yading@10: /*
yading@10:  * Common bit i/o utils
yading@10:  * Copyright (c) 2000, 2001 Fabrice Bellard
yading@10:  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
yading@10:  * Copyright (c) 2010 Loren Merritt
yading@10:  *
yading@10:  * alternative bitstream reader & writer by Michael Niedermayer <michaelni@gmx.at>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * bitstream api.
yading@10:  */
yading@10: 
yading@10: #include "libavutil/avassert.h"
yading@10: #include "avcodec.h"
yading@10: #include "mathops.h"
yading@10: #include "get_bits.h"
yading@10: #include "put_bits.h"
yading@10: 
yading@10: const uint8_t ff_log2_run[41]={
yading@10:  0, 0, 0, 0, 1, 1, 1, 1,
yading@10:  2, 2, 2, 2, 3, 3, 3, 3,
yading@10:  4, 4, 5, 5, 6, 6, 7, 7,
yading@10:  8, 9,10,11,12,13,14,15,
yading@10: 16,17,18,19,20,21,22,23,
yading@10: 24,
yading@10: };
yading@10: 
yading@10: void avpriv_align_put_bits(PutBitContext *s)
yading@10: {
yading@10:     put_bits(s,s->bit_left & 7,0);
yading@10: }
yading@10: 
yading@10: void avpriv_put_string(PutBitContext *pb, const char *string, int terminate_string)
yading@10: {
yading@10:     while(*string){
yading@10:         put_bits(pb, 8, *string);
yading@10:         string++;
yading@10:     }
yading@10:     if(terminate_string)
yading@10:         put_bits(pb, 8, 0);
yading@10: }
yading@10: 
yading@10: void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
yading@10: {
yading@10:     int words= length>>4;
yading@10:     int bits= length&15;
yading@10:     int i;
yading@10: 
yading@10:     if(length==0) return;
yading@10: 
yading@10:     if(CONFIG_SMALL || words < 16 || put_bits_count(pb)&7){
yading@10:         for(i=0; i<words; i++) put_bits(pb, 16, AV_RB16(src + 2*i));
yading@10:     }else{
yading@10:         for(i=0; put_bits_count(pb)&31; i++)
yading@10:             put_bits(pb, 8, src[i]);
yading@10:         flush_put_bits(pb);
yading@10:         memcpy(put_bits_ptr(pb), src+i, 2*words-i);
yading@10:         skip_put_bytes(pb, 2*words-i);
yading@10:     }
yading@10: 
yading@10:     put_bits(pb, bits, AV_RB16(src + 2*words)>>(16-bits));
yading@10: }
yading@10: 
yading@10: /* VLC decoding */
yading@10: 
yading@10: #define GET_DATA(v, table, i, wrap, size) \
yading@10: {\
yading@10:     const uint8_t *ptr = (const uint8_t *)table + i * wrap;\
yading@10:     switch(size) {\
yading@10:     case 1:\
yading@10:         v = *(const uint8_t *)ptr;\
yading@10:         break;\
yading@10:     case 2:\
yading@10:         v = *(const uint16_t *)ptr;\
yading@10:         break;\
yading@10:     default:\
yading@10:         v = *(const uint32_t *)ptr;\
yading@10:         break;\
yading@10:     }\
yading@10: }
yading@10: 
yading@10: 
yading@10: static int alloc_table(VLC *vlc, int size, int use_static)
yading@10: {
yading@10:     int index;
yading@10:     index = vlc->table_size;
yading@10:     vlc->table_size += size;
yading@10:     if (vlc->table_size > vlc->table_allocated) {
yading@10:         if(use_static)
yading@10:             abort(); // cannot do anything, init_vlc() is used with too little memory
yading@10:         vlc->table_allocated += (1 << vlc->bits);
yading@10:         vlc->table = av_realloc_f(vlc->table,
yading@10:                                   vlc->table_allocated, sizeof(VLC_TYPE) * 2);
yading@10:         if (!vlc->table)
yading@10:             return -1;
yading@10:     }
yading@10:     return index;
yading@10: }
yading@10: 
yading@10: static av_always_inline uint32_t bitswap_32(uint32_t x) {
yading@10:     return (uint32_t)ff_reverse[x&0xFF]<<24
yading@10:          | (uint32_t)ff_reverse[(x>>8)&0xFF]<<16
yading@10:          | (uint32_t)ff_reverse[(x>>16)&0xFF]<<8
yading@10:          | (uint32_t)ff_reverse[x>>24];
yading@10: }
yading@10: 
yading@10: typedef struct {
yading@10:     uint8_t bits;
yading@10:     uint16_t symbol;
yading@10:     /** codeword, with the first bit-to-be-read in the msb
yading@10:      * (even if intended for a little-endian bitstream reader) */
yading@10:     uint32_t code;
yading@10: } VLCcode;
yading@10: 
yading@10: static int compare_vlcspec(const void *a, const void *b)
yading@10: {
yading@10:     const VLCcode *sa=a, *sb=b;
yading@10:     return (sa->code >> 1) - (sb->code >> 1);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Build VLC decoding tables suitable for use with get_vlc().
yading@10:  *
yading@10:  * @param vlc            the context to be initted
yading@10:  *
yading@10:  * @param table_nb_bits  max length of vlc codes to store directly in this table
yading@10:  *                       (Longer codes are delegated to subtables.)
yading@10:  *
yading@10:  * @param nb_codes       number of elements in codes[]
yading@10:  *
yading@10:  * @param codes          descriptions of the vlc codes
yading@10:  *                       These must be ordered such that codes going into the same subtable are contiguous.
yading@10:  *                       Sorting by VLCcode.code is sufficient, though not necessary.
yading@10:  */
yading@10: static int build_table(VLC *vlc, int table_nb_bits, int nb_codes,
yading@10:                        VLCcode *codes, int flags)
yading@10: {
yading@10:     int table_size, table_index, index, code_prefix, symbol, subtable_bits;
yading@10:     int i, j, k, n, nb, inc;
yading@10:     uint32_t code;
yading@10:     VLC_TYPE (*table)[2];
yading@10: 
yading@10:     table_size = 1 << table_nb_bits;
yading@10:     if (table_nb_bits > 30)
yading@10:        return -1;
yading@10:     table_index = alloc_table(vlc, table_size, flags & INIT_VLC_USE_NEW_STATIC);
yading@10:     av_dlog(NULL, "new table index=%d size=%d\n", table_index, table_size);
yading@10:     if (table_index < 0)
yading@10:         return -1;
yading@10:     table = &vlc->table[table_index];
yading@10: 
yading@10:     for (i = 0; i < table_size; i++) {
yading@10:         table[i][1] = 0; //bits
yading@10:         table[i][0] = -1; //codes
yading@10:     }
yading@10: 
yading@10:     /* first pass: map codes and compute auxiliary table sizes */
yading@10:     for (i = 0; i < nb_codes; i++) {
yading@10:         n = codes[i].bits;
yading@10:         code = codes[i].code;
yading@10:         symbol = codes[i].symbol;
yading@10:         av_dlog(NULL, "i=%d n=%d code=0x%x\n", i, n, code);
yading@10:         if (n <= table_nb_bits) {
yading@10:             /* no need to add another table */
yading@10:             j = code >> (32 - table_nb_bits);
yading@10:             nb = 1 << (table_nb_bits - n);
yading@10:             inc = 1;
yading@10:             if (flags & INIT_VLC_LE) {
yading@10:                 j = bitswap_32(code);
yading@10:                 inc = 1 << n;
yading@10:             }
yading@10:             for (k = 0; k < nb; k++) {
yading@10:                 av_dlog(NULL, "%4x: code=%d n=%d\n", j, i, n);
yading@10:                 if (table[j][1] /*bits*/ != 0) {
yading@10:                     av_log(NULL, AV_LOG_ERROR, "incorrect codes\n");
yading@10:                     return -1;
yading@10:                 }
yading@10:                 table[j][1] = n; //bits
yading@10:                 table[j][0] = symbol;
yading@10:                 j += inc;
yading@10:             }
yading@10:         } else {
yading@10:             /* fill auxiliary table recursively */
yading@10:             n -= table_nb_bits;
yading@10:             code_prefix = code >> (32 - table_nb_bits);
yading@10:             subtable_bits = n;
yading@10:             codes[i].bits = n;
yading@10:             codes[i].code = code << table_nb_bits;
yading@10:             for (k = i+1; k < nb_codes; k++) {
yading@10:                 n = codes[k].bits - table_nb_bits;
yading@10:                 if (n <= 0)
yading@10:                     break;
yading@10:                 code = codes[k].code;
yading@10:                 if (code >> (32 - table_nb_bits) != code_prefix)
yading@10:                     break;
yading@10:                 codes[k].bits = n;
yading@10:                 codes[k].code = code << table_nb_bits;
yading@10:                 subtable_bits = FFMAX(subtable_bits, n);
yading@10:             }
yading@10:             subtable_bits = FFMIN(subtable_bits, table_nb_bits);
yading@10:             j = (flags & INIT_VLC_LE) ? bitswap_32(code_prefix) >> (32 - table_nb_bits) : code_prefix;
yading@10:             table[j][1] = -subtable_bits;
yading@10:             av_dlog(NULL, "%4x: n=%d (subtable)\n",
yading@10:                     j, codes[i].bits + table_nb_bits);
yading@10:             index = build_table(vlc, subtable_bits, k-i, codes+i, flags);
yading@10:             if (index < 0)
yading@10:                 return -1;
yading@10:             /* note: realloc has been done, so reload tables */
yading@10:             table = &vlc->table[table_index];
yading@10:             table[j][0] = index; //code
yading@10:             i = k-1;
yading@10:         }
yading@10:     }
yading@10:     return table_index;
yading@10: }
yading@10: 
yading@10: 
yading@10: /* Build VLC decoding tables suitable for use with get_vlc().
yading@10: 
yading@10:    'nb_bits' set thee decoding table size (2^nb_bits) entries. The
yading@10:    bigger it is, the faster is the decoding. But it should not be too
yading@10:    big to save memory and L1 cache. '9' is a good compromise.
yading@10: 
yading@10:    'nb_codes' : number of vlcs codes
yading@10: 
yading@10:    'bits' : table which gives the size (in bits) of each vlc code.
yading@10: 
yading@10:    'codes' : table which gives the bit pattern of of each vlc code.
yading@10: 
yading@10:    'symbols' : table which gives the values to be returned from get_vlc().
yading@10: 
yading@10:    'xxx_wrap' : give the number of bytes between each entry of the
yading@10:    'bits' or 'codes' tables.
yading@10: 
yading@10:    'xxx_size' : gives the number of bytes of each entry of the 'bits'
yading@10:    or 'codes' tables.
yading@10: 
yading@10:    'wrap' and 'size' allows to use any memory configuration and types
yading@10:    (byte/word/long) to store the 'bits', 'codes', and 'symbols' tables.
yading@10: 
yading@10:    'use_static' should be set to 1 for tables, which should be freed
yading@10:    with av_free_static(), 0 if ff_free_vlc() will be used.
yading@10: */
yading@10: int ff_init_vlc_sparse(VLC *vlc, int nb_bits, int nb_codes,
yading@10:              const void *bits, int bits_wrap, int bits_size,
yading@10:              const void *codes, int codes_wrap, int codes_size,
yading@10:              const void *symbols, int symbols_wrap, int symbols_size,
yading@10:              int flags)
yading@10: {
yading@10:     VLCcode *buf;
yading@10:     int i, j, ret;
yading@10: 
yading@10:     vlc->bits = nb_bits;
yading@10:     if(flags & INIT_VLC_USE_NEW_STATIC){
yading@10:         VLC dyn_vlc = *vlc;
yading@10: 
yading@10:         if (vlc->table_size)
yading@10:             return 0;
yading@10: 
yading@10:         ret = ff_init_vlc_sparse(&dyn_vlc, nb_bits, nb_codes,
yading@10:                                  bits, bits_wrap, bits_size,
yading@10:                                  codes, codes_wrap, codes_size,
yading@10:                                  symbols, symbols_wrap, symbols_size,
yading@10:                                  flags & ~INIT_VLC_USE_NEW_STATIC);
yading@10:         av_assert0(ret >= 0);
yading@10:         av_assert0(dyn_vlc.table_size <= vlc->table_allocated);
yading@10:         if(dyn_vlc.table_size < vlc->table_allocated)
yading@10:             av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", dyn_vlc.table_size, vlc->table_allocated);
yading@10:         memcpy(vlc->table, dyn_vlc.table, dyn_vlc.table_size * sizeof(*vlc->table));
yading@10:         vlc->table_size = dyn_vlc.table_size;
yading@10:         ff_free_vlc(&dyn_vlc);
yading@10:         return 0;
yading@10:     }else {
yading@10:         vlc->table = NULL;
yading@10:         vlc->table_allocated = 0;
yading@10:         vlc->table_size = 0;
yading@10:     }
yading@10: 
yading@10:     av_dlog(NULL, "build table nb_codes=%d\n", nb_codes);
yading@10: 
yading@10:     buf = av_malloc((nb_codes+1)*sizeof(VLCcode));
yading@10: 
yading@10:     av_assert0(symbols_size <= 2 || !symbols);
yading@10:     j = 0;
yading@10: #define COPY(condition)\
yading@10:     for (i = 0; i < nb_codes; i++) {\
yading@10:         GET_DATA(buf[j].bits, bits, i, bits_wrap, bits_size);\
yading@10:         if (!(condition))\
yading@10:             continue;\
yading@10:         if (buf[j].bits > 3*nb_bits || buf[j].bits>32) {\
yading@10:             av_log(NULL, AV_LOG_ERROR, "Too long VLC in init_vlc\n");\
yading@10:             return -1;\
yading@10:         }\
yading@10:         GET_DATA(buf[j].code, codes, i, codes_wrap, codes_size);\
yading@10:         if (buf[j].code >= (1LL<<buf[j].bits)) {\
yading@10:             av_log(NULL, AV_LOG_ERROR, "Invalid code in init_vlc\n");\
yading@10:             return -1;\
yading@10:         }\
yading@10:         if (flags & INIT_VLC_LE)\
yading@10:             buf[j].code = bitswap_32(buf[j].code);\
yading@10:         else\
yading@10:             buf[j].code <<= 32 - buf[j].bits;\
yading@10:         if (symbols)\
yading@10:             GET_DATA(buf[j].symbol, symbols, i, symbols_wrap, symbols_size)\
yading@10:         else\
yading@10:             buf[j].symbol = i;\
yading@10:         j++;\
yading@10:     }
yading@10:     COPY(buf[j].bits > nb_bits);
yading@10:     // qsort is the slowest part of init_vlc, and could probably be improved or avoided
yading@10:     qsort(buf, j, sizeof(VLCcode), compare_vlcspec);
yading@10:     COPY(buf[j].bits && buf[j].bits <= nb_bits);
yading@10:     nb_codes = j;
yading@10: 
yading@10:     ret = build_table(vlc, nb_bits, nb_codes, buf, flags);
yading@10: 
yading@10:     av_free(buf);
yading@10:     if (ret < 0) {
yading@10:         av_freep(&vlc->table);
yading@10:         return -1;
yading@10:     }
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: 
yading@10: void ff_free_vlc(VLC *vlc)
yading@10: {
yading@10:     av_freep(&vlc->table);
yading@10: }