4 * Functions used for decompression.
8 * Copyright (C) 2012 Eric Biggers
10 * This file is part of wimlib, a library for working with WIM files.
12 * wimlib is free software; you can redistribute it and/or modify it under the
13 * terms of the GNU Lesser General Public License as published by the Free
14 * Software Foundation; either version 2.1 of the License, or (at your option)
17 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
18 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
19 * A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
22 * You should have received a copy of the GNU Lesser General Public License
23 * along with wimlib; if not, see http://www.gnu.org/licenses/.
29 /* Reads @n bytes from the bitstream @stream into the location pointed to by @dest.
30 * The bitstream must be 16-bit aligned. */
31 int bitstream_read_bytes(struct input_bitstream *stream, size_t n, void *dest)
33 /* Precondition: The bitstream is 16-byte aligned. */
34 wimlib_assert(stream->bitsleft % 16 == 0);
38 /* Get the bytes currently in the buffer variable. */
39 while (stream->bitsleft != 0) {
42 *p++ = bitstream_peek_bits(stream, 8);
43 bitstream_remove_bits(stream, 8);
46 /* Get the rest directly from the pointer to the data. Of course, it's
47 * necessary to check there are really n bytes available. */
48 if (n > stream->data_bytes_left) {
49 ERROR("Unexpected end of input when "
50 "reading %zu bytes from bitstream "
51 "(only have %u bytes left)\n", n,
52 stream->data_bytes_left);
55 memcpy(p, stream->data, n);
57 stream->data_bytes_left -= n;
59 /* It's possible to copy an odd number of bytes and leave the stream in
60 * an inconsistent state. Fix it by reading the next byte, if it is
62 if ((n & 1) && stream->data_bytes_left != 0) {
64 stream->data_bytes_left--;
65 stream->bitbuf |= (input_bitbuf_t)(*stream->data) <<
66 (sizeof(input_bitbuf_t) * 8 - 8);
72 /* Aligns the bitstream on a 16-bit boundary.
74 * Note: M$'s idea of "alignment" means that for some reason, a 16-bit word
75 * should be skipped over if the buffer happens to already be aligned on such a
76 * boundary. This only applies for realigning the stream after the blocktype
77 * and length fields of an uncompressed block, however; it does not apply when
78 * realigning the stream after the end of the uncompressed block.
80 int align_input_bitstream(struct input_bitstream *stream,
81 bool skip_word_if_aligned)
84 if (stream->bitsleft % 16 != 0) {
85 bitstream_remove_bits(stream, stream->bitsleft % 16);
86 } else if (skip_word_if_aligned) {
87 if (stream->bitsleft == 0) {
88 ret = bitstream_ensure_bits(stream, 16);
90 ERROR("Unexpected end of input when "
91 "aligning bitstream!\n");
95 bitstream_remove_bits(stream, 16);
101 * Builds a fast huffman decoding table from a canonical huffman code lengths
102 * table. Based on code written by David Tritscher.
104 * @decode_table: The array in which to create the fast huffman decoding
105 * table. It must have a length of at least
106 * (2**num_bits) + 2 * num_syms to guarantee
107 * that there is enough space.
109 * @num_syms: Total number of symbols in the Huffman tree.
111 * @num_bits: Any symbols with a code length of num_bits or less can be
112 * decoded in one lookup of the table. 2**num_bits
113 * must be greater than or equal to @num_syms if there are
114 * any Huffman codes longer than @num_bits.
116 * @lens: An array of length @num_syms, indexable by symbol, that
117 * gives the length of that symbol. Because the Huffman
118 * tree is in canonical form, it can be reconstructed by
119 * only knowing the length of the code for each symbol.
121 * @make_codeword_len: An integer that gives the longest possible codeword
124 * Returns 0 on success; returns 1 if the length values do not correspond to a
125 * valid Huffman tree, or if there are codes of length greater than @num_bits
126 * but 2**num_bits < num_syms.
128 * What exactly is the format of the fast Huffman decoding table? The first
129 * (1 << num_bits) entries of the table are indexed by chunks of the input of
130 * size @num_bits. If the next Huffman code in the input happens to have a
131 * length of exactly @num_bits, the symbol is simply read directly from the
132 * decoding table. Alternatively, if the next Huffman code has length _less
133 * than_ @num_bits, the symbol is also read directly from the decode table; this
134 * is possible because every entry in the table that is indexed by an integer
135 * that has the shorter code as a binary prefix is filled in with the
136 * appropriate symbol. If a code has length n <= num_bits, it will have
137 * 2**(num_bits - n) possible suffixes, and thus that many entries in the
140 * It's a bit more complicated if the next Huffman code has length of more than
141 * @num_bits. The table entry indexed by the first @num_bits of that code
142 * cannot give the appropriate symbol directly, because that entry is guaranteed
143 * to be referenced by the Huffman codes for multiple symbols. And while the
144 * LZX compression format does not allow codes longer than 16 bits, a table of
145 * size (2 ** 16) = 65536 entries would be too slow to create.
147 * There are several different ways to make it possible to look up the symbols
148 * for codes longer than @num_bits. A common way is to make the entries for the
149 * prefixes of length @num_bits of those entries be pointers to additional
150 * decoding tables that are indexed by some number of additional bits of the
151 * code symbol. The technique used here is a bit simpler, however. We just
152 * store the needed subtrees of the Huffman tree in the decoding table after the
153 * lookup entries, beginning at index (2**num_bits). Real pointers are
154 * replaced by indices into the decoding table, and we distinguish symbol
155 * entries from pointers by the fact that values less than @num_syms must be
158 int make_huffman_decode_table(u16 decode_table[], uint num_syms,
159 uint num_bits, const u8 lens[],
162 /* Number of entries in the decode table. */
163 u32 table_num_entries = 1 << num_bits;
165 /* Current position in the decode table. */
166 u32 decode_table_pos = 0;
168 /* Fill entries for codes short enough for a direct mapping. Here we
169 * are taking advantage of the ordering of the codes, since they are for
170 * a canonical Huffman tree. It must be the case that all the codes of
171 * some length @code_length, zero-extended or one-extended, numerically
172 * precede all the codes of length @code_length + 1. Furthermore, if we
173 * have 2 symbols A and B, such that A is listed before B in the lens
174 * array, and both symbols have the same code length, then we know that
175 * the code for A numerically precedes the code for B.
177 for (uint code_len = 1; code_len <= num_bits; code_len++) {
179 /* Number of entries that a code of length @code_length would
181 u32 code_num_entries = 1 << (num_bits - code_len);
184 /* For each symbol of length @code_len, fill in its entries in
185 * the decode table. */
186 for (uint sym = 0; sym < num_syms; sym++) {
188 if (lens[sym] != code_len)
192 /* Check for table overrun. This can only happen if the
193 * given lengths do not correspond to a valid Huffman
195 if (decode_table_pos >= table_num_entries) {
196 ERROR("Huffman decoding table overrun: "
197 "pos = %u, num_entries = %u\n",
203 /* Fill all possible lookups of this symbol with
204 * the symbol itself. */
205 for (uint i = 0; i < code_num_entries; i++)
206 decode_table[decode_table_pos + i] = sym;
208 /* Increment the position in the decode table by
209 * the number of entries that were just filled
211 decode_table_pos += code_num_entries;
215 /* If all entries of the decode table have been filled in, there are no
216 * codes longer than num_bits, so we are done filling in the decode
218 if (decode_table_pos == table_num_entries)
221 /* Otherwise, fill in the remaining entries, which correspond to codes longer
225 /* First, zero out the rest of the entries; this is necessary so
226 * that the entries appear as "unallocated" in the next part. */
227 for (uint i = decode_table_pos; i < table_num_entries; i++)
230 /* Assert that 2**num_bits is at least num_syms. If this wasn't the
231 * case, we wouldn't be able to distinguish pointer entries from symbol
233 wimlib_assert((1 << num_bits) >= num_syms);
236 /* The current Huffman code. */
237 uint current_code = decode_table_pos;
239 /* The tree nodes are allocated starting at
240 * decode_table[table_num_entries]. Remember that the full size of the
241 * table, including the extra space for the tree nodes, is actually
242 * 2**num_bits + 2 * num_syms slots, while table_num_entries is only
244 uint next_free_tree_slot = table_num_entries;
246 /* Go through every codeword of length greater than @num_bits. Note:
247 * the LZX format guarantees that the codeword length can be at most 16
249 for (uint code_len = num_bits + 1; code_len <= max_code_len;
253 for (uint sym = 0; sym < num_syms; sym++) {
254 if (lens[sym] != code_len)
258 /* i is the index of the current node; find it from the
259 * prefix of the current Huffman code. */
260 uint i = current_code >> (code_len - num_bits);
262 if (i >= (1 << num_bits)) {
263 ERROR("Invalid canonical Huffman code!\n");
267 /* Go through each bit of the current Huffman code
268 * beyond the prefix of length num_bits and walk the
269 * tree, "allocating" slots that have not yet been
271 for (int bit_num = num_bits + 1; bit_num <= code_len; bit_num++) {
273 /* If the current tree node points to nowhere
274 * but we need to follow it, allocate a new node
275 * for it to point to. */
276 if (decode_table[i] == 0) {
277 decode_table[i] = next_free_tree_slot;
278 decode_table[next_free_tree_slot++] = 0;
279 decode_table[next_free_tree_slot++] = 0;
284 /* Is the next bit 0 or 1? If 0, go left;
285 * otherwise, go right (by incrementing i by 1) */
286 int bit_pos = code_len - bit_num;
288 int bit = (current_code & (1 << bit_pos)) >>
293 /* i is now the index of the leaf entry into which the
294 * actual symbol will go. */
295 decode_table[i] = sym;
297 /* Increment decode_table_pos only if the prefix of the
298 * Huffman code changes. */
299 if (current_code >> (code_len - num_bits) !=
300 (current_code + 1) >> (code_len - num_bits))
303 /* current_code is always incremented because this is
304 * how canonical Huffman codes are generated (add 1 for
305 * each code, then left shift whenever the code length
312 /* If the lengths really represented a valid Huffman tree, all
313 * @table_num_entries in the table will have been filled. However, it
314 * is also possible that the tree is completely empty (as noted
315 * earlier) with all 0 lengths, and this is expected to succeed. */
317 if (decode_table_pos != table_num_entries) {
319 for (uint i = 0; i < num_syms; i++) {
321 ERROR("Lengths do not form a valid "
322 "canonical Huffman tree "
323 "(only filled %u of %u decode "
324 "table slots)!\n", decode_table_pos,
333 /* Reads a Huffman-encoded symbol when it is known there are less than
334 * MAX_CODE_LEN bits remaining in the bitstream. */
335 static int read_huffsym_near_end_of_input(struct input_bitstream *istream,
336 const u16 decode_table[],
342 uint bitsleft = istream->bitsleft;
347 if (table_bits > bitsleft) {
350 key_bits = bitstream_peek_bits(istream, key_size) <<
351 (table_bits - key_size);
353 key_size = table_bits;
354 bitsleft -= table_bits;
355 key_bits = bitstream_peek_bits(istream, table_bits);
358 sym = decode_table[key_bits];
359 if (sym >= num_syms) {
360 bitstream_remove_bits(istream, key_size);
363 ERROR("Input stream exhausted!\n");
366 key_bits = sym + bitstream_peek_bits(istream, 1);
367 bitstream_remove_bits(istream, 1);
369 } while ((sym = decode_table[key_bits]) >= num_syms);
371 bitstream_remove_bits(istream, lens[sym]);
378 * Reads a Huffman-encoded symbol from a bitstream.
380 * This function may be called hundreds of millions of times when extracting a
381 * large WIM file. I'm not sure it could be made much faster that it is,
382 * especially since there isn't enough time to make a big table that allows
383 * decoding multiple symbols per lookup. But if extracting files to a hard
384 * disk, the IO will be the bottleneck anyway.
386 * @buf: The input buffer from which the symbol will be read.
387 * @decode_table: The fast Huffman decoding table for the Huffman tree.
388 * @lengths: The table that gives the length of the code for each
390 * @num_symbols: The number of symbols in the Huffman code.
391 * @table_bits: Huffman codes this length or less can be looked up
392 * directory in the decode_table, as the
393 * decode_table contains 2**table_bits entries.
395 int read_huffsym(struct input_bitstream *stream,
396 const u16 decode_table[],
398 unsigned num_symbols,
401 unsigned max_codeword_len)
403 /* In the most common case, there are at least max_codeword_len bits
404 * remaining in the stream. */
405 if (bitstream_ensure_bits(stream, max_codeword_len) == 0) {
407 /* Use the next table_bits of the input as an index into the
409 u16 key_bits = bitstream_peek_bits(stream, table_bits);
411 u16 sym = decode_table[key_bits];
413 /* If the entry in the decode table is not a valid symbol, it is
414 * the offset of the root of its Huffman subtree. */
415 if (sym >= num_symbols) {
416 bitstream_remove_bits(stream, table_bits);
418 key_bits = sym + bitstream_peek_bits(stream, 1);
419 bitstream_remove_bits(stream, 1);
421 wimlib_assert(key_bits < num_symbols * 2 +
423 } while ((sym = decode_table[key_bits]) >= num_symbols);
425 wimlib_assert(lengths[sym] <= table_bits);
426 bitstream_remove_bits(stream, lengths[sym]);
431 /* Otherwise, we must be careful to use only the bits that are
432 * actually remaining. Don't inline this part since it is very
434 return read_huffsym_near_end_of_input(stream, decode_table, lengths,
435 num_symbols, table_bits, n);