xref: /freebsd/contrib/xz/src/liblzma/lzma/lzma2_encoder.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       lzma2_encoder.c
4 /// \brief      LZMA2 encoder
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 #include "lz_encoder.h"
15 #include "lzma_encoder.h"
16 #include "fastpos.h"
17 #include "lzma2_encoder.h"
18 
19 
20 struct lzma_coder_s {
21 	enum {
22 		SEQ_INIT,
23 		SEQ_LZMA_ENCODE,
24 		SEQ_LZMA_COPY,
25 		SEQ_UNCOMPRESSED_HEADER,
26 		SEQ_UNCOMPRESSED_COPY,
27 	} sequence;
28 
29 	/// LZMA encoder
30 	lzma_coder *lzma;
31 
32 	/// LZMA options currently in use.
33 	lzma_options_lzma opt_cur;
34 
35 	bool need_properties;
36 	bool need_state_reset;
37 	bool need_dictionary_reset;
38 
39 	/// Uncompressed size of a chunk
40 	size_t uncompressed_size;
41 
42 	/// Compressed size of a chunk (excluding headers); this is also used
43 	/// to indicate the end of buf[] in SEQ_LZMA_COPY.
44 	size_t compressed_size;
45 
46 	/// Read position in buf[]
47 	size_t buf_pos;
48 
49 	/// Buffer to hold the chunk header and LZMA compressed data
50 	uint8_t buf[LZMA2_HEADER_MAX + LZMA2_CHUNK_MAX];
51 };
52 
53 
54 static void
55 lzma2_header_lzma(lzma_coder *coder)
56 {
57 	assert(coder->uncompressed_size > 0);
58 	assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX);
59 	assert(coder->compressed_size > 0);
60 	assert(coder->compressed_size <= LZMA2_CHUNK_MAX);
61 
62 	size_t pos;
63 
64 	if (coder->need_properties) {
65 		pos = 0;
66 
67 		if (coder->need_dictionary_reset)
68 			coder->buf[pos] = 0x80 + (3 << 5);
69 		else
70 			coder->buf[pos] = 0x80 + (2 << 5);
71 	} else {
72 		pos = 1;
73 
74 		if (coder->need_state_reset)
75 			coder->buf[pos] = 0x80 + (1 << 5);
76 		else
77 			coder->buf[pos] = 0x80;
78 	}
79 
80 	// Set the start position for copying.
81 	coder->buf_pos = pos;
82 
83 	// Uncompressed size
84 	size_t size = coder->uncompressed_size - 1;
85 	coder->buf[pos++] += size >> 16;
86 	coder->buf[pos++] = (size >> 8) & 0xFF;
87 	coder->buf[pos++] = size & 0xFF;
88 
89 	// Compressed size
90 	size = coder->compressed_size - 1;
91 	coder->buf[pos++] = size >> 8;
92 	coder->buf[pos++] = size & 0xFF;
93 
94 	// Properties, if needed
95 	if (coder->need_properties)
96 		lzma_lzma_lclppb_encode(&coder->opt_cur, coder->buf + pos);
97 
98 	coder->need_properties = false;
99 	coder->need_state_reset = false;
100 	coder->need_dictionary_reset = false;
101 
102 	// The copying code uses coder->compressed_size to indicate the end
103 	// of coder->buf[], so we need add the maximum size of the header here.
104 	coder->compressed_size += LZMA2_HEADER_MAX;
105 
106 	return;
107 }
108 
109 
110 static void
111 lzma2_header_uncompressed(lzma_coder *coder)
112 {
113 	assert(coder->uncompressed_size > 0);
114 	assert(coder->uncompressed_size <= LZMA2_CHUNK_MAX);
115 
116 	// If this is the first chunk, we need to include dictionary
117 	// reset indicator.
118 	if (coder->need_dictionary_reset)
119 		coder->buf[0] = 1;
120 	else
121 		coder->buf[0] = 2;
122 
123 	coder->need_dictionary_reset = false;
124 
125 	// "Compressed" size
126 	coder->buf[1] = (coder->uncompressed_size - 1) >> 8;
127 	coder->buf[2] = (coder->uncompressed_size - 1) & 0xFF;
128 
129 	// Set the start position for copying.
130 	coder->buf_pos = 0;
131 	return;
132 }
133 
134 
135 static lzma_ret
136 lzma2_encode(lzma_coder *restrict coder, lzma_mf *restrict mf,
137 		uint8_t *restrict out, size_t *restrict out_pos,
138 		size_t out_size)
139 {
140 	while (*out_pos < out_size)
141 	switch (coder->sequence) {
142 	case SEQ_INIT:
143 		// If there's no input left and we are flushing or finishing,
144 		// don't start a new chunk.
145 		if (mf_unencoded(mf) == 0) {
146 			// Write end of payload marker if finishing.
147 			if (mf->action == LZMA_FINISH)
148 				out[(*out_pos)++] = 0;
149 
150 			return mf->action == LZMA_RUN
151 					? LZMA_OK : LZMA_STREAM_END;
152 		}
153 
154 		if (coder->need_state_reset)
155 			return_if_error(lzma_lzma_encoder_reset(
156 					coder->lzma, &coder->opt_cur));
157 
158 		coder->uncompressed_size = 0;
159 		coder->compressed_size = 0;
160 		coder->sequence = SEQ_LZMA_ENCODE;
161 
162 	// Fall through
163 
164 	case SEQ_LZMA_ENCODE: {
165 		// Calculate how much more uncompressed data this chunk
166 		// could accept.
167 		const uint32_t left = LZMA2_UNCOMPRESSED_MAX
168 				- coder->uncompressed_size;
169 		uint32_t limit;
170 
171 		if (left < mf->match_len_max) {
172 			// Must flush immediately since the next LZMA symbol
173 			// could make the uncompressed size of the chunk too
174 			// big.
175 			limit = 0;
176 		} else {
177 			// Calculate maximum read_limit that is OK from point
178 			// of view of LZMA2 chunk size.
179 			limit = mf->read_pos - mf->read_ahead
180 					+ left - mf->match_len_max;
181 		}
182 
183 		// Save the start position so that we can update
184 		// coder->uncompressed_size.
185 		const uint32_t read_start = mf->read_pos - mf->read_ahead;
186 
187 		// Call the LZMA encoder until the chunk is finished.
188 		const lzma_ret ret = lzma_lzma_encode(coder->lzma, mf,
189 				coder->buf + LZMA2_HEADER_MAX,
190 				&coder->compressed_size,
191 				LZMA2_CHUNK_MAX, limit);
192 
193 		coder->uncompressed_size += mf->read_pos - mf->read_ahead
194 				- read_start;
195 
196 		assert(coder->compressed_size <= LZMA2_CHUNK_MAX);
197 		assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX);
198 
199 		if (ret != LZMA_STREAM_END)
200 			return LZMA_OK;
201 
202 		// See if the chunk compressed. If it didn't, we encode it
203 		// as uncompressed chunk. This saves a few bytes of space
204 		// and makes decoding faster.
205 		if (coder->compressed_size >= coder->uncompressed_size) {
206 			coder->uncompressed_size += mf->read_ahead;
207 			assert(coder->uncompressed_size
208 					<= LZMA2_UNCOMPRESSED_MAX);
209 			mf->read_ahead = 0;
210 			lzma2_header_uncompressed(coder);
211 			coder->need_state_reset = true;
212 			coder->sequence = SEQ_UNCOMPRESSED_HEADER;
213 			break;
214 		}
215 
216 		// The chunk did compress at least by one byte, so we store
217 		// the chunk as LZMA.
218 		lzma2_header_lzma(coder);
219 
220 		coder->sequence = SEQ_LZMA_COPY;
221 	}
222 
223 	// Fall through
224 
225 	case SEQ_LZMA_COPY:
226 		// Copy the compressed chunk along its headers to the
227 		// output buffer.
228 		lzma_bufcpy(coder->buf, &coder->buf_pos,
229 				coder->compressed_size,
230 				out, out_pos, out_size);
231 		if (coder->buf_pos != coder->compressed_size)
232 			return LZMA_OK;
233 
234 		coder->sequence = SEQ_INIT;
235 		break;
236 
237 	case SEQ_UNCOMPRESSED_HEADER:
238 		// Copy the three-byte header to indicate uncompressed chunk.
239 		lzma_bufcpy(coder->buf, &coder->buf_pos,
240 				LZMA2_HEADER_UNCOMPRESSED,
241 				out, out_pos, out_size);
242 		if (coder->buf_pos != LZMA2_HEADER_UNCOMPRESSED)
243 			return LZMA_OK;
244 
245 		coder->sequence = SEQ_UNCOMPRESSED_COPY;
246 
247 	// Fall through
248 
249 	case SEQ_UNCOMPRESSED_COPY:
250 		// Copy the uncompressed data as is from the dictionary
251 		// to the output buffer.
252 		mf_read(mf, out, out_pos, out_size, &coder->uncompressed_size);
253 		if (coder->uncompressed_size != 0)
254 			return LZMA_OK;
255 
256 		coder->sequence = SEQ_INIT;
257 		break;
258 	}
259 
260 	return LZMA_OK;
261 }
262 
263 
264 static void
265 lzma2_encoder_end(lzma_coder *coder, const lzma_allocator *allocator)
266 {
267 	lzma_free(coder->lzma, allocator);
268 	lzma_free(coder, allocator);
269 	return;
270 }
271 
272 
273 static lzma_ret
274 lzma2_encoder_options_update(lzma_coder *coder, const lzma_filter *filter)
275 {
276 	// New options can be set only when there is no incomplete chunk.
277 	// This is the case at the beginning of the raw stream and right
278 	// after LZMA_SYNC_FLUSH.
279 	if (filter->options == NULL || coder->sequence != SEQ_INIT)
280 		return LZMA_PROG_ERROR;
281 
282 	// Look if there are new options. At least for now,
283 	// only lc/lp/pb can be changed.
284 	const lzma_options_lzma *opt = filter->options;
285 	if (coder->opt_cur.lc != opt->lc || coder->opt_cur.lp != opt->lp
286 			|| coder->opt_cur.pb != opt->pb) {
287 		// Validate the options.
288 		if (opt->lc > LZMA_LCLP_MAX || opt->lp > LZMA_LCLP_MAX
289 				|| opt->lc + opt->lp > LZMA_LCLP_MAX
290 				|| opt->pb > LZMA_PB_MAX)
291 			return LZMA_OPTIONS_ERROR;
292 
293 		// The new options will be used when the encoder starts
294 		// a new LZMA2 chunk.
295 		coder->opt_cur.lc = opt->lc;
296 		coder->opt_cur.lp = opt->lp;
297 		coder->opt_cur.pb = opt->pb;
298 		coder->need_properties = true;
299 		coder->need_state_reset = true;
300 	}
301 
302 	return LZMA_OK;
303 }
304 
305 
306 static lzma_ret
307 lzma2_encoder_init(lzma_lz_encoder *lz, const lzma_allocator *allocator,
308 		const void *options, lzma_lz_options *lz_options)
309 {
310 	if (options == NULL)
311 		return LZMA_PROG_ERROR;
312 
313 	if (lz->coder == NULL) {
314 		lz->coder = lzma_alloc(sizeof(lzma_coder), allocator);
315 		if (lz->coder == NULL)
316 			return LZMA_MEM_ERROR;
317 
318 		lz->code = &lzma2_encode;
319 		lz->end = &lzma2_encoder_end;
320 		lz->options_update = &lzma2_encoder_options_update;
321 
322 		lz->coder->lzma = NULL;
323 	}
324 
325 	lz->coder->opt_cur = *(const lzma_options_lzma *)(options);
326 
327 	lz->coder->sequence = SEQ_INIT;
328 	lz->coder->need_properties = true;
329 	lz->coder->need_state_reset = false;
330 	lz->coder->need_dictionary_reset
331 			= lz->coder->opt_cur.preset_dict == NULL
332 			|| lz->coder->opt_cur.preset_dict_size == 0;
333 
334 	// Initialize LZMA encoder
335 	return_if_error(lzma_lzma_encoder_create(&lz->coder->lzma, allocator,
336 			&lz->coder->opt_cur, lz_options));
337 
338 	// Make sure that we will always have enough history available in
339 	// case we need to use uncompressed chunks. They are used when the
340 	// compressed size of a chunk is not smaller than the uncompressed
341 	// size, so we need to have at least LZMA2_COMPRESSED_MAX bytes
342 	// history available.
343 	if (lz_options->before_size + lz_options->dict_size < LZMA2_CHUNK_MAX)
344 		lz_options->before_size
345 				= LZMA2_CHUNK_MAX - lz_options->dict_size;
346 
347 	return LZMA_OK;
348 }
349 
350 
351 extern lzma_ret
352 lzma_lzma2_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
353 		const lzma_filter_info *filters)
354 {
355 	return lzma_lz_encoder_init(
356 			next, allocator, filters, &lzma2_encoder_init);
357 }
358 
359 
360 extern uint64_t
361 lzma_lzma2_encoder_memusage(const void *options)
362 {
363 	const uint64_t lzma_mem = lzma_lzma_encoder_memusage(options);
364 	if (lzma_mem == UINT64_MAX)
365 		return UINT64_MAX;
366 
367 	return sizeof(lzma_coder) + lzma_mem;
368 }
369 
370 
371 extern lzma_ret
372 lzma_lzma2_props_encode(const void *options, uint8_t *out)
373 {
374 	const lzma_options_lzma *const opt = options;
375 	uint32_t d = my_max(opt->dict_size, LZMA_DICT_SIZE_MIN);
376 
377 	// Round up to the next 2^n - 1 or 2^n + 2^(n - 1) - 1 depending
378 	// on which one is the next:
379 	--d;
380 	d |= d >> 2;
381 	d |= d >> 3;
382 	d |= d >> 4;
383 	d |= d >> 8;
384 	d |= d >> 16;
385 
386 	// Get the highest two bits using the proper encoding:
387 	if (d == UINT32_MAX)
388 		out[0] = 40;
389 	else
390 		out[0] = get_dist_slot(d + 1) - 24;
391 
392 	return LZMA_OK;
393 }
394 
395 
396 extern uint64_t
397 lzma_lzma2_block_size(const void *options)
398 {
399 	const lzma_options_lzma *const opt = options;
400 
401 	// Use at least 1 MiB to keep compression ratio better.
402 	return my_max((uint64_t)(opt->dict_size) * 3, UINT64_C(1) << 20);
403 }
404