2 * bzip2 is written by Julian Seward <jseward@bzip.org>.
3 * Adapted for busybox by Denys Vlasenko <vda.linux@googlemail.com>.
4 * See README and LICENSE files in this directory for more information.
7 /*-------------------------------------------------------------*/
8 /*--- Compression machinery (not incl block sorting) ---*/
10 /*-------------------------------------------------------------*/
12 /* ------------------------------------------------------------------
13 This file is part of bzip2/libbzip2, a program and library for
14 lossless, block-sorting data compression.
16 bzip2/libbzip2 version 1.0.4 of 20 December 2006
17 Copyright (C) 1996-2006 Julian Seward <jseward@bzip.org>
19 Please read the WARNING, DISCLAIMER and PATENTS sections in the
22 This program is released under the terms of the license contained
24 ------------------------------------------------------------------ */
27 * 0.9.0 -- original version.
28 * 0.9.0a/b -- no changes in this file.
29 * 0.9.0c -- changed setting of nGroups in sendMTFValues()
30 * so as to do a bit better on small files
33 /* #include "bzlib_private.h" */
35 /*---------------------------------------------------*/
36 /*--- Bit stream I/O ---*/
37 /*---------------------------------------------------*/
39 /*---------------------------------------------------*/
41 void BZ2_bsInitWrite(EState* s)
48 /*---------------------------------------------------*/
50 void bsFinishWrite(EState* s)
52 while (s->bsLive > 0) {
53 s->zbits[s->numZ] = (uint8_t)(s->bsBuff >> 24);
61 /*---------------------------------------------------*/
63 /* Helps only on level 5, on other levels hurts. ? */
64 #if CONFIG_BZIP2_FEATURE_SPEED >= 5
67 void bsW(EState* s, int32_t n, uint32_t v)
69 while (s->bsLive >= 8) {
70 s->zbits[s->numZ] = (uint8_t)(s->bsBuff >> 24);
75 s->bsBuff |= (v << (32 - s->bsLive - n));
80 /*---------------------------------------------------*/
82 void bsPutU32(EState* s, unsigned u)
84 bsW(s, 8, (u >> 24) & 0xff);
85 bsW(s, 8, (u >> 16) & 0xff);
86 bsW(s, 8, (u >> 8) & 0xff);
91 /*---------------------------------------------------*/
93 void bsPutU16(EState* s, unsigned u)
95 bsW(s, 8, (u >> 8) & 0xff);
100 /*---------------------------------------------------*/
101 /*--- The back end proper ---*/
102 /*---------------------------------------------------*/
104 /*---------------------------------------------------*/
106 void makeMaps_e(EState* s)
110 for (i = 0; i < 256; i++) {
112 s->unseqToSeq[i] = s->nInUse;
119 /*---------------------------------------------------*/
121 void generateMTFValues(EState* s)
130 * After sorting (eg, here),
131 * s->arr1[0 .. s->nblock-1] holds sorted order,
133 * ((uint8_t*)s->arr2)[0 .. s->nblock-1]
134 * holds the original block data.
136 * The first thing to do is generate the MTF values,
138 * ((uint16_t*)s->arr1)[0 .. s->nblock-1].
139 * Because there are strictly fewer or equal MTF values
140 * than block values, ptr values in this area are overwritten
141 * with MTF values only when they are no longer needed.
143 * The final compressed bitstream is generated into the
145 * &((uint8_t*)s->arr2)[s->nblock]
147 * These storage aliases are set up in bzCompressInit(),
148 * except for the last one, which is arranged in
151 uint32_t* ptr = s->ptr;
152 uint8_t* block = s->block;
153 uint16_t* mtfv = s->mtfv;
158 for (i = 0; i <= EOB; i++)
163 for (i = 0; i < s->nInUse; i++)
166 for (i = 0; i < s->nblock; i++) {
168 AssertD(wr <= i, "generateMTFValues(1)");
172 ll_i = s->unseqToSeq[block[j]];
173 AssertD(ll_i < s->nInUse, "generateMTFValues(2a)");
182 mtfv[wr] = BZ_RUNB; wr++;
183 s->mtfFreq[BZ_RUNB]++;
185 mtfv[wr] = BZ_RUNA; wr++;
186 s->mtfFreq[BZ_RUNA]++;
188 if (zPend < 2) break;
189 zPend = (zPend - 2) / 2;
194 register uint8_t rtmp;
195 register uint8_t* ryy_j;
196 register uint8_t rll_i;
201 while (rll_i != rtmp) {
202 register uint8_t rtmp2;
209 j = ryy_j - &(yy[0]);
222 mtfv[wr] = BZ_RUNB; wr++;
223 s->mtfFreq[BZ_RUNB]++;
225 mtfv[wr] = BZ_RUNA; wr++;
226 s->mtfFreq[BZ_RUNA]++;
230 zPend = (zPend - 2) / 2;
243 /*---------------------------------------------------*/
244 #define BZ_LESSER_ICOST 0
245 #define BZ_GREATER_ICOST 15
248 void sendMTFValues(EState* s)
250 int32_t v, t, i, j, gs, ge, totc, bt, bc, iter;
251 int32_t nSelectors, alphaSize, minLen, maxLen, selCtr;
252 int32_t nGroups, nBytes;
255 * uint8_t len[BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE];
256 * is a global since the decoder also needs it.
258 * int32_t code[BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE];
259 * int32_t rfreq[BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE];
260 * are also globals only used in this proc.
261 * Made global to keep stack frame size small.
264 uint16_t cost[BZ_N_GROUPS];
265 int32_t fave[BZ_N_GROUPS];
267 uint16_t* mtfv = s->mtfv;
269 alphaSize = s->nInUse+2;
270 for (t = 0; t < BZ_N_GROUPS; t++)
271 for (v = 0; v < alphaSize; v++)
272 s->len[t][v] = BZ_GREATER_ICOST;
274 /*--- Decide how many coding tables to use ---*/
275 AssertH(s->nMTF > 0, 3001);
276 if (s->nMTF < 200) nGroups = 2; else
277 if (s->nMTF < 600) nGroups = 3; else
278 if (s->nMTF < 1200) nGroups = 4; else
279 if (s->nMTF < 2400) nGroups = 5; else
282 /*--- Generate an initial set of coding tables ---*/
284 int32_t nPart, remF, tFreq, aFreq;
290 tFreq = remF / nPart;
293 while (aFreq < tFreq && ge < alphaSize-1) {
295 aFreq += s->mtfFreq[ge];
299 && nPart != nGroups && nPart != 1
300 && ((nGroups - nPart) % 2 == 1)
302 aFreq -= s->mtfFreq[ge];
306 for (v = 0; v < alphaSize; v++)
307 if (v >= gs && v <= ge)
308 s->len[nPart-1][v] = BZ_LESSER_ICOST;
310 s->len[nPart-1][v] = BZ_GREATER_ICOST;
319 * Iterate up to BZ_N_ITERS times to improve the tables.
321 for (iter = 0; iter < BZ_N_ITERS; iter++) {
322 for (t = 0; t < nGroups; t++)
325 for (t = 0; t < nGroups; t++)
326 for (v = 0; v < alphaSize; v++)
329 #if CONFIG_BZIP2_FEATURE_SPEED >= 5
331 * Set up an auxiliary length table which is used to fast-track
332 * the common case (nGroups == 6).
335 for (v = 0; v < alphaSize; v++) {
336 s->len_pack[v][0] = (s->len[1][v] << 16) | s->len[0][v];
337 s->len_pack[v][1] = (s->len[3][v] << 16) | s->len[2][v];
338 s->len_pack[v][2] = (s->len[5][v] << 16) | s->len[4][v];
346 /*--- Set group start & end marks. --*/
349 ge = gs + BZ_G_SIZE - 1;
354 * Calculate the cost of this group as coded
355 * by each of the coding tables.
357 for (t = 0; t < nGroups; t++)
359 #if CONFIG_BZIP2_FEATURE_SPEED >= 5
360 if (nGroups == 6 && 50 == ge-gs+1) {
361 /*--- fast track the common case ---*/
362 register uint32_t cost01, cost23, cost45;
363 register uint16_t icv;
364 cost01 = cost23 = cost45 = 0;
365 #define BZ_ITER(nn) \
366 icv = mtfv[gs+(nn)]; \
367 cost01 += s->len_pack[icv][0]; \
368 cost23 += s->len_pack[icv][1]; \
369 cost45 += s->len_pack[icv][2];
370 BZ_ITER(0); BZ_ITER(1); BZ_ITER(2); BZ_ITER(3); BZ_ITER(4);
371 BZ_ITER(5); BZ_ITER(6); BZ_ITER(7); BZ_ITER(8); BZ_ITER(9);
372 BZ_ITER(10); BZ_ITER(11); BZ_ITER(12); BZ_ITER(13); BZ_ITER(14);
373 BZ_ITER(15); BZ_ITER(16); BZ_ITER(17); BZ_ITER(18); BZ_ITER(19);
374 BZ_ITER(20); BZ_ITER(21); BZ_ITER(22); BZ_ITER(23); BZ_ITER(24);
375 BZ_ITER(25); BZ_ITER(26); BZ_ITER(27); BZ_ITER(28); BZ_ITER(29);
376 BZ_ITER(30); BZ_ITER(31); BZ_ITER(32); BZ_ITER(33); BZ_ITER(34);
377 BZ_ITER(35); BZ_ITER(36); BZ_ITER(37); BZ_ITER(38); BZ_ITER(39);
378 BZ_ITER(40); BZ_ITER(41); BZ_ITER(42); BZ_ITER(43); BZ_ITER(44);
379 BZ_ITER(45); BZ_ITER(46); BZ_ITER(47); BZ_ITER(48); BZ_ITER(49);
381 cost[0] = cost01 & 0xffff; cost[1] = cost01 >> 16;
382 cost[2] = cost23 & 0xffff; cost[3] = cost23 >> 16;
383 cost[4] = cost45 & 0xffff; cost[5] = cost45 >> 16;
388 /*--- slow version which correctly handles all situations ---*/
389 for (i = gs; i <= ge; i++) {
390 uint16_t icv = mtfv[i];
391 for (t = 0; t < nGroups; t++)
392 cost[t] += s->len[t][icv];
396 * Find the coding table which is best for this group,
397 * and record its identity in the selector table.
403 for (t = 1 /*0*/; t < nGroups; t++) {
411 s->selector[nSelectors] = bt;
415 * Increment the symbol frequencies for the selected table.
417 /* 1% faster compress. +800 bytes */
418 #if CONFIG_BZIP2_FEATURE_SPEED >= 4
419 if (nGroups == 6 && 50 == ge-gs+1) {
420 /*--- fast track the common case ---*/
421 #define BZ_ITUR(nn) s->rfreq[bt][mtfv[gs + (nn)]]++
422 BZ_ITUR(0); BZ_ITUR(1); BZ_ITUR(2); BZ_ITUR(3); BZ_ITUR(4);
423 BZ_ITUR(5); BZ_ITUR(6); BZ_ITUR(7); BZ_ITUR(8); BZ_ITUR(9);
424 BZ_ITUR(10); BZ_ITUR(11); BZ_ITUR(12); BZ_ITUR(13); BZ_ITUR(14);
425 BZ_ITUR(15); BZ_ITUR(16); BZ_ITUR(17); BZ_ITUR(18); BZ_ITUR(19);
426 BZ_ITUR(20); BZ_ITUR(21); BZ_ITUR(22); BZ_ITUR(23); BZ_ITUR(24);
427 BZ_ITUR(25); BZ_ITUR(26); BZ_ITUR(27); BZ_ITUR(28); BZ_ITUR(29);
428 BZ_ITUR(30); BZ_ITUR(31); BZ_ITUR(32); BZ_ITUR(33); BZ_ITUR(34);
429 BZ_ITUR(35); BZ_ITUR(36); BZ_ITUR(37); BZ_ITUR(38); BZ_ITUR(39);
430 BZ_ITUR(40); BZ_ITUR(41); BZ_ITUR(42); BZ_ITUR(43); BZ_ITUR(44);
431 BZ_ITUR(45); BZ_ITUR(46); BZ_ITUR(47); BZ_ITUR(48); BZ_ITUR(49);
437 /*--- slow version which correctly handles all situations ---*/
439 s->rfreq[bt][mtfv[gs]]++;
442 /* already is: gs = ge + 1; */
447 * Recompute the tables based on the accumulated frequencies.
449 /* maxLen was changed from 20 to 17 in bzip2-1.0.3. See
450 * comment in huffman.c for details. */
451 for (t = 0; t < nGroups; t++)
452 BZ2_hbMakeCodeLengths(&(s->len[t][0]), &(s->rfreq[t][0]), alphaSize, 17 /*20*/);
455 AssertH(nGroups < 8, 3002);
456 AssertH(nSelectors < 32768 && nSelectors <= (2 + (900000 / BZ_G_SIZE)), 3003);
458 /*--- Compute MTF values for the selectors. ---*/
460 uint8_t pos[BZ_N_GROUPS], ll_i, tmp2, tmp;
462 for (i = 0; i < nGroups; i++)
464 for (i = 0; i < nSelectors; i++) {
465 ll_i = s->selector[i];
468 while (ll_i != tmp) {
475 s->selectorMtf[i] = j;
479 /*--- Assign actual codes for the tables. --*/
480 for (t = 0; t < nGroups; t++) {
483 for (i = 0; i < alphaSize; i++) {
484 if (s->len[t][i] > maxLen) maxLen = s->len[t][i];
485 if (s->len[t][i] < minLen) minLen = s->len[t][i];
487 AssertH(!(maxLen > 17 /*20*/), 3004);
488 AssertH(!(minLen < 1), 3005);
489 BZ2_hbAssignCodes(&(s->code[t][0]), &(s->len[t][0]), minLen, maxLen, alphaSize);
492 /*--- Transmit the mapping table. ---*/
494 /* bbox: optimized a bit more than in bzip2 */
496 for (i = 0; i < 16; i++) {
497 if (sizeof(long) <= 4) {
498 inUse16 = inUse16*2 +
499 ((*(uint32_t*)&(s->inUse[i * 16 + 0])
500 | *(uint32_t*)&(s->inUse[i * 16 + 4])
501 | *(uint32_t*)&(s->inUse[i * 16 + 8])
502 | *(uint32_t*)&(s->inUse[i * 16 + 12])) != 0);
503 } else { /* Our CPU can do better */
504 inUse16 = inUse16*2 +
505 ((*(uint64_t*)&(s->inUse[i * 16 + 0])
506 | *(uint64_t*)&(s->inUse[i * 16 + 8])) != 0);
513 inUse16 <<= (sizeof(int)*8 - 16); /* move 15th bit into sign bit */
514 for (i = 0; i < 16; i++) {
517 for (j = 0; j < 16; j++)
518 v16 = v16*2 + s->inUse[i * 16 + j];
525 /*--- Now the selectors. ---*/
528 bsW(s, 15, nSelectors);
529 for (i = 0; i < nSelectors; i++) {
530 for (j = 0; j < s->selectorMtf[i]; j++)
535 /*--- Now the coding tables. ---*/
538 for (t = 0; t < nGroups; t++) {
539 int32_t curr = s->len[t][0];
541 for (i = 0; i < alphaSize; i++) {
542 while (curr < s->len[t][i]) { bsW(s, 2, 2); curr++; /* 10 */ };
543 while (curr > s->len[t][i]) { bsW(s, 2, 3); curr--; /* 11 */ };
548 /*--- And finally, the block data proper ---*/
555 ge = gs + BZ_G_SIZE - 1;
558 AssertH(s->selector[selCtr] < nGroups, 3006);
560 /* Costs 1300 bytes and is _slower_ (on Intel Core 2) */
562 if (nGroups == 6 && 50 == ge-gs+1) {
563 /*--- fast track the common case ---*/
565 uint8_t* s_len_sel_selCtr = &(s->len[s->selector[selCtr]][0]);
566 int32_t* s_code_sel_selCtr = &(s->code[s->selector[selCtr]][0]);
567 #define BZ_ITAH(nn) \
568 mtfv_i = mtfv[gs+(nn)]; \
569 bsW(s, s_len_sel_selCtr[mtfv_i], s_code_sel_selCtr[mtfv_i])
570 BZ_ITAH(0); BZ_ITAH(1); BZ_ITAH(2); BZ_ITAH(3); BZ_ITAH(4);
571 BZ_ITAH(5); BZ_ITAH(6); BZ_ITAH(7); BZ_ITAH(8); BZ_ITAH(9);
572 BZ_ITAH(10); BZ_ITAH(11); BZ_ITAH(12); BZ_ITAH(13); BZ_ITAH(14);
573 BZ_ITAH(15); BZ_ITAH(16); BZ_ITAH(17); BZ_ITAH(18); BZ_ITAH(19);
574 BZ_ITAH(20); BZ_ITAH(21); BZ_ITAH(22); BZ_ITAH(23); BZ_ITAH(24);
575 BZ_ITAH(25); BZ_ITAH(26); BZ_ITAH(27); BZ_ITAH(28); BZ_ITAH(29);
576 BZ_ITAH(30); BZ_ITAH(31); BZ_ITAH(32); BZ_ITAH(33); BZ_ITAH(34);
577 BZ_ITAH(35); BZ_ITAH(36); BZ_ITAH(37); BZ_ITAH(38); BZ_ITAH(39);
578 BZ_ITAH(40); BZ_ITAH(41); BZ_ITAH(42); BZ_ITAH(43); BZ_ITAH(44);
579 BZ_ITAH(45); BZ_ITAH(46); BZ_ITAH(47); BZ_ITAH(48); BZ_ITAH(49);
585 /*--- slow version which correctly handles all situations ---*/
586 /* code is bit bigger, but moves multiply out of the loop */
587 uint8_t* s_len_sel_selCtr = &(s->len [s->selector[selCtr]][0]);
588 int32_t* s_code_sel_selCtr = &(s->code[s->selector[selCtr]][0]);
591 s_len_sel_selCtr[mtfv[gs]],
592 s_code_sel_selCtr[mtfv[gs]]
596 /* already is: gs = ge+1; */
600 AssertH(selCtr == nSelectors, 3007);
604 /*---------------------------------------------------*/
606 void BZ2_compressBlock(EState* s, int is_last_block)
609 BZ_FINALISE_CRC(s->blockCRC);
610 s->combinedCRC = (s->combinedCRC << 1) | (s->combinedCRC >> 31);
611 s->combinedCRC ^= s->blockCRC;
618 s->zbits = &((uint8_t*)s->arr2)[s->nblock];
620 /*-- If this is the first block, create the stream header. --*/
621 if (s->blockNo == 1) {
623 /*bsPutU8(s, BZ_HDR_B);*/
624 /*bsPutU8(s, BZ_HDR_Z);*/
625 /*bsPutU8(s, BZ_HDR_h);*/
626 /*bsPutU8(s, BZ_HDR_0 + s->blockSize100k);*/
627 bsPutU32(s, BZ_HDR_BZh0 + s->blockSize100k);
631 /*bsPutU8(s, 0x31);*/
632 /*bsPutU8(s, 0x41);*/
633 /*bsPutU8(s, 0x59);*/
634 /*bsPutU8(s, 0x26);*/
635 bsPutU32(s, 0x31415926);
636 /*bsPutU8(s, 0x53);*/
637 /*bsPutU8(s, 0x59);*/
640 /*-- Now the block's CRC, so it is in a known place. --*/
641 bsPutU32(s, s->blockCRC);
644 * Now a single bit indicating (non-)randomisation.
645 * As of version 0.9.5, we use a better sorting algorithm
646 * which makes randomisation unnecessary. So always set
647 * the randomised bit to 'no'. Of course, the decoder
648 * still needs to be able to handle randomised blocks
649 * so as to maintain backwards compatibility with
650 * older versions of bzip2.
654 bsW(s, 24, s->origPtr);
655 generateMTFValues(s);
659 /*-- If this is the last block, add the stream trailer. --*/
661 /*bsPutU8(s, 0x17);*/
662 /*bsPutU8(s, 0x72);*/
663 /*bsPutU8(s, 0x45);*/
664 /*bsPutU8(s, 0x38);*/
665 bsPutU32(s, 0x17724538);
666 /*bsPutU8(s, 0x50);*/
667 /*bsPutU8(s, 0x90);*/
669 bsPutU32(s, s->combinedCRC);
675 /*-------------------------------------------------------------*/
676 /*--- end compress.c ---*/
677 /*-------------------------------------------------------------*/