edlib.cpp 63.5 KB
Newer Older
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
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
#include "edlib.h"

#include <stdint.h>
#include <cstdlib>
#include <algorithm>
#include <vector>
#include <cstring>
#include <string>

using namespace std;

typedef uint64_t Word;
static const int WORD_SIZE = sizeof(Word) * 8; // Size of Word in bits
static const Word WORD_1 = (Word)1;
static const Word HIGH_BIT_MASK = WORD_1 << (WORD_SIZE - 1);  // 100..00
static const int MAX_UCHAR = 255;

// Data needed to find alignment.
struct AlignmentData {
    Word* Ps;
    Word* Ms;
    int* scores;
    int* firstBlocks;
    int* lastBlocks;

    AlignmentData(int maxNumBlocks, int targetLength) {
        // We build a complete table and mark first and last block for each column
        // (because algorithm is banded so only part of each columns is used).
        // TODO: do not build a whole table, but just enough blocks for each column.
         Ps     = new Word[maxNumBlocks * targetLength];
         Ms     = new Word[maxNumBlocks * targetLength];
         scores = new  int[maxNumBlocks * targetLength];
         firstBlocks = new int[targetLength];
         lastBlocks  = new int[targetLength];
    }

    ~AlignmentData() {
        delete[] Ps;
        delete[] Ms;
        delete[] scores;
        delete[] firstBlocks;
        delete[] lastBlocks;
    }
};

struct Block {
    Word P;  // Pvin
    Word M;  // Mvin
    int score; // score of last cell in block;

    Block() {}
    Block(Word P, Word M, int score) :P(P), M(M), score(score) {}
};


/**
 * Defines equality relation on alphabet characters.
 * By default each character is always equal only to itself, but you can also provide additional equalities.
 */
class EqualityDefinition {
private:
    bool matrix[MAX_UCHAR + 1][MAX_UCHAR + 1];
public:
    EqualityDefinition(const string& alphabet,
                       const EdlibEqualityPair* additionalEqualities = NULL,
                       const int additionalEqualitiesLength = 0) {
        for (int i = 0; i < (int) alphabet.size(); i++) {
            for (int j = 0; j < (int) alphabet.size(); j++) {
                matrix[i][j] = (i == j);
            }
        }
        if (additionalEqualities != NULL) {
            for (int i = 0; i < additionalEqualitiesLength; i++) {
                size_t firstTransformed = alphabet.find(additionalEqualities[i].first);
                size_t secondTransformed = alphabet.find(additionalEqualities[i].second);
                if (firstTransformed != string::npos && secondTransformed != string::npos) {
                    matrix[firstTransformed][secondTransformed] = matrix[secondTransformed][firstTransformed] = true;
                }
            }
        }
    }

    /**
     * @param a  Element from transformed sequence.
     * @param b  Element from transformed sequence.
     * @return True if a and b are defined as equal, false otherwise.
     */
    bool areEqual(unsigned char a, unsigned char b) const {
        return matrix[a][b];
    }
};

static int myersCalcEditDistanceSemiGlobal(const Word* Peq, int W, int maxNumBlocks,
                                           int queryLength,
                                           const unsigned char* target, int targetLength,
                                           int k, EdlibAlignMode mode,
                                           int* bestScore_, int** positions_, int* numPositions_);

static int myersCalcEditDistanceNW(const Word* Peq, int W, int maxNumBlocks,
                                   int queryLength,
                                   const unsigned char* target, int targetLength,
                                   int k, int* bestScore_,
                                   int* position_, bool findAlignment,
                                   AlignmentData** alignData, int targetStopPosition);


static int obtainAlignment(
        const unsigned char* query, const unsigned char* rQuery, int queryLength,
        const unsigned char* target, const unsigned char* rTarget, int targetLength,
        const EqualityDefinition& equalityDefinition, int alphabetLength, int bestScore,
        unsigned char** alignment, int* alignmentLength);

static int obtainAlignmentHirschberg(
        const unsigned char* query, const unsigned char* rQuery, int queryLength,
        const unsigned char* target, const unsigned char* rTarget, int targetLength,
        const EqualityDefinition& equalityDefinition, int alphabetLength, int bestScore,
        unsigned char** alignment, int* alignmentLength);

static int obtainAlignmentTraceback(int queryLength, int targetLength,
                                    int bestScore, const AlignmentData* alignData,
                                    unsigned char** alignment, int* alignmentLength);

static string transformSequences(const char* queryOriginal, int queryLength,
                                 const char* targetOriginal, int targetLength,
                                 unsigned char** queryTransformed,
                                 unsigned char** targetTransformed);

static inline int ceilDiv(int x, int y);

static inline unsigned char* createReverseCopy(const unsigned char* seq, int length);

static inline Word* buildPeq(const int alphabetLength,
                             const unsigned char* query,
                             const int queryLength,
                             const EqualityDefinition& equalityDefinition);


/**
 * Main edlib method.
 */
extern "C" EdlibAlignResult edlibAlign(const char* const queryOriginal, const int queryLength,
                                       const char* const targetOriginal, const int targetLength,
                                       const EdlibAlignConfig config) {
    EdlibAlignResult result;
    result.status = EDLIB_STATUS_OK;
    result.editDistance = -1;
    result.endLocations = result.startLocations = NULL;
    result.numLocations = 0;
    result.alignment = NULL;
    result.alignmentLength = 0;
    result.alphabetLength = 0;

    /*------------ TRANSFORM SEQUENCES AND RECOGNIZE ALPHABET -----------*/
    unsigned char* query, * target;
    string alphabet = transformSequences(queryOriginal, queryLength, targetOriginal, targetLength,
                                         &query, &target);
    result.alphabetLength = (int) alphabet.size();
    /*-------------------------------------------------------*/

    /*--------------------- INITIALIZATION ------------------*/
    int maxNumBlocks = ceilDiv(queryLength, WORD_SIZE); // bmax in Myers
    int W = maxNumBlocks * WORD_SIZE - queryLength; // number of redundant cells in last level blocks
    EqualityDefinition equalityDefinition(alphabet, config.additionalEqualities, config.additionalEqualitiesLength);
    Word* Peq = buildPeq((int) alphabet.size(), query, queryLength, equalityDefinition);
    /*-------------------------------------------------------*/

    /*------------------ MAIN CALCULATION -------------------*/
    // TODO: Store alignment data only after k is determined? That could make things faster.
    int positionNW; // Used only when mode is NW.
    AlignmentData* alignData = NULL;
    bool dynamicK = false;
    int k = config.k;
    if (k < 0) { // If valid k is not given, auto-adjust k until solution is found.
        dynamicK = true;
        k = WORD_SIZE; // Gives better results than smaller k.
    }

    do {
        if (config.mode == EDLIB_MODE_HW || config.mode == EDLIB_MODE_SHW) {
            myersCalcEditDistanceSemiGlobal(Peq, W, maxNumBlocks,
                                            queryLength, target, targetLength,
                                            k, config.mode, &(result.editDistance),
                                            &(result.endLocations), &(result.numLocations));
        } else {  // mode == EDLIB_MODE_NW
            myersCalcEditDistanceNW(Peq, W, maxNumBlocks,
                                    queryLength, target, targetLength,
                                    k, &(result.editDistance), &positionNW,
                                    false, &alignData, -1);
        }
        k *= 2;
    } while(dynamicK && result.editDistance == -1);

    if (result.editDistance >= 0) {  // If there is solution.
        // If NW mode, set end location explicitly.
        if (config.mode == EDLIB_MODE_NW) {
            result.endLocations = (int *) malloc(sizeof(int) * 1);
            result.endLocations[0] = targetLength - 1;
            result.numLocations = 1;
        }

        // Find starting locations.
        if (config.task == EDLIB_TASK_LOC || config.task == EDLIB_TASK_PATH) {
            result.startLocations = (int*) malloc(result.numLocations * sizeof(int));
            if (config.mode == EDLIB_MODE_HW) {  // If HW, I need to calculate start locations.
                const unsigned char* rTarget = createReverseCopy(target, targetLength);
                const unsigned char* rQuery  = createReverseCopy(query, queryLength);
                // Peq for reversed query.
                Word* rPeq = buildPeq((int) alphabet.size(), rQuery, queryLength, equalityDefinition);
                for (int i = 0; i < result.numLocations; i++) {
                    int endLocation = result.endLocations[i];
                    if (endLocation == -1) {
                        // NOTE: Sometimes one of optimal solutions is that query starts before target, like this:
                        //                       AAGG <- target
                        //                   CCTT     <- query
                        //   It will never be only optimal solution and it does not happen often, however it is
                        //   possible and in that case end location will be -1. What should we do with that?
                        //   Should we just skip reporting such end location, although it is a solution?
                        //   If we do report it, what is the start location? -4? -1? Nothing?
                        // TODO: Figure this out. This has to do in general with how we think about start
                        //   and end locations.
                        //   Also, we have alignment later relying on this locations to limit the space of it's
                        //   search -> how can it do it right if these locations are negative or incorrect?
                        result.startLocations[i] = 0;  // I put 0 for now, but it does not make much sense.
                    } else {
                        int bestScoreSHW, numPositionsSHW;
                        int* positionsSHW;
                        myersCalcEditDistanceSemiGlobal(
                                rPeq, W, maxNumBlocks,
                                queryLength, rTarget + targetLength - endLocation - 1, endLocation + 1,
                                result.editDistance, EDLIB_MODE_SHW,
                                &bestScoreSHW, &positionsSHW, &numPositionsSHW);
                        // Taking last location as start ensures that alignment will not start with insertions
                        // if it can start with mismatches instead.
                        result.startLocations[i] = endLocation - positionsSHW[numPositionsSHW - 1];
                        free(positionsSHW);
                    }
                }
                delete[] rTarget;
                delete[] rQuery;
                delete[] rPeq;
            } else {  // If mode is SHW or NW
                for (int i = 0; i < result.numLocations; i++) {
                    result.startLocations[i] = 0;
                }
            }
        }

        // Find alignment -> all comes down to finding alignment for NW.
        // Currently we return alignment only for first pair of locations.
        if (config.task == EDLIB_TASK_PATH) {
            int alnStartLocation = result.startLocations[0];
            int alnEndLocation = result.endLocations[0];
            const unsigned char* alnTarget = target + alnStartLocation;
            const int alnTargetLength = alnEndLocation - alnStartLocation + 1;
            const unsigned char* rAlnTarget = createReverseCopy(alnTarget, alnTargetLength);
            const unsigned char* rQuery  = createReverseCopy(query, queryLength);
            obtainAlignment(query, rQuery, queryLength,
                            alnTarget, rAlnTarget, alnTargetLength,
                            equalityDefinition, (int) alphabet.size(), result.editDistance,
                            &(result.alignment), &(result.alignmentLength));
            delete[] rAlnTarget;
            delete[] rQuery;
        }
    }
    /*-------------------------------------------------------*/

    //--- Free memory ---//
    delete[] Peq;
    free(query);
    free(target);
    if (alignData) delete alignData;
    //-------------------//

    return result;
}

extern "C" char* edlibAlignmentToCigar(const unsigned char* const alignment, const int alignmentLength,
                                       const EdlibCigarFormat cigarFormat) {
    if (cigarFormat != EDLIB_CIGAR_EXTENDED && cigarFormat != EDLIB_CIGAR_STANDARD) {
        return 0;
    }

    // Maps move code from alignment to char in cigar.
    //                        0    1    2    3
    char moveCodeToChar[] = {'=', 'I', 'D', 'X'};
    if (cigarFormat == EDLIB_CIGAR_STANDARD) {
        moveCodeToChar[0] = moveCodeToChar[3] = 'M';
    }

    vector<char>* cigar = new vector<char>();
    char lastMove = 0;  // Char of last move. 0 if there was no previous move.
    int numOfSameMoves = 0;
    for (int i = 0; i <= alignmentLength; i++) {
        // if new sequence of same moves started
        if (i == alignmentLength || (moveCodeToChar[alignment[i]] != lastMove && lastMove != 0)) {
            // Write number of moves to cigar string.
            int numDigits = 0;
            for (; numOfSameMoves; numOfSameMoves /= 10) {
                cigar->push_back('0' + numOfSameMoves % 10);
                numDigits++;
            }
            reverse(cigar->end() - numDigits, cigar->end());
            // Write code of move to cigar string.
            cigar->push_back(lastMove);
            // If not at the end, start new sequence of moves.
            if (i < alignmentLength) {
                // Check if alignment has valid values.
                if (alignment[i] > 3) {
                    delete cigar;
                    return 0;
                }
                numOfSameMoves = 0;
            }
        }
        if (i < alignmentLength) {
            lastMove = moveCodeToChar[alignment[i]];
            numOfSameMoves++;
        }
    }
    cigar->push_back(0);  // Null character termination.
    char* cigar_ = (char*) malloc(cigar->size() * sizeof(char));
    memcpy(cigar_, &(*cigar)[0], cigar->size() * sizeof(char));
    delete cigar;

    return cigar_;
}

/**
 * Build Peq table for given query and alphabet.
 * Peq is table of dimensions alphabetLength+1 x maxNumBlocks.
 * Bit i of Peq[s * maxNumBlocks + b] is 1 if i-th symbol from block b of query equals symbol s, otherwise it is 0.
 * NOTICE: free returned array with delete[]!
 */
static inline Word* buildPeq(const int alphabetLength,
                             const unsigned char* const query,
                             const int queryLength,
                             const EqualityDefinition& equalityDefinition) {
    int maxNumBlocks = ceilDiv(queryLength, WORD_SIZE);
    // table of dimensions alphabetLength+1 x maxNumBlocks. Last symbol is wildcard.
    Word* Peq = new Word[(alphabetLength + 1) * maxNumBlocks];

    // Build Peq (1 is match, 0 is mismatch). NOTE: last column is wildcard(symbol that matches anything) with just 1s
    for (unsigned char symbol = 0; symbol <= alphabetLength; symbol++) {
        for (int b = 0; b < maxNumBlocks; b++) {
            if (symbol < alphabetLength) {
                Peq[symbol * maxNumBlocks + b] = 0;
                for (int r = (b+1) * WORD_SIZE - 1; r >= b * WORD_SIZE; r--) {
                    Peq[symbol * maxNumBlocks + b] <<= 1;
                    // NOTE: We pretend like query is padded at the end with W wildcard symbols
                    if (r >= queryLength || equalityDefinition.areEqual(query[r], symbol))
                        Peq[symbol * maxNumBlocks + b] += 1;
                }
            } else { // Last symbol is wildcard, so it is all 1s
                Peq[symbol * maxNumBlocks + b] = (Word)-1;
            }
        }
    }

    return Peq;
}


/**
 * Returns new sequence that is reverse of given sequence.
 * Free returned array with delete[].
 */
static inline unsigned char* createReverseCopy(const unsigned char* const seq, const int length) {
    unsigned char* rSeq = new unsigned char[length];
    for (int i = 0; i < length; i++) {
        rSeq[i] = seq[length - i - 1];
    }
    return rSeq;
}

/**
 * Corresponds to Advance_Block function from Myers.
 * Calculates one word(block), which is part of a column.
 * Highest bit of word (one most to the left) is most bottom cell of block from column.
 * Pv[i] and Mv[i] define vin of cell[i]: vin = cell[i] - cell[i-1].
 * @param [in] Pv  Bitset, Pv[i] == 1 if vin is +1, otherwise Pv[i] == 0.
 * @param [in] Mv  Bitset, Mv[i] == 1 if vin is -1, otherwise Mv[i] == 0.
 * @param [in] Eq  Bitset, Eq[i] == 1 if match, 0 if mismatch.
 * @param [in] hin  Will be +1, 0 or -1.
 * @param [out] PvOut  Bitset, PvOut[i] == 1 if vout is +1, otherwise PvOut[i] == 0.
 * @param [out] MvOut  Bitset, MvOut[i] == 1 if vout is -1, otherwise MvOut[i] == 0.
 * @param [out] hout  Will be +1, 0 or -1.
 */
static inline int calculateBlock(Word Pv, Word Mv, Word Eq, const int hin,
                                 Word &PvOut, Word &MvOut) {
    // hin can be 1, -1 or 0.
    // 1  -> 00...01
    // 0  -> 00...00
    // -1 -> 11...11 (2-complement)

    Word hinIsNeg = (Word)(hin >> 2) & WORD_1; // 00...001 if hin is -1, 00...000 if 0 or 1

    Word Xv = Eq | Mv;
    // This is instruction below written using 'if': if (hin < 0) Eq |= (Word)1;
    Eq |= hinIsNeg;
    Word Xh = (((Eq & Pv) + Pv) ^ Pv) | Eq;

    Word Ph = Mv | ~(Xh | Pv);
    Word Mh = Pv & Xh;

    int hout = 0;
    // This is instruction below written using 'if': if (Ph & HIGH_BIT_MASK) hout = 1;
    hout = (Ph & HIGH_BIT_MASK) >> (WORD_SIZE - 1);
    // This is instruction below written using 'if': if (Mh & HIGH_BIT_MASK) hout = -1;
    hout -= (Mh & HIGH_BIT_MASK) >> (WORD_SIZE - 1);

    Ph <<= 1;
    Mh <<= 1;

    // This is instruction below written using 'if': if (hin < 0) Mh |= (Word)1;
    Mh |= hinIsNeg;
    // This is instruction below written using 'if': if (hin > 0) Ph |= (Word)1;
    Ph |= (Word)((hin + 1) >> 1);

    PvOut = Mh | ~(Xv | Ph);
    MvOut = Ph & Xv;

    return hout;
}

/**
 * Does ceiling division x / y.
 * Note: x and y must be non-negative and x + y must not overflow.
 */
static inline int ceilDiv(const int x, const int y) {
    return x % y ? x / y + 1 : x / y;
}

static inline int min(const int x, const int y) {
    return x < y ? x : y;
}

static inline int max(const int x, const int y) {
    return x > y ? x : y;
}


/**
 * @param [in] block
 * @return Values of cells in block, starting with bottom cell in block.
 */
static inline vector<int> getBlockCellValues(const Block block) {
    vector<int> scores(WORD_SIZE);
    int score = block.score;
    Word mask = HIGH_BIT_MASK;
    for (int i = 0; i < WORD_SIZE - 1; i++) {
        scores[i] = score;
        if (block.P & mask) score--;
        if (block.M & mask) score++;
        mask >>= 1;
    }
    scores[WORD_SIZE - 1] = score;
    return scores;
}

/**
 * Writes values of cells in block into given array, starting with first/top cell.
 * @param [in] block
 * @param [out] dest  Array into which cell values are written. Must have size of at least WORD_SIZE.
 */
static inline void readBlock(const Block block, int* const dest) {
    int score = block.score;
    Word mask = HIGH_BIT_MASK;
    for (int i = 0; i < WORD_SIZE - 1; i++) {
        dest[WORD_SIZE - 1 - i] = score;
        if (block.P & mask) score--;
        if (block.M & mask) score++;
        mask >>= 1;
    }
    dest[0] = score;
}

/**
 * Writes values of cells in block into given array, starting with last/bottom cell.
 * @param [in] block
 * @param [out] dest  Array into which cell values are written. Must have size of at least WORD_SIZE.
 */
static inline void readBlockReverse(const Block block, int* const dest) {
    int score = block.score;
    Word mask = HIGH_BIT_MASK;
    for (int i = 0; i < WORD_SIZE - 1; i++) {
        dest[i] = score;
        if (block.P & mask) score--;
        if (block.M & mask) score++;
        mask >>= 1;
    }
    dest[WORD_SIZE - 1] = score;
}

/**
 * @param [in] block
 * @param [in] k
 * @return True if all cells in block have value larger than k, otherwise false.
 */
static inline bool allBlockCellsLarger(const Block block, const int k) {
    vector<int> scores = getBlockCellValues(block);
    for (int i = 0; i < WORD_SIZE; i++) {
        if (scores[i] <= k) return false;
    }
    return true;
}


/**
 * Uses Myers' bit-vector algorithm to find edit distance for one of semi-global alignment methods.
 * @param [in] Peq  Query profile.
 * @param [in] W  Size of padding in last block.
 *                TODO: Calculate this directly from query, instead of passing it.
 * @param [in] maxNumBlocks  Number of blocks needed to cover the whole query.
 *                           TODO: Calculate this directly from query, instead of passing it.
 * @param [in] queryLength
 * @param [in] target
 * @param [in] targetLength
 * @param [in] k
 * @param [in] mode  EDLIB_MODE_HW or EDLIB_MODE_SHW
 * @param [out] bestScore_  Edit distance.
 * @param [out] positions_  Array of 0-indexed positions in target at which best score was found.
                            Make sure to free this array with free().
 * @param [out] numPositions_  Number of positions in the positions_ array.
 * @return Status.
 */
static int myersCalcEditDistanceSemiGlobal(
        const Word* const Peq, const int W, const int maxNumBlocks,
        const int queryLength,
        const unsigned char* const target, const int targetLength,
        int k, const EdlibAlignMode mode,
        int* const bestScore_, int** const positions_, int* const numPositions_) {
    *positions_ = NULL;
    *numPositions_ = 0;

    // firstBlock is 0-based index of first block in Ukkonen band.
    // lastBlock is 0-based index of last block in Ukkonen band.
    int firstBlock = 0;
    int lastBlock = min(ceilDiv(k + 1, WORD_SIZE), maxNumBlocks) - 1; // y in Myers
    Block *bl; // Current block

    Block* blocks = new Block[maxNumBlocks];

    // For HW, solution will never be larger then queryLength.
    if (mode == EDLIB_MODE_HW) {
        k = min(queryLength, k);
    }

    // Each STRONG_REDUCE_NUM column is reduced in more expensive way.
    // This gives speed up of about 2 times for small k.
    const int STRONG_REDUCE_NUM = 2048;

    // Initialize P, M and score
    bl = blocks;
    for (int b = 0; b <= lastBlock; b++) {
        bl->score = (b + 1) * WORD_SIZE;
        bl->P = (Word)-1; // All 1s
        bl->M = (Word)0;
        bl++;
    }

    int bestScore = -1;
    vector<int> positions; // TODO: Maybe put this on heap?
    const int startHout = mode == EDLIB_MODE_HW ? 0 : 1; // If 0 then gap before query is not penalized;
    const unsigned char* targetChar = target;
    for (int c = 0; c < targetLength; c++) { // for each column
        const Word* Peq_c = Peq + (*targetChar) * maxNumBlocks;

        //----------------------- Calculate column -------------------------//
        int hout = startHout;
        bl = blocks + firstBlock;
        Peq_c += firstBlock;
        for (int b = firstBlock; b <= lastBlock; b++) {
            hout = calculateBlock(bl->P, bl->M, *Peq_c, hout, bl->P, bl->M);
            bl->score += hout;
            bl++; Peq_c++;
        }
        bl--; Peq_c--;
        //------------------------------------------------------------------//

        //---------- Adjust number of blocks according to Ukkonen ----------//
        if ((lastBlock < maxNumBlocks - 1) && (bl->score - hout <= k) // bl is pointing to last block
            && ((*(Peq_c + 1) & WORD_1) || hout < 0)) { // Peq_c is pointing to last block
            // If score of left block is not too big, calculate one more block
            lastBlock++; bl++; Peq_c++;
            bl->P = (Word)-1; // All 1s
            bl->M = (Word)0;
            bl->score = (bl - 1)->score - hout + WORD_SIZE + calculateBlock(bl->P, bl->M, *Peq_c, hout, bl->P, bl->M);
        } else {
            while (lastBlock >= firstBlock && bl->score >= k + WORD_SIZE) {
                lastBlock--; bl--; Peq_c--;
            }
        }

        // Every some columns, do some expensive but also more efficient block reducing.
        // This is important!
        //
        // Reduce the band by decreasing last block if possible.
        if (c % STRONG_REDUCE_NUM == 0) {
            while (lastBlock >= 0 && lastBlock >= firstBlock && allBlockCellsLarger(*bl, k)) {
                lastBlock--; bl--; Peq_c--;
            }
        }
        // For HW, even if all cells are > k, there still may be solution in next
        // column because starting conditions at upper boundary are 0.
        // That means that first block is always candidate for solution,
        // and we can never end calculation before last column.
        if (mode == EDLIB_MODE_HW && lastBlock == -1) {
            lastBlock++; bl++; Peq_c++;
        }

        // Reduce band by increasing first block if possible. Not applicable to HW.
        if (mode != EDLIB_MODE_HW) {
            while (firstBlock <= lastBlock && blocks[firstBlock].score >= k + WORD_SIZE) {
                firstBlock++;
            }
            if (c % STRONG_REDUCE_NUM == 0) { // Do strong reduction every some blocks
                while (firstBlock <= lastBlock && allBlockCellsLarger(blocks[firstBlock], k)) {
                    firstBlock++;
                }
            }
        }

        // If band stops to exist finish
        if (lastBlock < firstBlock) {
            *bestScore_ = bestScore;
            if (bestScore != -1) {
                *positions_ = (int *) malloc(sizeof(int) * (int) positions.size());
                *numPositions_ = (int) positions.size();
                copy(positions.begin(), positions.end(), *positions_);
            }
            delete[] blocks;
            return EDLIB_STATUS_OK;
        }
        //------------------------------------------------------------------//

        //------------------------- Update best score ----------------------//
        if (lastBlock == maxNumBlocks - 1) {
            int colScore = bl->score;
            if (colScore <= k) { // Scores > k dont have correct values (so we cannot use them), but are certainly > k.
                // NOTE: Score that I find in column c is actually score from column c-W
                if (bestScore == -1 || colScore <= bestScore) {
                    if (colScore != bestScore) {
                        positions.clear();
                        bestScore = colScore;
                        // Change k so we will look only for equal or better
                        // scores then the best found so far.
                        k = bestScore;
                    }
                    positions.push_back(c - W);
                }
            }
        }
        //------------------------------------------------------------------//

        targetChar++;
    }


    // Obtain results for last W columns from last column.
    if (lastBlock == maxNumBlocks - 1) {
        vector<int> blockScores = getBlockCellValues(*bl);
        for (int i = 0; i < W; i++) {
            int colScore = blockScores[i + 1];
            if (colScore <= k && (bestScore == -1 || colScore <= bestScore)) {
                if (colScore != bestScore) {
                    positions.clear();
                    k = bestScore = colScore;
                }
                positions.push_back(targetLength - W + i);
            }
        }
    }

    *bestScore_ = bestScore;
    if (bestScore != -1) {
        *positions_ = (int *) malloc(sizeof(int) * (int) positions.size());
        *numPositions_ = (int) positions.size();
        copy(positions.begin(), positions.end(), *positions_);
    }

    delete[] blocks;
    return EDLIB_STATUS_OK;
}


/**
 * Uses Myers' bit-vector algorithm to find edit distance for global(NW) alignment method.
 * @param [in] Peq  Query profile.
 * @param [in] W  Size of padding in last block.
 *                TODO: Calculate this directly from query, instead of passing it.
 * @param [in] maxNumBlocks  Number of blocks needed to cover the whole query.
 *                           TODO: Calculate this directly from query, instead of passing it.
 * @param [in] queryLength
 * @param [in] target
 * @param [in] targetLength
 * @param [in] k
 * @param [out] bestScore_  Edit distance.
 * @param [out] position_  0-indexed position in target at which best score was found.
 * @param [in] findAlignment  If true, whole matrix is remembered and alignment data is returned.
 *                            Quadratic amount of memory is consumed.
 * @param [out] alignData  Data needed for alignment traceback (for reconstruction of alignment).
 *                         Set only if findAlignment is set to true, otherwise it is NULL.
 *                         Make sure to free this array using delete[].
 * @param [out] targetStopPosition  If set to -1, whole calculation is performed normally, as expected.
 *         If set to p, calculation is performed up to position p in target (inclusive)
 *         and column p is returned as the only column in alignData.
 * @return Status.
 */
static int myersCalcEditDistanceNW(const Word* const Peq, const int W, const int maxNumBlocks,
                                   const int queryLength,
                                   const unsigned char* const target, const int targetLength,
                                   int k, int* const bestScore_,
                                   int* const position_, const bool findAlignment,
                                   AlignmentData** const alignData, const int targetStopPosition) {
    if (targetStopPosition > -1 && findAlignment) {
        // They can not be both set at the same time!
        return EDLIB_STATUS_ERROR;
    }

    // Each STRONG_REDUCE_NUM column is reduced in more expensive way.
    const int STRONG_REDUCE_NUM = 2048; // TODO: Choose this number dinamically (based on query and target lengths?), so it does not affect speed of computation

    if (k < abs(targetLength - queryLength)) {
        *bestScore_ = *position_ = -1;
        return EDLIB_STATUS_OK;
    }

    k = min(k, max(queryLength, targetLength));  // Upper bound for k

    // firstBlock is 0-based index of first block in Ukkonen band.
    // lastBlock is 0-based index of last block in Ukkonen band.
    int firstBlock = 0;
    // This is optimal now, by my formula.
    int lastBlock = min(maxNumBlocks, ceilDiv(min(k, (k + queryLength - targetLength) / 2) + 1, WORD_SIZE)) - 1;
    Block* bl; // Current block

    Block* blocks = new Block[maxNumBlocks];

    // Initialize P, M and score
    bl = blocks;
    for (int b = 0; b <= lastBlock; b++) {
        bl->score = (b + 1) * WORD_SIZE;
        bl->P = (Word)-1; // All 1s
        bl->M = (Word)0;
        bl++;
    }

    // If we want to find alignment, we have to store needed data.
    if (findAlignment)
        *alignData = new AlignmentData(maxNumBlocks, targetLength);
    else if (targetStopPosition > -1)
        *alignData = new AlignmentData(maxNumBlocks, 1);
    else
        *alignData = NULL;

    const unsigned char* targetChar = target;
    for (int c = 0; c < targetLength; c++) { // for each column
        const Word* Peq_c = Peq + *targetChar * maxNumBlocks;

        //----------------------- Calculate column -------------------------//
        int hout = 1;
        bl = blocks + firstBlock;
        for (int b = firstBlock; b <= lastBlock; b++) {
            hout = calculateBlock(bl->P, bl->M, Peq_c[b], hout, bl->P, bl->M);
            bl->score += hout;
            bl++;
        }
        bl--;
        //------------------------------------------------------------------//
        // bl now points to last block

        // Update k. I do it only on end of column because it would slow calculation too much otherwise.
        // NOTICE: I add W when in last block because it is actually result from W cells to the left and W cells up.
        k = min(k, bl->score
                + max(targetLength - c - 1, queryLength - ((1 + lastBlock) * WORD_SIZE - 1) - 1)
                + (lastBlock == maxNumBlocks - 1 ? W : 0));

        //---------- Adjust number of blocks according to Ukkonen ----------//
        //--- Adjust last block ---//
        // If block is not beneath band, calculate next block. Only next because others are certainly beneath band.
        if (lastBlock + 1 < maxNumBlocks
            && !(//score[lastBlock] >= k + WORD_SIZE ||  // NOTICE: this condition could be satisfied if above block also!
                 ((lastBlock + 1) * WORD_SIZE - 1
                  > k - bl->score + 2 * WORD_SIZE - 2 - targetLength + c + queryLength))) {
            lastBlock++; bl++;
            bl->P = (Word)-1; // All 1s
            bl->M = (Word)0;
            int newHout = calculateBlock(bl->P, bl->M, Peq_c[lastBlock], hout, bl->P, bl->M);
            bl->score = (bl - 1)->score - hout + WORD_SIZE + newHout;
            hout = newHout;
        }

        // While block is out of band, move one block up.
        // NOTE: Condition used here is more loose than the one from the article, since I simplified the max() part of it.
        // I could consider adding that max part, for optimal performance.
        while (lastBlock >= firstBlock
               && (bl->score >= k + WORD_SIZE
                   || ((lastBlock + 1) * WORD_SIZE - 1 >
                       // TODO: Does not work if do not put +1! Why???
                       k - bl->score + 2 * WORD_SIZE - 2 - targetLength + c + queryLength + 1))) {
            lastBlock--; bl--;
        }
        //-------------------------//

        //--- Adjust first block ---//
        // While outside of band, advance block
        while (firstBlock <= lastBlock
               && (blocks[firstBlock].score >= k + WORD_SIZE
                   || ((firstBlock + 1) * WORD_SIZE - 1 <
                       blocks[firstBlock].score - k - targetLength + queryLength + c))) {
            firstBlock++;
        }
        //--------------------------/


        // TODO: consider if this part is useful, it does not seem to help much
        if (c % STRONG_REDUCE_NUM == 0) { // Every some columns do more expensive but more efficient reduction
            while (lastBlock >= firstBlock) {
                // If all cells outside of band, remove block
                vector<int> scores = getBlockCellValues(*bl);
                int numCells = lastBlock == maxNumBlocks - 1 ? WORD_SIZE - W : WORD_SIZE;
                int r = lastBlock * WORD_SIZE + numCells - 1;
                bool reduce = true;
                for (int i = WORD_SIZE - numCells; i < WORD_SIZE; i++) {
                    // TODO: Does not work if do not put +1! Why???
                    if (scores[i] <= k && r <= k - scores[i] - targetLength + c + queryLength + 1) {
                        reduce = false;
                        break;
                    }
                    r--;
                }
                if (!reduce) break;
                lastBlock--; bl--;
            }

            while (firstBlock <= lastBlock) {
                // If all cells outside of band, remove block
                vector<int> scores = getBlockCellValues(blocks[firstBlock]);
                int numCells = firstBlock == maxNumBlocks - 1 ? WORD_SIZE - W : WORD_SIZE;
                int r = firstBlock * WORD_SIZE + numCells - 1;
                bool reduce = true;
                for (int i = WORD_SIZE - numCells; i < WORD_SIZE; i++) {
                    if (scores[i] <= k && r >= scores[i] - k - targetLength + c + queryLength) {
                        reduce = false;
                        break;
                    }
                    r--;
                }
                if (!reduce) break;
                firstBlock++;
            }
        }


        // If band stops to exist finish
        if (lastBlock < firstBlock) {
            *bestScore_ = *position_ = -1;
            delete[] blocks;
            return EDLIB_STATUS_OK;
        }
        //------------------------------------------------------------------//


        //---- Save column so it can be used for reconstruction ----//
        if (findAlignment && c < targetLength) {
            bl = blocks + firstBlock;
            for (int b = firstBlock; b <= lastBlock; b++) {
                (*alignData)->Ps[maxNumBlocks * c + b] = bl->P;
                (*alignData)->Ms[maxNumBlocks * c + b] = bl->M;
                (*alignData)->scores[maxNumBlocks * c + b] = bl->score;
                (*alignData)->firstBlocks[c] = firstBlock;
                (*alignData)->lastBlocks[c] = lastBlock;
                bl++;
            }
        }
        //----------------------------------------------------------//
        //---- If this is stop column, save it and finish ----//
        if (c == targetStopPosition) {
            for (int b = firstBlock; b <= lastBlock; b++) {
                (*alignData)->Ps[b] = (blocks + b)->P;
                (*alignData)->Ms[b] = (blocks + b)->M;
                (*alignData)->scores[b] = (blocks + b)->score;
                (*alignData)->firstBlocks[0] = firstBlock;
                (*alignData)->lastBlocks[0] = lastBlock;
            }
            *bestScore_ = -1;
            *position_ = targetStopPosition;
            delete[] blocks;
            return EDLIB_STATUS_OK;
        }
        //----------------------------------------------------//

        targetChar++;
    }

    if (lastBlock == maxNumBlocks - 1) { // If last block of last column was calculated
        // Obtain best score from block -> it is complicated because query is padded with W cells
        int bestScore = getBlockCellValues(blocks[lastBlock])[W];
        if (bestScore <= k) {
            *bestScore_ = bestScore;
            *position_ = targetLength - 1;
            delete[] blocks;
            return EDLIB_STATUS_OK;
        }
    }

    *bestScore_ = *position_ = -1;
    delete[] blocks;
    return EDLIB_STATUS_OK;
}


/**
 * Finds one possible alignment that gives optimal score by moving back through the dynamic programming matrix,
 * that is stored in alignData. Consumes large amount of memory: O(queryLength * targetLength).
 * @param [in] queryLength  Normal length, without W.
 * @param [in] targetLength  Normal length, without W.
 * @param [in] bestScore  Best score.
 * @param [in] alignData  Data obtained during finding best score that is useful for finding alignment.
 * @param [out] alignment  Alignment.
 * @param [out] alignmentLength  Length of alignment.
 * @return Status code.
 */
static int obtainAlignmentTraceback(const int queryLength, const int targetLength,
                                    const int bestScore, const AlignmentData* const alignData,
                                    unsigned char** const alignment, int* const alignmentLength) {
    const int maxNumBlocks = ceilDiv(queryLength, WORD_SIZE);
    const int W = maxNumBlocks * WORD_SIZE - queryLength;

    *alignment = (unsigned char*) malloc((queryLength + targetLength - 1) * sizeof(unsigned char));
    *alignmentLength = 0;
    int c = targetLength - 1; // index of column
    int b = maxNumBlocks - 1; // index of block in column
    int currScore = bestScore; // Score of current cell
    int lScore  = -1; // Score of left cell
    int uScore  = -1; // Score of upper cell
    int ulScore = -1; // Score of upper left cell
    Word currP = alignData->Ps[c * maxNumBlocks + b]; // P of current block
    Word currM = alignData->Ms[c * maxNumBlocks + b]; // M of current block
    // True if block to left exists and is in band
    bool thereIsLeftBlock = c > 0 && b >= alignData->firstBlocks[c-1] && b <= alignData->lastBlocks[c-1];
    // We set initial values of lP and lM to 0 only to avoid compiler warnings, they should not affect the
    // calculation as both lP and lM should be initialized at some moment later (but compiler can not
    // detect it since this initialization is guaranteed by "business" logic).
    Word lP = 0, lM = 0;
    if (thereIsLeftBlock) {
        lP = alignData->Ps[(c - 1) * maxNumBlocks + b]; // P of block to the left
        lM = alignData->Ms[(c - 1) * maxNumBlocks + b]; // M of block to the left
    }
    currP <<= W;
    currM <<= W;
    int blockPos = WORD_SIZE - W - 1; // 0 based index of current cell in blockPos

    // TODO(martin): refactor this whole piece of code. There are too many if-else statements,
    // it is too easy for a bug to hide and to hard to effectively cover all the edge-cases.
    // We need better separation of logic and responsibilities.
    while (true) {
        if (c == 0) {
            thereIsLeftBlock = true;
            lScore = b * WORD_SIZE + blockPos + 1;
            ulScore = lScore - 1;
        }

        // TODO: improvement: calculate only those cells that are needed,
        //       for example if I calculate upper cell and can move up,
        //       there is no need to calculate left and upper left cell
        //---------- Calculate scores ---------//
        if (lScore == -1 && thereIsLeftBlock) {
            lScore = alignData->scores[(c - 1) * maxNumBlocks + b]; // score of block to the left
            for (int i = 0; i < WORD_SIZE - blockPos - 1; i++) {
                if (lP & HIGH_BIT_MASK) lScore--;
                if (lM & HIGH_BIT_MASK) lScore++;
                lP <<= 1;
                lM <<= 1;
            }
        }
        if (ulScore == -1) {
            if (lScore != -1) {
                ulScore = lScore;
                if (lP & HIGH_BIT_MASK) ulScore--;
                if (lM & HIGH_BIT_MASK) ulScore++;
            }
            else if (c > 0 && b-1 >= alignData->firstBlocks[c-1] && b-1 <= alignData->lastBlocks[c-1]) {
                // This is the case when upper left cell is last cell in block,
                // and block to left is not in band so lScore is -1.
                ulScore = alignData->scores[(c - 1) * maxNumBlocks + b - 1];
            }
        }
        if (uScore == -1) {
            uScore = currScore;
            if (currP & HIGH_BIT_MASK) uScore--;
            if (currM & HIGH_BIT_MASK) uScore++;
            currP <<= 1;
            currM <<= 1;
        }
        //-------------------------------------//

        // TODO: should I check if there is upper block?

        //-------------- Move --------------//
        // Move up - insertion to target - deletion from query
        if (uScore != -1 && uScore + 1 == currScore) {
            currScore = uScore;
            lScore = ulScore;
            uScore = ulScore = -1;
            if (blockPos == 0) { // If entering new (upper) block
                if (b == 0) { // If there are no cells above (only boundary cells)
                    (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_INSERT; // Move up
                    for (int i = 0; i < c + 1; i++) // Move left until end
                        (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_DELETE;
                    break;
                } else {
                    blockPos = WORD_SIZE - 1;
                    b--;
                    currP = alignData->Ps[c * maxNumBlocks + b];
                    currM = alignData->Ms[c * maxNumBlocks + b];
                    if (c > 0 && b >= alignData->firstBlocks[c-1] && b <= alignData->lastBlocks[c-1]) {
                        thereIsLeftBlock = true;
                        lP = alignData->Ps[(c - 1) * maxNumBlocks + b]; // TODO: improve this, too many operations
                        lM = alignData->Ms[(c - 1) * maxNumBlocks + b];
                    } else {
                        thereIsLeftBlock = false;
                        // TODO(martin): There may not be left block, but there can be left boundary - do we
                        // handle this correctly then? Are l and ul score set correctly? I should check that / refactor this.
                    }
                }
            } else {
                blockPos--;
                lP <<= 1;
                lM <<= 1;
            }
            // Mark move
            (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_INSERT;
        }
        // Move left - deletion from target - insertion to query
        else if (lScore != -1 && lScore + 1 == currScore) {
            currScore = lScore;
            uScore = ulScore;
            lScore = ulScore = -1;
            c--;
            if (c == -1) { // If there are no cells to the left (only boundary cells)
                (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_DELETE; // Move left
                int numUp = b * WORD_SIZE + blockPos + 1;
                for (int i = 0; i < numUp; i++) // Move up until end
                    (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_INSERT;
                break;
            }
            currP = lP;
            currM = lM;
            if (c > 0 && b >= alignData->firstBlocks[c-1] && b <= alignData->lastBlocks[c-1]) {
                thereIsLeftBlock = true;
                lP = alignData->Ps[(c - 1) * maxNumBlocks + b];
                lM = alignData->Ms[(c - 1) * maxNumBlocks + b];
            } else {
                if (c == 0) { // If there are no cells to the left (only boundary cells)
                    thereIsLeftBlock = true;
                    lScore = b * WORD_SIZE + blockPos + 1;
                    ulScore = lScore - 1;
                } else {
                    thereIsLeftBlock = false;
                }
            }
            // Mark move
            (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_DELETE;
        }
        // Move up left - (mis)match
        else if (ulScore != -1) {
            unsigned char moveCode = ulScore == currScore ? EDLIB_EDOP_MATCH : EDLIB_EDOP_MISMATCH;
            currScore = ulScore;
            uScore = lScore = ulScore = -1;
            c--;
            if (c == -1) { // If there are no cells to the left (only boundary cells)
                (*alignment)[(*alignmentLength)++] = moveCode; // Move left
                int numUp = b * WORD_SIZE + blockPos;
                for (int i = 0; i < numUp; i++) // Move up until end
                    (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_INSERT;
                break;
            }
            if (blockPos == 0) { // If entering upper left block
                if (b == 0) { // If there are no more cells above (only boundary cells)
                    (*alignment)[(*alignmentLength)++] = moveCode; // Move up left
                    for (int i = 0; i < c + 1; i++) // Move left until end
                        (*alignment)[(*alignmentLength)++] = EDLIB_EDOP_DELETE;
                    break;
                }
                blockPos = WORD_SIZE - 1;
                b--;
                currP = alignData->Ps[c * maxNumBlocks + b];
                currM = alignData->Ms[c * maxNumBlocks + b];
            } else { // If entering left block
                blockPos--;
                currP = lP;
                currM = lM;
                currP <<= 1;
                currM <<= 1;
            }
            // Set new left block
            if (c > 0 && b >= alignData->firstBlocks[c-1] && b <= alignData->lastBlocks[c-1]) {
                thereIsLeftBlock = true;
                lP = alignData->Ps[(c - 1) * maxNumBlocks + b];
                lM = alignData->Ms[(c - 1) * maxNumBlocks + b];
            } else {
                if (c == 0) { // If there are no cells to the left (only boundary cells)
                    thereIsLeftBlock = true;
                    lScore = b * WORD_SIZE + blockPos + 1;
                    ulScore = lScore - 1;
                } else {
                    thereIsLeftBlock = false;
                }
            }
            // Mark move
            (*alignment)[(*alignmentLength)++] = moveCode;
        } else {
            // Reached end - finished!
            break;
        }
        //----------------------------------//
    }

    *alignment = (unsigned char*) realloc(*alignment, (*alignmentLength) * sizeof(unsigned char));
    reverse(*alignment, *alignment + (*alignmentLength));
    return EDLIB_STATUS_OK;
}


/**
 * Finds one possible alignment that gives optimal score (bestScore).
 * It will split problem into smaller problems using Hirschberg's algorithm and when they are small enough,
 * it will solve them using traceback algorithm.
 * @param [in] query
 * @param [in] rQuery  Reversed query.
 * @param [in] queryLength
 * @param [in] target
 * @param [in] rTarget  Reversed target.
 * @param [in] targetLength
 * @param [in] equalityDefinition
 * @param [in] alphabetLength
 * @param [in] bestScore  Best(optimal) score.
 * @param [out] alignment  Sequence of edit operations that make target equal to query.
 * @param [out] alignmentLength  Length of alignment.
 * @return Status code.
 */
static int obtainAlignment(
        const unsigned char* const query, const unsigned char* const rQuery, const int queryLength,
        const unsigned char* const target, const unsigned char* const rTarget, const int targetLength,
        const EqualityDefinition& equalityDefinition, const int alphabetLength, const int bestScore,
        unsigned char** const alignment, int* const alignmentLength) {

    // Handle special case when one of sequences has length of 0.
    if (queryLength == 0 || targetLength == 0) {
        *alignmentLength = targetLength + queryLength;
        *alignment = (unsigned char*) malloc((*alignmentLength) * sizeof(unsigned char));
        for (int i = 0; i < *alignmentLength; i++) {
            (*alignment)[i] = queryLength == 0 ? EDLIB_EDOP_DELETE : EDLIB_EDOP_INSERT;
        }
        return EDLIB_STATUS_OK;
    }

    const int maxNumBlocks = ceilDiv(queryLength, WORD_SIZE);
    const int W = maxNumBlocks * WORD_SIZE - queryLength;
    int statusCode;

    // TODO: think about reducing number of memory allocations in alignment functions, probably
    // by sharing some memory that is allocated only once. That refers to: Peq, columns in Hirschberg,
    // and it could also be done for alignments - we could have one big array for alignment that would be
    // sparsely populated by each of steps in recursion, and at the end we would just consolidate those results.

    // If estimated memory consumption for traceback algorithm is smaller than 1MB use it,
    // otherwise use Hirschberg's algorithm. By running few tests I choose boundary of 1MB as optimal.
    long long alignmentDataSize = (long long) (2 * sizeof(Word) + sizeof(int)) * maxNumBlocks * targetLength
        + (long long) 2 * sizeof(int) * targetLength;
    if (alignmentDataSize < 1024 * 1024) {
        int score_, endLocation_;  // Used only to call function.
        AlignmentData* alignData = NULL;
        Word* Peq = buildPeq(alphabetLength, query, queryLength, equalityDefinition);
        myersCalcEditDistanceNW(Peq, W, maxNumBlocks,
                                queryLength,
                                target, targetLength,
                                bestScore,
                                &score_, &endLocation_, true, &alignData, -1);
        //assert(score_ == bestScore);
        //assert(endLocation_ == targetLength - 1);

        statusCode = obtainAlignmentTraceback(queryLength, targetLength,
                                              bestScore, alignData, alignment, alignmentLength);
        delete alignData;
        delete[] Peq;
    } else {
        statusCode = obtainAlignmentHirschberg(query, rQuery, queryLength,
                                               target, rTarget, targetLength,
                                               equalityDefinition, alphabetLength, bestScore,
                                               alignment, alignmentLength);
    }
    return statusCode;
}


/**
 * Finds one possible alignment that gives optimal score (bestScore).
 * Uses Hirschberg's algorithm to split problem into two sub-problems, solve them and combine them together.
 * @param [in] query
 * @param [in] rQuery  Reversed query.
 * @param [in] queryLength
 * @param [in] target
 * @param [in] rTarget  Reversed target.
 * @param [in] targetLength
 * @param [in] alphabetLength
 * @param [in] bestScore  Best(optimal) score.
 * @param [out] alignment  Sequence of edit operations that make target equal to query.
 * @param [out] alignmentLength  Length of alignment.
 * @return Status code.
 */
static int obtainAlignmentHirschberg(
        const unsigned char* const query, const unsigned char* const rQuery, const int queryLength,
        const unsigned char* const target, const unsigned char* const rTarget, const int targetLength,
        const EqualityDefinition& equalityDefinition, const int alphabetLength, const int bestScore,
        unsigned char** const alignment, int* const alignmentLength) {

    const int maxNumBlocks = ceilDiv(queryLength, WORD_SIZE);
    const int W = maxNumBlocks * WORD_SIZE - queryLength;

    Word* Peq = buildPeq(alphabetLength, query, queryLength, equalityDefinition);
    Word* rPeq = buildPeq(alphabetLength, rQuery, queryLength, equalityDefinition);

    // Used only to call functions.
    int score_, endLocation_;

    // Divide dynamic matrix into two halfs, left and right.
    const int leftHalfWidth = targetLength / 2;
    const int rightHalfWidth = targetLength - leftHalfWidth;

    // Calculate left half.
    AlignmentData* alignDataLeftHalf = NULL;
    int leftHalfCalcStatus = myersCalcEditDistanceNW(
            Peq, W, maxNumBlocks, queryLength, target, targetLength, bestScore,
            &score_, &endLocation_, false, &alignDataLeftHalf, leftHalfWidth - 1);

    // Calculate right half.
    AlignmentData* alignDataRightHalf = NULL;
    int rightHalfCalcStatus = myersCalcEditDistanceNW(
            rPeq, W, maxNumBlocks, queryLength, rTarget, targetLength, bestScore,
            &score_, &endLocation_, false, &alignDataRightHalf, rightHalfWidth - 1);

    delete[] Peq;
    delete[] rPeq;

    if (leftHalfCalcStatus == EDLIB_STATUS_ERROR || rightHalfCalcStatus == EDLIB_STATUS_ERROR) {
        if (alignDataLeftHalf) delete alignDataLeftHalf;
        if (alignDataRightHalf) delete alignDataRightHalf;
        return EDLIB_STATUS_ERROR;
    }

    // Unwrap the left half.
    int firstBlockIdxLeft = alignDataLeftHalf->firstBlocks[0];
    int lastBlockIdxLeft = alignDataLeftHalf->lastBlocks[0];
    // TODO: avoid this allocation by using some shared array?
    // scoresLeft contains scores from left column, starting with scoresLeftStartIdx row (query index)
    // and ending with scoresLeftEndIdx row (0-indexed).
    int scoresLeftLength = (lastBlockIdxLeft - firstBlockIdxLeft + 1) * WORD_SIZE;
    int* scoresLeft = new int[scoresLeftLength];
    for (int blockIdx = firstBlockIdxLeft; blockIdx <= lastBlockIdxLeft; blockIdx++) {
        Block block(alignDataLeftHalf->Ps[blockIdx], alignDataLeftHalf->Ms[blockIdx],
                    alignDataLeftHalf->scores[blockIdx]);
        readBlock(block, scoresLeft + (blockIdx - firstBlockIdxLeft) * WORD_SIZE);
    }
    int scoresLeftStartIdx = firstBlockIdxLeft * WORD_SIZE;
    // If last block contains padding, shorten the length of scores for the length of padding.
    if (lastBlockIdxLeft == maxNumBlocks - 1) {
        scoresLeftLength -= W;
    }

    // Unwrap the right half (I also reverse it while unwraping).
    int firstBlockIdxRight = alignDataRightHalf->firstBlocks[0];
    int lastBlockIdxRight = alignDataRightHalf->lastBlocks[0];
    int scoresRightLength = (lastBlockIdxRight - firstBlockIdxRight + 1) * WORD_SIZE;
    int* scoresRight = new int[scoresRightLength];
    int* scoresRightOriginalStart = scoresRight;
    for (int blockIdx = firstBlockIdxRight; blockIdx <= lastBlockIdxRight; blockIdx++) {
        Block block(alignDataRightHalf->Ps[blockIdx], alignDataRightHalf->Ms[blockIdx],
                    alignDataRightHalf->scores[blockIdx]);
        readBlockReverse(block, scoresRight + (lastBlockIdxRight - blockIdx) * WORD_SIZE);
    }
    int scoresRightStartIdx = queryLength - (lastBlockIdxRight + 1) * WORD_SIZE;
    // If there is padding at the beginning of scoresRight (that can happen because of reversing that we do),
    // move pointer forward to remove the padding (that is why we remember originalStart).
    if (scoresRightStartIdx < 0) {
        //assert(scoresRightStartIdx == -1 * W);
        scoresRight += W;
        scoresRightStartIdx += W;
        scoresRightLength -= W;
    }

    delete alignDataLeftHalf;
    delete alignDataRightHalf;

    //--------------------- Find the best move ----------------//
    // Find the query/row index of cell in left column which together with its lower right neighbour
    // from right column gives the best score (when summed). We also have to consider boundary cells
    // (those cells at -1 indexes).
    //  x|
    //  -+-
    //   |x
    int queryIdxLeftStart = max(scoresLeftStartIdx, scoresRightStartIdx - 1);
    int queryIdxLeftEnd = min(scoresLeftStartIdx + scoresLeftLength - 1,
                          scoresRightStartIdx + scoresRightLength - 2);
    int leftScore, rightScore;
    int queryIdxLeftAlignment;  // Query/row index of cell in left column where alignment is passing through.
    bool queryIdxLeftAlignmentFound = false;
    for (int queryIdx = queryIdxLeftStart; queryIdx <= queryIdxLeftEnd; queryIdx++) {
        leftScore = scoresLeft[queryIdx - scoresLeftStartIdx];
        rightScore = scoresRight[queryIdx + 1 - scoresRightStartIdx];
        if (leftScore + rightScore == bestScore) {
            queryIdxLeftAlignment = queryIdx;
            queryIdxLeftAlignmentFound = true;
            break;
        }
    }
    // Check boundary cells.
    if (!queryIdxLeftAlignmentFound && scoresLeftStartIdx == 0 && scoresRightStartIdx == 0) {
        leftScore = leftHalfWidth;
        rightScore = scoresRight[0];
        if (leftScore + rightScore == bestScore) {
            queryIdxLeftAlignment = -1;
            queryIdxLeftAlignmentFound = true;
        }
    }
    if (!queryIdxLeftAlignmentFound && scoresLeftStartIdx + scoresLeftLength == queryLength
        && scoresRightStartIdx + scoresRightLength == queryLength) {
        leftScore = scoresLeft[scoresLeftLength - 1];
        rightScore = rightHalfWidth;
        if (leftScore + rightScore == bestScore) {
            queryIdxLeftAlignment = queryLength - 1;
            queryIdxLeftAlignmentFound = true;
        }
    }

    delete[] scoresLeft;
    delete[] scoresRightOriginalStart;

    if (queryIdxLeftAlignmentFound == false) {
        // If there was no move that is part of optimal alignment, then there is no such alignment
        // or given bestScore is not correct!
        return EDLIB_STATUS_ERROR;
    }
    //----------------------------------------------------------//

    // Calculate alignments for upper half of left half (upper left - ul)
    // and lower half of right half (lower right - lr).
    const int ulHeight = queryIdxLeftAlignment + 1;
    const int lrHeight = queryLength - ulHeight;
    const int ulWidth = leftHalfWidth;
    const int lrWidth = rightHalfWidth;
    unsigned char* ulAlignment = NULL; int ulAlignmentLength;
    int ulStatusCode = obtainAlignment(query, rQuery + lrHeight, ulHeight,
                                       target, rTarget + lrWidth, ulWidth,
                                       equalityDefinition, alphabetLength, leftScore,
                                       &ulAlignment, &ulAlignmentLength);
    unsigned char* lrAlignment = NULL; int lrAlignmentLength;
    int lrStatusCode = obtainAlignment(query + ulHeight, rQuery, lrHeight,
                                       target + ulWidth, rTarget, lrWidth,
                                       equalityDefinition, alphabetLength, rightScore,
                                       &lrAlignment, &lrAlignmentLength);
    if (ulStatusCode == EDLIB_STATUS_ERROR || lrStatusCode == EDLIB_STATUS_ERROR) {
        if (ulAlignment) free(ulAlignment);
        if (lrAlignment) free(lrAlignment);
        return EDLIB_STATUS_ERROR;
    }

    // Build alignment by concatenating upper left alignment with lower right alignment.
    *alignmentLength = ulAlignmentLength + lrAlignmentLength;
    *alignment = (unsigned char*) malloc((*alignmentLength) * sizeof(unsigned char));
    memcpy(*alignment, ulAlignment, ulAlignmentLength);
    memcpy(*alignment + ulAlignmentLength, lrAlignment, lrAlignmentLength);

    free(ulAlignment);
    free(lrAlignment);
    return EDLIB_STATUS_OK;
}


/**
 * Takes char query and char target, recognizes alphabet and transforms them into unsigned char sequences
 * where elements in sequences are not any more letters of alphabet, but their index in alphabet.
 * Most of internal edlib functions expect such transformed sequences.
 * This function will allocate queryTransformed and targetTransformed, so make sure to free them when done.
 * Example:
 *   Original sequences: "ACT" and "CGT".
 *   Alphabet would be recognized as "ACTG". Alphabet length = 4.
 *   Transformed sequences: [0, 1, 2] and [1, 3, 2].
 * @param [in] queryOriginal
 * @param [in] queryLength
 * @param [in] targetOriginal
 * @param [in] targetLength
 * @param [out] queryTransformed  It will contain values in range [0, alphabet length - 1].
 * @param [out] targetTransformed  It will contain values in range [0, alphabet length - 1].
 * @return  Alphabet as a string of unique characters, where index of each character is its value in transformed
 *          sequences.
 */
static string transformSequences(const char* const queryOriginal, const int queryLength,
                                 const char* const targetOriginal, const int targetLength,
                                 unsigned char** const queryTransformed,
                                 unsigned char** const targetTransformed) {
    // Alphabet is constructed from letters that are present in sequences.
    // Each letter is assigned an ordinal number, starting from 0 up to alphabetLength - 1,
    // and new query and target are created in which letters are replaced with their ordinal numbers.
    // This query and target are used in all the calculations later.
    *queryTransformed = (unsigned char *) malloc(sizeof(unsigned char) * queryLength);
    *targetTransformed = (unsigned char *) malloc(sizeof(unsigned char) * targetLength);

    string alphabet = "";

    // Alphabet information, it is constructed on fly while transforming sequences.
    // letterIdx[c] is index of letter c in alphabet.
    unsigned char letterIdx[MAX_UCHAR + 1];
    bool inAlphabet[MAX_UCHAR + 1]; // inAlphabet[c] is true if c is in alphabet
    for (int i = 0; i < MAX_UCHAR + 1; i++) inAlphabet[i] = false;

    for (int i = 0; i < queryLength; i++) {
        unsigned char c = static_cast<unsigned char>(queryOriginal[i]);
        if (!inAlphabet[c]) {
            inAlphabet[c] = true;
            letterIdx[c] = (unsigned char) alphabet.size();
            alphabet += queryOriginal[i];
        }
        (*queryTransformed)[i] = letterIdx[c];
    }
    for (int i = 0; i < targetLength; i++) {
        unsigned char c = static_cast<unsigned char>(targetOriginal[i]);
        if (!inAlphabet[c]) {
            inAlphabet[c] = true;
            letterIdx[c] = (unsigned char) alphabet.size();
            alphabet += targetOriginal[i];
        }
        (*targetTransformed)[i] = letterIdx[c];
    }

    return alphabet;
}


extern "C" EdlibAlignConfig edlibNewAlignConfig(int k, EdlibAlignMode mode, EdlibAlignTask task,
                                                EdlibEqualityPair* additionalEqualities,
                                                int additionalEqualitiesLength) {
    EdlibAlignConfig config;
    config.k = k;
    config.mode = mode;
    config.task = task;
    config.additionalEqualities = additionalEqualities;
    config.additionalEqualitiesLength = additionalEqualitiesLength;
    return config;
}

extern "C" EdlibAlignConfig edlibDefaultAlignConfig(void) {
    return edlibNewAlignConfig(-1, EDLIB_MODE_NW, EDLIB_TASK_DISTANCE, NULL, 0);
}

extern "C" void edlibFreeAlignResult(EdlibAlignResult result) {
    if (result.endLocations) free(result.endLocations);
    if (result.startLocations) free(result.startLocations);
    if (result.alignment) free(result.alignment);
}