1;
}
if(*rank == 0 || *rank > S){
MPI_Group_rank(NS_GROUP,NSgrouprank);
MPI_Comm_size(*NS_WORLD, P);
MPI_Comm_rank(*NS_WORLD, rank);
}
}
void deal_with_flex(int N, int finished,int rank, simulation_params *sim_params, MPI_Comm NS_WORLD, MPI_Comm FLEX_WORLD, ChainHash *chainhash, Chain* temporary,Chain *cpoints){
MPI_Status status;
int flex_status_for_ns;
int one = 1;
if(rank == 0){
int flex_ready = check_flex_ready(sim_params);
if(flex_ready == 1 || sim_params->iter == sim_params->iter_start){
if(flex_ready == 1){
flex_status_for_ns = 1;
MPI_Bcast(&flex_status_for_ns,1,MPI_INT,0,NS_WORLD);
MPI_Send(&(sim_params->logLstar),1,MPI_DOUBLE,one,10,FLEX_WORLD);
int number_to_recv = 0;
MPI_Recv(&number_to_recv,1,MPI_INT,one,11,FLEX_WORLD,&status);
MPI_Bcast(&number_to_recv,1,MPI_INT,0,NS_WORLD);
int receives;
for(receives = 0; receives < number_to_recv; receives++){
mpi_rec_chain(temporary, one, 0, &sim_params->logLstar, 0, FLEX_WORLD);
int overwrite;
overwrite = 1 + ((int)(rand()/(double)RAND_MAX * (N))) % (N);
chainhash[overwrite].ll = temporary->ll;
int where_to_send = chainhash[overwrite].processor;
MPI_Bcast(&where_to_send,1,MPI_INT,0,NS_WORLD);
if(where_to_send == 0){
copybetween(&cpoints[chainhash[overwrite].index],temporary);
}
else{
MPI_Send(&(chainhash[overwrite].index),1,MPI_INT, chainhash[overwrite].processor ,sim_params->iter,NS_WORLD);
mpi_send_chain(temporary,0,where_to_send,&sim_params->logLstar,sim_params->iter,NS_WORLD);
}
}
repairheap(chainhash,N);
}
else{
flex_status_for_ns = 0;
MPI_Bcast(&flex_status_for_ns,1,MPI_INT,0,NS_WORLD);
}
int tosend = 1;//1 + ((int)(rand()/(double)RAND_MAX * (N))) % (N);
int where_from = chainhash[tosend].processor;
MPI_Bcast(&where_from,1,MPI_INT,0,NS_WORLD);
if(where_from == 0){
copybetween(temporary,&cpoints[chainhash[tosend].index]);
}
else{
MPI_Send(&(chainhash[tosend].index),1,MPI_INT, chainhash[tosend].processor ,sim_params->iter,NS_WORLD);
mpi_rec_chain(temporary, chainhash[tosend].processor, 0, &sim_params->logLstar, sim_params->iter, NS_WORLD);
}
//chain sent to FLEX
mpi_send_chain(temporary,0,one,&sim_params->logLstar,1,FLEX_WORLD);
}
else{
flex_status_for_ns = -1; //done
MPI_Bcast(&flex_status_for_ns,1,MPI_INT,0,NS_WORLD);
}
}
else{ //not rank 0
MPI_Bcast(&flex_status_for_ns,1,MPI_INT,0,NS_WORLD);
if(flex_status_for_ns == 1){
int number_to_recv = 0;
MPI_Bcast(&number_to_recv,1,MPI_INT,0,NS_WORLD);
int receives;
for(receives = 0; receives < number_to_recv; receives++){
int want_it;
MPI_Bcast(&want_it,1,MPI_INT,0,NS_WORLD);
if(want_it == rank){
int which_index;
MPI_Recv(&(which_index),1,MPI_INT,0,sim_params->iter,NS_WORLD,&status);
mpi_rec_chain(&(cpoints[which_index]), 0, rank, &sim_params->logLstar , sim_params->iter,NS_WORLD);
}
}
}
if(flex_status_for_ns >= 0){
int where_from;
MPI_Bcast(&where_from,1,MPI_INT,0,NS_WORLD);
if(where_from == rank){
int which_index;
MPI_Recv(&which_index,1,MPI_INT, 0 ,sim_params->iter,NS_WORLD,&status);
mpi_send_chain(&cpoints[which_index],rank,0,&sim_params->logLstar,sim_params->iter,NS_WORLD);
}
}
}
if(finished == 0 && rank == 0){
char filename[DEFAULT_SHORT_STRING_LENGTH];
//strcpy(filename,sim_params->flex_params.output_path);
strcpy(filename,"finished");
FILE *fptr = fopen(filename,"w");
fclose(fptr);
}
}
#endif
void check_to_output_checkpoint_file(Chain* cpoints, int current_stored, int N, simulation_params *sim_params, int rank, void*comm_pointer){
//if outputting checkpoint file
// The output printing has to happen before updating the logX, weight, logZ etc., otherwise
// a) the logZ will contain the highest energy configuration's weight twice
// b) logX will have been advanced using the old run's alpha and we cannot correct for it any more (matters if P changes)
// This means we print the previous logX, logZ, logL*, H and amplitude
if(sim_params->checkpoint == 1){
sim_params->N = N;
if(sim_params->iter != sim_params->iter_start) {
//fprintf(stderr,"trying to print checkpoint\n");
output_checkpoint_file(cpoints, current_stored, sim_params,comm_pointer);
}
if (rank == 0) {
if( sim_params->iter % sim_params->num_NS_per_checkpoint == 0 ){
char *out = (char*)malloc(sizeof(char)*1010);
if(sim_params->outfile_name != NULL){
sprintf(out,"%s_%d",sim_params->outfile_name,sim_params->checkpoint_counter);
if (sim_params->outfile) fclose(sim_params->outfile);
sim_params->outfile = fopen(out, "w");
}
free(out);
}
}
}
}
void find_worst(simulation_params *sim_params, ChainHash *chainhash, int *heaplength, int N, int P ){
/*find P worst samples in positions, N,N-1,N-2,... */
for(int k = 0; k < P; k++){
copyhash(&(chainhash[0]),&(chainhash[1]));
copyhash(&(chainhash[1]),&(chainhash[*heaplength]));
copyhash(&(chainhash[*heaplength]),&(chainhash[0]));
(*heaplength)--;
heapifyhashout(chainhash,*heaplength);
}
sim_params->logLstar = chainhash[N-P+1].ll;
}
void update_NS_parameters(simulation_params *sim_params, int *converged, double *logZnew, double *lweight){
/*update weight = width * likelihood */
*lweight = sim_params->log_DeltaX + sim_params->logLstar*sim_params->thermobeta;
//fprintf(stderr,"LWJ %f %d\n",sim_params->log_DeltaX,sim_params->iter);
sim_params->logX = sim_params->logX_start + sim_params->Delta_logX*(sim_params->iter-sim_params->iter_start+1);
/* update evidence and information */
*logZnew = PLUS(sim_params->logZ,*lweight);
sim_params->H = exp(*lweight - *logZnew) * sim_params->logLstar * sim_params->thermobeta + exp(sim_params->logZ-*logZnew)*(sim_params->H+sim_params->logZ) - *logZnew;
/* control */
if(sim_params->checkpoint == 1 && sim_params->iter % sim_params->num_NS_per_checkpoint == 0 && sim_params->iter != sim_params->iter_start) {
fprintf(stderr,"%d %lf %lf %lf %lf %lf\n", sim_params->iter, sim_params->logX, sim_params->logLstar,sim_params->logZ,sim_params->H,sim_params->amplitude);
}
/*Check if converged */
if(sim_params->lowtemp == 1 && fabs(*logZnew - sim_params->logZ) < 1e-8) *converged = 0;
sim_params->logZ = *logZnew;
}
#ifdef PARALLEL
void MC_first(ChainHash *ChainHash,Chain *cpoints, Chaint* chaint, int current_stored, Biasmap *biasmap,simulation_params *sim_params, int rank, int N,int P, MPI_Comm *NSWORLD){
int i;
MPI_Status status;
double *newls = (double*)malloc(sizeof(double)*(1+N/P));
for(i = 0; i < current_stored; i++){
int iter;
double currE = -cpoints[i].ll;
for(iter= 0; iter < sim_params->number_initial_MC; iter++){
move(&(cpoints[i]),chaint,biasmap,0,&currE,0,sim_params);
}
cpoints[i].ll = -currE;
newls[i] = cpoints[i].ll;
}
if(rank == 0){
fprintf(stderr,"Beware: %d initial MC moves before NS started\n",sim_params->number_initial_MC);
for(i = 0; i < current_stored; i++){
ChainHash[1+i*P].ll = newls[i];
}
int proc;
if(P > 1){
for(proc = 1; proc < P; proc++){
MPI_Recv(newls,1+N/P,MPI_DOUBLE,proc,proc,*NSWORLD,&status);
for(i = 0; i < N/P+1; i++){
if(1+i*P+proc <= N) ChainHash[1+i*P+proc].ll = newls[i];
}
}
}
}
else{
MPI_Send(newls,1+N/P,MPI_DOUBLE,0,rank,*NSWORLD);
}
free(newls);
}
#ifndef FAST
void collect_chains(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int N,MPI_Comm *NS_WORLD,Instructions *instructions){
//first collect the P chains which will be the start of the MMC
int copies, got_all;
MPI_Status status;
int minus1 = -1;
//collect the chains
if (rank == 0) {
//random number generation would fail if N==P
if (N==P) stop("number of processors = number of active points. All chains would be scrapped at each MC step.");
if (N<P) stop("number of processors > number of active points. All chains would be scrapped at each MC step.");
//first collect the P chains which will be the start of the MMC
int * start_chains = (int*)malloc(sizeof(int)*P);
for(int k = 0; k < P; k++){
copies = 1 + ((int)(rand()/(double)RAND_MAX * (N-P))) % (N-P);
start_chains[k] = copies;
}
for(int k = 0; k < P; k++){
if(chainhash[start_chains[k]].processor == 0){
copybetween(&(chaincopies[k]),&(cpoints[chainhash[start_chains[k]].index]));
}
else{
MPI_Send(&(chainhash[start_chains[k]].index),1,MPI_INT,chainhash[start_chains[k]].processor,sim_params->iter,*NS_WORLD);
mpi_rec_chain(&(chaincopies[k]), chainhash[start_chains[k]].processor, 0,&sim_params->logLstar , sim_params->iter,*NS_WORLD);
}
}
//let all processors know we're done
for(int i = 1; i < P; i++){
MPI_Send(&minus1,1,MPI_INT,i,sim_params->iter,*NS_WORLD);
}
//then send one to each processor
for(int k = 1; k < P; k++){
mpi_send_chain(&(chaincopies[k]), 0, k,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
free(start_chains);
} //end rank 0
else{
//processor 0 is asking for a random P cpoints spread throughout all processors
//send any it wants
got_all = 0;
while(got_all != -1){
MPI_Recv(&got_all,1,MPI_INT,0,sim_params->iter,*NS_WORLD,&status);
if(got_all != -1) mpi_send_chain(&(cpoints[got_all]), rank,0, &sim_params->logLstar , sim_params->iter,*NS_WORLD);
}
//receive the chain I'm expected to MC
mpi_rec_chain(&chaincopies[0],0,rank,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
}
void return_and_reheap_chains(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int *heaplength, MPI_Comm *NS_WORLD, Instructions *instructions){
//recieve the P chains on the master processor
int minus1 = -1;
int got_all;
MPI_Status status;
if (rank == 0) {
for(int k = 1; k < P; k++){
mpi_rec_chain(&(chaincopies[k]), k, 0,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
}
else {
mpi_send_chain(&chaincopies[0],rank,0,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
if (rank == 0) {
//put them back in heap and send them to overwrite the worst ones
for(int k = 0; k < P; k++){
(*heaplength)++;
chainhash[*heaplength].ll = chaincopies[k].ll;
if(chainhash[*heaplength].processor == 0){
copybetween(&cpoints[chainhash[*heaplength].index],&chaincopies[k]);
}
else{
MPI_Send(&(chainhash[*heaplength].index),1,MPI_INT, chainhash[*heaplength].processor ,sim_params->iter,*NS_WORLD);
mpi_send_chain(&(chaincopies[k]),0, chainhash[*heaplength].processor,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
heapifyhashin(chainhash,*heaplength);
}
//tell processors we've given out all chains
for(int i = 1; i < P; i++){
MPI_Send(&minus1,1,MPI_INT,i,sim_params->iter,*NS_WORLD);
}
}
else {
//get given any of the new points to replace the worst P points in the collective cpoints arrays
got_all = 0;
while(got_all != -1){
MPI_Recv(&got_all,1,MPI_INT, 0 ,sim_params->iter,*NS_WORLD,&status);
if(got_all != -1) mpi_rec_chain(&(cpoints[got_all]),0,rank,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
}
}
#else
void initialize_instruction_set(Instructions * instruction, int P){
instruction->length = 6*P+1;
instruction->current_position = -1;
instruction->instructions = (int*)malloc(sizeof(int)*instruction->length);
}
void finalize_instruction_set(Instructions * instruction){
free(instruction->instructions);
}
void carry_out_first_instructions(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int N,MPI_Comm *NS_WORLD,Instructions *instructions){
instructions[0].current_position = -3;
int instruct; int proc;
int index;
do{
instructions[0].current_position+=3;
instruct = instructions[0].instructions[instructions[0].current_position];
proc = instructions[0].instructions[instructions[0].current_position+1];
index = instructions[0].instructions[instructions[0].current_position+2];
if(instruct == 1 ){ //send
mpi_send_chain(&(cpoints[index]), rank, proc,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
else if(instruct == -1){ //receive
mpi_rec_chain(&(chaincopies[0]), proc,rank,&sim_params->logLstar , sim_params->iter,*NS_WORLD);
}
}while(instruct != 0);
if(index != -1){
copybetween(&(chaincopies[0]),&(cpoints[index]));
}
instructions[0].current_position+=3;
}
void carry_out_last_instructions(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int N,MPI_Comm *NS_WORLD,Instructions *instructions){
instructions[0].current_position-=3;
int instruct; int proc;
int index;
do{
instructions[0].current_position+=3;
instruct = instructions[0].instructions[instructions[0].current_position];
proc = instructions[0].instructions[instructions[0].current_position+1];
index = instructions[0].instructions[instructions[0].current_position+2];
if(instruct == 1 ){ //send
mpi_send_chain(&(chaincopies[0]), rank, proc,&sim_params->logLstar,sim_params->iter,*NS_WORLD);
}
else if(instruct == -1){ //receive
mpi_rec_chain(&(cpoints[index]), proc,rank,&sim_params->logLstar , sim_params->iter,*NS_WORLD);
}
}while(instruct != 0);
if(index != -1){
copybetween(&(cpoints[index]),&(chaincopies[0]));
}
}
void output_instructions(Instructions *instructions, int rank){
fprintf(stderr, "Rank %d Number Instructions %d:\n",rank,instructions->current_position/3);
for(int i = 0; i < instructions->current_position/3; i++){
fprintf(stderr, "%d %d %d\n",instructions->instructions[3*i],instructions->instructions[3*i+1],instructions->instructions[3*i+2]);
}
}
void send_and_receive_instructions( Instructions * instructions, int rank, int P, MPI_Comm *NS_WORLD){
if(rank == 0){
//output_instructions(&(instructions[0]),0);
for(int k = 1; k < P; k++){
//output_instructions(&(instructions[k]),k);
MPI_Send(&(instructions[k].current_position),1,MPI_INT,k,25,*NS_WORLD);
MPI_Send(instructions[k].instructions,instructions[k].current_position,MPI_INT,k,50,*NS_WORLD);
}
}
else{
MPI_Status status;
MPI_Recv(&(instructions[0].current_position),1,MPI_INT,0,25,*NS_WORLD,&status);
MPI_Recv(instructions[0].instructions,instructions[0].current_position,MPI_INT,0,50,*NS_WORLD,&status);
}
}
void collect_chains(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int N,MPI_Comm *NS_WORLD,Instructions *instructions){
MPI_Bcast(&(sim_params->logLstar),1,MPI_DOUBLE,0,*NS_WORLD);
if (rank == 0) {
int *procs = malloc(sizeof(int)*P);
for(int k = 0; k < P; k++){
procs[k] = -1;
instructions[k].current_position = 0;
}
int *leftovers = malloc(sizeof(int)*(P*2));
int number_leftovers = 0;
int copies;
for(int k = 0; k < P; k++){
copies = 1 + ((int)(rand()/(double)RAND_MAX * (N-P))) % (N-P);
int which_proc = chainhash[copies].processor;
int which_index = chainhash[copies].index;
if(procs[which_proc] == -1){
procs[which_proc] = which_index;
}
else{
leftovers[number_leftovers*2] = which_proc;
leftovers[number_leftovers*2+1] = which_index;
number_leftovers++;
}
}
// static int number_chain = 0;
// number_chain += number_leftovers;
for(int k = 0; k < P; k++){
if(procs[k] == -1){
instructions[k].instructions[instructions[k].current_position] = -1;
instructions[k].instructions[instructions[k].current_position+1] = leftovers[(number_leftovers-1)*2];
instructions[k].instructions[instructions[k].current_position+2] = leftovers[(number_leftovers-1)*2+1];
instructions[k].current_position+=3;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position] = +1;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position+1] = k;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position+2] = leftovers[(number_leftovers-1)*2+1];
instructions[leftovers[(number_leftovers-1)*2]].current_position+=3;
number_leftovers--;
}
}
for(int k = 0; k < P; k++){
instructions[k].instructions[instructions[k].current_position] = 0;
instructions[k].instructions[instructions[k].current_position+1] = 0;
instructions[k].instructions[instructions[k].current_position+2] = procs[k];
instructions[k].current_position+=3;
}
//now sort out after MC move instructions
for(int k = 0; k < P; k++)procs[k] = -1;
number_leftovers = 0;
for(int k = 0; k < P; k++){
copies = N-k;
int which_proc = chainhash[copies].processor;
int which_index = chainhash[copies].index;
if(procs[which_proc] == -1){
procs[which_proc] = which_index;
}
else{
leftovers[number_leftovers*2] = which_proc;
leftovers[number_leftovers*2+1] = which_index;
number_leftovers++;
}
}
// number_chain += number_leftovers;
//fprintf(stderr,"NUMBER CHAINS %d\n",number_chain);
for(int k = 0; k < P; k++){
if(procs[k] == -1){
instructions[k].instructions[instructions[k].current_position] = 1;
instructions[k].instructions[instructions[k].current_position+1] = leftovers[(number_leftovers-1)*2];
instructions[k].instructions[instructions[k].current_position+2] = leftovers[(number_leftovers-1)*2+1];
instructions[k].current_position+=3;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position] = -1;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position+1] = k;
instructions[leftovers[(number_leftovers-1)*2]].instructions[instructions[leftovers[(number_leftovers-1)*2]].current_position+2] = leftovers[(number_leftovers-1)*2+1];
instructions[leftovers[(number_leftovers-1)*2]].current_position+=3;
number_leftovers--;
}
}
for(int k = 0; k < P; k++){
instructions[k].instructions[instructions[k].current_position] = 0;
instructions[k].instructions[instructions[k].current_position+1] = 0;
instructions[k].instructions[instructions[k].current_position+2] = procs[k];
instructions[k].current_position+=3;
}
free(procs);
free(leftovers);
}
send_and_receive_instructions(instructions,rank,P,NS_WORLD);
carry_out_first_instructions(chainhash,cpoints, chaincopies, sim_params, rank, P, N,NS_WORLD,instructions);
}
void return_and_reheap_chains(ChainHash *chainhash,Chain *cpoints, Chain*chaincopies, simulation_params *sim_params, int rank, int P, int *heaplength, MPI_Comm *NS_WORLD, Instructions *instructions){
carry_out_last_instructions(chainhash,cpoints, chaincopies, sim_params, rank, P, *heaplength,NS_WORLD,instructions);
if (rank == 0) {
MPI_Status status;
for(int k = 0; k < P; k++){
(*heaplength)++;
if(k == 0){
chainhash[*heaplength].ll = chaincopies[0].ll;
instructions[k].current_position+=2;
}
else{
instructions[k].current_position--;
MPI_Recv(&(chainhash[*heaplength].ll),1,MPI_DOUBLE,k,75,*NS_WORLD,&status);
}
//fprintf(stderr,"KK %d %d %d\n",k,instructions[k].current_position,instructions[k].instructions[instructions[k].current_position]); fflush(stderr);
if(instructions[k].instructions[instructions[k].current_position] == -1){
instructions[k].current_position -=3;
chainhash[*heaplength].processor = instructions[k].instructions[instructions[k].current_position-1];
}
else{
chainhash[*heaplength].processor = k;
}
chainhash[*heaplength].index = instructions[k].instructions[instructions[k].current_position];
heapifyhashin(chainhash,*heaplength);
}
}
else{
MPI_Send(&(chaincopies[0].ll),1,MPI_DOUBLE,0,75,*NS_WORLD);
}
}
#endif
#endif
/* Output NS point (in a serial run), or
collect the NS point through MPI and output it on the master node. */
void output_NS_point(ChainHash *chainhash, Chain *cpoints, Biasmap *biasmap, simulation_params *sim_params, int N,void*mpi_comm) {
Chain *chain_to_output = NULL;
int P = 1;
//get the chain to be output
#ifdef PARALLEL
MPI_Comm *MPI_COMM = mpi_comm;
int rank = 0;
MPI_Status status;
MPI_Comm_size(*MPI_COMM, &P);
MPI_Comm_rank(*MPI_COMM, &rank);
Chain temporary; //if the chain needs collecting from another processor
if (rank == 0) {
temporary.aa = NULL; temporary.xaa = NULL; temporary.erg = NULL; temporary.xaa_prev = NULL;
allocmem_chain(&temporary,cpoints[0].NAA,cpoints[0].Nchains);
//We need to update AA.id and AA.num, because those are not sent through MPI.
for (int i=0; i< sim_params->NAA; i++) {
temporary.aa[i].id = sim_params->seq[i];
temporary.aa[i].num = i;
}
//Need to actually get the chain first!
MPI_Bcast(&chainhash[N-P+1].processor,1,MPI_INT,0,*MPI_COMM);
if(chainhash[N-P+1].processor != 0){
MPI_Send(&chainhash[N-P+1].index,1,MPI_INT,chainhash[N-P+1].processor,sim_params->iter,*MPI_COMM );
mpi_rec_chain(&temporary, chainhash[N-P+1].processor, 0, &sim_params->logLstar,sim_params->iter,*MPI_COMM);
chain_to_output = &temporary;
}
else{
chain_to_output = &(cpoints[chainhash[N-P+1].index]);
}
}
else {
int whichP, whichindex;
//if it's my chain processor 0 wants then send it
MPI_Bcast(&whichP,1,MPI_INT,0,*MPI_COMM);
if(whichP == rank){
MPI_Recv(&whichindex,1,MPI_INT,0,sim_params->iter,*MPI_COMM,&status );
mpi_send_chain( &(cpoints[whichindex]), rank,0, &sim_params->logLstar,sim_params->iter,*MPI_COMM);
}
}
#else
chain_to_output = &(cpoints[chainhash[N-P+1].index]);
// chain_to_output = &(cpoints[chainhash[1].index]);
#endif
//do the tests on it
#ifdef PARALLEL
if (rank == 0) {
#endif
tests(chain_to_output,biasmap,sim_params->tmask, sim_params, 0x11, NULL);
// if((sim_params->tmask >> 15) & 0x1)fprintf(sim_params->outfile,"log(X): %f \t log(Evidence) estimate: %f\n",sim_params->logfactor*(sim_params->iter+1),sim_params->logZ);
if((sim_params->tmask >> 15) & 0x1)fprintf(sim_params->outfile,"log(X): %f \t log(Evidence) estimate: %f\n",sim_params->logX,sim_params->logZ);
if((sim_params->tmask >> 16) & 0x1)fprintf(sim_params->outfile,"Information estimate: %f\n",sim_params->H);
// fprintf(stderr,"%f ",sim_params->logfactor * (sim_params->iter+1));
// fprintf(stderr,"bad %f ",sim_params->Delta_logX * (sim_params->iter+1));
// fprintf(stderr,",%f + %f * (%d - %d + 1) = %f, ",sim_params->logX_start,sim_params->Delta_logX,sim_params->iter,sim_params->iter_start,sim_params->logX);
fprintf(stderr,"%f ",sim_params->logX);
#ifdef PARALLEL
freemem_chain(&temporary);
}
#endif
}
/* Nested Sampling routine.
Eventually, we want to remove the reading in routine from here and put it into main. */
void nestedsampling(int thinning, int maxiter, simulation_params *sim_params){
//set P and rank for serial job
int P = 1; //number of processors used for NS
int rank = 0; //master process
void *comm_pointer = NULL;
//NS points and their chaint-s
Chain* cpoints = NULL;
Chaint *chaint= (Chaint*)malloc(sizeof(Chaint));
chaint->aat = NULL; chaint->ergt = NULL; chaint->xaat = NULL; chaint->xaat_prev = NULL;
//biasmap
Biasmap *biasmap = (Biasmap*)malloc(sizeof(Biasmap));
biasmap->distb = NULL;
//temporary chain for reading in
Chain* temporary = NULL;
temporary = (Chain *)malloc(sizeof(Chain));
temporary->aa = NULL; temporary->xaa = NULL; temporary->erg = NULL; temporary->xaa_prev = NULL;
//number of chains
int N = 0;
//number of chains stored on this processor
int current_stored = 0;
time_t timer1 = 0;
time_t timer2 = 0;
#ifdef PARALLEL
timer1 = MPI_Wtime();
#else
timer1 = time(NULL);
#endif
#ifdef PARALLEL
MPI_Comm FLEX_WORLD, NS_WORLD;
int S = sim_params->flex_params.number_of_processors;
int SP,Srank, NSgrouprank,FLEXgrouprank;
/*If only reading in from checkpoint file then do not use FLEX processor, even if it was asked for*/
if(sim_params->num_NS_per_checkpoint == -1 && S != 0){
S=0;
fprintf(stderr,"Note, ignoring FLEX options as -C -1,FILENAME is used!\n");
}
int in_NS = 1;
setup_communicators(&FLEX_WORLD,&NS_WORLD,S,&P,&rank,&SP,&Srank,&NSgrouprank,&FLEXgrouprank,&in_NS);
/* P and rank now refer to the rank and number of processors in NS_WORLD and
* SP and srank refer to the rank and number of processors in FLEX_WORLD*/
if(rank == 0){
fprintf(stderr,"Nested Sampling using %d processors + %d FLEX processors\n",P,S);
}
if(in_NS == -1){
ns_for_flex_processor(FLEX_WORLD,FLEXgrouprank,biasmap,sim_params);
return;
}
else{
comm_pointer = (void*)(&NS_WORLD);
}
//parallel variable changes
maxiter /= P;
maxiter++;
thinning /= P;
if(thinning == 0) thinning = 1;
#endif
Instructions *instructions=NULL;
int copies = 0;
//in case restarting from checkpoint
int only_output_checkpoint = 0;
if(sim_params->num_NS_per_checkpoint == -1)
only_output_checkpoint = 1;
else sim_params->num_NS_per_checkpoint /= P;
if(sim_params->num_NS_per_checkpoint == 0) sim_params->num_NS_per_checkpoint = 1;
//initialising Nested Sampling variables
sim_params->iter_start = 0; //this is to make sure we restart after checkpoint at correct place
sim_params->iter = 0;
sim_params->H = 0.0; //information estimate
sim_params->logZ = -DBL_MAX; //evidence or partition function estimate
sim_params->amplitude = -M_PI; //MCMC crankshaft amplitude
sim_params->alpha = 1; //shrinkage ratio of X, the available prior space ratio
sim_params->logX = 0; //log of the available prior space ratio estimate
sim_params->logX_start = 0; //log of the available prior space ratio estimate -- this is important if we restart the simulation
sim_params->Delta_logX = 0; //step length along the logX axis: logX_{i-1} - logX_i = log alpha
sim_params->log_DeltaX = 0; //current width of the prior space weight: log( X_{i-1} - X_i ) = i * log(1 - alpha)
//local Nested Sampling variables
double currE, logZnew, lweight;
//the array where the chains are stored
//this stores the processor, index and ll of cpoints arrays from ALL processors
ChainHash* chainhash = NULL; //only used by master processor of parallel
//this is used to bring in and send out the P chains at each NS interation and for new sample point generation by MCMC
Chain* chaincopies = NULL;
//read in chains from either PDB file or checkpoint file
if(sim_params->restart_from_checkpoint == 1){
//note if only outputting from checkpoint then the tests are performed in here and no chains are stored or sent
current_stored = read_in_from_checkpoint(sim_params, &biasmap, temporary, &cpoints, &chaint, &chainhash, rank, P,comm_pointer,only_output_checkpoint);
} else {
//this will also map and initialise the chains
time_t timer_init1 = 0, timer_init2 = 0;
#ifdef PARALLEL
timer_init1 = MPI_Wtime();