4 * LZX compression routines, originally based on code written by Matthew T.
5 * Russotto (liblzxcomp), but heavily modified.
9 * Copyright (C) 2002 Matthew T. Russotto
10 * Copyright (C) 2012 Eric Biggers
12 * This file is part of wimlib, a library for working with WIM files.
14 * wimlib is free software; you can redistribute it and/or modify it under the
15 * terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 3 of the License, or (at your option)
19 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
20 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
21 * A PARTICULAR PURPOSE. See the GNU General Public License for more
24 * You should have received a copy of the GNU General Public License
25 * along with wimlib; if not, see http://www.gnu.org/licenses/.
30 * This file provides lzx_compress(), a function to compress an in-memory buffer
31 * of data using LZX compression, as used in the WIM file format.
33 * Please see the comments in lzx-decomp.c for more information about this
36 * One thing to keep in mind is that there is no sliding window, since the
37 * window is always the entirety of a WIM chunk, which is at most WIM_CHUNK_SIZE
40 * The basic compression algorithm used here should be familiar if you are
41 * familiar with Huffman trees and with other LZ77 and Huffman-based formats
42 * such as DEFLATE. Otherwise it can be quite tricky to understand. Basically
43 * it is the following:
45 * - Preprocess the input data (LZX-specific)
46 * - Go through the input data and determine matches. This part is based on
47 * code from zlib, and a hash table of 3-character strings is used to
48 * accelerate the process of finding matches.
49 * - Build the Huffman trees based on the frequencies of symbols determined
50 * while recording matches.
51 * - Output the block header, including the Huffman trees; then output the
52 * compressed stream of matches and literal characters.
54 * It is possible for a WIM chunk to include multiple LZX blocks, since for some
55 * input data this will produce a better compression ratio (especially since
56 * each block can include new Huffman codes). However, producing multiple LZX
57 * blocks from one input chunk is not yet implemented.
67 /* Structure to contain the Huffman codes for the main, length, and aligned
70 u16 main_codewords[LZX_MAINTREE_NUM_SYMBOLS];
71 u8 main_lens[LZX_MAINTREE_NUM_SYMBOLS];
73 u16 len_codewords[LZX_LENTREE_NUM_SYMBOLS];
74 u8 len_lens[LZX_LENTREE_NUM_SYMBOLS];
76 u16 aligned_codewords[LZX_ALIGNEDTREE_NUM_SYMBOLS];
77 u8 aligned_lens[LZX_ALIGNEDTREE_NUM_SYMBOLS];
80 struct lzx_freq_tables {
81 u32 main_freq_table[LZX_MAINTREE_NUM_SYMBOLS];
82 u32 len_freq_table[LZX_LENTREE_NUM_SYMBOLS];
83 u32 aligned_freq_table[LZX_ALIGNEDTREE_NUM_SYMBOLS];
86 /* Returns the LZX position slot that corresponds to a given formatted offset.
88 * Logically, this returns the smallest i such that
89 * formatted_offset >= lzx_position_base[i].
91 * The actual implementation below takes advantage of the regularity of the
92 * numbers in the lzx_position_base array to calculate the slot directly from
93 * the formatted offset without actually looking at the array.
95 static inline unsigned lzx_get_position_slot(unsigned formatted_offset)
99 * Slots 36-49 (formatted_offset >= 262144) can be found by
100 * (formatted_offset/131072) + 34 == (formatted_offset >> 17) + 34;
101 * however, this check for formatted_offset >= 262144 is commented out
102 * because WIM chunks cannot be that large.
104 if (formatted_offset >= 262144) {
105 return (formatted_offset >> 17) + 34;
109 /* Note: this part here only works if:
111 * 2 <= formatted_offset < 655360
113 * It is < 655360 because the frequency of the position bases
114 * increases starting at the 655360 entry, and it is >= 2
115 * because the below calculation fails if the most significant
116 * bit is lower than the 2's place. */
117 wimlib_assert(formatted_offset >= 2 && formatted_offset < 655360);
118 unsigned mssb_idx = bsr32(formatted_offset);
119 return (mssb_idx << 1) |
120 ((formatted_offset >> (mssb_idx - 1)) & 1);
124 static u32 lzx_record_literal(u8 literal, void *__main_freq_tab)
126 u32 *main_freq_tab = __main_freq_tab;
127 main_freq_tab[literal]++;
131 /* Equivalent to lzx_extra_bits[position_slot] except position_slot must be
132 * between 2 and 37 */
133 static inline unsigned lzx_get_num_extra_bits(unsigned position_slot)
136 return lzx_extra_bits[position_slot];
138 wimlib_assert(position_slot >= 2 && position_slot <= 37);
139 return (position_slot >> 1) - 1;
142 /* Constructs a match from an offset and a length, and updates the LRU queue and
143 * the frequency of symbols in the main, length, and aligned offset alphabets.
144 * The return value is a 32-bit number that provides the match in an
145 * intermediate representation documented below. */
146 static u32 lzx_record_match(unsigned match_offset, unsigned match_len,
147 void *__freq_tabs, void *__queue)
149 struct lzx_freq_tables *freq_tabs = __freq_tabs;
150 struct lru_queue *queue = __queue;
151 unsigned formatted_offset;
152 unsigned position_slot;
153 unsigned position_footer = 0;
158 unsigned adjusted_match_len;
160 wimlib_assert(match_len >= LZX_MIN_MATCH && match_len <= LZX_MAX_MATCH);
161 wimlib_assert(match_offset != 0);
163 /* If possible, encode this offset as a repeated offset. */
164 if (match_offset == queue->R0) {
165 formatted_offset = 0;
167 } else if (match_offset == queue->R1) {
168 swap(queue->R0, queue->R1);
169 formatted_offset = 1;
171 } else if (match_offset == queue->R2) {
172 swap(queue->R0, queue->R2);
173 formatted_offset = 2;
176 /* Not a repeated offset. */
178 /* offsets of 0, 1, and 2 are reserved for the repeated offset
179 * codes, so non-repeated offsets must be encoded as 3+. The
180 * minimum offset is 1, so encode the offsets offset by 2. */
181 formatted_offset = match_offset + LZX_MIN_MATCH;
183 queue->R2 = queue->R1;
184 queue->R1 = queue->R0;
185 queue->R0 = match_offset;
187 /* The (now-formatted) offset will actually be encoded as a
188 * small position slot number that maps to a certain hard-coded
189 * offset (position base), followed by a number of extra bits---
190 * the position footer--- that are added to the position base to
191 * get the original formatted offset. */
193 position_slot = lzx_get_position_slot(formatted_offset);
194 position_footer = formatted_offset &
195 ((1 << lzx_get_num_extra_bits(position_slot)) - 1);
198 adjusted_match_len = match_len - LZX_MIN_MATCH;
200 /* Pack the position slot, position footer, and match length into an
201 * intermediate representation.
204 * ---- -----------------------------------------------------------
206 * 31 1 if a match, 0 if a literal.
208 * 30-25 position slot. This can be at most 50, so it will fit in 6
211 * 8-24 position footer. This is the offset of the real formatted
212 * offset from the position base. This can be at most 17 bits
213 * (since lzx_extra_bits[LZX_NUM_POSITION_SLOTS - 1] is 17).
215 * 0-7 length of match, offset by 2. This can be at most
216 * (LZX_MAX_MATCH - 2) == 255, so it will fit in 8 bits. */
218 (position_slot << 25) |
219 (position_footer << 8) |
220 (adjusted_match_len);
222 /* The match length must be at least 2, so let the adjusted match length
223 * be the match length minus 2.
225 * If it is less than 7, the adjusted match length is encoded as a 3-bit
226 * number offset by 2. Otherwise, the 3-bit length header is all 1's
227 * and the actual adjusted length is given as a symbol encoded with the
228 * length tree, offset by 7.
230 if (adjusted_match_len < LZX_NUM_PRIMARY_LENS) {
231 len_header = adjusted_match_len;
233 len_header = LZX_NUM_PRIMARY_LENS;
234 len_footer = adjusted_match_len - LZX_NUM_PRIMARY_LENS;
235 freq_tabs->len_freq_table[len_footer]++;
237 len_pos_header = (position_slot << 3) | len_header;
239 wimlib_assert(len_pos_header < LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS);
241 freq_tabs->main_freq_table[len_pos_header + LZX_NUM_CHARS]++;
244 * if (lzx_extra_bits[position_slot] >= 3) */
245 if (position_slot >= 8)
246 freq_tabs->aligned_freq_table[position_footer & 7]++;
252 * Writes a compressed literal match to the output.
254 * @out: The output bitstream.
255 * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM)
256 * @match: The match, encoded as a 32-bit number.
257 * @codes: Pointer to a structure that contains the codewords for the
258 * main, length, and aligned offset Huffman codes.
260 static int lzx_write_match(struct output_bitstream *out, int block_type,
261 u32 match, const struct lzx_codes *codes)
263 /* low 8 bits are the match length minus 2 */
264 unsigned match_len_minus_2 = match & 0xff;
265 /* Next 17 bits are the position footer */
266 unsigned position_footer = (match >> 8) & 0x1ffff; /* 17 bits */
267 /* Next 6 bits are the position slot. */
268 unsigned position_slot = (match >> 25) & 0x3f; /* 6 bits */
271 unsigned len_pos_header;
272 unsigned main_symbol;
273 unsigned num_extra_bits;
274 unsigned verbatim_bits;
275 unsigned aligned_bits;
278 /* If the match length is less than MIN_MATCH (= 2) +
279 * NUM_PRIMARY_LENS (= 7), the length header contains
280 * the match length minus MIN_MATCH, and there is no
283 * Otherwise, the length header contains
284 * NUM_PRIMARY_LENS, and the length footer contains
285 * the match length minus NUM_PRIMARY_LENS minus
287 if (match_len_minus_2 < LZX_NUM_PRIMARY_LENS) {
288 len_header = match_len_minus_2;
289 /* No length footer-- mark it with a special
291 len_footer = (unsigned)(-1);
293 len_header = LZX_NUM_PRIMARY_LENS;
294 len_footer = match_len_minus_2 - LZX_NUM_PRIMARY_LENS;
297 /* Combine the position slot with the length header into
298 * a single symbol that will be encoded with the main
300 len_pos_header = (position_slot << 3) | len_header;
302 /* The actual main symbol is offset by LZX_NUM_CHARS because
303 * values under LZX_NUM_CHARS are used to indicate a literal
304 * byte rather than a match. */
305 main_symbol = len_pos_header + LZX_NUM_CHARS;
307 /* Output main symbol. */
308 ret = bitstream_put_bits(out, codes->main_codewords[main_symbol],
309 codes->main_lens[main_symbol]);
313 /* If there is a length footer, output it using the
314 * length Huffman code. */
315 if (len_footer != (unsigned)(-1)) {
316 ret = bitstream_put_bits(out, codes->len_codewords[len_footer],
317 codes->len_lens[len_footer]);
322 wimlib_assert(position_slot < LZX_NUM_POSITION_SLOTS);
324 num_extra_bits = lzx_extra_bits[position_slot];
326 /* For aligned offset blocks with at least 3 extra bits, output the
327 * verbatim bits literally, then the aligned bits encoded using the
328 * aligned offset tree. Otherwise, only the verbatim bits need to be
330 if ((block_type == LZX_BLOCKTYPE_ALIGNED) && (num_extra_bits >= 3)) {
332 verbatim_bits = position_footer >> 3;
333 ret = bitstream_put_bits(out, verbatim_bits,
338 aligned_bits = (position_footer & 7);
339 ret = bitstream_put_bits(out,
340 codes->aligned_codewords[aligned_bits],
341 codes->aligned_lens[aligned_bits]);
345 /* verbatim bits is the same as the position
346 * footer, in this case. */
347 ret = bitstream_put_bits(out, position_footer, num_extra_bits);
355 * Writes all compressed literals in a block, both matches and literal bytes, to
356 * the output bitstream.
358 * @out: The output bitstream.
359 * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM)
360 * @match_tab[]: The array of matches that will be output. It has length
361 * of @num_compressed_literals.
362 * @num_compressed_literals: Number of compressed literals to be output.
363 * @codes: Pointer to a structure that contains the codewords for the
364 * main, length, and aligned offset Huffman codes.
366 static int lzx_write_compressed_literals(struct output_bitstream *ostream,
368 const u32 match_tab[],
369 unsigned num_compressed_literals,
370 const struct lzx_codes *codes)
376 for (i = 0; i < num_compressed_literals; i++) {
377 match = match_tab[i];
379 /* High bit of the match indicates whether the match is an
380 * actual match (1) or a literal uncompressed byte (0) */
381 if (match & 0x80000000) {
383 ret = lzx_write_match(ostream, block_type, match,
389 wimlib_assert(match < LZX_NUM_CHARS);
390 ret = bitstream_put_bits(ostream,
391 codes->main_codewords[match],
392 codes->main_lens[match]);
401 * Writes a compressed Huffman tree to the output, preceded by the pretree for
404 * The Huffman tree is represented in the output as a series of path lengths
405 * from which the canonical Huffman code can be reconstructed. The path lengths
406 * themselves are compressed using a separate Huffman code, the pretree, which
407 * consists of LZX_PRETREE_NUM_SYMBOLS (= 20) symbols that cover all possible code
408 * lengths, plus extra codes for repeated lengths. The path lengths of the
409 * pretree precede the path lengths of the larger code and are uncompressed,
410 * consisting of 20 entries of 4 bits each.
412 * @out: The bitstream for the compressed output.
413 * @lens: The code lengths for the Huffman tree, indexed by symbol.
414 * @num_symbols: The number of symbols in the code.
416 static int lzx_write_compressed_tree(struct output_bitstream *out,
417 const u8 lens[], unsigned num_symbols)
419 /* Frequencies of the length symbols, including the RLE symbols (NOT the
420 * actual lengths themselves). */
421 unsigned pretree_freqs[LZX_PRETREE_NUM_SYMBOLS];
422 u8 pretree_lens[LZX_PRETREE_NUM_SYMBOLS];
423 u16 pretree_codewords[LZX_PRETREE_NUM_SYMBOLS];
424 u8 output_syms[num_symbols * 2];
425 unsigned output_syms_idx;
426 unsigned cur_run_len;
429 unsigned additional_bits;
433 ZERO_ARRAY(pretree_freqs);
435 /* Since the code word lengths use a form of RLE encoding, the goal here
436 * is to find each run of identical lengths when going through them in
437 * symbol order (including runs of length 1). For each run, as many
438 * lengths are encoded using RLE as possible, and the rest are output
441 * output_syms[] will be filled in with the length symbols that will be
442 * output, including RLE codes, not yet encoded using the pre-tree.
444 * cur_run_len keeps track of how many code word lengths are in the
445 * current run of identical lengths.
449 for (i = 1; i <= num_symbols; i++) {
451 if (i != num_symbols && lens[i] == lens[i - 1]) {
452 /* Still in a run--- keep going. */
457 /* Run ended! Check if it is a run of zeroes or a run of
460 /* The symbol that was repeated in the run--- not to be confused
461 * with the length *of* the run (cur_run_len) */
462 len_in_run = lens[i - 1];
464 if (len_in_run == 0) {
465 /* A run of 0's. Encode it in as few length
466 * codes as we can. */
468 /* The magic length 18 indicates a run of 20 + n zeroes,
469 * where n is an uncompressed literal 5-bit integer that
470 * follows the magic length. */
471 while (cur_run_len >= 20) {
473 additional_bits = min(cur_run_len - 20, 0x1f);
475 output_syms[output_syms_idx++] = 18;
476 output_syms[output_syms_idx++] = additional_bits;
477 cur_run_len -= 20 + additional_bits;
480 /* The magic length 17 indicates a run of 4 + n zeroes,
481 * where n is an uncompressed literal 4-bit integer that
482 * follows the magic length. */
483 while (cur_run_len >= 4) {
484 additional_bits = min(cur_run_len - 4, 0xf);
486 output_syms[output_syms_idx++] = 17;
487 output_syms[output_syms_idx++] = additional_bits;
488 cur_run_len -= 4 + additional_bits;
493 /* A run of nonzero lengths. */
495 /* The magic length 19 indicates a run of 4 + n
496 * nonzeroes, where n is a literal bit that follows the
497 * magic length, and where the value of the lengths in
498 * the run is given by an extra length symbol, encoded
499 * with the pretree, that follows the literal bit.
501 * The extra length symbol is encoded as a difference
502 * from the length of the codeword for the first symbol
503 * in the run in the previous tree.
505 while (cur_run_len >= 4) {
506 additional_bits = (cur_run_len > 4);
507 delta = -(char)len_in_run;
511 pretree_freqs[(unsigned char)delta]++;
512 output_syms[output_syms_idx++] = 19;
513 output_syms[output_syms_idx++] = additional_bits;
514 output_syms[output_syms_idx++] = delta;
515 cur_run_len -= 4 + additional_bits;
519 /* Any remaining lengths in the run are outputted without RLE,
520 * as a difference from the length of that codeword in the
522 while (cur_run_len--) {
523 delta = -(char)len_in_run;
527 pretree_freqs[(unsigned char)delta]++;
528 output_syms[output_syms_idx++] = delta;
534 wimlib_assert(output_syms_idx < ARRAY_LEN(output_syms));
536 /* Build the pretree from the frequencies of the length symbols. */
538 make_canonical_huffman_code(LZX_PRETREE_NUM_SYMBOLS,
539 LZX_MAX_CODEWORD_LEN,
540 pretree_freqs, pretree_lens,
543 /* Write the lengths of the pretree codes to the output. */
544 for (i = 0; i < LZX_PRETREE_NUM_SYMBOLS; i++)
545 bitstream_put_bits(out, pretree_lens[i],
546 LZX_PRETREE_ELEMENT_SIZE);
548 /* Write the length symbols, encoded with the pretree, to the output. */
551 while (i < output_syms_idx) {
552 pretree_sym = output_syms[i++];
554 bitstream_put_bits(out, pretree_codewords[pretree_sym],
555 pretree_lens[pretree_sym]);
556 switch (pretree_sym) {
558 bitstream_put_bits(out, output_syms[i++], 4);
561 bitstream_put_bits(out, output_syms[i++], 5);
564 bitstream_put_bits(out, output_syms[i++], 1);
565 bitstream_put_bits(out,
566 pretree_codewords[output_syms[i]],
567 pretree_lens[output_syms[i]]);
577 /* Builds the canonical Huffman code for the main tree, the length tree, and the
578 * aligned offset tree. */
579 static void lzx_make_huffman_codes(const struct lzx_freq_tables *freq_tabs,
580 struct lzx_codes *codes)
582 make_canonical_huffman_code(LZX_MAINTREE_NUM_SYMBOLS,
583 LZX_MAX_CODEWORD_LEN,
584 freq_tabs->main_freq_table,
586 codes->main_codewords);
588 make_canonical_huffman_code(LZX_LENTREE_NUM_SYMBOLS,
589 LZX_MAX_CODEWORD_LEN,
590 freq_tabs->len_freq_table,
592 codes->len_codewords);
594 make_canonical_huffman_code(LZX_ALIGNEDTREE_NUM_SYMBOLS, 8,
595 freq_tabs->aligned_freq_table,
597 codes->aligned_codewords);
600 /* Do the 'E8' preprocessing, where the targets of x86 CALL instructions were
601 * changed from relative offsets to absolute offsets. This type of
602 * preprocessing can be used on any binary data even if it is not actually
603 * machine code. It seems to always be used in WIM files, even though there is
604 * no bit to indicate that it actually is used, unlike in the LZX compressed
605 * format as used in other file formats such as the cabinet format, where a bit
606 * is reserved for that purpose. */
607 static void do_call_insn_preprocessing(u8 uncompressed_data[],
608 unsigned uncompressed_data_len)
611 int file_size = LZX_MAGIC_FILESIZE;
615 /* Not enabled in the last 6 bytes, which means the 5-byte call
616 * instruction cannot start in the last *10* bytes. */
617 while (i < uncompressed_data_len - 10) {
618 if (uncompressed_data[i] != 0xe8) {
622 rel_offset = le32_to_cpu(*(int32_t*)(uncompressed_data + i + 1));
624 if (rel_offset >= -i && rel_offset < file_size) {
625 if (rel_offset < file_size - i) {
626 /* "good translation" */
627 abs_offset = rel_offset + i;
629 /* "compensating translation" */
630 abs_offset = rel_offset - file_size;
632 *(int32_t*)(uncompressed_data + i + 1) = cpu_to_le32(abs_offset);
639 static const struct lz_params lzx_lz_params = {
641 /* LZX_MIN_MATCH == 2, but 2-character matches are rarely useful; the
642 * minimum match for compression is set to 3 instead. */
645 .max_match = LZX_MAX_MATCH,
646 .good_match = LZX_MAX_MATCH,
647 .nice_match = LZX_MAX_MATCH,
648 .max_chain_len = LZX_MAX_MATCH,
649 .max_lazy_match = LZX_MAX_MATCH,
654 * Performs LZX compression on a block of data.
656 * @__uncompressed_data: Pointer to the data to be compressed.
657 * @uncompressed_len: Length, in bytes, of the data to be compressed.
658 * @compressed_data: Pointer to a location at least (@uncompressed_len - 1)
659 * bytes long into which the compressed data may be
661 * @compressed_len_ret: A pointer to an unsigned int into which the length of
662 * the compressed data may be returned.
664 * Returns zero if compression was successfully performed. In that case
665 * @compressed_data and @compressed_len_ret will contain the compressed data and
666 * its length. A return value of nonzero means that compressing the data did
667 * not reduce its size, and @compressed_data will not contain the full
670 int lzx_compress(const void *__uncompressed_data, unsigned uncompressed_len,
671 void *compressed_data, unsigned *compressed_len_ret)
673 struct output_bitstream ostream;
674 u8 uncompressed_data[uncompressed_len + LZX_MAX_MATCH];
675 struct lzx_freq_tables freq_tabs;
676 struct lzx_codes codes;
677 u32 match_tab[uncompressed_len];
678 struct lru_queue queue;
679 unsigned num_matches;
680 unsigned compressed_len;
683 int block_type = LZX_BLOCKTYPE_ALIGNED;
685 if (uncompressed_len < 100)
688 memset(&freq_tabs, 0, sizeof(freq_tabs));
693 /* The input data must be preprocessed. To avoid changing the original
694 * input, copy it to a temporary buffer. */
695 memcpy(uncompressed_data, __uncompressed_data, uncompressed_len);
697 /* Before doing any actual compression, do the call instruction (0xe8
698 * byte) translation on the uncompressed data. */
699 do_call_insn_preprocessing(uncompressed_data, uncompressed_len);
701 /* Determine the sequence of matches and literals that will be output,
702 * and in the process, keep counts of the number of times each symbol
703 * will be output, so that the Huffman trees can be made. */
705 num_matches = lz_analyze_block(uncompressed_data, uncompressed_len,
706 match_tab, lzx_record_match,
707 lzx_record_literal, &freq_tabs,
708 &queue, freq_tabs.main_freq_table,
711 lzx_make_huffman_codes(&freq_tabs, &codes);
713 /* Initialize the output bitstream. */
714 init_output_bitstream(&ostream, compressed_data, uncompressed_len - 1);
716 /* The first three bits tell us what kind of block it is, and are one
717 * of the LZX_BLOCKTYPE_* values. */
718 bitstream_put_bits(&ostream, block_type, 3);
720 /* The next bit indicates whether the block size is the default (32768),
721 * indicated by a 1 bit, or whether the block size is given by the next
722 * 16 bits, indicated by a 0 bit. */
723 if (uncompressed_len == 32768) {
724 bitstream_put_bits(&ostream, 1, 1);
726 bitstream_put_bits(&ostream, 0, 1);
727 bitstream_put_bits(&ostream, uncompressed_len, 16);
730 /* Write out the aligned offset tree. Note that M$ lies and says that
731 * the aligned offset tree comes after the length tree, but that is
732 * wrong; it actually is before the main tree. */
733 if (block_type == LZX_BLOCKTYPE_ALIGNED)
734 for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++)
735 bitstream_put_bits(&ostream, codes.aligned_lens[i],
736 LZX_ALIGNEDTREE_ELEMENT_SIZE);
738 /* Write the pre-tree and lengths for the first LZX_NUM_CHARS symbols in the
740 ret = lzx_write_compressed_tree(&ostream, codes.main_lens,
745 /* Write the pre-tree and symbols for the rest of the main tree. */
746 ret = lzx_write_compressed_tree(&ostream, codes.main_lens +
748 LZX_MAINTREE_NUM_SYMBOLS -
753 /* Write the pre-tree and symbols for the length tree. */
754 ret = lzx_write_compressed_tree(&ostream, codes.len_lens,
755 LZX_LENTREE_NUM_SYMBOLS);
759 /* Write the compressed literals. */
760 ret = lzx_write_compressed_literals(&ostream, block_type,
761 match_tab, num_matches, &codes);
765 ret = flush_output_bitstream(&ostream);
769 compressed_len = ostream.bit_output - (u8*)compressed_data;
771 *compressed_len_ret = compressed_len;
773 #ifdef ENABLE_VERIFY_COMPRESSION
774 /* Verify that we really get the same thing back when decompressing. */
775 u8 buf[uncompressed_len];
776 ret = lzx_decompress(compressed_data, compressed_len, buf,
779 ERROR("lzx_compress(): Failed to decompress data we compressed");
783 for (i = 0; i < uncompressed_len; i++) {
784 if (buf[i] != *((u8*)__uncompressed_data + i)) {
785 ERROR("lzx_compress(): Data we compressed didn't "
786 "decompress to the original data (difference at "
787 "byte %u of %u)", i + 1, uncompressed_len);