3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * Copyright (c) 2010 Loren Merritt
7 * alternative bitstream reader & writer by Michael Niedermayer <michaelni@gmx.at>
9 * This file is part of FFmpeg.
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.
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.
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
31 #include "libavutil/atomic.h"
32 #include "libavutil/avassert.h"
39 const uint8_t ff_log2_run[41]={
40 0, 0, 0, 0, 1, 1, 1, 1,
41 2, 2, 2, 2, 3, 3, 3, 3,
42 4, 4, 5, 5, 6, 6, 7, 7,
43 8, 9,10,11,12,13,14,15,
44 16,17,18,19,20,21,22,23,
48 void avpriv_align_put_bits(PutBitContext *s)
50 put_bits(s, s->bit_left & 7, 0);
53 void avpriv_put_string(PutBitContext *pb, const char *string,
57 put_bits(pb, 8, *string);
64 void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
66 int words = length >> 4;
67 int bits = length & 15;
73 if (CONFIG_SMALL || words < 16 || put_bits_count(pb) & 7) {
74 for (i = 0; i < words; i++)
75 put_bits(pb, 16, AV_RB16(src + 2 * i));
77 for (i = 0; put_bits_count(pb) & 31; i++)
78 put_bits(pb, 8, src[i]);
80 memcpy(put_bits_ptr(pb), src + i, 2 * words - i);
81 skip_put_bytes(pb, 2 * words - i);
84 put_bits(pb, bits, AV_RB16(src + 2 * words) >> (16 - bits));
89 #define GET_DATA(v, table, i, wrap, size) \
91 const uint8_t *ptr = (const uint8_t *)table + i * wrap; \
94 v = *(const uint8_t *)ptr; \
97 v = *(const uint16_t *)ptr; \
100 v = *(const uint32_t *)ptr; \
106 static int alloc_table(VLC *vlc, int size, int use_static)
108 int index = vlc->table_size;
110 vlc->table_size += size;
111 if (vlc->table_size > vlc->table_allocated) {
113 abort(); // cannot do anything, init_vlc() is used with too little memory
114 vlc->table_allocated += (1 << vlc->bits);
115 vlc->table = av_realloc_f(vlc->table, vlc->table_allocated, sizeof(VLC_TYPE) * 2);
117 vlc->table_allocated = 0;
119 return AVERROR(ENOMEM);
121 memset(vlc->table + vlc->table_allocated - (1 << vlc->bits), 0, sizeof(VLC_TYPE) * 2 << vlc->bits);
126 static av_always_inline uint32_t bitswap_32(uint32_t x)
128 return (uint32_t)ff_reverse[ x & 0xFF] << 24 |
129 (uint32_t)ff_reverse[(x >> 8) & 0xFF] << 16 |
130 (uint32_t)ff_reverse[(x >> 16) & 0xFF] << 8 |
131 (uint32_t)ff_reverse[ x >> 24];
134 typedef struct VLCcode {
137 /** codeword, with the first bit-to-be-read in the msb
138 * (even if intended for a little-endian bitstream reader) */
142 static int compare_vlcspec(const void *a, const void *b)
144 const VLCcode *sa = a, *sb = b;
145 return (sa->code >> 1) - (sb->code >> 1);
148 * Build VLC decoding tables suitable for use with get_vlc().
150 * @param vlc the context to be initted
152 * @param table_nb_bits max length of vlc codes to store directly in this table
153 * (Longer codes are delegated to subtables.)
155 * @param nb_codes number of elements in codes[]
157 * @param codes descriptions of the vlc codes
158 * These must be ordered such that codes going into the same subtable are contiguous.
159 * Sorting by VLCcode.code is sufficient, though not necessary.
161 static int build_table(VLC *vlc, int table_nb_bits, int nb_codes,
162 VLCcode *codes, int flags)
164 int table_size, table_index, index, code_prefix, symbol, subtable_bits;
165 int i, j, k, n, nb, inc;
167 volatile VLC_TYPE (* volatile table)[2]; // the double volatile is needed to prevent a internal compiler error in gcc 4.2
169 table_size = 1 << table_nb_bits;
170 if (table_nb_bits > 30)
172 table_index = alloc_table(vlc, table_size, flags & INIT_VLC_USE_NEW_STATIC);
173 ff_dlog(NULL, "new table index=%d size=%d\n", table_index, table_size);
176 table = (volatile VLC_TYPE (*)[2])&vlc->table[table_index];
178 /* first pass: map codes and compute auxiliary table sizes */
179 for (i = 0; i < nb_codes; i++) {
181 code = codes[i].code;
182 symbol = codes[i].symbol;
183 ff_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);
189 if (flags & INIT_VLC_LE) {
190 j = bitswap_32(code);
193 for (k = 0; k < nb; k++) {
194 int bits = table[j][1];
195 ff_dlog(NULL, "%4x: code=%d n=%d\n", j, i, n);
196 if (bits != 0 && bits != n) {
197 av_log(NULL, AV_LOG_ERROR, "incorrect codes\n");
198 return AVERROR_INVALIDDATA;
200 table[j][1] = n; //bits
201 table[j][0] = symbol;
205 /* fill auxiliary table recursively */
207 code_prefix = code >> (32 - table_nb_bits);
210 codes[i].code = code << table_nb_bits;
211 for (k = i+1; k < nb_codes; k++) {
212 n = codes[k].bits - table_nb_bits;
215 code = codes[k].code;
216 if (code >> (32 - table_nb_bits) != code_prefix)
219 codes[k].code = code << table_nb_bits;
220 subtable_bits = FFMAX(subtable_bits, n);
222 subtable_bits = FFMIN(subtable_bits, table_nb_bits);
223 j = (flags & INIT_VLC_LE) ? bitswap_32(code_prefix) >> (32 - table_nb_bits) : code_prefix;
224 table[j][1] = -subtable_bits;
225 ff_dlog(NULL, "%4x: n=%d (subtable)\n",
226 j, codes[i].bits + table_nb_bits);
227 index = build_table(vlc, subtable_bits, k-i, codes+i, flags);
230 /* note: realloc has been done, so reload tables */
231 table = (volatile VLC_TYPE (*)[2])&vlc->table[table_index];
232 table[j][0] = index; //code
237 for (i = 0; i < table_size; i++) {
238 if (table[i][1] == 0) //bits
239 table[i][0] = -1; //codes
246 /* Build VLC decoding tables suitable for use with get_vlc().
248 'nb_bits' set thee decoding table size (2^nb_bits) entries. The
249 bigger it is, the faster is the decoding. But it should not be too
250 big to save memory and L1 cache. '9' is a good compromise.
252 'nb_codes' : number of vlcs codes
254 'bits' : table which gives the size (in bits) of each vlc code.
256 'codes' : table which gives the bit pattern of of each vlc code.
258 'symbols' : table which gives the values to be returned from get_vlc().
260 'xxx_wrap' : give the number of bytes between each entry of the
261 'bits' or 'codes' tables.
263 'xxx_size' : gives the number of bytes of each entry of the 'bits'
266 'wrap' and 'size' allows to use any memory configuration and types
267 (byte/word/long) to store the 'bits', 'codes', and 'symbols' tables.
269 'use_static' should be set to 1 for tables, which should be freed
270 with av_free_static(), 0 if ff_free_vlc() will be used.
272 int ff_init_vlc_sparse(VLC *vlc_arg, int nb_bits, int nb_codes,
273 const void *bits, int bits_wrap, int bits_size,
274 const void *codes, int codes_wrap, int codes_size,
275 const void *symbols, int symbols_wrap, int symbols_size,
280 VLCcode localbuf[1500]; // the maximum currently needed is 1296 by rv34
285 if (flags & INIT_VLC_USE_NEW_STATIC) {
286 av_assert0(nb_codes + 1 <= FF_ARRAY_ELEMS(localbuf));
293 vlc->table_allocated = 0;
296 buf = av_malloc_array((nb_codes + 1), sizeof(VLCcode));
298 return AVERROR(ENOMEM);
302 av_assert0(symbols_size <= 2 || !symbols);
304 #define COPY(condition)\
305 for (i = 0; i < nb_codes; i++) { \
306 GET_DATA(buf[j].bits, bits, i, bits_wrap, bits_size); \
309 if (buf[j].bits > 3*nb_bits || buf[j].bits>32) { \
310 av_log(NULL, AV_LOG_ERROR, "Too long VLC (%d) in init_vlc\n", buf[j].bits);\
311 if (!(flags & INIT_VLC_USE_NEW_STATIC)) \
315 GET_DATA(buf[j].code, codes, i, codes_wrap, codes_size); \
316 if (buf[j].code >= (1LL<<buf[j].bits)) { \
317 av_log(NULL, AV_LOG_ERROR, "Invalid code in init_vlc\n"); \
318 if (!(flags & INIT_VLC_USE_NEW_STATIC)) \
322 if (flags & INIT_VLC_LE) \
323 buf[j].code = bitswap_32(buf[j].code); \
325 buf[j].code <<= 32 - buf[j].bits; \
327 GET_DATA(buf[j].symbol, symbols, i, symbols_wrap, symbols_size) \
332 COPY(buf[j].bits > nb_bits);
333 // qsort is the slowest part of init_vlc, and could probably be improved or avoided
334 qsort(buf, j, sizeof(VLCcode), compare_vlcspec);
335 COPY(buf[j].bits && buf[j].bits <= nb_bits);
338 ret = build_table(vlc, nb_bits, nb_codes, buf, flags);
340 if (flags & INIT_VLC_USE_NEW_STATIC) {
341 if(vlc->table_size != vlc->table_allocated)
342 av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", vlc->table_size, vlc->table_allocated);
344 av_assert0(ret >= 0);
349 av_freep(&vlc->table);
357 void ff_free_vlc(VLC *vlc)
359 av_freep(&vlc->table);