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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
|
/*
* MrBayes 3
*
* (c) 2002-2010
*
* This file originally contributed by:
*
* Marc A. Suchard
* Department of Biomathematics
* University of California, Los Angeles
* Los Angeles, CA 90095
* msuchard@ucla.edu
*
* With important contributions by
*
* Maxim Teslenko (maxim.teslenko@nrm.se)
*
* and by many users (run 'acknowledgements' to see more info)
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details (www.gnu.org).
*
*/
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <math.h>
#include <string.h>
#include <ctype.h>
#include <assert.h>
#include "mb.h"
#include "mbbeagle.h"
#include "globals.h"
#include "utils.h"
#include "mcmc.h"
#include "model.h"
const char* const svnRevisionMbbeagleC="$Rev: 469 $"; /* Revision keyword which is expended/updated by svn on each commit/update*/
/* Functions and variables defined in mcmc.c that are not exported in mcmc.h */
void LaunchLogLikeForDivision(int chain, int d, MrBFlt* lnL);
void FlipCondLikeSpace (ModelInfo *m, int chain, int nodeIndex);
void FlipNodeScalerSpace (ModelInfo *m, int chain, int nodeIndex);
void FlipSiteScalerSpace (ModelInfo *m, int chain);
void ResetSiteScalers (ModelInfo *m, int chain);
void CopySiteScalers (ModelInfo *m, int chain);
int TreeCondLikes_Beagle (Tree *t, int division, int chain);
int TreeLikelihood_Beagle (Tree *t, int division, int chain, MrBFlt *lnL, int whichSitePats);
int TreeTiProbs_Beagle (Tree *t, int division, int chain);
int TreeCondLikes_Beagle_No_Rescale (Tree *t, int division, int chain);
int TreeCondLikes_Beagle_Rescale_All (Tree *t, int division, int chain);
MrBFlt GetRate (int division, int chain);
void FlipTiProbsSpace (ModelInfo *m, int chain, int nodeIndex);
extern int *chainId;
extern int numLocalChains;
//#define DEBUG_MB_BEAGLE_FLOW
#if defined (BEAGLE_ENABLED)
/*------------------------------------------------------------------------
|
| InitBeagleInstance: create and initialize a beagle instance
|
-------------------------------------------------------------------------*/
int InitBeagleInstance (ModelInfo *m, int division)
{
int i, j, k, c, s, *inStates, numPartAmbigTips;
double *inPartials;
SafeLong *charBits;
BeagleInstanceDetails details;
long preferedFlags, requiredFlags;
int resource;
if (m->useBeagle == NO)
return ERROR;
/* at least one eigen buffer needed */
if (m->nCijkParts == 0)
m->nCijkParts = 1;
/* allocate memory used by beagle */
m->logLikelihoods = (MrBFlt *) SafeCalloc ((numLocalChains)*m->numChars, sizeof(MrBFlt));
m->inRates = (MrBFlt *) SafeCalloc (m->numGammaCats, sizeof(MrBFlt));
m->branchLengths = (MrBFlt *) SafeCalloc (2*numLocalTaxa, sizeof(MrBFlt));
m->tiProbIndices = (int *) SafeCalloc (2*numLocalTaxa, sizeof(int));
m->inWeights = (MrBFlt *) SafeCalloc (m->numGammaCats*m->nCijkParts, sizeof(MrBFlt));
m->bufferIndices = (int *) SafeCalloc (m->nCijkParts, sizeof(int));
m->eigenIndices = (int *) SafeCalloc (m->nCijkParts, sizeof(int));
m->childBufferIndices = (int *) SafeCalloc (m->nCijkParts, sizeof(int));
m->childTiProbIndices = (int *) SafeCalloc (m->nCijkParts, sizeof(int));
m->cumulativeScaleIndices = (int *) SafeCalloc (m->nCijkParts, sizeof(int));
numPartAmbigTips = 0;
if (m->numStates != m->numModelStates)
numPartAmbigTips = numLocalTaxa;
else
{
for (i=0; i<numLocalTaxa; i++)
{
if (m->isPartAmbig[i] == YES)
numPartAmbigTips++;
}
}
if (beagleResourceCount == 0)
{
preferedFlags = beagleFlags;
}
else
{
resource = beagleResource[beagleInstanceCount % beagleResourceCount];
preferedFlags = beagleFlags;
}
requiredFlags = 0L;
if (beagleScalingScheme == MB_BEAGLE_SCALE_ALWAYS)
requiredFlags |= BEAGLE_FLAG_SCALERS_LOG; //BEAGLE_FLAG_SCALERS_RAW;
/* TODO: allocate fewer buffers when nCijkParts > 1 */
/* create beagle instance */
m->beagleInstance = beagleCreateInstance(numLocalTaxa,
m->numCondLikes * m->nCijkParts,
numLocalTaxa - numPartAmbigTips,
m->numModelStates,
m->numChars,
(numLocalChains + 1) * m->nCijkParts,
m->numTiProbs*m->nCijkParts,
m->numGammaCats,
m->numScalers * m->nCijkParts,
(beagleResourceCount == 0 ? NULL : &resource),
(beagleResourceCount == 0 ? 0 : 1),
preferedFlags,
requiredFlags,
&details);
if (m->beagleInstance < 0)
{
MrBayesPrint ("%s Failed to start BEAGLE instance\n", spacer);
return (ERROR);
}
else
{
MrBayesPrint( "\n%s Using BEAGLE resource %i for division %d:", spacer, details.resourceNumber, division+1);
#if defined (THREADS_ENABLED)
MrBayesPrint( " (%s)\n", (tryToUseThreads ? "threaded" : "non-threaded"));
#else
MrBayesPrint( " (non-threaded)\n");
#endif
MrBayesPrint( "%s Rsrc Name : %s\n", spacer, details.resourceName);
MrBayesPrint( "%s Impl Name : %s\n", spacer, details.implName);
MrBayesPrint( "%s Flags:", spacer);
BeaglePrintFlags(details.flags);
MrBayesPrint( "\n");
beagleInstanceCount++;
}
/* initialize tip data */
inStates = (int *) SafeMalloc (m->numChars * sizeof(int));
if (!inStates)
return ERROR;
inPartials = (double *) SafeMalloc (m->numChars * m->numModelStates * sizeof(double));
if (!inPartials)
return ERROR;
for (i=0; i<numLocalTaxa; i++)
{
if (m->isPartAmbig[i] == NO)
{
charBits = m->parsSets[i];
for (c=0; c<m->numChars; c++)
{
for (s=j=0; s<m->numModelStates; s++)
{
if (IsBitSet(s, charBits))
{
inStates[c] = s;
j++;
}
}
if (j == m->numModelStates)
inStates[c] = j;
else
assert(j==1);
charBits += m->nParsIntsPerSite;
}
beagleSetTipStates(m->beagleInstance, i, inStates);
}
else /* if (m->isPartAmbig == YES) */
{
k = 0;
charBits = m->parsSets[i];
for (c=0; c<m->numChars; c++)
{
for (s=0; s<m->numModelStates; s++)
{
if (IsBitSet(s%m->numStates, charBits))
inPartials[k++] = 1.0;
else
inPartials[k++] = 0.0;
}
charBits += m->nParsIntsPerSite;
}
beagleSetTipPartials(m->beagleInstance, i, inPartials);
}
}
free (inStates);
free (inPartials);
return NO_ERROR;
}
/*-----------------------------------------------------------------
|
| LaunchBEAGLELogLikeForDivision: calculate the log likelihood
| of the new state of the chain for a single division
|
-----------------------------------------------------------------*/
void LaunchBEAGLELogLikeForDivision(int chain, int d, ModelInfo* m, Tree* tree, MrBFlt* lnL) {
int i, rescaleFreqNew;
int *isScalerNode;
TreeNode *p;
if (beagleScalingScheme == MB_BEAGLE_SCALE_ALWAYS)
{
#if defined (DEBUG_MB_BEAGLE_FLOW)
printf("ALWAYS RESCALING\n");
#endif
/* Flip and copy or reset site scalers */
FlipSiteScalerSpace(m, chain);
if (m->upDateAll == YES) {
for (i=0; i<m->nCijkParts; i++) {
beagleResetScaleFactors(m->beagleInstance, m->siteScalerIndex[chain] + i);
}
}
else
CopySiteScalers(m, chain);
TreeTiProbs_Beagle(tree, d, chain);
TreeCondLikes_Beagle(tree, d, chain);
TreeLikelihood_Beagle(tree, d, chain, lnL, (chainId[chain] % chainParams.numChains));
}
else
{ /* MB_BEAGLE_SCALE_DYNAMIC */
/* This flag is only valid within this block */
m->rescaleBeagleAll = NO;
TreeTiProbs_Beagle(tree, d, chain);
if( m->succesCount[chain] > 1000 )
{
m->succesCount[chain] = 10;
m->rescaleFreq[chain]++; /* increase rescaleFreq independent of whether we accept or reject new state*/
m->rescaleFreqOld = rescaleFreqNew = m->rescaleFreq[chain];
for (i=0; i<tree->nIntNodes; i++)
{
p = tree->intDownPass[i];
if ( p->upDateCl == YES ) {
/* flip to the new workspace since TreeCondLikes_Beagle_Rescale_All() does not do it for
(p->upDateCl == YES) since it assumes that TreeCondLikes_Beagle_No_Rescale() did it */
FlipCondLikeSpace (m, chain, p->index);
}
}
goto rescale_all;
}
if( beagleScalingFrequency != 0 &&
m->beagleComputeCount[chain] % (beagleScalingFrequency) == 0 )
{
m->rescaleFreqOld = rescaleFreqNew = m->rescaleFreq[chain];
for (i=0; i<tree->nIntNodes; i++)
{
p = tree->intDownPass[i];
if ( p->upDateCl == YES ) {
/* flip to the new workspace since TreeCondLikes_Beagle_Rescale_All() does not do it for (p->upDateCl == YES) since it assumes that TreeCondLikes_Beagle_No_Rescale() did it*/
FlipCondLikeSpace (m, chain, p->index);
}
}
goto rescale_all;
}
TreeCondLikes_Beagle_No_Rescale(tree, d, chain);
/* Check if likelihood is valid */
if( TreeLikelihood_Beagle(tree, d, chain, lnL, (chainId[chain] % chainParams.numChains)) == BEAGLE_ERROR_FLOATING_POINT )
{
m->rescaleFreqOld = rescaleFreqNew = m->rescaleFreq[chain];
if(rescaleFreqNew > 1 && m->succesCount[chain] < 40)
{
if( m->succesCount[chain] < 10 )
{
if( m->succesCount[chain] < 4 )
{
rescaleFreqNew-= rescaleFreqNew >> 3; /* <== we cut up to 12,5% of rescaleFreq */
if( m->succesCount[chain] < 2 )
{
rescaleFreqNew-= rescaleFreqNew >> 3;
/* to avoid situation when we may stack at high rescaleFreq when new states do not get accepted because of low liklihood but there proposed frequency is high we reduce rescaleFreq even if we reject the last move*/
/* basically the higher probability of proposing of low liklihood state which needs smaller rescaleFreq would lead to higher probability of hitting this code which should reduce rescaleFreqOld thus reduce further probability of hitting this code */
/* at some point this negative feedback mechanism should get in balance with the mechanism of periodically increasing rescaleFreq when long sequence of successes is achieved*/
m->rescaleFreqOld-= m->rescaleFreqOld >> 3;
}
m->rescaleFreqOld-= m->rescaleFreqOld >> 3;
m->rescaleFreqOld--;
m->rescaleFreqOld = ( m->rescaleFreqOld ? m->rescaleFreqOld:1);
m->recalculateScalers = YES;
recalcScalers = YES;
}
}
rescaleFreqNew--;
rescaleFreqNew = ( rescaleFreqNew ? rescaleFreqNew:1);
}
m->succesCount[chain] = 0;
rescale_all:
#if defined (DEBUG_MB_BEAGLE_FLOW)
printf("NUMERICAL RESCALING\n");
#endif
m->rescaleBeagleAll = YES;
FlipSiteScalerSpace(m, chain);
isScalerNode = m->isScalerNode[chain];
while_loop:
ResetScalersPartition ( isScalerNode, tree, rescaleFreqNew );
for (i=0; i<m->nCijkParts; i++)
{
beagleResetScaleFactors(m->beagleInstance, m->siteScalerIndex[chain] + i);
}
TreeCondLikes_Beagle_Rescale_All (tree, d, chain);
if( TreeLikelihood_Beagle(tree, d, chain, lnL, (chainId[chain] % chainParams.numChains)) == BEAGLE_ERROR_FLOATING_POINT )
{
if( rescaleFreqNew > 1 )
{
/*Swap back scalers which were swapped in TreeCondLikes_Beagle_Rescale_All() */
for (i=0; i<tree->nIntNodes; i++)
{
p = tree->intDownPass[i];
if ( isScalerNode[p->index] == YES)
FlipNodeScalerSpace (m, chain, p->index);
}
rescaleFreqNew -= rescaleFreqNew >> 3; /*<== we cut up to 12,5% of rescaleFreq */
rescaleFreqNew--; /* we cut extra 1 of rescaleFreq */
goto while_loop;
}
}
m->rescaleFreq[chain] = rescaleFreqNew;
}
}
/* Count number of evaluations */
m->beagleComputeCount[chain]++;
}
void recalculateScalers(int chain)
{
int i, d, rescaleFreqNew;
int *isScalerNode;
ModelInfo* m;
Tree *tree;
for (d=0; d<numCurrentDivisions; d++)
{
m = &modelSettings[d];
if( m->recalculateScalers == YES)
{
m->recalculateScalers = NO;
tree = GetTree(m->brlens, chain, state[chain]);
rescaleFreqNew = m->rescaleFreq[chain];
isScalerNode = m->isScalerNode[chain];
ResetScalersPartition ( isScalerNode, tree, rescaleFreqNew );
for (i=0; i<m->nCijkParts; i++) {
beagleResetScaleFactors(m->beagleInstance, m->siteScalerIndex[chain] + i);
}
/* here it does not matter if we flip CL space or not */
TreeCondLikes_Beagle_Rescale_All (tree, d, chain);
}
}
}
void BeagleAddGPUDevicesToList(int **newResourceList, int *beagleResourceCount) {
BeagleResourceList* beagleResources;
int i, gpuCount;
beagleResources = beagleGetResourceList();
if (*newResourceList == NULL) {
*newResourceList = (int*) SafeCalloc(sizeof(int), beagleResources->length);
}
gpuCount = 0;
for (i = 0; i < beagleResources->length; i++) {
if (beagleResources->list[i].supportFlags & BEAGLE_FLAG_PROCESSOR_GPU) {
(*newResourceList)[gpuCount] = i;
gpuCount++;
}
}
*beagleResourceCount = gpuCount;
}
void BeagleRemoveGPUDevicesFromList(int **beagleResource, int *beagleResourceCount) {
*beagleResourceCount = 0;
}
/*-----
|
| BeaglePrintResources: outputs the available BEAGLE resources
|
----------*/
void BeaglePrintResources()
{
int i;
BeagleResourceList* beagleResources;
beagleResources = beagleGetResourceList();
MrBayesPrint ("%s Available resources reported by beagle library:\n", spacer);
for (i=0; i<beagleResources->length; i++)
{
MrBayesPrint ("\tResource %i:\n", i);
MrBayesPrint ("\tName: %s\n", beagleResources->list[i].name);
if (i > 0)
{
MrBayesPrint ("\tDesc: %s\n", beagleResources->list[i].description);
}
MrBayesPrint("\tFlags:");
BeaglePrintFlags(beagleResources->list[i].supportFlags);
MrBayesPrint("\n\n");
}
}
int BeagleCheckFlagCompatability(long inFlags) {
if (inFlags & BEAGLE_FLAG_PROCESSOR_GPU) {
if (inFlags & BEAGLE_FLAG_VECTOR_SSE) {
MrBayesPrint("%s Simultaneous use of GPU and SSE not available.\n", spacer);
return NO;
}
if (inFlags & BEAGLE_FLAG_THREADING_OPENMP) {
MrBayesPrint("%s Simultaneous use of GPU and OpenMP not available.\n", spacer);
return NO;
}
}
return YES;
}
/*-------------------
|
| BeaglePrintFlags: outputs beagle instance details
|
______________________*/
void BeaglePrintFlags(long inFlags)
{
int i, k;
char *names[] = { "PROCESSOR_CPU",
"PROCESSOR_GPU",
"PROCESSOR_FPGA",
"PROCESSOR_CELL",
"PRECISION_DOUBLE",
"PRECISION_SINGLE",
"COMPUTATION_ASYNCH",
"COMPUTATION_SYNCH",
"EIGEN_REAL",
"EIGEN_COMPLEX",
"SCALING_MANUAL",
"SCALING_AUTO",
"SCALING_ALWAYS",
"SCALING_DYNAMIC",
"SCALERS_RAW",
"SCALERS_LOG",
"VECTOR_NONE",
"VECTOR_SSE",
"THREADING_NONE",
"THREADING_OPENMP"
};
long flags[] = { BEAGLE_FLAG_PROCESSOR_CPU,
BEAGLE_FLAG_PROCESSOR_GPU,
BEAGLE_FLAG_PROCESSOR_FPGA,
BEAGLE_FLAG_PROCESSOR_CELL,
BEAGLE_FLAG_PRECISION_DOUBLE,
BEAGLE_FLAG_PRECISION_SINGLE,
BEAGLE_FLAG_COMPUTATION_ASYNCH,
BEAGLE_FLAG_COMPUTATION_SYNCH,
BEAGLE_FLAG_EIGEN_REAL,
BEAGLE_FLAG_EIGEN_COMPLEX,
BEAGLE_FLAG_SCALING_MANUAL,
BEAGLE_FLAG_SCALING_AUTO,
BEAGLE_FLAG_SCALING_ALWAYS,
BEAGLE_FLAG_SCALING_DYNAMIC,
BEAGLE_FLAG_SCALERS_RAW,
BEAGLE_FLAG_SCALERS_LOG,
BEAGLE_FLAG_VECTOR_NONE,
BEAGLE_FLAG_VECTOR_SSE,
BEAGLE_FLAG_THREADING_NONE,
BEAGLE_FLAG_THREADING_OPENMP
};
for (i=k=0; i<20; i++)
{
if (inFlags & flags[i])
{
if (k%4 == 0 && k > 0)
MrBayesPrint("\n%s ", spacer);
MrBayesPrint (" %s", names[i]);
k++;
}
}
}
int ScheduleLogLikeForAllDivisions() {
int d;
int divisionsToLaunch = 0;
ModelInfo *m;
if (numCurrentDivisions < 2) {
return 0;
}
for (d=0; d<numCurrentDivisions; d++) {
m = &modelSettings[d];
if (m->upDateCl == YES) {
divisionsToLaunch++;
}
}
return (divisionsToLaunch > 1);
}
#if defined(THREADS_ENABLED)
void *LaunchThreadLogLikeForDivision(void *arguments) {
int d, chain;
MrBFlt *lnL;
LaunchStruct* launchStruct;
launchStruct = (LaunchStruct*) arguments;
chain = launchStruct->chain;
d = launchStruct->division;
lnL = launchStruct->lnL;
LaunchLogLikeForDivision(chain, d, lnL);
return 0;
}
MrBFlt LaunchLogLikeForAllDivisionsInParallel(int chain) {
int d;
int threadError;
pthread_t* threads;
LaunchStruct* launchValues;
int* wait;
ModelInfo* m;
MrBFlt chainLnLike;
chainLnLike = 0.0;
/* TODO Initialize only once */
threads = (pthread_t*) SafeMalloc(sizeof(pthread_t) * numCurrentDivisions);
launchValues = (LaunchStruct*) SafeMalloc(sizeof(LaunchStruct) * numCurrentDivisions);
wait = (int*) SafeMalloc(sizeof(int) * numCurrentDivisions);
/* Cycle through divisions and recalculate tis and cond likes as necessary. */
/* Code below does not try to avoid recalculating ti probs for divisions */
/* that could share ti probs with other divisions. */
for (d=0; d<numCurrentDivisions; d++)
{
#if defined (BEST_MPI_ENABLED)
if (isDivisionActive[d] == NO)
continue;
#endif
m = &modelSettings[d];
if (m->upDateCl == YES)
{
launchValues[d].chain = chain;
launchValues[d].division = d;
launchValues[d].lnL = &(m->lnLike[2*chain + state[chain]]);
/* Fork */
threadError = pthread_create(&threads[d], NULL,
LaunchThreadLogLikeForDivision,
(void*) &launchValues[d]);
assert(0 == threadError);
wait[d] = 1;
}
else
{
wait[d] = 0;
}
}
for (d = 0; d < numCurrentDivisions; d++)
{
/* Join */
if (wait[d])
{
threadError = pthread_join(threads[d], NULL);
assert(0 == threadError);
}
m = &modelSettings[d];
chainLnLike += m->lnLike[2*chain + state[chain]];
}
/* TODO Free these once */
free(threads);
free(launchValues);
free(wait);
return chainLnLike;
}
#endif
/*----------------------------------------------------------------
|
| TreeCondLikes_Beagle: This routine updates all conditional
| (partial) likelihoods of a beagle instance while doing no rescaling.
| That potentialy can make final liklihood bad then calculation with rescaling needs to be done.
|
-----------------------------------------------------------------*/
int TreeCondLikes_Beagle_No_Rescale (Tree *t, int division, int chain)
{
int i, j, cumulativeScaleIndex;
BeagleOperation operations;
TreeNode *p;
ModelInfo *m;
unsigned chil1Step, chil2Step;
int *isScalerNode;
m = &modelSettings[division];
isScalerNode = m->isScalerNode[chain];
for (i=0; i<t->nIntNodes; i++)
{
p = t->intDownPass[i];
/* check if conditional likelihoods need updating */
if (p->upDateCl == YES)
{
/* flip to the new workspace */
FlipCondLikeSpace (m, chain, p->index);
/* update conditional likelihoods */
operations.destinationPartials = m->condLikeIndex[chain][p->index ];
operations.child1Partials = m->condLikeIndex[chain][p->left->index ];
operations.child1TransitionMatrix = m->tiProbsIndex [chain][p->left->index ];
operations.child2Partials = m->condLikeIndex[chain][p->right->index];
operations.child2TransitionMatrix = m->tiProbsIndex [chain][p->right->index];
/* All partials for tips are the same across omega catigoris, thus we are doing the following two if statments.*/
if(p->left->left== NULL )
chil1Step=0;
else
chil1Step=1;
if(p->right->left== NULL )
chil2Step=0;
else
chil2Step=1;
operations.destinationScaleWrite = BEAGLE_OP_NONE;
cumulativeScaleIndex = BEAGLE_OP_NONE;
if ( isScalerNode[p->index] == YES )
{
operations.destinationScaleRead = m->nodeScalerIndex[chain][p->index];
}
else
{
operations.destinationScaleRead = BEAGLE_OP_NONE;
}
for (j=0; j<m->nCijkParts; j++)
{
beagleUpdatePartials(m->beagleInstance,
&operations,
1,
cumulativeScaleIndex);
operations.destinationPartials++;
operations.child1Partials+=chil1Step;
operations.child1TransitionMatrix++;
operations.child2Partials+=chil2Step;
operations.child2TransitionMatrix++;
if ( isScalerNode[p->index] == YES )
operations.destinationScaleRead++;
}
}
}
return NO_ERROR;
}
/*----------------------------------------------------------------
|
| TreeCondLikes_Beagle: This routine updates all conditional
| (partial) likelihoods of a beagle instance while rescaling at every node.
| Note: all nodes get recalculated, not only tached by move.
|
-----------------------------------------------------------------*/
int TreeCondLikes_Beagle_Rescale_All (Tree *t, int division, int chain)
{
int i, j, cumulativeScaleIndex;
BeagleOperation operations;
TreeNode *p;
ModelInfo *m;
unsigned chil1Step, chil2Step;
int *isScalerNode;
m = &modelSettings[division];
isScalerNode = m->isScalerNode[chain];
for (i=0; i<t->nIntNodes; i++)
{
p = t->intDownPass[i];
if (p->upDateCl == NO ) {
//p->upDateCl = YES;
/* flip to the new workspace */
FlipCondLikeSpace (m, chain, p->index);
}
/* update conditional likelihoods */
operations.destinationPartials = m->condLikeIndex[chain][p->index ];
operations.child1Partials = m->condLikeIndex[chain][p->left->index ];
operations.child1TransitionMatrix = m->tiProbsIndex [chain][p->left->index ];
operations.child2Partials = m->condLikeIndex[chain][p->right->index];
operations.child2TransitionMatrix = m->tiProbsIndex [chain][p->right->index];
/* All partials for tips are the same across omega catigoris, thus we are doing the following two if statments.*/
if(p->left->left== NULL)
chil1Step=0;
else
chil1Step=1;
if(p->right->left== NULL)
chil2Step=0;
else
chil2Step=1;
operations.destinationScaleRead = BEAGLE_OP_NONE;
if ( isScalerNode[p->index] == YES )
{
FlipNodeScalerSpace (m, chain, p->index);
operations.destinationScaleWrite = m->nodeScalerIndex[chain][p->index];
cumulativeScaleIndex = m->siteScalerIndex[chain];
}
else
{
operations.destinationScaleWrite = BEAGLE_OP_NONE;
cumulativeScaleIndex = BEAGLE_OP_NONE;
}
for (j=0; j<m->nCijkParts; j++)
{
beagleUpdatePartials(m->beagleInstance,
&operations,
1,
cumulativeScaleIndex);
operations.destinationPartials++;
operations.child1Partials+=chil1Step;
operations.child1TransitionMatrix++;
operations.child2Partials+=chil2Step;
operations.child2TransitionMatrix++;
if ( isScalerNode[p->index] == YES ){
operations.destinationScaleWrite++;
cumulativeScaleIndex++;
}
}
}
return NO_ERROR;
}
/*----------------------------------------------------------------
|
| TreeCondLikes_Beagle: This routine updates all conditional
| (partial) likelihoods of a beagle instance.
|
-----------------------------------------------------------------*/
int TreeCondLikes_Beagle (Tree *t, int division, int chain)
{
int i, j, destinationScaleRead, cumulativeScaleIndex;
BeagleOperation operations;
TreeNode *p;
ModelInfo *m;
unsigned chil1Step, chil2Step;
m = &modelSettings[division];
for (i=0; i<t->nIntNodes; i++)
{
p = t->intDownPass[i];
/* check if conditional likelihoods need updating */
if (p->upDateCl == YES)
{
/* remove old scalers */
if (m->scalersSet[chain][p->index] == YES && m->upDateAll == NO)
{
destinationScaleRead = m->nodeScalerIndex[chain][p->index];
cumulativeScaleIndex = m->siteScalerIndex[chain];
for (j=0; j<m->nCijkParts; j++)
{
beagleRemoveScaleFactors(m->beagleInstance,
&destinationScaleRead,
1,
cumulativeScaleIndex);
destinationScaleRead++;
cumulativeScaleIndex++;
}
}
/* flip to the new workspace */
FlipCondLikeSpace (m, chain, p->index);
FlipNodeScalerSpace (m, chain, p->index);
m->scalersSet[chain][p->index] = NO;
/* update conditional likelihoods */
operations.destinationPartials = m->condLikeIndex[chain][p->index ];
operations.child1Partials = m->condLikeIndex[chain][p->left->index ];
operations.child1TransitionMatrix = m->tiProbsIndex [chain][p->left->index ];
operations.child2Partials = m->condLikeIndex[chain][p->right->index];
operations.child2TransitionMatrix = m->tiProbsIndex [chain][p->right->index];
/* All partials for tips are the same across omega catigoris, thus we are doing the following two if statments.*/
if(p->left->left== NULL && p->left->right== NULL)
chil1Step=0;
else
chil1Step=1;
if(p->right->left== NULL && p->right->right== NULL)
chil2Step=0;
else
chil2Step=1;
if (p->scalerNode == YES)
{
m->scalersSet[chain][p->index] = YES;
operations.destinationScaleWrite = m->nodeScalerIndex[chain][p->index];
cumulativeScaleIndex = m->siteScalerIndex[chain];
}
else
{
operations.destinationScaleWrite = BEAGLE_OP_NONE;
cumulativeScaleIndex = BEAGLE_OP_NONE;
}
operations.destinationScaleRead = BEAGLE_OP_NONE;
for (j=0; j<m->nCijkParts; j++)
{
beagleUpdatePartials(m->beagleInstance,
&operations,
1,
cumulativeScaleIndex);
operations.destinationPartials++;
operations.child1Partials+=chil1Step;
operations.child1TransitionMatrix++;
operations.child2Partials+=chil2Step;
operations.child2TransitionMatrix++;
if (p->scalerNode == YES)
{
operations.destinationScaleWrite++;
cumulativeScaleIndex++;
}
}
}
} /* end of for */
return NO_ERROR;
}
/**---------------------------------------------------------------------------
|
| TreeLikelihood_Beagle: Accumulate the log likelihoods calculated by Beagle
| at the root.
|
---------------------------------------- -------------------------------------*/
int TreeLikelihood_Beagle (Tree *t, int division, int chain, MrBFlt *lnL, int whichSitePats)
{
int i, j, c = 0, nStates, hasPInvar, beagleReturn;
MrBFlt *swr, s01, s10, probOn, probOff, covBF[40], pInvar=0.0, *bs, freq, likeI, lnLikeI, diff, *omegaCatFreq;
CLFlt *clInvar=NULL, *nSitesOfPat;
double *nSitesOfPat_Beagle;
TreeNode *p;
ModelInfo *m;
double pUnobserved;
#if defined (MB_PRINT_DYNAMIC_RESCALE_FAIL_STAT)
static unsigned countBeagleDynamicFail=0;
static unsigned countALL=0;
#endif
/* find root node */
p = t->root->left;
/* find model settings and nStates, pInvar, invar cond likes */
m = &modelSettings[division];
nStates = m->numModelStates;
if (m->pInvar == NULL)
{
hasPInvar = NO;
}
else
{
hasPInvar = YES;
pInvar = *(GetParamVals (m->pInvar, chain, state[chain]));
clInvar = m->invCondLikes;
}
/* find base frequencies */
bs = GetParamSubVals (m->stateFreq, chain, state[chain]);
/* if covarion model, adjust base frequencies */
if (m->switchRates != NULL)
{
/* find the stationary frequencies */
swr = GetParamVals(m->switchRates, chain, state[chain]);
s01 = swr[0];
s10 = swr[1];
probOn = s01 / (s01 + s10);
probOff = 1.0 - probOn;
/* now adjust the base frequencies; on-state stored first in cond likes */
for (j=0; j<nStates/2; j++)
{
covBF[j] = bs[j] * probOn;
covBF[j+nStates/2] = bs[j] * probOff;
}
/* finally set bs pointer to adjusted values */
bs = covBF;
}
if (m->upDateCijk == YES) { /* TODO Really only need to check if state frequencies have changed */
/* set base frequencies in BEAGLE instance */
for (i=0; i<m->nCijkParts; i++)
beagleSetStateFrequencies(m->beagleInstance,
m->cijkIndex[chain] + i,
bs);
}
/* find category frequencies */
if (hasPInvar == NO)
freq = 1.0 / m->numGammaCats;
else
freq = (1.0 - pInvar) / m->numGammaCats;
/* TODO: cat weights only need to be set when they change */
/* set category frequencies in beagle instance */
if (m->numOmegaCats > 1)
{
omegaCatFreq = GetParamSubVals(m->omega, chain, state[chain]);
for (i=0; i<m->nCijkParts; i++)
{
for (j=0; j<m->numGammaCats; j++)
m->inWeights[j] = freq * omegaCatFreq[i];
beagleSetCategoryWeights(m->beagleInstance,
m->cijkIndex[chain] + i,
m->inWeights);
}
}
else if (hasPInvar == YES)
{
for (i=0; i<m->numGammaCats; i++)
m->inWeights[i] = freq;
beagleSetCategoryWeights(m->beagleInstance,
m->cijkIndex[chain],
m->inWeights);
}
/* find nSitesOfPat */
nSitesOfPat = numSitesOfPat + (whichSitePats*numCompressedChars) + m->compCharStart;
/* TODO: pattern weights only need to be set when they change */
/* set pattern weights in beagle instance if using dynamic reweighting */
if (chainParams.weightScheme[0] + chainParams.weightScheme[1] > ETA)
{
nSitesOfPat_Beagle = (double *) SafeMalloc (m->numChars * sizeof(double));
for (c=0; c<m->numChars; c++)
nSitesOfPat_Beagle[c] = numSitesOfPat[m->compCharStart + c];
beagleSetPatternWeights(m->beagleInstance,
nSitesOfPat_Beagle);
SafeFree ((void **)(&nSitesOfPat_Beagle));
}
/* find root log likelihoods and scalers */
for (i=0; i<m->nCijkParts; i++)
{
m->bufferIndices[i] = m->condLikeIndex[chain][p->index] + i;
m->eigenIndices[i] = m->cijkIndex[chain] + i;
m->cumulativeScaleIndices[i] = m->siteScalerIndex[chain] + i;
if (t->isRooted == NO)
{
m->childBufferIndices[i] = m->condLikeIndex [chain][p->anc->index];
m->childTiProbIndices[i] = m->tiProbsIndex [chain][p->index] + i;
}
}
/* reset lnL */
*lnL = 0.0;
/* get root log likelihoods */
if (t->isRooted == YES)
{
beagleReturn = beagleCalculateRootLogLikelihoods(m->beagleInstance,
m->bufferIndices,
m->eigenIndices,
m->eigenIndices,
m->cumulativeScaleIndices,
m->nCijkParts,
lnL);
}
else
{
beagleReturn = beagleCalculateEdgeLogLikelihoods(m->beagleInstance,
m->bufferIndices,
m->childBufferIndices,
m->childTiProbIndices,
NULL,
NULL,
m->eigenIndices,
m->eigenIndices,
m->cumulativeScaleIndices,
m->nCijkParts,
lnL,
NULL,
NULL);
}
#if defined (MB_PRINT_DYNAMIC_RESCALE_FAIL_STAT)
countALL++;
#endif
if( beagleReturn == BEAGLE_ERROR_FLOATING_POINT )
{
#if defined (MB_PRINT_DYNAMIC_RESCALE_FAIL_STAT)
countBeagleDynamicFail++;
printf("#####DEBUG INFO (it is not an error)############## countBeagleDynamicFail:%d countALL:%d\n",countBeagleDynamicFail,countALL);
#endif
return beagleReturn;
}
assert(beagleReturn == BEAGLE_SUCCESS);
m->succesCount[chain]++;
/* accumulate logs across sites */
if (hasPInvar == NO)
{
if( m->dataType == RESTRICTION )
{
beagleGetSiteLogLikelihoods(m->beagleInstance, m->logLikelihoods);
(*lnL) = 0.0;
pUnobserved = 0.0;
for (c=0; c<m->numDummyChars; c++)
{
pUnobserved += exp((double)m->logLikelihoods[c]);
}
/* correct for absent characters */
(*lnL) -= log( 1-pUnobserved ) * (m->numUncompressedChars);
for (; c<m->numChars; c++)
{
(*lnL) += m->logLikelihoods[c] * nSitesOfPat[c];
}
}
/* already done, just check for numerical errors */
assert ((*lnL) == (*lnL));
}
else
{
/* has invariable category */
beagleGetSiteLogLikelihoods(m->beagleInstance,
m->logLikelihoods);
(*lnL) = 0.0;
for (c=0; c<m->numChars; c++)
{
likeI = 0.0;
for (j=0; j<nStates; j++)
likeI += (*(clInvar++)) * bs[j];
if (likeI != 0.0)
{
lnLikeI = log(likeI * pInvar);
diff = lnLikeI - m->logLikelihoods[c];
}
else
diff = -1000.0;
if (diff < -200.0)
(*lnL) += m->logLikelihoods[c] * nSitesOfPat[c];
else if (diff > 200.0)
(*lnL) += lnLikeI * nSitesOfPat[c];
else
{
(*lnL) += (m->logLikelihoods[c] + log(1.0 + exp(diff))) * nSitesOfPat[c];
}
/* check for numerical errors */
assert((*lnL) == (*lnL));
}
}
return (NO_ERROR);
}
/*----------------------------------------------------------------
|
| TreeTiProbs_Beagle: This routine updates all transition
| probability matrices of a beagle instance.
|
-----------------------------------------------------------------*/
int TreeTiProbs_Beagle (Tree *t, int division, int chain)
{
int i, j, k, count;
MrBFlt correctionFactor, theRate, baseRate, *catRate, length;
TreeNode *p;
ModelInfo *m;
/* get model settings */
m = &modelSettings[division];
/* find the correction factor to make branch lengths
in terms of expected number of substitutions per character */
correctionFactor = 1.0;
if (m->dataType == DNA || m->dataType == RNA)
{
if (m->nucModelId == NUCMODEL_DOUBLET)
correctionFactor = 2.0;
else if (m->nucModelId == NUCMODEL_CODON)
correctionFactor = 3.0;
}
/* get rate multipliers (for gamma & partition specific rates) */
theRate = 1.0;
baseRate = GetRate (division, chain);
/* compensate for invariable sites if appropriate */
if (m->pInvar != NULL)
baseRate /= ( 1.0 - ( *GetParamVals(m->pInvar, chain, state[chain])));
/* get category rates for gamma */
if (m->shape == NULL)
catRate = &theRate;
else
catRate = GetParamSubVals (m->shape, chain, state[chain]);
/* get effective category rates */
for (k=0; k<m->numGammaCats; k++)
m->inRates[k] = baseRate * catRate[k] * correctionFactor;
/* TODO: only need to set category rates when they change */
/* set category rates */
beagleSetCategoryRates(m->beagleInstance, m->inRates);
/* get ti prob indices and branch lengths to update */
for (i=count=0; i<t->nNodes; i++)
{
p = t->allDownPass[i];
/* check if transition probs need updating */
if (p->upDateTi == YES)
{
/* flip transition probability */
FlipTiProbsSpace (m, chain, p->index);
/* find length */
if (m->cppEvents != NULL)
{
length = GetParamSubVals (m->cppEvents, chain, state[chain])[p->index];
}
else if (m->tk02BranchRates != NULL)
{
length = GetParamSubVals (m->tk02BranchRates, chain, state[chain])[p->index];
}
else if (m->igrBranchRates != NULL)
{
length = GetParamSubVals (m->igrBranchRates, chain, state[chain])[p->index];
}
else
length = p->length;
/* numerical errors might ensue if we allow very large or very small branch lengths, which might
occur in relaxed clock models; an elegant solution would be to substitute the stationary
probs and initial probs but for now we truncate lengths at small or large values */
if (length > BRLENS_MAX)
length = BRLENS_MAX;
else if (length < BRLENS_MIN)
length = BRLENS_MIN;
m->branchLengths[count] = length;
/* find index */
m->tiProbIndices[count] = m->tiProbsIndex[chain][p->index];
count++;
}
}
/* TODO: only need to update branches that have changed */
/* calculate transition probabilities */
if (count > 0) {
for (i=0; i<m->nCijkParts; i++)
{
beagleUpdateTransitionMatrices(m->beagleInstance,
m->cijkIndex[chain] + i,
m->tiProbIndices,
NULL,
NULL,
m->branchLengths,
count);
for (j=0; j<count; j++)
m->tiProbIndices[j]++;
}
}
/* return success */
return NO_ERROR;
}
#endif // BEAGLE_ENABLED
void BeagleNotLinked()
{
MrBayesPrint("%s BEAGLE library is not linked to this executable.\n", spacer);
}
void BeagleThreadsNotLinked()
{
MrBayesPrint("%s Pthreads library is not linked to this executable.\n", spacer);
}
|
