WIMBoot: Update docs
[wimlib] / src / xpress-compress.c
1 /*
2  * xpress-compress.c
3  *
4  * XPRESS compression routines.
5  *
6  * See the comments in xpress-decompress.c about the XPRESS format.
7  */
8
9 /*
10  * Copyright (C) 2012, 2013 Eric Biggers
11  *
12  * This file is part of wimlib, a library for working with WIM files.
13  *
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)
17  * any later version.
18  *
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
22  * details.
23  *
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/.
26  */
27
28 #ifdef HAVE_CONFIG_H
29 #  include "config.h"
30 #endif
31
32 #include "wimlib.h"
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/lz_hash.h"
38 #include "wimlib/util.h"
39 #include "wimlib/xpress.h"
40
41 #include <string.h>
42
43 struct xpress_record_ctx {
44         input_idx_t freqs[XPRESS_NUM_SYMBOLS];
45         struct xpress_match *matches;
46 };
47
48 struct xpress_compressor {
49         u8 *window;
50         u32 max_window_size;
51         struct xpress_match *matches;
52         input_idx_t *prev_tab;
53         u16 codewords[XPRESS_NUM_SYMBOLS];
54         u8 lens[XPRESS_NUM_SYMBOLS];
55         struct xpress_record_ctx record_ctx;
56 };
57
58 /* Intermediate XPRESS match/literal representation.  */
59 struct xpress_match {
60         u16 adjusted_len;  /* Match length minus XPRESS_MIN_MATCH_LEN */
61         u16 offset;        /* Match offset */
62         /* For literals, offset == 0 and adjusted_len is the literal.  */
63 };
64
65 /*
66  * Writes @match, which is a match given in the intermediate representation for
67  * XPRESS matches, to the output stream @ostream.
68  *
69  * @codewords and @lens provide the Huffman code that is being used.
70  */
71 static void
72 xpress_write_match(struct xpress_match match,
73                    struct output_bitstream *restrict ostream,
74                    const u16 codewords[restrict],
75                    const u8 lens[restrict])
76 {
77         u8 len_hdr = min(match.adjusted_len, 0xf);
78         u8 offset_bsr = bsr32(match.offset);
79         unsigned sym = XPRESS_NUM_CHARS + ((offset_bsr << 4) | len_hdr);
80
81         bitstream_put_bits(ostream, codewords[sym], lens[sym]);
82
83         if (match.adjusted_len >= 0xf) {
84                 u8 byte1 = min(match.adjusted_len - 0xf, 0xff);
85                 bitstream_put_byte(ostream, byte1);
86                 if (byte1 == 0xff) {
87                         bitstream_put_byte(ostream, match.adjusted_len & 0xff);
88                         bitstream_put_byte(ostream, match.adjusted_len >> 8);
89                 }
90         }
91         bitstream_put_bits(ostream, match.offset ^ (1U << offset_bsr), offset_bsr);
92 }
93
94 static void
95 xpress_write_matches_and_literals(struct output_bitstream *ostream,
96                                   const struct xpress_match matches[restrict],
97                                   input_idx_t num_matches,
98                                   const u16 codewords[restrict],
99                                   const u8 lens[restrict])
100 {
101         for (input_idx_t i = 0; i < num_matches; i++) {
102                 if (matches[i].offset) {
103                         /* Real match  */
104                         xpress_write_match(matches[i], ostream, codewords, lens);
105                 } else {
106                         /* Literal byte  */
107                         u8 lit = matches[i].adjusted_len;
108                         bitstream_put_bits(ostream, codewords[lit], lens[lit]);
109                 }
110         }
111         bitstream_put_bits(ostream, codewords[XPRESS_END_OF_DATA], lens[XPRESS_END_OF_DATA]);
112 }
113
114 static void
115 xpress_record_literal(u8 lit, void *_ctx)
116 {
117         struct xpress_record_ctx *ctx = _ctx;
118         ctx->freqs[lit]++;
119         *(ctx->matches++) = (struct xpress_match) { .offset = 0, .adjusted_len = lit };
120 }
121
122 static void
123 xpress_record_match(unsigned len, unsigned offset, void *_ctx)
124 {
125         struct xpress_record_ctx *ctx = _ctx;
126
127         XPRESS_ASSERT(len >= XPRESS_MIN_MATCH_LEN);
128         XPRESS_ASSERT(len <= XPRESS_MAX_MATCH_LEN);
129         XPRESS_ASSERT(offset >= XPRESS_MIN_OFFSET);
130         XPRESS_ASSERT(offset <= XPRESS_MAX_OFFSET);
131
132         unsigned adjusted_len = len - XPRESS_MIN_MATCH_LEN;
133         unsigned len_hdr = min(adjusted_len, 0xf);
134         unsigned sym = XPRESS_NUM_CHARS + ((bsr32(offset) << 4) | len_hdr);
135
136         XPRESS_ASSERT(sym >= XPRESS_NUM_CHARS);
137         XPRESS_ASSERT(sym < XPRESS_NUM_SYMBOLS);
138
139         ctx->freqs[sym]++;
140         *(ctx->matches++) = (struct xpress_match) { .offset = offset,
141                                                     .adjusted_len = adjusted_len };
142 }
143
144 static const struct lz_params xpress_lz_params = {
145         .min_match      = XPRESS_MIN_MATCH_LEN,
146         .max_match      = XPRESS_MAX_MATCH_LEN,
147         .max_offset     = XPRESS_MAX_OFFSET,
148         .good_match     = 16,
149         .nice_match     = 32,
150         .max_chain_len  = 16,
151         .max_lazy_match = 16,
152         .too_far        = 4096,
153 };
154
155 static size_t
156 xpress_compress(const void *uncompressed_data, size_t uncompressed_size,
157                 void *compressed_data, size_t compressed_size_avail, void *_c)
158 {
159         struct xpress_compressor *c = _c;
160         u8 *cptr = compressed_data;
161         struct output_bitstream ostream;
162         input_idx_t num_matches;
163         input_idx_t i;
164         size_t compressed_size;
165
166         /* XPRESS requires 256 bytes of overhead for the Huffman code, so it's
167          * impossible to compress 256 bytes or less of data to less than the
168          * input size.
169          *
170          * +1 to take into account that the buffer for compressed data is 1 byte
171          * smaller than the buffer for uncompressed data.
172          *
173          * +4 to take into account that init_output_bitstream() requires at
174          * least 4 bytes of data.  */
175         if (compressed_size_avail < XPRESS_NUM_SYMBOLS / 2 + 1 + 4)
176                 return 0;
177
178         /* Copy the data to a temporary buffer, but only to avoid
179          * inconsequential accesses of uninitialized memory in
180          * lz_analyze_block().  */
181         memcpy(c->window, uncompressed_data, uncompressed_size);
182         memset(c->window + uncompressed_size, 0, 8);
183
184         /* Determine match/literal sequence to divide the data into.  */
185         memset(c->record_ctx.freqs, 0, sizeof(c->record_ctx.freqs));
186         c->record_ctx.matches = c->matches;
187         lz_analyze_block(c->window,
188                          uncompressed_size,
189                          xpress_record_match,
190                          xpress_record_literal,
191                          &c->record_ctx,
192                          &xpress_lz_params,
193                          c->prev_tab);
194
195         num_matches = (c->record_ctx.matches - c->matches);
196
197         /* Account for end of data symbol.  */
198         c->record_ctx.freqs[XPRESS_END_OF_DATA]++;
199
200         /* Build the Huffman code.  */
201         make_canonical_huffman_code(XPRESS_NUM_SYMBOLS, XPRESS_MAX_CODEWORD_LEN,
202                                     c->record_ctx.freqs, c->lens, c->codewords);
203
204         /* Output the Huffman code as a series of 512 4-bit lengths.  */
205         for (i = 0; i < XPRESS_NUM_SYMBOLS; i += 2)
206                 *cptr++ = (c->lens[i] & 0xf) | (c->lens[i + 1] << 4);
207
208         /* Output the encoded matches/literals.  */
209         init_output_bitstream(&ostream, cptr,
210                               compressed_size_avail - XPRESS_NUM_SYMBOLS / 2 - 1);
211
212         xpress_write_matches_and_literals(&ostream, c->matches,
213                                           num_matches, c->codewords, c->lens);
214
215         /* Flush any pending data and get the length of the compressed data.  */
216         compressed_size = flush_output_bitstream(&ostream);
217         if (compressed_size == ~(input_idx_t)0)
218                 return 0;
219
220         compressed_size += XPRESS_NUM_SYMBOLS / 2;
221
222 #if defined(ENABLE_XPRESS_DEBUG) || defined(ENABLE_VERIFY_COMPRESSION)
223         /* Verify that we really get the same thing back when decompressing.  */
224         {
225                 struct wimlib_decompressor *decompressor;
226
227                 if (0 == wimlib_create_decompressor(WIMLIB_COMPRESSION_TYPE_XPRESS,
228                                                     c->max_window_size,
229                                                     NULL,
230                                                     &decompressor))
231                 {
232                         int ret;
233                         ret = wimlib_decompress(compressed_data,
234                                                 compressed_size,
235                                                 c->window,
236                                                 uncompressed_size,
237                                                 decompressor);
238                         wimlib_free_decompressor(decompressor);
239
240                         if (ret) {
241                                 ERROR("Failed to decompress data we "
242                                       "compressed using XPRESS algorithm");
243                                 wimlib_assert(0);
244                                 return 0;
245                         }
246                         if (memcmp(uncompressed_data, c->window,
247                                    uncompressed_size))
248                         {
249                                 ERROR("Data we compressed using XPRESS algorithm "
250                                       "didn't decompress to original");
251                                 wimlib_assert(0);
252                                 return 0;
253                         }
254                 } else {
255                         WARNING("Failed to create decompressor for "
256                                 "data verification!");
257                 }
258         }
259 #endif
260
261         return compressed_size;
262 }
263
264 static void
265 xpress_free_compressor(void *_c)
266 {
267         struct xpress_compressor *c = _c;
268
269         if (c) {
270                 FREE(c->window);
271                 FREE(c->matches);
272                 FREE(c->prev_tab);
273                 FREE(c);
274         }
275 }
276
277 static int
278 xpress_create_compressor(size_t max_window_size,
279                          const struct wimlib_compressor_params_header *params,
280                          void **c_ret)
281 {
282         struct xpress_compressor *c;
283
284         if (max_window_size == 0 || max_window_size > (1U << 26))
285                 return WIMLIB_ERR_INVALID_PARAM;
286
287         c = CALLOC(1, sizeof(struct xpress_compressor));
288         if (c == NULL)
289                 goto oom;
290
291         c->window = MALLOC(max_window_size + 8);
292         if (c->window == NULL)
293                 goto oom;
294
295         c->max_window_size = max_window_size;
296
297         c->matches = MALLOC(max_window_size * sizeof(c->matches[0]));
298         if (c->matches == NULL)
299                 goto oom;
300
301         c->prev_tab = MALLOC(max_window_size * sizeof(c->prev_tab[0]));
302         if (c->prev_tab == NULL)
303                 goto oom;
304
305         *c_ret = c;
306         return 0;
307
308 oom:
309         xpress_free_compressor(c);
310         return WIMLIB_ERR_NOMEM;
311 }
312
313 static u64
314 xpress_get_needed_memory(size_t max_window_size,
315                          const struct wimlib_compressor_params_header *params)
316 {
317         u64 size = 0;
318
319         size += sizeof(struct xpress_compressor);
320         size += max_window_size + 8;
321         size += max_window_size * sizeof(((struct xpress_compressor*)0)->matches[0]);
322         size += max_window_size * sizeof(((struct xpress_compressor*)0)->prev_tab[0]);
323
324         return size;
325 }
326
327 const struct compressor_ops xpress_compressor_ops = {
328         .get_needed_memory  = xpress_get_needed_memory,
329         .create_compressor  = xpress_create_compressor,
330         .compress           = xpress_compress,
331         .free_compressor    = xpress_free_compressor,
332 };