Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4349 | Serge | 1 | /* |
2 | * Common bit i/o utils |
||
3 | * Copyright (c) 2000, 2001 Fabrice Bellard |
||
4 | * Copyright (c) 2002-2004 Michael Niedermayer |
||
5 | * Copyright (c) 2010 Loren Merritt |
||
6 | * |
||
7 | * alternative bitstream reader & writer by Michael Niedermayer |
||
8 | * |
||
9 | * This file is part of FFmpeg. |
||
10 | * |
||
11 | * FFmpeg is free software; you can redistribute it and/or |
||
12 | * modify it under the terms of the GNU Lesser General Public |
||
13 | * License as published by the Free Software Foundation; either |
||
14 | * version 2.1 of the License, or (at your option) any later version. |
||
15 | * |
||
16 | * FFmpeg is distributed in the hope that it will be useful, |
||
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
19 | * Lesser General Public License for more details. |
||
20 | * |
||
21 | * You should have received a copy of the GNU Lesser General Public |
||
22 | * License along with FFmpeg; if not, write to the Free Software |
||
23 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
||
24 | */ |
||
25 | |||
26 | /** |
||
27 | * @file |
||
28 | * bitstream api. |
||
29 | */ |
||
30 | |||
31 | #include "libavutil/atomic.h" |
||
32 | #include "libavutil/avassert.h" |
||
33 | #include "avcodec.h" |
||
34 | #include "mathops.h" |
||
35 | #include "get_bits.h" |
||
36 | #include "put_bits.h" |
||
37 | |||
38 | const uint8_t ff_log2_run[41]={ |
||
39 | 0, 0, 0, 0, 1, 1, 1, 1, |
||
40 | 2, 2, 2, 2, 3, 3, 3, 3, |
||
41 | 4, 4, 5, 5, 6, 6, 7, 7, |
||
42 | 8, 9,10,11,12,13,14,15, |
||
43 | 16,17,18,19,20,21,22,23, |
||
44 | 24, |
||
45 | }; |
||
46 | |||
47 | void avpriv_align_put_bits(PutBitContext *s) |
||
48 | { |
||
49 | put_bits(s, s->bit_left & 7, 0); |
||
50 | } |
||
51 | |||
52 | void avpriv_put_string(PutBitContext *pb, const char *string, |
||
53 | int terminate_string) |
||
54 | { |
||
55 | while (*string) { |
||
56 | put_bits(pb, 8, *string); |
||
57 | string++; |
||
58 | } |
||
59 | if (terminate_string) |
||
60 | put_bits(pb, 8, 0); |
||
61 | } |
||
62 | |||
63 | void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length) |
||
64 | { |
||
65 | int words = length >> 4; |
||
66 | int bits = length & 15; |
||
67 | int i; |
||
68 | |||
69 | if (length == 0) |
||
70 | return; |
||
71 | |||
72 | if (CONFIG_SMALL || words < 16 || put_bits_count(pb) & 7) { |
||
73 | for (i = 0; i < words; i++) |
||
74 | put_bits(pb, 16, AV_RB16(src + 2 * i)); |
||
75 | } else { |
||
76 | for (i = 0; put_bits_count(pb) & 31; i++) |
||
77 | put_bits(pb, 8, src[i]); |
||
78 | flush_put_bits(pb); |
||
79 | memcpy(put_bits_ptr(pb), src + i, 2 * words - i); |
||
80 | skip_put_bytes(pb, 2 * words - i); |
||
81 | } |
||
82 | |||
83 | put_bits(pb, bits, AV_RB16(src + 2 * words) >> (16 - bits)); |
||
84 | } |
||
85 | |||
86 | /* VLC decoding */ |
||
87 | |||
88 | #define GET_DATA(v, table, i, wrap, size) \ |
||
89 | { \ |
||
90 | const uint8_t *ptr = (const uint8_t *)table + i * wrap; \ |
||
91 | switch(size) { \ |
||
92 | case 1: \ |
||
93 | v = *(const uint8_t *)ptr; \ |
||
94 | break; \ |
||
95 | case 2: \ |
||
96 | v = *(const uint16_t *)ptr; \ |
||
97 | break; \ |
||
98 | default: \ |
||
99 | v = *(const uint32_t *)ptr; \ |
||
100 | break; \ |
||
101 | } \ |
||
102 | } |
||
103 | |||
104 | |||
105 | static int alloc_table(VLC *vlc, int size, int use_static) |
||
106 | { |
||
107 | int index = vlc->table_size; |
||
108 | |||
109 | vlc->table_size += size; |
||
110 | if (vlc->table_size > vlc->table_allocated) { |
||
111 | if (use_static) |
||
112 | abort(); // cannot do anything, init_vlc() is used with too little memory |
||
113 | vlc->table_allocated += (1 << vlc->bits); |
||
114 | vlc->table = av_realloc_f(vlc->table, vlc->table_allocated, sizeof(VLC_TYPE) * 2); |
||
115 | if (!vlc->table) |
||
116 | return AVERROR(ENOMEM); |
||
117 | } |
||
118 | return index; |
||
119 | } |
||
120 | |||
121 | static av_always_inline uint32_t bitswap_32(uint32_t x) |
||
122 | { |
||
123 | return (uint32_t)ff_reverse[ x & 0xFF] << 24 | |
||
124 | (uint32_t)ff_reverse[(x >> 8) & 0xFF] << 16 | |
||
125 | (uint32_t)ff_reverse[(x >> 16) & 0xFF] << 8 | |
||
126 | (uint32_t)ff_reverse[ x >> 24]; |
||
127 | } |
||
128 | |||
129 | typedef struct { |
||
130 | uint8_t bits; |
||
131 | uint16_t symbol; |
||
132 | /** codeword, with the first bit-to-be-read in the msb |
||
133 | * (even if intended for a little-endian bitstream reader) */ |
||
134 | uint32_t code; |
||
135 | } VLCcode; |
||
136 | |||
137 | static int compare_vlcspec(const void *a, const void *b) |
||
138 | { |
||
139 | const VLCcode *sa = a, *sb = b; |
||
140 | return (sa->code >> 1) - (sb->code >> 1); |
||
141 | } |
||
142 | /** |
||
143 | * Build VLC decoding tables suitable for use with get_vlc(). |
||
144 | * |
||
145 | * @param vlc the context to be initted |
||
146 | * |
||
147 | * @param table_nb_bits max length of vlc codes to store directly in this table |
||
148 | * (Longer codes are delegated to subtables.) |
||
149 | * |
||
150 | * @param nb_codes number of elements in codes[] |
||
151 | * |
||
152 | * @param codes descriptions of the vlc codes |
||
153 | * These must be ordered such that codes going into the same subtable are contiguous. |
||
154 | * Sorting by VLCcode.code is sufficient, though not necessary. |
||
155 | */ |
||
156 | static int build_table(VLC *vlc, int table_nb_bits, int nb_codes, |
||
157 | VLCcode *codes, int flags) |
||
158 | { |
||
159 | int table_size, table_index, index, code_prefix, symbol, subtable_bits; |
||
160 | int i, j, k, n, nb, inc; |
||
161 | uint32_t code; |
||
162 | VLC_TYPE (*table)[2]; |
||
163 | |||
164 | table_size = 1 << table_nb_bits; |
||
165 | if (table_nb_bits > 30) |
||
166 | return -1; |
||
167 | table_index = alloc_table(vlc, table_size, flags & INIT_VLC_USE_NEW_STATIC); |
||
168 | av_dlog(NULL, "new table index=%d size=%d\n", table_index, table_size); |
||
169 | if (table_index < 0) |
||
170 | return table_index; |
||
171 | table = &vlc->table[table_index]; |
||
172 | |||
173 | for (i = 0; i < table_size; i++) { |
||
174 | table[i][1] = 0; //bits |
||
175 | table[i][0] = -1; //codes |
||
176 | } |
||
177 | |||
178 | /* first pass: map codes and compute auxiliary table sizes */ |
||
179 | for (i = 0; i < nb_codes; i++) { |
||
180 | n = codes[i].bits; |
||
181 | code = codes[i].code; |
||
182 | symbol = codes[i].symbol; |
||
183 | av_dlog(NULL, "i=%d n=%d code=0x%x\n", i, n, code); |
||
184 | if (n <= table_nb_bits) { |
||
185 | /* no need to add another table */ |
||
186 | j = code >> (32 - table_nb_bits); |
||
187 | nb = 1 << (table_nb_bits - n); |
||
188 | inc = 1; |
||
189 | if (flags & INIT_VLC_LE) { |
||
190 | j = bitswap_32(code); |
||
191 | inc = 1 << n; |
||
192 | } |
||
193 | for (k = 0; k < nb; k++) { |
||
194 | av_dlog(NULL, "%4x: code=%d n=%d\n", j, i, n); |
||
195 | if (table[j][1] /*bits*/ != 0) { |
||
196 | av_log(NULL, AV_LOG_ERROR, "incorrect codes\n"); |
||
197 | return AVERROR_INVALIDDATA; |
||
198 | } |
||
199 | table[j][1] = n; //bits |
||
200 | table[j][0] = symbol; |
||
201 | j += inc; |
||
202 | } |
||
203 | } else { |
||
204 | /* fill auxiliary table recursively */ |
||
205 | n -= table_nb_bits; |
||
206 | code_prefix = code >> (32 - table_nb_bits); |
||
207 | subtable_bits = n; |
||
208 | codes[i].bits = n; |
||
209 | codes[i].code = code << table_nb_bits; |
||
210 | for (k = i+1; k < nb_codes; k++) { |
||
211 | n = codes[k].bits - table_nb_bits; |
||
212 | if (n <= 0) |
||
213 | break; |
||
214 | code = codes[k].code; |
||
215 | if (code >> (32 - table_nb_bits) != code_prefix) |
||
216 | break; |
||
217 | codes[k].bits = n; |
||
218 | codes[k].code = code << table_nb_bits; |
||
219 | subtable_bits = FFMAX(subtable_bits, n); |
||
220 | } |
||
221 | subtable_bits = FFMIN(subtable_bits, table_nb_bits); |
||
222 | j = (flags & INIT_VLC_LE) ? bitswap_32(code_prefix) >> (32 - table_nb_bits) : code_prefix; |
||
223 | table[j][1] = -subtable_bits; |
||
224 | av_dlog(NULL, "%4x: n=%d (subtable)\n", |
||
225 | j, codes[i].bits + table_nb_bits); |
||
226 | index = build_table(vlc, subtable_bits, k-i, codes+i, flags); |
||
227 | if (index < 0) |
||
228 | return index; |
||
229 | /* note: realloc has been done, so reload tables */ |
||
230 | table = &vlc->table[table_index]; |
||
231 | table[j][0] = index; //code |
||
232 | i = k-1; |
||
233 | } |
||
234 | } |
||
235 | return table_index; |
||
236 | } |
||
237 | |||
238 | |||
239 | /* Build VLC decoding tables suitable for use with get_vlc(). |
||
240 | |||
241 | 'nb_bits' set thee decoding table size (2^nb_bits) entries. The |
||
242 | bigger it is, the faster is the decoding. But it should not be too |
||
243 | big to save memory and L1 cache. '9' is a good compromise. |
||
244 | |||
245 | 'nb_codes' : number of vlcs codes |
||
246 | |||
247 | 'bits' : table which gives the size (in bits) of each vlc code. |
||
248 | |||
249 | 'codes' : table which gives the bit pattern of of each vlc code. |
||
250 | |||
251 | 'symbols' : table which gives the values to be returned from get_vlc(). |
||
252 | |||
253 | 'xxx_wrap' : give the number of bytes between each entry of the |
||
254 | 'bits' or 'codes' tables. |
||
255 | |||
256 | 'xxx_size' : gives the number of bytes of each entry of the 'bits' |
||
257 | or 'codes' tables. |
||
258 | |||
259 | 'wrap' and 'size' allows to use any memory configuration and types |
||
260 | (byte/word/long) to store the 'bits', 'codes', and 'symbols' tables. |
||
261 | |||
262 | 'use_static' should be set to 1 for tables, which should be freed |
||
263 | with av_free_static(), 0 if ff_free_vlc() will be used. |
||
264 | */ |
||
265 | int ff_init_vlc_sparse(VLC *vlc, int nb_bits, int nb_codes, |
||
266 | const void *bits, int bits_wrap, int bits_size, |
||
267 | const void *codes, int codes_wrap, int codes_size, |
||
268 | const void *symbols, int symbols_wrap, int symbols_size, |
||
269 | int flags) |
||
270 | { |
||
271 | VLCcode *buf; |
||
272 | int i, j, ret; |
||
273 | VLCcode localbuf[1500]; // the maximum currently needed is 1296 by rv34 |
||
274 | void *state; |
||
275 | |||
276 | vlc->bits = nb_bits; |
||
277 | if (flags & INIT_VLC_USE_NEW_STATIC) { |
||
278 | while (state = avpriv_atomic_ptr_cas(&vlc->init_state, NULL, vlc)) { |
||
279 | if (state == vlc + 1) { |
||
280 | av_assert0(vlc->table_size && vlc->table_size == vlc->table_allocated); |
||
281 | return 0; |
||
282 | } |
||
283 | } |
||
284 | av_assert0(!vlc->table_size); |
||
285 | av_assert0(nb_codes + 1 <= FF_ARRAY_ELEMS(localbuf)); |
||
286 | buf = localbuf; |
||
287 | } else { |
||
288 | vlc->table = NULL; |
||
289 | vlc->table_allocated = 0; |
||
290 | vlc->table_size = 0; |
||
291 | |||
292 | buf = av_malloc((nb_codes + 1) * sizeof(VLCcode)); |
||
293 | if (!buf) |
||
294 | return AVERROR(ENOMEM); |
||
295 | } |
||
296 | |||
297 | |||
298 | av_assert0(symbols_size <= 2 || !symbols); |
||
299 | j = 0; |
||
300 | #define COPY(condition)\ |
||
301 | for (i = 0; i < nb_codes; i++) { \ |
||
302 | GET_DATA(buf[j].bits, bits, i, bits_wrap, bits_size); \ |
||
303 | if (!(condition)) \ |
||
304 | continue; \ |
||
305 | if (buf[j].bits > 3*nb_bits || buf[j].bits>32) { \ |
||
306 | av_log(NULL, AV_LOG_ERROR, "Too long VLC (%d) in init_vlc\n", buf[j].bits);\ |
||
307 | if (!(flags & INIT_VLC_USE_NEW_STATIC)) \ |
||
308 | av_free(buf); \ |
||
309 | return -1; \ |
||
310 | } \ |
||
311 | GET_DATA(buf[j].code, codes, i, codes_wrap, codes_size); \ |
||
312 | if (buf[j].code >= (1LL< |
||
313 | av_log(NULL, AV_LOG_ERROR, "Invalid code in init_vlc\n"); \ |
||
314 | if (!(flags & INIT_VLC_USE_NEW_STATIC)) \ |
||
315 | av_free(buf); \ |
||
316 | return -1; \ |
||
317 | } \ |
||
318 | if (flags & INIT_VLC_LE) \ |
||
319 | buf[j].code = bitswap_32(buf[j].code); \ |
||
320 | else \ |
||
321 | buf[j].code <<= 32 - buf[j].bits; \ |
||
322 | if (symbols) \ |
||
323 | GET_DATA(buf[j].symbol, symbols, i, symbols_wrap, symbols_size) \ |
||
324 | else \ |
||
325 | buf[j].symbol = i; \ |
||
326 | j++; \ |
||
327 | } |
||
328 | COPY(buf[j].bits > nb_bits); |
||
329 | // qsort is the slowest part of init_vlc, and could probably be improved or avoided |
||
330 | qsort(buf, j, sizeof(VLCcode), compare_vlcspec); |
||
331 | COPY(buf[j].bits && buf[j].bits <= nb_bits); |
||
332 | nb_codes = j; |
||
333 | |||
334 | ret = build_table(vlc, nb_bits, nb_codes, buf, flags); |
||
335 | |||
336 | if (flags & INIT_VLC_USE_NEW_STATIC) { |
||
337 | if(vlc->table_size != vlc->table_allocated) |
||
338 | av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", vlc->table_size, vlc->table_allocated); |
||
339 | state = avpriv_atomic_ptr_cas(&vlc->init_state, vlc, vlc+1); |
||
340 | av_assert0(state == vlc); |
||
341 | av_assert0(ret >= 0); |
||
342 | } else { |
||
343 | av_free(buf); |
||
344 | if (ret < 0) { |
||
345 | av_freep(&vlc->table); |
||
346 | return ret; |
||
347 | } |
||
348 | } |
||
349 | return 0; |
||
350 | } |
||
351 | |||
352 | |||
353 | void ff_free_vlc(VLC *vlc) |
||
354 | { |
||
355 | av_freep(&vlc->table); |
||
356 | }>=>=><=> |