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
|
/*
* Copyright 2011, Ben Langmead <langmea@cs.jhu.edu>
*
* This file is part of Bowtie 2.
*
* Bowtie 2 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* Bowtie 2 is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with Bowtie 2. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef DIFF_SAMPLE_H_
#define DIFF_SAMPLE_H_
#ifdef WITH_TBB
#include <tbb/tbb.h>
#include <tbb/task_group.h>
#endif
#include <stdint.h>
#include <string.h>
#include "assert_helpers.h"
#include "multikey_qsort.h"
#include "timer.h"
#include "ds.h"
#include "mem_ids.h"
#include "ls.h"
#include "btypes.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)
extern struct sampleEntry clDCs[16];
extern bool clDCs_calced; /// 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
AutoArray<bool> covered(v, EBWT_CAT);
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;
}
}
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;
AutoArray<bool> diffs(v, EBWT_CAT);
// 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
AutoArray<T> ds(d, EBWT_CAT);
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);
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wmisleading-indentation"
if(!diffs[d1]) diffCnt++; diffs[d1] = true;
if(!diffs[d2]) diffCnt++; diffs[d2] = true;
#pragma GCC diagnostic pop
}
}
// 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;
}
} // next |D|
assert(ok);
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;
}
/**
* 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
AutoArray<bool> diffs(maxv, EBWT_CAT);
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);
}
}
}
clDCs_calced = true;
}
/**
* A precalculated list of difference covers.
*/
extern uint32_t dc0to64[65][10];
/**
* 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;
ret.clear();
// 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*/), EBWT_CAT);
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;
};
/**
* 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());
#ifndef NDEBUG
for(size_t i = 0; i < sorted.size()-1; i++) {
assert(sstr_suf_lt(this->text(), sorted[i], this->text(), sorted[i+1], false));
}
#endif
}
/**
* 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.resizeExact((size_t)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(TIndexOffU ti = 0; ti <= tlen; ti += v) {
for(uint32_t di = 0; di < d; di++) {
TIndexOffU tti = 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>
#ifdef WITH_TBB
class VSorting_worker {
void *vp;
public:
VSorting_worker(const VSorting_worker& W): vp(W.vp) {};
VSorting_worker(void *vp_):vp(vp_) {};
void operator()() const
{
#else
static void VSorting_worker(void *vp)
{
#endif
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, true);
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);
}
}
#ifdef WITH_TBB
};
#endif
/**
* 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'
EList<TIndexOffU> sPrime;
// Need to allocate 2 extra elements at the end of the sPrime and _isaPrime
// arrays. One element that's less than all others, and another that acts
// as needed padding for the Larsson-Sadakane sorting code.
size_t padding = 1;
VMSG_NL(" Building sPrime");
buildSPrime(sPrime, padding);
size_t sPrimeSz = sPrime.size() - padding;
assert_gt(sPrime.size(), padding);
assert_leq(sPrime.size(), t.length() + padding + 1);
TIndexOffU nextRank = 0;
{
VMSG_NL(" Building sPrimeOrder");
EList<TIndexOffU> sPrimeOrder;
sPrimeOrder.resizeExact(sPrimeSz);
for(TIndexOffU i = 0; i < sPrimeSz; 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]);
// 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
TIndexOffU *sOrig = NULL;
if(this->sanityCheck()) {
sOrig = new TIndexOffU[sPrimeSz];
memcpy(sOrig, sPrimeArr, OFF_SIZE * sPrimeSz);
}
mkeyQSortSuf2(t, sPrimeArr, sPrimeSz, sPrimeOrderArr, 4,
this->verbose(), false, query_depth, &boundaries);
if(boundaries.size() > 0) {
#ifdef WITH_TBB
tbb::task_group tbb_grp;
#else
AutoArray<tthread::thread*> threads(nthreads);
#endif
EList<VSortingParam<TStr> > tparams;
size_t cur = 0;
MUTEX_T mutex;
tparams.resize(nthreads);
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;
#ifdef WITH_TBB
tbb_grp.run(VSorting_worker<TStr>(((void*)&tparams[tid])));
}
tbb_grp.wait();
#else
threads[tid] = new tthread::thread(VSorting_worker<TStr>, (void*)&tparams[tid]);
}
for (int tid = 0; tid < nthreads; tid++) {
threads[tid]->join();
}
#endif
}
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.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
#ifndef NDEBUG
for(size_t i = 0; i < sPrimeSz; i++) {
assert_neq(OFF_MASK, _isaPrime[i]);
assert_lt(_isaPrime[i], sPrimeSz);
}
#endif
}
// sPrimeOrder is destroyed
// All the information we need is now in _isaPrime
}
_isaPrime[_isaPrime.size()-1] = (TIndexOffU)sPrimeSz;
sPrime[sPrime.size()-1] = (TIndexOffU)sPrimeSz;
// _isaPrime[_isaPrime.size()-1] and sPrime[sPrime.size()-1] are just
// spacer for the Larsson-Sadakane routine to use
{
Timer timer(cout, " Invoking Larsson-Sadakane on ranks time: ", this->verbose());
VMSG_NL(" Invoking Larsson-Sadakane on ranks");
if(sPrime.size() >= LS_SIZE) {
cerr << "Error; sPrime array has so many elements that it can't be converted to a signed array without overflow." << endl;
throw 1;
}
LarssonSadakane<TIndexOff> ls;
ls.suffixsort(
(TIndexOff*)_isaPrime.ptr(),
(TIndexOff*)sPrime.ptr(),
(TIndexOff)sPrimeSz,
(TIndexOff)sPrime.size(),
0);
}
// chop off final character of _isaPrime
_isaPrime.resizeExact(sPrimeSz);
for(size_t i = 0; i < _isaPrime.size(); i++) {
_isaPrime[i]--;
}
#ifndef NDEBUG
for(size_t i = 0; i < sPrimeSz-1; i++) {
assert_lt(_isaPrime[i], sPrimeSz);
assert(i == 0 || _isaPrime[i] != _isaPrime[i-1]);
}
#endif
VMSG_NL(" Sanity-checking and returning");
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.
*/
template <typename TStr>
bool DifferenceCoverSample<TStr>::isCovered(TIndexOffU i) const {
assert(built());
uint32_t modi = this->modv(i);
assert_lt(modi, _dInv.size());
return _dInv[modi] != 0xffffffff;
}
/**
* Given a text offset that's covered, return its lexicographical rank
* among the sample suffixes.
*/
template <typename TStr>
TIndexOffU DifferenceCoverSample<TStr>::rank(TIndexOffU i) const {
assert(built());
assert_lt(i, this->text().length());
uint32_t imodv = this->modv(i);
assert_neq(0xffffffff, _dInv[imodv]); // must be in the sample
TIndexOffU ioff = this->divv(i);
assert_lt(ioff, _doffs[_dInv[imodv]+1] - _doffs[_dInv[imodv]]);
TIndexOffU isaIIdx = _doffs[_dInv[imodv]] + ioff;
assert_lt(isaIIdx, _isaPrime.size());
TIndexOffU isaPrimeI = _isaPrime[isaIIdx];
assert_leq(isaPrimeI, _isaPrime.size());
return isaPrimeI;
}
/**
* Return: < 0 if suffix i is lexicographically less than suffix j; > 0
* if suffix j is lexicographically greater.
*/
template <typename TStr>
int64_t DifferenceCoverSample<TStr>::breakTie(TIndexOffU i, TIndexOffU j) const {
assert(built());
assert_neq(i, j);
assert_lt(i, this->text().length());
assert_lt(j, this->text().length());
uint32_t imodv = this->modv(i);
uint32_t jmodv = this->modv(j);
assert_neq(0xffffffff, _dInv[imodv]); // must be in the sample
assert_neq(0xffffffff, _dInv[jmodv]); // must be in the sample
uint32_t dimodv = _dInv[imodv];
uint32_t djmodv = _dInv[jmodv];
TIndexOffU ioff = this->divv(i);
TIndexOffU joff = this->divv(j);
assert_lt(dimodv+1, _doffs.size());
assert_lt(djmodv+1, _doffs.size());
// assert_lt: expected (32024) < (0)
assert_lt(ioff, _doffs[dimodv+1] - _doffs[dimodv]);
assert_lt(joff, _doffs[djmodv+1] - _doffs[djmodv]);
TIndexOffU isaIIdx = _doffs[dimodv] + ioff;
TIndexOffU isaJIdx = _doffs[djmodv] + joff;
assert_lt(isaIIdx, _isaPrime.size());
assert_lt(isaJIdx, _isaPrime.size());
assert_neq(isaIIdx, isaJIdx); // ranks must be unique
TIndexOffU isaPrimeI = _isaPrime[isaIIdx];
TIndexOffU isaPrimeJ = _isaPrime[isaJIdx];
assert_neq(isaPrimeI, isaPrimeJ); // ranks must be unique
assert_leq(isaPrimeI, _isaPrime.size());
assert_leq(isaPrimeJ, _isaPrime.size());
return (int64_t)isaPrimeI - (int64_t)isaPrimeJ;
}
/**
* Given i, j, return the number of additional characters that need to
* be compared before the difference cover can break the tie.
*/
template <typename TStr>
uint32_t DifferenceCoverSample<TStr>::tieBreakOff(TIndexOffU i, TIndexOffU j) const {
const TStr& t = this->text();
const EList<uint32_t>& dmap = this->dmap();
assert(built());
// It's actually convenient to allow this, but we're permitted to
// return nonsense in that case
if(t[i] != t[j]) return 0xffffffff;
//assert_eq(t[i], t[j]); // if they're unequal, there's no tie to break
uint32_t v = this->v();
assert_neq(i, j);
assert_lt(i, t.length());
assert_lt(j, t.length());
uint32_t imod = this->modv(i);
uint32_t jmod = this->modv(j);
uint32_t diffLeft = (jmod >= imod)? (jmod - imod) : (jmod + v - imod);
uint32_t diffRight = (imod >= jmod)? (imod - jmod) : (imod + v - jmod);
assert_lt(diffLeft, dmap.size());
assert_lt(diffRight, dmap.size());
uint32_t destLeft = dmap[diffLeft]; // offset where i needs to be
uint32_t destRight = dmap[diffRight]; // offset where i needs to be
assert(isCovered(destLeft));
assert(isCovered(destLeft+diffLeft));
assert(isCovered(destRight));
assert(isCovered(destRight+diffRight));
assert_lt(destLeft, v);
assert_lt(destRight, v);
uint32_t deltaLeft = (destLeft >= imod)? (destLeft - imod) : (destLeft + v - imod);
if(deltaLeft == v) deltaLeft = 0;
uint32_t deltaRight = (destRight >= jmod)? (destRight - jmod) : (destRight + v - jmod);
if(deltaRight == v) deltaRight = 0;
assert_lt(deltaLeft, v);
assert_lt(deltaRight, v);
assert(isCovered(i+deltaLeft));
assert(isCovered(j+deltaLeft));
assert(isCovered(i+deltaRight));
assert(isCovered(j+deltaRight));
return min(deltaLeft, deltaRight);
}
#endif /*DIFF_SAMPLE_H_*/
|