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 General Public License as published by the Free
14 * Software Foundation; either version 3 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 General Public License for more
22 * You should have received a copy of the GNU General Public License
23 * along with wimlib; if not, see http://www.gnu.org/licenses/.
26 #include "decompress.h"
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_assert2(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 reading %zu bytes from "
50 "bitstream (only have %u bytes left)",
51 n, stream->data_bytes_left);
54 memcpy(p, stream->data, n);
56 stream->data_bytes_left -= n;
58 /* It's possible to copy an odd number of bytes and leave the stream in
59 * an inconsistent state. Fix it by reading the next byte, if it is
61 if ((n & 1) && stream->data_bytes_left != 0) {
63 stream->data_bytes_left--;
64 stream->bitbuf |= (input_bitbuf_t)(*stream->data) <<
65 (sizeof(input_bitbuf_t) * 8 - 8);
72 * Builds a fast huffman decoding table from an array that gives the length of
73 * the codeword for each symbol in the alphabet. Originally based on code
74 * written by David Tritscher (taken the original LZX decompression code); also
75 * heavily modified to add some optimizations used in the zlib code, as well as
78 * @decode_table: The array in which to create the fast huffman decoding
79 * table. It must have a length of at least
80 * (2**table_bits) + 2 * num_syms to guarantee
81 * that there is enough space.
83 * @num_syms: Total number of symbols in the Huffman tree.
85 * @table_bits: Any symbols with a code length of table_bits or less can
86 * be decoded in one lookup of the table. 2**table_bits
87 * must be greater than or equal to @num_syms if there are
88 * any Huffman codes longer than @table_bits.
90 * @lens: An array of length @num_syms, indexable by symbol, that
91 * gives the length of the Huffman codeward for that
92 * symbol. Because the Huffman tree is in canonical form,
93 * it can be reconstructed by only knowing the length of
94 * the codeword for each symbol. It is assumed, but not
95 * checked, that every length is less than
98 * @max_codeword_len: The longest codeword length allowed in the compression
101 * Returns 0 on success; returns -1 if the length values do not correspond to a
102 * valid Huffman tree.
104 * The format of the Huffamn decoding table is as follows. The first (1 <<
105 * table_bits) entries of the table are indexed by chunks of the input of size
106 * @table_bits. If the next Huffman codeword in the input happens to have a
107 * length of exactly @table_bits, the symbol is simply read directly from the
108 * decoding table. Alternatively, if the next Huffman codeword has length _less
109 * than_ @table_bits, the symbol is also read directly from the decode table;
110 * this is possible because every entry in the table that is indexed by an
111 * integer that has the shorter codeword as a binary prefix is filled in with
112 * the appropriate symbol. If a codeword has length n <= table_bits, it will
113 * have 2**(table_bits - n) possible suffixes, and thus that many entries in the
116 * It's a bit more complicated if the next Huffman codeword has length of more
117 * than @table_bits. The table entry indexed by the first @table_bits of that
118 * codeword cannot give the appropriate symbol directly, because that entry is
119 * guaranteed to be referenced by the Huffman codewords of multiple symbols.
120 * And while the LZX compression format does not allow codes longer than 16
121 * bits, a table of size (2 ** 16) = 65536 entries would be too slow to create.
123 * There are several different ways to make it possible to look up the symbols
124 * for codewords longer than @table_bits. One way is to make the entries for
125 * the prefixes of length @table_bits of those entries be pointers to additional
126 * decoding tables that are indexed by some number of additional bits of the
127 * codeword. The technique used here is a bit simpler, however: just store the
128 * needed subtrees of the Huffman tree in the decoding table after the lookup
129 * entries, beginning at index (2**table_bits). Real pointers are replaced by
130 * indices into the decoding table, and symbol entries are distinguished from
131 * pointers by the fact that values less than @num_syms must be symbol values.
133 int make_huffman_decode_table(u16 decode_table[], unsigned num_syms,
134 unsigned table_bits, const u8 lens[],
135 unsigned max_codeword_len)
137 unsigned len_counts[max_codeword_len + 1];
138 u16 sorted_syms[num_syms];
139 unsigned offsets[max_codeword_len + 1];
140 const unsigned table_num_entries = 1 << table_bits;
142 /* accumulate lengths for codes */
143 for (unsigned i = 0; i <= max_codeword_len; i++)
146 for (unsigned sym = 0; sym < num_syms; sym++) {
147 wimlib_assert2(lens[sym] <= max_codeword_len);
148 len_counts[lens[sym]]++;
151 /* check for an over-subscribed or incomplete set of lengths */
153 for (unsigned len = 1; len <= max_codeword_len; len++) {
155 left -= len_counts[len];
156 if (left < 0) { /* over-subscribed */
157 ERROR("Invalid Huffman code (over-subscribed)");
161 if (left != 0) /* incomplete set */{
162 if (left == 1 << max_codeword_len) {
163 /* Empty code--- okay in XPRESS and LZX */
164 memset(decode_table, 0,
165 table_num_entries * sizeof(decode_table[0]));
168 ERROR("Invalid Huffman code (incomplete set)");
173 /* Generate offsets into symbol table for each length for sorting */
175 for (unsigned len = 1; len < max_codeword_len; len++)
176 offsets[len + 1] = offsets[len] + len_counts[len];
178 /* Sort symbols primarily by length and secondarily by symbol order.
179 * This is basically a count-sort over the codeword lengths.
180 * In the process, calculate the number of symbols that have nonzero
181 * length and are therefore used in the symbol stream. */
182 unsigned num_used_syms = 0;
183 for (unsigned sym = 0; sym < num_syms; sym++) {
184 if (lens[sym] != 0) {
185 sorted_syms[offsets[lens[sym]]++] = sym;
190 /* Fill entries for codewords short enough for a direct mapping. We can
191 * take advantage of the ordering of the codewords, since the Huffman
192 * code is canonical. It must be the case that all the codewords of
193 * some length L numerically precede all the codewords of length L + 1.
194 * Furthermore, if we have 2 symbols A and B with the same codeword
195 * length but symbol A is sorted before symbol B, then then we know that
196 * the codeword for A numerically precedes the codeword for B. */
197 unsigned decode_table_pos = 0;
200 wimlib_assert2(num_used_syms != 0);
202 unsigned sym = sorted_syms[i];
203 unsigned codeword_len = lens[sym];
204 if (codeword_len > table_bits)
207 unsigned num_entries = 1 << (table_bits - codeword_len);
209 (sizeof(unsigned long) / sizeof(decode_table[0])))
211 wimlib_assert2(decode_table_pos % 4 == 0);
212 BUILD_BUG_ON(sizeof(unsigned long) != 4 &&
213 sizeof(unsigned long) != 8);
215 unsigned long *p = (unsigned long *)&decode_table[decode_table_pos];
216 unsigned long n = num_entries /
217 (sizeof(unsigned long) /
218 sizeof(decode_table[0]));
219 unsigned long v = sym;
220 if (sizeof(unsigned long) >= 4)
222 if (sizeof(unsigned long) >= 8)
228 decode_table_pos += num_entries;
231 decode_table[decode_table_pos++] = sym;
232 } while (--num_entries);
234 wimlib_assert2(decode_table_pos <= table_num_entries);
235 if (++i == num_used_syms) {
236 wimlib_assert2(decode_table_pos == table_num_entries);
237 /* No codewords were longer than @table_bits, so the
238 * table is now entirely filled with the codewords. */
243 wimlib_assert2(i < num_used_syms);
244 wimlib_assert2(decode_table_pos < table_num_entries);
246 /* Fill in the remaining entries, which correspond to codes longer than
249 * First, zero out the rest of the entries. This is necessary so that
250 * the entries appear as "unallocated" in the next part. */
252 unsigned j = decode_table_pos;
255 } while (++j != table_num_entries);
258 /* Assert that 2**table_bits is at least num_syms. If this wasn't the
259 * case, we wouldn't be able to distinguish pointer entries from symbol
261 wimlib_assert2(table_num_entries >= num_syms);
263 /* The current Huffman codeword */
264 unsigned cur_codeword = decode_table_pos;
266 /* The tree nodes are allocated starting at decode_table[1 <<
267 * table_bits]. Remember that the full size of the table, including the
268 * extra space for the tree nodes, is actually 2**table_bits + 2 *
269 * num_syms slots, while table_num_entries is only 2**table_Bits. */
270 unsigned next_free_tree_slot = table_num_entries;
272 /* Go through every codeword of length greater than @table_bits,
273 * primarily in order of codeword length and secondarily in order of
275 unsigned prev_codeword_len = table_bits;
277 unsigned sym = sorted_syms[i];
278 unsigned codeword_len = lens[sym];
279 unsigned extra_bits = codeword_len - table_bits;
282 cur_codeword <<= (codeword_len - prev_codeword_len);
283 prev_codeword_len = codeword_len;
285 /* index of the current node; find it from the prefix of the
286 * current Huffman codeword. */
287 unsigned node_idx = cur_codeword >> extra_bits;
288 wimlib_assert2(node_idx < table_num_entries);
290 /* Go through each bit of the current Huffman codeword beyond
291 * the prefix of length @table_bits and walk the tree,
292 * allocating any slots that have not yet been allocated. */
295 /* If the current tree node points to nowhere
296 * but we need to follow it, allocate a new node
297 * for it to point to. */
298 if (decode_table[node_idx] == 0) {
299 decode_table[node_idx] = next_free_tree_slot;
300 decode_table[next_free_tree_slot++] = 0;
301 decode_table[next_free_tree_slot++] = 0;
302 wimlib_assert2(next_free_tree_slot <=
303 table_num_entries + 2 * num_syms);
306 /* Set node_idx to left child */
307 node_idx = decode_table[node_idx];
309 /* Is the next bit 0 or 1? If 0, go left (already done).
310 * If 1, go right by incrementing node_idx. */
312 node_idx += (cur_codeword >> extra_bits) & 1;
313 } while (extra_bits != 0);
315 /* node_idx is now the index of the leaf entry into which the
316 * actual symbol will go. */
317 decode_table[node_idx] = sym;
319 /* cur_codeword is always incremented because this is
320 * how canonical Huffman codes are generated (add 1 for
321 * each code, then left shift whenever the code length
324 } while (++i != num_used_syms);
328 /* Reads a Huffman-encoded symbol when it is known there are less than
329 * MAX_CODE_LEN bits remaining in the bitstream. */
330 int read_huffsym_near_end_of_input(struct input_bitstream *istream,
331 const u16 decode_table[],
337 unsigned bitsleft = istream->bitsleft;
342 if (table_bits > bitsleft) {
345 key_bits = bitstream_peek_bits(istream, key_size) <<
346 (table_bits - key_size);
348 key_size = table_bits;
349 bitsleft -= table_bits;
350 key_bits = bitstream_peek_bits(istream, table_bits);
353 sym = decode_table[key_bits];
354 if (sym >= num_syms) {
355 bitstream_remove_bits(istream, key_size);
358 ERROR("Input stream exhausted");
361 key_bits = sym + bitstream_peek_bits(istream, 1);
362 bitstream_remove_bits(istream, 1);
364 } while ((sym = decode_table[key_bits]) >= num_syms);
366 bitstream_remove_bits(istream, lens[sym]);