-
Notifications
You must be signed in to change notification settings - Fork 77
/
diff_sample.h
1109 lines (1062 loc) · 32.3 KB
/
diff_sample.h
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
#ifndef DIFF_SAMPLE_H_
#define DIFF_SAMPLE_H_
#if (__cplusplus >= 201103)
#include <thread>
#endif
#include <stdint.h>
#include "assert_helpers.h"
#include "btypes.h"
#include "ds.h"
#include "ls.h"
#include "multikey_qsort.h"
#include "timer.h"
#include "threading.h"
using namespace std;
#ifndef VMSG_NL
#define VMSG_NL(...) \
if(this->verbose()) { \
stringstream tmp; \
tmp << __VA_ARGS__ << endl; \
this->verbose(tmp.str()); \
}
#endif
#ifndef VMSG
#define VMSG(...) \
if(this->verbose()) { \
stringstream tmp; \
tmp << __VA_ARGS__; \
this->verbose(tmp.str()); \
}
#endif
/**
* Routines for calculating, sanity-checking, and dispensing difference
* cover samples to clients.
*/
/**
*
*/
struct sampleEntry {
uint32_t maxV;
uint32_t numSamples;
uint32_t samples[128];
};
/// Array of Colbourn and Ling calculated difference covers up to
/// r = 16 (maxV = 5953)
static struct sampleEntry clDCs[16];
static bool clDCs_calced = false; /// have clDCs been calculated?
/**
* Check that the given difference cover 'ds' actually covers all
* differences for a periodicity of v.
*/
template<typename T>
static bool dcRepOk(T v, EList<T>& ds) {
// diffs[] records all the differences observed
bool *covered = new bool[v];
for(T i = 1; i < v; i++) {
covered[i] = false;
}
for(T di = T(); di < ds.size(); di++) {
for(T dj = di+1; dj < ds.size(); dj++) {
assert_lt(ds[di], ds[dj]);
T d1 = (ds[dj] - ds[di]);
T d2 = (ds[di] + v - ds[dj]);
assert_lt(d1, v);
assert_lt(d2, v);
covered[d1] = true;
covered[d2] = true;
}
}
bool ok = true;
for(T i = 1; i < v; i++) {
if(covered[i] == false) {
ok = false;
break;
}
}
delete[] covered;
return ok;
}
/**
* Return true iff each element of ts (with length 'limit') is greater
* than the last.
*/
template<typename T>
static bool increasing(T* ts, size_t limit) {
for(size_t i = 0; i < limit-1; i++) {
if(ts[i+1] <= ts[i]) return false;
}
return true;
}
/**
* Return true iff the given difference cover covers difference 'diff'
* mod 'v'.
*/
template<typename T>
static inline bool hasDifference(T *ds, T d, T v, T diff) {
// diffs[] records all the differences observed
for(T di = T(); di < d; di++) {
for(T dj = di+1; dj < d; dj++) {
assert_lt(ds[di], ds[dj]);
T d1 = (ds[dj] - ds[di]);
T d2 = (ds[di] + v - ds[dj]);
assert_lt(d1, v);
assert_lt(d2, v);
if(d1 == diff || d2 == diff) return true;
}
}
return false;
}
/**
* Exhaustively calculate optimal difference cover samples for v = 4,
* 8, 16, 32, 64, 128, 256 and store results in p2DCs[]
*/
template<typename T>
void calcExhaustiveDC(T i, bool verbose = false, bool sanityCheck = false) {
T v = i;
bool *diffs = new bool[v];
// v is the target period
T ld = (T)ceil(sqrt(v));
// ud is the upper bound on |D|
T ud = v / 2;
// for all possible |D|s
bool ok = true;
T *ds = NULL;
T d;
for(d = ld; d <= ud+1; d++) {
// for all possible |D| samples
ds = new T[d];
for(T j = 0; j < d; j++) {
ds[j] = j;
}
assert(increasing(ds, d));
while(true) {
// reset diffs[]
for(T t = 1; t < v; t++) {
diffs[t] = false;
}
T diffCnt = 0;
// diffs[] records all the differences observed
for(T di = 0; di < d; di++) {
for(T dj = di+1; dj < d; dj++) {
assert_lt(ds[di], ds[dj]);
T d1 = (ds[dj] - ds[di]);
T d2 = (ds[di] + v - ds[dj]);
assert_lt(d1, v);
assert_lt(d2, v);
assert_gt(d1, 0);
assert_gt(d2, 0);
if(!diffs[d1]) {
diffCnt++;
diffs[d1] = true;
}
if(!diffs[d2]) {
diffCnt++;
diffs[d2] = true;
}
}
}
// Do we observe all possible differences (except 0)
ok = diffCnt == v-1;
if(ok) {
// Yes, all differences are covered
break;
} else {
// Advance ds
// (Following is commented out because it turns out
// it's slow)
// Find a missing difference
//uint32_t missing = 0xffffffff;
//for(uint32_t t = 1; t < v; t++) {
// if(diffs[t] == false) {
// missing = diffs[t];
// break;
// }
//}
//assert_neq(missing, 0xffffffff);
assert(increasing(ds, d));
bool advanced = false;
bool keepGoing = false;
do {
keepGoing = false;
for(T bd = d-1; bd > 1; bd--) {
T dif = (d-1)-bd;
if(ds[bd] < v-1-dif) {
ds[bd]++;
assert_neq(0, ds[bd]);
// Reset subsequent ones
for(T bdi = bd+1; bdi < d; bdi++) {
assert_eq(0, ds[bdi]);
ds[bdi] = ds[bdi-1]+1;
assert_gt(ds[bdi], ds[bdi-1]);
}
assert(increasing(ds, d));
// (Following is commented out because
// it turns out it's slow)
// See if the new DC has the missing value
//if(!hasDifference(ds, d, v, missing)) {
// keepGoing = true;
// break;
//}
advanced = true;
break;
} else {
ds[bd] = 0;
// keep going
}
}
} while(keepGoing);
// No solution for this |D|
if(!advanced) break;
assert(increasing(ds, d));
}
} // next sample assignment
if(ok) {
break;
}
delete[] ds;
} // next |D|
assert(ok);
delete[] diffs;
cout << "Did exhaustive v=" << v << " |D|=" << d << endl;
cout << " ";
for(T i = 0; i < d; i++) {
cout << ds[i];
if(i < d-1) cout << ",";
}
cout << endl;
delete[] ds;
}
/**
* Routune for calculating the elements of clDCs up to r = 16 using the
* technique of Colbourn and Ling.
*
* See http://citeseer.ist.psu.edu/211575.html
*/
template <typename T>
void calcColbournAndLingDCs(bool verbose = false, bool sanityCheck = false) {
for(T r = 0; r < 16; r++) {
T maxv = 24*r*r + 36*r + 13; // Corollary 2.3
T numsamp = 6*r + 4;
clDCs[r].maxV = maxv;
clDCs[r].numSamples = numsamp;
memset(clDCs[r].samples, 0, 4 * 128);
T i;
// clDCs[r].samples[0] = 0;
// Fill in the 1^r part of the B series
for(i = 1; i < r+1; i++) {
clDCs[r].samples[i] = clDCs[r].samples[i-1] + 1;
}
// Fill in the (r + 1)^1 part
clDCs[r].samples[r+1] = clDCs[r].samples[r] + r + 1;
// Fill in the (2r + 1)^r part
for(i = r+2; i < r+2+r; i++) {
clDCs[r].samples[i] = clDCs[r].samples[i-1] + 2*r + 1;
}
// Fill in the (4r + 3)^(2r + 1) part
for(i = r+2+r; i < r+2+r+2*r+1; i++) {
clDCs[r].samples[i] = clDCs[r].samples[i-1] + 4*r + 3;
}
// Fill in the (2r + 2)^(r + 1) part
for(i = r+2+r+2*r+1; i < r+2+r+2*r+1+r+1; i++) {
clDCs[r].samples[i] = clDCs[r].samples[i-1] + 2*r + 2;
}
// Fill in the last 1^r part
for(i = r+2+r+2*r+1+r+1; i < r+2+r+2*r+1+r+1+r; i++) {
clDCs[r].samples[i] = clDCs[r].samples[i-1] + 1;
}
assert_eq(i, numsamp);
assert_lt(i, 128);
if(sanityCheck) {
// diffs[] records all the differences observed
bool *diffs = new bool[maxv];
for(T i = 0; i < numsamp; i++) {
for(T j = i+1; j < numsamp; j++) {
T d1 = (clDCs[r].samples[j] - clDCs[r].samples[i]);
T d2 = (clDCs[r].samples[i] + maxv - clDCs[r].samples[j]);
assert_lt(d1, maxv);
assert_lt(d2, maxv);
diffs[d1] = true;
diffs[d2] = true;
}
}
// Should have observed all possible differences (except 0)
for(T i = 1; i < maxv; i++) {
if(diffs[i] == false) cout << r << ", " << i << endl;
assert(diffs[i] == true);
}
delete[] diffs;
}
}
clDCs_calced = true;
}
/**
* Entries 4-57 are transcribed from page 6 of Luk and Wong's paper
* "Two New Quorum Based Algorithms for Distributed Mutual Exclusion",
* which is also used and cited in the Burkhardt and Karkkainen's
* papers on difference covers for sorting. These samples are optimal
* according to Luk and Wong.
*
* All other entries are generated via the exhaustive algorithm in
* calcExhaustiveDC().
*
* The 0 is stored at the end of the sample as an end-of-list marker,
* but 0 is also an element of each.
*
* Note that every difference cover has a 0 and a 1. Intuitively,
* any optimal difference cover sample can be oriented (i.e. rotated)
* such that it includes 0 and 1 as elements.
*
* All samples in this list have been verified to be complete covers.
*
* A value of 0xffffffff in the first column indicates that there is no
* sample for that value of v. We do not keep samples for values of v
* less than 3, since they are trivial (and the caller probably didn't
* mean to ask for it).
*/
static uint32_t dc0to64[65][10] = {
{0xffffffff}, // 0
{0xffffffff}, // 1
{0xffffffff}, // 2
{1, 0}, // 3
{1, 2, 0}, // 4
{1, 2, 0}, // 5
{1, 3, 0}, // 6
{1, 3, 0}, // 7
{1, 2, 4, 0}, // 8
{1, 2, 4, 0}, // 9
{1, 2, 5, 0}, // 10
{1, 2, 5, 0}, // 11
{1, 3, 7, 0}, // 12
{1, 3, 9, 0}, // 13
{1, 2, 3, 7, 0}, // 14
{1, 2, 3, 7, 0}, // 15
{1, 2, 5, 8, 0}, // 16
{1, 2, 4, 12, 0}, // 17
{1, 2, 5, 11, 0}, // 18
{1, 2, 6, 9, 0}, // 19
{1, 2, 3, 6, 10, 0}, // 20
{1, 4, 14, 16, 0}, // 21
{1, 2, 3, 7, 11, 0}, // 22
{1, 2, 3, 7, 11, 0}, // 23
{1, 2, 3, 7, 15, 0}, // 24
{1, 2, 3, 8, 12, 0}, // 25
{1, 2, 5, 9, 15, 0}, // 26
{1, 2, 5, 13, 22, 0}, // 27
{1, 4, 15, 20, 22, 0}, // 28
{1, 2, 3, 4, 9, 14, 0}, // 29
{1, 2, 3, 4, 9, 19, 0}, // 30
{1, 3, 8, 12, 18, 0}, // 31
{1, 2, 3, 7, 11, 19, 0}, // 32
{1, 2, 3, 6, 16, 27, 0}, // 33
{1, 2, 3, 7, 12, 20, 0}, // 34
{1, 2, 3, 8, 12, 21, 0}, // 35
{1, 2, 5, 12, 14, 20, 0}, // 36
{1, 2, 4, 10, 15, 22, 0}, // 37
{1, 2, 3, 4, 8, 14, 23, 0}, // 38
{1, 2, 4, 13, 18, 33, 0}, // 39
{1, 2, 3, 4, 9, 14, 24, 0}, // 40
{1, 2, 3, 4, 9, 15, 25, 0}, // 41
{1, 2, 3, 4, 9, 15, 25, 0}, // 42
{1, 2, 3, 4, 10, 15, 26, 0}, // 43
{1, 2, 3, 6, 16, 27, 38, 0}, // 44
{1, 2, 3, 5, 12, 18, 26, 0}, // 45
{1, 2, 3, 6, 18, 25, 38, 0}, // 46
{1, 2, 3, 5, 16, 22, 40, 0}, // 47
{1, 2, 5, 9, 20, 26, 36, 0}, // 48
{1, 2, 5, 24, 33, 36, 44, 0}, // 49
{1, 3, 8, 17, 28, 32, 38, 0}, // 50
{1, 2, 5, 11, 18, 30, 38, 0}, // 51
{1, 2, 3, 4, 6, 14, 21, 30, 0}, // 52
{1, 2, 3, 4, 7, 21, 29, 44, 0}, // 53
{1, 2, 3, 4, 9, 15, 21, 31, 0}, // 54
{1, 2, 3, 4, 6, 19, 26, 47, 0}, // 55
{1, 2, 3, 4, 11, 16, 33, 39, 0}, // 56
{1, 3, 13, 32, 36, 43, 52, 0}, // 57
// Generated by calcExhaustiveDC()
{1, 2, 3, 7, 21, 33, 37, 50, 0}, // 58
{1, 2, 3, 6, 13, 21, 35, 44, 0}, // 59
{1, 2, 4, 9, 15, 25, 30, 42, 0}, // 60
{1, 2, 3, 7, 15, 25, 36, 45, 0}, // 61
{1, 2, 4, 10, 32, 39, 46, 51, 0}, // 62
{1, 2, 6, 8, 20, 38, 41, 54, 0}, // 63
{1, 2, 5, 14, 16, 34, 42, 59, 0} // 64
};
/**
* Get a difference cover for the requested periodicity v.
*/
template <typename T>
static EList<T> getDiffCover(T v,
bool verbose = false,
bool sanityCheck = false)
{
assert_gt(v, 2);
EList<T> ret;
// Can we look it up in our hardcoded array?
if(v <= 64 && dc0to64[v][0] == 0xffffffff) {
if(verbose) cout << "v in hardcoded area, but hardcoded entry was all-fs" << endl;
return ret;
} else if(v <= 64) {
ret.push_back(0);
for(size_t i = 0; i < 10; i++) {
if(dc0to64[v][i] == 0) break;
ret.push_back(dc0to64[v][i]);
}
if(sanityCheck) assert(dcRepOk(v, ret));
return ret;
}
// Can we look it up in our calcColbournAndLingDCs array?
if(!clDCs_calced) {
calcColbournAndLingDCs<uint32_t>(verbose, sanityCheck);
assert(clDCs_calced);
}
for(size_t i = 0; i < 16; i++) {
if(v <= clDCs[i].maxV) {
for(size_t j = 0; j < clDCs[i].numSamples; j++) {
T s = clDCs[i].samples[j];
if(s >= v) {
s %= v;
for(size_t k = 0; k < ret.size(); k++) {
if(s == ret[k]) break;
if(s < ret[k]) {
ret.insert(s, k);
break;
}
}
} else {
ret.push_back(s % v);
}
}
if(sanityCheck) assert(dcRepOk(v, ret));
return ret;
}
}
cerr << "Error: Could not find a difference cover sample for v=" << v << endl;
throw 1;
}
/**
* Calculate and return a delta map based on the given difference cover
* and periodicity v.
*/
template <typename T>
static EList<T> getDeltaMap(T v, const EList<T>& dc) {
// Declare anchor-map-related items
EList<T> amap;
size_t amapEnts = 1;
amap.resizeExact((size_t)v);
amap.fill(0xffffffff);
amap[0] = 0;
// Print out difference cover (and optionally calculate
// anchor map)
for(size_t i = 0; i < dc.size(); i++) {
for(size_t j = i+1; j < dc.size(); j++) {
assert_gt(dc[j], dc[i]);
T diffLeft = dc[j] - dc[i];
T diffRight = dc[i] + v - dc[j];
assert_lt(diffLeft, v);
assert_lt(diffRight, v);
if(amap[diffLeft] == 0xffffffff) {
amap[diffLeft] = dc[i];
amapEnts++;
}
if(amap[diffRight] == 0xffffffff) {
amap[diffRight] = dc[j];
amapEnts++;
}
}
}
return amap;
}
/**
* Return population count (count of all bits set to 1) of i.
*/
template<typename T>
static unsigned int popCount(T i) {
unsigned int cnt = 0;
for(size_t j = 0; j < sizeof(T)*8; j++) {
if(i & 1) cnt++;
i >>= 1;
}
return cnt;
}
/**
* Calculate log-base-2 of i
*/
template<typename T>
static unsigned int myLog2(T i) {
assert_eq(1, popCount(i)); // must be power of 2
for(size_t j = 0; j < sizeof(T)*8; j++) {
if(i & 1) return (int)j;
i >>= 1;
}
assert(false);
return 0xffffffff;
}
/**
*
*/
template<typename TStr>
class DifferenceCoverSample {
public:
DifferenceCoverSample(const TStr& __text,
uint32_t __v,
bool __verbose = false,
bool __sanity = false,
ostream& __logger = cout) :
_text(__text),
_v(__v),
_verbose(__verbose),
_sanity(__sanity),
_ds(getDiffCover(_v, _verbose, _sanity)),
_dmap(getDeltaMap(_v, _ds)),
_d((uint32_t)_ds.size()),
_doffs(),
_isaPrime(),
_dInv(),
_log2v(myLog2(_v)),
_vmask(OFF_MASK << _log2v),
_logger(__logger)
{
assert_gt(_d, 0);
assert_eq(1, popCount(_v)); // must be power of 2
// Build map from d's to idx's
_dInv.resizeExact((size_t)_v);
_dInv.fill(0xffffffff);
uint32_t lim = (uint32_t)_ds.size();
for(uint32_t i = 0; i < lim; i++) {
_dInv[_ds[i]] = i;
}
}
/**
* Allocate an amount of memory that simulates the peak memory
* usage of the DifferenceCoverSample with the given text and v.
* Throws bad_alloc if it's not going to fit in memory. Returns
* the approximate number of bytes the Cover takes at all times.
*/
static size_t simulateAllocs(const TStr& text, uint32_t v) {
EList<uint32_t> ds = getDiffCover(v, false /*verbose*/, false /*sanity*/);
size_t len = text.length();
size_t sPrimeSz = (len / v) * ds.size();
// sPrime, sPrimeOrder, _isaPrime all exist in memory at
// once and that's the peak
AutoArray<TIndexOffU> aa(sPrimeSz * 3 + (1024 * 1024 /*out of caution*/));
return sPrimeSz * 4; // sPrime array
}
uint32_t v() const { return _v; }
uint32_t log2v() const { return _log2v; }
uint32_t vmask() const { return _vmask; }
uint32_t modv(TIndexOffU i) const { return (uint32_t)(i & ~_vmask); }
TIndexOffU divv(TIndexOffU i) const { return i >> _log2v; }
uint32_t d() const { return _d; }
bool verbose() const { return _verbose; }
bool sanityCheck() const { return _sanity; }
const TStr& text() const { return _text; }
const EList<uint32_t>& ds() const { return _ds; }
const EList<uint32_t>& dmap() const { return _dmap; }
ostream& log() const { return _logger; }
void build(int nthreads);
uint32_t tieBreakOff(TIndexOffU i, TIndexOffU j) const;
int64_t breakTie(TIndexOffU i, TIndexOffU j) const;
bool isCovered(TIndexOffU i) const;
TIndexOffU rank(TIndexOffU i) const;
/**
* Print out the suffix array such that every sample offset has its
* rank filled in and every non-sample offset is shown as '-'.
*/
void print(ostream& out) {
for(size_t i = 0; i < _text.length(); i++) {
if(isCovered(i)) {
out << rank(i);
} else {
out << "-";
}
if(i < _text.length()-1) {
out << ",";
}
}
out << endl;
}
private:
void doBuiltSanityCheck() const;
void buildSPrime(EList<TIndexOffU>& sPrime, size_t padding);
bool built() const {
return _isaPrime.size() > 0;
}
void verbose(const string& s) const {
if(this->verbose()) {
this->log() << s.c_str();
this->log().flush();
}
}
const TStr& _text; // text to sample
uint32_t _v; // periodicity of sample
bool _verbose; //
bool _sanity; //
EList<uint32_t> _ds; // samples: idx -> d
EList<uint32_t> _dmap; // delta map
uint32_t _d; // |D| - size of sample
EList<TIndexOffU> _doffs; // offsets into sPrime/isaPrime for each d idx
EList<TIndexOffU> _isaPrime; // ISA' array
EList<uint32_t> _dInv; // Map from d -> idx
uint32_t _log2v;
TIndexOffU _vmask;
ostream& _logger;
};
/**
* Return true iff suffixes with offsets suf1 and suf2 out of host
* string 'host' are identical up to depth 'v'.
*/
template <typename TStr>
static inline bool suffixLt(const TStr& host, TIndexOffU suf1, TIndexOffU suf2) {
TIndexOffU hlen = (TIndexOffU)host.length();
assert_neq(suf1, suf2);
TIndexOffU i = 0;
while(suf1 + i < hlen && suf2 + i < hlen) {
if(host[suf1+i] < host[suf2+i]) return true;
if(host[suf1+i] > host[suf2+i]) return false;
i++;
}
if(suf1 + i == hlen) {
assert_lt(suf2 + i, hlen);
return false;
}
assert_eq(suf2 + i, hlen);
return true;
}
/**
* Sanity-check the difference cover by first inverting _isaPrime then
* checking that each successive suffix really is less than the next.
*/
template <typename TStr>
void DifferenceCoverSample<TStr>::doBuiltSanityCheck() const {
uint32_t v = this->v();
assert(built());
VMSG_NL(" Doing sanity check");
TIndexOffU added = 0;
EList<TIndexOffU> sorted;
sorted.resizeExact(_isaPrime.size());
sorted.fill(OFF_MASK);
for(size_t di = 0; di < this->d(); di++) {
uint32_t d = _ds[di];
size_t i = 0;
for(size_t doi = _doffs[di]; doi < _doffs[di+1]; doi++, i++) {
assert_eq(OFF_MASK, sorted[_isaPrime[doi]]);
// Maps the offset of the suffix to its rank
sorted[_isaPrime[doi]] = (TIndexOffU)(v*i + d);
added++;
}
}
assert_eq(added, _isaPrime.size());
for(size_t i = 0; i < sorted.size()-1; i++) {
assert(suffixLt(this->text(), sorted[i], sorted[i+1]));
}
}
/**
* Build the s' array by sampling suffixes (suffix offsets, actually)
* from t according to the difference-cover sample and pack them into
* an array of machine words in the order dictated by the "mu" mapping
* described in Burkhardt.
*
* Also builds _doffs map.
*/
template <typename TStr>
void DifferenceCoverSample<TStr>::buildSPrime(
EList<TIndexOffU>& sPrime,
size_t padding)
{
const TStr& t = this->text();
const EList<uint32_t>& ds = this->ds();
TIndexOffU tlen = (TIndexOffU)t.length();
uint32_t v = this->v();
uint32_t d = this->d();
assert_gt(v, 2);
assert_lt(d, v);
// Record where each d section should begin in sPrime
TIndexOffU tlenDivV = this->divv(tlen);
uint32_t tlenModV = this->modv(tlen);
TIndexOffU sPrimeSz = 0;
assert(_doffs.empty());
_doffs.resize(d+1);
for(uint32_t di = 0; di < d; di++) {
// mu mapping
TIndexOffU sz = tlenDivV + ((ds[di] <= tlenModV) ? 1 : 0);
assert_geq(sz, 0);
_doffs[di] = sPrimeSz;
sPrimeSz += sz;
}
_doffs[d] = sPrimeSz;
#ifndef NDEBUG
if(tlenDivV > 0) {
for(size_t i = 0; i < d; i++) {
assert_gt(_doffs[i+1], _doffs[i]);
TIndexOffU diff = _doffs[i+1] - _doffs[i];
assert(diff == tlenDivV || diff == tlenDivV+1);
}
}
#endif
assert_eq(_doffs.size(), d+1);
// Size sPrime appropriately
sPrime.resizeExact((size_t)sPrimeSz + padding);
sPrime.fill(OFF_MASK);
// Slot suffixes from text into sPrime according to the mu
// mapping; where the mapping would leave a blank, insert a 0
TIndexOffU added = 0;
TIndexOffU i = 0;
for(uint64_t ti = 0; ti <= tlen; ti += v) {
for(uint32_t di = 0; di < d; di++) {
TIndexOffU tti = (TIndexOffU)(ti + ds[di]);
if(tti > tlen) break;
TIndexOffU spi = _doffs[di] + i;
assert_lt(spi, _doffs[di+1]);
assert_leq(tti, tlen);
assert_lt(spi, sPrimeSz);
assert_eq(OFF_MASK, sPrime[spi]);
sPrime[spi] = tti; added++;
}
i++;
}
assert_eq(added, sPrimeSz);
}
/**
* Return true iff suffixes with offsets suf1 and suf2 out of host
* string 'host' are identical up to depth 'v'.
*/
template <typename TStr>
static inline bool suffixSameUpTo(
const TStr& host,
TIndexOffU suf1,
TIndexOffU suf2,
TIndexOffU v)
{
for(TIndexOffU i = 0; i < v; i++) {
bool endSuf1 = suf1+i >= host.length();
bool endSuf2 = suf2+i >= host.length();
if((endSuf1 && !endSuf2) || (!endSuf1 && endSuf2)) return false;
if(endSuf1 && endSuf2) return true;
if(host[suf1+i] != host[suf2+i]) return false;
}
return true;
}
template<typename TStr>
struct VSortingParam {
DifferenceCoverSample<TStr>* dcs;
TIndexOffU* sPrimeArr;
size_t sPrimeSz;
TIndexOffU* sPrimeOrderArr;
size_t depth;
const EList<size_t>* boundaries;
size_t* cur;
MUTEX_T* mutex;
};
template<typename TStr>
static void VSorting_worker(void *vp)
{
VSortingParam<TStr>* param = (VSortingParam<TStr>*)vp;
DifferenceCoverSample<TStr>* dcs = param->dcs;
const TStr& host = dcs->text();
const size_t hlen = host.length();
uint32_t v = dcs->v();
while(true) {
size_t cur = 0;
{
ThreadSafe ts(param->mutex);
cur = *(param->cur);
(*param->cur)++;
}
if(cur >= param->boundaries->size()) return;
size_t begin = (cur == 0 ? 0 : (*param->boundaries)[cur-1]);
size_t end = (*param->boundaries)[cur];
assert_leq(begin, end);
if(end - begin <= 1) continue;
mkeyQSortSuf2(
host,
hlen,
param->sPrimeArr,
param->sPrimeSz,
param->sPrimeOrderArr,
4,
begin,
end,
param->depth,
v);
}
}
/**
* Calculates a ranking of all suffixes in the sample and stores them,
* packed according to the mu mapping, in _isaPrime.
*/
template <typename TStr>
void DifferenceCoverSample<TStr>::build(int nthreads) {
// Local names for relevant types
VMSG_NL("Building DifferenceCoverSample");
// Local names for relevant data
const TStr& t = this->text();
uint32_t v = this->v();
assert_gt(v, 2);
// Build s'
size_t padding = 1;
EList<TIndexOffU> sPrime;
VMSG_NL(" Building sPrime");
buildSPrime(sPrime, padding);
size_t sPrimeSz = sPrime.size() - padding;
assert_gt(sPrime.size(), 0);
assert_leq(sPrime.size(), t.length()+1); // +1 is because of the end-cap
TIndexOffU nextRank = 0;
{
VMSG_NL(" Building sPrimeOrder");
EList<TIndexOffU> sPrimeOrder;
sPrimeOrder.resizeExact(sPrimeSz);
for(TIndexOffU i = 0; i < (TIndexOffU)sPrimeOrder.size(); i++) {
sPrimeOrder[i] = i;
}
// sPrime now holds suffix-offsets for DC samples.
{
Timer timer(cout, " V-Sorting samples time: ", this->verbose());
VMSG_NL(" V-Sorting samples");
// Extract backing-store array from sPrime and sPrimeOrder;
// the mkeyQSortSuf2 routine works on the array for maximum
// efficiency
TIndexOffU *sPrimeArr = (TIndexOffU*)sPrime.ptr();
assert_eq(sPrimeArr[0], sPrime[0]);
assert_eq(sPrimeArr[sPrimeSz-1], sPrime[sPrimeSz-1]);
TIndexOffU *sPrimeOrderArr = (TIndexOffU*)sPrimeOrder.ptr();
assert_eq(sPrimeOrderArr[0], sPrimeOrder[0]);
assert_eq(sPrimeOrderArr[sPrimeSz-1], sPrimeOrder[sPrimeSz-1]);
TIndexOffU *sOrig = NULL;
if(this->sanityCheck()) {
sOrig = new TIndexOffU[sPrimeSz];
memcpy(sOrig, sPrimeArr, OFF_SIZE * sPrimeSz);
}
// Sort sample suffixes up to the vth character using a
// multikey quicksort. Sort time is proportional to the
// number of samples times v. It isn't quadratic.
// sPrimeOrder is passed in as a swapping partner for
// sPrimeArr, i.e., every time the multikey qsort swaps
// elements in sPrime, it swaps the same elements in
// sPrimeOrder too. This allows us to easily reconstruct
// what the sort did.
if(nthreads == 1) {
mkeyQSortSuf2(t, sPrimeArr, sPrimeSz, sPrimeOrderArr, 4,
this->verbose(), this->sanityCheck(), v);
} else {
int query_depth = 0;
int tmp_nthreads = nthreads;
while(tmp_nthreads > 0) {
query_depth++;
tmp_nthreads >>= 1;
}
EList<size_t> boundaries; // bucket boundaries for parallelization
mkeyQSortSuf2(t, sPrimeArr, sPrimeSz, sPrimeOrderArr, 4,
this->verbose(), false, query_depth, &boundaries);
if(boundaries.size() > 0) {
#if (__cplusplus >= 201103)
AutoArray<std::thread*> threads(nthreads);
#else
AutoArray<tthread::thread*> threads(nthreads);
#endif
EList<VSortingParam<TStr> > tparams;
tparams.resize(nthreads);
size_t cur = 0;
MUTEX_T mutex;
for(int tid = 0; tid < nthreads; tid++) {
// Calculate bucket sizes by doing a binary search for each
// suffix and noting where it lands
tparams[tid].dcs = this;
tparams[tid].sPrimeArr = sPrimeArr;
tparams[tid].sPrimeSz = sPrimeSz;
tparams[tid].sPrimeOrderArr = sPrimeOrderArr;
tparams[tid].depth = query_depth;
tparams[tid].boundaries = &boundaries;
tparams[tid].cur = &cur;
tparams[tid].mutex = &mutex;
#if (__cplusplus >= 201103L)
threads[tid] = new std::thread(VSorting_worker<TStr>, (void*)&tparams[tid]);
#else
threads[tid] = new tthread::thread(VSorting_worker<TStr>, (void*)&tparams[tid]);
#endif
}
for (int tid = 0; tid < nthreads; tid++) {
threads[tid]->join();
}
}
}
if(this->sanityCheck()) {
sanityCheckOrderedSufs(t, t.length(), sPrimeArr, sPrimeSz, v);
for(size_t i = 0; i < sPrimeSz; i++) {
assert_eq(sPrimeArr[i], sOrig[sPrimeOrderArr[i]]);
}
delete[] sOrig;
}
// Make sure sPrime and sPrimeOrder are consistent with
// their respective backing-store arrays
assert_eq(sPrimeArr[0], sPrime[0]);
assert_eq(sPrimeArr[sPrimeSz-1], sPrime[sPrimeSz-1]);
assert_eq(sPrimeOrderArr[0], sPrimeOrder[0]);
assert_eq(sPrimeOrderArr[sPrimeSz-1], sPrimeOrder[sPrimeSz-1]);
}
// Now assign the ranking implied by the sorted sPrime/sPrimeOrder
// arrays back into sPrime.
VMSG_NL(" Allocating rank array");
_isaPrime.resize(sPrime.size());
_isaPrime.resizeExact(sPrime.size());
ASSERT_ONLY(_isaPrime.fill(OFF_MASK));
assert_gt(_isaPrime.size(), 0);
{
Timer timer(cout, " Ranking v-sort output time: ", this->verbose());
VMSG_NL(" Ranking v-sort output");
for(size_t i = 0; i < sPrimeSz-1; i++) {
// Place the appropriate ranking
_isaPrime[sPrimeOrder[i]] = nextRank;
// If sPrime[i] and sPrime[i+1] are identical up to v, then we
// should give the next suffix the same rank
if(!suffixSameUpTo(t, sPrime[i], sPrime[i+1], v)) nextRank++;
}
_isaPrime[sPrimeOrder[sPrimeSz-1]] = nextRank; // finish off
}
// sPrimeOrder is destroyed
// All the information we need is now in _isaPrime
}
#ifndef NDEBUG
// Check that all ranks are sane
for(size_t i = 0; i < sPrimeSz; i++) {
assert_neq(_isaPrime[i], OFF_MASK);
assert_lt(_isaPrime[i], sPrimeSz);
}
#endif
// Now pass the sPrimeRanks[] array to LarssonSadakane (in SeqAn).
_isaPrime[_isaPrime.size() - 1] = sPrimeSz;
sPrime[sPrime.size() - 1] = sPrimeSz;
{
Timer timer(cout, " Invoking Larsson-Sadakane on ranks time: ", this->verbose());
VMSG_NL(" Invoking Larsson-Sadakane on ranks");
LarssonSadakane<TIndexOff> ls;
ls.suffixsort(
(TIndexOff*)&_isaPrime[0],
(TIndexOff*)&sPrime[0],
(TIndexOff)sPrimeSz,
(TIndexOff)sPrime.size(),
0);
}
// sPrime now contains the suffix array (which we ignore)
assert_eq(_isaPrime.size(), sPrime.size());
assert_gt(_isaPrime.size(), 0);
// chop off final character of _isaPrime
_isaPrime.resize(sPrimeSz);
// Subtract 1 from each isaPrime (to adjust for LarssonSadakane
// always ranking the final suffix as lexicographically first)
for(size_t i = 0; i < _isaPrime.size(); i++) {
_isaPrime[i]--;
}
VMSG_NL(" Sanity-checking and returning");
// done!
//if(this->verbose()) print(cout);
if(this->sanityCheck()) doBuiltSanityCheck();
}
/**
* Return true iff index i within the text is covered by the difference
* cover sample. Allow i to be off the end of the text; simplifies
* logic elsewhere.