aboutsummaryrefslogtreecommitdiff
path: root/src/cmd/bzip2/lib/blocksort.c
blob: f43d569e6da61eca1add14dd1eb8cfe4d9284467 (plain)
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

/*-------------------------------------------------------------*/
/*--- Block sorting machinery                               ---*/
/*---                                           blocksort.c ---*/
/*-------------------------------------------------------------*/

/*--
  This file is a part of bzip2 and/or libbzip2, a program and
  library for lossless, block-sorting data compression.

  Copyright (C) 1996-2000 Julian R Seward.  All rights reserved.

  Redistribution and use in source and binary forms, with or without
  modification, are permitted provided that the following conditions
  are met:

  1. Redistributions of source code must retain the above copyright
     notice, this list of conditions and the following disclaimer.

  2. The origin of this software must not be misrepresented; you must 
     not claim that you wrote the original software.  If you use this 
     software in a product, an acknowledgment in the product 
     documentation would be appreciated but is not required.

  3. Altered source versions must be plainly marked as such, and must
     not be misrepresented as being the original software.

  4. The name of the author may not be used to endorse or promote 
     products derived from this software without specific prior written 
     permission.

  THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
  OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
  DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
  GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

  Julian Seward, Cambridge, UK.
  jseward@acm.org
  bzip2/libbzip2 version 1.0 of 21 March 2000

  This program is based on (at least) the work of:
     Mike Burrows
     David Wheeler
     Peter Fenwick
     Alistair Moffat
     Radford Neal
     Ian H. Witten
     Robert Sedgewick
     Jon L. Bentley

  For more information on these sources, see the manual.

  To get some idea how the block sorting algorithms in this file 
  work, read my paper 
     On the Performance of BWT Sorting Algorithms
  in Proceedings of the IEEE Data Compression Conference 2000,
  Snowbird, Utah, USA, 27-30 March 2000.  The main sort in this
  file implements the algorithm called  cache  in the paper.
--*/

#include "os.h"
#include "bzlib.h"
#include "bzlib_private.h"

/*---------------------------------------------*/
/*--- Fallback O(N log(N)^2) sorting        ---*/
/*--- algorithm, for repetitive blocks      ---*/
/*---------------------------------------------*/

/*---------------------------------------------*/
static 
__inline__
void fallbackSimpleSort ( UInt32* fmap, 
                          UInt32* eclass, 
                          Int32   lo, 
                          Int32   hi )
{
   Int32 i, j, tmp;
   UInt32 ec_tmp;

   if (lo == hi) return;

   if (hi - lo > 3) {
      for ( i = hi-4; i >= lo; i-- ) {
         tmp = fmap[i];
         ec_tmp = eclass[tmp];
         for ( j = i+4; j <= hi && ec_tmp > eclass[fmap[j]]; j += 4 )
            fmap[j-4] = fmap[j];
         fmap[j-4] = tmp;
      }
   }

   for ( i = hi-1; i >= lo; i-- ) {
      tmp = fmap[i];
      ec_tmp = eclass[tmp];
      for ( j = i+1; j <= hi && ec_tmp > eclass[fmap[j]]; j++ )
         fmap[j-1] = fmap[j];
      fmap[j-1] = tmp;
   }
}


/*---------------------------------------------*/
#define fswap(zz1, zz2) \
   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }

#define fvswap(zzp1, zzp2, zzn)       \
{                                     \
   Int32 yyp1 = (zzp1);               \
   Int32 yyp2 = (zzp2);               \
   Int32 yyn  = (zzn);                \
   while (yyn > 0) {                  \
      fswap(fmap[yyp1], fmap[yyp2]);  \
      yyp1++; yyp2++; yyn--;          \
   }                                  \
}


#define fmin(a,b) ((a) < (b)) ? (a) : (b)

#define fpush(lz,hz) { stackLo[sp] = lz; \
                       stackHi[sp] = hz; \
                       sp++; }

#define fpop(lz,hz) { sp--;              \
                      lz = stackLo[sp];  \
                      hz = stackHi[sp]; }

#define FALLBACK_QSORT_SMALL_THRESH 10
#define FALLBACK_QSORT_STACK_SIZE   100


static
void fallbackQSort3 ( UInt32* fmap, 
                      UInt32* eclass,
                      Int32   loSt, 
                      Int32   hiSt )
{
   Int32 unLo, unHi, ltLo, gtHi, n, m;
   Int32 sp, lo, hi;
   UInt32 med, r, r3;
   Int32 stackLo[FALLBACK_QSORT_STACK_SIZE];
   Int32 stackHi[FALLBACK_QSORT_STACK_SIZE];

   r = 0;

   sp = 0;
   fpush ( loSt, hiSt );

   while (sp > 0) {

      AssertH ( sp < FALLBACK_QSORT_STACK_SIZE, 1004 );

      fpop ( lo, hi );
      if (hi - lo < FALLBACK_QSORT_SMALL_THRESH) {
         fallbackSimpleSort ( fmap, eclass, lo, hi );
         continue;
      }

      /* Random partitioning.  Median of 3 sometimes fails to
         avoid bad cases.  Median of 9 seems to help but 
         looks rather expensive.  This too seems to work but
         is cheaper.  Guidance for the magic constants 
         7621 and 32768 is taken from Sedgewick's algorithms
         book, chapter 35.
      */
      r = ((r * 7621) + 1) % 32768;
      r3 = r % 3;
      if (r3 == 0) med = eclass[fmap[lo]]; else
      if (r3 == 1) med = eclass[fmap[(lo+hi)>>1]]; else
                   med = eclass[fmap[hi]];

      unLo = ltLo = lo;
      unHi = gtHi = hi;

      while (1) {
         while (1) {
            if (unLo > unHi) break;
            n = (Int32)eclass[fmap[unLo]] - (Int32)med;
            if (n == 0) { 
               fswap(fmap[unLo], fmap[ltLo]); 
               ltLo++; unLo++; 
               continue; 
            };
            if (n > 0) break;
            unLo++;
         }
         while (1) {
            if (unLo > unHi) break;
            n = (Int32)eclass[fmap[unHi]] - (Int32)med;
            if (n == 0) { 
               fswap(fmap[unHi], fmap[gtHi]); 
               gtHi--; unHi--; 
               continue; 
            };
            if (n < 0) break;
            unHi--;
         }
         if (unLo > unHi) break;
         fswap(fmap[unLo], fmap[unHi]); unLo++; unHi--;
      }

      AssertD ( unHi == unLo-1, "fallbackQSort3(2)" );

      if (gtHi < ltLo) continue;

      n = fmin(ltLo-lo, unLo-ltLo); fvswap(lo, unLo-n, n);
      m = fmin(hi-gtHi, gtHi-unHi); fvswap(unLo, hi-m+1, m);

      n = lo + unLo - ltLo - 1;
      m = hi - (gtHi - unHi) + 1;

      if (n - lo > hi - m) {
         fpush ( lo, n );
         fpush ( m, hi );
      } else {
         fpush ( m, hi );
         fpush ( lo, n );
      }
   }
}

#undef fmin
#undef fpush
#undef fpop
#undef fswap
#undef fvswap
#undef FALLBACK_QSORT_SMALL_THRESH
#undef FALLBACK_QSORT_STACK_SIZE


/*---------------------------------------------*/
/* Pre:
      nblock > 0
      eclass exists for [0 .. nblock-1]
      ((UChar*)eclass) [0 .. nblock-1] holds block
      ptr exists for [0 .. nblock-1]

   Post:
      ((UChar*)eclass) [0 .. nblock-1] holds block
      All other areas of eclass destroyed
      fmap [0 .. nblock-1] holds sorted order
      bhtab [ 0 .. 2+(nblock/32) ] destroyed
*/

#define       SET_BH(zz)  bhtab[(zz) >> 5] |= (1 << ((zz) & 31))
#define     CLEAR_BH(zz)  bhtab[(zz) >> 5] &= ~(1 << ((zz) & 31))
#define     ISSET_BH(zz)  (bhtab[(zz) >> 5] & (1 << ((zz) & 31)))
#define      WORD_BH(zz)  bhtab[(zz) >> 5]
#define UNALIGNED_BH(zz)  ((zz) & 0x01f)

static
void fallbackSort ( UInt32* fmap, 
                    UInt32* eclass, 
                    UInt32* bhtab,
                    Int32   nblock,
                    Int32   verb )
{
   Int32 ftab[257];
   Int32 ftabCopy[256];
   Int32 H, i, j, k, l, r, cc, cc1;
   Int32 nNotDone;
   Int32 nBhtab;
   UChar* eclass8 = (UChar*)eclass;

   /*--
      Initial 1-char radix sort to generate
      initial fmap and initial BH bits.
   --*/
   if (verb >= 4)
      VPrintf0 ( "        bucket sorting ...\n" );
   for (i = 0; i < 257;    i++) ftab[i] = 0;
   for (i = 0; i < nblock; i++) ftab[eclass8[i]]++;
   for (i = 0; i < 256;    i++) ftabCopy[i] = ftab[i];
   for (i = 1; i < 257;    i++) ftab[i] += ftab[i-1];

   for (i = 0; i < nblock; i++) {
      j = eclass8[i];
      k = ftab[j] - 1;
      ftab[j] = k;
      fmap[k] = i;
   }

   nBhtab = 2 + (nblock / 32);
   for (i = 0; i < nBhtab; i++) bhtab[i] = 0;
   for (i = 0; i < 256; i++) SET_BH(ftab[i]);

   /*--
      Inductively refine the buckets.  Kind-of an
      "exponential radix sort" (!), inspired by the
      Manber-Myers suffix array construction algorithm.
   --*/

   /*-- set sentinel bits for block-end detection --*/
   for (i = 0; i < 32; i++) { 
      SET_BH(nblock + 2*i);
      CLEAR_BH(nblock + 2*i + 1);
   }

   /*-- the log(N) loop --*/
   H = 1;
   while (1) {

      if (verb >= 4) 
         VPrintf1 ( "        depth %6d has ", H );

      j = 0;
      for (i = 0; i < nblock; i++) {
         if (ISSET_BH(i)) j = i;
         k = fmap[i] - H; if (k < 0) k += nblock;
         eclass[k] = j;
      }

      nNotDone = 0;
      r = -1;
      while (1) {

	 /*-- find the next non-singleton bucket --*/
         k = r + 1;
         while (ISSET_BH(k) && UNALIGNED_BH(k)) k++;
         if (ISSET_BH(k)) {
            while (WORD_BH(k) == 0xffffffff) k += 32;
            while (ISSET_BH(k)) k++;
         }
         l = k - 1;
         if (l >= nblock) break;
         while (!ISSET_BH(k) && UNALIGNED_BH(k)) k++;
         if (!ISSET_BH(k)) {
            while (WORD_BH(k) == 0x00000000) k += 32;
            while (!ISSET_BH(k)) k++;
         }
         r = k - 1;
         if (r >= nblock) break;

         /*-- now [l, r] bracket current bucket --*/
         if (r > l) {
            nNotDone += (r - l + 1);
            fallbackQSort3 ( fmap, eclass, l, r );

            /*-- scan bucket and generate header bits-- */
            cc = -1;
            for (i = l; i <= r; i++) {
               cc1 = eclass[fmap[i]];
               if (cc != cc1) { SET_BH(i); cc = cc1; };
            }
         }
      }

      if (verb >= 4) 
         VPrintf1 ( "%6d unresolved strings\n", nNotDone );

      H *= 2;
      if (H > nblock || nNotDone == 0) break;
   }

   /*-- 
      Reconstruct the original block in
      eclass8 [0 .. nblock-1], since the
      previous phase destroyed it.
   --*/
   if (verb >= 4)
      VPrintf0 ( "        reconstructing block ...\n" );
   j = 0;
   for (i = 0; i < nblock; i++) {
      while (ftabCopy[j] == 0) j++;
      ftabCopy[j]--;
      eclass8[fmap[i]] = (UChar)j;
   }
   AssertH ( j < 256, 1005 );
}

#undef       SET_BH
#undef     CLEAR_BH
#undef     ISSET_BH
#undef      WORD_BH
#undef UNALIGNED_BH


/*---------------------------------------------*/
/*--- The main, O(N^2 log(N)) sorting       ---*/
/*--- algorithm.  Faster for "normal"       ---*/
/*--- non-repetitive blocks.                ---*/
/*---------------------------------------------*/

/*---------------------------------------------*/
static
__inline__
Bool mainGtU ( UInt32  i1, 
               UInt32  i2,
               UChar*  block, 
               UInt16* quadrant,
               UInt32  nblock,
               Int32*  budget )
{
   Int32  k;
   UChar  c1, c2;
   UInt16 s1, s2;

   AssertD ( i1 != i2, "mainGtU" );
   /* 1 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 2 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 3 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 4 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 5 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 6 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 7 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 8 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 9 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 10 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 11 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;
   /* 12 */
   c1 = block[i1]; c2 = block[i2];
   if (c1 != c2) return (c1 > c2);
   i1++; i2++;

   k = nblock + 8;

   do {
      /* 1 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 2 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 3 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 4 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 5 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 6 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 7 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;
      /* 8 */
      c1 = block[i1]; c2 = block[i2];
      if (c1 != c2) return (c1 > c2);
      s1 = quadrant[i1]; s2 = quadrant[i2];
      if (s1 != s2) return (s1 > s2);
      i1++; i2++;

      if (i1 >= nblock) i1 -= nblock;
      if (i2 >= nblock) i2 -= nblock;

      k -= 8;
      (*budget)--;
   }
      while (k >= 0);

   return False;
}


/*---------------------------------------------*/
/*--
   Knuth's increments seem to work better
   than Incerpi-Sedgewick here.  Possibly
   because the number of elems to sort is
   usually small, typically <= 20.
--*/
static
Int32 incs[14] = { 1, 4, 13, 40, 121, 364, 1093, 3280,
                   9841, 29524, 88573, 265720,
                   797161, 2391484 };

static
void mainSimpleSort ( UInt32* ptr,
                      UChar*  block,
                      UInt16* quadrant,
                      Int32   nblock,
                      Int32   lo, 
                      Int32   hi, 
                      Int32   d,
                      Int32*  budget )
{
   Int32 i, j, h, bigN, hp;
   UInt32 v;

   bigN = hi - lo + 1;
   if (bigN < 2) return;

   hp = 0;
   while (incs[hp] < bigN) hp++;
   hp--;

   for (; hp >= 0; hp--) {
      h = incs[hp];

      i = lo + h;
      while (True) {

         /*-- copy 1 --*/
         if (i > hi) break;
         v = ptr[i];
         j = i;
         while ( mainGtU ( 
                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
                 ) ) {
            ptr[j] = ptr[j-h];
            j = j - h;
            if (j <= (lo + h - 1)) break;
         }
         ptr[j] = v;
         i++;

         /*-- copy 2 --*/
         if (i > hi) break;
         v = ptr[i];
         j = i;
         while ( mainGtU ( 
                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
                 ) ) {
            ptr[j] = ptr[j-h];
            j = j - h;
            if (j <= (lo + h - 1)) break;
         }
         ptr[j] = v;
         i++;

         /*-- copy 3 --*/
         if (i > hi) break;
         v = ptr[i];
         j = i;
         while ( mainGtU ( 
                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
                 ) ) {
            ptr[j] = ptr[j-h];
            j = j - h;
            if (j <= (lo + h - 1)) break;
         }
         ptr[j] = v;
         i++;

         if (*budget < 0) return;
      }
   }
}


/*---------------------------------------------*/
/*--
   The following is an implementation of
   an elegant 3-way quicksort for strings,
   described in a paper "Fast Algorithms for
   Sorting and Searching Strings", by Robert
   Sedgewick and Jon L. Bentley.
--*/

#define mswap(zz1, zz2) \
   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }

#define mvswap(zzp1, zzp2, zzn)       \
{                                     \
   Int32 yyp1 = (zzp1);               \
   Int32 yyp2 = (zzp2);               \
   Int32 yyn  = (zzn);                \
   while (yyn > 0) {                  \
      mswap(ptr[yyp1], ptr[yyp2]);    \
      yyp1++; yyp2++; yyn--;          \
   }                                  \
}

static 
__inline__
UChar mmed3 ( UChar a, UChar b, UChar c )
{
   UChar t;
   if (a > b) { t = a; a = b; b = t; };
   if (b > c) { 
      b = c;
      if (a > b) b = a;
   }
   return b;
}

#define mmin(a,b) ((a) < (b)) ? (a) : (b)

#define mpush(lz,hz,dz) { stackLo[sp] = lz; \
                          stackHi[sp] = hz; \
                          stackD [sp] = dz; \
                          sp++; }

#define mpop(lz,hz,dz) { sp--;             \
                         lz = stackLo[sp]; \
                         hz = stackHi[sp]; \
                         dz = stackD [sp]; }


#define mnextsize(az) (nextHi[az]-nextLo[az])

#define mnextswap(az,bz)                                        \
   { Int32 tz;                                                  \
     tz = nextLo[az]; nextLo[az] = nextLo[bz]; nextLo[bz] = tz; \
     tz = nextHi[az]; nextHi[az] = nextHi[bz]; nextHi[bz] = tz; \
     tz = nextD [az]; nextD [az] = nextD [bz]; nextD [bz] = tz; }


#define MAIN_QSORT_SMALL_THRESH 20
#define MAIN_QSORT_DEPTH_THRESH (BZ_N_RADIX + BZ_N_QSORT)
#define MAIN_QSORT_STACK_SIZE 100

static
void mainQSort3 ( UInt32* ptr,
                  UChar*  block,
                  UInt16* quadrant,
                  Int32   nblock,
                  Int32   loSt, 
                  Int32   hiSt, 
                  Int32   dSt,
                  Int32*  budget )
{
   Int32 unLo, unHi, ltLo, gtHi, n, m, med;
   Int32 sp, lo, hi, d;

   Int32 stackLo[MAIN_QSORT_STACK_SIZE];
   Int32 stackHi[MAIN_QSORT_STACK_SIZE];
   Int32 stackD [MAIN_QSORT_STACK_SIZE];

   Int32 nextLo[3];
   Int32 nextHi[3];
   Int32 nextD [3];

   sp = 0;
   mpush ( loSt, hiSt, dSt );

   while (sp > 0) {

      AssertH ( sp < MAIN_QSORT_STACK_SIZE, 1001 );

      mpop ( lo, hi, d );
      if (hi - lo < MAIN_QSORT_SMALL_THRESH || 
          d > MAIN_QSORT_DEPTH_THRESH) {
         mainSimpleSort ( ptr, block, quadrant, nblock, lo, hi, d, budget );
         if (*budget < 0) return;
         continue;
      }

      med = (Int32) 
            mmed3 ( block[ptr[ lo         ]+d],
                    block[ptr[ hi         ]+d],
                    block[ptr[ (lo+hi)>>1 ]+d] );

      unLo = ltLo = lo;
      unHi = gtHi = hi;

      while (True) {
         while (True) {
            if (unLo > unHi) break;
            n = ((Int32)block[ptr[unLo]+d]) - med;
            if (n == 0) { 
               mswap(ptr[unLo], ptr[ltLo]); 
               ltLo++; unLo++; continue; 
            };
            if (n >  0) break;
            unLo++;
         }
         while (True) {
            if (unLo > unHi) break;
            n = ((Int32)block[ptr[unHi]+d]) - med;
            if (n == 0) { 
               mswap(ptr[unHi], ptr[gtHi]); 
               gtHi--; unHi--; continue; 
            };
            if (n <  0) break;
            unHi--;
         }
         if (unLo > unHi) break;
         mswap(ptr[unLo], ptr[unHi]); unLo++; unHi--;
      }

      AssertD ( unHi == unLo-1, "mainQSort3(2)" );

      if (gtHi < ltLo) {
         mpush(lo, hi, d+1 );
         continue;
      }

      n = mmin(ltLo-lo, unLo-ltLo); mvswap(lo, unLo-n, n);
      m = mmin(hi-gtHi, gtHi-unHi); mvswap(unLo, hi-m+1, m);

      n = lo + unLo - ltLo - 1;
      m = hi - (gtHi - unHi) + 1;

      nextLo[0] = lo;  nextHi[0] = n;   nextD[0] = d;
      nextLo[1] = m;   nextHi[1] = hi;  nextD[1] = d;
      nextLo[2] = n+1; nextHi[2] = m-1; nextD[2] = d+1;

      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
      if (mnextsize(1) < mnextsize(2)) mnextswap(1,2);
      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);

      AssertD (mnextsize(0) >= mnextsize(1), "mainQSort3(8)" );
      AssertD (mnextsize(1) >= mnextsize(2), "mainQSort3(9)" );

      mpush (nextLo[0], nextHi[0], nextD[0]);
      mpush (nextLo[1], nextHi[1], nextD[1]);
      mpush (nextLo[2], nextHi[2], nextD[2]);
   }
}

#undef mswap
#undef mvswap
#undef mpush
#undef mpop
#undef mmin
#undef mnextsize
#undef mnextswap
#undef MAIN_QSORT_SMALL_THRESH
#undef MAIN_QSORT_DEPTH_THRESH
#undef MAIN_QSORT_STACK_SIZE


/*---------------------------------------------*/
/* Pre:
      nblock > N_OVERSHOOT
      block32 exists for [0 .. nblock-1 +N_OVERSHOOT]
      ((UChar*)block32) [0 .. nblock-1] holds block
      ptr exists for [0 .. nblock-1]

   Post:
      ((UChar*)block32) [0 .. nblock-1] holds block
      All other areas of block32 destroyed
      ftab [0 .. 65536 ] destroyed
      ptr [0 .. nblock-1] holds sorted order
      if (*budget < 0), sorting was abandoned
*/

#define BIGFREQ(b) (ftab[((b)+1) << 8] - ftab[(b) << 8])
#define SETMASK (1 << 21)
#define CLEARMASK (~(SETMASK))

static
void mainSort ( UInt32* ptr, 
                UChar*  block,
                UInt16* quadrant, 
                UInt32* ftab,
                Int32   nblock,
                Int32   verb,
                Int32*  budget )
{
   Int32  i, j, k, ss, sb;
   Int32  runningOrder[256];
   Bool   bigDone[256];
   Int32  copyStart[256];
   Int32  copyEnd  [256];
   UChar  c1;
   Int32  numQSorted;
   UInt16 s;
   if (verb >= 4) VPrintf0 ( "        main sort initialise ...\n" );

   /*-- set up the 2-byte frequency table --*/
   for (i = 65536; i >= 0; i--) ftab[i] = 0;

   j = block[0] << 8;
   i = nblock-1;
   for (; i >= 3; i -= 4) {
      quadrant[i] = 0;
      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
      ftab[j]++;
      quadrant[i-1] = 0;
      j = (j >> 8) | ( ((UInt16)block[i-1]) << 8);
      ftab[j]++;
      quadrant[i-2] = 0;
      j = (j >> 8) | ( ((UInt16)block[i-2]) << 8);
      ftab[j]++;
      quadrant[i-3] = 0;
      j = (j >> 8) | ( ((UInt16)block[i-3]) << 8);
      ftab[j]++;
   }
   for (; i >= 0; i--) {
      quadrant[i] = 0;
      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
      ftab[j]++;
   }

   /*-- (emphasises close relationship of block & quadrant) --*/
   for (i = 0; i < BZ_N_OVERSHOOT; i++) {
      block   [nblock+i] = block[i];
      quadrant[nblock+i] = 0;
   }

   if (verb >= 4) VPrintf0 ( "        bucket sorting ...\n" );

   /*-- Complete the initial radix sort --*/
   for (i = 1; i <= 65536; i++) ftab[i] += ftab[i-1];

   s = block[0] << 8;
   i = nblock-1;
   for (; i >= 3; i -= 4) {
      s = (s >> 8) | (block[i] << 8);
      j = ftab[s] -1;
      ftab[s] = j;
      ptr[j] = i;
      s = (s >> 8) | (block[i-1] << 8);
      j = ftab[s] -1;
      ftab[s] = j;
      ptr[j] = i-1;
      s = (s >> 8) | (block[i-2] << 8);
      j = ftab[s] -1;
      ftab[s] = j;
      ptr[j] = i-2;
      s = (s >> 8) | (block[i-3] << 8);
      j = ftab[s] -1;
      ftab[s] = j;
      ptr[j] = i-3;
   }
   for (; i >= 0; i--) {
      s = (s >> 8) | (block[i] << 8);
      j = ftab[s] -1;
      ftab[s] = j;
      ptr[j] = i;
   }

   /*--
      Now ftab contains the first loc of every small bucket.
      Calculate the running order, from smallest to largest
      big bucket.
   --*/
   for (i = 0; i <= 255; i++) {
      bigDone     [i] = False;
      runningOrder[i] = i;
   }

   {
      Int32 vv;
      Int32 h = 1;
      do h = 3 * h + 1; while (h <= 256);
      do {
         h = h / 3;
         for (i = h; i <= 255; i++) {
            vv = runningOrder[i];
            j = i;
            while ( BIGFREQ(runningOrder[j-h]) > BIGFREQ(vv) ) {
               runningOrder[j] = runningOrder[j-h];
               j = j - h;
               if (j <= (h - 1)) goto zero;
            }
            zero:
            runningOrder[j] = vv;
         }
      } while (h != 1);
   }

   /*--
      The main sorting loop.
   --*/

   numQSorted = 0;

   for (i = 0; i <= 255; i++) {

      /*--
         Process big buckets, starting with the least full.
         Basically this is a 3-step process in which we call
         mainQSort3 to sort the small buckets [ss, j], but
         also make a big effort to avoid the calls if we can.
      --*/
      ss = runningOrder[i];

      /*--
         Step 1:
         Complete the big bucket [ss] by quicksorting
         any unsorted small buckets [ss, j], for j != ss.  
         Hopefully previous pointer-scanning phases have already
         completed many of the small buckets [ss, j], so
         we don't have to sort them at all.
      --*/
      for (j = 0; j <= 255; j++) {
         if (j != ss) {
            sb = (ss << 8) + j;
            if ( ! (ftab[sb] & SETMASK) ) {
               Int32 lo = ftab[sb]   & CLEARMASK;
               Int32 hi = (ftab[sb+1] & CLEARMASK) - 1;
               if (hi > lo) {
                  if (verb >= 4)
                     VPrintf4 ( "        qsort [0x%x, 0x%x]   "
                                "done %d   this %d\n",
                                ss, j, numQSorted, hi - lo + 1 );
                  mainQSort3 ( 
                     ptr, block, quadrant, nblock, 
                     lo, hi, BZ_N_RADIX, budget 
                  );   
                  numQSorted += (hi - lo + 1);
                  if (*budget < 0) return;
               }
            }
            ftab[sb] |= SETMASK;
         }
      }

      AssertH ( !bigDone[ss], 1006 );

      /*--
         Step 2:
         Now scan this big bucket [ss] so as to synthesise the
         sorted order for small buckets [t, ss] for all t,
         including, magically, the bucket [ss,ss] too.
         This will avoid doing Real Work in subsequent Step 1's.
      --*/
      {
         for (j = 0; j <= 255; j++) {
            copyStart[j] =  ftab[(j << 8) + ss]     & CLEARMASK;
            copyEnd  [j] = (ftab[(j << 8) + ss + 1] & CLEARMASK) - 1;
         }
         for (j = ftab[ss << 8] & CLEARMASK; j < copyStart[ss]; j++) {
            k = ptr[j]-1; if (k < 0) k += nblock;
            c1 = block[k];
            if (!bigDone[c1])
               ptr[ copyStart[c1]++ ] = k;
         }
         for (j = (ftab[(ss+1) << 8] & CLEARMASK) - 1; j > copyEnd[ss]; j--) {
            k = ptr[j]-1; if (k < 0) k += nblock;
            c1 = block[k];
            if (!bigDone[c1]) 
               ptr[ copyEnd[c1]-- ] = k;
         }
      }

      AssertH ( copyStart[ss]-1 == copyEnd[ss], 1007 );

      for (j = 0; j <= 255; j++) ftab[(j << 8) + ss] |= SETMASK;

      /*--
         Step 3:
         The [ss] big bucket is now done.  Record this fact,
         and update the quadrant descriptors.  Remember to
         update quadrants in the overshoot area too, if
         necessary.  The "if (i < 255)" test merely skips
         this updating for the last bucket processed, since
         updating for the last bucket is pointless.

         The quadrant array provides a way to incrementally
         cache sort orderings, as they appear, so as to 
         make subsequent comparisons in fullGtU() complete
         faster.  For repetitive blocks this makes a big
         difference (but not big enough to be able to avoid
         the fallback sorting mechanism, exponential radix sort).

         The precise meaning is: at all times:

            for 0 <= i < nblock and 0 <= j <= nblock

            if block[i] != block[j], 

               then the relative values of quadrant[i] and 
                    quadrant[j] are meaningless.

               else {
                  if quadrant[i] < quadrant[j]
                     then the string starting at i lexicographically
                     precedes the string starting at j

                  else if quadrant[i] > quadrant[j]
                     then the string starting at j lexicographically
                     precedes the string starting at i

                  else
                     the relative ordering of the strings starting
                     at i and j has not yet been determined.
               }
      --*/
      bigDone[ss] = True;

      if (i < 255) {
         Int32 bbStart  = ftab[ss << 8] & CLEARMASK;
         Int32 bbSize   = (ftab[(ss+1) << 8] & CLEARMASK) - bbStart;
         Int32 shifts   = 0;

         while ((bbSize >> shifts) > 65534) shifts++;

         for (j = bbSize-1; j >= 0; j--) {
            Int32 a2update     = ptr[bbStart + j];
            UInt16 qVal        = (UInt16)(j >> shifts);
            quadrant[a2update] = qVal;
            if (a2update < BZ_N_OVERSHOOT)
               quadrant[a2update + nblock] = qVal;
         }
         AssertH ( ((bbSize-1) >> shifts) <= 65535, 1002 );
      }

   }

   if (verb >= 4)
      VPrintf3 ( "        %d pointers, %d sorted, %d scanned\n",
                 nblock, numQSorted, nblock - numQSorted );
}

#undef BIGFREQ
#undef SETMASK
#undef CLEARMASK


/*---------------------------------------------*/
/* Pre:
      nblock > 0
      arr2 exists for [0 .. nblock-1 +N_OVERSHOOT]
      ((UChar*)arr2)  [0 .. nblock-1] holds block
      arr1 exists for [0 .. nblock-1]

   Post:
      ((UChar*)arr2) [0 .. nblock-1] holds block
      All other areas of block destroyed
      ftab [ 0 .. 65536 ] destroyed
      arr1 [0 .. nblock-1] holds sorted order
*/
void BZ2_blockSort ( EState* s )
{
   UInt32* ptr    = s->ptr; 
   UChar*  block  = s->block;
   UInt32* ftab   = s->ftab;
   Int32   nblock = s->nblock;
   Int32   verb   = s->verbosity;
   Int32   wfact  = s->workFactor;
   UInt16* quadrant;
   Int32   budget;
   Int32   budgetInit;
   Int32   i;

   if (nblock < 10000) {
      fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
   } else {
      /* Calculate the location for quadrant, remembering to get
         the alignment right.  Assumes that &(block[0]) is at least
         2-byte aligned -- this should be ok since block is really
         the first section of arr2.
      */
      i = nblock+BZ_N_OVERSHOOT;
      if (i & 1) i++;
      quadrant = (UInt16*)(&(block[i]));

      /* (wfact-1) / 3 puts the default-factor-30
         transition point at very roughly the same place as 
         with v0.1 and v0.9.0.  
         Not that it particularly matters any more, since the
         resulting compressed stream is now the same regardless
         of whether or not we use the main sort or fallback sort.
      */
      if (wfact < 1  ) wfact = 1;
      if (wfact > 100) wfact = 100;
      budgetInit = nblock * ((wfact-1) / 3);
      budget = budgetInit;

      mainSort ( ptr, block, quadrant, ftab, nblock, verb, &budget );
      if (verb >= 3) 
         VPrintf3 ( "      %d work, %d block, ratio %5.2f\n",
                    budgetInit - budget,
                    nblock, 
                    (float)(budgetInit - budget) /
                    (float)(nblock==0 ? 1 : nblock) ); 
      if (budget < 0) {
         if (verb >= 2) 
            VPrintf0 ( "    too repetitive; using fallback"
                       " sorting algorithm\n" );
         fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
      }
   }

   s->origPtr = -1;
   for (i = 0; i < s->nblock; i++)
      if (ptr[i] == 0)
         { s->origPtr = i; break; };

   AssertH( s->origPtr != -1, 1003 );
}


/*-------------------------------------------------------------*/
/*--- end                                       blocksort.c ---*/
/*-------------------------------------------------------------*/