Mercurial > libavcodec.hg
annotate common.c @ 487:cee02f2f4a5a libavcodec
grayscale only encoding
| author | michaelni |
|---|---|
| date | Sun, 09 Jun 2002 14:04:36 +0000 |
| parents | 297eed976b20 |
| children | 5b33d11bd1fb |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Common bit i/o utils | |
| 429 | 3 * Copyright (c) 2000, 2001 Fabrice Bellard. |
| 0 | 4 * |
| 429 | 5 * This library is free software; you can redistribute it and/or |
| 6 * modify it under the terms of the GNU Lesser General Public | |
| 7 * License as published by the Free Software Foundation; either | |
| 8 * version 2 of the License, or (at your option) any later version. | |
| 0 | 9 * |
| 429 | 10 * This library is distributed in the hope that it will be useful, |
| 0 | 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 429 | 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 * Lesser General Public License for more details. | |
| 0 | 14 * |
| 429 | 15 * You should have received a copy of the GNU Lesser General Public |
| 16 * License along with this library; if not, write to the Free Software | |
| 17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
18 * |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
19 * alternative bitstream reader & writer by Michael Niedermayer <michaelni@gmx.at> |
| 0 | 20 */ |
| 64 | 21 #include "common.h" |
| 8 | 22 |
| 0 | 23 void init_put_bits(PutBitContext *s, |
| 24 UINT8 *buffer, int buffer_size, | |
| 25 void *opaque, | |
| 26 void (*write_data)(void *, UINT8 *, int)) | |
| 27 { | |
| 28 s->buf = buffer; | |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
29 s->buf_end = s->buf + buffer_size; |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
30 s->data_out_size = 0; |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
31 if(write_data!=NULL) |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
32 { |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
33 fprintf(stderr, "write Data callback is not supported\n"); |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
34 } |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
35 #ifdef ALT_BITSTREAM_WRITER |
|
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
36 s->index=0; |
|
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
37 ((uint32_t*)(s->buf))[0]=0; |
|
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
38 // memset(buffer, 0, buffer_size); |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
39 #else |
| 0 | 40 s->buf_ptr = s->buf; |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
41 s->bit_left=32; |
| 0 | 42 s->bit_buf=0; |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
43 #endif |
| 0 | 44 } |
| 45 | |
| 46 /* return the number of bits output */ | |
| 64 | 47 INT64 get_bit_count(PutBitContext *s) |
| 0 | 48 { |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
49 #ifdef ALT_BITSTREAM_WRITER |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
50 return s->data_out_size * 8 + s->index; |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
51 #else |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
52 return (s->buf_ptr - s->buf + s->data_out_size) * 8 + 32 - (INT64)s->bit_left; |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
53 #endif |
| 0 | 54 } |
| 55 | |
| 56 void align_put_bits(PutBitContext *s) | |
| 57 { | |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
58 #ifdef ALT_BITSTREAM_WRITER |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
59 put_bits(s,( - s->index) & 7,0); |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
60 #else |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
61 put_bits(s,s->bit_left & 7,0); |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
62 #endif |
| 0 | 63 } |
| 64 | |
| 65 /* pad the end of the output stream with zeros */ | |
| 66 void flush_put_bits(PutBitContext *s) | |
| 67 { | |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
68 #ifdef ALT_BITSTREAM_WRITER |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
69 align_put_bits(s); |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
70 #else |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
71 s->bit_buf<<= s->bit_left; |
|
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
72 while (s->bit_left < 32) { |
| 0 | 73 /* XXX: should test end of buffer */ |
| 74 *s->buf_ptr++=s->bit_buf >> 24; | |
| 75 s->bit_buf<<=8; | |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
76 s->bit_left+=8; |
| 0 | 77 } |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
78 s->bit_left=32; |
| 0 | 79 s->bit_buf=0; |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
80 #endif |
| 0 | 81 } |
| 82 | |
| 83 /* pad the end of the output stream with zeros */ | |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
84 #ifndef ALT_BITSTREAM_WRITER |
| 0 | 85 void jflush_put_bits(PutBitContext *s) |
| 86 { | |
| 87 unsigned int b; | |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
88 s->bit_buf<<= s->bit_left; |
|
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
89 s->bit_buf |= ~1U >> (32 - s->bit_left); /* set all the unused bits to one */ |
| 0 | 90 |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
91 while (s->bit_left < 32) { |
| 0 | 92 b = s->bit_buf >> 24; |
| 93 *s->buf_ptr++ = b; | |
| 94 if (b == 0xff) | |
| 95 *s->buf_ptr++ = 0; | |
| 96 s->bit_buf<<=8; | |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
97 s->bit_left+=8; |
| 0 | 98 } |
|
238
99a9f903f0e3
optimized the normal bitstream writer, its faster than the alternative one on p3 now ... lets hope its at least not slower on p4 & k7
michaelni
parents:
234
diff
changeset
|
99 s->bit_left=32; |
| 0 | 100 s->bit_buf=0; |
| 101 } | |
|
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
102 #else |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
103 void jflush_put_bits(PutBitContext *s) |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
104 { |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
105 int num= ( - s->index) & 7; |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
106 jput_bits(s, num,0xFF>>(8-num)); |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
107 } |
|
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
219
diff
changeset
|
108 #endif |
| 0 | 109 |
| 358 | 110 void put_string(PutBitContext * pbc, char *s) |
| 111 { | |
| 112 while(*s){ | |
| 113 put_bits(pbc, 8, *s); | |
| 114 s++; | |
| 115 } | |
| 116 put_bits(pbc, 8, 0); | |
| 117 } | |
| 118 | |
| 0 | 119 /* bit input functions */ |
| 120 | |
| 121 void init_get_bits(GetBitContext *s, | |
| 122 UINT8 *buffer, int buffer_size) | |
| 123 { | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
124 #ifdef ALT_BITSTREAM_READER |
|
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
125 s->index=0; |
|
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
126 s->buffer= buffer; |
|
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
127 #else |
| 0 | 128 s->buf = buffer; |
| 129 s->buf_ptr = buffer; | |
| 130 s->buf_end = buffer + buffer_size; | |
| 131 s->bit_cnt = 0; | |
| 132 s->bit_buf = 0; | |
| 133 while (s->buf_ptr < s->buf_end && | |
| 134 s->bit_cnt < 32) { | |
| 135 s->bit_buf |= (*s->buf_ptr++ << (24 - s->bit_cnt)); | |
| 136 s->bit_cnt += 8; | |
| 137 } | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
138 #endif |
|
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
264
diff
changeset
|
139 s->size= buffer_size; |
| 0 | 140 } |
| 141 | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
142 #ifndef ALT_BITSTREAM_READER |
| 0 | 143 /* n must be >= 1 and <= 32 */ |
|
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
144 /* also true: n > s->bit_cnt */ |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
145 unsigned int get_bits_long(GetBitContext *s, int n) |
| 0 | 146 { |
| 147 unsigned int val; | |
| 148 int bit_cnt; | |
| 149 unsigned int bit_buf; | |
| 150 | |
| 151 #ifdef STATS | |
| 152 st_bit_counts[st_current_index] += n; | |
| 153 #endif | |
| 154 | |
| 155 bit_buf = s->bit_buf; | |
|
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
156 bit_cnt = s->bit_cnt - n; |
| 0 | 157 |
|
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
158 // if (bit_cnt >= 0) { |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
159 // val = bit_buf >> (32 - n); |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
160 // bit_buf <<= n; |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
161 // } else |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
162 { |
|
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
163 UINT8 *buf_ptr; |
| 0 | 164 val = bit_buf >> (32 - n); |
| 165 buf_ptr = s->buf_ptr; | |
| 166 buf_ptr += 4; | |
| 167 /* handle common case: we can read everything */ | |
| 168 if (buf_ptr <= s->buf_end) { | |
| 8 | 169 #if ARCH_X86 |
| 170 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4]))); | |
| 171 #else | |
| 172 bit_buf = (buf_ptr[-4] << 24) | | |
| 173 (buf_ptr[-3] << 16) | | |
| 0 | 174 (buf_ptr[-2] << 8) | |
| 8 | 175 (buf_ptr[-1]); |
| 176 #endif | |
| 452 | 177 val |= bit_buf >> (32 + bit_cnt); |
| 178 bit_buf <<= - bit_cnt; | |
| 179 bit_cnt += 32; | |
| 0 | 180 } else { |
| 181 buf_ptr -= 4; | |
| 182 bit_buf = 0; | |
| 183 if (buf_ptr < s->buf_end) | |
| 184 bit_buf |= *buf_ptr++ << 24; | |
| 185 if (buf_ptr < s->buf_end) | |
| 186 bit_buf |= *buf_ptr++ << 16; | |
| 187 if (buf_ptr < s->buf_end) | |
| 188 bit_buf |= *buf_ptr++ << 8; | |
| 189 if (buf_ptr < s->buf_end) | |
| 190 bit_buf |= *buf_ptr++; | |
| 452 | 191 |
| 192 val |= bit_buf >> (32 + bit_cnt); | |
| 193 bit_buf <<= - bit_cnt; | |
| 194 bit_cnt += 8*(buf_ptr - s->buf_ptr); | |
| 195 if(bit_cnt<0) bit_cnt=0; | |
| 0 | 196 } |
| 197 s->buf_ptr = buf_ptr; | |
| 198 } | |
| 199 s->bit_buf = bit_buf; | |
| 200 s->bit_cnt = bit_cnt; | |
| 201 return val; | |
| 202 } | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
203 #endif |
| 0 | 204 |
| 205 void align_get_bits(GetBitContext *s) | |
| 206 { | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
207 #ifdef ALT_BITSTREAM_READER |
|
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
208 s->index= (s->index + 7) & (~7); |
|
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
209 #else |
| 0 | 210 int n; |
| 211 n = s->bit_cnt & 7; | |
| 212 if (n > 0) { | |
| 213 get_bits(s, n); | |
| 214 } | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
215 #endif |
| 0 | 216 } |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
217 |
|
264
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
218 int check_marker(GetBitContext *s, char *msg) |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
219 { |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
220 int bit= get_bits1(s); |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
221 if(!bit) printf("Marker bit missing %s\n", msg); |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
222 |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
223 return bit; |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
224 } |
|
28c5c62b1c4c
support decoding (with mplayer) of 3 .mp4 files from mphq
michaelni
parents:
238
diff
changeset
|
225 |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
226 #ifndef ALT_BITSTREAM_READER |
| 144 | 227 /* This function is identical to get_bits_long(), the */ |
| 228 /* only diference is that it doesn't touch the buffer */ | |
| 229 /* it is usefull to see the buffer. */ | |
| 230 | |
| 231 unsigned int show_bits_long(GetBitContext *s, int n) | |
| 232 { | |
| 233 unsigned int val; | |
| 234 int bit_cnt; | |
| 235 unsigned int bit_buf; | |
| 236 UINT8 *buf_ptr; | |
| 237 | |
| 238 bit_buf = s->bit_buf; | |
| 239 bit_cnt = s->bit_cnt - n; | |
| 240 | |
| 241 val = bit_buf >> (32 - n); | |
| 242 buf_ptr = s->buf_ptr; | |
| 243 buf_ptr += 4; | |
| 244 | |
| 245 /* handle common case: we can read everything */ | |
| 246 if (buf_ptr <= s->buf_end) { | |
| 247 #ifdef ARCH_X86 | |
| 248 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4]))); | |
| 249 #else | |
| 250 bit_buf = (buf_ptr[-4] << 24) | | |
| 251 (buf_ptr[-3] << 16) | | |
| 252 (buf_ptr[-2] << 8) | | |
| 253 (buf_ptr[-1]); | |
| 254 #endif | |
| 255 } else { | |
| 256 buf_ptr -= 4; | |
| 257 bit_buf = 0; | |
| 258 if (buf_ptr < s->buf_end) | |
| 259 bit_buf |= *buf_ptr++ << 24; | |
| 260 if (buf_ptr < s->buf_end) | |
| 261 bit_buf |= *buf_ptr++ << 16; | |
| 262 if (buf_ptr < s->buf_end) | |
| 263 bit_buf |= *buf_ptr++ << 8; | |
| 264 if (buf_ptr < s->buf_end) | |
| 265 bit_buf |= *buf_ptr++; | |
| 266 } | |
| 267 val |= bit_buf >> (32 + bit_cnt); | |
| 268 bit_buf <<= - bit_cnt; | |
| 269 bit_cnt += 32; | |
| 270 | |
| 271 return val; | |
| 272 } | |
|
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
273 #endif |
| 0 | 274 |
| 275 /* VLC decoding */ | |
| 276 | |
| 277 //#define DEBUG_VLC | |
| 278 | |
| 279 #define GET_DATA(v, table, i, wrap, size) \ | |
| 280 {\ | |
| 281 UINT8 *ptr = (UINT8 *)table + i * wrap;\ | |
| 282 switch(size) {\ | |
| 283 case 1:\ | |
| 284 v = *(UINT8 *)ptr;\ | |
| 285 break;\ | |
| 286 case 2:\ | |
| 287 v = *(UINT16 *)ptr;\ | |
| 288 break;\ | |
| 289 default:\ | |
| 290 v = *(UINT32 *)ptr;\ | |
| 291 break;\ | |
| 292 }\ | |
| 293 } | |
| 294 | |
| 295 | |
| 296 static int alloc_table(VLC *vlc, int size) | |
| 297 { | |
| 298 int index; | |
| 299 index = vlc->table_size; | |
| 300 vlc->table_size += size; | |
| 301 if (vlc->table_size > vlc->table_allocated) { | |
| 302 vlc->table_allocated += (1 << vlc->bits); | |
| 303 vlc->table_bits = realloc(vlc->table_bits, | |
| 304 sizeof(INT8) * vlc->table_allocated); | |
| 305 vlc->table_codes = realloc(vlc->table_codes, | |
| 306 sizeof(INT16) * vlc->table_allocated); | |
| 307 if (!vlc->table_bits || | |
| 308 !vlc->table_codes) | |
| 309 return -1; | |
| 310 } | |
| 311 return index; | |
| 312 } | |
| 313 | |
| 314 static int build_table(VLC *vlc, int table_nb_bits, | |
| 315 int nb_codes, | |
| 316 const void *bits, int bits_wrap, int bits_size, | |
| 317 const void *codes, int codes_wrap, int codes_size, | |
| 318 UINT32 code_prefix, int n_prefix) | |
| 319 { | |
| 320 int i, j, k, n, table_size, table_index, nb, n1, index; | |
| 321 UINT32 code; | |
| 322 INT8 *table_bits; | |
| 323 INT16 *table_codes; | |
| 324 | |
| 325 table_size = 1 << table_nb_bits; | |
| 326 table_index = alloc_table(vlc, table_size); | |
| 327 #ifdef DEBUG_VLC | |
| 328 printf("new table index=%d size=%d code_prefix=%x n=%d\n", | |
| 329 table_index, table_size, code_prefix, n_prefix); | |
| 330 #endif | |
| 331 if (table_index < 0) | |
| 332 return -1; | |
| 333 table_bits = &vlc->table_bits[table_index]; | |
| 334 table_codes = &vlc->table_codes[table_index]; | |
| 335 | |
| 336 for(i=0;i<table_size;i++) { | |
| 337 table_bits[i] = 0; | |
| 338 table_codes[i] = -1; | |
| 339 } | |
| 340 | |
| 341 /* first pass: map codes and compute auxillary table sizes */ | |
| 342 for(i=0;i<nb_codes;i++) { | |
| 343 GET_DATA(n, bits, i, bits_wrap, bits_size); | |
| 344 GET_DATA(code, codes, i, codes_wrap, codes_size); | |
| 345 /* we accept tables with holes */ | |
| 346 if (n <= 0) | |
| 347 continue; | |
| 348 #if defined(DEBUG_VLC) && 0 | |
| 349 printf("i=%d n=%d code=0x%x\n", i, n, code); | |
| 350 #endif | |
| 351 /* if code matches the prefix, it is in the table */ | |
| 352 n -= n_prefix; | |
| 353 if (n > 0 && (code >> n) == code_prefix) { | |
| 354 if (n <= table_nb_bits) { | |
| 355 /* no need to add another table */ | |
| 356 j = (code << (table_nb_bits - n)) & (table_size - 1); | |
| 357 nb = 1 << (table_nb_bits - n); | |
| 358 for(k=0;k<nb;k++) { | |
| 359 #ifdef DEBUG_VLC | |
| 360 printf("%4x: code=%d n=%d\n", | |
| 361 j, i, n); | |
| 362 #endif | |
| 363 if (table_bits[j] != 0) { | |
| 364 fprintf(stderr, "incorrect codes\n"); | |
| 365 exit(1); | |
| 366 } | |
| 367 table_bits[j] = n; | |
| 368 table_codes[j] = i; | |
| 369 j++; | |
| 370 } | |
| 371 } else { | |
| 372 n -= table_nb_bits; | |
| 373 j = (code >> n) & ((1 << table_nb_bits) - 1); | |
| 374 #ifdef DEBUG_VLC | |
| 375 printf("%4x: n=%d (subtable)\n", | |
| 376 j, n); | |
| 377 #endif | |
| 378 /* compute table size */ | |
| 379 n1 = -table_bits[j]; | |
| 380 if (n > n1) | |
| 381 n1 = n; | |
| 382 table_bits[j] = -n1; | |
| 383 } | |
| 384 } | |
| 385 } | |
| 386 | |
| 387 /* second pass : fill auxillary tables recursively */ | |
| 388 for(i=0;i<table_size;i++) { | |
| 389 n = table_bits[i]; | |
| 390 if (n < 0) { | |
| 391 n = -n; | |
| 392 if (n > table_nb_bits) { | |
| 393 n = table_nb_bits; | |
| 394 table_bits[i] = -n; | |
| 395 } | |
| 396 index = build_table(vlc, n, nb_codes, | |
| 397 bits, bits_wrap, bits_size, | |
| 398 codes, codes_wrap, codes_size, | |
| 399 (code_prefix << table_nb_bits) | i, | |
| 400 n_prefix + table_nb_bits); | |
| 401 if (index < 0) | |
| 402 return -1; | |
| 403 /* note: realloc has been done, so reload tables */ | |
| 404 table_bits = &vlc->table_bits[table_index]; | |
| 405 table_codes = &vlc->table_codes[table_index]; | |
| 406 table_codes[i] = index; | |
| 407 } | |
| 408 } | |
| 409 return table_index; | |
| 410 } | |
| 411 | |
| 412 | |
| 24 | 413 /* Build VLC decoding tables suitable for use with get_vlc(). |
| 414 | |
| 415 'nb_bits' set thee decoding table size (2^nb_bits) entries. The | |
| 416 bigger it is, the faster is the decoding. But it should not be too | |
| 417 big to save memory and L1 cache. '9' is a good compromise. | |
| 418 | |
| 419 'nb_codes' : number of vlcs codes | |
| 420 | |
| 421 'bits' : table which gives the size (in bits) of each vlc code. | |
| 422 | |
| 423 'codes' : table which gives the bit pattern of of each vlc code. | |
| 424 | |
| 425 'xxx_wrap' : give the number of bytes between each entry of the | |
| 426 'bits' or 'codes' tables. | |
| 427 | |
| 428 'xxx_size' : gives the number of bytes of each entry of the 'bits' | |
| 429 or 'codes' tables. | |
| 430 | |
| 431 'wrap' and 'size' allows to use any memory configuration and types | |
| 432 (byte/word/long) to store the 'bits' and 'codes' tables. | |
| 433 */ | |
| 0 | 434 int init_vlc(VLC *vlc, int nb_bits, int nb_codes, |
| 435 const void *bits, int bits_wrap, int bits_size, | |
| 436 const void *codes, int codes_wrap, int codes_size) | |
| 437 { | |
| 438 vlc->bits = nb_bits; | |
| 439 vlc->table_bits = NULL; | |
| 440 vlc->table_codes = NULL; | |
| 441 vlc->table_allocated = 0; | |
| 442 vlc->table_size = 0; | |
| 443 #ifdef DEBUG_VLC | |
| 444 printf("build table nb_codes=%d\n", nb_codes); | |
| 445 #endif | |
| 446 | |
| 447 if (build_table(vlc, nb_bits, nb_codes, | |
| 448 bits, bits_wrap, bits_size, | |
| 449 codes, codes_wrap, codes_size, | |
| 450 0, 0) < 0) { | |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
358
diff
changeset
|
451 av_free(vlc->table_bits); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
358
diff
changeset
|
452 av_free(vlc->table_codes); |
| 0 | 453 return -1; |
| 454 } | |
| 455 return 0; | |
| 456 } | |
| 457 | |
| 458 | |
| 459 void free_vlc(VLC *vlc) | |
| 460 { | |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
358
diff
changeset
|
461 av_free(vlc->table_bits); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
358
diff
changeset
|
462 av_free(vlc->table_codes); |
| 0 | 463 } |
| 464 | |
| 324 | 465 int ff_gcd(int a, int b){ |
| 466 if(b) return ff_gcd(b, a%b); | |
| 467 else return a; | |
| 468 } |
