001/* 002 * Licensed to the Apache Software Foundation (ASF) under one 003 * or more contributor license agreements. See the NOTICE file 004 * distributed with this work for additional information 005 * regarding copyright ownership. The ASF licenses this file 006 * to you under the Apache License, Version 2.0 (the 007 * "License"); you may not use this file except in compliance 008 * with the License. You may obtain a copy of the License at 009 * 010 * http://www.apache.org/licenses/LICENSE-2.0 011 * 012 * Unless required by applicable law or agreed to in writing, 013 * software distributed under the License is distributed on an 014 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY 015 * KIND, either express or implied. See the License for the 016 * specific language governing permissions and limitations 017 * under the License. 018 */ 019package org.apache.commons.compress.compressors.bzip2; 020 021import java.io.IOException; 022import java.io.OutputStream; 023 024import org.apache.commons.compress.compressors.CompressorOutputStream; 025 026/** 027 * An output stream that compresses into the BZip2 format into another stream. 028 * 029 * <p> 030 * The compression requires large amounts of memory. Thus you should call the 031 * {@link #close() close()} method as soon as possible, to force 032 * {@code BZip2CompressorOutputStream} to release the allocated memory. 033 * </p> 034 * 035 * <p> You can shrink the amount of allocated memory and maybe raise 036 * the compression speed by choosing a lower blocksize, which in turn 037 * may cause a lower compression ratio. You can avoid unnecessary 038 * memory allocation by avoiding using a blocksize which is bigger 039 * than the size of the input. </p> 040 * 041 * <p> You can compute the memory usage for compressing by the 042 * following formula: </p> 043 * 044 * <pre> 045 * <code>400k + (9 * blocksize)</code>. 046 * </pre> 047 * 048 * <p> To get the memory required for decompression by {@link 049 * BZip2CompressorInputStream} use </p> 050 * 051 * <pre> 052 * <code>65k + (5 * blocksize)</code>. 053 * </pre> 054 * 055 * <table width="100%" border="1" summary="Memory usage by blocksize"> 056 * <tr> 057 * <th colspan="3">Memory usage by blocksize</th> 058 * </tr> 059 * <tr> 060 * <th align="right">Blocksize</th> <th align="right">Compression<br> 061 * memory usage</th> <th align="right">Decompression<br> 062 * memory usage</th> 063 * </tr> 064 * <tr> 065 * <td align="right">100k</td> 066 * <td align="right">1300k</td> 067 * <td align="right">565k</td> 068 * </tr> 069 * <tr> 070 * <td align="right">200k</td> 071 * <td align="right">2200k</td> 072 * <td align="right">1065k</td> 073 * </tr> 074 * <tr> 075 * <td align="right">300k</td> 076 * <td align="right">3100k</td> 077 * <td align="right">1565k</td> 078 * </tr> 079 * <tr> 080 * <td align="right">400k</td> 081 * <td align="right">4000k</td> 082 * <td align="right">2065k</td> 083 * </tr> 084 * <tr> 085 * <td align="right">500k</td> 086 * <td align="right">4900k</td> 087 * <td align="right">2565k</td> 088 * </tr> 089 * <tr> 090 * <td align="right">600k</td> 091 * <td align="right">5800k</td> 092 * <td align="right">3065k</td> 093 * </tr> 094 * <tr> 095 * <td align="right">700k</td> 096 * <td align="right">6700k</td> 097 * <td align="right">3565k</td> 098 * </tr> 099 * <tr> 100 * <td align="right">800k</td> 101 * <td align="right">7600k</td> 102 * <td align="right">4065k</td> 103 * </tr> 104 * <tr> 105 * <td align="right">900k</td> 106 * <td align="right">8500k</td> 107 * <td align="right">4565k</td> 108 * </tr> 109 * </table> 110 * 111 * <p> 112 * For decompression {@code BZip2CompressorInputStream} allocates less memory if the 113 * bzipped input is smaller than one block. 114 * </p> 115 * 116 * <p> 117 * Instances of this class are not threadsafe. 118 * </p> 119 * 120 * <p> 121 * TODO: Update to BZip2 1.0.1 122 * </p> 123 * @NotThreadSafe 124 */ 125public class BZip2CompressorOutputStream extends CompressorOutputStream 126 implements BZip2Constants { 127 128 /** 129 * The minimum supported blocksize {@code == 1}. 130 */ 131 public static final int MIN_BLOCKSIZE = 1; 132 133 /** 134 * The maximum supported blocksize {@code == 9}. 135 */ 136 public static final int MAX_BLOCKSIZE = 9; 137 138 private static final int GREATER_ICOST = 15; 139 private static final int LESSER_ICOST = 0; 140 141 private static void hbMakeCodeLengths(final byte[] len, final int[] freq, 142 final Data dat, final int alphaSize, 143 final int maxLen) { 144 /* 145 * Nodes and heap entries run from 1. Entry 0 for both the heap and 146 * nodes is a sentinel. 147 */ 148 final int[] heap = dat.heap; 149 final int[] weight = dat.weight; 150 final int[] parent = dat.parent; 151 152 for (int i = alphaSize; --i >= 0;) { 153 weight[i + 1] = (freq[i] == 0 ? 1 : freq[i]) << 8; 154 } 155 156 for (boolean tooLong = true; tooLong;) { 157 tooLong = false; 158 159 int nNodes = alphaSize; 160 int nHeap = 0; 161 heap[0] = 0; 162 weight[0] = 0; 163 parent[0] = -2; 164 165 for (int i = 1; i <= alphaSize; i++) { 166 parent[i] = -1; 167 nHeap++; 168 heap[nHeap] = i; 169 170 int zz = nHeap; 171 int tmp = heap[zz]; 172 while (weight[tmp] < weight[heap[zz >> 1]]) { 173 heap[zz] = heap[zz >> 1]; 174 zz >>= 1; 175 } 176 heap[zz] = tmp; 177 } 178 179 while (nHeap > 1) { 180 int n1 = heap[1]; 181 heap[1] = heap[nHeap]; 182 nHeap--; 183 184 int yy = 0; 185 int zz = 1; 186 int tmp = heap[1]; 187 188 while (true) { 189 yy = zz << 1; 190 191 if (yy > nHeap) { 192 break; 193 } 194 195 if ((yy < nHeap) 196 && (weight[heap[yy + 1]] < weight[heap[yy]])) { 197 yy++; 198 } 199 200 if (weight[tmp] < weight[heap[yy]]) { 201 break; 202 } 203 204 heap[zz] = heap[yy]; 205 zz = yy; 206 } 207 208 heap[zz] = tmp; 209 210 int n2 = heap[1]; 211 heap[1] = heap[nHeap]; 212 nHeap--; 213 214 yy = 0; 215 zz = 1; 216 tmp = heap[1]; 217 218 while (true) { 219 yy = zz << 1; 220 221 if (yy > nHeap) { 222 break; 223 } 224 225 if ((yy < nHeap) 226 && (weight[heap[yy + 1]] < weight[heap[yy]])) { 227 yy++; 228 } 229 230 if (weight[tmp] < weight[heap[yy]]) { 231 break; 232 } 233 234 heap[zz] = heap[yy]; 235 zz = yy; 236 } 237 238 heap[zz] = tmp; 239 nNodes++; 240 parent[n1] = parent[n2] = nNodes; 241 242 final int weight_n1 = weight[n1]; 243 final int weight_n2 = weight[n2]; 244 weight[nNodes] = ((weight_n1 & 0xffffff00) 245 + (weight_n2 & 0xffffff00)) 246 | (1 + (((weight_n1 & 0x000000ff) 247 > (weight_n2 & 0x000000ff)) 248 ? (weight_n1 & 0x000000ff) 249 : (weight_n2 & 0x000000ff))); 250 251 parent[nNodes] = -1; 252 nHeap++; 253 heap[nHeap] = nNodes; 254 255 tmp = 0; 256 zz = nHeap; 257 tmp = heap[zz]; 258 final int weight_tmp = weight[tmp]; 259 while (weight_tmp < weight[heap[zz >> 1]]) { 260 heap[zz] = heap[zz >> 1]; 261 zz >>= 1; 262 } 263 heap[zz] = tmp; 264 265 } 266 267 for (int i = 1; i <= alphaSize; i++) { 268 int j = 0; 269 int k = i; 270 271 for (int parent_k; (parent_k = parent[k]) >= 0;) { 272 k = parent_k; 273 j++; 274 } 275 276 len[i - 1] = (byte) j; 277 if (j > maxLen) { 278 tooLong = true; 279 } 280 } 281 282 if (tooLong) { 283 for (int i = 1; i < alphaSize; i++) { 284 int j = weight[i] >> 8; 285 j = 1 + (j >> 1); 286 weight[i] = j << 8; 287 } 288 } 289 } 290 } 291 292 /** 293 * Index of the last char in the block, so the block size == last + 1. 294 */ 295 private int last; 296 297 /** 298 * Always: in the range 0 .. 9. The current block size is 100000 * this 299 * number. 300 */ 301 private final int blockSize100k; 302 303 private int bsBuff; 304 private int bsLive; 305 private final CRC crc = new CRC(); 306 307 private int nInUse; 308 309 private int nMTF; 310 311 private int currentChar = -1; 312 private int runLength = 0; 313 314 private int blockCRC; 315 private int combinedCRC; 316 private final int allowableBlockSize; 317 318 /** 319 * All memory intensive stuff. 320 */ 321 private Data data; 322 private BlockSort blockSorter; 323 324 private OutputStream out; 325 326 /** 327 * Chooses a blocksize based on the given length of the data to compress. 328 * 329 * @return The blocksize, between {@link #MIN_BLOCKSIZE} and 330 * {@link #MAX_BLOCKSIZE} both inclusive. For a negative 331 * {@code inputLength} this method returns {@code MAX_BLOCKSIZE} 332 * always. 333 * 334 * @param inputLength 335 * The length of the data which will be compressed by 336 * {@code BZip2CompressorOutputStream}. 337 */ 338 public static int chooseBlockSize(long inputLength) { 339 return (inputLength > 0) ? (int) Math 340 .min((inputLength / 132000) + 1, 9) : MAX_BLOCKSIZE; 341 } 342 343 /** 344 * Constructs a new {@code BZip2CompressorOutputStream} with a blocksize of 900k. 345 * 346 * @param out 347 * the destination stream. 348 * 349 * @throws IOException 350 * if an I/O error occurs in the specified stream. 351 * @throws NullPointerException 352 * if <code>out == null</code>. 353 */ 354 public BZip2CompressorOutputStream(final OutputStream out) 355 throws IOException { 356 this(out, MAX_BLOCKSIZE); 357 } 358 359 /** 360 * Constructs a new {@code BZip2CompressorOutputStream} with specified blocksize. 361 * 362 * @param out 363 * the destination stream. 364 * @param blockSize 365 * the blockSize as 100k units. 366 * 367 * @throws IOException 368 * if an I/O error occurs in the specified stream. 369 * @throws IllegalArgumentException 370 * if <code>(blockSize < 1) || (blockSize > 9)</code>. 371 * @throws NullPointerException 372 * if <code>out == null</code>. 373 * 374 * @see #MIN_BLOCKSIZE 375 * @see #MAX_BLOCKSIZE 376 */ 377 public BZip2CompressorOutputStream(final OutputStream out, final int blockSize) throws IOException { 378 if (blockSize < 1) { 379 throw new IllegalArgumentException("blockSize(" + blockSize + ") < 1"); 380 } 381 if (blockSize > 9) { 382 throw new IllegalArgumentException("blockSize(" + blockSize + ") > 9"); 383 } 384 385 this.blockSize100k = blockSize; 386 this.out = out; 387 388 /* 20 is just a paranoia constant */ 389 this.allowableBlockSize = (this.blockSize100k * BZip2Constants.BASEBLOCKSIZE) - 20; 390 init(); 391 } 392 393 @Override 394 public void write(final int b) throws IOException { 395 if (this.out != null) { 396 write0(b); 397 } else { 398 throw new IOException("closed"); 399 } 400 } 401 402 /** 403 * Writes the current byte to the buffer, run-length encoding it 404 * if it has been repeated at least four times (the first step 405 * RLEs sequences of four identical bytes). 406 * 407 * <p>Flushes the current block before writing data if it is 408 * full.</p> 409 * 410 * <p>"write to the buffer" means adding to data.buffer starting 411 * two steps "after" this.last - initially starting at index 1 412 * (not 0) - and updating this.last to point to the last index 413 * written minus 1.</p> 414 */ 415 private void writeRun() throws IOException { 416 final int lastShadow = this.last; 417 418 if (lastShadow < this.allowableBlockSize) { 419 final int currentCharShadow = this.currentChar; 420 final Data dataShadow = this.data; 421 dataShadow.inUse[currentCharShadow] = true; 422 final byte ch = (byte) currentCharShadow; 423 424 int runLengthShadow = this.runLength; 425 this.crc.updateCRC(currentCharShadow, runLengthShadow); 426 427 switch (runLengthShadow) { 428 case 1: 429 dataShadow.block[lastShadow + 2] = ch; 430 this.last = lastShadow + 1; 431 break; 432 433 case 2: 434 dataShadow.block[lastShadow + 2] = ch; 435 dataShadow.block[lastShadow + 3] = ch; 436 this.last = lastShadow + 2; 437 break; 438 439 case 3: { 440 final byte[] block = dataShadow.block; 441 block[lastShadow + 2] = ch; 442 block[lastShadow + 3] = ch; 443 block[lastShadow + 4] = ch; 444 this.last = lastShadow + 3; 445 } 446 break; 447 448 default: { 449 runLengthShadow -= 4; 450 dataShadow.inUse[runLengthShadow] = true; 451 final byte[] block = dataShadow.block; 452 block[lastShadow + 2] = ch; 453 block[lastShadow + 3] = ch; 454 block[lastShadow + 4] = ch; 455 block[lastShadow + 5] = ch; 456 block[lastShadow + 6] = (byte) runLengthShadow; 457 this.last = lastShadow + 5; 458 } 459 break; 460 461 } 462 } else { 463 endBlock(); 464 initBlock(); 465 writeRun(); 466 } 467 } 468 469 /** 470 * Overriden to close the stream. 471 */ 472 @Override 473 protected void finalize() throws Throwable { 474 finish(); 475 super.finalize(); 476 } 477 478 479 public void finish() throws IOException { 480 if (out != null) { 481 try { 482 if (this.runLength > 0) { 483 writeRun(); 484 } 485 this.currentChar = -1; 486 endBlock(); 487 endCompression(); 488 } finally { 489 this.out = null; 490 this.data = null; 491 this.blockSorter = null; 492 } 493 } 494 } 495 496 @Override 497 public void close() throws IOException { 498 if (out != null) { 499 OutputStream outShadow = this.out; 500 finish(); 501 outShadow.close(); 502 } 503 } 504 505 @Override 506 public void flush() throws IOException { 507 OutputStream outShadow = this.out; 508 if (outShadow != null) { 509 outShadow.flush(); 510 } 511 } 512 513 /** 514 * Writes magic bytes like BZ on the first position of the stream 515 * and bytes indiciating the file-format, which is 516 * huffmanised, followed by a digit indicating blockSize100k. 517 * @throws IOException if the magic bytes could not been written 518 */ 519 private void init() throws IOException { 520 bsPutUByte('B'); 521 bsPutUByte('Z'); 522 523 this.data = new Data(this.blockSize100k); 524 this.blockSorter = new BlockSort(this.data); 525 526 // huffmanised magic bytes 527 bsPutUByte('h'); 528 bsPutUByte('0' + this.blockSize100k); 529 530 this.combinedCRC = 0; 531 initBlock(); 532 } 533 534 private void initBlock() { 535 // blockNo++; 536 this.crc.initialiseCRC(); 537 this.last = -1; 538 // ch = 0; 539 540 boolean[] inUse = this.data.inUse; 541 for (int i = 256; --i >= 0;) { 542 inUse[i] = false; 543 } 544 545 } 546 547 private void endBlock() throws IOException { 548 this.blockCRC = this.crc.getFinalCRC(); 549 this.combinedCRC = (this.combinedCRC << 1) | (this.combinedCRC >>> 31); 550 this.combinedCRC ^= this.blockCRC; 551 552 // empty block at end of file 553 if (this.last == -1) { 554 return; 555 } 556 557 /* sort the block and establish posn of original string */ 558 blockSort(); 559 560 /* 561 * A 6-byte block header, the value chosen arbitrarily as 0x314159265359 562 * :-). A 32 bit value does not really give a strong enough guarantee 563 * that the value will not appear by chance in the compressed 564 * datastream. Worst-case probability of this event, for a 900k block, 565 * is about 2.0e-3 for 32 bits, 1.0e-5 for 40 bits and 4.0e-8 for 48 566 * bits. For a compressed file of size 100Gb -- about 100000 blocks -- 567 * only a 48-bit marker will do. NB: normal compression/ decompression 568 * donot rely on these statistical properties. They are only important 569 * when trying to recover blocks from damaged files. 570 */ 571 bsPutUByte(0x31); 572 bsPutUByte(0x41); 573 bsPutUByte(0x59); 574 bsPutUByte(0x26); 575 bsPutUByte(0x53); 576 bsPutUByte(0x59); 577 578 /* Now the block's CRC, so it is in a known place. */ 579 bsPutInt(this.blockCRC); 580 581 /* Now a single bit indicating no randomisation. */ 582 bsW(1, 0); 583 584 /* Finally, block's contents proper. */ 585 moveToFrontCodeAndSend(); 586 } 587 588 private void endCompression() throws IOException { 589 /* 590 * Now another magic 48-bit number, 0x177245385090, to indicate the end 591 * of the last block. (sqrt(pi), if you want to know. I did want to use 592 * e, but it contains too much repetition -- 27 18 28 18 28 46 -- for me 593 * to feel statistically comfortable. Call me paranoid.) 594 */ 595 bsPutUByte(0x17); 596 bsPutUByte(0x72); 597 bsPutUByte(0x45); 598 bsPutUByte(0x38); 599 bsPutUByte(0x50); 600 bsPutUByte(0x90); 601 602 bsPutInt(this.combinedCRC); 603 bsFinishedWithStream(); 604 } 605 606 /** 607 * Returns the blocksize parameter specified at construction time. 608 * @return the blocksize parameter specified at construction time 609 */ 610 public final int getBlockSize() { 611 return this.blockSize100k; 612 } 613 614 @Override 615 public void write(final byte[] buf, int offs, final int len) 616 throws IOException { 617 if (offs < 0) { 618 throw new IndexOutOfBoundsException("offs(" + offs + ") < 0."); 619 } 620 if (len < 0) { 621 throw new IndexOutOfBoundsException("len(" + len + ") < 0."); 622 } 623 if (offs + len > buf.length) { 624 throw new IndexOutOfBoundsException("offs(" + offs + ") + len(" 625 + len + ") > buf.length(" 626 + buf.length + ")."); 627 } 628 if (this.out == null) { 629 throw new IOException("stream closed"); 630 } 631 632 for (int hi = offs + len; offs < hi;) { 633 write0(buf[offs++]); 634 } 635 } 636 637 /** 638 * Keeps track of the last bytes written and implicitly performs 639 * run-length encoding as the first step of the bzip2 algorithm. 640 */ 641 private void write0(int b) throws IOException { 642 if (this.currentChar != -1) { 643 b &= 0xff; 644 if (this.currentChar == b) { 645 if (++this.runLength > 254) { 646 writeRun(); 647 this.currentChar = -1; 648 this.runLength = 0; 649 } 650 // else nothing to do 651 } else { 652 writeRun(); 653 this.runLength = 1; 654 this.currentChar = b; 655 } 656 } else { 657 this.currentChar = b & 0xff; 658 this.runLength++; 659 } 660 } 661 662 private static void hbAssignCodes(final int[] code, final byte[] length, 663 final int minLen, final int maxLen, 664 final int alphaSize) { 665 int vec = 0; 666 for (int n = minLen; n <= maxLen; n++) { 667 for (int i = 0; i < alphaSize; i++) { 668 if ((length[i] & 0xff) == n) { 669 code[i] = vec; 670 vec++; 671 } 672 } 673 vec <<= 1; 674 } 675 } 676 677 private void bsFinishedWithStream() throws IOException { 678 while (this.bsLive > 0) { 679 int ch = this.bsBuff >> 24; 680 this.out.write(ch); // write 8-bit 681 this.bsBuff <<= 8; 682 this.bsLive -= 8; 683 } 684 } 685 686 private void bsW(final int n, final int v) throws IOException { 687 final OutputStream outShadow = this.out; 688 int bsLiveShadow = this.bsLive; 689 int bsBuffShadow = this.bsBuff; 690 691 while (bsLiveShadow >= 8) { 692 outShadow.write(bsBuffShadow >> 24); // write 8-bit 693 bsBuffShadow <<= 8; 694 bsLiveShadow -= 8; 695 } 696 697 this.bsBuff = bsBuffShadow | (v << (32 - bsLiveShadow - n)); 698 this.bsLive = bsLiveShadow + n; 699 } 700 701 private void bsPutUByte(final int c) throws IOException { 702 bsW(8, c); 703 } 704 705 private void bsPutInt(final int u) throws IOException { 706 bsW(8, (u >> 24) & 0xff); 707 bsW(8, (u >> 16) & 0xff); 708 bsW(8, (u >> 8) & 0xff); 709 bsW(8, u & 0xff); 710 } 711 712 private void sendMTFValues() throws IOException { 713 final byte[][] len = this.data.sendMTFValues_len; 714 final int alphaSize = this.nInUse + 2; 715 716 for (int t = N_GROUPS; --t >= 0;) { 717 byte[] len_t = len[t]; 718 for (int v = alphaSize; --v >= 0;) { 719 len_t[v] = GREATER_ICOST; 720 } 721 } 722 723 /* Decide how many coding tables to use */ 724 // assert (this.nMTF > 0) : this.nMTF; 725 final int nGroups = (this.nMTF < 200) ? 2 : (this.nMTF < 600) ? 3 726 : (this.nMTF < 1200) ? 4 : (this.nMTF < 2400) ? 5 : 6; 727 728 /* Generate an initial set of coding tables */ 729 sendMTFValues0(nGroups, alphaSize); 730 731 /* 732 * Iterate up to N_ITERS times to improve the tables. 733 */ 734 final int nSelectors = sendMTFValues1(nGroups, alphaSize); 735 736 /* Compute MTF values for the selectors. */ 737 sendMTFValues2(nGroups, nSelectors); 738 739 /* Assign actual codes for the tables. */ 740 sendMTFValues3(nGroups, alphaSize); 741 742 /* Transmit the mapping table. */ 743 sendMTFValues4(); 744 745 /* Now the selectors. */ 746 sendMTFValues5(nGroups, nSelectors); 747 748 /* Now the coding tables. */ 749 sendMTFValues6(nGroups, alphaSize); 750 751 /* And finally, the block data proper */ 752 sendMTFValues7(); 753 } 754 755 private void sendMTFValues0(final int nGroups, final int alphaSize) { 756 final byte[][] len = this.data.sendMTFValues_len; 757 final int[] mtfFreq = this.data.mtfFreq; 758 759 int remF = this.nMTF; 760 int gs = 0; 761 762 for (int nPart = nGroups; nPart > 0; nPart--) { 763 final int tFreq = remF / nPart; 764 int ge = gs - 1; 765 int aFreq = 0; 766 767 for (final int a = alphaSize - 1; (aFreq < tFreq) && (ge < a);) { 768 aFreq += mtfFreq[++ge]; 769 } 770 771 if ((ge > gs) && (nPart != nGroups) && (nPart != 1) 772 && (((nGroups - nPart) & 1) != 0)) { 773 aFreq -= mtfFreq[ge--]; 774 } 775 776 final byte[] len_np = len[nPart - 1]; 777 for (int v = alphaSize; --v >= 0;) { 778 if ((v >= gs) && (v <= ge)) { 779 len_np[v] = LESSER_ICOST; 780 } else { 781 len_np[v] = GREATER_ICOST; 782 } 783 } 784 785 gs = ge + 1; 786 remF -= aFreq; 787 } 788 } 789 790 private int sendMTFValues1(final int nGroups, final int alphaSize) { 791 final Data dataShadow = this.data; 792 final int[][] rfreq = dataShadow.sendMTFValues_rfreq; 793 final int[] fave = dataShadow.sendMTFValues_fave; 794 final short[] cost = dataShadow.sendMTFValues_cost; 795 final char[] sfmap = dataShadow.sfmap; 796 final byte[] selector = dataShadow.selector; 797 final byte[][] len = dataShadow.sendMTFValues_len; 798 final byte[] len_0 = len[0]; 799 final byte[] len_1 = len[1]; 800 final byte[] len_2 = len[2]; 801 final byte[] len_3 = len[3]; 802 final byte[] len_4 = len[4]; 803 final byte[] len_5 = len[5]; 804 final int nMTFShadow = this.nMTF; 805 806 int nSelectors = 0; 807 808 for (int iter = 0; iter < N_ITERS; iter++) { 809 for (int t = nGroups; --t >= 0;) { 810 fave[t] = 0; 811 int[] rfreqt = rfreq[t]; 812 for (int i = alphaSize; --i >= 0;) { 813 rfreqt[i] = 0; 814 } 815 } 816 817 nSelectors = 0; 818 819 for (int gs = 0; gs < this.nMTF;) { 820 /* Set group start & end marks. */ 821 822 /* 823 * Calculate the cost of this group as coded by each of the 824 * coding tables. 825 */ 826 827 final int ge = Math.min(gs + G_SIZE - 1, nMTFShadow - 1); 828 829 if (nGroups == N_GROUPS) { 830 // unrolled version of the else-block 831 832 short cost0 = 0; 833 short cost1 = 0; 834 short cost2 = 0; 835 short cost3 = 0; 836 short cost4 = 0; 837 short cost5 = 0; 838 839 for (int i = gs; i <= ge; i++) { 840 final int icv = sfmap[i]; 841 cost0 += len_0[icv] & 0xff; 842 cost1 += len_1[icv] & 0xff; 843 cost2 += len_2[icv] & 0xff; 844 cost3 += len_3[icv] & 0xff; 845 cost4 += len_4[icv] & 0xff; 846 cost5 += len_5[icv] & 0xff; 847 } 848 849 cost[0] = cost0; 850 cost[1] = cost1; 851 cost[2] = cost2; 852 cost[3] = cost3; 853 cost[4] = cost4; 854 cost[5] = cost5; 855 856 } else { 857 for (int t = nGroups; --t >= 0;) { 858 cost[t] = 0; 859 } 860 861 for (int i = gs; i <= ge; i++) { 862 final int icv = sfmap[i]; 863 for (int t = nGroups; --t >= 0;) { 864 cost[t] += len[t][icv] & 0xff; 865 } 866 } 867 } 868 869 /* 870 * Find the coding table which is best for this group, and 871 * record its identity in the selector table. 872 */ 873 int bt = -1; 874 for (int t = nGroups, bc = 999999999; --t >= 0;) { 875 final int cost_t = cost[t]; 876 if (cost_t < bc) { 877 bc = cost_t; 878 bt = t; 879 } 880 } 881 882 fave[bt]++; 883 selector[nSelectors] = (byte) bt; 884 nSelectors++; 885 886 /* 887 * Increment the symbol frequencies for the selected table. 888 */ 889 final int[] rfreq_bt = rfreq[bt]; 890 for (int i = gs; i <= ge; i++) { 891 rfreq_bt[sfmap[i]]++; 892 } 893 894 gs = ge + 1; 895 } 896 897 /* 898 * Recompute the tables based on the accumulated frequencies. 899 */ 900 for (int t = 0; t < nGroups; t++) { 901 hbMakeCodeLengths(len[t], rfreq[t], this.data, alphaSize, 20); 902 } 903 } 904 905 return nSelectors; 906 } 907 908 private void sendMTFValues2(final int nGroups, final int nSelectors) { 909 // assert (nGroups < 8) : nGroups; 910 911 final Data dataShadow = this.data; 912 byte[] pos = dataShadow.sendMTFValues2_pos; 913 914 for (int i = nGroups; --i >= 0;) { 915 pos[i] = (byte) i; 916 } 917 918 for (int i = 0; i < nSelectors; i++) { 919 final byte ll_i = dataShadow.selector[i]; 920 byte tmp = pos[0]; 921 int j = 0; 922 923 while (ll_i != tmp) { 924 j++; 925 byte tmp2 = tmp; 926 tmp = pos[j]; 927 pos[j] = tmp2; 928 } 929 930 pos[0] = tmp; 931 dataShadow.selectorMtf[i] = (byte) j; 932 } 933 } 934 935 private void sendMTFValues3(final int nGroups, final int alphaSize) { 936 int[][] code = this.data.sendMTFValues_code; 937 byte[][] len = this.data.sendMTFValues_len; 938 939 for (int t = 0; t < nGroups; t++) { 940 int minLen = 32; 941 int maxLen = 0; 942 final byte[] len_t = len[t]; 943 for (int i = alphaSize; --i >= 0;) { 944 final int l = len_t[i] & 0xff; 945 if (l > maxLen) { 946 maxLen = l; 947 } 948 if (l < minLen) { 949 minLen = l; 950 } 951 } 952 953 // assert (maxLen <= 20) : maxLen; 954 // assert (minLen >= 1) : minLen; 955 956 hbAssignCodes(code[t], len[t], minLen, maxLen, alphaSize); 957 } 958 } 959 960 private void sendMTFValues4() throws IOException { 961 final boolean[] inUse = this.data.inUse; 962 final boolean[] inUse16 = this.data.sentMTFValues4_inUse16; 963 964 for (int i = 16; --i >= 0;) { 965 inUse16[i] = false; 966 final int i16 = i * 16; 967 for (int j = 16; --j >= 0;) { 968 if (inUse[i16 + j]) { 969 inUse16[i] = true; 970 } 971 } 972 } 973 974 for (int i = 0; i < 16; i++) { 975 bsW(1, inUse16[i] ? 1 : 0); 976 } 977 978 final OutputStream outShadow = this.out; 979 int bsLiveShadow = this.bsLive; 980 int bsBuffShadow = this.bsBuff; 981 982 for (int i = 0; i < 16; i++) { 983 if (inUse16[i]) { 984 final int i16 = i * 16; 985 for (int j = 0; j < 16; j++) { 986 // inlined: bsW(1, inUse[i16 + j] ? 1 : 0); 987 while (bsLiveShadow >= 8) { 988 outShadow.write(bsBuffShadow >> 24); // write 8-bit 989 bsBuffShadow <<= 8; 990 bsLiveShadow -= 8; 991 } 992 if (inUse[i16 + j]) { 993 bsBuffShadow |= 1 << (32 - bsLiveShadow - 1); 994 } 995 bsLiveShadow++; 996 } 997 } 998 } 999 1000 this.bsBuff = bsBuffShadow; 1001 this.bsLive = bsLiveShadow; 1002 } 1003 1004 private void sendMTFValues5(final int nGroups, final int nSelectors) 1005 throws IOException { 1006 bsW(3, nGroups); 1007 bsW(15, nSelectors); 1008 1009 final OutputStream outShadow = this.out; 1010 final byte[] selectorMtf = this.data.selectorMtf; 1011 1012 int bsLiveShadow = this.bsLive; 1013 int bsBuffShadow = this.bsBuff; 1014 1015 for (int i = 0; i < nSelectors; i++) { 1016 for (int j = 0, hj = selectorMtf[i] & 0xff; j < hj; j++) { 1017 // inlined: bsW(1, 1); 1018 while (bsLiveShadow >= 8) { 1019 outShadow.write(bsBuffShadow >> 24); 1020 bsBuffShadow <<= 8; 1021 bsLiveShadow -= 8; 1022 } 1023 bsBuffShadow |= 1 << (32 - bsLiveShadow - 1); 1024 bsLiveShadow++; 1025 } 1026 1027 // inlined: bsW(1, 0); 1028 while (bsLiveShadow >= 8) { 1029 outShadow.write(bsBuffShadow >> 24); 1030 bsBuffShadow <<= 8; 1031 bsLiveShadow -= 8; 1032 } 1033 // bsBuffShadow |= 0 << (32 - bsLiveShadow - 1); 1034 bsLiveShadow++; 1035 } 1036 1037 this.bsBuff = bsBuffShadow; 1038 this.bsLive = bsLiveShadow; 1039 } 1040 1041 private void sendMTFValues6(final int nGroups, final int alphaSize) 1042 throws IOException { 1043 final byte[][] len = this.data.sendMTFValues_len; 1044 final OutputStream outShadow = this.out; 1045 1046 int bsLiveShadow = this.bsLive; 1047 int bsBuffShadow = this.bsBuff; 1048 1049 for (int t = 0; t < nGroups; t++) { 1050 byte[] len_t = len[t]; 1051 int curr = len_t[0] & 0xff; 1052 1053 // inlined: bsW(5, curr); 1054 while (bsLiveShadow >= 8) { 1055 outShadow.write(bsBuffShadow >> 24); // write 8-bit 1056 bsBuffShadow <<= 8; 1057 bsLiveShadow -= 8; 1058 } 1059 bsBuffShadow |= curr << (32 - bsLiveShadow - 5); 1060 bsLiveShadow += 5; 1061 1062 for (int i = 0; i < alphaSize; i++) { 1063 int lti = len_t[i] & 0xff; 1064 while (curr < lti) { 1065 // inlined: bsW(2, 2); 1066 while (bsLiveShadow >= 8) { 1067 outShadow.write(bsBuffShadow >> 24); // write 8-bit 1068 bsBuffShadow <<= 8; 1069 bsLiveShadow -= 8; 1070 } 1071 bsBuffShadow |= 2 << (32 - bsLiveShadow - 2); 1072 bsLiveShadow += 2; 1073 1074 curr++; /* 10 */ 1075 } 1076 1077 while (curr > lti) { 1078 // inlined: bsW(2, 3); 1079 while (bsLiveShadow >= 8) { 1080 outShadow.write(bsBuffShadow >> 24); // write 8-bit 1081 bsBuffShadow <<= 8; 1082 bsLiveShadow -= 8; 1083 } 1084 bsBuffShadow |= 3 << (32 - bsLiveShadow - 2); 1085 bsLiveShadow += 2; 1086 1087 curr--; /* 11 */ 1088 } 1089 1090 // inlined: bsW(1, 0); 1091 while (bsLiveShadow >= 8) { 1092 outShadow.write(bsBuffShadow >> 24); // write 8-bit 1093 bsBuffShadow <<= 8; 1094 bsLiveShadow -= 8; 1095 } 1096 // bsBuffShadow |= 0 << (32 - bsLiveShadow - 1); 1097 bsLiveShadow++; 1098 } 1099 } 1100 1101 this.bsBuff = bsBuffShadow; 1102 this.bsLive = bsLiveShadow; 1103 } 1104 1105 private void sendMTFValues7() throws IOException { 1106 final Data dataShadow = this.data; 1107 final byte[][] len = dataShadow.sendMTFValues_len; 1108 final int[][] code = dataShadow.sendMTFValues_code; 1109 final OutputStream outShadow = this.out; 1110 final byte[] selector = dataShadow.selector; 1111 final char[] sfmap = dataShadow.sfmap; 1112 final int nMTFShadow = this.nMTF; 1113 1114 int selCtr = 0; 1115 1116 int bsLiveShadow = this.bsLive; 1117 int bsBuffShadow = this.bsBuff; 1118 1119 for (int gs = 0; gs < nMTFShadow;) { 1120 final int ge = Math.min(gs + G_SIZE - 1, nMTFShadow - 1); 1121 final int selector_selCtr = selector[selCtr] & 0xff; 1122 final int[] code_selCtr = code[selector_selCtr]; 1123 final byte[] len_selCtr = len[selector_selCtr]; 1124 1125 while (gs <= ge) { 1126 final int sfmap_i = sfmap[gs]; 1127 1128 // 1129 // inlined: bsW(len_selCtr[sfmap_i] & 0xff, 1130 // code_selCtr[sfmap_i]); 1131 // 1132 while (bsLiveShadow >= 8) { 1133 outShadow.write(bsBuffShadow >> 24); 1134 bsBuffShadow <<= 8; 1135 bsLiveShadow -= 8; 1136 } 1137 final int n = len_selCtr[sfmap_i] & 0xFF; 1138 bsBuffShadow |= code_selCtr[sfmap_i] << (32 - bsLiveShadow - n); 1139 bsLiveShadow += n; 1140 1141 gs++; 1142 } 1143 1144 gs = ge + 1; 1145 selCtr++; 1146 } 1147 1148 this.bsBuff = bsBuffShadow; 1149 this.bsLive = bsLiveShadow; 1150 } 1151 1152 private void moveToFrontCodeAndSend() throws IOException { 1153 bsW(24, this.data.origPtr); 1154 generateMTFValues(); 1155 sendMTFValues(); 1156 } 1157 1158 private void blockSort() { 1159 blockSorter.blockSort(data, last); 1160 } 1161 1162 /* 1163 * Performs Move-To-Front on the Burrows-Wheeler transformed 1164 * buffer, storing the MTFed data in data.sfmap in RUNA/RUNB 1165 * run-length-encoded form. 1166 * 1167 * <p>Keeps track of byte frequencies in data.mtfFreq at the same time.</p> 1168 */ 1169 private void generateMTFValues() { 1170 final int lastShadow = this.last; 1171 final Data dataShadow = this.data; 1172 final boolean[] inUse = dataShadow.inUse; 1173 final byte[] block = dataShadow.block; 1174 final int[] fmap = dataShadow.fmap; 1175 final char[] sfmap = dataShadow.sfmap; 1176 final int[] mtfFreq = dataShadow.mtfFreq; 1177 final byte[] unseqToSeq = dataShadow.unseqToSeq; 1178 final byte[] yy = dataShadow.generateMTFValues_yy; 1179 1180 // make maps 1181 int nInUseShadow = 0; 1182 for (int i = 0; i < 256; i++) { 1183 if (inUse[i]) { 1184 unseqToSeq[i] = (byte) nInUseShadow; 1185 nInUseShadow++; 1186 } 1187 } 1188 this.nInUse = nInUseShadow; 1189 1190 final int eob = nInUseShadow + 1; 1191 1192 for (int i = eob; i >= 0; i--) { 1193 mtfFreq[i] = 0; 1194 } 1195 1196 for (int i = nInUseShadow; --i >= 0;) { 1197 yy[i] = (byte) i; 1198 } 1199 1200 int wr = 0; 1201 int zPend = 0; 1202 1203 for (int i = 0; i <= lastShadow; i++) { 1204 final byte ll_i = unseqToSeq[block[fmap[i]] & 0xff]; 1205 byte tmp = yy[0]; 1206 int j = 0; 1207 1208 while (ll_i != tmp) { 1209 j++; 1210 byte tmp2 = tmp; 1211 tmp = yy[j]; 1212 yy[j] = tmp2; 1213 } 1214 yy[0] = tmp; 1215 1216 if (j == 0) { 1217 zPend++; 1218 } else { 1219 if (zPend > 0) { 1220 zPend--; 1221 while (true) { 1222 if ((zPend & 1) == 0) { 1223 sfmap[wr] = RUNA; 1224 wr++; 1225 mtfFreq[RUNA]++; 1226 } else { 1227 sfmap[wr] = RUNB; 1228 wr++; 1229 mtfFreq[RUNB]++; 1230 } 1231 1232 if (zPend >= 2) { 1233 zPend = (zPend - 2) >> 1; 1234 } else { 1235 break; 1236 } 1237 } 1238 zPend = 0; 1239 } 1240 sfmap[wr] = (char) (j + 1); 1241 wr++; 1242 mtfFreq[j + 1]++; 1243 } 1244 } 1245 1246 if (zPend > 0) { 1247 zPend--; 1248 while (true) { 1249 if ((zPend & 1) == 0) { 1250 sfmap[wr] = RUNA; 1251 wr++; 1252 mtfFreq[RUNA]++; 1253 } else { 1254 sfmap[wr] = RUNB; 1255 wr++; 1256 mtfFreq[RUNB]++; 1257 } 1258 1259 if (zPend >= 2) { 1260 zPend = (zPend - 2) >> 1; 1261 } else { 1262 break; 1263 } 1264 } 1265 } 1266 1267 sfmap[wr] = (char) eob; 1268 mtfFreq[eob]++; 1269 this.nMTF = wr + 1; 1270 } 1271 1272 static final class Data { 1273 1274 // with blockSize 900k 1275 /* maps unsigned byte => "does it occur in block" */ 1276 final boolean[] inUse = new boolean[256]; // 256 byte 1277 final byte[] unseqToSeq = new byte[256]; // 256 byte 1278 final int[] mtfFreq = new int[MAX_ALPHA_SIZE]; // 1032 byte 1279 final byte[] selector = new byte[MAX_SELECTORS]; // 18002 byte 1280 final byte[] selectorMtf = new byte[MAX_SELECTORS]; // 18002 byte 1281 1282 final byte[] generateMTFValues_yy = new byte[256]; // 256 byte 1283 final byte[][] sendMTFValues_len = new byte[N_GROUPS][MAX_ALPHA_SIZE]; // 1548 1284 // byte 1285 final int[][] sendMTFValues_rfreq = new int[N_GROUPS][MAX_ALPHA_SIZE]; // 6192 1286 // byte 1287 final int[] sendMTFValues_fave = new int[N_GROUPS]; // 24 byte 1288 final short[] sendMTFValues_cost = new short[N_GROUPS]; // 12 byte 1289 final int[][] sendMTFValues_code = new int[N_GROUPS][MAX_ALPHA_SIZE]; // 6192 1290 // byte 1291 final byte[] sendMTFValues2_pos = new byte[N_GROUPS]; // 6 byte 1292 final boolean[] sentMTFValues4_inUse16 = new boolean[16]; // 16 byte 1293 1294 final int[] heap = new int[MAX_ALPHA_SIZE + 2]; // 1040 byte 1295 final int[] weight = new int[MAX_ALPHA_SIZE * 2]; // 2064 byte 1296 final int[] parent = new int[MAX_ALPHA_SIZE * 2]; // 2064 byte 1297 1298 // ------------ 1299 // 333408 byte 1300 1301 /* holds the RLEd block of original data starting at index 1. 1302 * After sorting the last byte added to the buffer is at index 1303 * 0. */ 1304 final byte[] block; // 900021 byte 1305 /* maps index in Burrows-Wheeler transformed block => index of 1306 * byte in original block */ 1307 final int[] fmap; // 3600000 byte 1308 final char[] sfmap; // 3600000 byte 1309 // ------------ 1310 // 8433529 byte 1311 // ============ 1312 1313 /** 1314 * Index of original line in Burrows-Wheeler table. 1315 * 1316 * <p>This is the index in fmap that points to the last byte 1317 * of the original data.</p> 1318 */ 1319 int origPtr; 1320 1321 Data(int blockSize100k) { 1322 final int n = blockSize100k * BZip2Constants.BASEBLOCKSIZE; 1323 this.block = new byte[(n + 1 + NUM_OVERSHOOT_BYTES)]; 1324 this.fmap = new int[n]; 1325 this.sfmap = new char[2 * n]; 1326 } 1327 1328 } 1329 1330}