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/compressor_ops.h"
35 #include "wimlib/compress_common.h"
36 #include "wimlib/error.h"
37 #include "wimlib/util.h"
38 #include "wimlib/xpress.h"
46 struct xpress_record_ctx {
47 input_idx_t freqs[XPRESS_NUM_SYMBOLS];
48 struct xpress_match *matches;
51 struct xpress_compressor {
54 struct xpress_match *matches;
55 input_idx_t *prev_tab;
56 u16 codewords[XPRESS_NUM_SYMBOLS];
57 u8 lens[XPRESS_NUM_SYMBOLS];
58 struct xpress_record_ctx record_ctx;
61 /* Intermediate XPRESS match/literal representation. */
63 u16 adjusted_len; /* Match length minus XPRESS_MIN_MATCH_LEN */
64 u16 offset; /* Match offset */
65 /* For literals, offset == 0 and adjusted_len is the literal. */
69 * Writes @match, which is a match given in the intermediate representation for
70 * XPRESS matches, to the output stream @ostream.
72 * @codewords and @lens provide the Huffman code that is being used.
75 xpress_write_match(struct xpress_match match,
76 struct output_bitstream *restrict ostream,
77 const u16 codewords[restrict],
78 const u8 lens[restrict])
80 u8 len_hdr = min(match.adjusted_len, 0xf);
81 u8 offset_bsr = bsr32(match.offset);
82 unsigned sym = XPRESS_NUM_CHARS + ((offset_bsr << 4) | len_hdr);
84 bitstream_put_bits(ostream, codewords[sym], lens[sym]);
86 if (match.adjusted_len >= 0xf) {
87 u8 byte1 = min(match.adjusted_len - 0xf, 0xff);
88 bitstream_put_byte(ostream, byte1);
90 bitstream_put_byte(ostream, match.adjusted_len & 0xff);
91 bitstream_put_byte(ostream, match.adjusted_len >> 8);
94 bitstream_put_bits(ostream, match.offset ^ (1U << offset_bsr), offset_bsr);
98 xpress_write_matches_and_literals(struct output_bitstream *ostream,
99 const struct xpress_match matches[restrict],
100 input_idx_t num_matches,
101 const u16 codewords[restrict],
102 const u8 lens[restrict])
104 for (input_idx_t i = 0; i < num_matches; i++) {
105 if (matches[i].offset) {
107 xpress_write_match(matches[i], ostream, codewords, lens);
110 u8 lit = matches[i].adjusted_len;
111 bitstream_put_bits(ostream, codewords[lit], lens[lit]);
114 bitstream_put_bits(ostream, codewords[XPRESS_END_OF_DATA], lens[XPRESS_END_OF_DATA]);
118 xpress_record_literal(u8 lit, void *_ctx)
120 struct xpress_record_ctx *ctx = _ctx;
122 *(ctx->matches++) = (struct xpress_match) { .offset = 0, .adjusted_len = lit };
126 xpress_record_match(unsigned len, unsigned offset, void *_ctx)
128 struct xpress_record_ctx *ctx = _ctx;
130 XPRESS_ASSERT(len >= XPRESS_MIN_MATCH_LEN);
131 XPRESS_ASSERT(len <= XPRESS_MAX_MATCH_LEN);
132 XPRESS_ASSERT(offset >= XPRESS_MIN_OFFSET);
133 XPRESS_ASSERT(offset <= XPRESS_MAX_OFFSET);
135 unsigned adjusted_len = len - XPRESS_MIN_MATCH_LEN;
136 unsigned len_hdr = min(adjusted_len, 0xf);
137 unsigned sym = XPRESS_NUM_CHARS + ((bsr32(offset) << 4) | len_hdr);
139 XPRESS_ASSERT(sym >= XPRESS_NUM_CHARS);
140 XPRESS_ASSERT(sym < XPRESS_NUM_SYMBOLS);
143 *(ctx->matches++) = (struct xpress_match) { .offset = offset,
144 .adjusted_len = adjusted_len };
147 static const struct lz_params xpress_lz_params = {
148 .min_match = XPRESS_MIN_MATCH_LEN,
149 .max_match = XPRESS_MAX_MATCH_LEN,
150 .max_offset = XPRESS_MAX_OFFSET,
154 .max_lazy_match = 16,
159 xpress_compress(const void *uncompressed_data, size_t uncompressed_size,
160 void *compressed_data, size_t compressed_size_avail, void *_c)
162 struct xpress_compressor *c = _c;
163 u8 *cptr = compressed_data;
164 struct output_bitstream ostream;
165 input_idx_t num_matches;
167 size_t compressed_size;
169 /* XPRESS requires 256 bytes of overhead for the Huffman code, so it's
170 * impossible to compress 256 bytes or less of data to less than the
173 * +1 to take into account that the buffer for compressed data is 1 byte
174 * smaller than the buffer for uncompressed data.
176 * +4 to take into account that init_output_bitstream() requires at
177 * least 4 bytes of data. */
178 if (compressed_size_avail < XPRESS_NUM_SYMBOLS / 2 + 1 + 4)
181 /* Copy the data to a temporary buffer, but only to avoid
182 * inconsequential accesses of uninitialized memory in
183 * lz_analyze_block(). */
184 memcpy(c->window, uncompressed_data, uncompressed_size);
185 memset(c->window + uncompressed_size, 0, 8);
187 /* Determine match/literal sequence to divide the data into. */
188 memset(c->record_ctx.freqs, 0, sizeof(c->record_ctx.freqs));
189 c->record_ctx.matches = c->matches;
190 lz_analyze_block(c->window,
193 xpress_record_literal,
198 num_matches = (c->record_ctx.matches - c->matches);
200 /* Account for end of data symbol. */
201 c->record_ctx.freqs[XPRESS_END_OF_DATA]++;
203 /* Build the Huffman code. */
204 make_canonical_huffman_code(XPRESS_NUM_SYMBOLS, XPRESS_MAX_CODEWORD_LEN,
205 c->record_ctx.freqs, c->lens, c->codewords);
207 /* Output the Huffman code as a series of 512 4-bit lengths. */
208 for (i = 0; i < XPRESS_NUM_SYMBOLS; i += 2)
209 *cptr++ = (c->lens[i] & 0xf) | (c->lens[i + 1] << 4);
211 /* Output the encoded matches/literals. */
212 init_output_bitstream(&ostream, cptr,
213 compressed_size_avail - XPRESS_NUM_SYMBOLS / 2 - 1);
215 xpress_write_matches_and_literals(&ostream, c->matches,
216 num_matches, c->codewords, c->lens);
218 /* Flush any pending data and get the length of the compressed data. */
219 compressed_size = flush_output_bitstream(&ostream);
220 if (compressed_size == ~(input_idx_t)0)
223 compressed_size += XPRESS_NUM_SYMBOLS / 2;
225 #if defined(ENABLE_XPRESS_DEBUG) || defined(ENABLE_VERIFY_COMPRESSION)
226 /* Verify that we really get the same thing back when decompressing. */
228 struct wimlib_decompressor *decompressor;
230 if (0 == wimlib_create_decompressor(WIMLIB_COMPRESSION_TYPE_XPRESS,
236 ret = wimlib_decompress(compressed_data,
241 wimlib_free_decompressor(decompressor);
244 ERROR("Failed to decompress data we "
245 "compressed using XPRESS algorithm");
249 if (memcmp(uncompressed_data, c->window,
252 ERROR("Data we compressed using XPRESS algorithm "
253 "didn't decompress to original");
258 WARNING("Failed to create decompressor for "
259 "data verification!");
264 return compressed_size;
268 xpress_free_compressor(void *_c)
270 struct xpress_compressor *c = _c;
281 xpress_create_compressor(size_t max_window_size,
282 const struct wimlib_compressor_params_header *params,
285 struct xpress_compressor *c;
287 if (max_window_size == 0 || max_window_size > (1U << 26))
288 return WIMLIB_ERR_INVALID_PARAM;
290 c = CALLOC(1, sizeof(struct xpress_compressor));
294 c->window = MALLOC(max_window_size + 8);
295 if (c->window == NULL)
298 c->max_window_size = max_window_size;
300 c->matches = MALLOC(max_window_size * sizeof(c->matches[0]));
301 if (c->matches == NULL)
304 c->prev_tab = MALLOC(max_window_size * sizeof(c->prev_tab[0]));
305 if (c->prev_tab == NULL)
312 xpress_free_compressor(c);
313 return WIMLIB_ERR_NOMEM;
316 const struct compressor_ops xpress_compressor_ops = {
317 .create_compressor = xpress_create_compressor,
318 .compress = xpress_compress,
319 .free_compressor = xpress_free_compressor,