lz_encoder.c revision 278433
1///////////////////////////////////////////////////////////////////////////////
2//
3/// \file       lz_encoder.c
4/// \brief      LZ in window
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 "lz_encoder_hash.h"
16
17// See lz_encoder_hash.h. This is a bit hackish but avoids making
18// endianness a conditional in makefiles.
19#if defined(WORDS_BIGENDIAN) && !defined(HAVE_SMALL)
20#	include "lz_encoder_hash_table.h"
21#endif
22
23#include "memcmplen.h"
24
25
26struct lzma_coder_s {
27	/// LZ-based encoder e.g. LZMA
28	lzma_lz_encoder lz;
29
30	/// History buffer and match finder
31	lzma_mf mf;
32
33	/// Next coder in the chain
34	lzma_next_coder next;
35};
36
37
38/// \brief      Moves the data in the input window to free space for new data
39///
40/// mf->buffer is a sliding input window, which keeps mf->keep_size_before
41/// bytes of input history available all the time. Now and then we need to
42/// "slide" the buffer to make space for the new data to the end of the
43/// buffer. At the same time, data older than keep_size_before is dropped.
44///
45static void
46move_window(lzma_mf *mf)
47{
48	// Align the move to a multiple of 16 bytes. Some LZ-based encoders
49	// like LZMA use the lowest bits of mf->read_pos to know the
50	// alignment of the uncompressed data. We also get better speed
51	// for memmove() with aligned buffers.
52	assert(mf->read_pos > mf->keep_size_before);
53	const uint32_t move_offset
54		= (mf->read_pos - mf->keep_size_before) & ~UINT32_C(15);
55
56	assert(mf->write_pos > move_offset);
57	const size_t move_size = mf->write_pos - move_offset;
58
59	assert(move_offset + move_size <= mf->size);
60
61	memmove(mf->buffer, mf->buffer + move_offset, move_size);
62
63	mf->offset += move_offset;
64	mf->read_pos -= move_offset;
65	mf->read_limit -= move_offset;
66	mf->write_pos -= move_offset;
67
68	return;
69}
70
71
72/// \brief      Tries to fill the input window (mf->buffer)
73///
74/// If we are the last encoder in the chain, our input data is in in[].
75/// Otherwise we call the next filter in the chain to process in[] and
76/// write its output to mf->buffer.
77///
78/// This function must not be called once it has returned LZMA_STREAM_END.
79///
80static lzma_ret
81fill_window(lzma_coder *coder, const lzma_allocator *allocator,
82		const uint8_t *in, size_t *in_pos, size_t in_size,
83		lzma_action action)
84{
85	assert(coder->mf.read_pos <= coder->mf.write_pos);
86
87	// Move the sliding window if needed.
88	if (coder->mf.read_pos >= coder->mf.size - coder->mf.keep_size_after)
89		move_window(&coder->mf);
90
91	// Maybe this is ugly, but lzma_mf uses uint32_t for most things
92	// (which I find cleanest), but we need size_t here when filling
93	// the history window.
94	size_t write_pos = coder->mf.write_pos;
95	lzma_ret ret;
96	if (coder->next.code == NULL) {
97		// Not using a filter, simply memcpy() as much as possible.
98		lzma_bufcpy(in, in_pos, in_size, coder->mf.buffer,
99				&write_pos, coder->mf.size);
100
101		ret = action != LZMA_RUN && *in_pos == in_size
102				? LZMA_STREAM_END : LZMA_OK;
103
104	} else {
105		ret = coder->next.code(coder->next.coder, allocator,
106				in, in_pos, in_size,
107				coder->mf.buffer, &write_pos,
108				coder->mf.size, action);
109	}
110
111	coder->mf.write_pos = write_pos;
112
113	// If end of stream has been reached or flushing completed, we allow
114	// the encoder to process all the input (that is, read_pos is allowed
115	// to reach write_pos). Otherwise we keep keep_size_after bytes
116	// available as prebuffer.
117	if (ret == LZMA_STREAM_END) {
118		assert(*in_pos == in_size);
119		ret = LZMA_OK;
120		coder->mf.action = action;
121		coder->mf.read_limit = coder->mf.write_pos;
122
123	} else if (coder->mf.write_pos > coder->mf.keep_size_after) {
124		// This needs to be done conditionally, because if we got
125		// only little new input, there may be too little input
126		// to do any encoding yet.
127		coder->mf.read_limit = coder->mf.write_pos
128				- coder->mf.keep_size_after;
129	}
130
131	// Restart the match finder after finished LZMA_SYNC_FLUSH.
132	if (coder->mf.pending > 0
133			&& coder->mf.read_pos < coder->mf.read_limit) {
134		// Match finder may update coder->pending and expects it to
135		// start from zero, so use a temporary variable.
136		const size_t pending = coder->mf.pending;
137		coder->mf.pending = 0;
138
139		// Rewind read_pos so that the match finder can hash
140		// the pending bytes.
141		assert(coder->mf.read_pos >= pending);
142		coder->mf.read_pos -= pending;
143
144		// Call the skip function directly instead of using
145		// mf_skip(), since we don't want to touch mf->read_ahead.
146		coder->mf.skip(&coder->mf, pending);
147	}
148
149	return ret;
150}
151
152
153static lzma_ret
154lz_encode(lzma_coder *coder, const lzma_allocator *allocator,
155		const uint8_t *restrict in, size_t *restrict in_pos,
156		size_t in_size,
157		uint8_t *restrict out, size_t *restrict out_pos,
158		size_t out_size, lzma_action action)
159{
160	while (*out_pos < out_size
161			&& (*in_pos < in_size || action != LZMA_RUN)) {
162		// Read more data to coder->mf.buffer if needed.
163		if (coder->mf.action == LZMA_RUN && coder->mf.read_pos
164				>= coder->mf.read_limit)
165			return_if_error(fill_window(coder, allocator,
166					in, in_pos, in_size, action));
167
168		// Encode
169		const lzma_ret ret = coder->lz.code(coder->lz.coder,
170				&coder->mf, out, out_pos, out_size);
171		if (ret != LZMA_OK) {
172			// Setting this to LZMA_RUN for cases when we are
173			// flushing. It doesn't matter when finishing or if
174			// an error occurred.
175			coder->mf.action = LZMA_RUN;
176			return ret;
177		}
178	}
179
180	return LZMA_OK;
181}
182
183
184static bool
185lz_encoder_prepare(lzma_mf *mf, const lzma_allocator *allocator,
186		const lzma_lz_options *lz_options)
187{
188	// For now, the dictionary size is limited to 1.5 GiB. This may grow
189	// in the future if needed, but it needs a little more work than just
190	// changing this check.
191	if (lz_options->dict_size < LZMA_DICT_SIZE_MIN
192			|| lz_options->dict_size
193				> (UINT32_C(1) << 30) + (UINT32_C(1) << 29)
194			|| lz_options->nice_len > lz_options->match_len_max)
195		return true;
196
197	mf->keep_size_before = lz_options->before_size + lz_options->dict_size;
198
199	mf->keep_size_after = lz_options->after_size
200			+ lz_options->match_len_max;
201
202	// To avoid constant memmove()s, allocate some extra space. Since
203	// memmove()s become more expensive when the size of the buffer
204	// increases, we reserve more space when a large dictionary is
205	// used to make the memmove() calls rarer.
206	//
207	// This works with dictionaries up to about 3 GiB. If bigger
208	// dictionary is wanted, some extra work is needed:
209	//   - Several variables in lzma_mf have to be changed from uint32_t
210	//     to size_t.
211	//   - Memory usage calculation needs something too, e.g. use uint64_t
212	//     for mf->size.
213	uint32_t reserve = lz_options->dict_size / 2;
214	if (reserve > (UINT32_C(1) << 30))
215		reserve /= 2;
216
217	reserve += (lz_options->before_size + lz_options->match_len_max
218			+ lz_options->after_size) / 2 + (UINT32_C(1) << 19);
219
220	const uint32_t old_size = mf->size;
221	mf->size = mf->keep_size_before + reserve + mf->keep_size_after;
222
223	// Deallocate the old history buffer if it exists but has different
224	// size than what is needed now.
225	if (mf->buffer != NULL && old_size != mf->size) {
226		lzma_free(mf->buffer, allocator);
227		mf->buffer = NULL;
228	}
229
230	// Match finder options
231	mf->match_len_max = lz_options->match_len_max;
232	mf->nice_len = lz_options->nice_len;
233
234	// cyclic_size has to stay smaller than 2 Gi. Note that this doesn't
235	// mean limiting dictionary size to less than 2 GiB. With a match
236	// finder that uses multibyte resolution (hashes start at e.g. every
237	// fourth byte), cyclic_size would stay below 2 Gi even when
238	// dictionary size is greater than 2 GiB.
239	//
240	// It would be possible to allow cyclic_size >= 2 Gi, but then we
241	// would need to be careful to use 64-bit types in various places
242	// (size_t could do since we would need bigger than 32-bit address
243	// space anyway). It would also require either zeroing a multigigabyte
244	// buffer at initialization (waste of time and RAM) or allow
245	// normalization in lz_encoder_mf.c to access uninitialized
246	// memory to keep the code simpler. The current way is simple and
247	// still allows pretty big dictionaries, so I don't expect these
248	// limits to change.
249	mf->cyclic_size = lz_options->dict_size + 1;
250
251	// Validate the match finder ID and setup the function pointers.
252	switch (lz_options->match_finder) {
253#ifdef HAVE_MF_HC3
254	case LZMA_MF_HC3:
255		mf->find = &lzma_mf_hc3_find;
256		mf->skip = &lzma_mf_hc3_skip;
257		break;
258#endif
259#ifdef HAVE_MF_HC4
260	case LZMA_MF_HC4:
261		mf->find = &lzma_mf_hc4_find;
262		mf->skip = &lzma_mf_hc4_skip;
263		break;
264#endif
265#ifdef HAVE_MF_BT2
266	case LZMA_MF_BT2:
267		mf->find = &lzma_mf_bt2_find;
268		mf->skip = &lzma_mf_bt2_skip;
269		break;
270#endif
271#ifdef HAVE_MF_BT3
272	case LZMA_MF_BT3:
273		mf->find = &lzma_mf_bt3_find;
274		mf->skip = &lzma_mf_bt3_skip;
275		break;
276#endif
277#ifdef HAVE_MF_BT4
278	case LZMA_MF_BT4:
279		mf->find = &lzma_mf_bt4_find;
280		mf->skip = &lzma_mf_bt4_skip;
281		break;
282#endif
283
284	default:
285		return true;
286	}
287
288	// Calculate the sizes of mf->hash and mf->son and check that
289	// nice_len is big enough for the selected match finder.
290	const uint32_t hash_bytes = lz_options->match_finder & 0x0F;
291	if (hash_bytes > mf->nice_len)
292		return true;
293
294	const bool is_bt = (lz_options->match_finder & 0x10) != 0;
295	uint32_t hs;
296
297	if (hash_bytes == 2) {
298		hs = 0xFFFF;
299	} else {
300		// Round dictionary size up to the next 2^n - 1 so it can
301		// be used as a hash mask.
302		hs = lz_options->dict_size - 1;
303		hs |= hs >> 1;
304		hs |= hs >> 2;
305		hs |= hs >> 4;
306		hs |= hs >> 8;
307		hs >>= 1;
308		hs |= 0xFFFF;
309
310		if (hs > (UINT32_C(1) << 24)) {
311			if (hash_bytes == 3)
312				hs = (UINT32_C(1) << 24) - 1;
313			else
314				hs >>= 1;
315		}
316	}
317
318	mf->hash_mask = hs;
319
320	++hs;
321	if (hash_bytes > 2)
322		hs += HASH_2_SIZE;
323	if (hash_bytes > 3)
324		hs += HASH_3_SIZE;
325/*
326	No match finder uses this at the moment.
327	if (mf->hash_bytes > 4)
328		hs += HASH_4_SIZE;
329*/
330
331	const uint32_t old_hash_count = mf->hash_count;
332	const uint32_t old_sons_count = mf->sons_count;
333	mf->hash_count = hs;
334	mf->sons_count = mf->cyclic_size;
335	if (is_bt)
336		mf->sons_count *= 2;
337
338	// Deallocate the old hash array if it exists and has different size
339	// than what is needed now.
340	if (old_hash_count != mf->hash_count
341			|| old_sons_count != mf->sons_count) {
342		lzma_free(mf->hash, allocator);
343		mf->hash = NULL;
344
345		lzma_free(mf->son, allocator);
346		mf->son = NULL;
347	}
348
349	// Maximum number of match finder cycles
350	mf->depth = lz_options->depth;
351	if (mf->depth == 0) {
352		if (is_bt)
353			mf->depth = 16 + mf->nice_len / 2;
354		else
355			mf->depth = 4 + mf->nice_len / 4;
356	}
357
358	return false;
359}
360
361
362static bool
363lz_encoder_init(lzma_mf *mf, const lzma_allocator *allocator,
364		const lzma_lz_options *lz_options)
365{
366	// Allocate the history buffer.
367	if (mf->buffer == NULL) {
368		// lzma_memcmplen() is used for the dictionary buffer
369		// so we need to allocate a few extra bytes to prevent
370		// it from reading past the end of the buffer.
371		mf->buffer = lzma_alloc(mf->size + LZMA_MEMCMPLEN_EXTRA,
372				allocator);
373		if (mf->buffer == NULL)
374			return true;
375
376		// Keep Valgrind happy with lzma_memcmplen() and initialize
377		// the extra bytes whose value may get read but which will
378		// effectively get ignored.
379		memzero(mf->buffer + mf->size, LZMA_MEMCMPLEN_EXTRA);
380	}
381
382	// Use cyclic_size as initial mf->offset. This allows
383	// avoiding a few branches in the match finders. The downside is
384	// that match finder needs to be normalized more often, which may
385	// hurt performance with huge dictionaries.
386	mf->offset = mf->cyclic_size;
387	mf->read_pos = 0;
388	mf->read_ahead = 0;
389	mf->read_limit = 0;
390	mf->write_pos = 0;
391	mf->pending = 0;
392
393#if UINT32_MAX >= SIZE_MAX / 4
394	// Check for integer overflow. (Huge dictionaries are not
395	// possible on 32-bit CPU.)
396	if (mf->hash_count > SIZE_MAX / sizeof(uint32_t)
397			|| mf->sons_count > SIZE_MAX / sizeof(uint32_t))
398		return true;
399#endif
400
401	// Allocate and initialize the hash table. Since EMPTY_HASH_VALUE
402	// is zero, we can use lzma_alloc_zero() or memzero() for mf->hash.
403	//
404	// We don't need to initialize mf->son, but not doing that may
405	// make Valgrind complain in normalization (see normalize() in
406	// lz_encoder_mf.c). Skipping the initialization is *very* good
407	// when big dictionary is used but only small amount of data gets
408	// actually compressed: most of the mf->son won't get actually
409	// allocated by the kernel, so we avoid wasting RAM and improve
410	// initialization speed a lot.
411	if (mf->hash == NULL) {
412		mf->hash = lzma_alloc_zero(mf->hash_count * sizeof(uint32_t),
413				allocator);
414		mf->son = lzma_alloc(mf->sons_count * sizeof(uint32_t),
415				allocator);
416
417		if (mf->hash == NULL || mf->son == NULL) {
418			lzma_free(mf->hash, allocator);
419			mf->hash = NULL;
420
421			lzma_free(mf->son, allocator);
422			mf->son = NULL;
423
424			return true;
425		}
426	} else {
427/*
428		for (uint32_t i = 0; i < mf->hash_count; ++i)
429			mf->hash[i] = EMPTY_HASH_VALUE;
430*/
431		memzero(mf->hash, mf->hash_count * sizeof(uint32_t));
432	}
433
434	mf->cyclic_pos = 0;
435
436	// Handle preset dictionary.
437	if (lz_options->preset_dict != NULL
438			&& lz_options->preset_dict_size > 0) {
439		// If the preset dictionary is bigger than the actual
440		// dictionary, use only the tail.
441		mf->write_pos = my_min(lz_options->preset_dict_size, mf->size);
442		memcpy(mf->buffer, lz_options->preset_dict
443				+ lz_options->preset_dict_size - mf->write_pos,
444				mf->write_pos);
445		mf->action = LZMA_SYNC_FLUSH;
446		mf->skip(mf, mf->write_pos);
447	}
448
449	mf->action = LZMA_RUN;
450
451	return false;
452}
453
454
455extern uint64_t
456lzma_lz_encoder_memusage(const lzma_lz_options *lz_options)
457{
458	// Old buffers must not exist when calling lz_encoder_prepare().
459	lzma_mf mf = {
460		.buffer = NULL,
461		.hash = NULL,
462		.son = NULL,
463		.hash_count = 0,
464		.sons_count = 0,
465	};
466
467	// Setup the size information into mf.
468	if (lz_encoder_prepare(&mf, NULL, lz_options))
469		return UINT64_MAX;
470
471	// Calculate the memory usage.
472	return ((uint64_t)(mf.hash_count) + mf.sons_count) * sizeof(uint32_t)
473			+ mf.size + sizeof(lzma_coder);
474}
475
476
477static void
478lz_encoder_end(lzma_coder *coder, const lzma_allocator *allocator)
479{
480	lzma_next_end(&coder->next, allocator);
481
482	lzma_free(coder->mf.son, allocator);
483	lzma_free(coder->mf.hash, allocator);
484	lzma_free(coder->mf.buffer, allocator);
485
486	if (coder->lz.end != NULL)
487		coder->lz.end(coder->lz.coder, allocator);
488	else
489		lzma_free(coder->lz.coder, allocator);
490
491	lzma_free(coder, allocator);
492	return;
493}
494
495
496static lzma_ret
497lz_encoder_update(lzma_coder *coder, const lzma_allocator *allocator,
498		const lzma_filter *filters_null lzma_attribute((__unused__)),
499		const lzma_filter *reversed_filters)
500{
501	if (coder->lz.options_update == NULL)
502		return LZMA_PROG_ERROR;
503
504	return_if_error(coder->lz.options_update(
505			coder->lz.coder, reversed_filters));
506
507	return lzma_next_filter_update(
508			&coder->next, allocator, reversed_filters + 1);
509}
510
511
512extern lzma_ret
513lzma_lz_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
514		const lzma_filter_info *filters,
515		lzma_ret (*lz_init)(lzma_lz_encoder *lz,
516			const lzma_allocator *allocator, const void *options,
517			lzma_lz_options *lz_options))
518{
519#ifdef HAVE_SMALL
520	// We need that the CRC32 table has been initialized.
521	lzma_crc32_init();
522#endif
523
524	// Allocate and initialize the base data structure.
525	if (next->coder == NULL) {
526		next->coder = lzma_alloc(sizeof(lzma_coder), allocator);
527		if (next->coder == NULL)
528			return LZMA_MEM_ERROR;
529
530		next->code = &lz_encode;
531		next->end = &lz_encoder_end;
532		next->update = &lz_encoder_update;
533
534		next->coder->lz.coder = NULL;
535		next->coder->lz.code = NULL;
536		next->coder->lz.end = NULL;
537
538		next->coder->mf.buffer = NULL;
539		next->coder->mf.hash = NULL;
540		next->coder->mf.son = NULL;
541		next->coder->mf.hash_count = 0;
542		next->coder->mf.sons_count = 0;
543
544		next->coder->next = LZMA_NEXT_CODER_INIT;
545	}
546
547	// Initialize the LZ-based encoder.
548	lzma_lz_options lz_options;
549	return_if_error(lz_init(&next->coder->lz, allocator,
550			filters[0].options, &lz_options));
551
552	// Setup the size information into next->coder->mf and deallocate
553	// old buffers if they have wrong size.
554	if (lz_encoder_prepare(&next->coder->mf, allocator, &lz_options))
555		return LZMA_OPTIONS_ERROR;
556
557	// Allocate new buffers if needed, and do the rest of
558	// the initialization.
559	if (lz_encoder_init(&next->coder->mf, allocator, &lz_options))
560		return LZMA_MEM_ERROR;
561
562	// Initialize the next filter in the chain, if any.
563	return lzma_next_filter_init(&next->coder->next, allocator,
564			filters + 1);
565}
566
567
568extern LZMA_API(lzma_bool)
569lzma_mf_is_supported(lzma_match_finder mf)
570{
571	bool ret = false;
572
573#ifdef HAVE_MF_HC3
574	if (mf == LZMA_MF_HC3)
575		ret = true;
576#endif
577
578#ifdef HAVE_MF_HC4
579	if (mf == LZMA_MF_HC4)
580		ret = true;
581#endif
582
583#ifdef HAVE_MF_BT2
584	if (mf == LZMA_MF_BT2)
585		ret = true;
586#endif
587
588#ifdef HAVE_MF_BT3
589	if (mf == LZMA_MF_BT3)
590		ret = true;
591#endif
592
593#ifdef HAVE_MF_BT4
594	if (mf == LZMA_MF_BT4)
595		ret = true;
596#endif
597
598	return ret;
599}
600