4 * XPRESS compression routines.
6 * See the comments in xpress-decomp.c about the XPRESS format.
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/.
34 * Writes @match, which is a match given in the intermediate representation for
35 * XPRESS matches, to the output stream @ostream.
37 * @codewords and @lens provide the Huffman code that is being used.
39 static int xpress_write_match(struct output_bitstream *ostream, u32 match,
40 const u16 codewords[], const u8 lens[])
50 main_sym = (match & 0xff);
51 huff_sym = main_sym + XPRESS_NUM_CHARS;
52 ret = bitstream_put_bits(ostream, codewords[huff_sym], lens[huff_sym]);
56 offset_bsr = main_sym >> 4;
58 match_len = (match >> 8) & 0xff;
59 match_offset = (match >> 16);
62 match_len -= XPRESS_MIN_MATCH;
63 if (match_len >= 0xf) {
64 byte1 = (u8)(match_len - 0xf);
65 ret = bitstream_put_byte(ostream, byte1);
69 ret = bitstream_put_two_bytes(ostream, match_len);
74 return bitstream_put_bits(ostream, match_offset ^ (1 << offset_bsr),
78 static int xpress_write_compressed_literals(struct output_bitstream *ostream,
79 const u32 match_tab[],
81 const u16 codewords[],
88 for (i = 0; i < num_matches; i++) {
90 if (match >= XPRESS_NUM_CHARS) /* match */
91 ret = xpress_write_match(ostream, match, codewords,
93 else /* literal byte */
94 ret = bitstream_put_bits(ostream, codewords[match],
99 return bitstream_put_bits(ostream, codewords[256], lens[256]);
102 static u32 xpress_record_literal(u8 literal, void *__freq_tab)
104 u32 *freq_tab = __freq_tab;
109 static u32 xpress_record_match(uint match_offset, uint match_len,
110 void *__freq_tab, void *ignore)
112 u32 *freq_tab = __freq_tab;
117 wimlib_assert(match_len >= XPRESS_MIN_MATCH &&
118 match_len <= XPRESS_MAX_MATCH);
119 wimlib_assert(match_offset > 0);
121 len_hdr = min(match_len - XPRESS_MIN_MATCH, 15);
122 offset_bsr = bsr32(match_offset);
123 match = (offset_bsr << 4) | len_hdr;
124 freq_tab[match + XPRESS_NUM_CHARS]++;
125 match |= match_len << 8;
126 match |= match_offset << 16;
130 static const struct lz_params xpress_lz_params = {
132 .max_match = XPRESS_MAX_MATCH,
136 .max_lazy_match = 16,
141 * Performs XPRESS compression on a block of data.
143 * @__uncompressed_data: Pointer to the data to be compressed.
144 * @uncompressed_len: Length, in bytes, of the data to be compressed.
145 * @__compressed_data: Pointer to a location at least (@uncompressed_len - 1)
146 * bytes long into which the compressed data may be
148 * @compressed_len_ret: A pointer to an unsigned int into which the length of
149 * the compressed data may be returned.
151 * Returns zero if compression was successfully performed. In that case
152 * @compressed_data and @compressed_len_ret will contain the compressed data and
153 * its length. A return value of nonzero means that compressing the data did
154 * not reduce its size, and @compressed_data will not contain the full
157 int xpress_compress(const void *__uncompressed_data, uint uncompressed_len,
158 void *__compressed_data, uint *compressed_len_ret)
160 const u8 *uncompressed_data = __uncompressed_data;
161 u8 *compressed_data = __compressed_data;
162 struct output_bitstream ostream;
163 u32 match_tab[uncompressed_len];
164 u32 freq_tab[XPRESS_NUM_SYMBOLS];
165 u16 codewords[XPRESS_NUM_SYMBOLS];
166 u8 lens[XPRESS_NUM_SYMBOLS];
172 XPRESS_DEBUG("uncompressed_len = %u", uncompressed_len);
174 if (uncompressed_len < 300)
177 ZERO_ARRAY(freq_tab);
179 num_matches = lz_analyze_block(uncompressed_data, uncompressed_len,
180 match_tab, xpress_record_match,
181 xpress_record_literal, freq_tab,
185 XPRESS_DEBUG("using %u matches", num_matches);
189 make_canonical_huffman_code(XPRESS_NUM_SYMBOLS, XPRESS_MAX_CODEWORD_LEN,
190 freq_tab, lens, codewords);
194 * It's tempting to output the 512 Huffman codeword lengths using the
195 * bitstream_put_bits() function. However, this is NOT correct because
196 * bitstream_put_bits() will output 2 bytes at a time in little-endian
197 * order, which is the order that is needed for the compressed literals.
198 * However, the bytes in the lengths table are in order, so they need to
199 * be written one at a time without using bitstream_put_bits().
201 * Because of this, init_output_bitstream() is not called until after
202 * the lengths table is output.
204 for (i = 0; i < XPRESS_NUM_SYMBOLS; i += 2)
205 *compressed_data++ = (lens[i] & 0xf) | (lens[i + 1] << 4);
207 init_output_bitstream(&ostream, compressed_data,
208 uncompressed_len - XPRESS_NUM_SYMBOLS / 2 - 1);
210 ret = xpress_write_compressed_literals(&ostream, match_tab,
211 num_matches, codewords, lens);
215 /* Flush any bits that are buffered. */
216 ret = flush_output_bitstream(&ostream);
220 /* Assert that there are no output bytes between the ostream.output
221 * pointer and the ostream.next_bit_output pointer. This can only
222 * happen if bytes had been written at the ostream.output pointer before
223 * the last bit word was written to the stream. But, this does not
224 * occur since xpress_write_match() always finishes by writing some bits
225 * (a Huffman symbol), and the bitstream was just flushed. */
226 wimlib_assert(ostream.output - ostream.next_bit_output == 2);
229 * The length of the compressed data is supposed to be the value of the
230 * ostream.output pointer before flushing, which is now the
231 * output.next_bit_output pointer after flushing.
233 * There will be an extra 2 bytes at the ostream.bit_output pointer,
234 * which is zeroed out. (These 2 bytes may be either the last bytes in
235 * the compressed data, in which case they are actually unnecessary, or
236 * they may precede a number of bytes embedded into the bitstream.)
238 if (ostream.bit_output >
239 (const u8*)__compressed_data + uncompressed_len - 3)
241 *(u16*)ostream.bit_output = cpu_to_le16(0);
242 compressed_len = ostream.next_bit_output - (const u8*)__compressed_data;
244 wimlib_assert(compressed_len <= uncompressed_len - 1);
246 XPRESS_DEBUG("Compressed %u => %u bytes",
247 uncompressed_len, compressed_len);
249 *compressed_len_ret = compressed_len;
251 #ifdef ENABLE_VERIFY_COMPRESSION
252 /* Verify that we really get the same thing back when decompressing. */
253 XPRESS_DEBUG("Verifying the compressed data.");
254 u8 buf[uncompressed_len];
255 ret = xpress_decompress(__compressed_data, compressed_len, buf,
258 ERROR("xpress_compress(): Failed to decompress data we "
262 for (i = 0; i < uncompressed_len; i++) {
263 if (buf[i] != uncompressed_data[i]) {
264 ERROR("xpress_compress(): Data we compressed didn't "
265 "decompress to the original data (difference at "
266 "byte %u of %u)", i + 1, uncompressed_len);
270 XPRESS_DEBUG("Compression verified to be correct.");