4 * XPRESS compression routines.
6 * See the comments in xpress-decompress.c about the XPRESS format.
10 * Copyright (C) 2012, 2013 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/.
33 #include "wimlib/assert.h"
34 #include "wimlib/compress.h"
35 #include "wimlib/error.h"
36 #include "wimlib/util.h"
37 #include "wimlib/xpress.h"
43 * Writes @match, which is a match given in the intermediate representation for
44 * XPRESS matches, to the output stream @ostream.
46 * @codewords and @lens provide the Huffman code that is being used.
49 xpress_write_match(struct output_bitstream *ostream, u32 match,
50 const u16 codewords[], const u8 lens[])
52 u32 adjusted_match_len = match & 0xffff;
53 u32 match_offset = match >> 16;
54 u32 len_hdr = min(adjusted_match_len, 0xf);
55 u32 offset_bsr = bsr32(match_offset);
56 u32 sym = len_hdr | (offset_bsr << 4) | XPRESS_NUM_CHARS;
59 ret = bitstream_put_bits(ostream, codewords[sym], lens[sym]);
63 if (adjusted_match_len >= 0xf) {
64 u8 byte1 = min(adjusted_match_len - 0xf, 0xff);
65 ret = bitstream_put_byte(ostream, byte1);
69 ret = bitstream_put_two_bytes(ostream, adjusted_match_len);
74 return bitstream_put_bits(ostream,
75 match_offset ^ (1 << offset_bsr), offset_bsr);
79 xpress_write_compressed_literals(struct output_bitstream *ostream,
80 const u32 match_tab[],
82 const u16 codewords[],
85 for (unsigned i = 0; i < num_matches; i++) {
87 u32 match = match_tab[i];
89 if (match >= XPRESS_NUM_CHARS) /* match */
90 ret = xpress_write_match(ostream, match, codewords,
92 else /* literal byte */
93 ret = bitstream_put_bits(ostream, codewords[match],
98 return bitstream_put_bits(ostream, codewords[XPRESS_END_OF_DATA],
99 lens[XPRESS_END_OF_DATA]);
103 xpress_record_literal(u8 literal, void *__freq_tab)
105 freq_t *freq_tab = __freq_tab;
111 xpress_record_match(unsigned match_offset, unsigned match_len,
112 void *freq_tab, void *ignore)
114 wimlib_assert(match_len >= XPRESS_MIN_MATCH &&
115 match_len <= XPRESS_MAX_MATCH);
116 wimlib_assert(match_offset >= XPRESS_MIN_OFFSET &&
117 match_offset <= XPRESS_MAX_OFFSET);
120 * The intermediate representation of XPRESS matches is as follows:
123 * ---- -----------------------------------------------------------
125 * 16-31 match offset (XPRESS_MIN_OFFSET < x < XPRESS_MAX_OFFSET)
127 * 0-15 adjusted match length (0 <= x <= XPRESS_MAX_MATCH - XPRESS_MIN_MATCH)
129 * Literals are simply represented as themselves and can be
130 * distinguished from matches by the fact that only literals will have
131 * the upper three bytes completely clear. */
133 u32 adjusted_match_len = match_len - XPRESS_MIN_MATCH;
134 u32 len_hdr = min(adjusted_match_len, 0xf);
135 u32 offset_bsr = bsr32(match_offset);
136 u32 sym = len_hdr | (offset_bsr << 4) | XPRESS_NUM_CHARS;
137 ((freq_t*)freq_tab)[sym]++;
138 return adjusted_match_len | (match_offset << 16);
141 static const struct lz_params xpress_lz_params = {
142 .min_match = XPRESS_MIN_MATCH,
143 .max_match = XPRESS_MAX_MATCH,
147 .max_lazy_match = 16,
151 /* Documented in wimlib.h */
153 wimlib_xpress_compress(const void *__uncompressed_data,
154 unsigned uncompressed_len, void *__compressed_data)
156 u8 *compressed_data = __compressed_data;
157 struct output_bitstream ostream;
158 u32 match_tab[uncompressed_len];
159 freq_t freq_tab[XPRESS_NUM_SYMBOLS];
160 u16 codewords[XPRESS_NUM_SYMBOLS];
161 u8 lens[XPRESS_NUM_SYMBOLS];
162 unsigned num_matches;
163 unsigned compressed_len;
166 u8 uncompressed_data[uncompressed_len + 8];
168 memcpy(uncompressed_data, __uncompressed_data, uncompressed_len);
169 memset(uncompressed_data + uncompressed_len, 0, 8);
171 wimlib_assert(uncompressed_len <= 32768);
173 /* XPRESS requires 256 bytes of overhead for the Huffman tables, so it's
174 * impossible cannot compress 256 bytes or less of data to less than the
177 * +1 to take into account that the buffer for compressed data is 1 byte
178 * smaller than the buffer for uncompressed data.
180 * +4 to take into account that init_output_bitstream() requires at
181 * least 4 bytes of data. */
182 if (uncompressed_len < XPRESS_NUM_SYMBOLS / 2 + 1 + 4)
185 ZERO_ARRAY(freq_tab);
186 num_matches = lz_analyze_block(uncompressed_data, uncompressed_len,
187 match_tab, xpress_record_match,
188 xpress_record_literal, freq_tab,
192 freq_tab[XPRESS_END_OF_DATA]++;
194 make_canonical_huffman_code(XPRESS_NUM_SYMBOLS, XPRESS_MAX_CODEWORD_LEN,
195 freq_tab, lens, codewords);
199 * It's tempting to output the 512 Huffman codeword lengths using the
200 * bitstream_put_bits() function. However, this is NOT correct because
201 * bitstream_put_bits() will output 2 bytes at a time in little-endian
202 * order, which is the order that is needed for the compressed literals.
203 * However, the bytes in the lengths table are in order, so they need to
204 * be written one at a time without using bitstream_put_bits().
206 * Because of this, init_output_bitstream() is not called until after
207 * the lengths table is output.
209 for (i = 0; i < XPRESS_NUM_SYMBOLS; i += 2)
210 *compressed_data++ = (lens[i] & 0xf) | (lens[i + 1] << 4);
212 init_output_bitstream(&ostream, compressed_data,
213 uncompressed_len - XPRESS_NUM_SYMBOLS / 2 - 1);
215 ret = xpress_write_compressed_literals(&ostream, match_tab,
216 num_matches, codewords, lens);
220 /* Flush any bits that are buffered. */
221 ret = flush_output_bitstream(&ostream);
225 /* Assert that there are no output bytes between the ostream.output
226 * pointer and the ostream.next_bit_output pointer. This can only
227 * happen if bytes had been written at the ostream.output pointer before
228 * the last bit word was written to the stream. But, this does not
229 * occur since xpress_write_match() always finishes by writing some bits
230 * (a Huffman symbol), and the bitstream was just flushed. */
231 wimlib_assert(ostream.output - ostream.next_bit_output == 2);
233 /* The length of the compressed data is supposed to be the value of the
234 * ostream.output pointer before flushing, which is now the
235 * output.next_bit_output pointer after flushing.
237 * There will be an extra 2 bytes at the ostream.bit_output pointer,
238 * which is zeroed out. (These 2 bytes may be either the last bytes in
239 * the compressed data, in which case they are actually unnecessary, or
240 * they may precede a number of bytes embedded into the bitstream.) */
241 if (ostream.bit_output >
242 (const u8*)__compressed_data + uncompressed_len - 3)
244 *(u16*)ostream.bit_output = cpu_to_le16(0);
245 compressed_len = ostream.next_bit_output - (const u8*)__compressed_data;
247 wimlib_assert(compressed_len <= uncompressed_len - 1);
249 #ifdef ENABLE_VERIFY_COMPRESSION
250 /* Verify that we really get the same thing back when decompressing. */
252 u8 buf[uncompressed_len];
253 ret = wimlib_xpress_decompress(__compressed_data, compressed_len,
254 buf, uncompressed_len);
256 ERROR("xpress_compress(): Failed to decompress data we "
260 for (i = 0; i < uncompressed_len; i++) {
261 if (buf[i] != uncompressed_data[i]) {
262 ERROR("xpress_compress(): Data we compressed didn't "
263 "decompress to the original data (difference at "
264 "byte %u of %u)", i + 1, uncompressed_len);
270 return compressed_len;