1
2 /*-------------------------------------------------------------*/
3 /*--- Block sorting machinery ---*/
4 /*--- blocksort.c ---*/
5 /*-------------------------------------------------------------*/
6
7 /*--
8 This file is a part of bzip2 and/or libbzip2, a program and
9 library for lossless, block-sorting data compression.
10
11 Copyright (C) 1996-2000 Julian R Seward. All rights reserved.
12
13 Redistribution and use in source and binary forms, with or without
14 modification, are permitted provided that the following conditions
15 are met:
16
17 1. Redistributions of source code must retain the above copyright
18 notice, this list of conditions and the following disclaimer.
19
20 2. The origin of this software must not be misrepresented; you must
21 not claim that you wrote the original software. If you use this
22 software in a product, an acknowledgment in the product
23 documentation would be appreciated but is not required.
24
25 3. Altered source versions must be plainly marked as such, and must
26 not be misrepresented as being the original software.
27
28 4. The name of the author may not be used to endorse or promote
29 products derived from this software without specific prior written
30 permission.
31
32 THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
33 OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
34 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35 ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
36 DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37 DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
38 GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
39 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
40 WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
41 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
42 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
43
44 Julian Seward, Cambridge, UK.
45 jseward@acm.org
46 bzip2/libbzip2 version 1.0 of 21 March 2000
47
48 This program is based on (at least) the work of:
49 Mike Burrows
50 David Wheeler
51 Peter Fenwick
52 Alistair Moffat
53 Radford Neal
54 Ian H. Witten
55 Robert Sedgewick
56 Jon L. Bentley
57
58 For more information on these sources, see the manual.
59
60 To get some idea how the block sorting algorithms in this file
61 work, read my paper
62 On the Performance of BWT Sorting Algorithms
63 in Proceedings of the IEEE Data Compression Conference 2000,
64 Snowbird, Utah, USA, 27-30 March 2000. The main sort in this
65 file implements the algorithm called cache in the paper.
66 --*/
67
68 #include "os.h"
69 #include "bzlib.h"
70 #include "bzlib_private.h"
71
72 /*---------------------------------------------*/
73 /*--- Fallback O(N log(N)^2) sorting ---*/
74 /*--- algorithm, for repetitive blocks ---*/
75 /*---------------------------------------------*/
76
77 /*---------------------------------------------*/
78 static
79 __inline__
fallbackSimpleSort(UInt32 * fmap,UInt32 * eclass,Int32 lo,Int32 hi)80 void fallbackSimpleSort ( UInt32* fmap,
81 UInt32* eclass,
82 Int32 lo,
83 Int32 hi )
84 {
85 Int32 i, j, tmp;
86 UInt32 ec_tmp;
87
88 if (lo == hi) return;
89
90 if (hi - lo > 3) {
91 for ( i = hi-4; i >= lo; i-- ) {
92 tmp = fmap[i];
93 ec_tmp = eclass[tmp];
94 for ( j = i+4; j <= hi && ec_tmp > eclass[fmap[j]]; j += 4 )
95 fmap[j-4] = fmap[j];
96 fmap[j-4] = tmp;
97 }
98 }
99
100 for ( i = hi-1; i >= lo; i-- ) {
101 tmp = fmap[i];
102 ec_tmp = eclass[tmp];
103 for ( j = i+1; j <= hi && ec_tmp > eclass[fmap[j]]; j++ )
104 fmap[j-1] = fmap[j];
105 fmap[j-1] = tmp;
106 }
107 }
108
109
110 /*---------------------------------------------*/
111 #define fswap(zz1, zz2) \
112 { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
113
114 #define fvswap(zzp1, zzp2, zzn) \
115 { \
116 Int32 yyp1 = (zzp1); \
117 Int32 yyp2 = (zzp2); \
118 Int32 yyn = (zzn); \
119 while (yyn > 0) { \
120 fswap(fmap[yyp1], fmap[yyp2]); \
121 yyp1++; yyp2++; yyn--; \
122 } \
123 }
124
125
126 #define fmin(a,b) ((a) < (b)) ? (a) : (b)
127
128 #define fpush(lz,hz) { stackLo[sp] = lz; \
129 stackHi[sp] = hz; \
130 sp++; }
131
132 #define fpop(lz,hz) { sp--; \
133 lz = stackLo[sp]; \
134 hz = stackHi[sp]; }
135
136 #define FALLBACK_QSORT_SMALL_THRESH 10
137 #define FALLBACK_QSORT_STACK_SIZE 100
138
139
140 static
fallbackQSort3(UInt32 * fmap,UInt32 * eclass,Int32 loSt,Int32 hiSt)141 void fallbackQSort3 ( UInt32* fmap,
142 UInt32* eclass,
143 Int32 loSt,
144 Int32 hiSt )
145 {
146 Int32 unLo, unHi, ltLo, gtHi, n, m;
147 Int32 sp, lo, hi;
148 UInt32 med, r, r3;
149 Int32 stackLo[FALLBACK_QSORT_STACK_SIZE];
150 Int32 stackHi[FALLBACK_QSORT_STACK_SIZE];
151
152 r = 0;
153
154 sp = 0;
155 fpush ( loSt, hiSt );
156
157 while (sp > 0) {
158
159 AssertH ( sp < FALLBACK_QSORT_STACK_SIZE, 1004 );
160
161 fpop ( lo, hi );
162 if (hi - lo < FALLBACK_QSORT_SMALL_THRESH) {
163 fallbackSimpleSort ( fmap, eclass, lo, hi );
164 continue;
165 }
166
167 /* Random partitioning. Median of 3 sometimes fails to
168 avoid bad cases. Median of 9 seems to help but
169 looks rather expensive. This too seems to work but
170 is cheaper. Guidance for the magic constants
171 7621 and 32768 is taken from Sedgewick's algorithms
172 book, chapter 35.
173 */
174 r = ((r * 7621) + 1) % 32768;
175 r3 = r % 3;
176 if (r3 == 0) med = eclass[fmap[lo]]; else
177 if (r3 == 1) med = eclass[fmap[(lo+hi)>>1]]; else
178 med = eclass[fmap[hi]];
179
180 unLo = ltLo = lo;
181 unHi = gtHi = hi;
182
183 while (1) {
184 while (1) {
185 if (unLo > unHi) break;
186 n = (Int32)eclass[fmap[unLo]] - (Int32)med;
187 if (n == 0) {
188 fswap(fmap[unLo], fmap[ltLo]);
189 ltLo++; unLo++;
190 continue;
191 };
192 if (n > 0) break;
193 unLo++;
194 }
195 while (1) {
196 if (unLo > unHi) break;
197 n = (Int32)eclass[fmap[unHi]] - (Int32)med;
198 if (n == 0) {
199 fswap(fmap[unHi], fmap[gtHi]);
200 gtHi--; unHi--;
201 continue;
202 };
203 if (n < 0) break;
204 unHi--;
205 }
206 if (unLo > unHi) break;
207 fswap(fmap[unLo], fmap[unHi]); unLo++; unHi--;
208 }
209
210 AssertD ( unHi == unLo-1, "fallbackQSort3(2)" );
211
212 if (gtHi < ltLo) continue;
213
214 n = fmin(ltLo-lo, unLo-ltLo); fvswap(lo, unLo-n, n);
215 m = fmin(hi-gtHi, gtHi-unHi); fvswap(unLo, hi-m+1, m);
216
217 n = lo + unLo - ltLo - 1;
218 m = hi - (gtHi - unHi) + 1;
219
220 if (n - lo > hi - m) {
221 fpush ( lo, n );
222 fpush ( m, hi );
223 } else {
224 fpush ( m, hi );
225 fpush ( lo, n );
226 }
227 }
228 }
229
230 #undef fmin
231 #undef fpush
232 #undef fpop
233 #undef fswap
234 #undef fvswap
235 #undef FALLBACK_QSORT_SMALL_THRESH
236 #undef FALLBACK_QSORT_STACK_SIZE
237
238
239 /*---------------------------------------------*/
240 /* Pre:
241 nblock > 0
242 eclass exists for [0 .. nblock-1]
243 ((UChar*)eclass) [0 .. nblock-1] holds block
244 ptr exists for [0 .. nblock-1]
245
246 Post:
247 ((UChar*)eclass) [0 .. nblock-1] holds block
248 All other areas of eclass destroyed
249 fmap [0 .. nblock-1] holds sorted order
250 bhtab [ 0 .. 2+(nblock/32) ] destroyed
251 */
252
253 #define SET_BH(zz) bhtab[(zz) >> 5] |= (1 << ((zz) & 31))
254 #define CLEAR_BH(zz) bhtab[(zz) >> 5] &= ~(1 << ((zz) & 31))
255 #define ISSET_BH(zz) (bhtab[(zz) >> 5] & (1 << ((zz) & 31)))
256 #define WORD_BH(zz) bhtab[(zz) >> 5]
257 #define UNALIGNED_BH(zz) ((zz) & 0x01f)
258
259 static
fallbackSort(UInt32 * fmap,UInt32 * eclass,UInt32 * bhtab,Int32 nblock,Int32 verb)260 void fallbackSort ( UInt32* fmap,
261 UInt32* eclass,
262 UInt32* bhtab,
263 Int32 nblock,
264 Int32 verb )
265 {
266 Int32 ftab[257];
267 Int32 ftabCopy[256];
268 Int32 H, i, j, k, l, r, cc, cc1;
269 Int32 nNotDone;
270 Int32 nBhtab;
271 UChar* eclass8 = (UChar*)eclass;
272
273 /*--
274 Initial 1-char radix sort to generate
275 initial fmap and initial BH bits.
276 --*/
277 if (verb >= 4)
278 VPrintf0 ( " bucket sorting ...\n" );
279 for (i = 0; i < 257; i++) ftab[i] = 0;
280 for (i = 0; i < nblock; i++) ftab[eclass8[i]]++;
281 for (i = 0; i < 256; i++) ftabCopy[i] = ftab[i];
282 for (i = 1; i < 257; i++) ftab[i] += ftab[i-1];
283
284 for (i = 0; i < nblock; i++) {
285 j = eclass8[i];
286 k = ftab[j] - 1;
287 ftab[j] = k;
288 fmap[k] = i;
289 }
290
291 nBhtab = 2 + (nblock / 32);
292 for (i = 0; i < nBhtab; i++) bhtab[i] = 0;
293 for (i = 0; i < 256; i++) SET_BH(ftab[i]);
294
295 /*--
296 Inductively refine the buckets. Kind-of an
297 "exponential radix sort" (!), inspired by the
298 Manber-Myers suffix array construction algorithm.
299 --*/
300
301 /*-- set sentinel bits for block-end detection --*/
302 for (i = 0; i < 32; i++) {
303 SET_BH(nblock + 2*i);
304 CLEAR_BH(nblock + 2*i + 1);
305 }
306
307 /*-- the log(N) loop --*/
308 H = 1;
309 while (1) {
310
311 if (verb >= 4)
312 VPrintf1 ( " depth %6d has ", H );
313
314 j = 0;
315 for (i = 0; i < nblock; i++) {
316 if (ISSET_BH(i)) j = i;
317 k = fmap[i] - H; if (k < 0) k += nblock;
318 eclass[k] = j;
319 }
320
321 nNotDone = 0;
322 r = -1;
323 while (1) {
324
325 /*-- find the next non-singleton bucket --*/
326 k = r + 1;
327 while (ISSET_BH(k) && UNALIGNED_BH(k)) k++;
328 if (ISSET_BH(k)) {
329 while (WORD_BH(k) == 0xffffffff) k += 32;
330 while (ISSET_BH(k)) k++;
331 }
332 l = k - 1;
333 if (l >= nblock) break;
334 while (!ISSET_BH(k) && UNALIGNED_BH(k)) k++;
335 if (!ISSET_BH(k)) {
336 while (WORD_BH(k) == 0x00000000) k += 32;
337 while (!ISSET_BH(k)) k++;
338 }
339 r = k - 1;
340 if (r >= nblock) break;
341
342 /*-- now [l, r] bracket current bucket --*/
343 if (r > l) {
344 nNotDone += (r - l + 1);
345 fallbackQSort3 ( fmap, eclass, l, r );
346
347 /*-- scan bucket and generate header bits-- */
348 cc = -1;
349 for (i = l; i <= r; i++) {
350 cc1 = eclass[fmap[i]];
351 if (cc != cc1) { SET_BH(i); cc = cc1; };
352 }
353 }
354 }
355
356 if (verb >= 4)
357 VPrintf1 ( "%6d unresolved strings\n", nNotDone );
358
359 H *= 2;
360 if (H > nblock || nNotDone == 0) break;
361 }
362
363 /*--
364 Reconstruct the original block in
365 eclass8 [0 .. nblock-1], since the
366 previous phase destroyed it.
367 --*/
368 if (verb >= 4)
369 VPrintf0 ( " reconstructing block ...\n" );
370 j = 0;
371 for (i = 0; i < nblock; i++) {
372 while (ftabCopy[j] == 0) j++;
373 ftabCopy[j]--;
374 eclass8[fmap[i]] = (UChar)j;
375 }
376 AssertH ( j < 256, 1005 );
377 }
378
379 #undef SET_BH
380 #undef CLEAR_BH
381 #undef ISSET_BH
382 #undef WORD_BH
383 #undef UNALIGNED_BH
384
385
386 /*---------------------------------------------*/
387 /*--- The main, O(N^2 log(N)) sorting ---*/
388 /*--- algorithm. Faster for "normal" ---*/
389 /*--- non-repetitive blocks. ---*/
390 /*---------------------------------------------*/
391
392 /*---------------------------------------------*/
393 static
394 __inline__
mainGtU(UInt32 i1,UInt32 i2,UChar * block,UInt16 * quadrant,UInt32 nblock,Int32 * budget)395 Bool mainGtU ( UInt32 i1,
396 UInt32 i2,
397 UChar* block,
398 UInt16* quadrant,
399 UInt32 nblock,
400 Int32* budget )
401 {
402 Int32 k;
403 UChar c1, c2;
404 UInt16 s1, s2;
405
406 AssertD ( i1 != i2, "mainGtU" );
407 /* 1 */
408 c1 = block[i1]; c2 = block[i2];
409 if (c1 != c2) return (c1 > c2);
410 i1++; i2++;
411 /* 2 */
412 c1 = block[i1]; c2 = block[i2];
413 if (c1 != c2) return (c1 > c2);
414 i1++; i2++;
415 /* 3 */
416 c1 = block[i1]; c2 = block[i2];
417 if (c1 != c2) return (c1 > c2);
418 i1++; i2++;
419 /* 4 */
420 c1 = block[i1]; c2 = block[i2];
421 if (c1 != c2) return (c1 > c2);
422 i1++; i2++;
423 /* 5 */
424 c1 = block[i1]; c2 = block[i2];
425 if (c1 != c2) return (c1 > c2);
426 i1++; i2++;
427 /* 6 */
428 c1 = block[i1]; c2 = block[i2];
429 if (c1 != c2) return (c1 > c2);
430 i1++; i2++;
431 /* 7 */
432 c1 = block[i1]; c2 = block[i2];
433 if (c1 != c2) return (c1 > c2);
434 i1++; i2++;
435 /* 8 */
436 c1 = block[i1]; c2 = block[i2];
437 if (c1 != c2) return (c1 > c2);
438 i1++; i2++;
439 /* 9 */
440 c1 = block[i1]; c2 = block[i2];
441 if (c1 != c2) return (c1 > c2);
442 i1++; i2++;
443 /* 10 */
444 c1 = block[i1]; c2 = block[i2];
445 if (c1 != c2) return (c1 > c2);
446 i1++; i2++;
447 /* 11 */
448 c1 = block[i1]; c2 = block[i2];
449 if (c1 != c2) return (c1 > c2);
450 i1++; i2++;
451 /* 12 */
452 c1 = block[i1]; c2 = block[i2];
453 if (c1 != c2) return (c1 > c2);
454 i1++; i2++;
455
456 k = nblock + 8;
457
458 do {
459 /* 1 */
460 c1 = block[i1]; c2 = block[i2];
461 if (c1 != c2) return (c1 > c2);
462 s1 = quadrant[i1]; s2 = quadrant[i2];
463 if (s1 != s2) return (s1 > s2);
464 i1++; i2++;
465 /* 2 */
466 c1 = block[i1]; c2 = block[i2];
467 if (c1 != c2) return (c1 > c2);
468 s1 = quadrant[i1]; s2 = quadrant[i2];
469 if (s1 != s2) return (s1 > s2);
470 i1++; i2++;
471 /* 3 */
472 c1 = block[i1]; c2 = block[i2];
473 if (c1 != c2) return (c1 > c2);
474 s1 = quadrant[i1]; s2 = quadrant[i2];
475 if (s1 != s2) return (s1 > s2);
476 i1++; i2++;
477 /* 4 */
478 c1 = block[i1]; c2 = block[i2];
479 if (c1 != c2) return (c1 > c2);
480 s1 = quadrant[i1]; s2 = quadrant[i2];
481 if (s1 != s2) return (s1 > s2);
482 i1++; i2++;
483 /* 5 */
484 c1 = block[i1]; c2 = block[i2];
485 if (c1 != c2) return (c1 > c2);
486 s1 = quadrant[i1]; s2 = quadrant[i2];
487 if (s1 != s2) return (s1 > s2);
488 i1++; i2++;
489 /* 6 */
490 c1 = block[i1]; c2 = block[i2];
491 if (c1 != c2) return (c1 > c2);
492 s1 = quadrant[i1]; s2 = quadrant[i2];
493 if (s1 != s2) return (s1 > s2);
494 i1++; i2++;
495 /* 7 */
496 c1 = block[i1]; c2 = block[i2];
497 if (c1 != c2) return (c1 > c2);
498 s1 = quadrant[i1]; s2 = quadrant[i2];
499 if (s1 != s2) return (s1 > s2);
500 i1++; i2++;
501 /* 8 */
502 c1 = block[i1]; c2 = block[i2];
503 if (c1 != c2) return (c1 > c2);
504 s1 = quadrant[i1]; s2 = quadrant[i2];
505 if (s1 != s2) return (s1 > s2);
506 i1++; i2++;
507
508 if (i1 >= nblock) i1 -= nblock;
509 if (i2 >= nblock) i2 -= nblock;
510
511 k -= 8;
512 (*budget)--;
513 }
514 while (k >= 0);
515
516 return False;
517 }
518
519
520 /*---------------------------------------------*/
521 /*--
522 Knuth's increments seem to work better
523 than Incerpi-Sedgewick here. Possibly
524 because the number of elems to sort is
525 usually small, typically <= 20.
526 --*/
527 static
528 Int32 incs[14] = { 1, 4, 13, 40, 121, 364, 1093, 3280,
529 9841, 29524, 88573, 265720,
530 797161, 2391484 };
531
532 static
mainSimpleSort(UInt32 * ptr,UChar * block,UInt16 * quadrant,Int32 nblock,Int32 lo,Int32 hi,Int32 d,Int32 * budget)533 void mainSimpleSort ( UInt32* ptr,
534 UChar* block,
535 UInt16* quadrant,
536 Int32 nblock,
537 Int32 lo,
538 Int32 hi,
539 Int32 d,
540 Int32* budget )
541 {
542 Int32 i, j, h, bigN, hp;
543 UInt32 v;
544
545 bigN = hi - lo + 1;
546 if (bigN < 2) return;
547
548 hp = 0;
549 while (incs[hp] < bigN) hp++;
550 hp--;
551
552 for (; hp >= 0; hp--) {
553 h = incs[hp];
554
555 i = lo + h;
556 while (True) {
557
558 /*-- copy 1 --*/
559 if (i > hi) break;
560 v = ptr[i];
561 j = i;
562 while ( mainGtU (
563 ptr[j-h]+d, v+d, block, quadrant, nblock, budget
564 ) ) {
565 ptr[j] = ptr[j-h];
566 j = j - h;
567 if (j <= (lo + h - 1)) break;
568 }
569 ptr[j] = v;
570 i++;
571
572 /*-- copy 2 --*/
573 if (i > hi) break;
574 v = ptr[i];
575 j = i;
576 while ( mainGtU (
577 ptr[j-h]+d, v+d, block, quadrant, nblock, budget
578 ) ) {
579 ptr[j] = ptr[j-h];
580 j = j - h;
581 if (j <= (lo + h - 1)) break;
582 }
583 ptr[j] = v;
584 i++;
585
586 /*-- copy 3 --*/
587 if (i > hi) break;
588 v = ptr[i];
589 j = i;
590 while ( mainGtU (
591 ptr[j-h]+d, v+d, block, quadrant, nblock, budget
592 ) ) {
593 ptr[j] = ptr[j-h];
594 j = j - h;
595 if (j <= (lo + h - 1)) break;
596 }
597 ptr[j] = v;
598 i++;
599
600 if (*budget < 0) return;
601 }
602 }
603 }
604
605
606 /*---------------------------------------------*/
607 /*--
608 The following is an implementation of
609 an elegant 3-way quicksort for strings,
610 described in a paper "Fast Algorithms for
611 Sorting and Searching Strings", by Robert
612 Sedgewick and Jon L. Bentley.
613 --*/
614
615 #define mswap(zz1, zz2) \
616 { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
617
618 #define mvswap(zzp1, zzp2, zzn) \
619 { \
620 Int32 yyp1 = (zzp1); \
621 Int32 yyp2 = (zzp2); \
622 Int32 yyn = (zzn); \
623 while (yyn > 0) { \
624 mswap(ptr[yyp1], ptr[yyp2]); \
625 yyp1++; yyp2++; yyn--; \
626 } \
627 }
628
629 static
630 __inline__
mmed3(UChar a,UChar b,UChar c)631 UChar mmed3 ( UChar a, UChar b, UChar c )
632 {
633 UChar t;
634 if (a > b) { t = a; a = b; b = t; };
635 if (b > c) {
636 b = c;
637 if (a > b) b = a;
638 }
639 return b;
640 }
641
642 #define mmin(a,b) ((a) < (b)) ? (a) : (b)
643
644 #define mpush(lz,hz,dz) { stackLo[sp] = lz; \
645 stackHi[sp] = hz; \
646 stackD [sp] = dz; \
647 sp++; }
648
649 #define mpop(lz,hz,dz) { sp--; \
650 lz = stackLo[sp]; \
651 hz = stackHi[sp]; \
652 dz = stackD [sp]; }
653
654
655 #define mnextsize(az) (nextHi[az]-nextLo[az])
656
657 #define mnextswap(az,bz) \
658 { Int32 tz; \
659 tz = nextLo[az]; nextLo[az] = nextLo[bz]; nextLo[bz] = tz; \
660 tz = nextHi[az]; nextHi[az] = nextHi[bz]; nextHi[bz] = tz; \
661 tz = nextD [az]; nextD [az] = nextD [bz]; nextD [bz] = tz; }
662
663
664 #define MAIN_QSORT_SMALL_THRESH 20
665 #define MAIN_QSORT_DEPTH_THRESH (BZ_N_RADIX + BZ_N_QSORT)
666 #define MAIN_QSORT_STACK_SIZE 100
667
668 static
mainQSort3(UInt32 * ptr,UChar * block,UInt16 * quadrant,Int32 nblock,Int32 loSt,Int32 hiSt,Int32 dSt,Int32 * budget)669 void mainQSort3 ( UInt32* ptr,
670 UChar* block,
671 UInt16* quadrant,
672 Int32 nblock,
673 Int32 loSt,
674 Int32 hiSt,
675 Int32 dSt,
676 Int32* budget )
677 {
678 Int32 unLo, unHi, ltLo, gtHi, n, m, med;
679 Int32 sp, lo, hi, d;
680
681 Int32 stackLo[MAIN_QSORT_STACK_SIZE];
682 Int32 stackHi[MAIN_QSORT_STACK_SIZE];
683 Int32 stackD [MAIN_QSORT_STACK_SIZE];
684
685 Int32 nextLo[3];
686 Int32 nextHi[3];
687 Int32 nextD [3];
688
689 sp = 0;
690 mpush ( loSt, hiSt, dSt );
691
692 while (sp > 0) {
693
694 AssertH ( sp < MAIN_QSORT_STACK_SIZE, 1001 );
695
696 mpop ( lo, hi, d );
697 if (hi - lo < MAIN_QSORT_SMALL_THRESH ||
698 d > MAIN_QSORT_DEPTH_THRESH) {
699 mainSimpleSort ( ptr, block, quadrant, nblock, lo, hi, d, budget );
700 if (*budget < 0) return;
701 continue;
702 }
703
704 med = (Int32)
705 mmed3 ( block[ptr[ lo ]+d],
706 block[ptr[ hi ]+d],
707 block[ptr[ (lo+hi)>>1 ]+d] );
708
709 unLo = ltLo = lo;
710 unHi = gtHi = hi;
711
712 while (True) {
713 while (True) {
714 if (unLo > unHi) break;
715 n = ((Int32)block[ptr[unLo]+d]) - med;
716 if (n == 0) {
717 mswap(ptr[unLo], ptr[ltLo]);
718 ltLo++; unLo++; continue;
719 };
720 if (n > 0) break;
721 unLo++;
722 }
723 while (True) {
724 if (unLo > unHi) break;
725 n = ((Int32)block[ptr[unHi]+d]) - med;
726 if (n == 0) {
727 mswap(ptr[unHi], ptr[gtHi]);
728 gtHi--; unHi--; continue;
729 };
730 if (n < 0) break;
731 unHi--;
732 }
733 if (unLo > unHi) break;
734 mswap(ptr[unLo], ptr[unHi]); unLo++; unHi--;
735 }
736
737 AssertD ( unHi == unLo-1, "mainQSort3(2)" );
738
739 if (gtHi < ltLo) {
740 mpush(lo, hi, d+1 );
741 continue;
742 }
743
744 n = mmin(ltLo-lo, unLo-ltLo); mvswap(lo, unLo-n, n);
745 m = mmin(hi-gtHi, gtHi-unHi); mvswap(unLo, hi-m+1, m);
746
747 n = lo + unLo - ltLo - 1;
748 m = hi - (gtHi - unHi) + 1;
749
750 nextLo[0] = lo; nextHi[0] = n; nextD[0] = d;
751 nextLo[1] = m; nextHi[1] = hi; nextD[1] = d;
752 nextLo[2] = n+1; nextHi[2] = m-1; nextD[2] = d+1;
753
754 if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
755 if (mnextsize(1) < mnextsize(2)) mnextswap(1,2);
756 if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
757
758 AssertD (mnextsize(0) >= mnextsize(1), "mainQSort3(8)" );
759 AssertD (mnextsize(1) >= mnextsize(2), "mainQSort3(9)" );
760
761 mpush (nextLo[0], nextHi[0], nextD[0]);
762 mpush (nextLo[1], nextHi[1], nextD[1]);
763 mpush (nextLo[2], nextHi[2], nextD[2]);
764 }
765 }
766
767 #undef mswap
768 #undef mvswap
769 #undef mpush
770 #undef mpop
771 #undef mmin
772 #undef mnextsize
773 #undef mnextswap
774 #undef MAIN_QSORT_SMALL_THRESH
775 #undef MAIN_QSORT_DEPTH_THRESH
776 #undef MAIN_QSORT_STACK_SIZE
777
778
779 /*---------------------------------------------*/
780 /* Pre:
781 nblock > N_OVERSHOOT
782 block32 exists for [0 .. nblock-1 +N_OVERSHOOT]
783 ((UChar*)block32) [0 .. nblock-1] holds block
784 ptr exists for [0 .. nblock-1]
785
786 Post:
787 ((UChar*)block32) [0 .. nblock-1] holds block
788 All other areas of block32 destroyed
789 ftab [0 .. 65536 ] destroyed
790 ptr [0 .. nblock-1] holds sorted order
791 if (*budget < 0), sorting was abandoned
792 */
793
794 #define BIGFREQ(b) (ftab[((b)+1) << 8] - ftab[(b) << 8])
795 #define SETMASK (1 << 21)
796 #define CLEARMASK (~(SETMASK))
797
798 static
mainSort(UInt32 * ptr,UChar * block,UInt16 * quadrant,UInt32 * ftab,Int32 nblock,Int32 verb,Int32 * budget)799 void mainSort ( UInt32* ptr,
800 UChar* block,
801 UInt16* quadrant,
802 UInt32* ftab,
803 Int32 nblock,
804 Int32 verb,
805 Int32* budget )
806 {
807 Int32 i, j, k, ss, sb;
808 Int32 runningOrder[256];
809 Bool bigDone[256];
810 Int32 copyStart[256];
811 Int32 copyEnd [256];
812 UChar c1;
813 Int32 numQSorted;
814 UInt16 s;
815 if (verb >= 4) VPrintf0 ( " main sort initialise ...\n" );
816
817 /*-- set up the 2-byte frequency table --*/
818 for (i = 65536; i >= 0; i--) ftab[i] = 0;
819
820 j = block[0] << 8;
821 i = nblock-1;
822 for (; i >= 3; i -= 4) {
823 quadrant[i] = 0;
824 j = (j >> 8) | ( ((UInt16)block[i]) << 8);
825 ftab[j]++;
826 quadrant[i-1] = 0;
827 j = (j >> 8) | ( ((UInt16)block[i-1]) << 8);
828 ftab[j]++;
829 quadrant[i-2] = 0;
830 j = (j >> 8) | ( ((UInt16)block[i-2]) << 8);
831 ftab[j]++;
832 quadrant[i-3] = 0;
833 j = (j >> 8) | ( ((UInt16)block[i-3]) << 8);
834 ftab[j]++;
835 }
836 for (; i >= 0; i--) {
837 quadrant[i] = 0;
838 j = (j >> 8) | ( ((UInt16)block[i]) << 8);
839 ftab[j]++;
840 }
841
842 /*-- (emphasises close relationship of block & quadrant) --*/
843 for (i = 0; i < BZ_N_OVERSHOOT; i++) {
844 block [nblock+i] = block[i];
845 quadrant[nblock+i] = 0;
846 }
847
848 if (verb >= 4) VPrintf0 ( " bucket sorting ...\n" );
849
850 /*-- Complete the initial radix sort --*/
851 for (i = 1; i <= 65536; i++) ftab[i] += ftab[i-1];
852
853 s = block[0] << 8;
854 i = nblock-1;
855 for (; i >= 3; i -= 4) {
856 s = (s >> 8) | (block[i] << 8);
857 j = ftab[s] -1;
858 ftab[s] = j;
859 ptr[j] = i;
860 s = (s >> 8) | (block[i-1] << 8);
861 j = ftab[s] -1;
862 ftab[s] = j;
863 ptr[j] = i-1;
864 s = (s >> 8) | (block[i-2] << 8);
865 j = ftab[s] -1;
866 ftab[s] = j;
867 ptr[j] = i-2;
868 s = (s >> 8) | (block[i-3] << 8);
869 j = ftab[s] -1;
870 ftab[s] = j;
871 ptr[j] = i-3;
872 }
873 for (; i >= 0; i--) {
874 s = (s >> 8) | (block[i] << 8);
875 j = ftab[s] -1;
876 ftab[s] = j;
877 ptr[j] = i;
878 }
879
880 /*--
881 Now ftab contains the first loc of every small bucket.
882 Calculate the running order, from smallest to largest
883 big bucket.
884 --*/
885 for (i = 0; i <= 255; i++) {
886 bigDone [i] = False;
887 runningOrder[i] = i;
888 }
889
890 {
891 Int32 vv;
892 Int32 h = 1;
893 do h = 3 * h + 1; while (h <= 256);
894 do {
895 h = h / 3;
896 for (i = h; i <= 255; i++) {
897 vv = runningOrder[i];
898 j = i;
899 while ( BIGFREQ(runningOrder[j-h]) > BIGFREQ(vv) ) {
900 runningOrder[j] = runningOrder[j-h];
901 j = j - h;
902 if (j <= (h - 1)) goto zero;
903 }
904 zero:
905 runningOrder[j] = vv;
906 }
907 } while (h != 1);
908 }
909
910 /*--
911 The main sorting loop.
912 --*/
913
914 numQSorted = 0;
915
916 for (i = 0; i <= 255; i++) {
917
918 /*--
919 Process big buckets, starting with the least full.
920 Basically this is a 3-step process in which we call
921 mainQSort3 to sort the small buckets [ss, j], but
922 also make a big effort to avoid the calls if we can.
923 --*/
924 ss = runningOrder[i];
925
926 /*--
927 Step 1:
928 Complete the big bucket [ss] by quicksorting
929 any unsorted small buckets [ss, j], for j != ss.
930 Hopefully previous pointer-scanning phases have already
931 completed many of the small buckets [ss, j], so
932 we don't have to sort them at all.
933 --*/
934 for (j = 0; j <= 255; j++) {
935 if (j != ss) {
936 sb = (ss << 8) + j;
937 if ( ! (ftab[sb] & SETMASK) ) {
938 Int32 lo = ftab[sb] & CLEARMASK;
939 Int32 hi = (ftab[sb+1] & CLEARMASK) - 1;
940 if (hi > lo) {
941 if (verb >= 4)
942 VPrintf4 ( " qsort [0x%x, 0x%x] "
943 "done %d this %d\n",
944 ss, j, numQSorted, hi - lo + 1 );
945 mainQSort3 (
946 ptr, block, quadrant, nblock,
947 lo, hi, BZ_N_RADIX, budget
948 );
949 numQSorted += (hi - lo + 1);
950 if (*budget < 0) return;
951 }
952 }
953 ftab[sb] |= SETMASK;
954 }
955 }
956
957 AssertH ( !bigDone[ss], 1006 );
958
959 /*--
960 Step 2:
961 Now scan this big bucket [ss] so as to synthesise the
962 sorted order for small buckets [t, ss] for all t,
963 including, magically, the bucket [ss,ss] too.
964 This will avoid doing Real Work in subsequent Step 1's.
965 --*/
966 {
967 for (j = 0; j <= 255; j++) {
968 copyStart[j] = ftab[(j << 8) + ss] & CLEARMASK;
969 copyEnd [j] = (ftab[(j << 8) + ss + 1] & CLEARMASK) - 1;
970 }
971 for (j = ftab[ss << 8] & CLEARMASK; j < copyStart[ss]; j++) {
972 k = ptr[j]-1; if (k < 0) k += nblock;
973 c1 = block[k];
974 if (!bigDone[c1])
975 ptr[ copyStart[c1]++ ] = k;
976 }
977 for (j = (ftab[(ss+1) << 8] & CLEARMASK) - 1; j > copyEnd[ss]; j--) {
978 k = ptr[j]-1; if (k < 0) k += nblock;
979 c1 = block[k];
980 if (!bigDone[c1])
981 ptr[ copyEnd[c1]-- ] = k;
982 }
983 }
984
985 AssertH ( copyStart[ss]-1 == copyEnd[ss], 1007 );
986
987 for (j = 0; j <= 255; j++) ftab[(j << 8) + ss] |= SETMASK;
988
989 /*--
990 Step 3:
991 The [ss] big bucket is now done. Record this fact,
992 and update the quadrant descriptors. Remember to
993 update quadrants in the overshoot area too, if
994 necessary. The "if (i < 255)" test merely skips
995 this updating for the last bucket processed, since
996 updating for the last bucket is pointless.
997
998 The quadrant array provides a way to incrementally
999 cache sort orderings, as they appear, so as to
1000 make subsequent comparisons in fullGtU() complete
1001 faster. For repetitive blocks this makes a big
1002 difference (but not big enough to be able to avoid
1003 the fallback sorting mechanism, exponential radix sort).
1004
1005 The precise meaning is: at all times:
1006
1007 for 0 <= i < nblock and 0 <= j <= nblock
1008
1009 if block[i] != block[j],
1010
1011 then the relative values of quadrant[i] and
1012 quadrant[j] are meaningless.
1013
1014 else {
1015 if quadrant[i] < quadrant[j]
1016 then the string starting at i lexicographically
1017 precedes the string starting at j
1018
1019 else if quadrant[i] > quadrant[j]
1020 then the string starting at j lexicographically
1021 precedes the string starting at i
1022
1023 else
1024 the relative ordering of the strings starting
1025 at i and j has not yet been determined.
1026 }
1027 --*/
1028 bigDone[ss] = True;
1029
1030 if (i < 255) {
1031 Int32 bbStart = ftab[ss << 8] & CLEARMASK;
1032 Int32 bbSize = (ftab[(ss+1) << 8] & CLEARMASK) - bbStart;
1033 Int32 shifts = 0;
1034
1035 while ((bbSize >> shifts) > 65534) shifts++;
1036
1037 for (j = bbSize-1; j >= 0; j--) {
1038 Int32 a2update = ptr[bbStart + j];
1039 UInt16 qVal = (UInt16)(j >> shifts);
1040 quadrant[a2update] = qVal;
1041 if (a2update < BZ_N_OVERSHOOT)
1042 quadrant[a2update + nblock] = qVal;
1043 }
1044 AssertH ( ((bbSize-1) >> shifts) <= 65535, 1002 );
1045 }
1046
1047 }
1048
1049 if (verb >= 4)
1050 VPrintf3 ( " %d pointers, %d sorted, %d scanned\n",
1051 nblock, numQSorted, nblock - numQSorted );
1052 }
1053
1054 #undef BIGFREQ
1055 #undef SETMASK
1056 #undef CLEARMASK
1057
1058
1059 /*---------------------------------------------*/
1060 /* Pre:
1061 nblock > 0
1062 arr2 exists for [0 .. nblock-1 +N_OVERSHOOT]
1063 ((UChar*)arr2) [0 .. nblock-1] holds block
1064 arr1 exists for [0 .. nblock-1]
1065
1066 Post:
1067 ((UChar*)arr2) [0 .. nblock-1] holds block
1068 All other areas of block destroyed
1069 ftab [ 0 .. 65536 ] destroyed
1070 arr1 [0 .. nblock-1] holds sorted order
1071 */
BZ2_blockSort(EState * s)1072 void BZ2_blockSort ( EState* s )
1073 {
1074 UInt32* ptr = s->ptr;
1075 UChar* block = s->block;
1076 UInt32* ftab = s->ftab;
1077 Int32 nblock = s->nblock;
1078 Int32 verb = s->verbosity;
1079 Int32 wfact = s->workFactor;
1080 UInt16* quadrant;
1081 Int32 budget;
1082 Int32 budgetInit;
1083 Int32 i;
1084
1085 if (nblock < 10000) {
1086 fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
1087 } else {
1088 /* Calculate the location for quadrant, remembering to get
1089 the alignment right. Assumes that &(block[0]) is at least
1090 2-byte aligned -- this should be ok since block is really
1091 the first section of arr2.
1092 */
1093 i = nblock+BZ_N_OVERSHOOT;
1094 if (i & 1) i++;
1095 quadrant = (UInt16*)(&(block[i]));
1096
1097 /* (wfact-1) / 3 puts the default-factor-30
1098 transition point at very roughly the same place as
1099 with v0.1 and v0.9.0.
1100 Not that it particularly matters any more, since the
1101 resulting compressed stream is now the same regardless
1102 of whether or not we use the main sort or fallback sort.
1103 */
1104 if (wfact < 1 ) wfact = 1;
1105 if (wfact > 100) wfact = 100;
1106 budgetInit = nblock * ((wfact-1) / 3);
1107 budget = budgetInit;
1108
1109 mainSort ( ptr, block, quadrant, ftab, nblock, verb, &budget );
1110 if (verb >= 3)
1111 VPrintf3 ( " %d work, %d block, ratio %5.2f\n",
1112 budgetInit - budget,
1113 nblock,
1114 (float)(budgetInit - budget) /
1115 (float)(nblock==0 ? 1 : nblock) );
1116 if (budget < 0) {
1117 if (verb >= 2)
1118 VPrintf0 ( " too repetitive; using fallback"
1119 " sorting algorithm\n" );
1120 fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
1121 }
1122 }
1123
1124 s->origPtr = -1;
1125 for (i = 0; i < s->nblock; i++)
1126 if (ptr[i] == 0)
1127 { s->origPtr = i; break; };
1128
1129 AssertH( s->origPtr != -1, 1003 );
1130 }
1131
1132
1133 /*-------------------------------------------------------------*/
1134 /*--- end blocksort.c ---*/
1135 /*-------------------------------------------------------------*/
1136