xref: /freebsd/contrib/xz/src/liblzma/lz/lz_encoder.h (revision 06c3fb2749bda94cb5201f81ffdb8fa6c3161b2e)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       lz_encoder.h
4 /// \brief      LZ in window and match finder API
5 ///
6 //  Authors:    Igor Pavlov
7 //              Lasse Collin
8 //
9 //  This file has been put into the public domain.
10 //  You can do whatever you want with this file.
11 //
12 ///////////////////////////////////////////////////////////////////////////////
13 
14 #ifndef LZMA_LZ_ENCODER_H
15 #define LZMA_LZ_ENCODER_H
16 
17 #include "common.h"
18 
19 
20 /// A table of these is used by the LZ-based encoder to hold
21 /// the length-distance pairs found by the match finder.
22 typedef struct {
23 	uint32_t len;
24 	uint32_t dist;
25 } lzma_match;
26 
27 
28 typedef struct lzma_mf_s lzma_mf;
29 struct lzma_mf_s {
30 	///////////////
31 	// In Window //
32 	///////////////
33 
34 	/// Pointer to buffer with data to be compressed
35 	uint8_t *buffer;
36 
37 	/// Total size of the allocated buffer (that is, including all
38 	/// the extra space)
39 	uint32_t size;
40 
41 	/// Number of bytes that must be kept available in our input history.
42 	/// That is, once keep_size_before bytes have been processed,
43 	/// buffer[read_pos - keep_size_before] is the oldest byte that
44 	/// must be available for reading.
45 	uint32_t keep_size_before;
46 
47 	/// Number of bytes that must be kept in buffer after read_pos.
48 	/// That is, read_pos <= write_pos - keep_size_after as long as
49 	/// action is LZMA_RUN; when action != LZMA_RUN, read_pos is allowed
50 	/// to reach write_pos so that the last bytes get encoded too.
51 	uint32_t keep_size_after;
52 
53 	/// Match finders store locations of matches using 32-bit integers.
54 	/// To avoid adjusting several megabytes of integers every time the
55 	/// input window is moved with move_window, we only adjust the
56 	/// offset of the buffer. Thus, buffer[value_in_hash_table - offset]
57 	/// is the byte pointed by value_in_hash_table.
58 	uint32_t offset;
59 
60 	/// buffer[read_pos] is the next byte to run through the match
61 	/// finder. This is incremented in the match finder once the byte
62 	/// has been processed.
63 	uint32_t read_pos;
64 
65 	/// Number of bytes that have been ran through the match finder, but
66 	/// which haven't been encoded by the LZ-based encoder yet.
67 	uint32_t read_ahead;
68 
69 	/// As long as read_pos is less than read_limit, there is enough
70 	/// input available in buffer for at least one encoding loop.
71 	///
72 	/// Because of the stateful API, read_limit may and will get greater
73 	/// than read_pos quite often. This is taken into account when
74 	/// calculating the value for keep_size_after.
75 	uint32_t read_limit;
76 
77 	/// buffer[write_pos] is the first byte that doesn't contain valid
78 	/// uncompressed data; that is, the next input byte will be copied
79 	/// to buffer[write_pos].
80 	uint32_t write_pos;
81 
82 	/// Number of bytes not hashed before read_pos. This is needed to
83 	/// restart the match finder after LZMA_SYNC_FLUSH.
84 	uint32_t pending;
85 
86 	//////////////////
87 	// Match Finder //
88 	//////////////////
89 
90 	/// Find matches. Returns the number of distance-length pairs written
91 	/// to the matches array. This is called only via lzma_mf_find().
92 	uint32_t (*find)(lzma_mf *mf, lzma_match *matches);
93 
94 	/// Skips num bytes. This is like find() but doesn't make the
95 	/// distance-length pairs available, thus being a little faster.
96 	/// This is called only via mf_skip().
97 	void (*skip)(lzma_mf *mf, uint32_t num);
98 
99 	uint32_t *hash;
100 	uint32_t *son;
101 	uint32_t cyclic_pos;
102 	uint32_t cyclic_size; // Must be dictionary size + 1.
103 	uint32_t hash_mask;
104 
105 	/// Maximum number of loops in the match finder
106 	uint32_t depth;
107 
108 	/// Maximum length of a match that the match finder will try to find.
109 	uint32_t nice_len;
110 
111 	/// Maximum length of a match supported by the LZ-based encoder.
112 	/// If the longest match found by the match finder is nice_len,
113 	/// mf_find() tries to expand it up to match_len_max bytes.
114 	uint32_t match_len_max;
115 
116 	/// When running out of input, binary tree match finders need to know
117 	/// if it is due to flushing or finishing. The action is used also
118 	/// by the LZ-based encoders themselves.
119 	lzma_action action;
120 
121 	/// Number of elements in hash[]
122 	uint32_t hash_count;
123 
124 	/// Number of elements in son[]
125 	uint32_t sons_count;
126 };
127 
128 
129 typedef struct {
130 	/// Extra amount of data to keep available before the "actual"
131 	/// dictionary.
132 	size_t before_size;
133 
134 	/// Size of the history buffer
135 	size_t dict_size;
136 
137 	/// Extra amount of data to keep available after the "actual"
138 	/// dictionary.
139 	size_t after_size;
140 
141 	/// Maximum length of a match that the LZ-based encoder can accept.
142 	/// This is used to extend matches of length nice_len to the
143 	/// maximum possible length.
144 	size_t match_len_max;
145 
146 	/// Match finder will search matches up to this length.
147 	/// This must be less than or equal to match_len_max.
148 	size_t nice_len;
149 
150 	/// Type of the match finder to use
151 	lzma_match_finder match_finder;
152 
153 	/// Maximum search depth
154 	uint32_t depth;
155 
156 	/// TODO: Comment
157 	const uint8_t *preset_dict;
158 
159 	uint32_t preset_dict_size;
160 
161 } lzma_lz_options;
162 
163 
164 // The total usable buffer space at any moment outside the match finder:
165 // before_size + dict_size + after_size + match_len_max
166 //
167 // In reality, there's some extra space allocated to prevent the number of
168 // memmove() calls reasonable. The bigger the dict_size is, the bigger
169 // this extra buffer will be since with bigger dictionaries memmove() would
170 // also take longer.
171 //
172 // A single encoder loop in the LZ-based encoder may call the match finder
173 // (mf_find() or mf_skip()) at most after_size times. In other words,
174 // a single encoder loop may increment lzma_mf.read_pos at most after_size
175 // times. Since matches are looked up to
176 // lzma_mf.buffer[lzma_mf.read_pos + match_len_max - 1], the total
177 // amount of extra buffer needed after dict_size becomes
178 // after_size + match_len_max.
179 //
180 // before_size has two uses. The first one is to keep literals available
181 // in cases when the LZ-based encoder has made some read ahead.
182 // TODO: Maybe this could be changed by making the LZ-based encoders to
183 // store the actual literals as they do with length-distance pairs.
184 //
185 // Algorithms such as LZMA2 first try to compress a chunk, and then check
186 // if the encoded result is smaller than the uncompressed one. If the chunk
187 // was incompressible, it is better to store it in uncompressed form in
188 // the output stream. To do this, the whole uncompressed chunk has to be
189 // still available in the history buffer. before_size achieves that.
190 
191 
192 typedef struct {
193 	/// Data specific to the LZ-based encoder
194 	void *coder;
195 
196 	/// Function to encode from *dict to out[]
197 	lzma_ret (*code)(void *coder,
198 			lzma_mf *restrict mf, uint8_t *restrict out,
199 			size_t *restrict out_pos, size_t out_size);
200 
201 	/// Free allocated resources
202 	void (*end)(void *coder, const lzma_allocator *allocator);
203 
204 	/// Update the options in the middle of the encoding.
205 	lzma_ret (*options_update)(void *coder, const lzma_filter *filter);
206 
207 	/// Set maximum allowed output size
208 	lzma_ret (*set_out_limit)(void *coder, uint64_t *uncomp_size,
209 			uint64_t out_limit);
210 
211 } lzma_lz_encoder;
212 
213 
214 // Basic steps:
215 //  1. Input gets copied into the dictionary.
216 //  2. Data in dictionary gets run through the match finder byte by byte.
217 //  3. The literals and matches are encoded using e.g. LZMA.
218 //
219 // The bytes that have been ran through the match finder, but not encoded yet,
220 // are called `read ahead'.
221 
222 
223 /// Get how many bytes the match finder hashes in its initial step.
224 /// This is also the minimum nice_len value with the match finder.
225 static inline uint32_t
226 mf_get_hash_bytes(lzma_match_finder match_finder)
227 {
228 	return (uint32_t)match_finder & 0x0F;
229 }
230 
231 
232 /// Get pointer to the first byte not ran through the match finder
233 static inline const uint8_t *
234 mf_ptr(const lzma_mf *mf)
235 {
236 	return mf->buffer + mf->read_pos;
237 }
238 
239 
240 /// Get the number of bytes that haven't been ran through the match finder yet.
241 static inline uint32_t
242 mf_avail(const lzma_mf *mf)
243 {
244 	return mf->write_pos - mf->read_pos;
245 }
246 
247 
248 /// Get the number of bytes that haven't been encoded yet (some of these
249 /// bytes may have been ran through the match finder though).
250 static inline uint32_t
251 mf_unencoded(const lzma_mf *mf)
252 {
253 	return mf->write_pos - mf->read_pos + mf->read_ahead;
254 }
255 
256 
257 /// Calculate the absolute offset from the beginning of the most recent
258 /// dictionary reset. Only the lowest four bits are important, so there's no
259 /// problem that we don't know the 64-bit size of the data encoded so far.
260 ///
261 /// NOTE: When moving the input window, we need to do it so that the lowest
262 /// bits of dict->read_pos are not modified to keep this macro working
263 /// as intended.
264 static inline uint32_t
265 mf_position(const lzma_mf *mf)
266 {
267 	return mf->read_pos - mf->read_ahead;
268 }
269 
270 
271 /// Since everything else begins with mf_, use it also for lzma_mf_find().
272 #define mf_find lzma_mf_find
273 
274 
275 /// Skip the given number of bytes. This is used when a good match was found.
276 /// For example, if mf_find() finds a match of 200 bytes long, the first byte
277 /// of that match was already consumed by mf_find(), and the rest 199 bytes
278 /// have to be skipped with mf_skip(mf, 199).
279 static inline void
280 mf_skip(lzma_mf *mf, uint32_t amount)
281 {
282 	if (amount != 0) {
283 		mf->skip(mf, amount);
284 		mf->read_ahead += amount;
285 	}
286 }
287 
288 
289 /// Copies at most *left number of bytes from the history buffer
290 /// to out[]. This is needed by LZMA2 to encode uncompressed chunks.
291 static inline void
292 mf_read(lzma_mf *mf, uint8_t *out, size_t *out_pos, size_t out_size,
293 		size_t *left)
294 {
295 	const size_t out_avail = out_size - *out_pos;
296 	const size_t copy_size = my_min(out_avail, *left);
297 
298 	assert(mf->read_ahead == 0);
299 	assert(mf->read_pos >= *left);
300 
301 	memcpy(out + *out_pos, mf->buffer + mf->read_pos - *left,
302 			copy_size);
303 
304 	*out_pos += copy_size;
305 	*left -= copy_size;
306 	return;
307 }
308 
309 
310 extern lzma_ret lzma_lz_encoder_init(
311 		lzma_next_coder *next, const lzma_allocator *allocator,
312 		const lzma_filter_info *filters,
313 		lzma_ret (*lz_init)(lzma_lz_encoder *lz,
314 			const lzma_allocator *allocator,
315 			lzma_vli id, const void *options,
316 			lzma_lz_options *lz_options));
317 
318 
319 extern uint64_t lzma_lz_encoder_memusage(const lzma_lz_options *lz_options);
320 
321 
322 // These are only for LZ encoder's internal use.
323 extern uint32_t lzma_mf_find(
324 		lzma_mf *mf, uint32_t *count, lzma_match *matches);
325 
326 extern uint32_t lzma_mf_hc3_find(lzma_mf *dict, lzma_match *matches);
327 extern void lzma_mf_hc3_skip(lzma_mf *dict, uint32_t amount);
328 
329 extern uint32_t lzma_mf_hc4_find(lzma_mf *dict, lzma_match *matches);
330 extern void lzma_mf_hc4_skip(lzma_mf *dict, uint32_t amount);
331 
332 extern uint32_t lzma_mf_bt2_find(lzma_mf *dict, lzma_match *matches);
333 extern void lzma_mf_bt2_skip(lzma_mf *dict, uint32_t amount);
334 
335 extern uint32_t lzma_mf_bt3_find(lzma_mf *dict, lzma_match *matches);
336 extern void lzma_mf_bt3_skip(lzma_mf *dict, uint32_t amount);
337 
338 extern uint32_t lzma_mf_bt4_find(lzma_mf *dict, lzma_match *matches);
339 extern void lzma_mf_bt4_skip(lzma_mf *dict, uint32_t amount);
340 
341 #endif
342