1 /* vi: set sw=4 ts=4: */
3 * Small lzma deflate implementation.
4 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
6 * Based on LzmaDecode.c from the LZMA SDK 4.22 (http://www.7-zip.org/)
7 * Copyright (C) 1999-2005 Igor Pavlov
9 * Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
13 #include "unarchive.h"
15 #if ENABLE_FEATURE_LZMA_FAST
16 # define speed_inline ALWAYS_INLINE
26 /* Was keeping rc on stack in unlzma and separately allocating buffer,
27 * but with "buffer 'attached to' allocated rc" code is smaller: */
28 /* uint8_t *buffer; */
29 #define RC_BUFFER ((uint8_t*)(rc+1))
33 /* Had provisions for variable buffer, but we don't need it here */
34 /* int buffer_size; */
35 #define RC_BUFFER_SIZE 0x10000
42 #define RC_TOP_BITS 24
43 #define RC_MOVE_BITS 5
44 #define RC_MODEL_TOTAL_BITS 11
47 /* Called twice: once at startup and once in rc_normalize() */
48 static void rc_read(rc_t * rc)
50 int buffer_size = safe_read(rc->fd, RC_BUFFER, RC_BUFFER_SIZE);
52 bb_error_msg_and_die("unexpected EOF");
54 rc->buffer_end = RC_BUFFER + buffer_size;
58 static rc_t* rc_init(int fd) /*, int buffer_size) */
63 rc = xmalloc(sizeof(rc_t) + RC_BUFFER_SIZE);
66 /* rc->buffer_size = buffer_size; */
67 rc->buffer_end = RC_BUFFER + RC_BUFFER_SIZE;
68 rc->ptr = rc->buffer_end;
71 rc->range = 0xFFFFFFFF;
72 for (i = 0; i < 5; i++) {
73 if (rc->ptr >= rc->buffer_end)
75 rc->code = (rc->code << 8) | *rc->ptr++;
81 static ALWAYS_INLINE void rc_free(rc_t * rc)
83 if (ENABLE_FEATURE_CLEAN_UP)
87 /* Called twice, but one callsite is in speed_inline'd rc_is_bit_0_helper() */
88 static void rc_do_normalize(rc_t * rc)
90 if (rc->ptr >= rc->buffer_end)
93 rc->code = (rc->code << 8) | *rc->ptr++;
95 static ALWAYS_INLINE void rc_normalize(rc_t * rc)
97 if (rc->range < (1 << RC_TOP_BITS)) {
102 /* rc_is_bit_0 is called 9 times */
103 /* Why rc_is_bit_0_helper exists?
104 * Because we want to always expose (rc->code < rc->bound) to optimizer.
105 * Thus rc_is_bit_0 is always inlined, and rc_is_bit_0_helper is inlined
106 * only if we compile for speed.
108 static speed_inline uint32_t rc_is_bit_0_helper(rc_t * rc, uint16_t * p)
111 rc->bound = *p * (rc->range >> RC_MODEL_TOTAL_BITS);
114 static ALWAYS_INLINE int rc_is_bit_0(rc_t * rc, uint16_t * p)
116 uint32_t t = rc_is_bit_0_helper(rc, p);
120 /* Called ~10 times, but very small, thus inlined */
121 static speed_inline void rc_update_bit_0(rc_t * rc, uint16_t * p)
123 rc->range = rc->bound;
124 *p += ((1 << RC_MODEL_TOTAL_BITS) - *p) >> RC_MOVE_BITS;
126 static speed_inline void rc_update_bit_1(rc_t * rc, uint16_t * p)
128 rc->range -= rc->bound;
129 rc->code -= rc->bound;
130 *p -= *p >> RC_MOVE_BITS;
133 /* Called 4 times in unlzma loop */
134 static int rc_get_bit(rc_t * rc, uint16_t * p, int *symbol)
136 if (rc_is_bit_0(rc, p)) {
137 rc_update_bit_0(rc, p);
141 rc_update_bit_1(rc, p);
142 *symbol = *symbol * 2 + 1;
148 static ALWAYS_INLINE int rc_direct_bit(rc_t * rc)
152 if (rc->code >= rc->range) {
153 rc->code -= rc->range;
160 static speed_inline void
161 rc_bit_tree_decode(rc_t * rc, uint16_t * p, int num_levels, int *symbol)
167 rc_get_bit(rc, p + *symbol, symbol);
168 *symbol -= 1 << num_levels;
176 } __attribute__ ((packed)) lzma_header_t;
179 /* #defines will force compiler to compute/optimize each one with each usage.
180 * Have heart and use enum instead. */
182 LZMA_BASE_SIZE = 1846,
185 LZMA_NUM_POS_BITS_MAX = 4,
187 LZMA_LEN_NUM_LOW_BITS = 3,
188 LZMA_LEN_NUM_MID_BITS = 3,
189 LZMA_LEN_NUM_HIGH_BITS = 8,
192 LZMA_LEN_CHOICE_2 = (LZMA_LEN_CHOICE + 1),
193 LZMA_LEN_LOW = (LZMA_LEN_CHOICE_2 + 1),
194 LZMA_LEN_MID = (LZMA_LEN_LOW \
195 + (1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_LOW_BITS))),
196 LZMA_LEN_HIGH = (LZMA_LEN_MID \
197 + (1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_MID_BITS))),
198 LZMA_NUM_LEN_PROBS = (LZMA_LEN_HIGH + (1 << LZMA_LEN_NUM_HIGH_BITS)),
200 LZMA_NUM_STATES = 12,
201 LZMA_NUM_LIT_STATES = 7,
203 LZMA_START_POS_MODEL_INDEX = 4,
204 LZMA_END_POS_MODEL_INDEX = 14,
205 LZMA_NUM_FULL_DISTANCES = (1 << (LZMA_END_POS_MODEL_INDEX >> 1)),
207 LZMA_NUM_POS_SLOT_BITS = 6,
208 LZMA_NUM_LEN_TO_POS_STATES = 4,
210 LZMA_NUM_ALIGN_BITS = 4,
212 LZMA_MATCH_MIN_LEN = 2,
215 LZMA_IS_REP = (LZMA_IS_MATCH + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX)),
216 LZMA_IS_REP_G0 = (LZMA_IS_REP + LZMA_NUM_STATES),
217 LZMA_IS_REP_G1 = (LZMA_IS_REP_G0 + LZMA_NUM_STATES),
218 LZMA_IS_REP_G2 = (LZMA_IS_REP_G1 + LZMA_NUM_STATES),
219 LZMA_IS_REP_0_LONG = (LZMA_IS_REP_G2 + LZMA_NUM_STATES),
220 LZMA_POS_SLOT = (LZMA_IS_REP_0_LONG \
221 + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX)),
222 LZMA_SPEC_POS = (LZMA_POS_SLOT \
223 + (LZMA_NUM_LEN_TO_POS_STATES << LZMA_NUM_POS_SLOT_BITS)),
224 LZMA_ALIGN = (LZMA_SPEC_POS \
225 + LZMA_NUM_FULL_DISTANCES - LZMA_END_POS_MODEL_INDEX),
226 LZMA_LEN_CODER = (LZMA_ALIGN + (1 << LZMA_NUM_ALIGN_BITS)),
227 LZMA_REP_LEN_CODER = (LZMA_LEN_CODER + LZMA_NUM_LEN_PROBS),
228 LZMA_LITERAL = (LZMA_REP_LEN_CODER + LZMA_NUM_LEN_PROBS),
232 USE_DESKTOP(long long) int FAST_FUNC
233 unpack_lzma_stream(int src_fd, int dst_fd)
235 USE_DESKTOP(long long total_written = 0;)
236 lzma_header_t header;
238 uint32_t pos_state_mask;
239 uint32_t literal_pos_mask;
249 uint8_t previous_byte = 0;
250 size_t buffer_pos = 0, global_pos = 0;
253 uint32_t rep0 = 1, rep1 = 1, rep2 = 1, rep3 = 1;
255 xread(src_fd, &header, sizeof(header));
257 if (header.pos >= (9 * 5 * 5))
258 bb_error_msg_and_die("bad header");
263 pos_state_mask = (1 << pb) - 1;
264 literal_pos_mask = (1 << lp) - 1;
266 header.dict_size = SWAP_LE32(header.dict_size);
267 header.dst_size = SWAP_LE64(header.dst_size);
269 if (header.dict_size == 0)
270 header.dict_size = 1;
272 buffer = xmalloc(MIN(header.dst_size, header.dict_size));
274 num_probs = LZMA_BASE_SIZE + (LZMA_LIT_SIZE << (lc + lp));
275 p = xmalloc(num_probs * sizeof(*p));
276 num_probs = LZMA_LITERAL + (LZMA_LIT_SIZE << (lc + lp));
277 for (i = 0; i < num_probs; i++)
278 p[i] = (1 << RC_MODEL_TOTAL_BITS) >> 1;
280 rc = rc_init(src_fd); /*, RC_BUFFER_SIZE); */
282 while (global_pos + buffer_pos < header.dst_size) {
283 int pos_state = (buffer_pos + global_pos) & pos_state_mask;
285 prob = p + LZMA_IS_MATCH + (state << LZMA_NUM_POS_BITS_MAX) + pos_state;
286 if (rc_is_bit_0(rc, prob)) {
288 rc_update_bit_0(rc, prob);
289 prob = (p + LZMA_LITERAL
290 + (LZMA_LIT_SIZE * ((((buffer_pos + global_pos) & literal_pos_mask) << lc)
291 + (previous_byte >> (8 - lc))
296 if (state >= LZMA_NUM_LIT_STATES) {
299 pos = buffer_pos - rep0;
300 while (pos >= header.dict_size)
301 pos += header.dict_size;
302 match_byte = buffer[pos];
307 bit = match_byte & 0x100;
308 prob_lit = prob + 0x100 + bit + mi;
309 bit ^= (rc_get_bit(rc, prob_lit, &mi) << 8); /* 0x100 or 0 */
312 } while (mi < 0x100);
315 prob_lit = prob + mi;
316 rc_get_bit(rc, prob_lit, &mi);
325 previous_byte = (uint8_t) mi;
326 #if ENABLE_FEATURE_LZMA_FAST
328 buffer[buffer_pos++] = previous_byte;
329 if (buffer_pos == header.dict_size) {
331 global_pos += header.dict_size;
332 if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
334 USE_DESKTOP(total_written += header.dict_size;)
344 rc_update_bit_1(rc, prob);
345 prob = p + LZMA_IS_REP + state;
346 if (rc_is_bit_0(rc, prob)) {
347 rc_update_bit_0(rc, prob);
351 state = state < LZMA_NUM_LIT_STATES ? 0 : 3;
352 prob = p + LZMA_LEN_CODER;
354 rc_update_bit_1(rc, prob);
355 prob = p + LZMA_IS_REP_G0 + state;
356 if (rc_is_bit_0(rc, prob)) {
357 rc_update_bit_0(rc, prob);
358 prob = (p + LZMA_IS_REP_0_LONG
359 + (state << LZMA_NUM_POS_BITS_MAX)
362 if (rc_is_bit_0(rc, prob)) {
363 rc_update_bit_0(rc, prob);
365 state = state < LZMA_NUM_LIT_STATES ? 9 : 11;
366 #if ENABLE_FEATURE_LZMA_FAST
367 pos = buffer_pos - rep0;
368 while (pos >= header.dict_size)
369 pos += header.dict_size;
370 previous_byte = buffer[pos];
377 rc_update_bit_1(rc, prob);
382 rc_update_bit_1(rc, prob);
383 prob = p + LZMA_IS_REP_G1 + state;
384 if (rc_is_bit_0(rc, prob)) {
385 rc_update_bit_0(rc, prob);
388 rc_update_bit_1(rc, prob);
389 prob = p + LZMA_IS_REP_G2 + state;
390 if (rc_is_bit_0(rc, prob)) {
391 rc_update_bit_0(rc, prob);
394 rc_update_bit_1(rc, prob);
403 state = state < LZMA_NUM_LIT_STATES ? 8 : 11;
404 prob = p + LZMA_REP_LEN_CODER;
407 prob_len = prob + LZMA_LEN_CHOICE;
408 if (rc_is_bit_0(rc, prob_len)) {
409 rc_update_bit_0(rc, prob_len);
410 prob_len = (prob + LZMA_LEN_LOW
411 + (pos_state << LZMA_LEN_NUM_LOW_BITS));
413 num_bits = LZMA_LEN_NUM_LOW_BITS;
415 rc_update_bit_1(rc, prob_len);
416 prob_len = prob + LZMA_LEN_CHOICE_2;
417 if (rc_is_bit_0(rc, prob_len)) {
418 rc_update_bit_0(rc, prob_len);
419 prob_len = (prob + LZMA_LEN_MID
420 + (pos_state << LZMA_LEN_NUM_MID_BITS));
421 offset = 1 << LZMA_LEN_NUM_LOW_BITS;
422 num_bits = LZMA_LEN_NUM_MID_BITS;
424 rc_update_bit_1(rc, prob_len);
425 prob_len = prob + LZMA_LEN_HIGH;
426 offset = ((1 << LZMA_LEN_NUM_LOW_BITS)
427 + (1 << LZMA_LEN_NUM_MID_BITS));
428 num_bits = LZMA_LEN_NUM_HIGH_BITS;
431 rc_bit_tree_decode(rc, prob_len, num_bits, &len);
437 state += LZMA_NUM_LIT_STATES;
438 prob = p + LZMA_POS_SLOT +
439 ((len < LZMA_NUM_LEN_TO_POS_STATES ? len :
440 LZMA_NUM_LEN_TO_POS_STATES - 1)
441 << LZMA_NUM_POS_SLOT_BITS);
442 rc_bit_tree_decode(rc, prob, LZMA_NUM_POS_SLOT_BITS,
444 if (pos_slot >= LZMA_START_POS_MODEL_INDEX) {
445 num_bits = (pos_slot >> 1) - 1;
446 rep0 = 2 | (pos_slot & 1);
447 if (pos_slot < LZMA_END_POS_MODEL_INDEX) {
449 prob = p + LZMA_SPEC_POS + rep0 - pos_slot - 1;
451 num_bits -= LZMA_NUM_ALIGN_BITS;
453 rep0 = (rep0 << 1) | rc_direct_bit(rc);
454 prob = p + LZMA_ALIGN;
455 rep0 <<= LZMA_NUM_ALIGN_BITS;
456 num_bits = LZMA_NUM_ALIGN_BITS;
461 if (rc_get_bit(rc, prob + mi, &mi))
471 len += LZMA_MATCH_MIN_LEN;
472 SKIP_FEATURE_LZMA_FAST(string:)
474 pos = buffer_pos - rep0;
475 while (pos >= header.dict_size)
476 pos += header.dict_size;
477 previous_byte = buffer[pos];
478 SKIP_FEATURE_LZMA_FAST(one_byte2:)
479 buffer[buffer_pos++] = previous_byte;
480 if (buffer_pos == header.dict_size) {
482 global_pos += header.dict_size;
483 if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
485 USE_DESKTOP(total_written += header.dict_size;)
488 } while (len != 0 && buffer_pos < header.dst_size);
492 if (full_write(dst_fd, buffer, buffer_pos) != (ssize_t)buffer_pos) {
498 USE_DESKTOP(total_written += buffer_pos;)
499 return USE_DESKTOP(total_written) + 0;